jackkuo commited on
Commit
4863c38
·
verified ·
1 Parent(s): 265a654

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +27 -0
  2. 19AzT4oBgHgl3EQfRvso/content/tmp_files/2301.01219v1.pdf.txt +2130 -0
  3. 19AzT4oBgHgl3EQfRvso/content/tmp_files/load_file.txt +0 -0
  4. 3NAzT4oBgHgl3EQfR_tE/content/tmp_files/2301.01224v1.pdf.txt +1827 -0
  5. 3NAzT4oBgHgl3EQfR_tE/content/tmp_files/load_file.txt +0 -0
  6. 59AyT4oBgHgl3EQf2fl-/vector_store/index.pkl +3 -0
  7. 69E5T4oBgHgl3EQfPw7T/content/2301.05508v1.pdf +3 -0
  8. 69E5T4oBgHgl3EQfPw7T/vector_store/index.pkl +3 -0
  9. 7tAzT4oBgHgl3EQfSPsA/content/tmp_files/2301.01227v1.pdf.txt +0 -0
  10. 7tAzT4oBgHgl3EQfSPsA/content/tmp_files/load_file.txt +0 -0
  11. 8NFAT4oBgHgl3EQfox1h/content/tmp_files/2301.08636v1.pdf.txt +987 -0
  12. 8NFAT4oBgHgl3EQfox1h/content/tmp_files/load_file.txt +0 -0
  13. 9NFJT4oBgHgl3EQfoSxe/content/tmp_files/2301.11595v1.pdf.txt +1617 -0
  14. 9NFJT4oBgHgl3EQfoSxe/content/tmp_files/load_file.txt +0 -0
  15. AtAyT4oBgHgl3EQfd_hU/content/2301.00312v1.pdf +3 -0
  16. AtAyT4oBgHgl3EQfd_hU/vector_store/index.pkl +3 -0
  17. BNE0T4oBgHgl3EQfxwK1/vector_store/index.pkl +3 -0
  18. BdE0T4oBgHgl3EQfyAIA/content/tmp_files/2301.02652v1.pdf.txt +1334 -0
  19. BdE0T4oBgHgl3EQfyAIA/content/tmp_files/load_file.txt +0 -0
  20. BdE3T4oBgHgl3EQftAtl/content/tmp_files/2301.04672v1.pdf.txt +897 -0
  21. BdE3T4oBgHgl3EQftAtl/content/tmp_files/load_file.txt +0 -0
  22. BtFAT4oBgHgl3EQfsR7g/vector_store/index.faiss +3 -0
  23. CNFRT4oBgHgl3EQfvzii/content/2301.13636v1.pdf +3 -0
  24. CNFRT4oBgHgl3EQfvzii/vector_store/index.faiss +3 -0
  25. CNFRT4oBgHgl3EQfvzii/vector_store/index.pkl +3 -0
  26. CtAyT4oBgHgl3EQfR_f1/vector_store/index.faiss +3 -0
  27. DdAzT4oBgHgl3EQfif0c/vector_store/index.faiss +3 -0
  28. E9E0T4oBgHgl3EQfQwCg/content/tmp_files/2301.02198v1.pdf.txt +1971 -0
  29. E9E0T4oBgHgl3EQfQwCg/content/tmp_files/load_file.txt +0 -0
  30. EtAzT4oBgHgl3EQfUPx6/vector_store/index.faiss +3 -0
  31. F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf +0 -0
  32. F9E3T4oBgHgl3EQfWAol/content/tmp_files/2301.04464v1.pdf.txt +176 -0
  33. F9E3T4oBgHgl3EQfWAol/content/tmp_files/load_file.txt +107 -0
  34. FNAyT4oBgHgl3EQf4_q8/vector_store/index.faiss +3 -0
  35. GNE1T4oBgHgl3EQfFANy/content/tmp_files/2301.02897v1.pdf.txt +1197 -0
  36. GNE1T4oBgHgl3EQfFANy/content/tmp_files/load_file.txt +0 -0
  37. GtE5T4oBgHgl3EQfWA9Z/content/tmp_files/2301.05555v1.pdf.txt +1575 -0
  38. GtE5T4oBgHgl3EQfWA9Z/content/tmp_files/load_file.txt +0 -0
  39. HtAzT4oBgHgl3EQfjf0-/content/2301.01516v1.pdf +3 -0
  40. ItFLT4oBgHgl3EQfJS9D/content/2301.12003v1.pdf +3 -0
  41. ItFLT4oBgHgl3EQfJS9D/vector_store/index.pkl +3 -0
  42. JNFAT4oBgHgl3EQfvB44/content/tmp_files/2301.08673v1.pdf.txt +887 -0
  43. JNFAT4oBgHgl3EQfvB44/content/tmp_files/load_file.txt +0 -0
  44. M9AzT4oBgHgl3EQfWPwO/content/tmp_files/2301.01296v1.pdf.txt +1618 -0
  45. M9AzT4oBgHgl3EQfWPwO/content/tmp_files/load_file.txt +0 -0
  46. P9AzT4oBgHgl3EQflf2o/content/2301.01549v1.pdf +3 -0
  47. P9AzT4oBgHgl3EQflf2o/vector_store/index.faiss +3 -0
  48. P9AzT4oBgHgl3EQflf2o/vector_store/index.pkl +3 -0
  49. PdAzT4oBgHgl3EQfzv5V/content/2301.01773v1.pdf +3 -0
  50. PdAzT4oBgHgl3EQfzv5V/vector_store/index.pkl +3 -0
.gitattributes CHANGED
@@ -284,3 +284,30 @@ NNAyT4oBgHgl3EQfgvgA/content/2301.00362v1.pdf filter=lfs diff=lfs merge=lfs -tex
284
  NNE3T4oBgHgl3EQfBglT/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
285
  FtE2T4oBgHgl3EQfTAfF/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
286
  FtE2T4oBgHgl3EQfTAfF/content/2301.03799v1.pdf filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
284
  NNE3T4oBgHgl3EQfBglT/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
285
  FtE2T4oBgHgl3EQfTAfF/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
286
  FtE2T4oBgHgl3EQfTAfF/content/2301.03799v1.pdf filter=lfs diff=lfs merge=lfs -text
287
+ DdAzT4oBgHgl3EQfif0c/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
288
+ h9AzT4oBgHgl3EQfbPwA/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
289
+ zdE2T4oBgHgl3EQfMgbP/content/2301.03726v1.pdf filter=lfs diff=lfs merge=lfs -text
290
+ zdE2T4oBgHgl3EQfMgbP/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
291
+ P9AzT4oBgHgl3EQflf2o/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
292
+ ddAzT4oBgHgl3EQfn_0r/content/2301.01588v1.pdf filter=lfs diff=lfs merge=lfs -text
293
+ VtFKT4oBgHgl3EQfmy5j/content/2301.11859v1.pdf filter=lfs diff=lfs merge=lfs -text
294
+ RtFRT4oBgHgl3EQfLTfA/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
295
+ CNFRT4oBgHgl3EQfvzii/content/2301.13636v1.pdf filter=lfs diff=lfs merge=lfs -text
296
+ CNFRT4oBgHgl3EQfvzii/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
297
+ 69E5T4oBgHgl3EQfPw7T/content/2301.05508v1.pdf filter=lfs diff=lfs merge=lfs -text
298
+ EtAzT4oBgHgl3EQfUPx6/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
299
+ xtE5T4oBgHgl3EQfNA53/content/2301.05486v1.pdf filter=lfs diff=lfs merge=lfs -text
300
+ mtAyT4oBgHgl3EQflPh9/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
301
+ RtFRT4oBgHgl3EQfLTfA/content/2301.13502v1.pdf filter=lfs diff=lfs merge=lfs -text
302
+ CtAyT4oBgHgl3EQfR_f1/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
303
+ VtFKT4oBgHgl3EQfmy5j/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
304
+ P9AzT4oBgHgl3EQflf2o/content/2301.01549v1.pdf filter=lfs diff=lfs merge=lfs -text
305
+ PdAzT4oBgHgl3EQfzv5V/content/2301.01773v1.pdf filter=lfs diff=lfs merge=lfs -text
306
+ btA0T4oBgHgl3EQfGf-a/content/2301.02048v1.pdf filter=lfs diff=lfs merge=lfs -text
307
+ pNE0T4oBgHgl3EQfqgGa/content/2301.02554v1.pdf filter=lfs diff=lfs merge=lfs -text
308
+ BtFAT4oBgHgl3EQfsR7g/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
309
+ FNAyT4oBgHgl3EQf4_q8/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
310
+ btA0T4oBgHgl3EQfGf-a/vector_store/index.faiss filter=lfs diff=lfs merge=lfs -text
311
+ AtAyT4oBgHgl3EQfd_hU/content/2301.00312v1.pdf filter=lfs diff=lfs merge=lfs -text
312
+ ItFLT4oBgHgl3EQfJS9D/content/2301.12003v1.pdf filter=lfs diff=lfs merge=lfs -text
313
+ HtAzT4oBgHgl3EQfjf0-/content/2301.01516v1.pdf filter=lfs diff=lfs merge=lfs -text
19AzT4oBgHgl3EQfRvso/content/tmp_files/2301.01219v1.pdf.txt ADDED
@@ -0,0 +1,2130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Task-Guided IRL in POMDPs that Scales
2
+ Franck Djeumou∗, Christian Ellis∗∗, Murat Cubuktepe∗, Craig Lennon, Ufuk Topcu∗
3
+ Abstract
4
+ In inverse reinforcement learning (IRL), a learning agent infers a reward function en-
5
+ coding the underlying task using demonstrations from experts. However, many ex-
6
+ isting IRL techniques make the often unrealistic assumption that the agent has access
7
+ to full information about the environment. We remove this assumption by developing
8
+ an algorithm for IRL in partially observable Markov decision processes (POMDPs).
9
+ We address two limitations of existing IRL techniques. First, they require an exces-
10
+ sive amount of data due to the information asymmetry between the expert and the
11
+ learner. Second, most of these IRL techniques require solving the computationally in-
12
+ tractable forward problem—computing an optimal policy given a reward function—in
13
+ POMDPs. The developed algorithm reduces the information asymmetry while increas-
14
+ ing the data efficiency by incorporating task specifications expressed in temporal logic
15
+ into IRL. Such specifications may be interpreted as side information available to the
16
+ learner a priori in addition to the demonstrations. Further, the algorithm avoids a com-
17
+ mon source of algorithmic complexity by building on causal entropy as the measure of
18
+ the likelihood of the demonstrations as opposed to entropy. Nevertheless, the resulting
19
+ problem is nonconvex due to the so-called forward problem. We solve the intrinsic
20
+ nonconvexity of the forward problem in a scalable manner through a sequential linear
21
+ programming scheme that guarantees to converge to a locally optimal policy. In a series
22
+ of examples, including experiments in a high-fidelity Unity simulator, we demonstrate
23
+ that even with a limited amount of data and POMDPs with tens of thousands of states,
24
+ our algorithm learns reward functions and policies that satisfy the task while inducing
25
+ similar behavior to the expert by leveraging the provided side information.
26
+ 1. Introduction
27
+ A robot can satisfy certain human-specified tasks by describing desired behavior
28
+ through a reward function. However, the design of such a reward function is a non-
29
+ trivial task. Inverse reinforcement learning (IRL) is an established technique that in-
30
+ fers a reward function encoding the underlying task using expert demonstrations. IRL
31
+ ∗The University of Texas at Austin
32
+ ∗∗The University of Massachusetts: Dartmouth
33
+ United States Army Research Laboratory
34
+ Email addresses: fdjeumou@utexas.edu (Franck Djeumou), cellis3@umassd.edu
35
+ (Christian Ellis), mcubuktepe@utexas.edu (Murat Cubuktepe),
36
+ craig.t.lennon.civ@army.mil (Craig Lennon), utopcu@utexas.edu (Ufuk Topcu)
37
+ Preprint submitted to Elsevier
38
+ January 4, 2023
39
+ arXiv:2301.01219v1 [cs.LG] 30 Dec 2022
40
+
41
+ techniques have found a wide range of applications in various domains such as ac-
42
+ robatic helicopter flight [1], inferring future actions of people [2], human-autonomy
43
+ interaction [3, 4], robotic surgery [5, 6], and robotic manipulation tasks [7]. Most
44
+ existing work [1, 8, 9, 10, 3, 7] has focused on Markov decision processes (MDPs),
45
+ assuming that the learner can fully observe the state of the environment and expert
46
+ demonstrations. However, the learner will not have access to full state observations in
47
+ many applications. For example, a robot will never know everything about its envi-
48
+ ronment [11, 12, 13] and may not observe the internal states of a human with whom it
49
+ works [14, 15]. Such information limitations violate the intrinsic assumptions made in
50
+ most existing IRL techniques.
51
+ We investigate IRL in partially observable Markov decision processes (POMDPs),
52
+ a widely used model for decision-making under imperfect information. Partial observ-
53
+ ability brings two key challenges in IRL. The first challenge is due to the so-called
54
+ information asymmetry between the expert and the learner. The expert typically has
55
+ either full or partial information about the environment, while the learner has only a
56
+ partial view of the state and the expert’s demonstrations. Even in the hypothetical
57
+ case in which the underlying reward function is known to the learner, its optimal pol-
58
+ icy under limited information may not yield the same behavior as an expert with full
59
+ information due to such information asymmetry.
60
+ The second challenge is due to the computational complexity of policy synthesis in
61
+ POMDPs. Indeed, many standard IRL techniques rely on a subroutine that solves the
62
+ so-called forward problem, i.e., computing an optimal policy for a given reward. Solv-
63
+ ing the forward problem for POMDPs is significantly more challenging than MDPs,
64
+ both theoretically and practically [16]. Optimal policies for POMDPs may require infi-
65
+ nite memory of observations [17], whereas memoryless policies are enough for MDPs.
66
+ An additional limitation in existing IRL techniques is due to the limited expressiv-
67
+ ity and often impracticability of state-based reward functions in representing complex
68
+ tasks [18]. For example, it will be tremendously difficult to define a merely state-based
69
+ reward function to describe requirements such as “do not steer off the road while reach-
70
+ ing the target location and coming back to home” or “monitor multiple locations with
71
+ a certain order”. However, such requirements can be concisely and precisely speci-
72
+ fied in temporal logic [19, 20]. Therefore, recent work has demonstrated the utility of
73
+ incorporating temporal logic specifications into IRL in MDPs [21, 22].
74
+ In this work, we address these challenges and limitations in state-of-the-art IRL
75
+ techniques by investigating the following problem.
76
+ Task-Guided IRL in POMDPs: Given a POMDP, a set of expert demonstrations,
77
+ and, if available, a task specification expressed in temporal logic, learn a policy
78
+ along with the underlying reward function that maximizes the causal entropy of
79
+ the induced stochastic process, induces a behavior similar to the expert’s, and
80
+ ensures the satisfaction of the specification.
81
+ We highlight two parts of the problem statement. Using causal entropy as an opti-
82
+ mization criterion instead of traditional entropy results in a least-committal policy that
83
+ induces a behavior obtaining the same accumulated reward as the expert’s demonstra-
84
+ tions while making no additional assumptions about the demonstrations. Task specifi-
85
+ 2
86
+
87
+ cations given as task requirements guide the learning process by describing the feasible
88
+ behaviors and allow the learner to learn performant policies with respect to the task re-
89
+ quirements. Such specifications can be interpreted as side information available to
90
+ the learner a priori in addition to the demonstrations aimed at partially alleviating the
91
+ information asymmetry between the expert and the learner.
92
+ Specifically, we tackle the IRL on POMDPs problem by a reformulation into a
93
+ maximum causal entropy (MCE) problem. Then, we develop a new solver for the
94
+ MCE problem that improves computational tractability over existing approaches. The
95
+ developed solver can enforce prior task knowledge expressed as temporal logic specifi-
96
+ cations, which guides the learning, improves the data efficiency, and partially alleviates
97
+ the information asymmetry problem.
98
+ Most existing work on IRL relies on entropy as a measure of the likelihood of the
99
+ demonstrations, yet, when applied to stochastic MDPs, has to deal with nonconvex
100
+ optimization problems [8, 10]. On the other hand, IRL techniques that adopt causal
101
+ entropy as the measure of likelihood enjoy formulations based on convex optimiza-
102
+ tion [9, 10, 23]. We show similar algorithmic benefits in maximum-causal-entropy
103
+ IRL carry over from MDPs to POMDPs.
104
+ A major difference between MDPs and POMDPs in maximum-causal-entropy IRL
105
+ is, though, due to the intrinsic nonconvexity of policy synthesis in POMDPs, which
106
+ yields a formulation of the task-guided IRL problem as a nonconvex optimization
107
+ problem. It is known that this nonconvexity severely limits the scalability for syn-
108
+ thesis in POMDPs [16]. We develop an iterative algorithm that solves the resulting
109
+ nonconvex problem in a scalable manner by adapting sequential convex programming
110
+ (SCP) [24, 25].
111
+ In each iteration, it linearizes the underlying nonconvex problem
112
+ around the solution from the previous iteration. The algorithm introduces several ex-
113
+ tensions to alleviate the errors resulting from the linearization. One of these extensions
114
+ is a verification step not present in existing SCP schemes. We show that the proposed
115
+ algorithm computes a sound and locally optimal solution to the task-guided problem.
116
+ Additionally, we empirically demonstrate that the algorithm scales to POMDPs
117
+ with tens of thousands of states as opposed to tens of states in most existing work.
118
+ In POMDPs, finite-memory policies that are functions of the history of the observa-
119
+ tions outperform memoryless policies [26]. Besides, computing a finite-memory pol-
120
+ icy for a POMDP is equivalent to computing a memoryless policy on a larger product
121
+ POMDP [27]. Thus, we leverage the scalability of our algorithm to compute more per-
122
+ formant policies that incorporate memory using finite-state controllers [28, 29]. On the
123
+ other hand, the existing IRL techniques on POMDPs aforementioned cannot effectively
124
+ utilize memory, as they do not scale to large POMDPs.
125
+ We demonstrate the applicability of the approach through several examples, in-
126
+ cluding a simulated wheeled ground robot operating in a high-fidelity, continuous, 3-
127
+ D Unity simulation. We show that, without task specifications, the developed algo-
128
+ rithm can compute more performant policies than a straight adaptation of the original
129
+ GAIL [30] to POMDPs. Then, we demonstrate that by incorporating task specifications
130
+ into the IRL procedure, the learned reward function and policy accurately describe
131
+ the behavior of the expert while outperforming the policy obtained without the task
132
+ specifications. We observe that with more limited data, the performance gap becomes
133
+ more prominent between the learned policies with and without using task specifica-
134
+ 3
135
+
136
+ tions. Most importantly, we empirically demonstrate the scalability of our approach
137
+ for solving the forward problem through extensive comparisons with several state-of-
138
+ the-art POMDP solvers and show that on larger POMDPs, the algorithm can compute
139
+ more performant policies in significantly less time.
140
+ 2. Preliminaries
141
+ The following section provides a review of prerequisite understanding for POMDPs,
142
+ their accompanying policies and how a POMDP’s belief over states is updated using
143
+ Bayesian techniques.
144
+ Notation. We denote the set of nonnegative real numbers by R+, the set of all proba-
145
+ bility distributions over a finite or countably infinite set X by Distr(X), the set of all
146
+ (infinite or empty) sequences x0, x1, . . . , x∞ with xi ∈ X by (X)∗ for some set X,
147
+ and the expectation of a function g of jointly distributed random variables X and Y by
148
+ EX,Y [g(X, Y )].
149
+ 2.1. Partially Observable Markov Decision Process
150
+ A partially observable Markov decision process (POMDP) is a framework for mod-
151
+ eling sequential interaction between an agent and a partially observable environment,
152
+ where the agent cannot perceive its underlying state but must infer it based on the given
153
+ noisy observation.
154
+ POMDPs. We define a POMDP by a tuple M = (S, A, P, Z, O, R, µ0, γ), where S,
155
+ A, and Z are finite sets of states, actions, and observations, respectively. The function
156
+ µ0 : S �→ R+ provides the initial distribution of the agent’s state and γ ∈ [0, 1) is
157
+ a discount factor over a possibly infinite planning horizon. At each decision time, an
158
+ agent selects an action α ∈ A and the transition function P : S × A �→ Distr(S)
159
+ defines the probability P(s′|s, α) of reaching state s′ ∈ S given the current state s ∈ S
160
+ and action α. After the state transition, the agent receives an observation z′ ∈ Z
161
+ according to the function O : S �→ Distr(Z), which defines the probability O(z′|s′)
162
+ of perceiving z′ at state s′. The agent also receives a reward function R(s, α) from the
163
+ function R : S × A �→ R encoding the task specification. In the following, without
164
+ loss of generality, we consider infinite-horizon problems.
165
+ Policies. An observation-based policy σ : (Z × A)∗ × Z �→ Distr(A) for a POMDP
166
+ M maps a sequence of observations and actions to a distribution over actions. A M-
167
+ finite-state controller (M-FSC) is a tuple C = (Q, qI, η, δ), where Q = {q1, q2, . . . , qM}
168
+ is a finite set of memory states, qI is the initial memory state, η : Q×Z �→ Distr(A) is
169
+ a decision function, and δ : Q × Z × A �→ Distr(Q) is a memory transition function.
170
+ The action mapping η(n, z) takes a FSC memory state n and an observation z ∈ Z,
171
+ and returns a distribution over the POMDP actions. The memory update δ(n, z, α) re-
172
+ turns a distribution over memory states and is a function of the action α selected by η.
173
+ An FSC induces an observation-based policy by following a joint execution of these
174
+ two functions upon a trace of the POMDP. An FSC is memoryless if there is a single
175
+ 4
176
+
177
+ memory state. Memoryless FSCs, denoted by σ: Z → Distr(A), are observation-
178
+ based policies, where σ(α|z) = σz,α is the probability of taking the action α given
179
+ solely observation z.
180
+ Remark 1 (REDUCTION TO MEMORYLESS POLICIES). In the remainder of the pa-
181
+ per, for ease of notation, we synthesize optimal M-FSCs for POMDPs (so-called for-
182
+ ward problem) by computing memoryless policies σ on theoretically-justified larger
183
+ POMDPs obtained from the so-called product of the memory update δ and the original
184
+ POMDPs. Indeed, the authors of [27] provide product POMDPs, whose sizes grow
185
+ polynomially only with the size of the domain of δ.
186
+ Belief Update. Given a history on the POMDP M as the perceived observation and
187
+ executed action sequence τ = {(z0, α0), (z1, α1), . . . , (zT , αT )}, where zi ∈ Z, αi ∈
188
+ A, i ∈ {0, . . . , T} and T is the length of the trajectory, the belief state specifies the
189
+ probability of being in each state of the POMDP given an initial belief b0 = µ0. Such
190
+ a belief state can be updated at each time step using the following Bayes rule
191
+ bt+1(s′) =
192
+ O(zt|s′) �
193
+ s∈S P(s′|s, αt)bt(s)
194
+
195
+ s′′∈S O(zt|s′′) �
196
+ s∈S P(s′′|s, αt)bt(s).
197
+ (1)
198
+ 2.2. Causal Entropy in POMDPs.
199
+ For a POMDP M, a policy σ induces the stochastic processes Sσ
200
+ 0:∞ := (Sσ
201
+ 0 , . . . , Sσ
202
+ ∞),
203
+
204
+ 0:∞ := (Aσ
205
+ 0, . . . , Aσ
206
+ ∞), and Zσ
207
+ 0:∞ := (Zσ
208
+ 0 , . . . , Zσ
209
+ ∞). At each time index t, the ran-
210
+ dom variables Sσ
211
+ t , Aσ
212
+ t , and Zσ
213
+ t take values st ∈ S, αt ∈ A, and zt ∈ Z, respec-
214
+ tively. The probability P(A0:T ||S0:T ) of A0:T causally-conditioned on S0:T , given
215
+ by [10, 31, 32] P(A0:T ||S0:T ) := �T
216
+ t=0 P(At|S0:t, A0:t−1), defines a correlation be-
217
+ tween the stochastic processes, where each variable At is conditionally influenced by
218
+ only the earlier predicted variables S0:t, A0:t−1, and not the future variables St+1:T .
219
+ Let H(A|S) ≜ EA,S[− log P(A|S)] be the conditional entropy of a random variable
220
+ A given a random variable S. In the finite-horizon setting, the causal entropy Hσ in-
221
+ duced by a given policy σ is defined as Hσ := EAσ
222
+ 0:T ,Sσ
223
+ 0:T [− log P(Aσ
224
+ 0:T ||Sσ
225
+ 0:T )] =
226
+ �T
227
+ t=0 H(Aσ
228
+ t |Sσ
229
+ 0:t, Aσ
230
+ 0:t−1). Then, the causal entropy in the infinite-horizon setting,
231
+ namely the discounted causal entropy [9, 33], is defined as
232
+
233
+ σ :=
234
+ �∞
235
+ t=0 γtH(Aσ
236
+ t |Sσ
237
+ 0:t, Aσ
238
+ 0:t−1) =
239
+ �∞
240
+ t=0 γtEAσ
241
+ t ,Sσ
242
+ t [− log P(Aσ
243
+ t |Sσ
244
+ t )],
245
+ (2)
246
+ where the second equality is due to the Markov property.
247
+ Remark 2. The entropy of POMDPs (or MDPs) involves the future policy decisions [8],
248
+ i.e., Sσ
249
+ t+1:T , at a time index t, as opposed to the causal entropy in POMDPs (or MDPs).
250
+ Thus, the authors of [8] show that the problem of computing a policy that maximizes
251
+ the entropy is nonconvex, even in MDPs. Inverse reinforcement learning techniques
252
+ that maximize the entropy of the policy rely on approximations or assume that the tran-
253
+ sition function of the MDP is deterministic. On the other hand, computing a policy that
254
+ maximizes the causal entropy can be formulated as a convex optimization problem in
255
+ MDPs [10, 9].
256
+ 5
257
+
258
+ 2.3. LTL Specifications.
259
+ We use general linear temporal logic (LTL) to express complex task specifications
260
+ on the POMDP M. Given a set AP of atomic propositions, i.e., Boolean variables
261
+ with truth values for a given state s or observation z, LTL formulae are constructed
262
+ inductively as following:
263
+ ϕ := true | a | ¬ϕ | ϕ1 ∧ ϕ2 | Xϕ | ϕ1Uϕ2,
264
+ where a ∈ AP, ϕ, ϕ1, and ϕ2 are LTL formulae, ¬ and ∧ are the logic negation and
265
+ conjunction, and X and U are the next and until temporal operators. Besides, temporal
266
+ operators such as always (G) and eventually (F) are derived as Fϕ := trueUϕ and
267
+ Gϕ := ¬F¬ϕ. We denote by Prσ
268
+ M(ϕ) the probability of satisfying the LTL formula ϕ
269
+ when following the policy σ on the POMDP M. A detailed description of the syntax
270
+ and semantics of LTL is beyond the scope of this paper and can be found in [20, 19].
271
+ 3. Problem Formulation
272
+ In this section, we formulate the problem of task-guided inverse reinforcement
273
+ learning (IRL) in POMDPs. Given a POMDP M with an unknown reward function
274
+ R, we seek to learn a reward function R along with an underlying policy σ that in-
275
+ duces a behavior similar to the expert demonstrations.
276
+ We define an expert trajectory on the POMDP M as the perceived observation and
277
+ executed action sequence τ = {(z0, α0), (z1, α1), . . . , (zT , αT )}, where zi ∈ Z and
278
+ αi ∈ A for all i ∈ {0, . . . , T}, and T denotes the length of the trajectory. Similarly to
279
+ [34], we assume given or we can construct from τ (via Bayesian belief updates (1)) the
280
+ belief trajectory bτ = {b0 := µ0, . . . , bT }, where bi(s) is the estimated probability of
281
+ being at state s at time index i. In the following, we assume that we are given a set of
282
+ belief trajectories D = {bτ1, . . . , bτN } from trajectories τ1, . . . , τN, where N denotes
283
+ the total number of underlying trajectories.
284
+ We parameterize the unknown reward function R by a differentiable function (with
285
+ respect to the parameter) Rθ : S × A �→ Rd, where θ ∈ RF is a parameter that defines
286
+ uniquely the reward function. Such an encoding includes traditional representations of
287
+ the reward such as Rθ(s, α) = gθ(φ(s, α)), where φ : S × A �→ Rd is a known vector
288
+ of basis functions with components referred to as feature functions, d is the number
289
+ of features, and gθ can be any function approximator such as neural networks. For
290
+ example, in the traditional linear encoding, we have gθ(z) = θTz.
291
+ Specifically, we seek for a parameter θ defining Rθ and a policy σ such that its
292
+ discounted return expectation Rθ
293
+ σ matches an empirical discounted return expectation
294
+ ¯Rθ of the expert demonstration D. That is, we have that Rθ
295
+ σ = ¯Rθ, where
296
+
297
+ σ :=
298
+
299
+
300
+ t=0
301
+ γtESσ
302
+ t ,Aσ
303
+ t [Rθ(Sσ
304
+ t , Aσ
305
+ t )|σ] and ¯Rθ = 1
306
+ N
307
+
308
+ bτ ∈D
309
+
310
+ bi∈bτ
311
+ γi �
312
+ s∈S
313
+ bi(s)Rθ(s, αi).
314
+ In the case of linear encoding of the reward, the above condition is called feature match-
315
+ ing expectation, and it can be simplified by replacing Rθ with the feature function φ.
316
+ 6
317
+
318
+ Nevertheless, the problem is ill-posed and there may be infinitely many reward
319
+ functions and policies that can satisfy the above matching condition. To resolve the
320
+ ambiguities, we seek for a policy σ that also maximizes the discounted causal entropy
321
+
322
+ σ. We now define the problem of interest.
323
+ Problem 1. Given a reward-free POMDP M, a demonstration set D, and a feature φ,
324
+ compute a policy σ and weight θ such that (a) The matching condition holds; (b) The
325
+ causal entropy Hγ
326
+ σ given by (2) is maximized by σ.
327
+ Furthermore, we seek to incorporate, if available, a priori high-level side informa-
328
+ tion on the task demonstrated by the expert in the design of the reward and policy.
329
+ Problem 2. Given a linear temporal logic formula ϕ, compute a policy σ and weight
330
+ θ such that the constraints (a) and (b) in Problem 1 are satisfied, and Prσ
331
+ M(ϕ) ��� λ for
332
+ a given parameter λ ≥ 0.
333
+ Although the parameter λ that specifies the threshold for satisfaction of ϕ is as-
334
+ sumed to be given, the approach can easily be adapted to compute the optimal λ.
335
+ 4. Nonconvex Formulation for IRL in POMDPs
336
+ In this section, we formulate Problem 1 and Problem 2 as finding saddle points of
337
+ a nonconvex functions. Then, we propose an algorithm based on solving a nonconvex
338
+ optimization problem to compute such saddle points. We emphasize (see Remark 1)
339
+ that we compute M-FSC for POMDPs by computing memoryless policies σ on larger
340
+ product POMDPs. Indeed, in the remainder of the paper, we reason directly on the
341
+ product POMDP, which is the product of a POMDP and an FSC, and it yields a POMDP
342
+ with state memory pairs [27].
343
+ Substituting Visitation Counts. We eliminate the (infinite) time dependency in Hγ
344
+ σ
345
+ and the matching condition by a substitution of variables involving the policy-induced
346
+ discounted state visitation count µγ
347
+ σ : S �→ R+ and state-action visitation count νγ
348
+ σ :
349
+ S×A �→ R+. For a policy σ, state s, and action α, the discounted state and state-action
350
+ visitation counts are defined by
351
+ µγ
352
+ σ(s) := ESt[
353
+
354
+
355
+ t=1
356
+ γt1{St=s}|σ] and νγ
357
+ σ(s, α) := EAt,St[
358
+
359
+
360
+ t=1
361
+ γt1{St=s,At=α}|σ],
362
+ where 1{·} is the indicator function. From these definitions, it is straightforward to
363
+ deduce that νγ
364
+ σ(s, α) = πs,αµγ
365
+ σ(s), where πs,α = P[At = a|St = s]. It is also
366
+ straightforward to check that for all s ∈ S and α ∈ A, µγ
367
+ σ(s) ≥ 0, νγ
368
+ σ(s, α) ≥ 0, and
369
+ µγ
370
+ σ(s) = �
371
+ α∈A νγ
372
+ σ(s, α).
373
+ We first provide a concave expression for the discounted causal entropy Hγ
374
+ σ as a
375
+ 7
376
+
377
+ function of the visitation counts µγ
378
+ σ and νγ
379
+ σ:
380
+
381
+ σ :=
382
+ �∞
383
+ t=0 γtESσ
384
+ t ,Aσ
385
+ t [− log(πst,αt)]
386
+ =
387
+ �∞
388
+ t=0
389
+
390
+ (s,α)∈S×A −(log πs,α)πs,αγtP[Sσ
391
+ t = s]
392
+ =
393
+
394
+ (s,α)∈S×A −(log πs,α)πs,αµγ
395
+ σ(s)
396
+ =
397
+
398
+ (s,α)∈S×A − log νγ
399
+ σ(s, α)
400
+ µγ
401
+ σ(s) νγ
402
+ σ(s, α),
403
+ (3)
404
+ where the first equality is due to the definition of the discounted causal entropy Hγ
405
+ σ,
406
+ the second equality is obtained by expanding the expectation. The third and fourth
407
+ equalities follow by the definition of the state visitation count µγ
408
+ σ, and the state-action
409
+ visitation count νγ
410
+ σ. We prove in the appendix that the above expression is indeed
411
+ concave in the visitation counts. Next, we obtain a linear expression in νγ
412
+ σ for the
413
+ discounted return expectation Rθ
414
+ σ as:
415
+
416
+ σ =
417
+
418
+
419
+ t=0
420
+
421
+ (s,α)∈S×A
422
+ Rθ(s, α)γtP[Sσ
423
+ t = s, Aσ
424
+ t = α]
425
+ =
426
+
427
+ (s,α)∈S×A
428
+ Rθ(s, α)νγ
429
+ σ(s, α),
430
+ (4)
431
+ where the second equality is obtained by the definition of the visitation count νγ
432
+ σ. The
433
+ following nonconvex constraint in µγ
434
+ σ(s) and σz,α ensures observation-based policies:
435
+ νγ
436
+ σ(s, α) = µγ
437
+ σ(s)
438
+
439
+ z∈Z O(z|s)σz,α.
440
+ (5)
441
+ Finally, the variables for the discounted visitation counts must satisfy the so-called
442
+ Bellman flow constraint [9] to ensure that the policy is well-defined. For each state
443
+ s ∈ S,
444
+ µγ
445
+ σ(s) = µ0(s) + γ
446
+
447
+ s′∈S
448
+
449
+ α∈A
450
+ P(s|s′, α)νγ
451
+ σ(s′, α).
452
+ (6)
453
+ Saddle Point Formulation. Computing a policy σ that satisfies the return matching
454
+ constraint Rθ
455
+ σ = ¯Rθ might be infeasible due to ¯Rθ being an empirical estimate from
456
+ the finite set of demonstrations D. Additionally, the feature matching constraint might
457
+ also be infeasible due to the information asymmetry between the expert and the learner,
458
+ e.g., the expert has full observation.
459
+ We build on a saddle point computation problem to incorporate the return matching
460
+ constraints into the objective of the forward problem, similar to other IRL algorithms
461
+ in the literature. Specifically, the desired weight vector θ and policy σ of Problem 1
462
+ and Problem 2 are the solutions of minθ f(θ) := maxσ Hγ
463
+ σ +(Rθ
464
+ σ − ¯Rθ). The function
465
+ f corresponds to the inner optimization problem when the reward parameter is fixed.
466
+ That is, f(θ) computes a policy σ that maximizes the sum Hγ
467
+ σ + Rθ
468
+ σ of the causal
469
+ 8
470
+
471
+ Algorithm 1 Compute the weight vector θ and policy σ solution of the Lagrangian
472
+ relaxation of the IRL problem.
473
+ Input: Feature expectation ¯Rφ from D, initial weight θ0, step size η : N �→ R+, and
474
+ (if available) a priori side information ϕ and λ ∈ [0, 1] imposing Prσ
475
+ M(ϕ) ≥ λ .
476
+ 1: σ0 ← uniform policy
477
+ ▷ Initialize uniform policy
478
+ 2: for k = 0, 1, . . . , do
479
+ ▷ Compute θ via gradient descent
480
+ 3:
481
+ σk+1 ← SCPForward(θk, σk, ϕ, λ)
482
+ ▷ Solve the forward problem (7)–(9)
483
+ with optional ϕ and λ
484
+ 4:
485
+ θk+1 ← θk − η(k)∇θf(θk; σk+1)
486
+ ▷ Gradient step
487
+ 5: end for
488
+ 6: return σk, θk
489
+ entropy and the current estimate of the reward function. In other words, f(θ) returns
490
+ the solution to the forward problem, i.e., finding optimal policy on the POMDP when
491
+ the entropy term is removed.
492
+ Algorithm 1 updates the reward weights by using gradient descent. Initially, the
493
+ policy σ0 is a random uniform variable and the weight θ0 is a nonzero vector. At
494
+ iteration k ≥ 0, the policy σk+1 = arg maxσ Hγ
495
+ σ + (Rθk
496
+ σ − ¯Rθk) is the optimal policy
497
+ on the POMDP under the current reward estimate Rθk given by θk. That is, σk+1 is the
498
+ solution to the forward problem. Then, to update the weight θ, Algorithm 1 computes
499
+ the gradient ∇θf with respect to θ as follows:
500
+ ∇θf(θ; σ) =
501
+
502
+ s,α∈S×A
503
+ νγ
504
+ σ(s, α)∇θRθ(s, α) − 1
505
+ N
506
+
507
+ bτ ∈D
508
+
509
+ bi∈bτ
510
+ γi �
511
+ s∈S
512
+ bi(s)∇θRθ(s, αi).
513
+ We develop the algorithm SCPForward, presented in next section, to solve the
514
+ forward problem, i.e., computing σk+1 given θk, in an efficient and scalable manner
515
+ while incorporating high-level task specifications to guide the learning.
516
+ Nonconvex Formulation of the Forward Problem. Given a weight vector θk, we take
517
+ advantage of the obtained substitution by the expected visitation counts to formulate
518
+ the forward problem associated to Problem 1 as the nonconvex optimization problem:
519
+ maximize
520
+ µγ
521
+ σ,νγ
522
+ σ,σ
523
+
524
+ (s,α)∈S×A
525
+ − log νγ
526
+ σ(s, α)
527
+ µγ
528
+ σ(s) νγ
529
+ σ(s, α) +
530
+
531
+ (s,α)∈S×A
532
+ Rθk(s, α)νγ
533
+ σ(s, α)
534
+ (7)
535
+ subject to
536
+ (5) − (6),
537
+ ∀(s, α) ∈ S × A, µγ
538
+ σ(s) ≥ 0, νγ
539
+ σ(s, α) ≥ 0,
540
+ (8)
541
+ ∀(s, α) ∈ S × A, µγ
542
+ σ(s) =
543
+
544
+ α∈A νγ
545
+ σ(s, α),
546
+ (9)
547
+ where the source of nonconvexity is from (5), and we remove the constant − ¯Rθk from
548
+ the cost function of the above optimization problem.
549
+ 9
550
+
551
+ 5. Sequential Linear Programming Formulation
552
+ We develop an algorithm, SCPForward, adapting a sequential convex program-
553
+ ming (SCP) scheme to efficiently solve the nonconvex forward problem (7)–(9). In-
554
+ deed, SCPForward involves a verification step to compute sound policies and visi-
555
+ tation counts, which is not present in the existing SCP schemes. Additionally, we de-
556
+ scribe in the next section how to take advantage of high-level task specification (Prob-
557
+ lem 2) through slight modifications of the obtained optimization problem solved by
558
+ SCPForward.
559
+ 5.1. Linearizing Nonconvex Optimization Problem
560
+ SCPForward iteratively linearizes the nonconvex constraints in (5) around a pre-
561
+ vious solution. However, the linearization may result in an infeasible or unbounded
562
+ linear subproblem [25]. We first add slack variables to the linearized constraints to
563
+ ensure feasibility. The linearized problem may not accurately approximate the non-
564
+ convex problem if the solutions to this problem deviate significantly from the previous
565
+ solution. Thus, we utilize trust region constraints [25] to ensure that the linearization is
566
+ accurate to the nonconvex problem. At each iteration, we introduce a verification step
567
+ to ensure that the computed policy and visitation counts are not just approximations but
568
+ actually satisfy the nonconvex policy constraint (5), improves the realized cost function
569
+ over past iterations, and satisfy the temporal logic specifications, if available.
570
+ Linearizing Nonconvex Constraints and Adding Slack Variables. We linearize the
571
+ nonconvex constraint (5), which is quadratic in µγ
572
+ σ(s) and σz,α, around the previously
573
+ computed solution denoted by ˆσ, µγ
574
+ ˆσ, and νγ
575
+ ˆσ. However, the linearized constraints may
576
+ be infeasible. We alleviate this drawback by adding slack variables ks,α ∈ R for
577
+ (s, α) ∈ S × A, which results in the affine constraint:
578
+ νγ
579
+ σ(s, α) + ks,α = µγ
580
+ ˆσ(s)
581
+
582
+ z∈Z O(z|s)σz,α +
583
+ (10)
584
+
585
+ µγ
586
+ σ(s) − µγ
587
+ ˆσ(s)
588
+ � �
589
+ z∈Z O(z|s)ˆσz,α.
590
+ Trust Region Constraints. The linearization may be inaccurate if the solution deviates
591
+ significantly from the previous solution. We add following trust region constraints to
592
+ alleviate this drawback:
593
+ ∀(z, α) ∈ Z × A,
594
+ ˆσz,α/ρ ≤ σz,α ≤ ˆσz,αρ,
595
+ (11)
596
+ where ρ is the size of the trust region to restrict the set of allowed policies in the lin-
597
+ earized problem. We augment the cost function in (7) with the term −β �
598
+ (s,α)∈S×A ks,α
599
+ to ensure that we minimize the violation of the linearized constraints, where β is a large
600
+ positive constant.
601
+ 10
602
+
603
+ Linearized Problem. Finally, by differentiating x �→ x log x and y �→ x log(x/y),
604
+ we obtain the coefficients required to linearize the convex causal entropy cost function
605
+ in (7). Thus, we obtain the following linear program (LP):
606
+ maximize
607
+ µγ
608
+ σ,νγ
609
+ σ,σ
610
+
611
+ (s,α)∈S×A −
612
+
613
+ βks,α −
614
+ �νγ
615
+ ˆσ(s, α)
616
+ µγ
617
+ ˆσ(s)
618
+
619
+ µγ
620
+ σ(s)
621
+ +
622
+
623
+ log νγ
624
+ ˆσ(s, α)
625
+ µγ
626
+ ˆσ(s)
627
+ + 1
628
+
629
+ νγ
630
+ σ(s, α)
631
+
632
+ +
633
+
634
+ (s,α)∈S×A
635
+ Rθk(s, α)νγ
636
+ σ(s, α) (12)
637
+ subject to
638
+ (6), (8) − (11).
639
+ Verification Step. After each iteration, the linearization might be inaccurate, i.e, the
640
+ resulting policy ˜σ and potentially inaccurate visitation counts ˜νγ
641
+ ˜σ, ˜µγ
642
+ ˜σ might not be fea-
643
+ sible to the nonconvex policy constraint (5). As a consequence of the potential infea-
644
+ sibility, the currently attained (linearized) optimal cost might significantly differ from
645
+ the realized cost by the feasible visiation counts for the ˜σ. Additionally, existing SCP
646
+ schemes linearizes the nonconvex problem around the previously inaccurate solutions
647
+ for ˜νγ
648
+ ˜σ, and ˜µγ
649
+ ˜σ, further propagating the inaccuracy. The proposed verification step
650
+ solves these issues. Given the computed policy ˜σ, SCPForward computes the unique
651
+ and sound solution for the visitation count µγ
652
+ ˜σ by solving the corresponding Bellman
653
+ flow constraints:
654
+ µγ
655
+ ˜σ(s) =µ0(s) + γ
656
+
657
+ s′∈S
658
+
659
+ α∈A
660
+ P(s|s′, α)µγ
661
+ ˜σ(s′)
662
+
663
+ z∈Z
664
+ O(z|s)˜σz,α,
665
+ (13)
666
+ for all s ∈ S, and where µγ
667
+ ˜σ ≥ 0 is the only variable of the linear program. Then,
668
+ SCPForward computes νγ
669
+ ˜σ(s, α) = µγ
670
+ ˜σ(s′) �
671
+ z∈Z O(z|s)˜σz,α and the realized cost
672
+ at the current iteration is defined by
673
+ C(˜σ, θk) =
674
+
675
+ (s,α)∈S×A
676
+ − log νγ
677
+ ˜σ(s, α)
678
+ µγ
679
+ ˜σ
680
+ νγ
681
+ ˜σ(s, α) +
682
+
683
+ (s,α)∈S×A
684
+ Rθk(s, α)νγ
685
+ ˜σ(s, α),
686
+ (14)
687
+ where we assume 0 log 0 = 0. Finally, if the realized cost C(˜σ, θk) does not improve
688
+ over the previous cost C(ˆσ, θk), the verification step rejects the obtained policy ˜σ, con-
689
+ tracts the trust region, and SCPForward iterates with the previous solutions ˆσ, µγ
690
+ ˆσ,
691
+ and νγ
692
+ ˆσ . Otherwise, the linearization is sufficiently accurate, the trust region is ex-
693
+ panded, and SCPForward iterates with ˜σ, µγ
694
+ ˜σ and νγ
695
+ ˜σ. By incorporating this verifica-
696
+ tion step, we ensure that SCPForward always linearizes the nonconvex optimization
697
+ problem around a solution that satisfies the nonconvex constraint (5).
698
+ 5.2. Incorporating High-Level Task Specifications
699
+ Given high-level side information on the agent tasks as the LTL formula ϕ, we first
700
+ compute the product of the POMDP and the ω-automaton representing ϕ to find the
701
+ set T ⊆ S of states, called target or reach states, satisfying ϕ with probability 1 by
702
+ 11
703
+
704
+ using standard graph-based algorithms as a part of preprocessing step. We refer the
705
+ reader to [19] for a detailed introduction on how LTL specifications can be reduced to
706
+ reachability specifications given by T .
707
+ As a consequence, the probability of satisfying ϕ is the sum of the probability of
708
+ reaching the target states s ∈ T , which are given by the undiscounted state visitation
709
+ count µsp
710
+ σ . That is, Prσ
711
+ M(ϕ) = �
712
+ s∈T µsp
713
+ σ (s). Unless γ = 1, µsp
714
+ σ
715
+ ̸= µγ
716
+ σ. Thus,
717
+ we introduce new variables µsp
718
+ σ , νsp
719
+ σ , and the adequate constraints in the linearized
720
+ problem (12).
721
+ Incorporating Undiscounted Visitation Variables to Linearized Problem. We append
722
+ new constraints, similar to (8), (9), and (10), into the linearized problem (12), where
723
+ the variables µγ
724
+ σ, νγ
725
+ σ, ks,α, µγ
726
+ ˆσ, νγ
727
+ ˆσ are replaced by µsp
728
+ σ , νsp
729
+ σ , ksp
730
+ s,α, µsp
731
+ ˆσ , νsp
732
+ ˆσ , respectively.
733
+ Further, we add the constraint
734
+ µsp
735
+ σ (s) = µ0(s) +
736
+
737
+ s′∈S\T
738
+
739
+ α∈A
740
+ P(s|s′, α)νsp
741
+ σ (s′, α),
742
+ (15)
743
+ which is a modification of the Bellman flow constraints such that µsp
744
+ σ (s) for all s ∈ T
745
+ only counts transitions from non-target states. Finally, we penalize the introduced slack
746
+ variables for feasibility of the linearization by augmenting the cost function with the
747
+ term −β �
748
+ (s,α)∈S×A ksp
749
+ s,α.
750
+ Relaxing Specification Constraints. To incorporate the probability of satisfying the
751
+ specifications, We add the following constraint to the linearized problem:
752
+ (spec) :=
753
+
754
+ s∈T
755
+ µsp
756
+ σ (s) + Γsp ≥ λ,
757
+ (16)
758
+ where we introduce Γsp ≥ 0 as a slack variable ensuring that the linearized problem
759
+ is always feasible. Further, we augment the cost function with −βspΓsp to penalize
760
+ violating ϕ, where βsp is a positive hyperparameter.
761
+ Updating Verification Step. We modify the previously-introduced realized cost C(˜σ, θk)
762
+ to penalize when the obtained policy does not satisfy the specification ϕ. This cost also
763
+ accounts for the linearization inaccuracy of the new policy constraint due to σ, µsp
764
+ σ ,
765
+ and νsp
766
+ σ . At each iteration, SCPForward computes the accurate µsp
767
+ ˜σ of current pol-
768
+ icy ˜σ through solving a feasibility LP with constraints given by the modified Bellman
769
+ flow constraints (15). Then, it augments Csp
770
+ ˜σ = min{0, (�
771
+ s∈T µsp
772
+ ˜σ (s) − λ)βsp} to the
773
+ realized cost to take the specification constraints into account.
774
+ Convergence to Local Optimum Solution. The convergence guarantees of the pro-
775
+ posed sequential convex scheme with trust regions follow straightforwardly from the
776
+ general convergence of sequential convex programming (SCP) schemes as proved in
777
+ Theorem 3.14 and Theorem 4.7 of [25]. Specifically, weak convergence is ensured as
778
+ the SCP algorithm generates a set of convergent subsequences, all of which satisfy the
779
+ first-order conditions [25]. This is not convergence in its strict sense due to potential
780
+ oscillation between several limit points. Still, surprisingly most of the convergence
781
+ 12
782
+
783
+ Algorithm 2 SCPForward: Linear programming-based algorithm to solve the for-
784
+ ward problem (7)–(9), i.e., compute a policy σk+1 that maximizes the causal entropy,
785
+ considers the matching constraint, and satisfies the specifications, if available.
786
+ Input: Current weight estimate θk, current best policy ˆσ = σk, side information ϕ
787
+ and λ, trust region ρ > 1, penalization coefficients β, βsp ≥ 0, constant ρ0 to
788
+ expand or contract trust region, and a threshold ρlim for trust region contraction.
789
+ 1: Find µγ
790
+ ˆσ via linear constraint (13) and νγ
791
+ ˆσ = µγ
792
+ ˆσ(s′) �
793
+ z∈Z O(z|s)ˆσz,α, given ˆσ ▷
794
+ Realized visitation counts
795
+ 2: Find µsp
796
+ ˆσ via linear constraint (15) with νsp
797
+ ˆσ = µsp
798
+ ˆσ (s′) �
799
+ z∈Z O(z|s)ˆσz,α, given ˆσ
800
+ ▷ If ϕ is available
801
+ 3: Compute the realized cost C(ˆσ, θk) ← (14) + Csp
802
+ ˆσ , given ˆσ ▷ Add specifications’
803
+ violation
804
+ 4: while ρ > ρlim do
805
+ ▷ Trust region threshold
806
+ 5:
807
+ Find optimal ˜σ to the augmented LP (12) via ˆσ, µγ
808
+ ˆσ, νγ
809
+ ˆσ, µsp
810
+ ˆσ , νsp
811
+ ˆσ
812
+ ▷ We
813
+ augment the LP with constraints (8), (9), (10), (15), and (16) induced by µsp
814
+ σ , νsp
815
+ σ ,
816
+ and by adding −β �
817
+ (s,α)∈S×A ksp
818
+ s,α − βspΓsp to the cost (12).
819
+ 6:
820
+ Compute the realized µγ
821
+ ˜σ, νγ
822
+ ˜σ,µsp
823
+ ˜σ , νsp
824
+ ˜σ , and C(˜σ, θk) via ˜σ as in lines 1–3
825
+ 7:
826
+ {ˆσ ← ˜σ; ρ ← ρρ0} if C(˜σ, θk) ≥ C(ˆσ, θk) else {ρ ← ρ/ρ0}
827
+ ▷ Verification
828
+ step
829
+ 8: end while
830
+ 9: return σk+1 := ˆσ
831
+ claims of nonlinear optimization schemes fall into this category. Furthermore, under
832
+ the right regularity assumptions on the cost function, the authors of [25] proved that
833
+ SCP schemes with trust regions can converge to a local optimum solution with a super-
834
+ linear convergence rate.
835
+ 6. Numerical Experiments
836
+ We evaluate the proposed IRL algorithm on several POMDP instances from [35],
837
+ and a simulated wheeled ground robot operating in a high-fidelity, continuous, and 3-D
838
+ Unity simulation. We first compare our IRL algorithm with a straightforward variant
839
+ of GAIL [30] adapted for POMDPs. Then, we provide results on the data-efficiency
840
+ of the proposed approach when taking advantage of side information. Finally, we
841
+ demonstrate the scalability of the routine SCPForward for solving the forward prob-
842
+ lem through comparisons with state-of-the-art solvers such as SolvePOMDP [36],
843
+ SARSOP [37], PRISM-POMDP [38]. We provide the code for reproducibility of the
844
+ results in this paper at https://github.com/wuwushrek/MCE IRL POMDPS.
845
+ 6.1. Simulation on Hand-Crafted POMDP Instances
846
+ We first evaluate the proposed IRL algorithm on several POMDP instances ex-
847
+ tracted from the work [35].
848
+ 13
849
+
850
+ 1
851
+ 2
852
+ 3
853
+ 4
854
+ 5
855
+ 6
856
+ 9
857
+ 12
858
+ 7
859
+ 10
860
+ 13
861
+ 8
862
+ 11
863
+ 14
864
+ Figure 1: Some examples from the benchmark set provided in [35]. From left to right, we have the Maze,
865
+ Avoid, and Evade environments, respectively.
866
+ Benchmark Set. The POMDP instances are as follows. Evade is a turn-based game
867
+ where the agent must reach a destination without being intercepted by a faster player.
868
+ In Avoid, the agent must avoid being detected by two other moving players following
869
+ certain preset, yet unknown routes. In Intercept, the agent must intercept another player
870
+ who is trying to exit a gridworld. In Rocks, the agents must sample at least one good
871
+ rock over the several rocks without any failures. In Obstacle, an agent must find an exit
872
+ in a gridworld without colliding with any static obstacles. In these instances, the agent
873
+ only observes a fixed radius around its current position, see Figure 1. Finally, in Maze,
874
+ the agent must exit a maze as fast as possible while observing only the walls around it
875
+ and should not get stuck in any of the trap states.
876
+ Variants of Learned Policies and Experts. We refer to four types of policies. The
877
+ type of policy depends on whether it uses side information from a temporal specifi-
878
+ cation ϕ or not, and whether it uses a memory size M = 1 or M = 10. We also
879
+ consider two types of experts. The first expert has full information about the envi-
880
+ ronment and computes an optimal policy in the underlying MDP. The second expert
881
+ has partial observation and computes a locally optimal policy in the POMDP with a
882
+ memory size of M = 15. Recall that the agent always has partial information. There-
883
+ fore, the first type of expert corresponds to having information asymmetry between the
884
+ learning agent and expert. Besides, we consider as a baseline a variant of GAIL where
885
+ we learn the policy on the MDP without side information, and extend it to POMDPs
886
+ via an offline computation of the belief in the states. Specifically, we find the optimal
887
+ policy on the MDP by solving the convex optimization problem corresponding to the
888
+ forward problem on MDPs. The resulting policy is a state-based policy that needs to
889
+ be transformed in order to act on a POMDP. The transformation is done by exploiting
890
+ the expert demonstrations to construct a belief state. That is, the trajectories τ of the
891
+ expert are used in a Bayesian belief updates (1) to estimate the probability of being in
892
+ each state of the POMDP. Thus, by combining the computed belief and the state-based
893
+ policy, we obtain an observation-based policy for the POMDP. Doing so could provide
894
+ a significant advantage to the GAIL variant since the state-based policy is the optimal
895
+ policy on the MDP. However, despite the high performance in practice, the policy on
896
+ the POMDP is generally suboptimal, even if the MDP policy were optimal.
897
+ We discuss the effect of side information and memory in the corresponding policies.
898
+ While we detail only on the Maze example, where the agent must exit a maze as fast as
899
+ possible, we observe similar patterns for other examples. Detailed results for the other
900
+ examples are provided in the appendix.
901
+ 14
902
+
903
+ A low state-space Avoid instance
904
+ 0
905
+ 1
906
+ 2
907
+ 3
908
+ 4
909
+ 5
910
+ x=0,y=0
911
+ 0
912
+ X=2,y=2,d=E
913
+ X=0,y=4,d=E
914
+ 1
915
+ west
916
+ east
917
+ 2
918
+ north
919
+ south
920
+ 3
921
+ adv
922
+ placement
923
+ 5
924
+ XA low state-space Evade instance
925
+ 0
926
+ 1
927
+ 2
928
+ 3
929
+ 4
930
+ 5
931
+ x=1,y=0
932
+ 0
933
+ x=2,y=3
934
+ 1
935
+ scan
936
+ adv
937
+ 2
938
+ north
939
+ east
940
+ 3
941
+ placement
942
+ west
943
+ south
944
+ 4
945
+ 5
946
+ XNo information asymmetry
947
+ Under information asymmetry
948
+ GAIL
949
+ 0
950
+ 25
951
+ 50
952
+ 75
953
+ 100
954
+ −20
955
+ 0
956
+ 20
957
+ 40
958
+ 60
959
+ Finite-memory policy
960
+ Without side
961
+ information
962
+
963
+ σ
964
+ 0
965
+ 25
966
+ 50
967
+ 75
968
+ 100
969
+ Memoryless policy
970
+ 0
971
+ 25
972
+ 50
973
+ 75
974
+ 100
975
+ −20
976
+ 0
977
+ 20
978
+ 40
979
+ 60
980
+ Time Steps
981
+ With side
982
+ information
983
+
984
+ σ
985
+ 0
986
+ 25
987
+ 50
988
+ 75
989
+ 100
990
+ Time Steps
991
+ Figure 2: Representative results on the Maze example; each sub-figure represents the average accumulated
992
+ reward under the true reward function (Rθ
993
+ σ) over 1000 runs as a function of time. Compare the two rows:
994
+ The policies in the top row that do not utilize side information suffer a performance drop under information
995
+ asymmetry. On the other hand, in the bottom row, the performance of policies incorporating side information
996
+ into learning does not decrease under information asymmetry. Compare the two columns: The performance
997
+ of the finite-memory policies in the left column is significantly better than memoryless policies. Except for
998
+ the memoryless policies without side information, our algorithm outperforms GAIL. The expert reward on
999
+ the MDP is in average 48.22, while we obtain the value 47.83 for an expert acting on the POMDP.
1000
+ 6.1.1. Maze Example
1001
+ The POMDP M is specified by S = {s1, . . . , s14} corresponding to the cell labels
1002
+ in Figure 1. An agent in the maze only observes whether or not there is a wall (in blue)
1003
+ in a neighboring cell. That is, the set of observations is O = {o1, . . . , o6, o7}. For
1004
+ example, o1 corresponds to observing west and north walls (s1), o2 to north and south
1005
+ walls (s2, s4), and o5 to east and west walls (s6, s7, s8, s9, s10, s11). The observations
1006
+ o6 and o7 denote the target state (s13) and bad states(s12, s14). The transition model is
1007
+ stochastic with a probability of slipping p = 0.1. Further, the states s13 and s14 lead to
1008
+ the end of the simulation (trapping states).
1009
+ In the IRL experiments, we consider three feature functions. We penalize taking
1010
+ more steps with φtime(s, α) = −1 for all s, α. We provide a positive reward when
1011
+ reaching s13 with φtarget(s, α) = 1 if s = s13 and φtarget(s, α) = 0 otherwise. We
1012
+ penalize bad states s12 and s14 with φbad(s, α) = −1 if s = s12 or s = s14, and
1013
+ φbad(s, α) = 0 otherwise. Finally, we have the LTL formula ϕ = G ¬ bad as the
1014
+ task specification, where bad is an atomic proposition that is true if the current state
1015
+ s = s12 or s = s14. We constrain the learned policy to satisfy Prσ
1016
+ M(G ¬ bad) ≥ 0.9.
1017
+ Side Information Alleviates the Information Asymmetry. Figure 2 shows that if there
1018
+ is an information asymmetry between the learning agent and the expert, the policies
1019
+ that do not utilize side information suffer a significant performance drop. The policies
1020
+ 15
1021
+
1022
+ With side information
1023
+ Without side information
1024
+ GAIL
1025
+ 0
1026
+ 75
1027
+ 150
1028
+ 225
1029
+ 300
1030
+ −20
1031
+ 0
1032
+ 20
1033
+ 40
1034
+ Time Steps
1035
+ Total Reward
1036
+ Figure 3: Representative results on the Avoid example showing the reward of the policies under the true
1037
+ reward function (Rθ
1038
+ σ) versus the time steps.
1039
+ that do not incorporate side information into learning obtain a lower performance by
1040
+ 57% under information asymmetry, as shown in the top row of Figure 2. On the other
1041
+ hand, as seen in the bottom row of Figure 2, the performance of the policies that use
1042
+ side information is almost unaffected by the information asymmetry.
1043
+ Memory Leads to More Performant Policies. The results in Figure 2 demonstrate that
1044
+ incorporating memory into the policies improves the performance, i.e., the attained
1045
+ reward, in all examples, both in solving the forward problem and learning policies
1046
+ from expert demonstrations. Incorporating memory partially alleviates the effects of
1047
+ information asymmetry, as the performance of the finite-memory policy decreases by
1048
+ 18% under information asymmetry as opposed to 57% for the memoryless policy.
1049
+ We see that in Table 1, incorporating memory into policy on the Maze and Rocks
1050
+ benchmarks, allows SCPForward to compute policies that are almost optimal, evi-
1051
+ denced by obtaining almost the same reward as the solver SARSOP.
1052
+ Side Information Improves Data Efficiency. Figure 4 shows that even on a low data
1053
+ regime, learning with task specifications achieves significantly better performance than
1054
+ without the task specifications.
1055
+ 5
1056
+ 10
1057
+ 15
1058
+ 20
1059
+ 30
1060
+ 40
1061
+ Number of trajectories
1062
+ Total reward
1063
+ Without LTL
1064
+ With LTL
1065
+ Opt. Rew. POMDP
1066
+ 5
1067
+ 10
1068
+ 15
1069
+ 40
1070
+ 42
1071
+ 44
1072
+ 46
1073
+ Number of trajectories
1074
+ Figure 4: We show the data efficiency of the proposed approach through the total reward obtained by the
1075
+ learned policies as a function of the number of expert demonstrations (No information asymmetry). The
1076
+ figure on the left shows the performance of learning memoryless policies, while the figure on the right shows
1077
+ the performance of a 5-FSC.
1078
+ 16
1079
+
1080
+ SCPForward
1081
+ SARSOP
1082
+ SolvePOMDP
1083
+ Problem
1084
+ |S|
1085
+ |S × O|
1086
+ |O|
1087
+
1088
+ σ
1089
+ Time (s)
1090
+
1091
+ σ
1092
+ Time (s)
1093
+
1094
+ σ
1095
+ Time (s)
1096
+ Maze
1097
+ 17
1098
+ 162
1099
+ 11
1100
+ 39.24
1101
+ 0.1
1102
+ 47.83
1103
+ 0.24
1104
+ 47.83
1105
+ 0.33
1106
+ Maze (3-FSC)
1107
+ 49
1108
+ 777
1109
+ 31
1110
+ 44.98
1111
+ 0.6
1112
+ NA
1113
+ NA
1114
+ NA
1115
+ NA
1116
+ Maze (10-FSC)
1117
+ 161
1118
+ 2891
1119
+ 101
1120
+ 46.32
1121
+ 2.04
1122
+ NA
1123
+ NA
1124
+ NA
1125
+ NA
1126
+ Obstacle[10]
1127
+ 102
1128
+ 1126
1129
+ 5
1130
+ 19.71
1131
+ 8.79
1132
+ 19.8
1133
+ 0.02
1134
+ 5.05
1135
+ 3600
1136
+ Obstacle[10](5-FSC)
1137
+ 679
1138
+ 7545
1139
+ 31
1140
+ 19.77
1141
+ 38
1142
+ NA
1143
+ NA
1144
+ NA
1145
+ NA
1146
+ Obstacle[25]
1147
+ 627
1148
+ 7306
1149
+ 5
1150
+ 19.59
1151
+ 14.22
1152
+ 19.8
1153
+ 0.1
1154
+ 5.05
1155
+ 3600
1156
+ Rock
1157
+ 550
1158
+ 4643
1159
+ 67
1160
+ 19.68
1161
+ 12.2
1162
+ 19.83
1163
+ 0.05
1164
+
1165
+
1166
+ Rock (3-FSC)
1167
+ 1648
1168
+ 23203
1169
+ 199
1170
+ 19.8
1171
+ 15.25
1172
+ NA
1173
+ NA
1174
+
1175
+
1176
+ Rock (5-FSC)
1177
+ 2746
1178
+ 41759
1179
+ 331
1180
+ 19.82
1181
+ 97.84
1182
+ NA
1183
+ NA
1184
+
1185
+
1186
+ Intercept[5, 2, 0]
1187
+ 1321
1188
+ 5021
1189
+ 1025
1190
+ 19.83
1191
+ 10.28
1192
+ 19.83
1193
+ 13.71
1194
+
1195
+
1196
+ Intercept[5, 2, 0.1]
1197
+ 1321
1198
+ 7041
1199
+ 1025
1200
+ 19.81
1201
+ 13.18
1202
+ 19.81
1203
+ 81.19
1204
+
1205
+
1206
+ Evade[5, 2, 0]
1207
+ 2081
1208
+ 13561
1209
+ 1089
1210
+ 97.3
1211
+ 26.25
1212
+ 97.3
1213
+ 3600
1214
+
1215
+
1216
+ Evade[5, 2, 0.1]
1217
+ 2081
1218
+ 16761
1219
+ 1089
1220
+ 96.79
1221
+ 26.25
1222
+ 95.28
1223
+ 3600
1224
+
1225
+
1226
+ Evade[10, 2, 0]
1227
+ 36361
1228
+ 341121
1229
+ 18383
1230
+ 94.97
1231
+ 3600
1232
+
1233
+
1234
+
1235
+
1236
+ Avoid[4, 2, 0]
1237
+ 2241
1238
+ 5697
1239
+ 1956
1240
+ 9.86
1241
+ 34.74
1242
+ 9.86
1243
+ 9.19
1244
+
1245
+
1246
+ Avoid[4, 2, 0.1]
1247
+ 2241
1248
+ 8833
1249
+ 1956
1250
+ 9.86
1251
+ 14.63
1252
+ 9.86
1253
+ 210.47
1254
+
1255
+
1256
+ Avoid[7, 2, 0]
1257
+ 19797
1258
+ 62133
1259
+ 3164
1260
+ 9.72
1261
+ 3503
1262
+
1263
+
1264
+
1265
+
1266
+ Table 1: Results for the benchmark sets for solving the forward problem. On larger benchmarks (e.g., Evade
1267
+ and Avoid), SCPForward can compute locally optimal policies, while the other solvers fail to provide
1268
+ solutions in the given time limit. In the environments Obstacle[n], Intercept[n, r, slip], Evade[n, r, slip],
1269
+ and Avoid[n, r, slip], the parameters n, r, and slip are the size of the gridworld, the view radius of the agent,
1270
+ and the probability of slippery, respectively. We set the time-out to 3600 seconds. An empty cell (denoted by
1271
+ −) represents the solver failed to compute any policy before the time-out, while NA refers to not applicable
1272
+ due to the approach being based on belief updates.
1273
+ Side Information Improves Performance. Besides, in a more complicated environ-
1274
+ ment such as Avoid, Figure 3 shows that task specifications are crucial to hope even
1275
+ to learn the task. Specifically, Avoid[n, r, slip] is a turn-based game, where the agent
1276
+ must reach an exit point while avoiding being detected by two other moving players
1277
+ following certain predefined yet unknown routes. The agent can only observe the play-
1278
+ ers if they are within a fixed radius from the agent’s current position when the action
1279
+ scan is performed. Besides, with the players’ speed being uncertain, their position in
1280
+ the routes can not be inferred by the agent. The parameters n, r, and slip specify the
1281
+ dimension of the grid, the view radius, and the slippery probability, respectively.
1282
+ We consider four feature functions to parameterize the unknown reward. The first
1283
+ feature provides a positive reward to the agent upon reaching the exit point. The second
1284
+ feature penalizes the agent if it collides with a player. The third feature penalizes the
1285
+ agent if it is detected by a player. The fourth feature imposes a penalty cost for each
1286
+ action taken. We encode the side information as the temporal logic task specification
1287
+ avoid being detected until reaching the exit point with probability greater than 0.98.
1288
+ Figure 3 shows that the algorithm is unable to learn without side information while
1289
+ side information induces a learned policy that is optimal. Specifically, the learned
1290
+ policy without side information seems to only focus on avoiding being detected and
1291
+ collision as the corresponding learned features were close to zero.
1292
+ 17
1293
+
1294
+ Figure 5: Left: A simulated Clearpath Warthog operating in a Unity simulation. Right: A demonstration
1295
+ provided by an expert.
1296
+ 6.1.2. SCPForward Yields Better Scalability
1297
+ We highlight three observations regarding the scalability of SCPForward. First,
1298
+ the results in Table 1 show that only SARSOP is competitive with SCPForward on
1299
+ larger POMDPs. SolvePOMDP runs out of time in all but the smallest benchmarks,
1300
+ and PrismPOMDP runs out of memory in all benchmarks. Most of these approaches
1301
+ are based on updating a belief over the states, which for a large state space can become
1302
+ extremely computationally expensive.
1303
+ Second, in the benchmarks with smaller state spaces, e.g., Maze and Rock, SARSOP
1304
+ can compute policies that yield better performance in less time. This is due to the effi-
1305
+ ciency of belief-based approaches on small-size problems. On the other hand, SARSOP
1306
+ does not scale to larger POMDPs with a larger number of states and observations. For
1307
+ example, by increasing the number of transitions in Intercept benchmark from 5021 to
1308
+ 7041, the computation time for SARSOP increases by 516%. On the other hand, the
1309
+ increase of the computation time of SCPForward is only 28%.
1310
+ Third, on the largest benchmarks, including tens of thousands of states and obser-
1311
+ vations, SARSOP fails to compute any policy before time-out, while SCPForward
1312
+ found a solution. Finally, we also note that SCPForward can also compute a policy
1313
+ that maximizes the causal entropy and satisfies an LTL specification, unlike SARSOP.
1314
+ 6.2. Simulation on a Ground Robot
1315
+ We demonstrate an application of the proposed algorithm in a continuous 3-D Unity
1316
+ environment containing a ClearPath warthog operating in a semi-structured village. A
1317
+ screen shot of the robot operating in this environment and its corresponding trajectory
1318
+ can be seen in Figure 5. This environment contains a variety of obstacles including
1319
+ buildings, trees, and vehicles as well as three terrain types describing our features, φ,
1320
+ grass, gravel, and road. The simulated environment operates in a state space consisting
1321
+ of 3350 states, 33254 transitions and 944 total observations. This simulation is used to
1322
+ 18
1323
+
1324
+ 0
1325
+ 5
1326
+ 10
1327
+ 15
1328
+ 20
1329
+ 25
1330
+ 30
1331
+ 0
1332
+ 10
1333
+ 20
1334
+ 30
1335
+ grass
1336
+ gravel
1337
+ road
1338
+ unknown
1339
+ Figure 6: Gridworld representation of the environment. The figure shows the area of the unity environment
1340
+ where we applied the developed algorithm.
1341
+ gather data for training, and test an agent’s ability to follow a policy from the learned
1342
+ reward function in two experimental scenarios. In this experiment, we demonstrate
1343
+ the agent’s ability to learn a reward function from demonstrations that are sub-optimal
1344
+ with respect to a known, true reward function. We also show how the learned policies
1345
+ perform compared to the optimal policies with full and partial observations obtained
1346
+ by solving the MDP or POMDP problem with the true reward function.
1347
+ The ground vehicle contains an autonomy stack consisting of three main subsys-
1348
+ tems—mapping, perception, and planning. The mapping subsystem based on Omni-
1349
+ Mapper[? ] performs simultaneous localization and mapping (SLAM) using LiDAR
1350
+ and IMU sensors, providing a map used during planning. The perception subsystem
1351
+ provides pixel level semantic segmentation for each image in a video stream from a
1352
+ RGB camera to an ontology of terrain and object classes. Each semantic image is
1353
+ passed to a terrain projection algorithm which builds N binary occupancy feature maps
1354
+ of the known environment used for reward learning where N is the number of features.
1355
+ The planning subsystem uses the maps produced from previous subsystems and the
1356
+ trajectory from a learned policy to autonomously navigate to a waypoint.
1357
+ Expert Demonstrations and Reward Feature Encoding. We collected 10 demonstra-
1358
+ tions of an expert teleoperating a robot to a predetermined waypoint (see Figure 6).
1359
+ The expert has an implicit preference to traverse the road followed by grass, and lastly
1360
+ gravel. Consequently, we encode the unknown reward function as a linear combination
1361
+ of known features: Rθ = θ1φroad + θ2φgravel + θ3φgrass + θ4φtime + θ5φgoal, where
1362
+ φi returns a value of 0 when the feature of the corresponding state is not feature i, or
1363
+ 1 otherwise. In order to incentivize the shortest path, the feature time penalizes the
1364
+ number of actions taken in the environment before reaching the waypoint. Further-
1365
+ 19
1366
+
1367
+ (a) The trajectories resulting from executing each policy
1368
+ with and without task specifications. The learner exploit-
1369
+ ing task specifications (orange) is able to reach one of the
1370
+ target states, while avoiding the gravel along the path. In
1371
+ contrast, the learner without side information (purple) fails
1372
+ to avoid the gravel.
1373
+ 0
1374
+ 100
1375
+ 200
1376
+ 300
1377
+ −20
1378
+ 0
1379
+ 20
1380
+ Expert MDP
1381
+ Expert POMDP
1382
+ With LTL
1383
+ Without LTL
1384
+ (b) Evolution of the cumulative reward obtained by the
1385
+ learner as a function of the number of environment inter-
1386
+ actions.
1387
+ Expert MDP and Expert POMDP are the opti-
1388
+ mal policies on the MDP and POMDP, respectively for the
1389
+ ground truth reward function.
1390
+ Figure 7: Impact of incorporating task specifications into reward learning.
1391
+ more, goal provides a positive reward upon reaching the waypoint. For comparisons
1392
+ of the learned policy, we use the values θ = [0.2, −30, −2, −0.5, 50] as the ground
1393
+ truth reward weight vector. We emphasize that the demonstrations are sub-optimal
1394
+ with respect to the above ground truth reward as the vehicle often traverses gravel,
1395
+ corresponding to a high penalty reward.
1396
+ Modeling Robot Dynamics as POMDPs. From a ground truth map of the environment
1397
+ in the simulation, we obtain a high-level MDP abstraction of the learner’s behavior on
1398
+ the entire state space. Then, we impose a partial observability of the robot as follows:
1399
+ The robot does not see the entire map of the world but only see a fixed radius r = 4
1400
+ (in terms of the number of grid cells) around its current position. Furthermore, we also
1401
+ incorporate uncertainty on the sensor classification of terrain features such that with
1402
+ probability p = 0.9 the prediction is correct.
1403
+ Task Specifications. In addition to the expert demonstrations, we constrain the learned
1404
+ policy to satisfy Prσ
1405
+ M(¬ gravel U goal) ≥ 0.9, where gravel is an atomic proposition
1406
+ that is true for states having gravel as its feature, and goal is an atomic proposition that
1407
+ is true at each target state. Note that this side information does not necessarily enforce
1408
+ that the learner should reach the set of target states. Instead, if the learner reaches the
1409
+ target state, it should not drive on gravel with probability at least
1410
+ Results. Figure 7a shows how the learner with side information avoids the gravel com-
1411
+ pared to the learner without side information. Figure 7b further illustrates this result by
1412
+ empirically demonstrating that the proposed approach can efficiently take advantage
1413
+ of side information to compute policies that matches the expert’s desired behavior.
1414
+ Specifically, Figure 7b shows that the gain in the total reward of a learner without side
1415
+ 20
1416
+
1417
+ information increases by 294% with respect to a learner with side information. Ad-
1418
+ ditionally, it is important to note in Figure 6 how the initial state distribution of the
1419
+ demonstrator trajectories is different from the initial state distribution during the eval-
1420
+ uation of the learned policies (Figure 7a). Nevertheless, despite these distinctions, the
1421
+ learned policies can effectively navigate toward points present in the expert demonstra-
1422
+ tions and then maximally mimic these trajectories.
1423
+ 7. Related work.
1424
+ The closest work to ours is by [34], where they extend classical maximum-margin-
1425
+ based IRL techniques for MDPs to POMDPs. However, even on MDPs, maximum-
1426
+ margin-based approaches cannot resolve the ambiguity caused by suboptimal demon-
1427
+ strations, and they work well when there is a single reward function that is clearly better
1428
+ than alternatives [39]. In contrast, we adopt causal entropy that has been shown [39, 10]
1429
+ to alleviate these limitations on MDPs. Besides, [34] rely on efficient off-the-shelf
1430
+ solvers to the forward problem. Instead, this paper also develops an algorithm that
1431
+ outperforms off-the-shelf solvers and can scale to POMDPs that are orders of magni-
1432
+ tude larger compared to the examples in [34]. Further, [34] do not incorporate task
1433
+ specifications in their formulations.
1434
+ One of the basic challenges in IRL, is that finding a reward function and a policy
1435
+ that induces a similar behavior to the expert is an ill-defined problem. Prior work has
1436
+ addressed this challenge using maximum margin formulations [40, 41, 42], as well as
1437
+ probabilistic models to compute a likelihood of the expert demonstrations [43, 8, 10].
1438
+ We build on the latter approach and build on the maximum-causal-entropy IRL [9,
1439
+ 10, 23], which brings algorithmic benefits to IRL in POMDPs as mentioned in the
1440
+ introduction. We note that these maximum-causal-entropy IRL techniques assume that
1441
+ both the expert and the agent can fully observe the environment, and these approaches
1442
+ only apply for MDPs as opposed to POMDPs.
1443
+ IRL under partial information has been studied in prior work [2, 44, 45, 46, 47].
1444
+ Reference [44] considers the setting where the features of the reward function are par-
1445
+ tially specified as opposed to having partial information over the state of the environ-
1446
+ ment. The work in [2] considers a special case of POMDPs. It only infers a distribution
1447
+ over the future trajectories of the expert given demonstrations as opposed to computing
1448
+ a policy that induces a similar behavior to the expert. The works in [45, 46, 47] assume
1449
+ that the states of the environment are either fully observable, or fully hidden to the
1450
+ learning agent. Therefore, these approaches also consider a special case of POMDPs,
1451
+ like in [2]. We also note that none of these methods incorporate side information into
1452
+ IRL and do not provide guarantees on the performance of the policy with respect to a
1453
+ task specification.
1454
+ The idea of using side information expressed in temporal logic to guide and aug-
1455
+ ment IRL has been explored in some previous work. In [48, 22], the authors incor-
1456
+ porate side information as in temporal logic specification to learn policies that induce
1457
+ a behavior similar to the expert demonstrations and satisfies the specification. Refer-
1458
+ ence [21] iteratively infers an underlying task specification that is consistent with the
1459
+ expert demonstrations and learns a policy and a reward function that satisfies the task
1460
+ 21
1461
+
1462
+ specification. However, these methods also assume full information for both the expert
1463
+ and the agent.
1464
+ 8. Conclusion
1465
+ We develop an algorithm for inverse reinforcement learning under partial obser-
1466
+ vation. We empirically demonstrate that by incorporating task specifications into the
1467
+ learning process, we can alleviate the information asymmetry between the expert and
1468
+ the learner while increasing the data efficiency of the learning scheme. Further, we
1469
+ empirically demonstrate that our main routine SCPForward, used inside the IRL al-
1470
+ gorithm, solves the forward problem in a scalable manner and outperforms state-of-
1471
+ the-art POMDP solvers on instances with a large number of states, observations, and
1472
+ transitions.
1473
+ Work Limitations. This work assumes that the transition and observation functions of
1474
+ the POMDP are known to the algorithm. Future work will investigate removing this
1475
+ assumption and developing model-free-based approaches. We will also integrate the
1476
+ framework with more expressive neural-network-based reward functions.
1477
+ Acknowledgements.. Research was sponsored by the Army Research Laboratory and
1478
+ Office of Naval Research accomplished under cooperative agreement number(s) ARL
1479
+ W911NF-20-2-0132, ARL W911NF-19-2-0285 and ONR N00014-22-1-2254. The
1480
+ views and conclusions contained in this document are those of the authors and should
1481
+ not be interpreted as representing the official policies; either expressed or implied,
1482
+ of the Army Research Laboratory, Office of Naval Research, or the U.S. Government.
1483
+ The U.S. Government is authorized to reproduce and distribute reprints for Government
1484
+ purposed notwithstanding any copyright notation herein.
1485
+ References
1486
+ [1] P. Abbeel, A. Coates, A. Y. Ng, Autonomous Helicopter Aerobatics Through
1487
+ Apprenticeship Learning, The International Journal of Robotics Research 29 (13)
1488
+ (2010) 1608–1639.
1489
+ [2] K. M. Kitani, B. D. Ziebart, J. A. Bagnell, M. Hebert, Activity Forecasting, in:
1490
+ European Conference on Computer Vision, Springer, 2012, pp. 201–214.
1491
+ [3] D. Hadfield-Menell, S. J. Russell, P. Abbeel, A. D. Dragan, Cooperative Inverse
1492
+ Reinforcement Learning, in: NIPS, 2016.
1493
+ [4] A. D. Dragan, S. S. Srinivasa, A Policy-Blending Formalism for Shared Control,
1494
+ The International Journal of Robotics Research 32 (7) (2013) 790–805.
1495
+ [5] T. Osa, N. Sugita, M. Mitsuishi, Online Trajectory Planning in Dynamic Envi-
1496
+ ronments for Surgical Task Automation., in: Robotics: Science and Systems,
1497
+ Citeseer, 2014, pp. 1–9.
1498
+ 22
1499
+
1500
+ [6] T. Osa, N. Sugita, M. Mitsuishi, Online Trajectory Planning and Force Control
1501
+ for Automation of Surgical Tasks, IEEE Transactions on Automation Science and
1502
+ Engineering 15 (2) (2017) 675–691.
1503
+ [7] C. Finn, S. Levine, P. Abbeel, Guided Cost Learning: Deep Inverse Optimal Con-
1504
+ trol via Policy Optimization, in: International conference on machine learning,
1505
+ PMLR, 2016, pp. 49–58.
1506
+ [8] B. D. Ziebart, A. L. Maas, J. A. Bagnell, A. K. Dey, Maximum Entropy Inverse
1507
+ Reinforcement Learning, in: AAAI, Vol. 8, 2008, pp. 1433–1438.
1508
+ [9] Z. Zhou, M. Bloem, N. Bambos, Infinite Time Horizon Maximum Causal Entropy
1509
+ Inverse Reinforcement Learning, IEEE Transactions on Automatic Control 63 (9)
1510
+ (2017) 2787–2802.
1511
+ [10] B. D. Ziebart, J. A. Bagnell, A. K. Dey, Modeling Interaction via the Principle of
1512
+ Maximum Causal Entropy.
1513
+ [11] S. C. Ong, S. W. Png, D. Hsu, W. S. Lee, Pomdps for robotic tasks with mixed
1514
+ observability, in: Robotics: Science and Systems, Vol. 5, 2009, p. 4.
1515
+ [12] H. Bai, D. Hsu, W. S. Lee, Integrated Perception and Planning in the Continuous
1516
+ Space: A POMDP Approach, The International Journal of Robotics Research
1517
+ 33 (9) (2014) 1288–1302.
1518
+ [13] S. Zhang, J. Sinapov, S. Wei, P. Stone, Robot Behavioral Exploration and Mul-
1519
+ timodal Perception using POMDPs, in: AAAI Spring Symposium on Interactive
1520
+ Multisensory Object Perception for Embodied Agents, 2017.
1521
+ [14] K. Akash, K. Polson, T. Reid, N. Jain, Improving Human-Machine Collabora-
1522
+ tion Through Transparency-based Feedback–Part I: Human Trust and Workload
1523
+ Model, IFAC-PapersOnLine 51 (34) (2019) 315–321.
1524
+ [15] X. Liu, A. Datta, Modeling Context Aware Dynamic Trust Using Hidden Markov
1525
+ Model, in: Proceedings of the AAAI Conference on Artificial Intelligence,
1526
+ Vol. 26, 2012.
1527
+ [16] N. Vlassis, M. L. Littman, D. Barber, On the computational complexity of
1528
+ stochastic controller optimization in pomdps, ACM Trans. Comput. Theory 4 (4).
1529
+ [17] O. Madani, S. Hanks, A. Condon, On the Undecidability of Probabilistic Planning
1530
+ and Infinite-Horizon Partially Observable Markov Decision Problems, in: AAAI,
1531
+ AAAI Press, 1999, pp. 541–548.
1532
+ [18] M. L. Littman, U. Topcu, J. Fu, C. Isbell, M. Wen, J. MacGlashan, Environment-
1533
+ Independent Task Specifications via GLTL, arXiv preprint arXiv:1704.04341.
1534
+ [19] C. Baier, J.-P. Katoen, Principles of Model Checking, The MIT Press, 2008.
1535
+ 23
1536
+
1537
+ [20] A. Pnueli, The Temporal Logic of Programs, in: Proc. 18th Annu. Symp. Found.
1538
+ Computer Sci., Washington, D.C., USA, 1977, pp. 46–57.
1539
+ doi:10.1109/
1540
+ SFCS.1977.32.
1541
+ [21] F. Memarian, Z. Xu, B. Wu, M. Wen, U. Topcu, Active Task-Inference-Guided
1542
+ Deep Inverse Reinforcement Learning, in: 2020 59th IEEE Conference on Deci-
1543
+ sion and Control (CDC), IEEE, 2020, pp. 1932–1938.
1544
+ [22] M. Wen, I. Papusha, U. Topcu, Learning from Demonstrations with High-Level
1545
+ Side Information, in: Proceedings of the Twenty-Sixth International Joint Con-
1546
+ ference on Artificial Intelligence, 2017.
1547
+ [23] B. D. Ziebart, J. A. Bagnell, A. K. Dey, The Principle of Maximum Causal En-
1548
+ tropy for Estimating Interacting Processes, IEEE Transactions on Information
1549
+ Theory 59 (4) (2013) 1966–1980.
1550
+ [24] Y.-x. Yuan, Recent Advances in Trust Region Algorithms, Mathematical Pro-
1551
+ gramming 151 (1) (2015) 249–281.
1552
+ [25] Y. Mao, M. Szmuk, X. Xu, B. Ac¸ikmese, Successive convexification: A super-
1553
+ linearly convergent algorithm for non-convex optimal control problems, arXiv
1554
+ preprint arXiv:1804.06539.
1555
+ [26] H. Yu, D. P. Bertsekas, On Near Optimality of the Set of Finite-State Controllers
1556
+ for Average Cost POMDP, Mathematics of Operations Research 33 (1) (2008)
1557
+ 1–11.
1558
+ [27] S. Junges, N. Jansen, R. Wimmer, T. Quatmann, L. Winterer, J.-P. Katoen,
1559
+ B. Becker, Finite-State Controllers of POMDPs using Parameter Synthesis, in:
1560
+ Proceedings of UAI, 2018, pp. 519–529.
1561
+ [28] N. Meuleau, K.-E. Kim, L. P. Kaelbling, A. R. Cassandra, Solving POMDPs by
1562
+ searching the space of finite policies, in: UAI, 1999, pp. 417–426.
1563
+ [29] C. Amato, D. S. Bernstein, S. Zilberstein, Optimizing Fixed-Size Stochastic Con-
1564
+ trollers for POMDPs and Decentralized POMDPs, AAMAS 21 (3) (2010) 293–
1565
+ 320.
1566
+ [30] J. Ho, S. Ermon, Generative adversarial imitation learning, Advances in neural
1567
+ information processing systems 29 (2016) 4565–4573.
1568
+ [31] J. Massey, Causality, feedback and directed information, in: Proc. Int. Symp. Inf.
1569
+ Theory Applic.(ISITA-90), Citeseer, 1990, pp. 303–305.
1570
+ [32] G. Kramer, Directed Information for Channels with Feedback, Citeseer, 1998.
1571
+ [33] L. P. Hansen, T. J. Sargent, G. Turmuhambetova, N. Williams, Robust Control
1572
+ and Model Misspecification, Journal of Economic Theory 128 (1) (2006) 45–90.
1573
+ [34] J. Choi, K.-E. Kim, Inverse Reinforcement Learning in Partially Observable En-
1574
+ vironments, Journal of Machine Learning Research 12 (2011) 691–730.
1575
+ 24
1576
+
1577
+ [35] S. Junges, N. Jansen, S. A. Seshia, Enforcing almost-sure reachability in pomdps,
1578
+ in: International Conference on Computer Aided Verification, 2020.
1579
+ [36] E. Walraven, M. Spaan, Accelerated Vector Pruning for Optimal POMDP Solvers,
1580
+ in: Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 31, 2017.
1581
+ [37] H. Kurniawati, D. Hsu, W. S. Lee, Sarsop: Efficient point-based pomdp planning
1582
+ by approximating optimally reachable belief spaces., in: Robotics: Science and
1583
+ systems, Vol. 2008, Citeseer, 2008.
1584
+ [38] G. Norman, D. Parker, X. Zou, Verification and control of partially observable
1585
+ probabilistic systems, Real-Time Systems 53 (3) (2017) 354–402.
1586
+ [39] T. Osa, J. Pajarinen, G. Neumann, J. Bagnell, P. Abbeel, J. Peters, An Algorithmic
1587
+ Perspective on Imitation Learning, Foundations and Trends in Robotics 7 (1-2)
1588
+ (2018) 1–179.
1589
+ [40] N. D. Ratliff, J. A. Bagnell, M. A. Zinkevich, Maximum Margin Planning, in:
1590
+ Proceedings of the 23rd International Conference on Machine Learning, 2006,
1591
+ pp. 729–736.
1592
+ [41] P. Abbeel, A. Y. Ng, Apprenticeship Learning via Inverse Reinforcement Learn-
1593
+ ing, in: Proceedings of the 21st International Conference on Machine Learning,
1594
+ 2004, p. 1.
1595
+ [42] A. Y. Ng, S. J. Russell, et al., Algorithms for Inverse Reinforcement Learning., in:
1596
+ Proceedings of the 17th International Conference on Machine Learning, Vol. 1,
1597
+ 2000, p. 2.
1598
+ [43] D. Ramachandran, E. Amir, Bayesian Inverse Reinforcement Learning., in: IJ-
1599
+ CAI, Vol. 7, 2007, pp. 2586–2591.
1600
+ [44] A. Boularias, O. Kr¨omer, J. Peters, Structured Apprenticeship Learning, in:
1601
+ Joint European Conference on Machine Learning and Knowledge Discovery in
1602
+ Databases, Springer, 2012, pp. 227–242.
1603
+ [45] K. Bogert, P. Doshi, Multi-Robot Inverse Reinforcement Learning under Occlu-
1604
+ sion with Interactions, in: Proceedings of the 2014 International Conference on
1605
+ Autonomous Agents and Multi-Agent Systems, Citeseer, 2014, pp. 173–180.
1606
+ [46] K. Bogert, P. Doshi, Toward Estimating Others’ Transition Models under Occlu-
1607
+ sion for Multi-Robot IRL, in: Twenty-Fourth International Joint Conference on
1608
+ Artificial Intelligence, 2015.
1609
+ [47] K. Bogert, J. F.-S. Lin, P. Doshi, D. Kulic, Expectation-Maximization for Inverse
1610
+ Reinforcement Learning with Hidden Data, in: Proceedings of the 2016 Inter-
1611
+ national Conference on Autonomous Agents & Multiagent Systems, 2016, pp.
1612
+ 1034–1042.
1613
+ 25
1614
+
1615
+ [48] I. Papusha, M. Wen, U. Topcu, Inverse Optimal Control with Regular Language
1616
+ Specifications, in: 2018 Annual American Control Conference (ACC), IEEE,
1617
+ 2018, pp. 770–777.
1618
+ Appendices
1619
+ In this appendix, we provide supplementary derivations for the results in the paper and
1620
+ more details on the numerical experiments.
1621
+ A. Concavity of Causal Entropy and Derivations of the Bellman Constraints
1622
+ In this section, we first recall the obtained expression of the causal entropy Hγ
1623
+ σ as a
1624
+ function of the visitation counts µγ
1625
+ σ and νγ
1626
+ σ. We then prove the concavity of the causal
1627
+ entropy, which enables convex-optimization-based formulation of the task-guided in-
1628
+ verse reinforcement learning (IRL) problem. Then, we provide additional details on
1629
+ the derivation of the affine constraint implied by the Bellman flow constraint.
1630
+ Concave Causal Entropy. We first recall the definitions of the state and state-action
1631
+ visitation counts. For a policy σ, state s, and action α, the discounted state visitation
1632
+ counts are defined by µγ
1633
+ σ(s) ≜ ESt[�∞
1634
+ t=1 γt1{St=s}] and the discounted state-action
1635
+ visitation counts are defined by νγ
1636
+ σ(s, α) ≜ EAt,St[�∞
1637
+ t=1 γt1{St=s,At=α}], where 1{·}
1638
+ is the indicator function and t is the time step. From the definitions of the state and
1639
+ state-action visitation counts µγ
1640
+ σ and νγ
1641
+ σ, it is straightforward to deduce that νγ
1642
+ σ(s, α) =
1643
+ σs,αµγ
1644
+ σ(s), where σs,α = P[At = a|St = s]. We use the visitation counts to prove in
1645
+ Section 4 that
1646
+
1647
+ σ =
1648
+
1649
+ (s,α)∈S×A
1650
+ −(log πs,α)πs,αµγ
1651
+ σ(s) =
1652
+
1653
+ (s,α)∈S×A
1654
+ − log νγ
1655
+ σ(s, α)
1656
+ µγ
1657
+ σ(s) νγ
1658
+ σ(s, α),
1659
+ where the last inequality is obtained by using that πs,α = νγ
1660
+ σ(s, α)/µγ
1661
+ σ(s). We claim
1662
+ that Hγ
1663
+ σ is a concave fucntion of the visitation counts. Thus, we want to show that
1664
+ the function f(νγ
1665
+ σ, µγ
1666
+ σ) = �
1667
+ (s,α)∈S×A − log νγ
1668
+ σ(s,α)
1669
+ µγ
1670
+ σ(s) νγ
1671
+ σ(s, α) is concave. To this end,
1672
+ consider any λ ∈ (0, 1) and the two sets of variables νγ
1673
+ σ, µγ
1674
+ σ and ¯νγ
1675
+ σ, ¯µγ
1676
+ σ. Then, we have
1677
+ 26
1678
+
1679
+ the following result:
1680
+ f(λνγ
1681
+ σ + (1 − λ)¯νγ
1682
+ σ, λ¯µγ
1683
+ σ + (1 − λ)¯µγ
1684
+ σ)
1685
+ =
1686
+
1687
+ (s,α)∈S×A
1688
+ − log λνγ
1689
+ σ(s, α) + (1 − λ)¯νγ
1690
+ σ(s, α)
1691
+ λµγ
1692
+ σ(s) + (1 − λ)¯µγ
1693
+ σ(s, α) (λνγ
1694
+ σ(s, α) + (1 − λ)¯νγ
1695
+ σ(s, α))
1696
+
1697
+
1698
+ (s,α)∈S×A
1699
+ −λνγ
1700
+ σ(s, α) log λνγ
1701
+ σ(s, α)
1702
+ λµγ
1703
+ σ(s, α) − (1 − λ)¯νγ
1704
+ σ(s, α) log (1 − λ)¯νγ
1705
+ σ(s, α)
1706
+ (1 − λ)¯µγ
1707
+ σ(s, α)
1708
+ =
1709
+
1710
+ (s,α)∈S×A
1711
+ −λνγ
1712
+ σ(s, α) log νγ
1713
+ σ(s, α)
1714
+ µγ
1715
+ σ(s, α) − (1 − λ)¯νγ
1716
+ σ(s, α) log ¯νγ
1717
+ σ(s, α)
1718
+ ¯µγ
1719
+ σ(s, α)
1720
+ = λf(νγ
1721
+ σ, µγ
1722
+ σ) + (1 − λ)f(¯νγ
1723
+ σ, ¯µγ
1724
+ σ),
1725
+ where the first inequality is obtained by applying to the well-known log-sum inequality,
1726
+ i.e.,
1727
+ x1 log x1
1728
+ y1
1729
+ + x2 log x2
1730
+ y2
1731
+ ≥ (x1 + x2) log x1 + x2
1732
+ y1 + y2
1733
+ ,
1734
+ for nonnegative numbers x1, x2, y1, y2. Specifically, we apply the substitution x1 =
1735
+ λνγ
1736
+ σ, y1 = λµγ
1737
+ σ, x2 = (1 − λ)¯νγ
1738
+ σ, and y2 = (1 − λ)¯µγ
1739
+ σ. Note that the inequality
1740
+ f(λνγ
1741
+ σ + (1 − λ)¯νγ
1742
+ σ, λ¯µγ
1743
+ σ + (1 − λ)¯µγ
1744
+ σ) ≥ λf(νγ
1745
+ σ, µγ
1746
+ σ) + (1 − λ)f(¯νγ
1747
+ σ, ¯µγ
1748
+ σ)
1749
+ implies that f(νγ
1750
+ σ, µγ
1751
+ σ) is concave in νγ
1752
+ σ, and µγ
1753
+ σ.
1754
+ Bellman Flow Constraint. For the visitation count variables to correspond to a valid
1755
+ policy generating actions in the POMDP M , νγ
1756
+ σ and µγ
1757
+ σ must satisfy the bellman flow
1758
+ constraint given by
1759
+ µγ
1760
+ σ(s) = ESσ
1761
+ t
1762
+ � ∞
1763
+
1764
+ t=0
1765
+ γt1{Sσ
1766
+ t =s}
1767
+
1768
+ = µ0(s) + γESσ
1769
+ t
1770
+ � ∞
1771
+
1772
+ t=0
1773
+ γt1{Sσ
1774
+ t+1=s}
1775
+
1776
+ = µ0(s) + γ
1777
+
1778
+
1779
+ t=0
1780
+
1781
+ s′∈S
1782
+
1783
+ α∈A
1784
+ γtP(s|s′, α)P[Sσ
1785
+ t = s′, Aσ
1786
+ t = α]
1787
+ = µ0(s) + γ
1788
+
1789
+ s′∈S
1790
+
1791
+ α∈A
1792
+ P(s|s′, α)νγ
1793
+ σ(s′, α).
1794
+ B. Experimental Tasks
1795
+ In this section, we first provide a detailed description of the POMDP models used
1796
+ in the benchmark. The simulations on the benchmark examples empirically demon-
1797
+ strate that side information alleviates the information asymmetry, and more memory
1798
+ leads to more performant policies. Then, we provide additional numerical simulations
1799
+ supporting the claim that SCPForward is sound and yields better scalability than
1800
+ off-the-shelf solvers for the forward problem, i.e., computing an optimal policy on a
1801
+ POMDP for a given reward function.
1802
+ 27
1803
+
1804
+ B.1. Computation Resources and External Assets
1805
+ All the experiments of this paper were performed on a computer with an Intel Core
1806
+ i9-9900 CPU 3.1GHz ×16 processors and 31.2 Gb of RAM. All the implementations
1807
+ are written and tested in Python 3.8, and we attach the code with the supplementary
1808
+ material.
1809
+ Required Tools. . Our implementation requires Stormpy of Storm [? ] and Gurobipy
1810
+ of Gurobi 9.1 [? ]. On one hand, we use Storm, a tool for model checking, to parse
1811
+ POMDP file specifications, to compute the product POMDP with the finite state con-
1812
+ troller in order to reduce the synthesis problem to the synthesis of memoryless policies,
1813
+ and to compute the set T of target states satisfying a specification ϕ via graph prepro-
1814
+ cessing. On the other hand, we use Gurobi to solve both the linearized problem in (7)
1815
+ and the feasible solution of the Bellman flow constraint needed for the verification step.
1816
+ Off-The-Shelf Solvers for Forward Problem.
1817
+ . In order to show the scalability of
1818
+ the developed algorithm SCPForward, we compare it to state-of-the-art POMDP
1819
+ solvers SolvePOMDP [36], SARSOP [37], and PRISM-POMDP [38].
1820
+ The solver
1821
+ SolvePOMDP implements both exact and approximate value iterations via incremen-
1822
+ tal pruning [? ] combined with state-of-the-art vector pruning methods [36]. Finally,
1823
+ PrismPOMDP discretizes the belief state and adopts a finite memory strategy to find
1824
+ an approximate solution of the forward problem. For all the solvers above, we use the
1825
+ default settings except from the timeout enforced to be 3600 seconds. These solvers
1826
+ are not provided with our implementation. However, we provide the POMDP models
1827
+ that each of the solvers can straightforwardly use. Further details are provided in the
1828
+ readme files of our implementation.
1829
+ B.2. Benchmark Set
1830
+ We evaluate the proposed learning algorithm on several POMDP instances adapted
1831
+ from [35]. We attached the modified instances in our code with the automatically
1832
+ generated models for each off-the-shelf solver that the reader can straightforwardly
1833
+ use to reproduce Table 1. The reader can refer to Table 1 for the number of states,
1834
+ observations, and transitions of each environment of the benchmark set. In all the
1835
+ examples, we gather 10 trajectories from an expert that can fully observe its current
1836
+ state in the environment and an expert having partial observation of the environment.
1837
+ Our algorithm learns reward functions from these trajectories under different memory
1838
+ policies and high-level side information.
1839
+ 28
1840
+
1841
+ Rocks Instance. In the environment Rocks, an agent navigates in a gridworld to sam-
1842
+ ple at least one valuable rock (if a valuable rock is in the grid) over the two possibly
1843
+ dangerous rocks, without any failures. When at least one valuable rock has been col-
1844
+ lected, or the agent realizes that all the rocks are dangerous, it needs to get to an exit
1845
+ point to terminate the mission. The partial observability is due to the agent can only
1846
+ observe if its current location is an exit point or a dangerous rock. Furthermore, the
1847
+ agent has noisy sensors enabling sampling neighbor cells.
1848
+ We consider three feature functions. The first feature provides a positive reward
1849
+ when reaching the exit point with at least one valuable rock or no rocks when all of
1850
+ them are dangerous. The second feature provides a negative reward when the agent
1851
+ is at the location of a dangerous rock. Finally, the third feature penalizes each action
1852
+ taken with a negative reward to promote reaching the exit point as soon as possible.
1853
+ No information asymmetry
1854
+ Under information asymmetry
1855
+ GAIL
1856
+ 0
1857
+ 75
1858
+ 150
1859
+ 225
1860
+ 300
1861
+ 0
1862
+ 50
1863
+ 100
1864
+ Finite-memory policy
1865
+ Without side
1866
+ information
1867
+
1868
+ σ
1869
+ 0
1870
+ 75
1871
+ 150
1872
+ 225
1873
+ 300
1874
+ 0
1875
+ 50
1876
+ 100
1877
+ Memoryless policy
1878
+
1879
+ σ
1880
+ 0
1881
+ 75
1882
+ 150
1883
+ 225
1884
+ 300
1885
+ 0
1886
+ 50
1887
+ 100
1888
+ Time Steps
1889
+ With side
1890
+ information
1891
+
1892
+ σ
1893
+ 0
1894
+ 75
1895
+ 150
1896
+ 225
1897
+ 300
1898
+ 0
1899
+ 50
1900
+ 100
1901
+ Time Steps
1902
+
1903
+ σ
1904
+ Figure 8: Representative results on the Rock example showing the reward of the policies under the true
1905
+ reward function (Rφ
1906
+ σ) versus the time steps.
1907
+ We compare scenarios with no side information and the a priori knowledge on the
1908
+ task such as the agent eventually reaches an exit point with a probability greater than
1909
+ 0.995. Figure 8 supports our claim that side information indeed alleviates the informa-
1910
+ tion asymmetry between the expert and the agent. Additionally, we also observe that
1911
+ incorporating memory leads to more performant policies in terms of the mean accumu-
1912
+ lated reward.
1913
+ 29
1914
+
1915
+ Obstacle Instance. . In the environment Obstacle[n], an agent must find an exit in a
1916
+ gridworld without colliding with any of the five static obstacles in the grid. The agent
1917
+ only observes whether the current position is an obstacle or an exit state. The parameter
1918
+ n specifies the dimension of the grid.
1919
+ Similar to the Rocks example, the agent receives a positive reward if it successfully
1920
+ exits the gridworld and a negative reward for every taken action or colliding with an
1921
+ obstacle.
1922
+ As for the side information, we specify in temporal logic that while learning the
1923
+ reward, the agent should not collide any obstacles until it reaches an exit point with a
1924
+ probability greater than 0.9.
1925
+ No information asymmetry
1926
+ Under information asymmetry
1927
+ 0
1928
+ 25
1929
+ 50
1930
+ 75
1931
+ 100
1932
+ −200
1933
+ 0
1934
+ 200
1935
+ 400
1936
+ Finite-memory policy
1937
+ Without side
1938
+ information
1939
+
1940
+ σ
1941
+ 0
1942
+ 25
1943
+ 50
1944
+ 75
1945
+ 100
1946
+ −200
1947
+ 0
1948
+ 200
1949
+ 400
1950
+ Memoryless policy
1951
+
1952
+ σ
1953
+ 0
1954
+ 25
1955
+ 50
1956
+ 75
1957
+ 100
1958
+ −200
1959
+ 0
1960
+ 200
1961
+ 400
1962
+ Time Steps
1963
+ With side
1964
+ information
1965
+
1966
+ σ
1967
+ 0
1968
+ 25
1969
+ 50
1970
+ 75
1971
+ 100
1972
+ −200
1973
+ 0
1974
+ 200
1975
+ 400
1976
+ Time Steps
1977
+
1978
+ σ
1979
+ Figure 9: Representative results on the Obstacle example showing the reward of the policies under the true
1980
+ reward function (Rφ
1981
+ σ) versus the time steps.
1982
+ Similar to the Maze and Rock examples, Figure 9 supports our claim that side in-
1983
+ formation alleviates the information asymmetry and memory leads to more performant
1984
+ policies.
1985
+ 30
1986
+
1987
+ Evade Instance. Evade[n, r, slip] is a turn-based game where the agent must reach a
1988
+ destination without being intercepted by a faster player. The player cannot access the
1989
+ top row of the grid. Further, the agent can only observe the player if it is within a fixed
1990
+ radius from its current location and upon calling the action scan. The parameters n, r,
1991
+ and slip specify the dimension of the grid, the view radius, and the slippery probability,
1992
+ respectively.
1993
+ The feature functions are defined such that the agent receives a positive reward if
1994
+ at the destination, a high negative reward if it is intercepted by the player, and a small
1995
+ negative reward for each action taken, including the scan action.
1996
+ With side information
1997
+ Without side information
1998
+ GAIL
1999
+ 0
2000
+ 25
2001
+ 50
2002
+ 75
2003
+ 100
2004
+ −10
2005
+ 0
2006
+ 10
2007
+ 20
2008
+ Time Steps
2009
+ Mean accumulated reward
2010
+ Figure 10: Representative results on the Evade example showing the reward of the policies under the true
2011
+ reward function (Rφ
2012
+ σ) versus the time steps.
2013
+ As for the side information, we specify in temporal logic that while learning the
2014
+ reward, the agent must reach an exit point with probability greater than 0.98.
2015
+ Figure 10 shows that learning with side information provides higher reward than
2016
+ without side information. Besides, there is less randomness in the policy with side
2017
+ information compared to the policy without side information. Specifically, the standard
2018
+ deviation of the policy with side information is significantly smaller than the policy
2019
+ without side information.
2020
+ We did not discuss the impact of different memory size policies in this example
2021
+ since the performance of the memoryless policy is already near-optimal, as the policy
2022
+ obtains the same reward as SARSOP (see Table 1 for a reference. Specifically, we
2023
+ observe that the optimal policy on the underlying MDP yields comparable policies to
2024
+ the optimal memoryless policy on the POMDP. As a consequence, we observe that the
2025
+ information asymmetry between the expert and the agent does not hold here either, and
2026
+ the learned policies obtain a similar performance.
2027
+ 31
2028
+
2029
+ Intercept Instance. Intercept[n, r, slip] is a variant of Evade where the agent must
2030
+ intercept another player who is trying to exit the gridworld. The agent can move in 8
2031
+ directions and can only observe the player if it is within a fixed radius from the agent’s
2032
+ current position when the action scan is performed. Besides, the agent has a camera
2033
+ that enables it to observe all cells from west to east from the center of the gridworld. In
2034
+ contrast, the player can only move in 4 directions. The parameters n, r, and slip specify
2035
+ the dimension of the grid, the view radius, and the slippery probability, respectively.
2036
+ We consider three feature functions to parameterize the unknown reward. The first
2037
+ feature provides a positive reward to the agent upon intercepting the player. The second
2038
+ feature penalizes the agent if the player exits the gridworld. The third feature imposes
2039
+ a penalty cost for each action taken.
2040
+ With side information
2041
+ Without side information
2042
+ GAIL
2043
+ 0
2044
+ 25
2045
+ 50
2046
+ 75
2047
+ 100
2048
+ −10
2049
+ 0
2050
+ 10
2051
+ 20
2052
+ Time Steps
2053
+ Mean accumulated reward
2054
+ Figure 11: Representative results on the Intercept example showing the reward of the policies under the
2055
+ true reward function (Rφ
2056
+ σ) versus the time steps.
2057
+ We encode the high-level side information as the temporal logic task specification
2058
+ Eventually intercept the player with probability greater than 0.98, i.e., the agent should
2059
+ eventually reach an observation where its location coincides with the player’s location.
2060
+ Figure 11 demonstrates that side information does not improve the performance
2061
+ of the policy. This result is because memoryless policies are optimal in this example,
2062
+ and a combination of the given reward features can perfectly encode the temporal logic
2063
+ specifications, similar to the Evade example.
2064
+ B.3. Effects of Side Information
2065
+ In this section, we provide additional experiments on how the side information
2066
+ speeds up the learning process in terms of computation time and convergence to the
2067
+ optimal policy and reward parameters. Then, we quantify the effects of the side infor-
2068
+ mation by solving the POMDPs using only the task specifications and no demonstra-
2069
+ tions. All these experiments are performed on the Maze example, which is a relatively
2070
+ low-size POMDP example.
2071
+ 32
2072
+
2073
+ Convergence of the Learning With and Without Side Information. In the Maze ex-
2074
+ periments, we empirically observe that side information enables the learning algorithm
2075
+ to converge with eight times less number of iterations compared to learning without
2076
+ side information. The number of iterations here denotes both the number of lineariza-
2077
+ tions in the sequential convex scheme and the number of gradient steps during reward
2078
+ updates. However, the gain in computation time is not as prominent as the gain in
2079
+ the number of iterations. In fact, learning with side information is only approximately
2080
+ three times faster than learning without side information due to how side information
2081
+ almost doubles the number of variables in the convex optimization problem.
2082
+ Effects of Side Information Without Any Demonstrations.
2083
+ • Experiment 1.
2084
+ We consider the exact setting of the Maze example with no
2085
+ demonstrations and the LTL specification Prσ
2086
+ M(G ¬ bad) ≥ 0.9. Essentially,
2087
+ we seek policies that avoid the trapping states with high probability. Without any
2088
+ additional reward, the optimal policy is exactly what we expect: High probabil-
2089
+ ity for action enforcing no movements and low probability for the others. With
2090
+ respect to the true reward, this is clearly suboptimal as the reward will keep de-
2091
+ creasing due to no minimization of the amount of time spent in the environment.
2092
+ Now, we optimize the problem for a policy that satisfies the LTL specifications
2093
+ while penalizing spending time in the environment according to the ground truth
2094
+ reward for taking more steps in the environment. The obtained policy has the op-
2095
+ timal reward of 47.83, which corresponds to optimizing the ground truth reward
2096
+ in the POMDP. Indeed, the fastest way to clear the Maze is to exit through a goal
2097
+ state while not getting trapped.
2098
+ • Experiment 2.
2099
+ We consider the exact setting of the Maze example with no
2100
+ demonstrations and the LTL specification Prσ
2101
+ M(E target) ≥ 0.95. Essentially,
2102
+ we seek policies that eventually reach the target state (exit of the Maze) with
2103
+ high probability. Without any additional reward, the optimal policy is subopti-
2104
+ mal with respect to the optimal policy on the POMDP, given the ground truth
2105
+ reward. Indeed, the policy can reach the target with an optimal reward of 28.63
2106
+ due to the amount of time spent to reach the goal. By adding the reward on time
2107
+ spent in the environment to the LTL specifications, we can obtain the optimal
2108
+ reward on the POMDP again.
2109
+ B.4. Summary of the Results
2110
+ Side Information Alleviates the Information Asymmetry . As mentioned in the sub-
2111
+ mitted manuscript, side information can indeed alleviate the information asymmetry.
2112
+ Specifically, we observe that if there is an information asymmetry in the forward prob-
2113
+ lem, i.e., the obtained reward from an optimal policy on the underlying POMDP is
2114
+ lower than from an optimal policy on the underlying fully observable MDP, incor-
2115
+ porating side information in temporal logic specifications alleviates the information
2116
+ asymmetry between the expert and the agent. For example, we can see the effects of
2117
+ such information asymmetry in the Maze, Rocks, Obstacle, and Avoid examples. In
2118
+ 33
2119
+
2120
+ these examples, having partial observability reduces the obtained reward in the for-
2121
+ ward problem. The policies that do not incorporate side information into the learning
2122
+ procedure also obtain a lower reward under information asymmetry.
2123
+ Memory Leads to More Performance Policies. Similarly to the side information, we
2124
+ also observe that if incorporating memory improves the performance of the learned
2125
+ policies, if it also improves the obtained reward in the forward problem, as seen in the
2126
+ Maze, Rocks, and Obstacle instances. In Table 1, we can also see that incorporating
2127
+ memory helps to compute a better optimal policy in these examples, unlike computing
2128
+ a memoryless policy.
2129
+ 34
2130
+
19AzT4oBgHgl3EQfRvso/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
3NAzT4oBgHgl3EQfR_tE/content/tmp_files/2301.01224v1.pdf.txt ADDED
@@ -0,0 +1,1827 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ An Empirical Investigation into the Use of Image
2
+ Captioning for Automated Software Documentation
3
+ Kevin Moran†, Ali Yachnes∗, George Purnell∗, Junayed Mahmud†,
4
+ Michele Tufano‡, Carlos Bernal Cardenas‡, Denys Poshyvanyk∗, Zach H’Doubler∗
5
+ †George Mason University, VA, USA, ∗William & Mary, VA, USA, ‡Microsoft, WA, USA
6
+ kpmoran@gmu.edu, ayachnes@email.wm.edu, gwpurnell@email.wm.edu, jmahmud@gmu.edu
7
+ michele.tufano@microsoft.com, carlosbe@microsoft.com, denys@cs.wm.edu, pzhdoubler@email.wm.edu
8
+ Abstract—Existing automated techniques for software docu-
9
+ mentation typically attempt to reason between two main sources
10
+ of information: code and natural language. However, this reason-
11
+ ing process is often complicated by the lexical gap between more
12
+ abstract natural language and more structured programming
13
+ languages. One potential bridge for this gap is the Graphical User
14
+ Interface (GUI), as GUIs inherently encode salient information
15
+ about underlying program functionality into rich, pixel-based
16
+ data representations. This paper offers one of the first com-
17
+ prehensive empirical investigations into the connection between
18
+ GUIs and functional, natural language descriptions of software.
19
+ First, we collect, analyze, and open source a large dataset of
20
+ functional GUI descriptions consisting of 45,998 descriptions
21
+ for 10,204 screenshots from popular Android applications. The
22
+ descriptions were obtained from human labelers and underwent
23
+ several quality control mechanisms. To gain insight into the
24
+ representational potential of GUIs, we investigate the ability of
25
+ four Neural Image Captioning models to predict natural language
26
+ descriptions of varying granularity when provided a screenshot
27
+ as input. We evaluate these models quantitatively, using common
28
+ machine translation metrics, and qualitatively through a large-
29
+ scale user study. Finally, we offer learned lessons and a discussion
30
+ of the potential shown by multimodal models to enhance future
31
+ techniques for automated software documentation.
32
+ Index Terms—Software Documentation, Image Captioning,
33
+ Deep Learning
34
+ I. INTRODUCTION & MOTIVATION
35
+ Proper documentation is generally considered to be an inte-
36
+ gral component of building and distributing modern software
37
+ systems. In fact, past studies have illustrated the general ben-
38
+ efits of documentation during the development lifecycle [1],
39
+ [2], [3], [4] and the importance of technical documentation
40
+ to software maintenance and evolution [5]. However, despite
41
+ the value of well-documented systems, modern development
42
+ processes and constraints often lead to the disregard or aban-
43
+ donment of a range of documentation tasks [6], [5], [2], [7],
44
+ [8], [1]. These difficulties have given rise to a wealth of
45
+ research on automated techniques that aim to ease the burden
46
+ on stakeholders by generating various types of documentation
47
+ for a given task. For example, existing approaches have been
48
+ developed to automatically generate natural language sum-
49
+ maries and documentation for code [9], [10], [11], [12], [13],
50
+ [14], [15], APIs [16], [17], unit tests [18], bug reports [19],
51
+ [20], release notes [21], [22], and commit messages [23], [24],
52
+ among other artifacts [25], [26].
53
+ Generally, existing techniques for automated software doc-
54
+ umentation have been concerned with modeling relationships
55
+ that exist between two primary information modalities: code
56
+ and natural language (NL). Unfortunately, reasoning between
57
+ these two information sources is difficult due to the lexical
58
+ gap resulting from the often disparate conceptual associations
59
+ that connect source code lexicon and the more abstract words
60
+ and phrases used in NL descriptions
61
+ [27], [28]. Recently,
62
+ this lexical gap was acknowledged as an information inference
63
+ problem in a report made by Robillard et al. [29], wherein
64
+ key research challenges exist in (i) inferring undocumented
65
+ program properties, and (ii) discovering latent abstractions
66
+ and rationales. These challenges suggest that overcoming the
67
+ semantic disconnect between code and NL may require new
68
+ knowledge sources that encode distinct program properties
69
+ typically absent from traditional software or NL lexicon.
70
+ One source of information which has been left largely
71
+ unexplored for the purposes of automated documentation is
72
+ visual software data encoded into Graphical User Interfaces
73
+ (GUIs). GUI-based applications predominate modern user-
74
+ facing software, as can be readily seen in the popularity of
75
+ desktop and mobile apps [30]. Furthermore, high quality ap-
76
+ plications with well-designed GUIs allow technically-inclined
77
+ users to instinctively understand underlying program features.
78
+ Thus, intuitively, certain functional properties of applications
79
+ are encoded into the visual, pixel-based representation of the
80
+ GUI such that cognitive human processes can determine the
81
+ computing tasks provided by the interface. This suggests that
82
+ there are latent patterns that exist within visual GUI data
83
+ which indicate the presence of natural use cases capturing core
84
+ functionality [31].
85
+ Given the inherent representational power of GUIs in con-
86
+ veying program related information, we set forth the following
87
+ hypothesis that serves as the basis for work in this paper:
88
+ The representational power of graphical user interfaces to
89
+ convey program-related information can be meaningfully
90
+ leveraged to support automated techniques for software doc-
91
+ umentation.
92
+ While most existing work on automated documentation con-
93
+ cerns itself with the dichotomy between code and NL, we posit
94
+ that the latent information encoded within GUIs can aid in
95
+ bridging the existing semantic documentation gap by providing
96
+ arXiv:2301.01224v1 [cs.SE] 3 Jan 2023
97
+
98
+ an additional source of knowledge that inherently reflects pro-
99
+ gram functionality. In fact, GUI-based representations of soft-
100
+ ware have the potential to address the two challenges set forth
101
+ by Robillard et al. [29]. More specifically, GUIs can aid in
102
+ inferring undocumented program properties that are inherently
103
+ represented within the design of GUI controls or widgets (e.g.,
104
+ capturing a feature which is otherwise poorly represented by
105
+ low-quality code identifiers/comments). Further, GUIs could
106
+ be used as source to mine abstractions or rationales that
107
+ would otherwise remain obscure (e.g., providing a use case-
108
+ based explanation of a block of code connected to a GUI
109
+ screen). In overcoming these challenges, we see GUI-centric
110
+ documentation having an impact on the following types of
111
+ software documentation:
112
+ Technical Documentation: Developers utilize technical docu-
113
+ mentation, such as code comments or READMEs, in order
114
+ to learn about the functionality and interfaces of software
115
+ to support engineering tasks. Automatically generating such
116
+ documentation accurately is a challenging inference problem.
117
+ However, it has been shown that GUI-related code can com-
118
+ prise as much as half of the code in user facing programs [32].
119
+ This means that graphical software data is connected in some
120
+ way to large portions of GUI-based software projects i.e.,
121
+ through GUI-event handlers, or code stipulating GUI layouts
122
+ such as html. Therefore, if automated techniques are able to
123
+ effectively infer salient functionality from the GUIs, they could
124
+ be combined with existing techniques and leveraged to provide
125
+ automation to developers, such as comment generation or code
126
+ summarization with greater feature-based context awareness.
127
+ As we illustrate in this paper, GUI code/metadata appears to
128
+ encode orthogonal information compared to visual GUI data
129
+ (i.e., screenshots), which suggests that we may be able to infer
130
+ documentation information from visual GUI data that likely
131
+ can’t be inferred from GUI code alone.
132
+ User Documentation: Developers typically provide users with
133
+ documentation such as tutorials or walkthroughs to help
134
+ clearly illustrate software features. While some experienced
135
+ users can infer functionality directly from a GUI, end-users
136
+ exhibit a range of technological expertise, and many rely upon
137
+ various forms of end-user documentation [33]. Thus, building
138
+ techniques capable of automatically generating such documen-
139
+ tation would free up development effort for other critical tasks,
140
+ such as bug fixing. Beyond typical user facing software aids,
141
+ GUI-centric program documentation could also enable entirely
142
+ new classes of automated accessibility features, which are
143
+ sorely needed for mobile apps [34]. For example, rather than a
144
+ typical text-to-speech engine, one could envision a screen-to-
145
+ functionality engine that could aid a motor-impaired user with
146
+ navigating the software, without extra development effort.
147
+ To investigate the potential of automated GUI-centric soft-
148
+ ware documentation, we offer one of the first comprehensive
149
+ empirical investigations into this new research direction’s most
150
+ fundamental task: generating a natural language descrip-
151
+ tion given a screenshot (or screen-related information) of
152
+ a software GUI. Given that this task underlies the various
153
+ potential applications discussed above, we view this as a
154
+ logical first step towards investigating the feasibility of fu-
155
+ ture techniques. To accomplish this, we collect and analyze
156
+ a dataset for Comprehending visuaL semAntics to pRedict
157
+ applicatIon functionalTY (the CLARITY dataset) consisting
158
+ of 45,998 functional descriptions of 10,204 screenshots of
159
+ popular Android apps available on Google Play. We provide
160
+ a descriptive analysis of this dataset that investigates the
161
+ “naturalness” and semantic topics of the collected descriptions
162
+ by measuring cross-entropy compared to other corpora and
163
+ performing a topic modeling analysis. To learn functional
164
+ descriptions of the screens from this dataset, we customize,
165
+ train, and test four Deep Learning (DL) models for neural
166
+ image captioning—three that learn from image data and one
167
+ that learns from textual GUI metadata—to predict functional
168
+ descriptions of software at different granularities. We evaluate
169
+ the efficacy of these models both quantitatively, by measuring
170
+ the widely used BLEU metric, and qualitatively through a
171
+ large-scale user study. In summary, this paper’s contributions
172
+ are as follows:
173
+ • We collect the CLARITY dataset of GUIs annotated
174
+ with 45,998 functional, NL descriptions from 10,204
175
+ screenshots of popular Android apps. The NL captions
176
+ were obtained from human labelers, underwent several
177
+ quality control mechanisms, and contain both high- and
178
+ low-level descriptions of screen functionality. While other
179
+ GUI datasets exist [35], [36], the CLARITY dataset differs
180
+ by providing an extensively labeled set of screens, akin
181
+ to Flickr8K [37] or MSCOCO [38];
182
+ • We illustrate the underlying, natural patterns that exist in
183
+ the CLARITY dataset through topic modeling.
184
+ • We provide an extensive quantitative and qualitative eval-
185
+ uation of four tailored DL models for image captioning
186
+ using standard metrics and a large scale user study;
187
+ • We offer an online appendix with examples of model-
188
+ generated descriptions and experimental data [39]. Our
189
+ dataset, trained models, code, and evaluation scripts are
190
+ open source and accessible via the appendix.
191
+ II. BACKGROUND
192
+ A. The Connection between Images and NL
193
+ The task of image captioning is much more difficult than
194
+ that of classification or labeling, as an effective model must
195
+ be able to both learn salient features from images automati-
196
+ cally and semantically equate these features with the proper
197
+ NL words and grammar that describe them. This task of
198
+ semantically aligning two completely different modalities of
199
+ information has led to the development of multimodal DL
200
+ architectures that jointly embed NL and pixel-based infor-
201
+ mation in order to predict an appropriate description of a
202
+ given input image. These techniques are typically trained
203
+ on large-scale datasets that contain images annotated with
204
+ multiple captions, such as MSCOCO [38], and have largely
205
+ drawn inspiration from encoder-decoder neural language mod-
206
+ els traditionally applied to machine translation tasks. In this
207
+ 2
208
+
209
+
210
+
211
+
212
+ Input Image
213
+ CNN or RCNN
214
+ BRNN
215
+ or LSTM
216
+ xt
217
+ yt
218
+ W
219
+ st
220
+ v
221
+ Image “Encoder”
222
+ NL “Decoder”
223
+ Fig. 1: Generalized overview of multimodal DL architectures
224
+ for image captioning (with RCNN)
225
+ paper, we adapt three recent architectures for image caption-
226
+ ing, neuraltalk2 [40], the im2txt [41], and the show,
227
+ attend and tell (SAT) [42] frameworks to predict func-
228
+ tional descriptions of software screenshots through the use of
229
+ custom pre-training and fine-tuning procedures. Additionally,
230
+ we explore the seq2seq neural language model.
231
+ DL models for image captioning build upon the success
232
+ of encoder-decoder neural language models. The im2txt
233
+ framework treats image captioning as a machine translation
234
+ problem, wherein the source “sentence” is an image, and
235
+ the target “translation” is a NL description. The generalized
236
+ architecture of such models is shown in Fig. 1. As illustrated,
237
+ these architectures replace the encoder RNN with a Convolu-
238
+ tional Neural Network (CNN), which have been shown to be
239
+ highly capable of learning rich image features [43], [44], [45].
240
+ Google’s implementation of im2txt uses a Long-Short Term
241
+ Memory (LSTM) RNN [46] for the “decoder” module, which
242
+ has also proven extremely effective when applied to machine
243
+ translation tasks. The decoder module of the neuraltalk2
244
+ architecture is composed of a Bidirectional RNN (BRNN) [47]
245
+ as opposed to an LSTM. Finally, the show, attend, &
246
+ tell (SAT) model [42] uses an LSTM decoder but with
247
+ the addition of an attention mechanism that can “attend” to
248
+ salient parts of the image representation by combining “hard”
249
+ and “soft” attention mechanisms.
250
+ III. OVERVIEW
251
+ In this section, we provide an “at-a-glance” overview of the
252
+ data-collection procedures and various analyses performed in
253
+ this paper. Figure 2 illustrates the four major components of
254
+ the paper. The first major task of our study is to derive a
255
+ suitable dataset of screenshot-caption pairs. We describe this
256
+ process in two parts: (i) the collection of screenshots (Sec.
257
+ IV-A), and (ii) the collection of captions from human workers
258
+ (Sec. IV-B). The result of this data-collection effort is the
259
+ CLARITY dataset, which contains 45,998 captions of 10,204
260
+ Android screenshots. Next, we aim to understand the lexical
261
+ properties of our captions through an empirical analysis in
262
+ order to better understand how easily they might be modeled
263
+ (Sec. V). Thus, we perform both a comparison of the the
264
+ cross-entropy of language models trained our caption corpus
265
+ to other popular SE corpora, and perform an LDA-based
266
+ topic analysis. Next, we discuss the process of configuring
267
+ and training three neural image captioning models, and one
268
+ 1 Clarity Dataset Collection
269
+ (Screenshots + GUI metadata + Captions)
270
+ 2 Naturalness & Topic Analysis
271
+ Cross-Entropy
272
+ Analysis
273
+ LDA-based
274
+ Topic Analysis
275
+
276
+
277
+
278
+ Input Image
279
+ CNN or RCNN
280
+ BRNN
281
+ or LSTM
282
+ xt
283
+ yt
284
+ W
285
+ st
286
+ v
287
+ Image “Encoder”
288
+ NL “Decoder”
289
+ 3 Train Image-Captioning and
290
+ Metadata Captioning Models
291
+ Image-Captioning
292
+ Models
293
+ Metadata-Captioning
294
+ Models
295
+ 4 !antitative and !alitative
296
+ Model Evaluations
297
+ “Ground Truth”
298
+ Captions
299
+ “Predicted”
300
+ Captions
301
+ +
302
+ Screens
303
+ Large-Scale
304
+ Human Evaluation
305
+ !antitative
306
+ Evaluation
307
+ with BLEU
308
+ Fig. 2: Overview of Dataset Collection and Analysis
309
+ sequence-based model to predict functional descriptions of
310
+ software GUIs (Sec. VI). Finally, we conclude our analysis
311
+ by measuring the accuracy of our trained models according to
312
+ both automated reference-based metrics (i.e., BLEU@n) and
313
+ via a large-scale human evaluation. (Sec. VII)
314
+ IV. DATASET COLLECTION
315
+ A. Screen & GUI Metadata Collection
316
+ The first step in deriving the CLARITY dataset is the
317
+ collection of a sizable and diverse dataset of screenshots and
318
+ GUI-metadata. We chose to focus this dataset derivation on the
319
+ Android platform for three main reasons: (i) Android is cur-
320
+ rently the most popular OS in the world [30], (ii) Android apps
321
+ are highly GUI-and gesture driven, making them a suitable
322
+ target for our investigation, and (iii) the Android screencap
323
+ and uiautomator tools facilitate the extraction of screenshots
324
+ and GUI-metadata from running apps. Fortunately, large-scale
325
+ datasets of Android screenshots and metadata are publicly
326
+ available in related literature [48], [35]. For this work, we took
327
+ advantage of the REDRAW [48], [36] dataset which contains
328
+ nearly 17k unique screenshots from 8,655 of the top-rated
329
+ apps from the Google Play Store. It should be noted that
330
+ another large-scale Android GUI dataset that contains a larger
331
+ number of screenshots, RICO, is also available [35]. However,
332
+ we chose to utilize the REDRAW dataset as it aligned with
333
+ one of our primary study objectives. That is, we aim to learn
334
+ latent feature information from both screenshots and GUI-
335
+ metadata. However, for the GUI-metadata to properly align
336
+ with the displayed content on a screen, the app must make use
337
+ of native Android components. Therefore, apps that primarily
338
+ display their information using web technologies, so-called
339
+ hybrid apps, would obscure the GUI-metadata and impact
340
+ our study findings. The REDRAW dataset contains a set of
341
+ screenshots that underwent several stages of filtering to remove
342
+ instances of hybrid apps along with other noise. Furthermore,
343
+ the REDRAW dataset contains a set of GUI-component images
344
+ 3
345
+
346
+ 2:04
347
+ Q Search
348
+ Stories
349
+ Play All
350
+ Add
351
+ Your Story
352
+ George
353
+ Amanda
354
+ Colby
355
+ [因
356
+ What's on your mind?
357
+ Photo
358
+ Guillermo Moreno with Josephine
359
+ Williams and 2 others.
360
+ Yesterday at 10:14 PM ·
361
+ Good friends, good food and a lot of laughs
362
+ Colby Harris and 23 others
363
+ 2 CommentsHello
364
+ 4
365
+ again!
366
+ Password
367
+ I forgot
368
+ ENTER
369
+ Don't have an account?Register
370
+ wordsearch
371
+ Animals-Countries-Cifies
372
+ PLAY
373
+ Uspresidents-Trademarks0000
374
+ 三12:34
375
+ Kids A-Z
376
+ Teacher Username
377
+ No Username? Start Here!日#
378
+ 12:27
379
+ The Dollar in Mexico
380
+ =
381
+ updated: Jun 19, 2017
382
+ Order by:
383
+ Sell
384
+ V
385
+ Select the bank of your choice
386
+ BAsE
387
+ Banco
388
+ Buy: 17.4129
389
+ Sell: 17.8129
390
+ BANCODE MEXICO
391
+ Buy: 17.9895
392
+ Sell: 17.9945
393
+ *Interbank dollar to 48 hours
394
+ OBANCO AZTECA
395
+ Buy:
396
+ 16.95
397
+ Sell:
398
+ 18.01
399
+ HSBC
400
+ Buy:
401
+ 17.68
402
+ Sell:
403
+ 18.17
404
+ BANORTE
405
+ Buy:
406
+ 16.85
407
+ Sell:
408
+ 18.25
409
+ Ixe
410
+ Buy:
411
+ 16.85
412
+ Sell:
413
+ 18.25
414
+ monex
415
+ Buy:
416
+ 17.67
417
+ Sell:
418
+ 18.28
419
+ Banamex
420
+ Buy:
421
+ 17.50
422
+ Sell:
423
+ 18.30
424
+ INBURSA
425
+ Buy:
426
+ 17.70
427
+ Grupo Financierc
428
+ Sell:
429
+ 18.30
430
+ Santander
431
+ Buy:
432
+ 17.50
433
+ Sell:
434
+ 18.30
435
+ BBVA
436
+ Bancomer
437
+ Buy:
438
+ 17.16
439
+ Sell:
440
+ 18.35
441
+ B BANCO DEL BAJIO
442
+ Buy:
443
+ 17.40
444
+ Sell:
445
+ 18.40
446
+ Scotiabank
447
+ Buy:
448
+ 16.80
449
+ Sell:
450
+ 18.42
451
+ Bx+
452
+ Buy:
453
+ 17.50
454
+ Sell:
455
+ 18.50
456
+ BANREGIO
457
+ Buy:
458
+ 17.40
459
+ Sell:
460
+ 18.502:04
461
+ Q Search
462
+ Stories
463
+ Play All
464
+ Add
465
+ Your Story
466
+ George
467
+ Amanda
468
+ Colby
469
+ [因
470
+ What's on your mind?
471
+ Photo
472
+ Guillermo Moreno with Josephine
473
+ Williams and 2 others.
474
+ Yesterday at 10:14 PM ·
475
+ Good friends, good food and a lot of laughs
476
+ Colby Harris and 23 others
477
+ 2 Comments000
478
+ </>AHigh Level Caption
479
+ The screen allows the user to
480
+ look at clothing categories
481
+ Low Level Captions
482
+ The top le! icon allows the user
483
+ to access the menu
484
+ The top right icon allows the
485
+ user to access the shopping cart
486
+ The center list of categories allows
487
+ the user to make a selection
488
+ The heart icon to the le! of the
489
+ shopping cart allows the user to
490
+ view favorites
491
+ Fig. 3: Example of captions from the CLARITY dataset.
492
+ labeled with their corresponding types (e.g., Button) which
493
+ we utilize later in our study (Sec. VI). The end result of this
494
+ filtering process was a total set of 17,203 candidate screens for
495
+ labeling. We refer readers to the REDRAW paper for complete
496
+ details of the filtering process [48].
497
+ B. Collection of Functional Descriptions
498
+ Once we derived a suitable set of screens, we needed to
499
+ manually label these screens with functional captions. This
500
+ process occurred in two steps: (i) first, we conducted a pilot
501
+ labeling study in order to develop and prove out a tagging
502
+ methodology suitable for large scale caption collection; (ii)
503
+ second, we performed a full scale data collection study using
504
+ Amazon’s Mechanical Turk Crowd-worker platform to collect
505
+ over 10k screens with functional descriptions.
506
+ 1) Caption Granularity: Intuitively, GUIs encode func-
507
+ tional information at multiple levels of granularity. For exam-
508
+ ple, if you were to ask a user or developer what the high-
509
+ level purpose of a given screen is, they might say “This
510
+ screen allows users to browse clothing categories”, as shown
511
+ in Fig. 3. These types of descriptions constitute the “high-
512
+ level” functionality of a given screen. However, a single screen
513
+ rarely implements only one functionality, and there may be
514
+ multiple functional properties that enable the screen’s high-
515
+ level functional purpose. User descriptions of these types
516
+ of functional properties are typically centered around the
517
+ interactive components of a screen, since these represent the
518
+ instances of actions (e.g., users “doing something”) that are
519
+ easily attributed to implemented functions. For example, in
520
+ the screen in Fig. 3, underlying functions include viewing
521
+ favorites, accessing a shopping cart, or selecting an item from
522
+ a list. These types of “low-level” screen properties centered
523
+ around GUI-components describe key constituent functional-
524
+ ity. Hence, in order to capture a holistic functional view of
525
+ each screen, we tasked participants with labeling each screen
526
+ with one “high-level” functional caption, and up to four “low-
527
+ level” functional captions. Fig. 3 shows these two categories
528
+ using actual captions collected as part of the CLARITY dataset.
529
+ 2) Pilot Data Collection Study: We developed an initial
530
+ image caption collection platform using a Java-based web
531
+ application. Using this system, the authors manually labeled
532
+ 743 screens with the caption granularities described earlier.
533
+ During this study, we discovered some instances of screens
534
+ with relatively little information displayed on them, making
535
+ it difficult to label them with functional attributes, even after
536
+ the filtering techniques discussed previously. Therefore, before
537
+ moving onto the large-scale caption collection with Mechani-
538
+ cal Turk (MTurk), at least one author manually inspected each
539
+ of the 17,203 candidate screens, and discarded those with a
540
+ severe lack of functionality. This resulted in a set of 16,311
541
+ candidate screens for the next phase of the study.
542
+ 3) Mechanical Turk Data Collection Study: To set up our
543
+ large-scale data-collection process, we adapted our web ap-
544
+ plication caption collection mechanism to work with MTurk’s
545
+ crowd worker platform. This involved configuring a Human
546
+ Intelligence Task (HIT) that provided workers with a set of
547
+ detailed instructions, displaying a screenshot from our dataset
548
+ alongside text entry boxes for one high-level functional caption
549
+ and up to four low-level functional captions (a limit was
550
+ imposed to normalize the amount of time workers would spend
551
+ on the HIT). This study was approved by the Institutional
552
+ Review Board of the authors’ affiliated institution.
553
+ Given that we aimed to collect high-quality functional
554
+ descriptions of screens in natural English, we targeted MTurk
555
+ users from primarily English speaking countries that had
556
+ completed at least 1,000 HITs and had a HIT approval rate
557
+ of at least 90%. We provided a detailed set of instructions
558
+ for labeling images with captions that clearly explained the
559
+ concept of high-level and low-level captions with examples,
560
+ and provided users with explicit instructions as well as DOs
561
+ and DONTs for the labeling task. The full set of instructions is
562
+ available in our online appendix [39]. With regard to caption
563
+ quality, we specifically had three major requirements: (i) that
564
+ the caption describes the perceived functionality of a screen
565
+ and not simply its appearance, (ii) that spatial references are
566
+ given for low-level captions (e.g., “the button in the top-left
567
+ corner of the screen”), and (iii) that captions be written in
568
+ complete English sentences with reasonably proper grammar.
569
+ We published batches of HIT tasks by sampling unique
570
+ screens from our set of 16,311 candidate screens, ensuring
571
+ that no user was assigned the same screen twice. The quality
572
+ of work from crowd-sourced tasks is not always optimal,
573
+ so as captions were submitted, they needed to be vetted for
574
+ quality. Thus, the captions for each screen were examined by
575
+ at least one author for the three quality attributes mentioned
576
+ above. If an author was unsure about whether a screen met
577
+ these quality attributes, it was reviewed by at least one other
578
+ author to reach a consensus. In total, 2,419 screens were
579
+ rejected and republished as new HITs due to quality issues.
580
+ In summary, 2,150 MTurk workers collected 45,998 captions
581
+ (across granularities) for 10,204 screens (≈5 screens per
582
+ participant), and over $2,400 was paid out.
583
+ V. EMPIRICAL DATASET ANALYSIS
584
+ The CLARITY dataset provides a rich source of data for
585
+ exploring the relationship between GUI-based and lexical
586
+ 4
587
+
588
+ #?
589
+ 日9:47
590
+ asos
591
+
592
+ HOME
593
+ CATEGORIES
594
+ NEWIN:CLOTHING
595
+ ACTIVEWEAR
596
+ TALL
597
+ JEANS
598
+ SHOES&SNEAKERS
599
+ -SHIRTS
600
+ SUNGIASSESTABLE I: LDA topics learned over high-level captions k = 15
601
+ Assigned Label
602
+ Top 7 Words
603
+ ”color options”
604
+ screen show app option color book differ
605
+ ”login or create acccount”
606
+ user screen allow account log creat app
607
+ ”select image from a list”
608
+ user screen allow select view list imag
609
+ ”map search by location”
610
+ screen locat search map user show find
611
+ TABLE II: LDA Topics learned on low-level captions k = 25
612
+ Assigned Label
613
+ Top 7 Words
614
+ ”page button”
615
+ page button top center bottom side left
616
+ ”select date”
617
+ avail date select one option theme present
618
+ ”camera button”
619
+ video imag photo pictur bottom camera
620
+ ”privacy policy banner”
621
+ titl just term blue banner privaci polici
622
+ software data. However, it is important to investigate the
623
+ semantic makeup of the collected captions in order to better
624
+ understand: (i) the latent topics they capture as well as (ii)
625
+ their naturalness and, hence, predictability. In this section we
626
+ carry out an empirical analysis of this phenomena guided by
627
+ the following two Research Questions (RQs):
628
+ • RQ1: What are the latent topics captured within the high-
629
+ and low-level captions in the CLARITY dataset?
630
+ • RQ2: How natural (i.e., predictable) are the high- and
631
+ low-level captions in the CLARITY dataset?
632
+ A. Analysis Methodology
633
+ 1) RQ1: Investigating Dataset Topics: To investigate the
634
+ latent topics in the CLARITY dataset, we learned topic models
635
+ over caption corpora representing different granularities of
636
+ functional descriptions. More specifically, we applied Latent
637
+ Dirichlet Allocation (LDA) [49] to both segmented high-
638
+ and low- level captions from the CLARITY dataset. In our
639
+ analysis, the set of captions for a specific screenshot in the
640
+ CLARITY dataset represents a document, and the entire set
641
+ of captions across screenshots for a given granularity (i.e.,
642
+ high or low level) constitutes a corpus. LDA has several
643
+ configurable hyper-parameters that impact the smoothing of
644
+ generated topics. These include k, the number of topics,
645
+ n which denotes the number of iterations of the sampling
646
+ algorithm (Gibbs sampling [50], in our case), as well as α
647
+ and β which impact topic distributions. We set α and β to
648
+ standard values for NL corpora, set n to 500, which proved to
649
+ be a sufficient for model convergence, and varied k between
650
+ 15, 25, 50, and 75 topics.
651
+ 2) RQ2: Analyzing the Naturalness of GUI Descriptions:
652
+ Past work has pioneered the notion of the naturalness of
653
+ software [51], which illustrated the fact that software, even
654
+ more so than NL, exhibits repetitive patterns that make it
655
+ predictable. This finding was recently further investigated and
656
+ the existence of certain natural patterns was confirmed [52]. To
657
+ illustrate naturalness, these past studies have learned statistical
658
+ n-gram language models over software corpora, and measured
659
+ the “perplexity” (or a log-transformed version, cross-entropy)
660
+ of these models, which represents the degree to which a model
661
+ is “surprised” by the patterns on a test corpus when trained on
662
+ a corpus from the same domain. A model with lower measured
663
+ 1
664
+ 2
665
+ 3
666
+ 4
667
+ 5
668
+ 6
669
+ 7
670
+ 8
671
+ 9
672
+ 10
673
+ 1
674
+ 2
675
+ 3
676
+ 4
677
+ 5
678
+ 6
679
+ 7
680
+ 8
681
+ 9
682
+ 10
683
+ 11
684
+ 12
685
+ N-gram order
686
+ Cross-Entropy
687
+ High-Level Captions
688
+ Low-Level Captions
689
+ Java Raw Code
690
+ Java w/o Syntax Tokens
691
+ Stack Overflow
692
+ Guntenberg
693
+ Fig. 4: Cross-entropy of the CLARITY dataset’s high and low-
694
+ Level captions compared to other corpora.
695
+ cross-entropy represents a higher predictive power, and thus,
696
+ a more natural underlying corpus.
697
+ We follow the methodology of these past studies to explore
698
+ the naturalness of the CLARITY dataset captions. Thus, similar
699
+ to the methodology for the previous RQ, we split the collected
700
+ captions into two corpora, one for the high-level descriptions,
701
+ and one for the low-level descriptions. We then learned inter-
702
+ polated n-gram models, using the mitlm [53] implementation
703
+ of Kneser-Ney smoothing [54], which has been shown to be
704
+ the most effective n-gram smoothing method [51], following
705
+ a ten-fold cross-validation procedure. We report the average
706
+ cross-entropy values across these experiments for both the high
707
+ and low-level corpora, compared to prior results [51], [52] for
708
+ other NL and software corpora.
709
+ B. Analysis Results
710
+ 1) RQ1: Results of Dataset Topic Modeling: We present
711
+ selected results of some of the most representative topics in
712
+ Tables I & II, complete with descriptive labels that we provide
713
+ for readability, and include all the results in our appendix [39].
714
+ These topics help to provide a descriptive illustration of some
715
+ of the latent patterns that exist in both the high and low level
716
+ CLARITY captions. The high-level captions illustrate several
717
+ screen-level topics, including searching on a map and adjusting
718
+ app settings. The low-level captions conversely capture topics
719
+ that describe component-level functionality, such as date selec-
720
+ tors, camera buttons, and back buttons. These results indicate
721
+ the existence of logical topics specific to the domain of GUIs
722
+ in our collected captions.
723
+ 2) RQ2: The Naturalness of Clarity Descriptions: The
724
+ results of our naturalness analysis are illustrated in Figure 4.
725
+ This figure shows the average cross entropy of the high- and
726
+ low- level captions from the CLARITY dataset compared to
727
+ several other corpora as calculated by Rahman et al. [52].
728
+ More specifically, the graph depicts the average ten-fold cross
729
+ entropy for: (i) The Gutenberg corpus containing over 3k
730
+ English books written by over a hundred different authors,
731
+ (ii) Java code from over 134 open source projects on GitHub,
732
+ (iii) Java without Syntax Tokens (i.e., separators, keywords,
733
+ and operators), and (iv) a Stack Overflow corpus consisting of
734
+ only the English descriptions from over 200k posts.
735
+ 5
736
+
737
+ 13
738
+ high
739
+ 12
740
+ low
741
+ 11
742
+ javaraw
743
+ java withoutsyntaxtokens
744
+ 10
745
+ stack overflow
746
+ 9
747
+ gutenberg
748
+ entropy
749
+ 8
750
+ cross
751
+ 6
752
+ 5
753
+ 4
754
+ 3
755
+ 2
756
+ 1
757
+ 1
758
+ 2
759
+ m
760
+ 4
761
+ 5
762
+ 6
763
+ 7
764
+ 8
765
+ 9
766
+ 10
767
+ n-gram lengthAs described earlier, the lower the cross-entropy is for a
768
+ particular dataset, the more natural it is. That is, the corpora
769
+ that exhibit lower cross entropy tend to exhibit stronger latent
770
+ patterns that can be effectively modeled and predicted. As we
771
+ see from Fig. 4, the CLARITY high and low level captions are
772
+ more natural than every dataset excluding raw Java code. It
773
+ should be noted that, comparatively, there are several factors
774
+ that could account for the observed lower cross entropy of the
775
+ CLARITY captions. For instance, such factors could include
776
+ other corpora having a larger size or having a more diverse
777
+ set of human authors and writing styles. However, we mainly
778
+ provide entropy measures of other datasets to provide context
779
+ for the predictability of the CLARITY dataset compared to
780
+ other popular corpora. Regardless of dataset differences, the
781
+ average ≈ 5 bits of entropy measured for the two datasets of
782
+ CLARITY captions signals that our collected descriptions ex-
783
+ hibit strong semantic patterns that can be effectively modeled
784
+ for prediction. Additionally, we observe that the cross-entropy
785
+ for the high and low-level captions are surprisingly similar.
786
+ Intuitively, one might expect that the low-level CLARITY
787
+ captions would exhibit more prevalent patterns due to the
788
+ repetitive use cases of certain GUI-components such as menu
789
+ buttons. This indicates the tendency of both datasets to exhibit
790
+ patterns that can be appropriately modeled. However, as we
791
+ illustrate in Sec. VII the ability for GUI-related information
792
+ to predict captions differs according to granularity.
793
+ VI. DEEP LEARNING FUNCTIONAL DESCRIPTIONS FROM
794
+ SOFTWARE GUIS
795
+ The results of the analysis from the previous section demon-
796
+ strate the presence of the latent patterns in the CLARITY
797
+ dataset of screenshots and captions. In this section, we detail
798
+ our methodology for investigating the capability of different
799
+ customized DL models to learn these patterns to predict
800
+ functional descriptions from two GUI representations.
801
+ A. Clarity Dataset Segmentation
802
+ We collected two different granularities of captions from
803
+ users to derive the CLARITY dataset (Sec. IV-B). For the
804
+ experiments in this section, we want to explore the model’s
805
+ ability to learn both high- and low-level functional descrip-
806
+ tions. Thus, we split the CLARITY dataset into two groups,
807
+ one containing only high level captions, and one containing
808
+ only low level ones. We also created a third dataset combining
809
+ the high and low level captions, in order to explore whether the
810
+ predictive capabilities of the models improved by aggregating
811
+ multiple granularities. It should be noted that each low-level
812
+ caption was treated as a single caption (i.e. each low-level
813
+ caption was treated as a separate data point) as is convention
814
+ with datasets containing multiple captions [38]. Each screen
815
+ in the dataset has both an associated screenshot and a GUI-
816
+ metadata file. In order to make for a fair comparison of
817
+ performance across various model configurations, we created
818
+ consistent training, validation and test partitions (80%, 10%,
819
+ 10% according to the number of images/GUI metadata files) to
820
+ be used across models. The NL text used as input to the models
821
+ TABLE III: Image Captioning Model Configs. used in Study
822
+ Model
823
+ Identifier
824
+ Caption Config.
825
+ Model Config.
826
+ im2txt-h-imgnet
827
+ High
828
+ im2txt-l-imgnet
829
+ Low
830
+ im2txt-c-imgnet
831
+ Combined
832
+ inception v3 trained on
833
+ imagenet
834
+ im2txt-h-comp
835
+ High
836
+ im2txt-l-comp
837
+ Low
838
+ im2txt-c-comp
839
+ Combined
840
+ Inception v3 fine-tuned on
841
+ Component Dataset
842
+ im2txt-h-fs
843
+ High
844
+ im2txt-l-fs
845
+ Low
846
+ im2txt
847
+ im2txt-c-fs
848
+ Combined
849
+ Inception v3 fine-tuned on
850
+ Full Screen Dataset
851
+ ntk2-h-imgnet
852
+ High
853
+ ntk2-l-imgnet
854
+ Low
855
+ ntk2-c-imgnet
856
+ Combined
857
+ VGGNet pre-trained on
858
+ ImageNet
859
+ ntk2-h-ft
860
+ High
861
+ ntk2-l-ft
862
+ Low
863
+ NeuralTalk 2
864
+ ntk2-c-ft
865
+ Combined
866
+ VGGNet pre-trained on
867
+ ImageNet with Fine
868
+ Tuning
869
+ sat-h
870
+ High
871
+ sat-l
872
+ Low
873
+ SAT
874
+ sat-c
875
+ Combined
876
+ VGGNet pre-trained on
877
+ ImageNet
878
+ TABLE IV: Subset of Model Hyper-paramters
879
+ Hyperparameter
880
+ im2txt
881
+ NeuralTalk2
882
+ SAT
883
+ Seq2Seq
884
+ Batch Size
885
+ 64
886
+ 16
887
+ 17
888
+ 64
889
+ Embedding Size
890
+ 512
891
+ 512
892
+ 512
893
+ 128
894
+ Decoder RNN Size/Units
895
+ 512
896
+ 512
897
+ 1024
898
+ 128
899
+ Optimizer
900
+ SGD
901
+ SGD
902
+ Adam
903
+ Adam
904
+ Initial Learning Rate
905
+ 2
906
+ 2
907
+ 0.001
908
+ 0.0001
909
+ Dropout Probability
910
+ 0.7
911
+ 0.7
912
+ 0.3
913
+ 0.8
914
+ was preprocessed according to the specific requirements for
915
+ each model implementation [55], [56], [57].
916
+ B. Image Captioning Model Configurations
917
+ We customize, train, and test the three neural image
918
+ captioning models, im2txt, neuraltalk2, and show,
919
+ attend, & tell (SAT) (Sec. II-A), on the screenshots
920
+ and captions of the CLARITY dataset. We choose to explore
921
+ these three models due to their different underlying design
922
+ decisions related to the type of utilized CNNs and RNNs
923
+ (Sec. II-A), as these differences may affect their performance
924
+ in our domain. It should be noted that in the course of our
925
+ experiments, we make several customizations to these models
926
+ through adaptions to pre-training and fine-tuning procedures.
927
+ However, given the typical number of parameters that consti-
928
+ tute these models, the training time can be quite prohibitive,
929
+ even on modern hardware. Thus, to control our experimental
930
+ complexity and investigate a number of model configurations
931
+ that can be trained in a reasonable amount of time, we fix
932
+ the values of the hyper-parameters for each model in our
933
+ experiments. We derived our utilized hyper-parameter values
934
+ by conducting random searches for optimal values of certain
935
+ parameters, and chose optimal parameters reported in prior
936
+ work for others. While we fix the hyper-parameters for these
937
+ models, we instead customize the configurations of our image
938
+ captioning models at the architectural level. Specifically, we
939
+ investigate how training the “encoder” CNN using different
940
+ datasets and training procedures effects the efficacy of the
941
+ model predictions. This type of analysis allows us to more
942
+ effectively flush out broader patterns related to the benefits and
943
+ drawbacks of model design decisions. In the end, we trained
944
+ more than 15 different configurations of the models (see Table
945
+ III) over several machine months of computation.
946
+ 1) im2txt
947
+ Model
948
+ Configurations
949
+ &
950
+ Training:
951
+ For
952
+ im2txt, we adapted Google’s open source implementa-
953
+ 6
954
+
955
+ tion of the model in TensorFlow [55]. Given the incredibly
956
+ large number of parameters that need to be trained for the
957
+ im2txt model, performing even relatively simple hyper-
958
+ paramter searches proved to be computationally prohibitive
959
+ for our experiments. Therefore, for this model we utilized the
960
+ optimal set of parameters reported by Vinyals et al. [41] on
961
+ similarly sized datasets. A subset of these hyper-parameter
962
+ values are given in Table IV, whereas full configuration details
963
+ can be found in our appendix. The publicly available imple-
964
+ mentation of Google’s im2txt model utilizes the Inception
965
+ v3 [58] image captioning architecture as its encoder CNN.
966
+ In past work, the inception model weights were initialized
967
+ by training on the large-scale image classification dataset
968
+ ImageNet [59], which contains “commonplace” image cate-
969
+ goires. However, given that we are applying these models to
970
+ very particular domain (predicting descriptions of software)
971
+ it is unclear if an Inception v3 model trained on the broader
972
+ ImageNet dataset would capture subtle semantic patterns in
973
+ the CLARITY dataset. Therefore, we explored three different
974
+ model configurations to explore this phenomena: one with
975
+ Inception v3 pre-trained on ImageNet, and two with Inception
976
+ v3 fine-tuned on domain specific-datasets. The first domain
977
+ specific image dataset we utilize is the ReDraw cropped
978
+ image dataset outlined in Sec. IV-A, which contains over 190k
979
+ images of native Android GUI-components labeled with their
980
+ type (e.g., Button, TextView). The second domain specific
981
+ image dataset we use consists of the full screenshots from the
982
+ CLARITY dataset, labeled with their Google Play categories.
983
+ 2) NeuralTalk2 Model Configurations & Training: For
984
+ neuraltalk2, we adapted Karpathy et al.’s implementation
985
+ written in Torch and lua [56]. We performed a brief random-
986
+ ized hyper-parameter search for this model, given its more
987
+ efficient training time, using the optimal im2txt parameters as
988
+ a starting point. The optimal values resulting from this search
989
+ are provided in Table IV. For its CNN decoder, neuraltalk2
990
+ makes use of a VGGNet [44] architecture pre-trained on
991
+ the ImageNet [59] dataset. Unlike our im2txt configurations,
992
+ we explore the effect of jointly fine-tuning neuraltalk2’s
993
+ CNN and RNN. Thus, we explore two configurations of
994
+ neuraltalk2, one that jointly fine tunes the pre-trained
995
+ VGGNet on the CLARITY dataset, and one that does not
996
+ perform fine-tuning. We followed a training procedure similar
997
+ to that of our im2txt models, in that we trained our models
998
+ on the high, low, and combined CLARITY caption training
999
+ data for 500K iterations, saving model checkpoints every 2K
1000
+ iterations.
1001
+ 3) Show, Attend and Tell Model Configurations & Train-
1002
+ ing: For the SAT model, we adapted the open-source imple-
1003
+ mentation of the model in Tensorflow [60]. The hyperparam-
1004
+ eters that we used to train our model are shown in Table IV.
1005
+ The implementation used VGG16 [44] as its encoder CNN.
1006
+ We trained the SAT model on the CLARITY dataset for the
1007
+ low, high and combined captions for 500K iterations and kept
1008
+ the checkpoints after every 1K iterations. Note that due to
1009
+ the prohibitive training cost of this model, we did not explore
1010
+ using a fine-tuned VGGNet as we did with neuraltalk2.
1011
+ TABLE V: Metadata Captioning Model Congfigurations
1012
+ Model
1013
+ Identifier
1014
+ Caption Config.
1015
+ Model Config.
1016
+ seq2seq-h-type
1017
+ High
1018
+ seq2seq-l-type
1019
+ Low
1020
+ seq2seq-c-type
1021
+ Combined
1022
+ Trained on GUI
1023
+ Component Types
1024
+ seq2seq-h-text
1025
+ High
1026
+ seq2seq-l-text
1027
+ Low
1028
+ seq2seq-c-text
1029
+ Combined
1030
+ Trained on
1031
+ GUI-Component Text
1032
+ seq2seq-h-tt
1033
+ High
1034
+ seq2seq-l-tt
1035
+ Low
1036
+ seq2seq-c-tt
1037
+ Combined
1038
+ Trained on
1039
+ GUI-Component Type +
1040
+ Text
1041
+ seq2seq-h-ttl
1042
+ High
1043
+ seq2seq-l-ttl
1044
+ Low
1045
+ Seq2Seq
1046
+ seq2seq-c-ttl
1047
+ Combined
1048
+ Trained on
1049
+ GUI-component Type +
1050
+ Text + Location
1051
+ C. Metadata Captioning Model Configurations
1052
+ To explore the ability to translate between the lexical
1053
+ representations of GUI-metadata and NL functional descrip-
1054
+ tions, we train and test an encoder-decoder neural language
1055
+ model using Google’s seq2seq [57] framework. Note that
1056
+ recent work has proposed new models that take advantage of
1057
+ structural text properties [61], however, implementations of
1058
+ such models are generally not available, hence we leave the
1059
+ study of more advanced models for future work. We chose
1060
+ to utilize the default general-purpose architecture and hyper-
1061
+ parameters for this model, as they have been shown to be
1062
+ effective across a wide-range of machine translation tasks [62].
1063
+ More specifically, our encoder network consists of a BRNN
1064
+ with Gated Recurrent Units (GRUs) and our decoder network
1065
+ consists of an RNN with LSTM units; hyperparameters are
1066
+ listed in Table IV.
1067
+ To investigate the representative power of different attributes
1068
+ included in Android GUI-metadata, we create four config-
1069
+ urations of GUI-metadata consisting of different attribute
1070
+ combinations (Table V). We chose to utilize these attribute
1071
+ combinations as they represent (i) the attributes that are most
1072
+ likely to have values, and (ii) represent a wide range of
1073
+ information types (e.g., displayed text, component types, and
1074
+ spatial information). Note that seq2seq did not consistently
1075
+ converge for the high level caption dataset, thus we do not
1076
+ report these results. Consistent with the training of the other
1077
+ models, our implementation of the seq2seq model was
1078
+ trained to 500k iterations, with checkpoints every 2k iterations.
1079
+ VII. DEEP LEARNING MODEL EVALUATION
1080
+ To explore our core hypothesis set forth at the beginning
1081
+ of this paper, and evaluate our DL models described in
1082
+ Sec. VI, we perform a comprehensive empirical evaluation
1083
+ with two main goals: (i) intrinsically evaluate the predictive
1084
+ power of the models according to a well accepted machine
1085
+ translation effectiveness metric, and (ii) extrinsically evaluate
1086
+ the models by examining and rating the quality of the pred-
1087
+ icated functional NL descriptions. The quality focus of this
1088
+ evaluation is our studied models’ ability to effectively predict
1089
+ accurate, concise, and complete functional descriptions. To aid
1090
+ in achieving our study goals, we define the following RQs:
1091
+ • RQ3: How accurate are our model’s predicted NL de-
1092
+ scriptions?
1093
+ • RQ4: How accurate, complete, & understandable are our
1094
+ model’s predicted NL descriptions from the viewpoint of
1095
+ evaluators?
1096
+ 7
1097
+
1098
+ TABLE VI: BLEU Score Evaluation Results for Models
1099
+ Model
1100
+ Capt.
1101
+ Model Type
1102
+ Bc
1103
+ B1
1104
+ B2
1105
+ B3
1106
+ B4
1107
+ High
1108
+ im2txt-h-fs
1109
+ 12.4
1110
+ 24.8
1111
+ 12.6
1112
+ 6.7
1113
+ 5.3
1114
+ Low
1115
+ im2txt-l-comp
1116
+ 27.0
1117
+ 45.6
1118
+ 31.8
1119
+ 20.0
1120
+ 10.1
1121
+ im2txt
1122
+ Comb.
1123
+ im2txt-c-comp
1124
+ 30.3
1125
+ 51.7
1126
+ 35.9
1127
+ 22.1
1128
+ 11.6
1129
+ High
1130
+ ntk2-h-imgnet
1131
+ 13.3
1132
+ 27.4
1133
+ 13.5
1134
+ 7.3
1135
+ 5.3
1136
+ Low
1137
+ ntk2-l-ft
1138
+ 27.4
1139
+ 47.5
1140
+ 32.8
1141
+ 19.5
1142
+ 9.6
1143
+ NeuralTalk2
1144
+ Comb.
1145
+ ntk2-c-ft
1146
+ 30.1
1147
+ 52.1
1148
+ 36.0
1149
+ 21.8
1150
+ 10.8
1151
+ Low
1152
+ seq2seq-l-type
1153
+ 18.1
1154
+ 44.6
1155
+ 17.0
1156
+ 7.9
1157
+ 0.24
1158
+ seq2seq
1159
+ Comb.
1160
+ seq2seq-c-type
1161
+ 16.9
1162
+ 38.9
1163
+ 14.7
1164
+ 6.0
1165
+ 0.08
1166
+ High
1167
+ sat-h
1168
+ 17.7
1169
+ 30.1
1170
+ 18.3
1171
+ 12.9
1172
+ 9.8
1173
+ Low
1174
+ sat-l
1175
+ 35.0
1176
+ 52.5
1177
+ 38.7
1178
+ 28.1
1179
+ 20.7
1180
+ SAT
1181
+ Comb.
1182
+ sat-c
1183
+ 37.7
1184
+ 56.8
1185
+ 42.0
1186
+ 30.5
1187
+ 22.0
1188
+ NeuralTalk2
1189
+ Trained on Flickr8K
1190
+ 34.0
1191
+ 57.9
1192
+ 38.3
1193
+ 24.5
1194
+ 16.0
1195
+ NeuralTalk2
1196
+ Trained on MSCOCO
1197
+ 40.7
1198
+ 62.5
1199
+ 45.0
1200
+ 32.1
1201
+ 23.0
1202
+ im2txt
1203
+ 42.6
1204
+ 66.6
1205
+ 46.1
1206
+ 32.9
1207
+ 24.6
1208
+ SAT
1209
+ 45.7
1210
+ 71.8
1211
+ 50.4
1212
+ 35.7
1213
+ 25.0
1214
+ A. Evaluation Methodology
1215
+ 1) RQ3: Empirically Evaluating Model Accuracy: To
1216
+ evaluate the accuracy of our trained model’s generated cap-
1217
+ tions, we follow past work [40], [41] and report BLEU
1218
+ scores [63] of the predicted captions on the shared CLARITY
1219
+ test set of images and GUI-metadata. The BLEU score is a
1220
+ standard metric used in machine translation research that mea-
1221
+ sures the textual similarity between a predicted caption (the
1222
+ output from a model) and a reference caption (the collected
1223
+ descriptions from humans in the CLARITY test set). The BLEU
1224
+ score can be measured according to the similarity of different
1225
+ subsequence lengths (i.e., BLEUn), and we report BLEU1
1226
+ through BLEU4, as well as a composite score calculated as the
1227
+ average of these, as is convention [40], [41]. For the image
1228
+ captioning models, we use the coco-caption implementation
1229
+ of the BLEU score adapted for the CLARITY test set. For
1230
+ each test image across all image captioning models, three
1231
+ captions were generated using a beam width of 3 for the
1232
+ beam search across candidate predictions. The seq2seq models
1233
+ were evaluated in the same manner. We chose to utilize a
1234
+ beam width of 3 as an initial qualitative examination of our
1235
+ models’ predictions showed this size to achieve a reasonable
1236
+ balance between prediction accuracy and model confidence.
1237
+ For the high-level captions, the three candidate captions were
1238
+ compared to the reference, and the overall average BLEUn
1239
+ scores were calculated for each model. For the low-level and
1240
+ combined captions, the predicted captions and reference cap-
1241
+ tions were compared in a pairwise manner and overall average
1242
+ BLEUn scores were calculated for each model configuration.
1243
+ 2) RQ4: Human Perceptions of Predicted Captions: To
1244
+ qualitatively evaluate our studied model’s generated captions,
1245
+ we performed a large-scale study involving an additional 220
1246
+ participants recruited from MTurk. We randomly sampled
1247
+ 220 screens from the CLARITY test set, and then predicted
1248
+ high, low, and combined captions for them using the opti-
1249
+ mal configurations of im2txt, NeuralTalk2, and seq2seq
1250
+ according to the composite BLEU score for each model
1251
+ and caption level combination. The SAT captions were not
1252
+ included in this study due to time constraints related to the
1253
+ model’s training. We created a HIT wherein each participant
1254
+ viewed 11 screenshots paired with captions. Two of the 11
1255
+ captions were reference high and low to serve as a control,
1256
+ while the other 9 captions came from the model predictions.
1257
+ Screens and caption pairs were arranged into HITs such that
1258
+ 1) no single HIT had two of the same screenshot, 2) each
1259
+ of the 11 types of captions (2 reference, 9 model) were
1260
+ included only once per HIT. The order of these captions was
1261
+ randomized per HIT to prevent bias introduced by identical
1262
+ caption ordering between HITs. By this arrangement, each
1263
+ screen-caption pair was evaluated by 11 participants. After
1264
+ viewing these screenshot-caption pairs, participants were asked
1265
+ to answer six evaluation questions. Three of these questions
1266
+ (EQ1-EQ3) were adapted from prior work that assessed the
1267
+ quality of automatically generated code summaries [21], and
1268
+ inquired about accuracy, completeness, and understandability,
1269
+ respectively. The three remaining questions (EQ4-EQ6), were
1270
+ free response and asked participants to explain accuracies,
1271
+ inaccuracies, and improvements. The full set of questions and
1272
+ HIT are in our online appendix [39]. Similar to the CLARITY
1273
+ dataset collection, each participant’s response was thoroughly
1274
+ vetted by at least one author, and discarded if the answers
1275
+ were incomplete. Responses were collected until 220 HITs
1276
+ were completed by unique respondents.
1277
+ B. Evaluation Results
1278
+ 1) RQ3 Results: Evaluating BLEU Scores: We illustrate
1279
+ the BLEU score results for the most effective model config-
1280
+ uration and checkpoint across all of our trained models in
1281
+ Table VI, whereas the results for other model configurations
1282
+ can be found in our online appendix [39] in addition to
1283
+ caption examples. The cells highlighted in blue illustrate the
1284
+ highest performing model configuration for each caption type.
1285
+ In general we observe that SAT exhibits the highest overall
1286
+ BLEU scores across all caption granularities. We speculate
1287
+ that this is attributable to the addition of the advanced attention
1288
+ mechanism in this model that is able to “focus” on varying
1289
+ image regions or features to effectively handle multiple caption
1290
+ granularities. In general, the seq2seq model performed quite
1291
+ poorly across the varying caption types, indicating a lower
1292
+ tendency for rich representation. Perhaps most interestingly,
1293
+ we see that the optimal model configurations for the im2txt
1294
+ framework were those where the CNN was conditioned on
1295
+ domain specific datasets. More specifically, the best high-level
1296
+ caption model was conditioned on full screenshots and the
1297
+ best low-level caption was conditioned on the cropped GUI-
1298
+ component screenshots.
1299
+ Another general trend that emerges is the low-level and
1300
+ combined caption models tend to exhibit higher overall BELU
1301
+ scores compared to the high-level captions. This is somewhat
1302
+ intuitive, as it indicates that there are more natural connections
1303
+ between visual GUI and lexical patterns in the low-level
1304
+ captions, compared to the high-level captions that reflect more
1305
+ abstract functional descriptions. When examining the captions
1306
+ generated by the optimal configurations of each model, it is
1307
+ clear that im2txt and SAT produces a more diverse set of out-
1308
+ put captions than neuraltalk2, which could be considered
1309
+ as more useful in many software documentation tasks.
1310
+ Finally, it is worth discussing how the BLEU scores of our
1311
+ models compare to those of the same models trained on the
1312
+ 8
1313
+
1314
+ None
1315
+ Some
1316
+ A Lot
1317
+ im2txt high
1318
+ im2txt low
1319
+ im2txt combined
1320
+ Easy to Read
1321
+ Somewhat
1322
+ Readable
1323
+ Hard to
1324
+ Read
1325
+ im2txt high
1326
+ im2txt low
1327
+ im2txt combined
1328
+ EQ3: Understandability
1329
+ EQ2: Unnecessary Information
1330
+ im2txt high
1331
+ im2txt low
1332
+ im2txt combined
1333
+ Strongly
1334
+ Disagree
1335
+ Disagree
1336
+ Neutral
1337
+ Agree
1338
+ Strongly
1339
+ Agree
1340
+ EQ1: Accuracy
1341
+ seq2seq high
1342
+ seq2seq low
1343
+ seq2seq combined
1344
+ Fig. 5: Responses across models for EQ1-EQ3
1345
+ more traditional Flickr8k [37] and MSCOCO [38] datasets
1346
+ given at the bottom of Table VI. Given the data-intensive
1347
+ nature of our DL models, and the much larger size of the
1348
+ MSCOCO dataset (≈123k images, each with 5 captions), we
1349
+ did not expect our models trained on the CLARITY dataset to
1350
+ outperform those trained on MSCOCO. Thus, unsurprisingly,
1351
+ we observe that on average, im2txt, neuraltalk2, and SAT
1352
+ models trained on the MSCOCO dataset outperform the same
1353
+ models trained on the CLARITY datasets by ≈ 10 BLEU score
1354
+ points for the combined and low level captions, and ≈ 27
1355
+ points on high-level captions. However, when we examine
1356
+ the performance of Neuraltalk2 on the more similarly sized
1357
+ Flickr8K dataset (≈ 8K images, each with 5 captions) we
1358
+ observe comparable performance to the CLARITY low-level
1359
+ and combined datasets, with the SAT model narrowly outper-
1360
+ forming the Flickr8K neuraltalk2 model, with a slightly
1361
+ bigger discrepancy for the high-level captions. Overall, these
1362
+ results indicate that when compared with datasets of similar
1363
+ size, DL models trained on the CLARITY dataset exhibit
1364
+ similar performance.
1365
+ 2) RQ4 Results: Human Evaluations: The results of EQ1-
1366
+ EQ3 for the model configurations with the best performance
1367
+ during the human study, in addition to the seq2seq accuracy
1368
+ scores, are summarized in Fig. 5. Complete results across all
1369
+ model configurations can be found in our online appendix. The
1370
+ responses to EQ4-EQ6 varied by the type of caption, and are
1371
+ provided in our appendix in full. Generally, im2txt fared the
1372
+ best in terms of accuracy, and was followed by neuraltalk2
1373
+ and seq2seq respectively. For im2txt, despite mixed reac-
1374
+ tions from participants, in many cases respondents verified that
1375
+ the caption was accurate (e.g., ”The description accurately
1376
+ describes the screen, it is in fact a terms and conditions
1377
+ screen.”) and suggested minor improvements similarly to the
1378
+ reference captions (e.g., ”It could add specifics about what the
1379
+ settings pertain to (i.e. security)”). As illustrated in Fig. 5 the
1380
+ im2txt predictions were consistently rated as being readable
1381
+ and containing relevant information. It is also interesting to
1382
+ note that there appears to a mismatch between the performance
1383
+ as indicated by BLEU scores, and human perceptions, with the
1384
+ participants consistently rating the im2txt captions better
1385
+ than other models across EQ1-EQ3, despite neuraltalk2
1386
+ achieving a higher BLEU score for two model configurations.
1387
+ VIII. DISCUSSION & LEARNED LESSONS
1388
+ Lesson 1: Functional Descriptions of GUIs exhibit a
1389
+ high degree of naturalness and can be modeled using DL
1390
+ techniques. We observed that DL models trained on the low-
1391
+ level and combined datasets exhibit similar performance to
1392
+ models trained on general image captioning datasets of similar
1393
+ size (e.g., Flickr8K). This indicates that GUI screenshots could
1394
+ be used to augment approaches for automated documentation.
1395
+ Lesson 2: GUI-centric software documentation mod-
1396
+ els benefit from being pre-trained on domain specific
1397
+ GUI data, as opposed to general image datasets (e.g.,
1398
+ MSCOCO) The qualitative results of our model analysis
1399
+ illustrate that for im2txt, the most effective configurations
1400
+ were those trained on domain specific CNN datasets. This
1401
+ suggests a perceptible difference between the utility of image
1402
+ features learned from general datasets, compared to those
1403
+ learned on datasets more specific to software. This suggests
1404
+ that future work aiming to leverage DL models for GUI-
1405
+ centric program documentation should look to collecting and
1406
+ extracting features from large-scale GUI-related datasets.
1407
+ Lesson 3: Future automated approaches for GUI-centric
1408
+ program documentation would likely benefit from com-
1409
+ bining the orthogonal semantics of screenshots and GUI-
1410
+ metadata. Our evaluation in this paper illustrates that the rep-
1411
+ resentational power of screenshots appears to be superior when
1412
+ applied to a software documentation task. However, given stark
1413
+ differences between these two modalities of information, we
1414
+ also observed that they encode orthogonal semantic patterns
1415
+ that could be combined for more effective documentation
1416
+ generation. One property we observed of certain captions
1417
+ generated by the image-based models was the effect of their
1418
+ limited vocabulary. For example, certain predicted captions
1419
+ similar to the following: “The screen allows the user to select
1420
+ a <UNK>”, wherein the UNK token represents missing token,
1421
+ which should be mapped to some unobserved app property,
1422
+ such as a “album cover” or “store location”. However, such
1423
+ predictions could be combined with the vocabulary present in
1424
+ GUI metadata to help predict more complete, and accurate
1425
+ descriptions. Thus, a promising direction for future work is to
1426
+ jointly encode both screenshots and lexical GUI-metadata.
1427
+ Lesson 4: Training image captioning models to predict
1428
+ specific or diverse pieces of functionality is difficult.
1429
+ Practical models for GUI-centric documentation should able
1430
+ to predict both specific pieces of information (e.g. the func-
1431
+ tionality of a particular button for a given method handler),
1432
+ and diverse functionality (being able to generate descriptions
1433
+ of functionality anywhere on a given screen). However, one
1434
+ aspect we observed across our models is that the most
1435
+ common observed types of functionality (e.g., back buttons,
1436
+ menu buttons) corresponded to the functionalities that our
1437
+ 9
1438
+
1439
+ seq2seq high
1440
+ seq2seq low
1441
+ 0
1442
+ 0
1443
+ seq2seq combined
1444
+ 0
1445
+ 0models predicted most often and most confidently on unseen
1446
+ screenshots. This is somewhat expected, as the models saw the
1447
+ most examples of such functionalities during training. Thus,
1448
+ the diversity of predictions is an open problem for future
1449
+ research. This problem can be partially mitigated by larger,
1450
+ more diverse datasets with specifically curated descriptions
1451
+ (such as extensions to the CLARITY dataset). However, it is
1452
+ likely that domain-specific models, or ensembles of models,
1453
+ may be required to more effectively predict diverse features.
1454
+ Lesson 5: Future studies that evaluate automated GUI-
1455
+ centric documentation approaches should include human
1456
+ studies, as human perceptions of models may differ from
1457
+ automated reference-based metrics. One of the more surpris-
1458
+ ing results of our study is that there seems to be a mismatch
1459
+ between humans perceptions of the captions generated by our
1460
+ DL models and the BLEU score metrics typically used to asses
1461
+ the accuracy of model predictions. This signifies that there are
1462
+ aspects of human perception that are not effectively captured
1463
+ in the BLEU metric, and possibly other translation metrics.
1464
+ IX. LIMITATIONS & THREATS TO VALIDITY
1465
+ Internal Validity. Threats to internal validity correspond to
1466
+ unexpected factors in the experiments that may contribute to
1467
+ observed results. To derive our dataset we rely on MTurk and
1468
+ its workers to extract the high- and low-level descriptions per
1469
+ each screenshot. It should be noted that we did not ask MTurk
1470
+ workers to provide technical software documentation descrip-
1471
+ tions, but rather general descriptions of screen functionality at
1472
+ differing granularities. To minimize low quality captions we
1473
+ published the jobs for workers with more than 1k HITS, from
1474
+ English speaking countries, and HIT approval rate of more
1475
+ than 90%. Also, each successfully completed HIT was vetted
1476
+ by at least one of the authors to assure quality. If there was
1477
+ any question related to caption quality, at least one of the other
1478
+ authors stepped in to resolve the ambiguity. As a result 2,429
1479
+ HITs were rejected due to low quality descriptions.
1480
+ External Validity. Threats to external validity concern the
1481
+ generalization of the results. As with any collected dataset,
1482
+ there is a threat to external validity about the generalizability
1483
+ of the CLARITY dataset. However, we used a diverse set
1484
+ of popular apps from the Android domain, extracted popular
1485
+ screenshots from these apps, and the apps were captioned by
1486
+ a large and diverse set of MTurk workers. During our data
1487
+ collection process, we only collected 4 low-level captions per
1488
+ each screen in order to make the task feasible for MTurk
1489
+ workers as workers tend to abandon or perform poorly on long
1490
+ tasks. This means that, for certain screens with many GUI-
1491
+ components, some components may lack natural language
1492
+ descriptions. However, given the size of our dataset and the
1493
+ diversity of our screenshots and captions, we assert that our
1494
+ low-level captions are reasonably representative.
1495
+ X. RELATED WORK
1496
+ DL for Image Captioning and GUIs. Hossain et al. [64]
1497
+ recently performed a wide-ranging study on DL models for
1498
+ image captioning, surveying the many different architectures
1499
+ and datasets used to evaluate them. However, this survey
1500
+ did not examine the ability of any image captioning model
1501
+ to predict functional descriptions of software. There have
1502
+ been a limited number of papers in the SE community that
1503
+ have applied DL techniques to GUI related data. Chen et
1504
+ al. [65] designed an approach that uses an NMT to translate
1505
+ an Android screenshot into a GUI-skeleton. However, their
1506
+ technique is able to predict GUI structure given an image, not
1507
+ functional natural language descriptions. Recently, Zhang et.
1508
+ al. [66] created a dataset of iOS image captions to train a
1509
+ model for captioning accessibility data. However, the authors
1510
+ do not make their dataset publicly available and target a
1511
+ different goal of accessibility data compared our goal of
1512
+ generating functional captions. Chen et al. investigated the
1513
+ use of DL image captioning models for applying labels to
1514
+ GUI-components in mobile apps [67], however, this approach
1515
+ only aims to predict short labels for a limited subset of
1516
+ GUI-components, whereas our study focuses upon predicting
1517
+ functional descriptions consisting of complete sentences for
1518
+ both individual GUI-components and entire screenshots.
1519
+ GUI-based Analysis of Mobile Apps. GVT and GCat analyze
1520
+ the visual properties of GUIs to detect design violations and
1521
+ evolutionary changes [68], [69]. In contrast, we focus solely on
1522
+ image captioning techniques to provide functional program de-
1523
+ scriptions of screenshots. Approaches such as REMAUI [70],
1524
+ REDRAW [48], and pix2code [71] aim to automatically
1525
+ generate mobile app code given an app screenshot. Conversely,
1526
+ we leverage DL techniques to generate functional descriptions
1527
+ rather than source code using a pixel-based image as input.
1528
+ Chen et al. [72] introduced StoryDroid, for automatically
1529
+ generating visual storyboards of Android apps to help aid
1530
+ in the app design process. However, their approach is not
1531
+ capable of generating a functional description of an application
1532
+ from GUI data. Furthermore, Deka et al. showed how the
1533
+ Rico dataset could be navigated via semantic search using
1534
+ autoencoders
1535
+ [35]. UiRef [73] is an approach for resolving
1536
+ security and privacy concerns by considering semantics of
1537
+ GUI-components that request user’s inputs. Moreover, Liu et
1538
+ al. [74] presented an approach for automatically classifying
1539
+ mobile app icons according to semantic GUI patterns. Xiao et
1540
+ al. proposed IconIntent that combines program analysis and
1541
+ icon classification to detect privacy sensitive GUI-components
1542
+ [75]. Different from this body of work, we aim to predict
1543
+ functional descriptions of GUIs for software documentation.
1544
+ XI. CONCLUSION
1545
+ In this paper, we have conducted one of the first com-
1546
+ prehensive empirical investigations into the connection be-
1547
+ tween GUI-related information, and functional descriptions
1548
+ of programs. We have derived the CLARITY dataset of GUI
1549
+ screenshots/metadata and NL captions, trained DL models
1550
+ on this dataset, and demonstrated their ability to bridge the
1551
+ semantic gap between visual and lexical program information.
1552
+ ACKNOWLEDGMENT
1553
+ This work was supported by the NSF CCF-2007246 &
1554
+ CCF-1955853 grants. Any opinions, findings, and conclusions
1555
+ expressed herein are the authors’ and do not necessarily reflect
1556
+ those of the sponsors.
1557
+ 10
1558
+
1559
+ REFERENCES
1560
+ [1] J.-C. Chen and S.-J. Huang, “An empirical analysis of the impact of
1561
+ software development problem factors on software maintainability,” J.
1562
+ Syst. Softw., vol. 82, pp. 981–992, June 2009.
1563
+ [2] M. Kajko-Mattsson, “A survey of documentation practice within correc-
1564
+ tive maintenance,” Empirical Software Engineering, vol. 10, pp. 31–55,
1565
+ Jan 2005.
1566
+ [3] B. Dagenais and M. P. Robillard, “Creating and evolving developer
1567
+ documentation: Understanding the decisions of open source contribu-
1568
+ tors,” in Proceedings of the Eighteenth ACM SIGSOFT International
1569
+ Symposium on Foundations of Software Engineering, FSE ’10, (New
1570
+ York, NY, USA), pp. 127–136, ACM, 2010.
1571
+ [4] G. Garousi, V. Garousi-Yusifoglu, G. Ruhe, J. Zhi, M. Moussavi, and
1572
+ B. Smith, “Usage and usefulness of technical software documentation:
1573
+ An industrial case study,” Information and Software Technology, vol. 57,
1574
+ pp. 664 – 682, 2015.
1575
+ [5] A. Forward and T. C. Lethbridge, “The relevance of software documen-
1576
+ tation, tools and technologies: A survey,” in Proceedings of the 2002
1577
+ ACM Symposium on Document Engineering, DocEng ’02, (New York,
1578
+ NY, USA), pp. 26–33, ACM, 2002.
1579
+ [6] “The agile manfiesto http://agilemanifesto.org.”
1580
+ [7] J. Zhi, V. Garousi-Yusiflu, B. Sun, G. Garousi, S. Shahnewaz, and
1581
+ G. Ruhe, “Cost, benefits and quality of software development docu-
1582
+ mentation,” J. Sys. Sof., Jan. 2015.
1583
+ [8] B. Fluri, M. Wursch, and H. C. Gall, “Do code and comments co-
1584
+ evolve? on the relation between source code and comment changes,” in
1585
+ Proceedings of the 14th Working Conference on Reverse Engineering,
1586
+ WCRE ’07, (Washington, DC, USA), pp. 70–79, 2007.
1587
+ [9] L. Moreno, A. Marcus, L. Pollock, and K. Vijay-Shanker, “JSummarizer:
1588
+ An automatic generator of natural language summaries for Java classes,”
1589
+ in 2013 21st International Conference on Program Comprehension,
1590
+ ICPC’13, pp. 230–232, May 2013.
1591
+ [10] L. Moreno, J. Aponte, G. Sridhara, A. Marcus, L. Pollock, and K. Vijay-
1592
+ Shanker, “Automatic generation of natural language summaries for Java
1593
+ classes,” in 2013 21st International Conference on Program Compre-
1594
+ hension, ICPC’13, pp. 23–32, May 2013.
1595
+ [11] P. C. Rigby and M. P. Robillard, “Discovering Essential Code Elements
1596
+ in Informal Documentation,” in Proceedings of the 2013 International
1597
+ Conference on Software Engineering, ICSE’13, (San Francisco, CA,
1598
+ USA), pp. 832–841, 2013.
1599
+ [12] A. T. T. Ying and M. P. Robillard, “Selection and Presentation Prac-
1600
+ tices for Code Example Summarization,” in Proceedings of the 22Nd
1601
+ ACM SIGSOFT International Symposium on Foundations of Software
1602
+ Engineering, FSE’14, (Hong Kong, China), pp. 460–471, 2014.
1603
+ [13] A. T. T. Ying and M. P. Robillard, “Code Fragment Summarization,” in
1604
+ Proceedings of the 2013 9th Joint Meeting on Foundations of Software
1605
+ Engineering, FSE’13, pp. 655–658, 2013.
1606
+ [14] I. K. Ratol and M. P. Robillard, “Detecting fragile comments,” in
1607
+ 2017 32nd IEEE/ACM International Conference on Automated Software
1608
+ Engineering, ASE’17, pp. 112–122, Oct. 2017.
1609
+ [15] M. J. Howard, S. Gupta, L. Pollock, and K. Vijay-Shanker, “Automati-
1610
+ cally mining software-based, semantically-similar words from comment-
1611
+ code mappings,” in 2013 10th Working Conference on Mining Software
1612
+ Repositories, MSR’13, pp. 377–386, may 2013.
1613
+ [16] C. Treude and M. P. Robillard, “Augmenting API Documentation with
1614
+ Insights from Stack Overflow,” in Proceedings of the 38th International
1615
+ Conference on Software Engineering, ICSE ’16, (New York, NY, USA),
1616
+ pp. 392–403, 2016.
1617
+ [17] G. Petrosyan, M. P. Robillard, and R. De Mori, “Discovering Information
1618
+ Explaining API Types Using Text Classification,” in Proceedings of
1619
+ the 37th International Conference on Software Engineering - Volume
1620
+ 1, ICSE ’15, (Florence, Italy), pp. 869–879, 2015.
1621
+ [18] B. Li, C. Vendome, M. Linares-V´asquez, D. Poshyvanyk, and N. A.
1622
+ Kraft, “Automatically Documenting Unit Test Cases,” in 2016 IEEE In-
1623
+ ternational Conference on Software Testing, Verification and Validation,
1624
+ ICST’16, pp. 341–352, Apr. 2016.
1625
+ [19] L. Moreno, W. Bandara, S. Haiduc, and A. Marcus, “On the Relationship
1626
+ between the Vocabulary of Bug Reports and Source Code,” in 2013 IEEE
1627
+ International Conference on Software Maintenance, ICSM’13, pp. 452–
1628
+ 455, Sept. 2013.
1629
+ [20] K. Moran, M. Linares-V´asquez, C. Bernal-C´ardenas, and D. Poshy-
1630
+ vanyk, “Auto-completing Bug Reports for Android Applications,” in
1631
+ Proceedings of the 2015 10th Joint Meeting on Foundations of Software
1632
+ Engineering, FSE’15, (Bergamo, Italy), pp. 673–686, 2015.
1633
+ [21] L. Moreno, G. Bavota, M. D. Penta, R. Oliveto, A. Marcus, and
1634
+ G. Canfora, “Arena: An approach for the automated generation of release
1635
+ notes,” IEEE Transactions on Software Engineering, vol. 43, pp. 106–
1636
+ 127, Feb 2017.
1637
+ [22] L. Moreno, G. Bavota, M. Di Penta, R. Oliveto, A. Marcus, and
1638
+ G. Canfora, “Automatic Generation of Release Notes,” in Proceedings
1639
+ of the 22Nd ACM SIGSOFT International Symposium on Foundations
1640
+ of Software Engineering, FSE’14, pp. 484–495, 2014.
1641
+ [23] S. Jiang, A. Armaly, and C. McMillan, “Automatically generating
1642
+ commit messages from diffs using neural machine translation,” in
1643
+ 2017 32nd IEEE/ACM International Conference on Automated Software
1644
+ Engineering, ASE’17, pp. 135–146, oct 2017.
1645
+ [24] L. F. Cort´es-Coy, M. Linares-V´asquez, J. Aponte, and D. Poshyvanyk,
1646
+ “On automatically generating commit messages via summarization of
1647
+ source code changes,” in 2014 IEEE 14th International Working Con-
1648
+ ference on Source Code Analysis and Manipulation, SCAM’14, Sep.
1649
+ 2014.
1650
+ [25] P. Chatterjee, B. Gause, H. Hedinger, and L. Pollock, “Extracting
1651
+ Code Segments and Their Descriptions from Research Articles,” in
1652
+ Proceedings of the 14th International Conference on Mining Software
1653
+ Repositories, MSR ’17, (Buenos Aires, Argentina), pp. 91–101, 2017.
1654
+ [26] M. Linares-V´asquez, B. Li, C. Vendome, and D. Poshyvanyk, “Doc-
1655
+ umenting database usages and schema constraints in database-centric
1656
+ applications,” in Proceedings of the 25th International Symposium on
1657
+ Software Testing and Analysis, ISSTA 2016, (New York, NY, USA),
1658
+ pp. 270–281, 2016.
1659
+ [27] T. J. Biggerstaff, B. G. Mitbander, and D. E. Webster, “Program
1660
+ understanding and the concept assignment problem,” Commun. ACM,
1661
+ vol. 37, pp. 72–82, May 1994.
1662
+ [28] J. Guo, J. Cheng, and J. Cleland-Huang, “Semantically enhanced soft-
1663
+ ware traceability using deep learning techniques,” in Proceedings of
1664
+ the 39th International Conference on Software Engineering, ICSE ’17,
1665
+ (Piscataway, NJ, USA), pp. 3–14, 2017.
1666
+ [29] M. P. Robillard, A. Marcus, C. Treude, G. Bavota, O. Chaparro, N. Ernst,
1667
+ M. A. Gerosa, M. Godfrey, M. Lanza, M. Linares-V´asquez, G. C.
1668
+ Murphy, L. Moreno, D. Shepherd, and E. Wong, “On-demand Developer
1669
+ Documentation,” in 2017 IEEE International Conference on Software
1670
+ Maintenance and Evolution, ICSME’17, pp. 479–483, Sept. 2017.
1671
+ [30] “Android inches ahead of windows as most popular os - https://www.
1672
+ cnet.com/news/android-most-popular-os-beats-windows-statcounter/.”
1673
+ [31] K. Moran, C. Bernal-C´ardenas, M. Linares-V´asquez, and D. Poshy-
1674
+ vanyk, “Overcoming lanaguge dichotomies: Toward effective program
1675
+ comprehension for mobile app development,” in 26th IEEE International
1676
+ Conference on Program Compreehnsion, May 2018. To Appear.
1677
+ [32] A. M. Memon, A comprehensive framework for testing graphical user
1678
+ interfaces. Ph.D., 2001. Advisors: Mary Lou Soffa and Martha Pol-
1679
+ lack; Committee members: Prof. Rajiv Gupta (University of Arizona),
1680
+ Prof. Adele E. Howe (Colorado State University), Prof. Lori Pollock
1681
+ (University of Delaware).
1682
+ [33] A. Solem, “Designing computer documentation that will be used: Un-
1683
+ derstanding computer user attitudes,” in Proceedings of the 4th Annual
1684
+ International Conference on Systems Documentation, SIGDOC ’85,
1685
+ (New York, NY, USA), p. 55–56, Association for Computing Machinery,
1686
+ 1986.
1687
+ [34] A. Alshayban, I. Ahmed, and S. Malek, “Accessibility issues in android
1688
+ apps: State of affairs, sentiments, and ways forward,” in Proceedings
1689
+ of the 42nd International Conference on Software Engineering, ICSE
1690
+ 2020, (New York, NY, USA), p. to appear, ACM, 2020.
1691
+ [35] B. Deka, Z. Huang, C. Franzen, J. Hibschman, D. Afergan, Y. Li,
1692
+ J. Nichols, and R. Kumar, “Rico: A mobile app dataset for building
1693
+ data-driven design applications,” in Proceedings of the 30th Annual
1694
+ Symposium on User Interface Software and Technology, UIST ’17, 2017.
1695
+ [36] K. Moran, C. Bernal-Cardenas, M. Curcio, R. Bonett, and D. Poshy-
1696
+ vanyk, “The ReDraw Dataset: A Set of Android Screenshots, GUI Meta-
1697
+ data, and Labeled Images of GUI Components,” Jan. 2019. Supported
1698
+ by NSF Grant CCF-1815186.
1699
+ [37] M. Hodosh, P. Young, and J. Hockenmaier, “Framing image description
1700
+ as a ranking task: Data, models and evaluation metrics,” J. Artif. Int.
1701
+ Res., vol. 47, no. 1, pp. 853–899, 2013.
1702
+ [38] T. Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan,
1703
+ P. Doll´ar, and C. L. Zitnick, “Microsoft COCO: Common Objects in
1704
+ Context,” in Lecture Notes in Computer Science (including subseries
1705
+ Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinfor-
1706
+ matics), vol. 8693 LNCS, pp. 740–755, 2014.
1707
+ 11
1708
+
1709
+ [39] “Online appendix http://sagelab.io/Clarity/.”
1710
+ [40] A. Karpathy and L. Fei-Fei, “Deep visual-semantic alignments for
1711
+ generating image descriptions,” IEEE Trans. Pattern Anal. Mach. Intell.,
1712
+ vol. 39, pp. 664–676, Apr. 2017.
1713
+ [41] O. Vinyals, A. Toshev, S. Bengio, and D. Erhan, “Show and tell: Lessons
1714
+ learned from the 2015 mscoco image captioning challenge,” IEEE Trans.
1715
+ Pattern Anal. Mach. Intell., vol. 39, pp. 652–663, Apr. 2017.
1716
+ [42] K. Xu, J. Ba, R. Kiros, K. Cho, A. C. Courville, R. Salakhutdinov, R. S.
1717
+ Zemel, and Y. Bengio, “Show, attend and tell: Neural image caption
1718
+ generation with visual attention,” in ICML, 2015.
1719
+ [43] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification
1720
+ with Deep Convolutional Neural Networks,” in Advances in Neural In-
1721
+ formation Processing Systems 25 (F. Pereira, C. J. C. Burges, L. Bottou,
1722
+ and K. Q. Weinberger, eds.), pp. 1097–1105, 2012.
1723
+ [44] K. Simonyan and A. Zisserman, “Very deep convolutional networks for
1724
+ large-scale image recognition,” International Conference on Learning
1725
+ Represen., vol. abs/1409.1556, 2014.
1726
+ [45] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan,
1727
+ V. Vanhoucke, and A. Rabinovich, “Going Deeper with Convolutions,”
1728
+ in Computer Vision and Pattern Recognition, 2015.
1729
+ [46] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural
1730
+ Comput., vol. 9, no. 8, pp. 1735–1780, 1997.
1731
+ [47] M. Schuster and K. Paliwal, “Bidirectional recurrent neural networks,”
1732
+ Trans. Sig. Proc., vol. 45, pp. 2673–2681, Nov. 1997.
1733
+ [48] K. P. Moran, C. Bernal-C´ardenas, M. Curcio, R. Bonett, and D. Poshy-
1734
+ vanyk, “Machine learning-based prototyping of graphical user interfaces
1735
+ for mobile apps,” IEEE Transactions on Software Engineering, pp. 1–1,
1736
+ 2018.
1737
+ [49] D. M. Blei, A. Y. Ng, and M. I. Jordan, “Latent dirichlet allocation,” J.
1738
+ Mach. Learn. Res., vol. 3, pp. 993–1022, Mar. 2003.
1739
+ [50] I. Porteous, D. Newman, A. Ihler, A. Asuncion, P. Smyth, and
1740
+ M. Welling, “Fast collapsed gibbs sampling for latent dirichlet al-
1741
+ location,” in Proceedings of the 14th ACM SIGKDD International
1742
+ Conference on Knowledge Discovery and Data Mining, KDD ’08, (New
1743
+ York, NY, USA), pp. 569–577, 2008.
1744
+ [51] A. Hindle, E. T. Barr, Z. Su, M. Gabel, and P. Devanbu, “On the
1745
+ Naturalness of Software,” in International Conference on Software
1746
+ Engineering, ICSE’12, pp. 837–847, 2012.
1747
+ [52] M. Rahman, D. Palani, and P. Rigby, “Natural software revisted,” in Pro-
1748
+ ceedings of the 41st International Conference on Software Engineering
1749
+ Companion, ICSE ’19, (Montreal, QC Canada), p. to appear, 2019.
1750
+ [53] “Mit language modeling toolkit https://github.com/mitlm/mitlm.”
1751
+ [54] P. Koehn, Statistical Machine Translation. 1st ed., 2010.
1752
+ [55] “Google’s im2txt implementation https://github.com/tensorflow/models/
1753
+ tree/master/research/im2txt.”
1754
+ [56] “Neuraltalk2 implementation https://github.com/karpathy/neuraltalk2.”
1755
+ [57] “Google seq2seq framework https://github.com/google/seq2seq.”
1756
+ [58] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking
1757
+ the inception architecture for computer vision,” in 2016 IEEE Confer-
1758
+ ence on Computer Vision and Pattern Recognition, June 2016.
1759
+ [59] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma,
1760
+ Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and
1761
+ L. Fei-Fei, “ImageNet Large Scale Visual Recognition Challenge,”
1762
+ International Journal of Computer Vision, vol. 115, no. 3, pp. 211–252,
1763
+ 2015.
1764
+ [60] “Show, attend and tell implementation https://github.com/coldmanck/
1765
+ show-attend-and-tell.”
1766
+ [61] L. Sha, L. Mou, T. Liu, P. Poupart, S. Li, B. Chang, and Z. Sui,
1767
+ “Order-planning neural text generation from structured data,” The
1768
+ Thirty-Second AAAI Conference on Artificial Intelligence (AAAI’17),
1769
+ vol. abs/1709.00155, 2017.
1770
+ [62] D. Britz, A. Goldie, T. Luong, and Q. Le, “Massive Exploration of
1771
+ Neural Machine Translation Architectures,” ArXiv e-prints, Mar. 2017.
1772
+ [63] K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “Bleu: A method for
1773
+ automatic evaluation of machine translation,” in Proceedings of the 40th
1774
+ Annual Meeting on Association for Computational Linguistics, ACL ’02,
1775
+ (Stroudsburg, PA, USA), pp. 311–318, 2002.
1776
+ [64] M. Z. Hossain, F. Sohel, M. F. Shiratuddin, and H. Laga, “A compre-
1777
+ hensive survey of deep learning for image captioning,” ACM Comput.
1778
+ Surv., vol. 51, pp. 118:1–118:36, Feb. 2019.
1779
+ [65] S. Chen, L. Fan, C. Chen, T. Su, W. Li, Y. Liu, and L. Xu, “Storydroid:
1780
+ Automated generation of storyboard for android apps,” in Proceedings of
1781
+ the 41st International Conference on Software Engineering, ICSE ’19,
1782
+ (Piscataway, NJ, USA), pp. 596–607, IEEE Press, 2019.
1783
+ [66] X. Zhang, L. de Greef, A. Swearngin, S. White, K. Murray, L. Yu,
1784
+ Q. Shan, J. Nichols, J. Wu, C. Fleizach, A. Everitt, and J. P. Bigham,
1785
+ Screen Recognition: Creating Accessibility Metadata for Mobile Appli-
1786
+ cations from Pixels. New York, NY, USA: Association for Computing
1787
+ Machinery, 2021.
1788
+ [67] J. Chen, C. Chen, Z. Xing, X. Xu, L. Zhu, G. Li, and J. Wang, “Unblind
1789
+ your apps: Predicting natural-language labels for mobile gui components
1790
+ by deep learning,” in Proceedings of the 42nd International Conference
1791
+ on Software Engineering, ICSE 2020, (New York, NY, USA), p. to
1792
+ appear, ACM, 2020.
1793
+ [68] K. Moran, C. Watson, J. Hoskins, G. Purnell, and D. Poshyvanyk,
1794
+ “Detecting and summarizing gui changes in evolving mobile apps,”
1795
+ in Proceedings of the 33rd ACM/IEEE International Conference on
1796
+ Automated Software Engineering, ASE 2018, (New York, NY, USA),
1797
+ pp. 543–553, 2018.
1798
+ [69] K. Moran, B. Li, C. Bernal-C´ardenas, D. Jelf, and D. Poshyvanyk,
1799
+ “Automated Reporting of GUI Design Violations in Mobile Apps,” in
1800
+ Proceedings of the 40th International Conference on Software Engineer-
1801
+ ing Companion, ICSE ’18, (Gothenburg, Sweden), p. to appear, 2018.
1802
+ [70] T. A. Nguyen and C. Csallner, “Reverse Engineering Mobile Application
1803
+ User Interfaces with REMAUI,” in Proceedings of the 2015 30th
1804
+ IEEE/ACM International Conference on Automated Software Engineer-
1805
+ ing, ASE’15, (Washington, DC, USA), pp. 248–259, 2015.
1806
+ [71] T. Beltramelli, “Pix2code: Generating Code from a Graphical User
1807
+ Interface Screenshot,” CoRR, vol. abs/1705.07962, 2017.
1808
+ [72] Q. Chen and M. Zhou, “A neural framework for retrieval and summariza-
1809
+ tion of source code,” in Proceedings of the 33rd ACM/IEEE International
1810
+ Conference on Automated Software Engineering, ASE 2018, (New York,
1811
+ NY, USA), pp. 826–831, ACM, 2018.
1812
+ [73] B. Andow, A. Acharya, D. Li, W. Enck, K. Singh, and T. Xie, “UiRef:
1813
+ analysis of sensitive user inputs in Android applications,” in Proceedings
1814
+ of the 10th ACM Conference on Security and Privacy in Wireless and
1815
+ Mobile Networks, WiSec’17, (New York, New York, USA), pp. 23–34,
1816
+ 2017.
1817
+ [74] T. F. Liu, M. Craft, J. Situ, E. Yumer, R. Mech, and R. Kumar, “Learning
1818
+ design semantics for mobile apps,” in The 31st Annual ACM Symposium
1819
+ on User Interface Software and Technology, UIST ’18, (New York, NY,
1820
+ USA), pp. 569–579, 2018.
1821
+ [75] X. Xiao, X. Wang, Z. Cao, H. Wang, and P. Gao, “Iconintent: Automatic
1822
+ identification of sensitive ui widgets based on icon classification for
1823
+ android apps,” in Proceedings of the 41st International Conference on
1824
+ Software Engineering Companion, ICSE ’19, (Montreal, QC Canada),
1825
+ p. to appear, 2019.
1826
+ 12
1827
+
3NAzT4oBgHgl3EQfR_tE/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
59AyT4oBgHgl3EQf2fl-/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:822b9e4c15fb82315031971c642ab1d6b0e8408dc11577eec11f886b5df38249
3
+ size 231937
69E5T4oBgHgl3EQfPw7T/content/2301.05508v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:581e91aaecbf299e62728ada4709d56eac549f36caf731506a73172c0d11a8c1
3
+ size 293173
69E5T4oBgHgl3EQfPw7T/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5598c7d1ac76fa45c072c6d8a8df2ea74a6e087873402e8c4ca5015284dede8
3
+ size 132698
7tAzT4oBgHgl3EQfSPsA/content/tmp_files/2301.01227v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
7tAzT4oBgHgl3EQfSPsA/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
8NFAT4oBgHgl3EQfox1h/content/tmp_files/2301.08636v1.pdf.txt ADDED
@@ -0,0 +1,987 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.08636v1 [math.NA] 20 Jan 2023
2
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE
3
+ PROBLEM
4
+ Hajri Imen 1 Fethi Ben Belgacem 2
5
+ Abstract. In this article, we introduce and study three numerical methods for the Dirichlet
6
+ Monge-Ampère equation in two dimensions. The approaches consist in considering new equivalent
7
+ problems. The latter are discretized by a wide stencil finite difference discretization and monotone
8
+ schemes are obtained. Hence, we apply the Barles-Souganidis theory to prove the convergence of
9
+ the schemes and the Damped Newtons method is used to compute the solutions of the schemes.
10
+ Finally, some numerical results are illustrated.
11
+ Monge-Ampere, Monotone scheme, Newton method.
12
+ 1. Introduction
13
+ We are interested in the numerical solution of the Monge-Ampère equation with Dirichlet bound-
14
+ ary condition
15
+ (1.1)
16
+ (MAD)
17
+
18
+
19
+
20
+
21
+
22
+ det
23
+
24
+ D2u (x)
25
+
26
+ = f (x) , for x in Ω,
27
+ u (x) = ϕ (x) , for x on ∂Ω,
28
+ u is convex.
29
+ Where Ω is a convex bounded domain in R2, with boundary ∂Ω, (D2u) , is the Hessian of the
30
+ function u, f and ϕ are given functions.
31
+ We take the simplest boundary conditions. For more general operator of Monge-Ampère and
32
+ other boundary conditions, we mention for instance [12]. The convexity constraint is crucial for
33
+ the (MAD). It is required for the Monge-Ampère equation to be degenerate elliptic and for (MAD)
34
+ to have a unique solution. It is also needed for numerical stability. The Monge-Ampère equation,
35
+ has extensive applications, it is strictly related to the “prescribed Gauss curvature” problem, see
36
+ for instance [12]. It appears also in affine geometry, precisely, in the affine sphere problem and the
37
+ affine maximal surfaces problem, this was discussed in [5, 6, 25, 27, 28, 29]. Other applications
38
+ appear in fluid mechanics, geometric optics, and meteorology : for example, in semigeostrophic
39
+ equations, the Monge-Ampère equation is coupled with a transport equation, this is pointed out
40
+ in [12]. The analysis of the regularity of the Monge-Ampere equation is essential in the study
41
+ 1Higher
42
+ Institute
43
+ of
44
+ Applied
45
+ Studies
46
+ in
47
+ Humanities
48
+ of
49
+ Mahdia,5121
50
+ Mahdia,
51
+ Tunisia,
52
+ Email:hajri.imene2017@gmail.com.
53
+ 2Laboratory of partial differential equations (LR03ES04), ISIMM, University of Monastir, El Manar, TUNISIA.
54
+ Email: fethi.benbelgacem@isimm.rnu.tn
55
+ 1
56
+
57
+ 2
58
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
59
+ of the regularity of the transp ort problem. This, latter, has been employed in many areas. We
60
+ only briefly mention [8, 9, 4] for mesh geneartion,[15, 16, 17]for image registration, and [12] for
61
+ reflector design. Developing an efficient numerical method has aroused a lot of interest, and large
62
+ standard techniques have been proposed. A first method to do so was introduced in [24] by using
63
+ a discretization of the geometric Alexandrov-Bakelman interpretation of solutions. Variational
64
+ approaches have been presented in [10, 11], more precisely, the augmented Lagrangian approach
65
+ and the least-squares approach. But these methods needed more regularity than can be predicted
66
+ for solutions. A different approach was studied in [18], using the vanishing moment method. The
67
+ periodic case was treated in [14].
68
+ Although, the standard techniques, mentioned above, work well for smooth solutions, and they
69
+ fail for singular solutions, for more details, see, for instance, the discussion in [2]. To overcome
70
+ these difficulties, we have to use the notion of viscosity solution or Alexsandrov solution. In two
71
+ dimension, a numerical method was introduced in [24], which is geometric in nature, and converges
72
+ to the Alexsandrov solution. The method introduced in [22]„ in two dimension and improved in
73
+ [19] for higher dimension, uses the wide stencil scheme that converges to the viscosity solution,
74
+ which we briefly describe for this reason in the end of this section.
75
+ The following variant of the AM-GM inequalities, is the keystone of our formulation introduced
76
+ here.
77
+ For A and B two symmetric matrices, such that, A, B ≥ 0. We have the following inequality
78
+ 2
79
+
80
+ det (AB) ≤ Tr (AB) .
81
+ Where for symmetric matrices M ≥ 0 means xT Mx ≥ 0.
82
+ Remark 1. We can deduce from the above inequality that for a smooth convex solution u of (1.1),
83
+ one can deduce the following inequality
84
+ ∆u − 2
85
+
86
+ f ≥ 0.
87
+ Let us define the function
88
+ ˜g := ∆u − 2
89
+
90
+ f.
91
+ It is then straightforward to check that if u is a smooth solution of (1.1), then is indeed a solution
92
+ of the linear Dirichlet Poisson problem
93
+ (1.2)
94
+
95
+ P˜g� �
96
+ ∆u = 2√f + ˜g,
97
+ u|Γ = ϕ,
98
+ which can be easily descretized by any method of choice if the function ˜g is known.
99
+ We finish this remark by mentioning that the convexity constraint is essential to ensure unique-
100
+ ness (for example, u and −u are both solution of the Monge Ampère equation). For viscosity
101
+ solution, this constraint can be required by the equation
102
+ (1.3)
103
+ λ1
104
+
105
+ D2u
106
+
107
+ ≥ 0,
108
+
109
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
110
+ 3
111
+ in the viscosity sense, see for instance [21, 22], where λ1 (D2u) is the smallest eigenvalue of the
112
+ Hessian of u. However, for a twice continuously differentiable function u, the convexity restriction
113
+ is equivalent to requiring that the eigenvalues of the Hessian, D2u, are positives, which is approved
114
+ by considering the linear Poisson Dirichlet problem
115
+
116
+ P˜g�
117
+ .
118
+ The approaches that we follow, in the present paper, are inspired by the idea developed in [3]
119
+ and the wide stencil finite difference discretization introduced in [22] and [19] for viscosity solution
120
+ of M-A equation in two and higher dimensions that relies on a framework developed in [1]. For
121
+ clarity, we recall the full result in the next section.
122
+ 2. Viscosity solution and convergence theory of approximation schemes
123
+ 2.1. Degenerate elliptic equations. Let F (x, r, p, X) be a continuous real valued function de-
124
+ fined on Ω × R × Rn × Sn, with Sn being the space of symmetric n × n matrices. Consider the
125
+ nonlinear, partial differential equation with Dirichlet boundary conditions,
126
+
127
+ F (x, u (x) , Du (x) , D2u (x)) (x) = 0
128
+ for x in Ω
129
+ u (x) = g (x)
130
+ for x in ∂Ω.
131
+ Where Ω is a domain in Rn, Du and D2u denote the gradient and Hessian of u, respectively.
132
+ Definition 2. [19]The equation F is degenerate elliptic if
133
+ F (x, r, p, X) ≤ F (x, s, p, Y ) whenever r ≤ s and Y ≤ X.
134
+ Where Y ≤ X means that Y − X is a nonnegative definite symmetric matrix.
135
+ The viscosity solution for the Monge-Ampère equation is defined in [22].
136
+ Definition 3. Let u ∈ C (Ω) be convex and f ≥ 0 be continuous. The function u is a viscosity
137
+ subsolution (supersolution) of the Monge-Ampère equation in Ω if whenever convex ϕ ∈ C2 (Ω)
138
+ and x0 ∈ Ω are such that (u − ϕ) (x) ≤ (≥) (u − ϕ) (x0) for all x in a neighborhood of x0, then we
139
+ must have
140
+ det
141
+
142
+ D2φ (x0)
143
+
144
+ ≥ (≤) f (x0) .
145
+ The function u is a viscosity solution if it is both a viscosity subsolution and supersolution.
146
+ For the existence and uniqueness of viscosity solution for (1.1), we mention the next result in
147
+ [7],
148
+ Theorem 4. Let Ω ⊆ Rd be abounded and strictly convex, g ∈ C (∂Ω) , f ∈ C (Ω) , with f ≥ 0.
149
+ Then there exists a unique convex viscosity solution u ∈ C
150
+ � ¯Ω
151
+
152
+ of the problem (1.1).
153
+ The advantage of considering viscosity solutions come from the following fundamental theorem,
154
+ obtained in [1], which gives conditions for convergence of approximation schemes to viscosity
155
+ solution.
156
+
157
+ 4
158
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
159
+ Theorem 5. (Convergence of Approximation Schemes). Consider a degenerate elliptic equation,
160
+ for which there exist unique viscosity solutions. A consistent, stable approximation scheme con-
161
+ verges uniformly on compact subsets to the viscosity solution, provided it is monotone.
162
+ By the previous theorem, we need just a way to build a monotone finite difference schemes,
163
+ which represents a new challenge. In the sequel, we recall here the basic framework introduced in
164
+ [20], for building a monotone scheme.
165
+ Firstly, a finite difference equation take the form
166
+ F i [u] = F i (ui, ui − uj|i̸=j) .
167
+ We say that a scheme is degenerate elliptic if the following holds [20]:
168
+ Definition 6. The scheme F is degenerate elliptic if F i is non-decreasing in each variable.
169
+ We are now ready to present the following theorem in [20]:
170
+ Theorem 7. Under mild analytic conditions, degenerate elliptic schemes are monotone, and non-
171
+ expansive in the uniform norm. The iteration
172
+ (2.1)
173
+ um+1 = um + dtF (um) ,
174
+ is a contraction in L∞ provided dt ≤ K (F)−1 , where K (F) is the Lipschitz constant of the scheme,
175
+ regarded as a function from RN −→ RN.
176
+ We end this paragraphre by the next result, proven in [20]
177
+ Theorem 8. A proper, locally Lipschitz continuous degenerate elliptic scheme has a unique solu-
178
+ tion which is stable in the l∞ norm.
179
+ 2.2. Wide stencil schemes. We finish this section by noting that wide stencil schemes are re-
180
+ quired to build consistent, monotone schemes of degenerate second order PDEs (see discussion in
181
+ [22]). Wide stencil schemes were built for the two-dimensional Monge-Ampère equation in [22]
182
+ and for the convex envelope in [21]. Each approach considered here is a function of eigenvalues of
183
+ the Hessian. To fully discretize the equation (4.1) for the eigenvalues of the Hessian on a finite
184
+ difference grid, we approximate the second derivatives by centered finite differences; this is the
185
+ spatial discretization, with parameter h. We consider also a finite number of possible directions ν
186
+ that lie on the grid; this is the directional discretization, with parameter dθ. The spatial resolution
187
+ is improved by using more grid points, the directional resolution is improved by increasing the size
188
+ of the stencil. So, a wide stencil is needed (see Fig 2.2)
189
+ 3. First Formulation of the (MAD) in two dimensions (method A)
190
+ 3.1. An equivalent problem. Let us begin with a simple approach to illustrate the ideas. We
191
+ can rephrase, for instance, the (MAD) as the following:
192
+ (3.1)
193
+
194
+ Find a positive function g, such that
195
+ det (D2ug) = λ1 [D2ug] × λ2 [D2ug] = f,
196
+
197
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
198
+ 5
199
+ Figure 2.1. Grid for wide stencil 17 points, in two dimension
200
+ where: ug is the solution of�
201
+ ∆ug = λ1 [D2ug] + λ2 [D2ug]
202
+ = 2√f + g,
203
+ ug
204
+
205
+ = ϕ.
206
+ We are now ready to state a first example of our approches.
207
+ Lemma 9. Provided the solution, u, of (1.1) is in H2, there exists a unique positive function
208
+ ˜g ∈ L2, such that u = u˜g, where u˜g is the solution of (P˜g). Conversely, if u¯g is solution of (3.1)
209
+ for some ¯g > 0, then u¯g = u.
210
+ Proof. Let u be a solution of (1.1). From the above, one can see easily, that u = u˜g.
211
+ Conversely, if u¯g is a solution of (3.1), we can clearly see that
212
+
213
+ det (D2u¯g) = f > 0
214
+ ∆u¯g ≥ 0.
215
+ It follows that u¯g is convex and satisfies (1.1).
216
+
217
+ Remark 10. We notice that according to the result in [26], we have equivalence of viscosity and
218
+ weak solutions for the Poisson problem. This motivates us to build a convergent scheme to the
219
+ viscosity solution of Poisson problem
220
+
221
+ P˜g�
222
+ through the discretization of the (MAD) problem. The
223
+ viscosity solution u˜g of
224
+
225
+ P˜g�
226
+ will be equivalent to the weak solution of (MAD) problem in the
227
+ distributional sense.
228
+ 4. Discretization of the problem (3.1)
229
+ Let us consider a regular and uniform cartesian grid, consider the stencil at the reference point
230
+ x0 consist of the neighbors x1, ..., xN (as in Figure 1). We can define vi in polar coordinates by
231
+ vi = xi − x0 = hivθi.
232
+
233
+ 6
234
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
235
+ We assume that the stencil is symetric and we define the local spatial resolution and the directional
236
+ resolution respectively by
237
+ ¯h (x0) = max
238
+ i
239
+ hi
240
+ and
241
+ dθ = max
242
+ θ∈[−π,π] min
243
+ i
244
+ |θ − θi|.
245
+ First, the problem (3.1) can be written as function of the eigenvalues of the Hessian. We will
246
+ then start by discretizing λ1 and λ2. Hence by a simple substitution we obtain the scheme for (3.1).
247
+ We recall that the smallest and the largest eigenvalues of a symmetric matrix can be represented
248
+ respectively by the Rayleigh-Ritz formula
249
+ (4.1)
250
+ λ1
251
+
252
+ D2u
253
+
254
+ (x) = min
255
+ θ
256
+ d2u
257
+ dν2
258
+ θ
259
+ ,
260
+ λ2
261
+
262
+ D2u
263
+
264
+ (x) = max
265
+ θ
266
+ d2u
267
+ dν2
268
+ θ
269
+ ,
270
+ where νθ = (cos θ, sin θ)is the unit vector in the direction of the angle θ.
271
+ This formula was used in [22] to build a monotone scheme in two dimension for the (MAD).
272
+ We begin by building monotone schemes for λ1 and λ2 on a wide stencil uniform grid. These
273
+ operators are used to give schemes for all formulations in this paper.
274
+ We discretize the eigenvalues of the Hessian by the following formula.
275
+ (4.2)
276
+ λh,dθ
277
+ 1
278
+
279
+ D2ug�
280
+ (x) = min
281
+ i
282
+ ug (x + vi) − 2ug (x) + ug (x − vi)
283
+ |vi|2
284
+ and
285
+ (4.3)
286
+ λh,dθ
287
+ 2
288
+
289
+ D2ug�
290
+ (x) = max
291
+ i
292
+ ug (x + vi) − 2ug (x) + ug (x − vi)
293
+ |vi|2
294
+ .
295
+ Lemma 11. The schemes (4.2) and (4.3) are degenerate elliptic.
296
+ Proof. We follow the same as in [22].
297
+ Since each discrete second derivative in the direction vi is the average of the terms which have
298
+ the form ug
299
+ j − ug
300
+ i , they are non-decreasing in ug
301
+ j − ug
302
+ i . Taking a minimum (or maximum) of non-
303
+ decreasing functions furnishes a non-decreasing function.
304
+
305
+ We finally substitute (4.2) and (4.3) in (3.1) to obtain the wide stencil finite difference scheme
306
+ of (3.1)
307
+ (4.4)
308
+
309
+ Find a positive function gi, such that
310
+ λh,dθ
311
+ 1
312
+
313
+ D2ugi�
314
+ × λh,dθ
315
+ 2
316
+
317
+ D2ugi�
318
+ = f i,
319
+
320
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
321
+ 7
322
+ with
323
+
324
+ λh,dθ
325
+ 1
326
+
327
+ D2ugi�
328
+ + λh,dθ
329
+ 2
330
+
331
+ D2ugi�
332
+ = 2
333
+
334
+ f i + gi.
335
+ ug
336
+
337
+ = ϕ.
338
+ Where f i = f (xi) and gi = g (xi) .
339
+ Lemma 12. The scheme (4.4) is degenerate elliptic.
340
+ Proof. From the properties of nondecreasing functions, obtained in [22],
341
+ that if G : R2 → R is a nondecreasing function, and if F1 and F2 are degenerate elliptic finite
342
+ difference schemes, then so is F = G (F1, F2) . It is also clear that the discretization f i = f (xi)
343
+ and gi = g (xi) does affect the ordering properties. We conclude that (4.4) is degenerate elliptic.
344
+
345
+ In the following, for simplicity, we omit the index i when there is no ambiguity.
346
+ Definition 13. We say the scheme Hh,dθ is consistent with the equation (MAD) at x0 if for every
347
+ twice continuously differentiable function ϕ (x) defined in a neighborhood of x0, Hh,dθ(ϕ) (x0) →
348
+ H (ϕ) (x0) as h, dθ → 0. The global scheme defined on Ω is consistent if the limit above holds
349
+ uniformly for all x ∈ Ω. (The domain is assumed to be closed and bounded).
350
+ Lemma 14. The consistency holds for (4.2) and (4.3) and so for (4.4).
351
+ Proof. Let x0 be a reference point with neighbors x1, ..., xN, and direction vectors vi = xi − x0, for
352
+ i = 1, ..., N, arranged symmetrically, if vi is a direction vector, then so is −vi. By Taylor series one
353
+ has
354
+ ug (x0 + vi) − 2ug (x0) + ug (x0 − vi)
355
+ |vi|2
356
+ = d2ug
357
+ dv2
358
+ i
359
+ + O
360
+
361
+ h2
362
+ i
363
+
364
+ .
365
+ Let M given symetric 2 × 2 matrix, that we can take it diagonal. Set vθ a unit vector. It follows
366
+ from [22] (Lemma 3) that
367
+ min
368
+ θ∈{θ1,...,θN} vT
369
+ θ Mvθ = λ1 + (λ2 − λ1) O
370
+
371
+ θ2�
372
+ .
373
+ Which implies that
374
+ λ1 (ϕ) (x0) − λh,dθ
375
+ 1
376
+ (ϕ) (x0) = O
377
+ �¯h2 + (λ2 − λ1) dθ2�
378
+ and thus consistency holds for (4.2).
379
+ Similar argument gives consistency for (4.3) and so for
380
+ (4.4).
381
+
382
+ Theorem 15. Suppose that unique viscosity solutions exist for the equation (3.1) Then the finite
383
+ difference scheme given by (4.4) converges uniformly on compacts subsets of Ω to the unique
384
+ viscosity solution of the equation.
385
+ Proof. We need to verify consistency and monotonicity. Consistency follows from Lemma 14 and
386
+ monotonicity follows from Lemma 12.
387
+
388
+
389
+ 8
390
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
391
+ Finally, the scheme yields a fully nonlinear equation defined on grid functions. We perform the
392
+ iteration (2.1) and by Theorem 7 will converge to a fixed point which is a solution of the equation.
393
+ This approach is used in [22].
394
+ 5. TWO METHODS OF FIXED POINT
395
+ 5.1. The first method (Method B). Notice that from Lemma 9 if u is a solution of (1.1)
396
+ u (x, y) = u˜g (x, y) it follows that det (D2u) = det
397
+
398
+ D2u˜g�
399
+ , where u˜g is the solution of (1.2) for
400
+ ˜g ∈ L2.
401
+ By writing
402
+ △u˜g = 2
403
+
404
+ f + ˜g =
405
+
406
+ (∆u˜g)2 + 2 (f − det (D2u˜g))
407
+ and expanding
408
+
409
+ ∆u˜g�2 =
410
+
411
+ u˜g
412
+ xx
413
+ �2 +
414
+
415
+ u˜g
416
+ yy
417
+ �2 + 2u˜g
418
+ xxu˜g
419
+ yy we have
420
+ △u˜g =
421
+ ��
422
+ u˜g
423
+ xx
424
+ �2
425
+ +
426
+
427
+ u˜g
428
+ yy
429
+ �2
430
+ + 2
431
+
432
+ u˜g
433
+ xy
434
+ �2
435
+ + 2f = 2
436
+
437
+ f + ˜g
438
+ Let us define the operator Q : L2 (Ω) → L2 (Ω) for Ω ⊂ R2 by
439
+ Q (g) :=
440
+
441
+ (ug
442
+ xx)2 + (ug
443
+ yy)2 + 2 (ug
444
+ xy)2 + 2f − 2
445
+
446
+ f,
447
+ with ug solution of (Pg) . So, one has
448
+ Lemma 16. ˜g is a fixed point of Q.
449
+ Proof. It follows from above expansions.
450
+
451
+ 5.1.1. The scheme. We consider the following scheme
452
+ gn+1 = Q (gn) =
453
+
454
+ (ugn
455
+ xx)2 + (ugn
456
+ yy)2 + 2 (ugn
457
+ xy)2 + 2f − 2
458
+
459
+ f.
460
+ With initial value g0 > 0 close to zero and ug0 is the solution of (P g0) .
461
+ Remark 17. The advantage of this method by comparing it to that in [19] and [2] is that it
462
+ guarantees, at least, at each iteration that tr (D2ugn (x)) > 0, which is necessary to check the
463
+ convexity.
464
+ Although this method turns out to be simple to implement is well suited in the case where ug
465
+ is in H2 (Ω) . If not, the method may not converge.
466
+ 5.1.2. Algorithm.
467
+ • g0 ≥ 0 (close to 0), solve (P g0) , ((ug)0 = ug0being known),
468
+ • For n ≥ 0, compute gn+1 and (ug)n+1 as follows
469
+ gn+1 = Q (gn) ,
470
+ (ug)n+1 solution of
471
+
472
+ P gn+1�
473
+ .
474
+
475
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
476
+ 9
477
+ Where, the method involves simply discretising the second derivatives using standard central dif-
478
+ ferences on a uniform Cartesian grid, as a result
479
+ D2
480
+ xxuij
481
+ =
482
+ 1
483
+ h2 (ui+1,,j + ui−1,j−, 2ui,,j) ,
484
+ D2
485
+ yyuij
486
+ =
487
+ 1
488
+ h2 (ui,,j+1 + ui,,j−1 − 2ui,,j) ,
489
+ D2
490
+ xxuij
491
+ =
492
+ 1
493
+ 4h2 (ui+1,,j+1 + ui−1,,j−1 − ui−1,,j+1 − ui+1,,j−1) .
494
+ 5.2. The second method (Method C). In the same setting we define the next operator.
495
+ Definition 18. Let Ω a bounded domain in R2. Define the operator F : L2 (Ω) → L2 (Ω) ,
496
+ by
497
+ (5.1)
498
+ F (g) =
499
+
500
+ |det [D2ug] − f| + g,
501
+ where ug is a solution of
502
+ (5.2)
503
+ (Pg)
504
+
505
+ ∆u = 2√f + g,
506
+ u|Γ = ϕ.
507
+ For g ∈ L2 (Ω), the operator F is well defined and it is easy to verify that
508
+ Lemma 19. �g is a fixed point of the operator F.
509
+ Proof. Let u a smooth solution of (1.1). It follows from Lemma 9 that u = u�g. Which implies that
510
+ det
511
+
512
+ D2u�g�
513
+ = det [D2u] = f and therefore, F (�g) = �g.
514
+
515
+ 5.3. The scheme. We define the following scheme
516
+ gn+1 = F (gn) =
517
+
518
+ |det [D2ugn] − f| + gn.
519
+ With initial value g0 > 0 close to zero and ug0 is the solution of (P g0) .
520
+ Remark 20. The method is advantageous, it simply involves evaluating derivatives and solving the
521
+ Poisson equation that preserves the convexity constraint.
522
+ 5.3.1. Algorithm.
523
+ • g0 ≥ 0 (close to 0), solve (P g0) , ((ug)0 = ug0being known),
524
+ • For n ≥ 0, compute gn+1 and (ug)n+1 as follows
525
+ gn+1 = α
526
+
527
+ |det [D2ugn] − f| + gn,
528
+ with 0 < α < 1.
529
+ (ug)n+1 solution of
530
+
531
+ P gn+1�
532
+ .
533
+ As in the above method, second derivatives are descretized using standard central differences on a
534
+ uniform Cartesian grid.
535
+
536
+ 10
537
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
538
+ N
539
+ Results in [19]
540
+ Method A
541
+ Method B
542
+ Method C
543
+ 31
544
+ 2.44 × 10−4
545
+ 2.965 × 10−4
546
+ 4.226 × 10−4
547
+ 18 × 10−4
548
+ 45
549
+ 1.52 × 10−4
550
+ 3.052 × 10−4
551
+ 2.202 × 10−4
552
+ 18 × 10−4
553
+ 63
554
+ 9.06 × 10−5
555
+ 2.801 × 10−4
556
+ 1.190 × 10−4
557
+ 17 × 10−4
558
+ 89
559
+ 5.32 × 10−5
560
+ 8.035 × 10−4
561
+ 6.494 × 10−5
562
+ 17 × 10−4
563
+ 127
564
+ 3.06 × 10−5
565
+ 2.015 × 10−4
566
+ 3.888 × 10−5
567
+ 17 × 10−4
568
+ Table 1. Errors
569
+ ��u − uN��
570
+ ∞ for the exact solution of the first example on an N ×N
571
+ grid. We include results from the wide stencil methods of [19] on seventeen point
572
+ stencils.
573
+ −1
574
+ −0.5
575
+ 0
576
+ 0.5
577
+ 1
578
+ −1
579
+ −0.5
580
+ 0
581
+ 0.5
582
+ 1
583
+ 1
584
+ 1.5
585
+ 2
586
+ 2.5
587
+ 3
588
+ 30
589
+ 40
590
+ 50
591
+ 60
592
+ 70
593
+ 80
594
+ 90
595
+ 100
596
+ 110
597
+ 120
598
+ 130
599
+ 0
600
+ 20
601
+ 40
602
+ 60
603
+ 80
604
+ 100
605
+ 120
606
+ 140
607
+ 160
608
+ 180
609
+ N
610
+ CPU Time
611
+
612
+
613
+ Method A
614
+ Method B
615
+ Method C
616
+ Figure 6.1. Results for example 1 on an N × N grid and total CPU time versus
617
+ N for the methods A, B and C.
618
+ 6. Numerical experiments
619
+ The three methods are tested on three different examples (smooth or singular solutions). The
620
+ discretization is done in the wide stencil Finite Difference method with 17- points (see Figure 2.2).
621
+ The number of noeuds meshing is equal to N ∗ N with N = 31, 45, 63, 89, 127, the step of meshing
622
+ h = L/N, with L is the length of the side of the rectangular domain Ω. The results obtained are
623
+ compared with those in [19].
624
+ In the first example we study the regular solution given by :
625
+ u (x, y) = exp
626
+ �(x2 + y2)
627
+ 2
628
+
629
+ with f (x, y) =
630
+
631
+ x2 + y2 + 1
632
+
633
+ exp
634
+
635
+ x2 + y2�
636
+ .
637
+ The Table 1 summarizes the obtained results for different meshing.
638
+ In Figure 6 we show the surface plot of the solution and the total CPU time versus N for the
639
+ methods A, B and C.
640
+
641
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
642
+ 11
643
+ N
644
+ Results in [19]
645
+ Method A
646
+ Method B
647
+ Method C
648
+ 31
649
+ 1.22 × 10−3
650
+ 5.806 × 10−4
651
+ 6.853 × 10−4
652
+ 8.794 × 10−4
653
+ 45
654
+ 5.9 × 10−4
655
+ 4.92 × 10−4
656
+ 6.719 × 10−4
657
+ 8.727 × 10−4
658
+ 63
659
+ 4.2 × 10−4
660
+ 4.914 × 10−4
661
+ 2.733 × 10−4
662
+ 8.601 × 10−4
663
+ 89
664
+ 2.6 × 10−4
665
+ 4.085 × 10−4
666
+ 2.09 × 10−5
667
+ 8.173 × 10−4
668
+ 127
669
+ 2.0 × 10−4
670
+ 4.056 × 10−4
671
+ 1.08 × 10−5
672
+ 8.164 × 10−4
673
+ Table 2. Errors
674
+ ��u − uN��
675
+ ∞ for the exact solution of the second example on an
676
+ N × N grid. We include results from the wide stencil methods of [19] on seventeen
677
+ point stencils.
678
+ 0
679
+ 0.2
680
+ 0.4
681
+ 0.6
682
+ 0.8
683
+ 1
684
+ 0
685
+ 0.5
686
+ 1
687
+ 0
688
+ 0.05
689
+ 0.1
690
+ 0.15
691
+ 0.2
692
+ 30
693
+ 40
694
+ 50
695
+ 60
696
+ 70
697
+ 80
698
+ 90
699
+ 100
700
+ 110
701
+ 120
702
+ 130
703
+ 0
704
+ 50
705
+ 100
706
+ 150
707
+ 200
708
+ 250
709
+ 300
710
+ N
711
+ CPU Time
712
+
713
+
714
+ Method A
715
+ Method B
716
+ Method C
717
+ Figure 6.2. Results for example 2 on an N × N grid and total CPU time versus
718
+ N for the methods A, B and C.
719
+ As a second example, which is C1 , we take the one considered in [19] which is given by
720
+ u(x, y) = 1
721
+ 2((
722
+
723
+ (x − 0.5)2 + (y − 0.5)2 − 0.2)+)2 with f(x, y) = (1 −
724
+ 0.2
725
+
726
+ (x − 0.5)2 + (y − 0.5)2)+.
727
+ The results are in Table 2.
728
+ Finally, we consider a third example which is singular at the bord of the domain Ω = [0, 1] ×
729
+ [0.1] ,defined by
730
+ u(x, y) = −
731
+
732
+ (2 − x2 − y2) where f(x, y) =
733
+ 2
734
+ (2 − x2 − y2)2.
735
+ .
736
+ The results are illustrated in Table 3 and
737
+
738
+ 12
739
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
740
+ N
741
+ Results in [19]
742
+ Method A
743
+ Method B
744
+ Method C
745
+ 31
746
+ 1.74 × 10−3
747
+ 1.7 × 10−3
748
+ 5.1 × 10−3
749
+ 5.7 × 10−3
750
+ 45
751
+ 9.8 × 10−4
752
+ 1.5 × 10−3
753
+ 4.8 × 10−3
754
+ 5.5 × 10−3
755
+ 63
756
+ 5.9 × 10−4
757
+ 8.9 × 10−4
758
+ 3.9 × 10−3
759
+ 5.5 × 10−3
760
+ 89
761
+ 3.5 × 10−4
762
+ 8.9 × 10−4
763
+ 3.1 × 10−3
764
+ 5.5 × 10−3
765
+ 127
766
+ 2.0 × 10−4
767
+ 8.2 × 10−4
768
+ 2.4 × 10−3
769
+ 5.5 × 10−3
770
+ Table 3. Errors
771
+ ��u − uN��
772
+ ∞ for the exact solution of the third example on an N ×N
773
+ grid. We include results from the wide stencil methods of [19] on seventeen point
774
+ stencils.
775
+ 0
776
+ 0.2
777
+ 0.4
778
+ 0.6
779
+ 0.8
780
+ 1
781
+ 0
782
+ 0.5
783
+ 1
784
+ −1.5
785
+ −1
786
+ −0.5
787
+ 0
788
+ 30
789
+ 40
790
+ 50
791
+ 60
792
+ 70
793
+ 80
794
+ 90
795
+ 100
796
+ 110
797
+ 120
798
+ 130
799
+ 0
800
+ 200
801
+ 400
802
+ 600
803
+ 800
804
+ 1000
805
+ 1200
806
+ 1400
807
+ 1600
808
+ 1800
809
+ N
810
+ CPU Time
811
+
812
+
813
+ Method A
814
+ Method B
815
+ Method C
816
+ Figure 6.3. Results for example 2 on an N × N grid and total CPU time versus
817
+ N for the methods A, B and C.
818
+ Acknowledgments
819
+ We are indebted to Pr. Pierre-Emmanuelle Jabin for his relevant remarks and his impressive
820
+ comments which have greatly improved this work.
821
+ References
822
+ [1] Guy Barles and Panagiotis E. Souganidis. Convergence of approximation schemes for fully nonlinear second
823
+ order equations. Asymptotic Anal., 4(3):271–283, 1991.
824
+ [2] Jean-David Benamou, Brittany D. Froese, and Adam M. Oberman. Two numerical methods for the elliptic
825
+ Monge-Ampère equation. ESAIM: Math. Model. Numer. Anal., 44(4), 2010.
826
+ [3] Fethi Ben Belgacem, Optimization approach for the Monge-Ampère equation, Acta Mathematica Scientia, Vol.
827
+ 38, Issu 4 (2018), 1285-1295.
828
+ [4] C. J. Budd and J. F. Williams. Moving mesh generation using the parabolic Monge-Ampère equation. SIAM
829
+ J. Sci. Comput., 31(5):3438–3465, 2009.
830
+ [5] Eugenio Calabi, Complete affine hyperspheres. I, Symposia Mathematica, Vol. X (Convegno di Geometria
831
+ Differenziale, INDAM, Rome, 1971), Academic Press, London, 1972, pp. 19–38. MR0365607 (51 #1859)
832
+
833
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
834
+ 13
835
+ [6] Shiu Yuen Cheng and Shing-Tung Yau, Complete affine hypersurfaces. I. The com- pleteness of affine metrics,
836
+ Comm. Pure Appl. Math. 39 (1986), no. 6, 839–866, DOI 10.1002/cpa.3160390606. MR859275 (87k:53127)
837
+ [7] Cristian E. Gutiérrez. The Monge-Ampère equation. Progress in Nonlinear Differential Equations and their
838
+ Applications, 44. Birkhäuser Boston Inc., Boston, MA, 2001.
839
+ [8] G. L. Delzanno, L. Chacón, J. M. Finn, Y. Chung, and G. Lapenta. An optimal robust equidistribution
840
+ method for two-dimensional grid adaptation based on Monge-Kantorovich optimization. J. Comput. Phys.,
841
+ 227(23):9841–9864, 2008
842
+ [9]
843
+ J. M. Finn, G. L. Delzanno, and L. Chacón. Grid generation and adaptation by Monge- Kantorovich opti-
844
+ mization in two and three dimensions. In Proceedings of the 17th Interna- tional Meshing Roundtable, pages
845
+ 551–568, 2008
846
+ [10] E. J. Dean and R. Glowinski. An augmented Lagrangian approach to the numerical solution of the Dirichlet
847
+ problem for the elliptic Monge-Ampère equation in two dimensions. Electron. Trans. Numer. Anal., 22:71–96
848
+ (electronic), 2006.
849
+ [11] E. J. Dean and Roland Glowinski. On the numerical solution of the elliptic Monge- Ampère equation in
850
+ dimension two: a least-squares approach. In Partial differential equations, volume 16 of Comput. Methods
851
+ Appl. Sci., pages 43–63. Springer, Dordrecht, 2008.
852
+ [12] G. De Philippis and A. Figalli, The Monge-Ampère equation and its link to optimal trans- portation, Bull.
853
+ Amer. Math. Soc. (N.S.) 51 (2014), no. 4, 527–580. MR3237759
854
+ [13] T. Glimm and V. Oliker. Optical design of single reflector systems and the Monge-Kantorovich mass transfer
855
+ problem. J. Math. Sci. (N. Y.), 117(3):4096–4108, 2003. Nonlinear problems and function theory.
856
+ [14] Grégoire Loeper and Francesca Rapetti. Numerical solution of the Monge-Ampére equation by a Newton’s
857
+ algorithm. C. R. Math. Acad. Sci. Paris, 340(4):319–324, 2005.
858
+ [15] T. ur Rehman, E. Haber, G. Pryor, J. Melonakos, and A. Tannenbaum. 3D nonrigid regis- tration via optimal
859
+ mass transport on the GPU. Med Image Anal, 13(6):931–40, 12 2009.
860
+ [16]
861
+ Steven Haker, Allen Tannenbaum, and Ron Kikinis. Mass preserving mappings and image registration. In
862
+ MICCAI ’01: Proceedings of the 4th International Conference on Medical Image Computing and Computer-
863
+ Assisted Intervention, pages 120–127, London, UK, 2001. Springer-Verlag
864
+ [17] . Steven Haker, Lei Zhu, Allen Tannenbaum, and Sigurd Angenent. Optimal mass transport for registration
865
+ and warping. Int. J. Comput. Vision, 60(3):225–240, 2004.
866
+ [18] X. Feng and Michael Neilan. Mixed finite element methods for the fully nonlinear Monge-Ampère equation
867
+ based on the vanishing moment method. SIAM J. Numer. Anal., 47(2):1226–1250, 2009.
868
+ [19] B. D. Froese and A. M. Oberman, Convergent finite difference solvers for viscosity solutions of the elliptic
869
+ Monge-Ampère equation in dimensions two and higher, SIAM J. Numer. Anal. 49 (2011), no. 4, 1692–1714.
870
+ MR2831067
871
+ [20] Adam M. Oberman. Convergent difference schemes for degenerate elliptic and parabolic equations: Hamilton-
872
+ Jacobi equations and free boundary problems. SIAM J. Numer. Anal., 44(2):879–895 (electronic), 2006.
873
+ [21] Adam M. Oberman. Computing the convex envelope using a nonlinear partial differential equation. Math.
874
+ Models Methods Appl. Sci., 18(5):759–780, 2008.
875
+ [22] Adam M. Oberman. Wide stencil finite difference schemes for the elliptic Monge-Ampère equation and functions
876
+ of the eigenvalues of the Hessian. Discrete Contin. Dyn. Syst. Ser. B, 10(1):221–238, 2008
877
+ [23] Adam M. Oberman and Luis Silvestre. The Dirichlet problem for the convex envelope. Trans. Amer. Math.
878
+ Soc. (to appear), 2010 http://arxiv.org/abs/1007.0773
879
+ [24] V. I. Oliker and L. D. Prussner. On the numerical solution of the equation (∂ 2 z/∂x 2 )(∂ 2 z/∂y 2 ) − (∂ 2
880
+ z/∂x∂y) 2 = f and its discretizations, I. Numer. Math., 54(3):271– 293, 1988.
881
+ [25] A. V. Pogorelov, On the improper convex affine hyperspheres, Geometriae Dedicata 1 (1972), no. 1, 33–46.
882
+ MR0319126 (47 #7672)
883
+
884
+ 14
885
+ CONVERGENT APPROACHES FOR THE DIRICHLET MONGE-AMPÈRE PROBLEM
886
+ [26] Siltakoski, J. Equivalence of viscosity and weak solutions for the normalized p(x)-Laplacian. Calc. Var. 57, 95
887
+ (2018). https://doi.org/10.1007/s00526-018-1375-1
888
+ [27] Neil S. Trudinger and Xu-Jia Wang, The Bernstein problem for affine maximal hypersur- faces, Invent. Math.
889
+ 140 (2000), no. 2, 399–422, DOI 10.1007/s002220000059. MR1757001 (2001h:53016)
890
+ [28] Neil S. Trudinger and Xu-Jia Wang, Affine complete locally convex hypersurfaces, Invent. Math. 150 (2002),
891
+ no. 1, 45–60, DOI 10.1007/s00222-002-0229-8. MR1930881 (2003h:53012)
892
+ [29] Neil S. Trudinger and Xu-Jia Wang, The affine Plateau problem, J. Amer. Math. Soc. 18 (2005), no. 2, 253–289,
893
+ DOI 10.1090/S0894-0347-05-00475-3. MR2137978 (2006e:53071)
894
+ [30] V. Zheligovsky, O. Podvigina, and U. Frisch. The Monge-Ampère equation: Various forms and numerical
895
+ solution. J. Comput. Phys., 229(13):5043–5061, 2010.
896
+ .
897
+
898
+ 40
899
+ 50
900
+ 60
901
+ 70
902
+ 80
903
+ 90
904
+ 100
905
+ 110
906
+ N
907
+
908
+ 40
909
+ 50
910
+ 60
911
+ 70
912
+ 80
913
+ 90
914
+ 100
915
+ 110
916
+ N
917
+
918
+ 40
919
+ 50
920
+ 60
921
+ 70
922
+ 80
923
+ 90
924
+ 100
925
+ 110
926
+ N
927
+
928
+ 40
929
+ 50
930
+ 60
931
+ 70
932
+ 80
933
+ 90
934
+ 100
935
+ 110
936
+ N
937
+
938
+ 10
939
+ 20
940
+ 30
941
+ 40
942
+ 50
943
+ 60
944
+ 70
945
+
946
+ 40
947
+ 50
948
+ 60
949
+ 70
950
+ 80
951
+ 90
952
+ 100
953
+ 110
954
+ N
955
+
956
+ 40
957
+ 50
958
+ 60
959
+ 70
960
+ 80
961
+ 90
962
+ 100
963
+ 110
964
+ N
965
+
966
+ 20
967
+ 40
968
+ 60
969
+ 80
970
+ 100
971
+
972
+ 20
973
+ 40
974
+ 60
975
+ 80
976
+ 100
977
+
978
+ 40
979
+ 50
980
+ 60
981
+ 70
982
+ 80
983
+ 90
984
+ 100
985
+ 110
986
+ N
987
+
8NFAT4oBgHgl3EQfox1h/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
9NFJT4oBgHgl3EQfoSxe/content/tmp_files/2301.11595v1.pdf.txt ADDED
@@ -0,0 +1,1617 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.11595v1 [math-ph] 27 Jan 2023
2
+ Exact solutions of Maxwell equations in homogeneous
3
+ spaces with group of motions G3(IX)
4
+ V. V. Obukhov
5
+ Institute of Scietific Research and Development, Tomsk State Pedagogical University
6
+ (TSPU). Tomsk State Pedagogical University, 60 Kievskaya St., Tomsk, 634041, Russia;
7
+ Laboratory for Theoretical Cosmology, International Center of Gravity and Cosmos, Tomsk
8
+ State University of Control Systems and Radio Electronics (TUSUR), 36, Lenin Avenue, Tomsk,
9
+ 634050, Russia
10
+ Keywords: Maxwell equations, Klein-Gordon-Fock equation, algebra of symmetry operators,
11
+ theory of symmetry, linear partial differential equations.
12
+ Exact solutions of Maxwell equations in homogeneous spaces with group of motions G3(IX)
13
+ 1
14
+ Introduction
15
+ All known methods of integration of main differential equations of mathematical physics are
16
+ based on complete reduction of these equations to a system of ordinary differential equations.
17
+ Reduction is carried out using symmetry operators. For the equations of motion of classical
18
+ or quantum sample particle in external electromagnetic and gravitational fields the symmetry
19
+ operators are integrals of motion. It is known that a necessary condition for the existence of
20
+ integrals of motion is the existence of spacetime symmetry given by the Killing fields.
21
+ Thus the problem of exact integration is closely related to the study of space-time symmetry.
22
+ At present two methods of exact integration of equations of motion are known. These are
23
+ methods of commutative (CIM) and noncommutative (NCIM) integration. The first method is
24
+ based on the theory of complete separation of variables, and it is applicable in stackel spaces.
25
+ Stackel spaces admit complete sets consisting of mutually commuting Killing fields. Theory
26
+ of Stackel spaces was developed in [1], [2], [3], [4], [5], [6],[7].
27
+ A description of the theory
28
+ and a detailed bibliography can be found in [8], [9] [10], [13] (see also [12]). Solutions of field
29
+ equations, which are still used widely in the theory of gravitation, have been constructed on
30
+ the basis of Stackel spaces. These solutions are often used in the study of various effects in
31
+ gravitational fields (see, for example,[14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25],
32
+ [26]).
33
+ The second method (NCI method) is based on the use of noncommutative algebras of
34
+ symmetry operators linear in moments and constructed using vector Killing fields. The method
35
+ was proposed in [27]. The development of the method and its application to gravity theory can
36
+ be found in [28], [29], [30], [31].
37
+ 1
38
+
39
+ As in stackel spaces in the spaces with a noncommutative group of motions the equations
40
+ of motion of a test particle admit the complete reduction to a system of ordinary differential
41
+ equations.
42
+ Therefore, we will call space-time manifolds admitting noncommutative groups
43
+ Gr, r ≥ 3 as post-Stackel spaces (PSS).
44
+ By analogy with stackel spaces we will call the PSS non-isotropic if a group Gr (or its
45
+ subgroup of rank 3) acts transitive on a non-isotropic hypersurface of spacetime, or isotropic,
46
+ if the hypersurface is isotropic. For non-isotropic post-stackel spaces we will also use the term
47
+ ”homogeneous post-stackel space (HPSS)”.
48
+ The same classification problems can be considered for the PSS as for the stackel spaces.
49
+ For example, in the papers [9] [10], [11] a complete classification is given for the case when
50
+ the Hamilton-Jacobi equation for a charged test particle admits the complete separation of
51
+ variables in the external electromagnetic field. A similar classification problem has been solved
52
+ for PSS-spaces as well. In [32] PSS-spaces with transitive four-parameter groups of motions
53
+ are considered; in [33] HPSS-spaces are considered (see also [34]); in [35] PSS-spaces with
54
+ groups acting on isotropic hypersurfaces of transitivity are considered. PSS-spaces with four-
55
+ parameter groups of motions are considered in [36], provided that these groups have transitive
56
+ three-parameter subgroups. Thus, one has found the potentials of all admissible electromag-
57
+ netic fields, for which the Hamilton-Jacobi and Klein-Gordon-Fock equations have algebras of
58
+ symmetry operators given by groups of motions of post-stackel spaces. It was proved, that
59
+ the Klein-Gordon-Fock equation admits the algebra of symmetry operators given by groups of
60
+ motions of PSS if and only if the Hamilton-Jacobi equations admits the appropriate algebra of
61
+ integrals of motion.
62
+ Next classification problem is the classification of electrovacuum solutions of the Einstein-
63
+ Maxwell equations for the case, when CIM and NCIM methods are applicable. During the
64
+ century-long history of General relativity, many exact solutions of the vacuum and electrovac-
65
+ uum Einstein equations have been found (see, for example,[42] ). Nevertheless, this problem has
66
+ not lost its relevance up to now. The main purpose of the classification is not so much to find
67
+ new exact solutions, as to list all gravitational and electromagnetic fields, in which equations
68
+ of motion of test particles can be exactly integrated or at least reduced to systems of ordinary
69
+ differential equations. This problem divided into two stages.
70
+ At the first stage all non-equivalent classes of solutions of the vacuum Maxwell equations for
71
+ the potentials of admissible electromagnetic fields are found. At the second stage the obtained
72
+ classification is used to classify the corresponding electrovacuum spaces. Historically, for Stackel
73
+ spaces this problem was solved before the problem of the first stage (see the bibliography given
74
+ in [9], [10], [11]). The present article is devoted to solving the first stage of this classification
75
+ problem. All non-equivalent solutions of empty Maxwell equations in homogeneous spaces of
76
+ type IX according to Bianchi’s classification are found.
77
+ 2
78
+ Admissible electromagnetic fields in homogeneous spaces
79
+ There are two definitions of homogeneous spaces.
80
+ According to the first a spacetime
81
+ V4
82
+ is homogeneous if its subspace
83
+ V3,
84
+ endowed with the Euclidean space signature, admits
85
+ coordinate transformations (forming the group G3(N) of motions of spaces V4), that allow to
86
+ connect any two points in
87
+ V3
88
+ (see [38]). This definition directly implies that metric of the
89
+ 2
90
+
91
+ V4 in the semi-geodesic coordinate system
92
+ [ui]
93
+ can be represented as follows:
94
+ ds2 = −du02 + ηabla
95
+ αlb
96
+ βduαduβ,
97
+ gij = −δ0
98
+ i δ0
99
+ j + δa
100
+ i δb
101
+ jea
102
+ αeb
103
+ βηab(u0),
104
+ det|ηab| > 0
105
+ ea
106
+ α,0 = 0.
107
+ (1)
108
+ The coordinate indices of the variables of the semi-geodesic coordinate system are denoted
109
+ by lower case Latin letters:
110
+ i, j, k = 0, 1 . . . 3.
111
+ The coordinate indices of the variables of the
112
+ local coordinate system on the hypersurface
113
+ V3
114
+ are denoted by lower case Greek letters:
115
+ α, β, γ = 1, . . . 3.
116
+ the time variable is denoted by a 0 index. Group indices and indices of
117
+ nonholonomic frame are denoted by
118
+ a, b, c = 1, . . . 3.
119
+ Summation is performed over repeated
120
+ upper and lower indices within the index range.
121
+ The 1-form
122
+ ea
123
+ αduα
124
+ is invariant under the acting of the group G3(N). The vectors of the
125
+ frame ea
126
+ α define a non-holonomic coordinate system in V3. The dual triplet of vectors
127
+
128
+ a,
129
+
130
+ aeb
131
+ α = δb
132
+ a,
133
+
134
+ aea
135
+ β = δα
136
+ β
137
+ defines the operators of the group algebra:
138
+ ˆYa = eα
139
+ a∂a,
140
+ [ ˆYa, ˆYb] = Cc
141
+ ab ˆYc.
142
+ (2)
143
+ According to another definition, space-time
144
+ V4
145
+ is homogeneous if it admits a three-parameter
146
+ group of motions
147
+ G3(N),
148
+ whose hypersurface
149
+ V3
150
+ of transitivity has the Euclidean space
151
+ signature. The Killing vector fields ξα
152
+ a
153
+ and their dual vector fields
154
+ ξa
155
+ α
156
+ form another frame
157
+ of the space V3 and another representation of the algebra of the group G3. The vector fields
158
+ ξα
159
+ a
160
+ satisfy the Killing equations:
161
+ gαβ
162
+ ,γ ξγ
163
+ a = gαγξβ
164
+ a,γ + gβγξα
165
+ a,γ,
166
+ (3)
167
+ and sets the infinitesimal group operators of the algebra G3
168
+ ˆXa = ξα
169
+ a ∂α,
170
+ [ ˆXa, ˆXb] = Cc
171
+ ab ˆXc.
172
+ (4)
173
+ Let us consider electromagnetic field with potential Ai. For a charged test particle, moving in
174
+ this external electromagnetic field, it has been proved, that the Hamilton-Jacobi equation:
175
+ gijPiPj = m,
176
+ Pi = pi + Ai.
177
+ (5)
178
+ and the Klein-Gordon-Fock equation:
179
+ ˆHϕ = (gij ˆPi ˆPj)ϕ = m2ϕ,
180
+ ˆPk = ˆpk + Ak.
181
+ (6)
182
+ admit the integrals of motion, which are given by Killing vectors:
183
+ ˜Xα = ξi
184
+ αpi
185
+ (or
186
+ ˆ˜Xα = ξi
187
+ αˆpi),
188
+ if and only if the conditions:
189
+ ξα
190
+ a ( ˜A),α = Cc
191
+ ab ˜A
192
+ (7)
193
+ are satisfied (see papers [33]). Here
194
+ pi = ∂iϕ;
195
+ ˆpk = −ı ˆ∇k;
196
+ ( ˆ∇k is the covariant deriva-
197
+ tive operator corresponding to the partial derivative operator -
198
+ ˆ∂i
199
+ in the coordinate field
200
+ ui),
201
+ ϕ
202
+ is a scalar function of particle with mass
203
+ m;
204
+ ˜Aa = ξα
205
+ a Aα.
206
+ 3
207
+
208
+ The electromagnetic field whose potential satisfies condition (7) is called admissible. All
209
+ admissible electromagnetic fields for groups of motion
210
+ Gr(N)
211
+ (r ≤ 4),
212
+ acting transitively
213
+ on hypersurfaces of the spacetime, have been found in [33], [35], [36].
214
+ Solutions of the set of equations (7) for HPSS of type IX have the form:
215
+ Aα = αa(u0)la
216
+ α ⇒ Aa = lα
217
+ aAα = αa(u0).
218
+ (8)
219
+ To prove this let’s find the frame vector. We will use the metric tensor of IX-type space by
220
+ Bianchi, found in Petrov’s book [39]. As it is known, the Bianchi type IX metric contains as
221
+ a special case the space of constant positive curvature and therefore is of special interest for
222
+ cosmology.
223
+ ds2 = du12[a11 − (a12 cos 2u3 + a22 sin 2u3)] + 2du1du3((a13 cos u3 − a23 sin u3)+
224
+ (9)
225
+ +2du1du2[(a13 cos u3 − a23 sin u3) cos u1 + (a12 cos 2u3 − a22 sin 2u3) sin u1]
226
+ +du22[a33cos u12 + (a23 cos u3 + a13 sin u3) sin 2u1 + (a12 sin 2u3 + a22 cos 2u3 + a11)sin u12]
227
+ 2du2du3(a33 cos u1 + (a23 cos u3 + a13 sin u3) sin u1) + du32a33 + edu02.
228
+ aab are arbitrary functions on u0.
229
+ To obtain the functions
230
+
231
+ a
232
+ , it is sufficient to consider the components
233
+ g13, g23
234
+ from the system (3). The solution has the form:
235
+ la
236
+ α = δp
237
+ αla
238
+ p(u1, u3) + δ3
239
+ αδa
240
+ 3
241
+ .
242
+ From the equations:
243
+ g13 = a13 cos u3 − a23 sin u3 = η3ala
244
+ 1,
245
+ g23 = a33 cos u1 + (a23 cos u3 + a13 sin u3) sin u1 = η3ala
246
+ 1
247
+ it follows:
248
+ la
249
+ α =
250
+
251
+
252
+ cos u3
253
+ − sin u3
254
+ 0
255
+ sin u1 sin u3
256
+ sin u1 cos u3
257
+ cos u1
258
+ 0
259
+ 0
260
+ 1
261
+
262
+ � , lα
263
+ a =
264
+
265
+
266
+ cos u3
267
+ sin u3
268
+ sin u1
269
+ −cos u1 sin u3
270
+ sin u1
271
+ − sin u3
272
+ cos u3
273
+ sin u1
274
+ −cos u1 cos u3
275
+ sin u1
276
+ 0
277
+ 0
278
+ 1
279
+
280
+ � ,
281
+ (10)
282
+ la
283
+ αlα
284
+ b = δa
285
+ b .
286
+ The lower index numbers the lines. One can show that the vector fields (10) satisfy the
287
+ equations (1), (2): We present the components of the vectors ξα
288
+ a in the form of a matrix:
289
+ ||ξα
290
+ a || =
291
+
292
+
293
+ 0
294
+ 1
295
+ 0
296
+ cos u2
297
+ −cos u1 sin u2
298
+ sin u1
299
+ sin u2
300
+ sin u1
301
+ − sin u2
302
+ −cos u1 cos u2
303
+ sin u1
304
+ cos u2
305
+ sin u1
306
+
307
+
308
+ The components ˜Aα can be expressed through Aα as follows:
309
+ ˜Aa = Zb
310
+ aAb,
311
+ 4
312
+
313
+ where
314
+ ||Zb
315
+ a = ξα
316
+ a lb
317
+ α|| =
318
+
319
+
320
+ sin u1 sin u3
321
+ sin u1 cos u3
322
+ cos u1
323
+ (cos u2 cos u3 − sin u2 sin u3 cos u1)
324
+ −(cos u2 sin u3 + sin u2 cos u3 cos u1)
325
+ sin u1 sin u2
326
+ −(sin u2 cos u3 + cos u2 sin u3 cos u1)
327
+ (sin u2 sin u3 − cos u2 cos u3 cos u1)
328
+ cos u2 sin u1
329
+
330
+ � .
331
+ It can be shown by direct calculation that the elements of the matrix Zb
332
+ a satisfy the equation:
333
+ Zb
334
+ a|c = Ca1
335
+ caZb
336
+ a1,
337
+ |a = lα
338
+ a∂α.
339
+ (11)
340
+ Therefore, the equation (7) can be reduced to the form:
341
+ ξα
342
+ a Ab,α = 0 ⇒ Aa = αa(u0).
343
+ (12)
344
+ 3
345
+ Maxwell’s equations with zero electromagnetic field
346
+ sources in a homogeneous spacetime
347
+ All exact solutions of vacuum Maxwell equations for solvable groups have been found in the
348
+ papers [40], [41]. In the present paper the problem is solved for the group G3(IX).
349
+ We will use the first definition of homogeneous spaces. Note, that for the space-time with
350
+ the groups of motions G3(I) − G3(V I), G3(IX) both definitions are equivalent
351
+ Consider the Maxwell equations with zero electromagnetic field sources in homogeneous
352
+ space in the presence of an electromagnetic field invariant with respect to the group Gr:
353
+ 1
354
+ √−g(√−gF ij),j = 0,
355
+ (13)
356
+ The metric tensor is defined by relations (1), the electromagnetic potential by the relations (7).
357
+ When i = 0, from the set of equations (13) it follows:
358
+ 1
359
+ √−g(√−ggαβF0β)α = 1
360
+ l (llα
361
+ aηab ˙αb),α = ηabρa ˙αb = 0.
362
+ (14)
363
+ Here it is denoted g = − det ||gαβ|| = −(ηl)2,
364
+ where
365
+ η2 = det ||ηαβ||,
366
+ l = det ||la
367
+ α||,
368
+ ρa =
369
+
370
+ a,α + l|a/l,
371
+ the dots means the time derivatives. Let
372
+ i = α.
373
+ Then from the equation (13)
374
+ it follows:
375
+ 1
376
+ η(ηgαβF0β),0 = 1
377
+ l (lgνβgαγFβγ),ν ⇒ 1
378
+ η(ηηablα
379
+ a ˙αb),0 = 1
380
+ l (llν
381
+ alβ
382
+ b ηablα
383
+ ˜alγ
384
+ ˜b η˜a˜bFβγ),ν ⇒
385
+ (15)
386
+ (ηηab ˙αb),0 = ηla
387
+ α
388
+ l (llβ
389
+ b lα
390
+ ˜a1lγ
391
+ ˜b Fβγ)|a1ηa1bη˜a˜b.
392
+ (16)
393
+ Let us find components of Fαβ, using the relations (8).
394
+ Fαβ = (la
395
+ β,α − la
396
+ β,α)αa = lc
397
+ βlγ
398
+ c ld
399
+ αlν
400
+ d(la
401
+ γ,ν − la
402
+ ν,γ)αa = lb
403
+ βla
404
+ αlc
405
+ γ(lγ
406
+ a|b − lγ
407
+ b|a)αc = lb
408
+ βla
409
+ αCc
410
+ baαc.
411
+ (17)
412
+ Then
413
+ (lF αβ),β = ηabη˜a˜bCd
414
+ ˜bbαd((llα
415
+ a)|˜a + llα
416
+ alγ
417
+ ˜a,γ).
418
+ (18)
419
+ 5
420
+
421
+ Structural constants of a group
422
+ G3
423
+ can be present in the form:
424
+ Cc
425
+ ab = Cc
426
+ 12ε12
427
+ ˜a˜b + Cc
428
+ 13ε13
429
+ ˜a˜b + Cc
430
+ 23ε23
431
+ ˜a˜b,
432
+ εAB
433
+ ab = δA
434
+ a δB
435
+ b − δA
436
+ b δB
437
+ a .
438
+ (19)
439
+ Using the notations:
440
+ σ1 = Ca
441
+ 23αa,
442
+ σ2 = Ca
443
+ 31αa,
444
+ σ3 = Ca
445
+ 12αa,
446
+ γ1 = σ1η11 + σ2η12 + σ3η13,
447
+ γ2 = σ1η12 + σ2η22 + σ3η23,
448
+ γ3 = σ1η13 + σ2η23 + σ3η33,
449
+ let us reduce Maxwell’s equations (13) to the form:
450
+ η(ηab ˙αb),0 = δa
451
+ 1(γ1(C1
452
+ 32) − γ2(C1
453
+ 31 + ρ3) + γ3(C1
454
+ 21 + ρ2)) + δa
455
+ 2(γ1(C2
456
+ 32 + ρ3)+
457
+ (20)
458
+ γ2C2
459
+ 13 − γ3(C2
460
+ 12ρ1)) + δa
461
+ 3(−γ1(C3
462
+ 23 + ρ2) + γ2(C3
463
+ 13 + ρ1) + γ3C3
464
+ 21),
465
+ The order of the equations (20) can be decreased by introducing a new independent functions:
466
+ βa = βa = ηηab ˙αb
467
+
468
+ η ˙αa = ηabβb.
469
+ (21)
470
+ Let us consider the Maxwell equations for the group G3(IX). As in this case
471
+ non zero
472
+ structural constants are following:
473
+ C3
474
+ 12 = C2
475
+ 31 = C1
476
+ 23 = 1,
477
+ functions
478
+ σa, γ1
479
+ have the form:
480
+ σ1 = α1,
481
+ σ2 = α2,
482
+ σ3 = α3.
483
+ γ1 = α1η11 + α2η12 + α3η13,
484
+ γ2 = α1η12 + α2η22 + α3η23,
485
+ γ1 = α1η13 + α2η23 + α3η33.
486
+ Using these relations, we obtain Maxwell’s equations (14), (20) as a system of linear algebraic
487
+ equations on the unknown functions
488
+ nab:
489
+ nab = ηab
490
+ η ⇒ η =
491
+ 1
492
+ det nab
493
+ .
494
+ (22)
495
+ ˆW ˆn = ˆω,
496
+ (23)
497
+ where
498
+ ˆW =
499
+
500
+
501
+
502
+
503
+
504
+
505
+
506
+
507
+ α1
508
+ α2
509
+ α3
510
+ 0
511
+ 0
512
+ 0
513
+ β1
514
+ β2
515
+ β3
516
+ 0
517
+ 0
518
+ 0
519
+ 0
520
+ α1
521
+ 0
522
+ α2
523
+ α3
524
+ 0
525
+ 0
526
+ β1
527
+ 0
528
+ β2
529
+ β3
530
+ 0
531
+ 0
532
+ 0
533
+ α1
534
+ 0
535
+ α2
536
+ α3
537
+ 0
538
+ 0
539
+ β1
540
+ 0
541
+ β2
542
+ β3
543
+
544
+
545
+
546
+
547
+
548
+
549
+
550
+
551
+ ,
552
+ (24)
553
+ ˆnT = (n11, n12, n13, n22, n23, n33);
554
+ ˆωT = (− ˙β1, ˙α1, − ˙β2, ˙α2, − ˙β3, ˙α3),
555
+ 6
556
+
557
+ index T means the transposition of a matrix. Let us find the algebraic complement of the
558
+ matrix ˆW :
559
+ ˆV =
560
+
561
+
562
+
563
+
564
+
565
+
566
+
567
+
568
+ β1V 2
569
+ 1
570
+ −α1V 2
571
+ 1
572
+ β2V 2
573
+ 1
574
+ −α2V 2
575
+ 1
576
+ β3V 2
577
+ 1
578
+ −α3V 2
579
+ 1
580
+ β1V1V2
581
+ −α1V1V2
582
+ β2V1V2
583
+ −α2V1V2
584
+ β3V1V2
585
+ −α3V1V2
586
+ β1V1V3
587
+ −α1V1V3
588
+ β2V1V3
589
+ −α2V1V3
590
+ β3V1V3
591
+ −α3V1V3
592
+ β1V 2
593
+ 2
594
+ −α1V 2
595
+ 2
596
+ β2V 2
597
+ 2
598
+ −α2V 2
599
+ 2
600
+ β3V 2
601
+ 2
602
+ −α3V 2
603
+ 2
604
+ β1V2V3
605
+ −α1V2V3
606
+ β2V2V3
607
+ −α2V2V3
608
+ β3V2V3
609
+ −α3V2V3
610
+ β1V 2
611
+ 3
612
+ −α1V 2
613
+ 3
614
+ β2V 2
615
+ 3
616
+ −α2V 2
617
+ 3
618
+ β3V 2
619
+ 3
620
+ −α3V 2
621
+ 3
622
+
623
+
624
+
625
+
626
+
627
+
628
+
629
+
630
+ (25)
631
+ As ˆW is singular matrix, ˆV is the annulling matrix for ˆW:
632
+ ˆV ˆW = 0.
633
+ (26)
634
+ Therefore, one of the equations from the system (23) can be replaced by the equation:
635
+ δab( ˙αa ˙αb + ˙βa ˙βb) ⇒ δab(αaαb + βaβb) = c2 = const.
636
+ (27)
637
+ Depending on the rank of the matrix ˆW, one or more functions nab are independent. The
638
+ remaining functions nab can be expressed through them and through the functions αa, βa. For
639
+ classification it is necessary to find non-equivalent solutions of the system (23). Obviously, this
640
+ system are symmetric with respect to the transposition
641
+
642
+ 1 ↔ lα
643
+ 2 . Therefore the reference
644
+ indices a = 1 and a = 2 can be interchanged. Taking this observation into account, let us
645
+ consider all non-equivalent options.
646
+ 4
647
+ Solutions of Maxwell equations
648
+ 1.
649
+ a1V1 ̸= 0 ⇒ the minor ˆW12 and its inverse matrix ˆΩ = ˆW −1
650
+ 12 have the form:
651
+ ˆW12 =
652
+
653
+
654
+
655
+
656
+
657
+
658
+ α2
659
+ α3
660
+ 0
661
+ 0
662
+ 0
663
+ α1
664
+ 0
665
+ α2
666
+ α3
667
+ 0
668
+ β1
669
+ 0
670
+ β2
671
+ β3
672
+ 0
673
+ 0
674
+ α1
675
+ 0
676
+ α2
677
+ α3
678
+ 0
679
+ β1
680
+ 0
681
+ β2
682
+ β3
683
+
684
+
685
+
686
+
687
+
688
+
689
+ ,
690
+ (28)
691
+ ˆΩ1 =
692
+
693
+
694
+
695
+
696
+
697
+
698
+
699
+
700
+
701
+ − V2
702
+ α1V1
703
+ − α3β2
704
+ α1V1
705
+ α2α3
706
+ α1V1
707
+ − α3β3
708
+ α1V1
709
+ α2
710
+ 3
711
+ α1V1
712
+ − V3
713
+ α1V1
714
+ α2β2
715
+ α1V1
716
+ − α2
717
+ 2
718
+ α1V1
719
+ α2β3
720
+ α1V1
721
+ −α2α3
722
+ α1V1
723
+ − V 2
724
+ 2
725
+ α1V 2
726
+ 1
727
+ (α3β1V1−α2β3V3)
728
+ α1V 2
729
+ 1
730
+ α3(α2V2−α1V1)
731
+ α1V 2
732
+ 1
733
+ −α3β3V2
734
+ α1V 2
735
+ 1
736
+ α2
737
+ 2V2
738
+ α1V 2
739
+ 1
740
+ − V2V3
741
+ α1V 2
742
+ 1
743
+ α2β2V2
744
+ α1V 2
745
+ 1
746
+ − α2
747
+ 2V2
748
+ α1V 2
749
+ 1
750
+ −α3β3‘V3
751
+ α1V 2
752
+ 1
753
+ α2
754
+ 3V3
755
+ α1V 2
756
+ 1
757
+ − V 2
758
+ 3
759
+ α1V 2
760
+ 1
761
+ α2β2V3
762
+ α1V 2
763
+ 1
764
+ − α2
765
+ 2V3
766
+ α1V 2
767
+ 1
768
+ (α3β2V3−α2β1V1)
769
+ α1V 2
770
+ 1
771
+ α2(α1V1−α3V3)
772
+ α1V 2
773
+ 1
774
+
775
+
776
+
777
+
778
+
779
+
780
+
781
+
782
+
783
+ (29)
784
+ Then the solution of equation (23) can be represented as:
785
+ ˆn1 = ˆΩ1ˆω1,
786
+ (30)
787
+ were
788
+ ˆnT
789
+ 1 = (n12, n13, n22, n23, n33);
790
+ ˆωT
791
+ 1 = (−( ˙β1 + α1n11), − ˙β2, ˙α2, −β3, ˙α3),
792
+ 7
793
+
794
+ Function
795
+ n11,
796
+ as well as the functions
797
+ αa,
798
+ βa
799
+ are arbitrary functions of
800
+ u0,
801
+ that
802
+ obey the condition (27).
803
+ 2.
804
+ α2V1 ̸= 0, ⇒ α1 = 0 ⇒ the minor ˆW −1
805
+ 14 and its inverse matrix ˆΩ2 = ˆW −1
806
+ 14 have the
807
+ form:
808
+ ˆW14 =
809
+
810
+
811
+
812
+
813
+
814
+
815
+ α2
816
+ α3
817
+ 0
818
+ 0
819
+ 0
820
+ β2
821
+ β3
822
+ 0
823
+ 0
824
+ 0
825
+ 0
826
+ 0
827
+ α2
828
+ α3
829
+ 0
830
+ 0
831
+ 0
832
+ 0
833
+ α2
834
+ α3
835
+ 0
836
+ β1
837
+ 0
838
+ β2
839
+ β3
840
+
841
+
842
+
843
+
844
+
845
+
846
+ ,
847
+ ˆΩ2 =
848
+
849
+
850
+
851
+
852
+
853
+
854
+
855
+
856
+ β3
857
+ V1
858
+ −α3
859
+ V1
860
+ 0
861
+ 0
862
+ 0
863
+ − β2
864
+ V1
865
+ α2
866
+ V1
867
+ 0
868
+ 0
869
+ 0
870
+ a2
871
+ 3β1β2
872
+ α2V 2
873
+ 1
874
+ −α2
875
+ 3β1
876
+ V 2
877
+ 1
878
+ 1
879
+ α2
880
+ − α3β3
881
+ α2V1
882
+ a2
883
+ 3
884
+ α2V1
885
+ −a3β1β2
886
+ V 2
887
+ 1
888
+ α2α3β1
889
+ V 2
890
+ 1
891
+ 0
892
+ β3
893
+ V1
894
+ − a3
895
+ V1
896
+ a2β1β2
897
+ V1
898
+ −α2
899
+ 2β1
900
+ V1
901
+ 0
902
+ − β2
903
+ V1
904
+ α2
905
+ V1
906
+
907
+
908
+
909
+
910
+
911
+
912
+
913
+
914
+ (31)
915
+ Solution of the equation (23) can be represented as:
916
+ ˆn2 = ˆΩˆω2,
917
+ (32)
918
+ were
919
+ ˆnT
920
+ 2 = (n12, n13, n22, n23, n33);
921
+ ˆω2 = (− ˙β1, −β1n11, − ˙β2, − ˙β3, ˙α3)
922
+ Function
923
+ n11,
924
+ as well as the functions
925
+ αa,
926
+ βa
927
+ are arbitrary functions of
928
+ u0,
929
+ that
930
+ obey the condition (27).
931
+ 3.
932
+ a3V1 ̸= 0, ⇒ a1 = a2 = 0 ⇒ the minor ˆW −1
933
+ 16 and its inverse matrix ˆΩ3 = ˆW −1
934
+ 16 have the
935
+ form:
936
+ ˆW16 =
937
+
938
+
939
+
940
+
941
+
942
+
943
+ 0
944
+ a3
945
+ 0
946
+ 0
947
+ 0
948
+ β2
949
+ β3
950
+ 0
951
+ 0
952
+ 0
953
+ 0
954
+ 0
955
+ 0
956
+ a3
957
+ 0
958
+ β1
959
+ 0
960
+ β2
961
+ β3
962
+ 0
963
+ 0
964
+ 0
965
+ 0
966
+ 0
967
+ a3
968
+
969
+
970
+
971
+
972
+
973
+
974
+ ,
975
+ ˆΩ3 =
976
+
977
+
978
+
979
+
980
+
981
+
982
+
983
+ − β3
984
+ a3β2
985
+ 1
986
+ β3
987
+ 0
988
+ 0
989
+ 0
990
+ 1
991
+ a3
992
+ 0
993
+ 0
994
+ 0
995
+ 0
996
+ β1β3
997
+ a3β2
998
+ 2
999
+ − β1
1000
+ β2
1001
+ 2
1002
+ − β3
1003
+ β2a3
1004
+ 1
1005
+ β2
1006
+ 0
1007
+ 0
1008
+ 0
1009
+ 1
1010
+ a3
1011
+ 0
1012
+ 0
1013
+ 0
1014
+ 0
1015
+ 0
1016
+ 0
1017
+ 1
1018
+ a3
1019
+
1020
+
1021
+
1022
+
1023
+
1024
+
1025
+
1026
+ (33)
1027
+ Then the solution of equation (23) can be represented as:
1028
+ ˆn3 = ˆΩ3ˆω3,
1029
+ (34)
1030
+ were
1031
+ ˆnT
1032
+ 3 = (n12, n13, n22, n23, n33);
1033
+ ˆωT
1034
+ 3 = (− ˙β1, −β1n11, − ˙β2, 0, − ˙β3)
1035
+ Function
1036
+ n11,
1037
+ as well as the functions
1038
+ α3,
1039
+ βa
1040
+ are arbitrary functions of
1041
+ u0,
1042
+ that
1043
+ obey the condition (27).
1044
+ 4.
1045
+ a1V3 ̸= 0. ⇒ V1 = V2 = 0,
1046
+ otherwise, we get a solution equivalent to the previous
1047
+ ones. As
1048
+ V3 ̸= 0 ⇒
1049
+ α3 = β3 = 0.
1050
+ The minor ˆW62 and its inverse matrix ˆΩ4 = ˆW −1
1051
+ 62 have
1052
+ 8
1053
+
1054
+ the form:
1055
+ ˆW26 =
1056
+
1057
+
1058
+
1059
+
1060
+
1061
+
1062
+ α1
1063
+ α2
1064
+ 0
1065
+ 0
1066
+ 0
1067
+ 0
1068
+ α1
1069
+ 0
1070
+ a2
1071
+ 0
1072
+ 0
1073
+ β1
1074
+ 0
1075
+ β2
1076
+ 0
1077
+ 0
1078
+ 0
1079
+ α1
1080
+ 0
1081
+ α2
1082
+ 0
1083
+ 0
1084
+ β1
1085
+ 0
1086
+ β2
1087
+
1088
+
1089
+
1090
+
1091
+
1092
+
1093
+ ,
1094
+ ˆΩ4 =
1095
+
1096
+
1097
+
1098
+
1099
+
1100
+
1101
+
1102
+ 1
1103
+ α1
1104
+ − α2β2
1105
+ α1V3
1106
+ α2
1107
+ 2
1108
+ α1V3
1109
+ 0
1110
+ 0
1111
+ 0
1112
+ β2
1113
+ V3
1114
+ −α2
1115
+ V3
1116
+ 0
1117
+ 0
1118
+ 0
1119
+ 0
1120
+ 0
1121
+ β2
1122
+ V3
1123
+ −α2
1124
+ V3
1125
+ 0
1126
+ − β1
1127
+ V3
1128
+ α1
1129
+ V3
1130
+ 0
1131
+ 0
1132
+ 0
1133
+ 0
1134
+ 0
1135
+ − β1
1136
+ V3
1137
+ α1
1138
+ V3
1139
+
1140
+
1141
+
1142
+
1143
+
1144
+
1145
+
1146
+ (35)
1147
+ Then the solution of equation (23) can be represented as:
1148
+ ˆn4 = ˆΩ4ˆω4.
1149
+ (36)
1150
+ were
1151
+ ˆnT
1152
+ 4 = (n11, n12, n13, n22, n23);
1153
+ ˆωT
1154
+ 4 = (− ˙β1, − ˙β2, ˙α2, 0, 0).
1155
+ Function
1156
+ n33,
1157
+ as well as the functions
1158
+ α1,
1159
+ α2
1160
+ βa
1161
+ are arbitrary functions of
1162
+ u0,
1163
+ that obey the condition (27).
1164
+ 5.
1165
+ Va = 0.
1166
+ Let us represent the system of Maxwell equations in the form:
1167
+ ˆQIˆnI = ˆωI
1168
+ were
1169
+ ˆQ =
1170
+
1171
+
1172
+
1173
+
1174
+
1175
+
1176
+
1177
+
1178
+ α1
1179
+ α2
1180
+ α3
1181
+ 0
1182
+ 0
1183
+ 0
1184
+ 0
1185
+ α1
1186
+ 0
1187
+ α2
1188
+ α3
1189
+ 0
1190
+ 0
1191
+ 0
1192
+ α1
1193
+ 0
1194
+ α2
1195
+ α3
1196
+ β1
1197
+ β2
1198
+ β3
1199
+ 0
1200
+ 0
1201
+ 0
1202
+ 0
1203
+ β1
1204
+ 0
1205
+ β2
1206
+ β3
1207
+ 0
1208
+ 0
1209
+ 0
1210
+ β1
1211
+ 0
1212
+ β2
1213
+ β3
1214
+
1215
+
1216
+
1217
+
1218
+
1219
+
1220
+
1221
+
1222
+ ,
1223
+ (37)
1224
+ ˆωI = (ˆωβ, ˆωα);
1225
+ ˆωβ = −( ˙β1, ˙β2, ˙β3),
1226
+ ˆωα = ( ˙α1, ˙α2, ˙α3)
1227
+ ˆnI = (ˆnα, ˆnβ);
1228
+ ˆnα = (n11, n12, n13),
1229
+ ˆnβ = (n22, n23, n33).
1230
+ To provide the classification, it is sufficient to consider the options:
1231
+ 1)
1232
+ a1 ̸= 0,
1233
+ 2)
1234
+ a3 ̸=
1235
+ 0,
1236
+ a1 = a2 = 0.
1237
+ a) a1 ̸= 0 ⇒ βa = αaβ1
1238
+ α1 .
1239
+ ˆWIˆnα = (ˆωβ − ˆQ1ˆnβ) ⇒ ˆnα = ˆW −1
1240
+ I (ˆωβ − ˆQ1ˆnβ),
1241
+ β1 ˆWIˆnα = α1ˆωα − β1 ˆQ1ˆnβ ⇒ β1ˆωβ − α1ˆωα = 0 ⇒
1242
+
1243
+
1244
+
1245
+ α1 ˙α2 + β1 ˙β2 = 0,
1246
+ α1 ˙α3 + β1 ˙β3 = 0,
1247
+ α1 ˙α1 + β1 ˙β1 = 0.
1248
+
1249
+
1250
+
1251
+
1252
+ α1 = e sin ϕ,
1253
+ β1 = e cos ϕ,
1254
+ e = const,
1255
+ α2 = ec2 sin ϕ,
1256
+ β1 = ec2 cos ϕ,
1257
+ e, c2 = const,
1258
+ α3 = ec3 sin ϕ,
1259
+ β1 = ec3 cos ϕ,
1260
+ e, c3 = const.
1261
+ (38)
1262
+ 9
1263
+
1264
+ Here:
1265
+ ˆWI =
1266
+
1267
+
1268
+ α1
1269
+ α2
1270
+ α3
1271
+ 0
1272
+ α1
1273
+ 0
1274
+ 0
1275
+ 0
1276
+ α1
1277
+
1278
+ � , ˆW −1
1279
+ I
1280
+ =
1281
+
1282
+
1283
+ 1
1284
+ α1
1285
+ −α2
1286
+ α2
1287
+ 1
1288
+ −α3
1289
+ α2
1290
+ 1
1291
+ 0
1292
+ 1
1293
+ α1
1294
+ 0
1295
+ 0
1296
+ 0
1297
+ 1
1298
+ α1
1299
+
1300
+ � , ˆQI =
1301
+
1302
+
1303
+ 0
1304
+ 0
1305
+ 0
1306
+ α2
1307
+ α3
1308
+ 0
1309
+ 0
1310
+ α2
1311
+ α3,
1312
+
1313
+
1314
+ α1 = e sin ϕ,
1315
+ β1 = e cos ϕ,
1316
+ e, ca = const,
1317
+ Then matrices ˆWI, ˆW −1
1318
+ I , ˆQI and lines ˆωT take the form:
1319
+ ˆWI = sin ϕ ˆP,
1320
+ ˆW −1
1321
+ I
1322
+ =
1323
+ 1
1324
+ sin ϕ
1325
+ ˆP −1,
1326
+ ˆQI = sin ϕ ˆQ.
1327
+ ˆP =
1328
+
1329
+
1330
+ 1
1331
+ c2
1332
+ c3
1333
+ 0
1334
+ 1
1335
+ 0
1336
+ 0
1337
+ 0
1338
+ 1
1339
+
1340
+ � ,
1341
+ ˆP −1 =
1342
+
1343
+
1344
+ 1
1345
+ −c2
1346
+ −c3
1347
+ 0
1348
+ 1
1349
+ 0
1350
+ 0
1351
+ 0
1352
+ 1
1353
+
1354
+ � ,
1355
+ ˆQ =
1356
+
1357
+
1358
+ 0
1359
+ 0
1360
+ 0
1361
+ c2
1362
+ c3
1363
+ 0
1364
+ 0
1365
+ c2
1366
+ c3,
1367
+
1368
+
1369
+ ˆωT
1370
+ α = ˆωT
1371
+ β = ˙ϕ sin ϕ ˆCT = ˙ϕ sin ϕ(1, c2, c3),
1372
+ ˆnα = ˆw−1( ˙ϕ ˆCT − ˆQˆnβ)
1373
+ Function
1374
+ n22, n23, n33,
1375
+ as well as the function
1376
+ ϕ
1377
+ are arbitrary functions of
1378
+ u0.
1379
+ b) Va = 0,
1380
+ α3 ̸= 0.
1381
+ ⇒ β1 = β2 = 0.
1382
+ The system of Maxwell equations has the form:
1383
+ α3n13 = α3n23 = 0,
1384
+ α3n33 = − ˙β3,
1385
+ β3n33 = ˙α3.
1386
+ a3 ˙a3 + β3 ˙β3 = 0 ⇒ a3 = c sin ϕ,
1387
+ β3 = cos ϕ
1388
+ From here:
1389
+ n33 = ˙ϕ,
1390
+ n13 = n23 = α1 = α2 = β1 = β2 = 0,
1391
+ α3 = c sin ϕ,
1392
+ β3 = c cos ϕ.
1393
+ Functions
1394
+ ϕ,
1395
+ n11,
1396
+ n12,
1397
+ n22 - are arbitrary functions on
1398
+ u0,
1399
+ c = const.
1400
+ 5
1401
+ Conclusion
1402
+ It is known that homogeneous spaces of IV and IX types according to Bianchi classification
1403
+ include as special cases the spaces of constant curvature.This causes a special interest to them
1404
+ in cosmology. In the Universe with the metric of homogeneous space all physical fields are
1405
+ invariant with respect to the group of motions of the space-time. Therefore, exactly such fields
1406
+ should be considered in the first place when solving the self-consistent Einstein equations,
1407
+ in particular the Einstein-Maxwell equations.
1408
+ The final goal of classification of PSS with
1409
+ admissible electromagnetic fields is to enumerate all electrovacuum solutions of the Einstein-
1410
+ Maxwell equations. In [40], [41] the complete classification of vacuum solutions of the Maxwell
1411
+ equations for homogeneous spaces with solvable groups of motions has been carried out. In the
1412
+ present paper the same problem is solved for HPSS of IX-type. For the final decision of the
1413
+ first stage of the classification problem it remains to consider HPSS V III-type, which will be
1414
+ 10
1415
+
1416
+ done in the next paper. The results obtained will be used in the second stage for integration
1417
+ of the corresponding Einstein-Maxwell equations.
1418
+ FUNDING: The work is supported by Russian Science Foundation, project number N 23-
1419
+ 21-00275.
1420
+ INSTITUTIONAL REVIEW BOARD STATEMENT: Not applicable.
1421
+ INFORMED CONSENT STATEMENT: Not applicable.
1422
+ DATA AVAILABILITY STATEMENT: The data that support the findings of this study
1423
+ are available within the article.
1424
+ CONFLICTS OF INTEREST: The author declares no conflict of interest.
1425
+ References
1426
+ [1] Stackel. P. Uber die intagration der Hamiltonschen differentialechung mittels separation
1427
+ der variablen. Math. Ann. 1897. 49, (145-147 pp.);
1428
+ [2] Eisenhart L.P. Separable systems of stackel. Ann.Math. 1934, 35, (284-305 pp).;
1429
+ [3] Levi-Civita T. Sulla Integraziome Della Equazione Di Hamilton-Jacobi Per Separazione
1430
+ Di Variabili. Math.Ann. 1904 59, (383-397 pp.);
1431
+ [4] Jarov-Jrovoy
1432
+ M.S.
1433
+ Integration
1434
+ of
1435
+ Hamilton-Jacobi
1436
+ equation
1437
+ by
1438
+ complete
1439
+ sep-
1440
+ aration
1441
+ of
1442
+ variables
1443
+ method.
1444
+ J.Appl.Math.Mech..
1445
+ 1963,
1446
+ 27,
1447
+ No
1448
+ 6,
1449
+ (173-219
1450
+ pp).https://doi.org/10.1016/0021-8928(63)90122-9;
1451
+ [5] Carter B. New family of Einstein spaces. Phys.Lett. 1968, A.25, No 9 (399-400pp.),
1452
+ doi.org/10.1016/0375-9601(68)90240-5
1453
+ [6] Shapovalov V.N., Symmetry and separation of variables in the Hamilton-Jacobi equation.
1454
+ Sov. Phys.J.. 1978, 21, 1124-1132pp. doi: 10.1007/BF00894560;
1455
+ [7] Shapovalov V.N.,
1456
+ Stackel‘s spaces. Sib. Math. J. 1979,
1457
+ 20,
1458
+ (1117-1130pp.),
1459
+ doi:
1460
+ org/10.1007/BF00971844;
1461
+ [8] W.
1462
+ Miller.
1463
+ Symmetry
1464
+ And
1465
+ Separation
1466
+ Of
1467
+ Variables.
1468
+ Cambridge
1469
+ University
1470
+ Press:Cambridge. 1984, (318 p.p.);
1471
+ [9] Obukhov V.V. Hamilton-Jacobi equation for a charged test particle in the Stackel space
1472
+ of type (2.0). Symmetry, 12, 2020, 12891291. doi: 10.3390/sym12081289.
1473
+ [10] Obukhov V.V. Hamilton-Jacobi equation for a charged test particle in the Stackel
1474
+ space of type (2.1).Int. J. Geom. Meth. Mod. Phys, 2020, 17, 14, 2050186. doi:
1475
+ 10.1142/S0219887820501868.
1476
+ [11] Obukhov V.V. Separation of variables in Hamilton-Jacobi and Klein-Gordon-Fock equa-
1477
+ tions for a charged test particle in the stackel spaces of type (1.1). Int. J. Geom. Meth.
1478
+ Mod. Phys.2021, 18, 03, (2150036); doi:10.1142/S0219887821500365
1479
+ 11
1480
+
1481
+ [12] Odintsov, S.D. Editorial for Special Issue Feature Papers 2020. Symmetry. 2023, 15, 8.
1482
+ doi.org/10.3390/sym15010008
1483
+ [13] McLenaghan R. G., Rastelli G. and Valero C. Complete separability of the Hamilton-
1484
+ Jacobi equation for the charged particle orbits in a Lienard-Wiehert field /em J. Math.
1485
+ Phys., /bf2020,61, (122903). doi.org/10.1063/5.0030305.
1486
+ [14] Mitsopoulos A., Tsamparlis M., Leon G. A.Paliathanasis. A.Paliathanasis. New conserva-
1487
+ tion laws and exact cosmological solutions in Brans-Dicke cosmology with an extra scalar
1488
+ field. Symmetry. 2021, 13(8):1364, doi: 10.3390/sym13081364
1489
+ [15] Claudio Dappiaggi, Benito A. Juarez-Aubry, Alessio Marta Ground. State for the Klein-
1490
+ Gordon field in anti-de Sitter spacetime with dynamical Wentzell boundary conditions.
1491
+ Phys. Rev. D. 2022, 105, 105017 doi.org/10.1103/PhysRevD.105.105017.
1492
+ [16] Francisco Astorga, J Felix Salazar and Thomas Zannias On the integrability of the geodesic
1493
+ flow on a Friedmann-Robertson-Walker spacetime Physica Scripta, 2018, 93, 8, 93 085205
1494
+ doi. 10.1088/1402-4896/aacd44
1495
+ [17] Capozziello S., De Laurentis M., Odintsov D. Hamiltonian dynamics and Noether sym-
1496
+ metries in extended gravity cosmology. Eur.Phys.J. 2012, C72, 2068 (22 pp.), doi:
1497
+ 10.1140/epjc/s10052-012-2068-0;
1498
+ [18] Salih
1499
+ Kibaroglu,
1500
+ Oktay
1501
+ Cebecioglu.
1502
+ Generalized
1503
+ cosmological
1504
+ constant
1505
+ from
1506
+ gauging
1507
+ Maxwell-conformal
1508
+ algebra,
1509
+ Phys.Lett.B,
1510
+ 2020,
1511
+ 803,
1512
+ 135295.
1513
+ doi.org/10.1016/j.physletb.2020.135295
1514
+ [19] Nojiri
1515
+ Shin’ichi,
1516
+ Odintsov
1517
+ Sergei
1518
+ D.,
1519
+ Faraoni
1520
+ Valerio,
1521
+ Searching
1522
+ for
1523
+ dynamical
1524
+ black
1525
+ holes
1526
+ in
1527
+ various
1528
+ theories
1529
+ of
1530
+ gravity.
1531
+ Phys.Rev.D.
1532
+ 2021,
1533
+ 103,
1534
+ 4,
1535
+ 044055.
1536
+ doi:10.1103/PhysRevD.103.044055
1537
+ [20] Epp V. Pervukhina O. The Stormer problem for an aligned rotator. MNRAS, 2018 474,
1538
+ (5330-5339) doi.org/10.1093/mnras/stx3102.
1539
+ [21] Epp V., Masterova M.A. Effective potential energy for relativistic particles in the field
1540
+ of inclined rotating magnetized sphere. Astrophys Space Sci, 2014, 353, (473-483).
1541
+ doi.org/10.1007/s10509-014-2066-9.
1542
+ [22] Kumaran Y.; Ovgun A. Deflection angle and shadow of the reissner-nordstrom black
1543
+ hole with higher-order magnetic correction in einstein-nonlinear-maxwell fields. Symmetry.
1544
+ 2022, 14, 2054. https://doi.org/10.3390/sym14102054
1545
+ [23] Osetrin K.; Osetrin E. Shapovalov wave-like spacetimes. Symmetry 2020, 12, 1372.
1546
+ https://doi.org/10.3390/SYM12081372.
1547
+ [24] Osetrin E.; Osetrin K.; Filippov A. Plane Gravitational Waves in Spatially-Homogeneous
1548
+ Models of type-(3.1) Stackel Spaces. Russian Physics Journal 2019, 62,
1549
+ 292-301.
1550
+ https://doi.org/10.1007/s11182-019-01711-1.
1551
+ 12
1552
+
1553
+ [25] Osetrin K., Osetrin E. and Osetrina E. Geodesic deviation and tidal acceleration in the
1554
+ gravitational wave of the Bianchi type IV universe. Eur. Phys. J. Plus. 2022, 137, (856).
1555
+ https://doi.org/10.1140/epjp/s13360-022-03061-3
1556
+ [26] Osetrin, K.; Osetrin, E.; Osetrina, E. Gravitational wave of the Bianchi VII universe:
1557
+ Particle trajectories, geodesic deviation and tidal accelerations. Eur. Phys. J. C 2022,
1558
+ 82, 1–16. doi.org/10.1140/epjc/s10052-022-10852-6.
1559
+ [27] Shapovalov, A.V., Shirokov I.V. Noncommutative integration method for linear partial
1560
+ differential equations. functional algebras and dimensional reduction. Theoret. And Math.
1561
+ Phys. 1996, 106:1, (1-10). doi.org/10.4213/tmf1093.
1562
+ [28] Shapovalov, A.; Breev, A. Harmonic Oscillator Coherent States from the Standpoint of
1563
+ Orbit Theory.Symmetry 2023, 15, 282. doi.org/10.3390/sym15020282
1564
+ [29] Breev, A.I., Shapovalov, A.V. Non-commutative integration of the Dirac equation in ho-
1565
+ mogeneous spaces. Symmetry 2020, 12, 1867.
1566
+ [30] Breev, A.I.; Shapovalov, A.V. Yang–Mills gauge fields conserving the symmetry algebra of
1567
+ the Dirac equation in a homogeneous space. J. Phys.: Conf. Ser. 2014, 563, 012004.
1568
+ [31] Magazev
1569
+ A.A.,Boldyreva
1570
+ M.N.
1571
+ Schrodinger
1572
+ equations
1573
+ in
1574
+ electromagnetic
1575
+ fields:
1576
+ symmetries
1577
+ and
1578
+ noncommutative
1579
+ integration,
1580
+ Symmetry.
1581
+ 2021,
1582
+ 13,
1583
+ (1527).
1584
+ https://doi.org/10.3390/sym13081527
1585
+ [32] Magazev A.A. Integrating Klein-Gordon-Fock equations in an extremal electromagnetic
1586
+ field on Lie groups. Theor.and Math.Phys., 2012, 173:3, (1654-1667). doi: 10.1007/s11232-
1587
+ 012-0139-x, arxiv.org/abs/1406.5698;
1588
+ [33] Obukhov V.V. Algebra of symmetry operators for Klein-Gordon-Fock Equation. Symme-
1589
+ try. 2021, 13, 727 (15p.). https://doi.org/10.3390/sym13040727.
1590
+ [34] Odintsov,
1591
+ S.D. Editorial for Feature Papers 2021-2022. Symmetry. 2023, 15, 32.
1592
+ doi.org/10.3390/sym15010032
1593
+ [35] Obukhov V.V. Algebra of the symmetry operators of the Klein-Gordon-Fock equation for
1594
+ the case when groups of motions G3 act transitively on null subsurfaces of spacetime.
1595
+ Symmetry. 2022, 14, (346). https://doi.org/10.3390/sym14020346
1596
+ [36] Obukhov V.V. Algebras of integrals of motion for the Hamilton-Jacobi and Klein-
1597
+ Gordon-Fock equations in spacetime with a four-parameter groups of motions in the
1598
+ presence of an external electromagnetic field J. Math. Phys. 2022, 63,
1599
+ Issue 2.
1600
+ https://doi.org/10.1063/5.0080703
1601
+ [37] Stephani, H., Kramer, D., MacCallum, M., Hoenselaers, C., Herlt, E. Exact Solu-
1602
+ tions of Einstein’s Field Equations (2nd ed., Cambridge Monographs on Mathematical
1603
+ Physics).2003 Cambridge: Cambridge University Press. doi:10.1017/CBO9780511535185
1604
+ [38] Landau L.D., Lifshits E.M. Theoretical physics, Field theory. 7th ed. Moskow.(vol. II):
1605
+ Science. Ch. ed. Phys.-Math. lit., 1988. - (512 p). ISBN 5-02-014420-7
1606
+ 13
1607
+
1608
+ [39] Petrov A. Z. Einstein Spaces, Oxford, 1969.
1609
+ [40] Obukhov V.V. Maxwell Equations in Homogeneous Spaces for Admissible Electromagnetic
1610
+ Fields.Universe. 2022, 8, (245). https://doi.org/10.3390/universe8040245
1611
+ [41] Obukhov, V.V. Maxwell Equations in Homogeneous Spaces with Solvable Groups of Mo-
1612
+ tions. Symmetry. 2022, 14, (2595).https://doi.org/10.3390/sym14122595
1613
+ [42] Stephani H., Kramer D., MacCallum M. A. H., Hoenselaers C., Herlt E. Exact Solu-
1614
+ tions of Einstein’s Field Equations Second Edition, 2003, Cambrige university press.
1615
+ doi.org/10.1017/CBO9780511535185
1616
+ 14
1617
+
9NFJT4oBgHgl3EQfoSxe/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
AtAyT4oBgHgl3EQfd_hU/content/2301.00312v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85ff01e4afcfd815466ba709f50a9c56f2cf3234cb25d1fa4e5b4d9d696cc842
3
+ size 1557622
AtAyT4oBgHgl3EQfd_hU/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd9d0350675397ee7798f1507a39a04c70a6735eee63fb82710631ae8bcda6ca
3
+ size 101954
BNE0T4oBgHgl3EQfxwK1/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fdf45de1c585fd45d4f6d0fd1ed53602817fae67409208cf1ad0e4a6f0bd7c3a
3
+ size 169124
BdE0T4oBgHgl3EQfyAIA/content/tmp_files/2301.02652v1.pdf.txt ADDED
@@ -0,0 +1,1334 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Draft version January 9, 2023
2
+ Typeset using LATEX twocolumn style in AASTeX631
3
+ Diverse Carbonates in Exoplanet Oceans Promote the Carbon Cycle
4
+ Kaustubh Hakim
5
+ ,1 Meng Tian
6
+ ,1 Dan J. Bower
7
+ ,1 and Kevin Heng
8
+ 2, 3, 4
9
+ 1University of Bern, Center for Space and Habitability, Gesellschaftsstrasse 6, CH-3012 Bern, Switzerland
10
+ 2Ludwig Maximilian University, University Observatory Munich, Scheinerstrasse 1, Munich D-81679, Germany
11
+ 3University of Warwick, Department of Physics, Astronomy & Astrophysics Group, Coventry CV4 7AL, United Kingdom
12
+ 4University of Bern, ARTORG Center for Biomedical Engineering Research, Murtenstrasse 50, CH-3008, Bern, Switzerland
13
+ ABSTRACT
14
+ Carbonate precipitation in oceans is essential for the carbonate-silicate cycle (inorganic carbon cycle)
15
+ to maintain temperate climates.
16
+ By considering the thermodynamics of carbonate chemistry, we
17
+ demonstrate that the ocean pH decreases by approximately 0.5 for a factor of 10 increase in the
18
+ atmospheric carbon dioxide content. The upper and lower limits of ocean pH are within 1–4 of each
19
+ other, where the upper limit is buffered by carbonate precipitation and defines the ocean pH when
20
+ the carbon cycle operates. If the carbonate compensation depth (CCD) resides above the ocean floor,
21
+ then carbonate precipitation and the carbon cycle cease to operate.
22
+ The CCD is deep (>40 km)
23
+ for high ocean temperature and high atmospheric carbon dioxide content. Key divalent carbonates of
24
+ magnesium, calcium and iron produce an increasingly wider parameter space of deep CCDs, suggesting
25
+ that chemical diversity promotes the carbon cycle. The search for life from exoplanets will benefit by
26
+ including chemically more diverse targets than Earth twins.
27
+ Keywords: Extrasolar rocky planets (511); Carbon dioxide (196); Habitable zone (696); Ocean-
28
+ atmosphere interactions (1150); Geological processes (2288); (Unified Astronomy Thesaurus)
29
+ 1. INTRODUCTION
30
+ The carbonate-silicate cycle, also known as the in-
31
+ organic carbon cycle, is a negative climate feedback
32
+ mechanism that stabilises the surface temperature via
33
+ the greenhouse effect of carbon dioxide in response
34
+ to changes in volcanism rates, stellar luminosity, at-
35
+ mospheric composition and opacity, planetary orbital
36
+ movements and spin axis tilt (Berner 2004; Catling
37
+ & Kasting 2017).
38
+ Continental silicate rocks and at-
39
+ mospheric carbon dioxide react with water in a pro-
40
+ cess known as silicate weathering to produce carbonate-
41
+ forming ions that precipitate as carbonates onto the
42
+ ocean floor (Walker et al. 1981).
43
+ The carbon cycle
44
+ is completed when carbonates are transferred into the
45
+ mantle for deep storage or carbon is eventually re-
46
+ leased back into the atmosphere by volcanism (Holland
47
+ 1978; Sleep & Zahnle 2001), although the degassing ef-
48
+ Corresponding author: Kaustubh Hakim
49
+ kaustubh.hakim@unibe.ch
50
+ ficiency is debated (Kelemen & Manning 2015; Foley
51
+ 2015). Silicate weathering and carbonate precipitation
52
+ are traditionally represented by the net chemical reac-
53
+ tion (Walker et al. 1981),
54
+ CaSiO3 + CO2 → CaCO3 + SiO2,
55
+ (1)
56
+ where wollastonite (CaSiO3), which serves as a proxy for
57
+ silicate rocks, is converted into calcite (CaCO3). Cal-
58
+ cium thus plays a crucial role in silicate weathering and
59
+ carbonate precipitation and is present as Ca2+ cations
60
+ in oceans (Sect. 2).
61
+ The existence of habitable zones assumes that the
62
+ carbon cycle operates on Earth analogues to stabilise
63
+ their atmospheric carbon dioxide content (Kasting et al.
64
+ 1993).
65
+ Implicitly, this assumes not only that silicate
66
+ weathering operates, but that ocean floor precipitation
67
+ and deep storage of carbonates also occur. There exists
68
+ a critical ocean depth known as the carbonate compen-
69
+ sation depth (CCD), below which carbonates are unable
70
+ to exist in their solid form because carbonate solubility
71
+ increases with pressure in the ocean (Zeebe & West-
72
+ broek 2003, see also Sect. 2.3, Figure 1). In modern
73
+ arXiv:2301.02652v1 [astro-ph.EP] 6 Jan 2023
74
+
75
+ ID2
76
+ CCD
77
+ !"#$#
78
+ %&'(
79
+ Ocean Depth
80
+ %)*+,-
81
+ %&'( ≤ %,/0
82
+ 1)*+,- = 13456 = 1
83
+ !78'(,#$#
84
+ Carbonates
85
+ Silicates
86
+ Dissolved ions
87
+ Figure 1. Model parameters, nDtot where D = Ca, Mg or
88
+ Fe, nSiO2,tot, PCO2, Patm, Poc and T. See Sect. 2 and Table 1
89
+ for a full list of output quantities and description.
90
+ Earth oceans, the CCD is located between 4–5 km, be-
91
+ low the average ocean depth of about 3.8 km (Zeebe
92
+ 2012). If the CCD resides at a depth above the ocean
93
+ floor, then carbonates are unable to settle. This leads
94
+ to the disruption of the carbon cycle—at least, as it is
95
+ understood to operate on Earth. Moreover, there are
96
+ currently no theoretical constraints on exoplanet ocean
97
+ chemistry. We investigate the interplay between atmo-
98
+ spheric carbon dioxide content, ocean acidity (pH) and
99
+ carbonate precipitation.
100
+ We then calculate the CCD
101
+ over a broad range of physical conditions.
102
+ 2. METHODS
103
+ 2.1. Ocean chemistry model
104
+ 2.1.1. Ca system
105
+ Ocean chemistry is modelled by considering thermo-
106
+ chemical equilibrium for pure Ca, Mg, or Fe systems.
107
+ The CO2 partial pressure PCO2, ocean–surface temper-
108
+ ature T and local ocean pressure Poc are control pa-
109
+ rameters (Figure 1, Table 1). In the Ca system, there
110
+ are 13 unknowns, the number density n of H+, OH−,
111
+ H2O, HCO−
112
+ 3 , CO2−
113
+ 3 , CO2(aq), Catot, Ca2+, SiO2,tot,
114
+ SiO2(aq), quartz SiO2(s), wollastonite CaSiO3(s) and
115
+ calcite CaCO3(s). Out of the 13 unknowns, 2 are conti-
116
+ nental silicate weathering products, nCatot and nSiO2,tot,
117
+ that depend on PCO2 and T (Sect.
118
+ 2.2).
119
+ There are
120
+ 11 remaining unknowns. We solve for 3 mass conserva-
121
+ tion equations (for H, Ca and SiO2), 1 charge balance
122
+ equation, and 7 equations from 7 chemical reactions pro-
123
+ viding relations between equilibrium constants (that de-
124
+ pend on Poc and T), reactants and products.
125
+ Table 1. Parameters and output quantities.
126
+ Symbol
127
+ Description
128
+ Reference
129
+ Parameters for ocean chemistry
130
+ T
131
+ Ocean–Surface temperature
132
+ 288 K
133
+ PCO2
134
+ CO2 partial pressure
135
+ 0.3 mbar
136
+ Patm
137
+ Atmospheric pressure
138
+ 1 bar
139
+ Poc
140
+ Ocean layer pressure
141
+ 1 bar
142
+ Parameters for weathering
143
+ nDtot,0
144
+ Ca, Mg or Fe ref. number density
145
+ 1 m−3
146
+ β
147
+ Weathering power-law exponent
148
+ 0.3
149
+ Te
150
+ e-folding temperature
151
+ 13.7 K
152
+ Output quantities
153
+ nX
154
+ Number density of X [m−3]
155
+ pH
156
+ –log10(nH+/n0); n0 = 103 m−3
157
+ These 3 mass-conservation equations, 1 charge balance
158
+ equation and 7 reactions (water dissociation, Henry’s
159
+ law/physical CO2 dissolution, chemical CO2 dissolu-
160
+ tion, bicarbonate ion dissociation, calcite precipitation,
161
+ quartz precipitation and wollastonite precipitation) are
162
+ specified below. Henry’s law gives the amount of CO2
163
+ physically dissolved in ocean water in equilibrium with
164
+ PCO2:
165
+ CO2(g) ⇌ CO2(aq).
166
+ (2)
167
+ The chemical dissolution or dissociation of CO2 in ocean
168
+ water leads to the production of HCO−
169
+ 3 and H+ ions and
170
+ thereby increases the ocean acidity (and decreases ocean
171
+ pH = − log10(nH+/n0), where the standard number den-
172
+ sity n0 = 1 m−3) by the following reaction:
173
+ CO2(g) + H2O ⇌ H+ + HCO−
174
+ 3 .
175
+ (3)
176
+ To maintain the charge balance in ocean water, the ad-
177
+ dition of Ca2+ to oceans decreases the number density
178
+ of H+ and hence increases the ocean pH. The charge
179
+ balance equation is given by:
180
+ 2nCa2+ + nH+ = nHCO−
181
+ 3 + 2nCO2−
182
+ 3
183
+ + nOH−,
184
+ (4)
185
+ where CO2−
186
+ 3
187
+ is produced due to the bicarbonate disso-
188
+ ciation reaction:
189
+ HCO−
190
+ 3 ⇌ CO2−
191
+ 3
192
+ + H+,
193
+ (5)
194
+ and where OH− is produced due to the water dissocia-
195
+ tion reaction:
196
+ H2O ⇌ H+ + OH−.
197
+ (6)
198
+
199
+ 3
200
+ The mass conservation of H is given by
201
+ nHtot = 2nH2O + nH+ + nHCO−
202
+ 3 .
203
+ (7)
204
+ Catot partitions into Ca2+, calcite and wollastonite
205
+ which is accounted for by mass conservation:
206
+ nCatot = nCa2+ + nCal + nWo.
207
+ (8)
208
+ Calcite precipitation occurs when nCa2+ is saturated to
209
+ a certain value determined by the equilibrium constant
210
+ of the calcite precipitation reaction and the abundance
211
+ of nCO2−
212
+ 3 :
213
+ Ca2+ + CO2−
214
+ 3
215
+ ⇌ CaCO3(s).
216
+ (9)
217
+ SiO2,tot partitions into aqueous silica SiO2(aq), quartz
218
+ SiO2(s) and wollastonite CaSiO3(s). The mass conser-
219
+ vation for SiO2 is given by:
220
+ nSiO2,tot = nSiO2(aq) + nQz + nWo.
221
+ (10)
222
+ The quartz precipitation reaction is:
223
+ SiO2(aq) ⇌ SiO2(s).
224
+ (11)
225
+ The reaction of wollastonite precipitation is given by:
226
+ Ca2+ + SiO2(aq) + H2O ⇌ 2H+ + CaSiO3(s).
227
+ (12)
228
+ These equilibrium chemistry calculations are per-
229
+ formed using Reaktoro v2 (Leal 2015), a multi-phase
230
+ (aqueous, gas and solid mineral phases) chemistry soft-
231
+ ware.
232
+ This software implements the extended law of
233
+ mass action including the determination of stable and
234
+ unstable species for a given set of species in the system
235
+ (Leal et al. 2017).
236
+ We use the SUPCRTBL database
237
+ for thermodynamic data (Johnson et al. 1992; Zimmer
238
+ et al. 2016), the Peng-Robinson activity model for gases
239
+ (Peng & Robinson 1976), the HKF activity model for
240
+ water (Helgeson et al. 1981) and the Drummond activ-
241
+ ity model for CO2(aq) (Drummond 1981).
242
+ 2.1.2. Mg and Fe systems
243
+ In the Mg system, Ca is replaced by Mg, calcite
244
+ by magnesite MgCO3(s) and wollastonite by enstatite
245
+ Mg2Si2O6(s). This includes replacing equilibrium con-
246
+ stants of all reactions including Mg. Similarly, in the
247
+ Fe system, Ca is replaced by Fe, calcite by siderite
248
+ FeCO3(s) and wollastonite by fayalite Fe2SiO4(s). We
249
+ limit our calculations to Fe2+ although its oxidation has
250
+ inhibited the formation of siderite during Earth’s his-
251
+ tory, particularly since the great oxidation event (Rye
252
+ et al. 1995).
253
+ 2.2. Weathering model
254
+ The introduction of carbonate-producing divalent
255
+ cations in oceans is dictated by silicate weathering. Sil-
256
+ icate weathering and therefore the total number density
257
+ of divalent cations D2+ (D = Ca, Mg or Fe) must depend
258
+ on the CO2 partial pressure PCO2 and surface tempera-
259
+ ture T (Walker et al. 1981; Hakim et al. 2021),
260
+ nDtot = fW (PCO2, T) = nDtot,0
261
+ � PCO2
262
+ PCO2,0
263
+ �β
264
+ exp
265
+ �T − T0
266
+ Te
267
+
268
+ ,
269
+ (13)
270
+ where ‘0’ represents the Earth reference values (Table 1),
271
+ Te = 13.7 K is the e-folding temperature and β = 0.3 is
272
+ the weathering power-law exponent (Walker et al. 1981).
273
+ However, not all added Ca (or Mg, Fe) in oceans re-
274
+ mains in the form of divalent cations, a fraction of it
275
+ precipitates as carbonates on the ocean floor and an-
276
+ other fraction as silicates. For this reason, we perform
277
+ partitioning calculations of Ca (or Mg, Fe) in different
278
+ phases following the ocean chemistry model (Sect. 2.1).
279
+ 2.3. CCD model
280
+ Carbonates are deposited onto the ocean floor as part
281
+ of sediments. The transition from calcite-rich to calcite-
282
+ free sediments is gradual. The carbonate compensation
283
+ depth (CCD) for the Earth ocean is normally defined
284
+ as the depth at which the dissolution flux of calcite bal-
285
+ ances the precipitation flux (Zeebe 2012). The depth
286
+ at which the rapid dissolution of calcite-rich sediments
287
+ begins is known as the lysocline, which is a sediment
288
+ property (Zeebe & Westbroek 2003). The lysocline and
289
+ CCD serve as bounds on the transition zone (∼0.5 km)
290
+ between calcite-rich and calcite-free sediments. Other
291
+ definitions for the CCD exist (Berger et al. 1976; Ridg-
292
+ well & Zeebe 2005; Zeebe 2012). The depth of ocean d
293
+ [km] in terms of ocean pressure Poc [bar] at the equator
294
+ is given by (Leroy & Parthiot 1998)
295
+ d =
296
+ 1
297
+ 9.7803 × 103 + 0.011Poc (97.266Poc − 2.512 × 10−3P 2
298
+ oc
299
+ + 2.28 × 10−7P 3
300
+ oc − 1.8 × 10−11P 4
301
+ oc).
302
+ (14)
303
+ We consider the CCD to be the depth dCCD (equiv-
304
+ alent to the ocean pressure where Poc = PCCD) at
305
+ which 99.9% of near-surface (Poc = Psurf) Ca, Mg or
306
+ Fe-carbonates dissolve,
307
+ nCarb,CCD = 0.001 nCarb,surf.
308
+ (15)
309
+ Our calculations of CCD are performed up to dCCD =
310
+ 45 km because of the availability of thermodynamic data
311
+ up to the pressure of 5000 bar (Zimmer et al. 2016). This
312
+ limitation does not affect our conclusions.
313
+
314
+ 4
315
+ 2.4. Analytical solution of ocean pH
316
+ Upper limit of ocean pH. For calcite precipitation, all
317
+ reactions in Section 2 need to be satisfied. However, two
318
+ of these reactions can be used to analytically constrain
319
+ ocean pH: Equations 9 and 16 where Equation 16 is a
320
+ combination of Equations 3 and 5,
321
+ CO2(g) + H2O ⇌ 2H+ + CO2−
322
+ 3 .
323
+ (16)
324
+ The ocean pH can be written as a function of PCO2,
325
+ nCa2+ and equilibrium constants of Equations 9 and 16
326
+ (Appendix A):
327
+ pH = −1
328
+ 2
329
+
330
+ log PCO2 + log K9K16 + log nCa2+
331
+ n0
332
+
333
+ . (17)
334
+ This equation demonstrates the reason for the slope of
335
+ approximately −0.5 for the upper limit of ocean pH as a
336
+ function of the logarithm (base 10) of PCO2. Because K9
337
+ and K16 are constants at a fixed T and P, pH becomes
338
+ a function of only PCO2 and nCa2+ in Equation 17. As a
339
+ function of PCO2, nCa2+ at the limit of carbonate satura-
340
+ tion varies between ∼0.1 m−3 (at PCO2 = 0.01 µbar) and
341
+ ∼6 m−3 (at PCO2 = 0.3 bar). This additional increase
342
+ in nCa2+ of less than two orders of magnitude over seven
343
+ orders of magnitude increase in PCO2, makes the slope
344
+ of ocean pH slightly steeper than −0.5 (see Fig. A1).
345
+ Using nCa2+ from the numerical solution in Equation 17
346
+ results in a semi-analytical solution matching with the
347
+ numerical solution until PCO2 = 0.1 bar, beyond which
348
+ non-ideal effects accounted in the numerical solution ex-
349
+ hibit a small deviation from the analytical equation.
350
+ Lower limit of ocean pH. In the absence of divalent
351
+ cations in ocean, the ocean pH is largely governed by the
352
+ conversion of CO2 to protons (Equation 3). For PCO2 >
353
+ 1 µbar, the ocean is acidic, where the number density of
354
+ H+ is larger than that of OH− and the number density
355
+ of HCO−
356
+ 3 is larger than CO2−
357
+ 3
358
+ (bicarbonate-carbonate-
359
+ water equilibria, Wolf-Gladrow et al. 2007). Therefore,
360
+ the charge balance equation can be approximated as
361
+ nH+ = nHCO−
362
+ 3 .
363
+ (18)
364
+ In terms of the equilibrium constant of Equation 3, this
365
+ leads to (Appendix A)
366
+ pH = −1
367
+ 2 (log PCO2 + log K3) .
368
+ (19)
369
+ At a fixed T and P, K3 is constant and thus the ocean
370
+ pH exhibits a slope of −0.5 for PCO2 > 1 µbar (Fig. A1).
371
+ For PCO2 < 1 µbar, the analytical solution does not
372
+ hold because the number density of OH− is significant
373
+ enough to make the charge balance approximation in
374
+ Equation 18 invalid.
375
+ The lower limit of ocean pH is
376
+ independent of the Ca, Mg or Fe systems considered.
377
+ 10
378
+ 8
379
+ 10
380
+ 7
381
+ 10
382
+ 6
383
+ 10
384
+ 5
385
+ 10
386
+ 4
387
+ 10
388
+ 3
389
+ 10
390
+ 2
391
+ 10
392
+ 1
393
+ PCO2 [bar]
394
+ 4
395
+ 5
396
+ 6
397
+ 7
398
+ 8
399
+ 9
400
+ 10
401
+ 11
402
+ Ocean pH
403
+ (a)
404
+ Carbon Cycle
405
+ No Carbon Cycle
406
+ Modern
407
+ Earth pH
408
+ Forbidden
409
+ Forbidden
410
+ Ca
411
+ nCa, tot = fW(PCO2)
412
+ 10
413
+ 8
414
+ 10
415
+ 7
416
+ 10
417
+ 6
418
+ 10
419
+ 5
420
+ 10
421
+ 4
422
+ 10
423
+ 3
424
+ 10
425
+ 2
426
+ 10
427
+ 1
428
+ PCO2 [bar]
429
+ 4
430
+ 5
431
+ 6
432
+ 7
433
+ 8
434
+ 9
435
+ 10
436
+ 11
437
+ Ocean pH
438
+ (b)
439
+ Carbon Cycle
440
+ No Carbon Cycle
441
+ Modern
442
+ Earth pH
443
+ Forbidden
444
+ Forbidden
445
+ Mg
446
+ nMg, tot = fW(PCO2)
447
+ 10
448
+ 8
449
+ 10
450
+ 7
451
+ 10
452
+ 6
453
+ 10
454
+ 5
455
+ 10
456
+ 4
457
+ 10
458
+ 3
459
+ 10
460
+ 2
461
+ 10
462
+ 1
463
+ PCO2 [bar]
464
+ 4
465
+ 5
466
+ 6
467
+ 7
468
+ 8
469
+ 9
470
+ 10
471
+ 11
472
+ Ocean pH
473
+ (c)
474
+ Carbon Cycle
475
+ No Carbon Cycle
476
+ Modern
477
+ Earth pH
478
+ Forbidden
479
+ Forbidden
480
+ Fe
481
+ nFe, tot = fW(PCO2)
482
+ Figure 2.
483
+ Sensitivity of ocean pH to PCO2 at T = 288 K for
484
+ pure (a) Ca, (b) Mg, (c) Fe systems. Upper and lower bounds
485
+ of ocean pH are represented by the blue shaded region. Pink
486
+ shaded regions are forbidden.
487
+
488
+ 5
489
+ 3. RESULTS AND DISCUSSION
490
+ We consider the ocean pH to be determined by the
491
+ chemical dissolution of atmospheric carbon dioxide in
492
+ a well-mixed ocean, which occurs at the atmosphere–
493
+ ocean interface. The chemical dissolution of CO2 is gov-
494
+ erned by the reaction between water and CO2 to produce
495
+ H+, HCO−
496
+ 3 and CO2−
497
+ 3
498
+ ions (Sect. 2). As PCO2 increases,
499
+ the ocean becomes more acidic. We consider an atmo-
500
+ spheric surface pressure of 1 bar, but allow the atmo-
501
+ spheric carbon dioxide content to vary via PCO2. Atmo-
502
+ spheric surface pressures up to 100 bar have a negligible
503
+ effect on our results and those between 100–1000 bar
504
+ exhibit a small effect (Fig. A2a).
505
+ For a given value of PCO2, the ocean pH is bounded
506
+ between two limits (Fig. 2a). The ocean pH is restricted
507
+ to a narrow range between 7–11 at PCO2 = 0.01 µbar
508
+ and 4–7 for PCO2 = 0.1 bar. These ocean pH ranges
509
+ are consistent with the inferences for Earth’s history,
510
+ transitioning from an acidic ocean during the Archean
511
+ at high PCO2 to an alkaline ocean at present-day PCO2
512
+ (Halevy & Bachan 2017; Krissansen-Totton et al. 2018).
513
+ The lower limit corresponds to the complete absence of
514
+ divalent cations and thus it is independent of the car-
515
+ bonate system under investigation (Sect. 2). The upper
516
+ limit corresponds to the saturation of calcium cations
517
+ in ocean water such that more weathering does not pro-
518
+ duce further changes in pH and simply produces more
519
+ calcite. This upper limit is buffered by the precipita-
520
+ tion of carbonates and hence it results in one solution
521
+ of ocean pH when the carbon cycle is operational for
522
+ a given carbonate system and PCO2. Both upper and
523
+ lower limits of ocean pH follow a slope of approximately
524
+ –0.5 as a function of PCO2 (see Sect.
525
+ 2.4).
526
+ Between
527
+ these two limits, the number density of calcium cations
528
+ is below the threshold to precipitate carbonates onto the
529
+ ocean floor; thus, the carbon cycle is not operational.
530
+ Due to their high condensation temperatures, the
531
+ relative abundances of refractory elements observed in
532
+ the photosphere of stars are expected to be mirrored
533
+ in the rocky exoplanets they host (Bond et al. 2010;
534
+ Thiabaud et al. 2015).
535
+ For example, the calcium-to-
536
+ magnesium ratio of the solar photosphere and Earth are
537
+ 0.062 and 0.066, respectively (Lodders 2003; Elser et al.
538
+ 2012). The relative abundances of Ca, Mg and Fe, mea-
539
+ sured from the spectra of stars, vary by up to an or-
540
+ der of magnitude. For example, Ca/Mg=0.02–0.2 and
541
+ Ca/Fe=0.04–0.2 in the Hypatia catalogue of more than
542
+ 7000 stars (Hinkel et al. 2014). Furthermore, carbonates
543
+ involving Mg and Fe are known to have formed during
544
+ Earth’s history: e.g., magnesite (MgCO3) and siderite
545
+ (FeCO3); these carbonates have dissolution properties
546
+ that differ from those of calcite.
547
+ Siderite could have
548
+ 10
549
+ 8
550
+ 10
551
+ 7
552
+ 10
553
+ 6
554
+ 10
555
+ 5
556
+ 10
557
+ 4
558
+ 10
559
+ 3
560
+ 10
561
+ 2
562
+ 10
563
+ 1
564
+ PCO2 [bar]
565
+ 280
566
+ 300
567
+ 320
568
+ 340
569
+ 360
570
+ T [K]
571
+ (a)
572
+ nCa, tot = 100 m
573
+ 3
574
+ nCa, tot = 1 m
575
+ 3
576
+ Carbon Cycle
577
+ No Carbon Cycle
578
+ (cations consumed
579
+ by silicates)
580
+ No Carbon Cycle
581
+ (too little CO2)
582
+ No Carbon Cycle
583
+ (too acidic)
584
+ nCa, tot = fW(PCO2, T)
585
+ Ca-CCD
586
+ 1
587
+ 2
588
+ 4
589
+ 10
590
+ 20
591
+ 40
592
+ CCD [km]
593
+ 10
594
+ 8
595
+ 10
596
+ 7
597
+ 10
598
+ 6
599
+ 10
600
+ 5
601
+ 10
602
+ 4
603
+ 10
604
+ 3
605
+ 10
606
+ 2
607
+ 10
608
+ 1
609
+ PCO2 [bar]
610
+ 280
611
+ 300
612
+ 320
613
+ 340
614
+ 360
615
+ T [K]
616
+ (b)
617
+ nMg, tot = 100 m
618
+ 3
619
+ nMg, tot = 1 m
620
+ 3
621
+ Carbon Cycle
622
+ No Carbon Cycle
623
+ (cations consumed
624
+ by slicates)
625
+ No Carbon Cycle
626
+ (too little CO2)
627
+ No Carbon Cycle
628
+ (too acidic)
629
+ nMg, tot = fW(PCO2, T)
630
+ Mg-CCD
631
+ 1
632
+ 2
633
+ 4
634
+ 10
635
+ 20
636
+ 40
637
+ CCD [km]
638
+ 10
639
+ 8
640
+ 10
641
+ 7
642
+ 10
643
+ 6
644
+ 10
645
+ 5
646
+ 10
647
+ 4
648
+ 10
649
+ 3
650
+ 10
651
+ 2
652
+ 10
653
+ 1
654
+ PCO2 [bar]
655
+ 280
656
+ 300
657
+ 320
658
+ 340
659
+ 360
660
+ T [K]
661
+ (c)
662
+ nFe, tot = 100 m
663
+ 3
664
+ nFe, tot = 1 m
665
+ 3
666
+ Carbon Cycle
667
+ No Carbon Cycle
668
+ (cations consumed
669
+ by slicates)
670
+ nFe, tot = fW(PCO2, T)
671
+ Fe-CCD
672
+ 1
673
+ 2
674
+ 4
675
+ 10
676
+ 20
677
+ 40
678
+ CCD [km]
679
+ Figure 3. Carbonate compensation depth (CCD) as a func-
680
+ tion of PCO2 and T (Patm = 1 bar) for (a) Ca, (b) Mg and
681
+ (c) Fe systems.
682
+ Gray contours represent the weathering-
683
+ dependent cation number density as a function of PCO2 and
684
+ T (Eq. 13). Gray disc denotes modern Earth PCO2 and T.
685
+
686
+ 6
687
+ 10
688
+ 8
689
+ 10
690
+ 7
691
+ 10
692
+ 6
693
+ 10
694
+ 5
695
+ 10
696
+ 4
697
+ 10
698
+ 3
699
+ 10
700
+ 2
701
+ 10
702
+ 1
703
+ PCO2 [bar]
704
+ 10
705
+ 1
706
+ 100
707
+ 101
708
+ n [m
709
+ 3]
710
+ (a)
711
+ nCa, tot = fW(PCO2)
712
+ Ca Partitioning
713
+ Ca++
714
+ Calcite
715
+ Silicates
716
+ 10
717
+ 8
718
+ 10
719
+ 7
720
+ 10
721
+ 6
722
+ 10
723
+ 5
724
+ 10
725
+ 4
726
+ 10
727
+ 3
728
+ 10
729
+ 2
730
+ 10
731
+ 1
732
+ PCO2 [bar]
733
+ 10
734
+ 1
735
+ 100
736
+ 101
737
+ n [m
738
+ 3]
739
+ (b)
740
+ nMg, tot = fW(PCO2)
741
+ Mg Partitioning
742
+ Mg++
743
+ Magnesite
744
+ Silicates
745
+ 10
746
+ 8
747
+ 10
748
+ 7
749
+ 10
750
+ 6
751
+ 10
752
+ 5
753
+ 10
754
+ 4
755
+ 10
756
+ 3
757
+ 10
758
+ 2
759
+ 10
760
+ 1
761
+ PCO2 [bar]
762
+ 10
763
+ 1
764
+ 100
765
+ 101
766
+ n [m
767
+ 3]
768
+ (c)
769
+ nFe, tot = fW(PCO2)
770
+ Fe Partitioning
771
+ Fe++
772
+ Siderite
773
+ Silicates
774
+ Figure 4.
775
+ Partitioning of (a) Ca, (b) Mg and (c) Fe in
776
+ aqueous, carbonate and silicate phases as a function of PCO2
777
+ at T = 310 K (Patm = Poc = 1 bar) in pure Ca, Mg and Fe
778
+ systems, respectively.
779
+ played a key role in locking up CO2 in carbonates on
780
+ Earth during the Archean (Rye et al. 1995; Sverjensky
781
+ & Lee 2010). We calculate ocean pH for the pure Mg and
782
+ Fe systems in addition to the Ca system (Fig. 2b,c). The
783
+ upper limit of ocean pH for a given PCO2 varies when
784
+ considering systems with purely Ca, Mg or Fe as the
785
+ source of weathering cations. The upper limit of ocean
786
+ pH for the Mg system is only 0.2 higher than for the Ca
787
+ system, whereas it is more than unity lower for the Fe
788
+ system.
789
+ For PCO2 < 10 µbar, ocean chemistry and hence
790
+ the CCD is sensitive to the addition of aqueous silica
791
+ (SiO2) in the ocean (Fig. 3). Silica is another product
792
+ of silicate weathering, which enables the locking up of
793
+ cations in silicate minerals instead of carbonate minerals
794
+ (Walker et al. 1981; Hakim et al. 2021). For instance,
795
+ for T > 300 K and PCO2 < 0.1 µbar in the Ca sys-
796
+ tem in the presence of aqueous silica, silicates impinge
797
+ on the stability of calcite (Fig. 4a) and prevent carbon-
798
+ ate precipitation at all depths (Fig. 3a).
799
+ In contrast,
800
+ when no silica is present in the ocean for T > 300 K
801
+ and PCO2 < 0.1 µbar, calcite is stable (Fig. B2a) and
802
+ deep CCDs are produced (Fig. B1a), thereby increasing
803
+ the parameter-space where the carbon cycle is stable.
804
+ Similarly, in the Mg and Fe systems, silicates are more
805
+ stable than carbonates for PCO2 < 10 µbar (Fig. 4b,c).
806
+ PCO2 > 10 µbar favours the thermodynamic stability of
807
+ carbonates over silicates.
808
+ Carbon cycle box models of exoplanets often omit self-
809
+ consistent modelling of ocean chemistry and precipita-
810
+ tion of carbonates. Carbonate precipitation is implicitly
811
+ assumed to persist and is not expected to be a bottle-
812
+ neck for carbon cycling.
813
+ Our ocean chemistry model
814
+ can be incorporated directly into carbon cycle box mod-
815
+ els for exoplanets, which can couple via key parameters,
816
+ PCO2, T, and the carbonate chemistry. Thermochemi-
817
+ cal equilibrium calculations of our ocean model can be
818
+ used to determine the carbon fluxes into or out of the
819
+ near-surface reservoirs.
820
+ The carbon cycle box models
821
+ can also be informed of the effect of ocean chemistry
822
+ and ocean depth on the efficiency of carbon degassing
823
+ and recycling.
824
+ Upcoming observations of terrestrial exoplanets from
825
+ the James Webb Space Telescope, Atmospheric Remote-
826
+ sensing Infrared Exoplanet Large-survey and Extremely
827
+ Large Telescopes will put constraints on their atmo-
828
+ spheric composition, for instance, the volume mixing
829
+ ratio of atmospheric carbon dioxide (PCO2/P). Deter-
830
+ mining the partial pressure of carbon dioxide (PCO2)
831
+ requires the atmospheric surface pressure (P) which is
832
+ not easily constrained. Nonetheless, our thermodynamic
833
+ calculations provide strong constraints on ocean chem-
834
+
835
+ 7
836
+ istry in the presence or absence of magnesium, calcium
837
+ or iron carbonates; the relative abundances of these
838
+ carbonate-forming elements in planetary systems can
839
+ be deduced from observations of stellar photospheres.
840
+ Our results suggest that the carbon cycle will oper-
841
+ ate robustly on chemically-diverse terrestrial exoplanets
842
+ exhibiting silicate weathering.
843
+ This implies that the
844
+ search for life from exoplanets with temperate climates
845
+ or biospheres will benefit by broadening the target list
846
+ to planets that are more chemically diverse than Earth.
847
+ We acknowledge financial support from the European
848
+ Research Council via Consolidator Grant (ERC-2017-
849
+ CoG-771620-EXOKLEIN, awarded to K. Heng) and the
850
+ Center for Space and Habitability, University of Bern.
851
+ We thank Allan Leal for the support with Reaktoro.
852
+ DATA AVAILABILITY
853
+ All data generated or analysed during this study are
854
+ included in the published article.
855
+ CODE AVAILABILITY
856
+ OCRA (Ocean Chemistry with Reaktoro And beyond):
857
+ the open-source code developed in this work is hosted
858
+ at https://github.com/kaustubhhakim/ocra. OCRA v1.0
859
+ was used in this study and is also available on Zenodo
860
+ (Hakim 2022).
861
+ Software:
862
+ numpy (Harris et al. 2020), scipy (Vir-
863
+ tanen et al. 2020), pandas (The pandas development
864
+ team 2020), astropy (Astropy Collaboration et al.
865
+ 2013, 2022), matplotlib (Hunter 2007), Reaktoro (Leal
866
+ 2015)
867
+ APPENDIX
868
+ A. ANALYTICAL SOLUTION OF OCEAN PH AND
869
+ P–T SENSITIVITY
870
+ The analytical solution for the upper limit of ocean
871
+ pH is derived from the relations between the equilibrium
872
+ constants and reactants and products (assuming water
873
+ activity to be unity in diluted solutions) of reactions
874
+ described by Equations 9 and 16,
875
+ K9 =
876
+ n2
877
+ 0
878
+ nCa2+nCO2−
879
+ 3
880
+ ,
881
+ (A1)
882
+ K16 =
883
+ n2
884
+ H+nCO2−
885
+ 3
886
+ PCO2n3
887
+ 0
888
+ .
889
+ (A2)
890
+ By eliminating the carbonate ion number density from
891
+ these two equations, proton number density is
892
+ nH+
893
+ n0
894
+ =
895
+
896
+ PCO2K9K16
897
+ nCa2+
898
+ n0
899
+ �1/2
900
+ (A3)
901
+ Because the pH is given by
902
+ pH = − log(nH+/n0),
903
+ (A4)
904
+ the analytical upper limit of ocean pH is Equation 17.
905
+ The analytical solution for the lower limit of ocean
906
+ pH is derived from the the equilibrium constant of the
907
+ reaction described by Equation 3,
908
+ K3 =
909
+ nH+nHCO−
910
+ 3
911
+ PCO2n2
912
+ 0
913
+ .
914
+ (A5)
915
+ Then the proton number density is
916
+ nH+
917
+ n0
918
+ = K3PCO2n0
919
+ nHCO−
920
+ 3
921
+ (A6)
922
+ 10
923
+ 8
924
+ 10
925
+ 7
926
+ 10
927
+ 6
928
+ 10
929
+ 5
930
+ 10
931
+ 4
932
+ 10
933
+ 3
934
+ 10
935
+ 2
936
+ 10
937
+ 1
938
+ PCO2 [bar]
939
+ 4
940
+ 5
941
+ 6
942
+ 7
943
+ 8
944
+ 9
945
+ 10
946
+ 11
947
+ Ocean pH
948
+ Carbon Cycle
949
+ No Carbon Cycle
950
+ Modern
951
+ Earth pH
952
+ Forbidden
953
+ Ca
954
+ Up (numerical)
955
+ Up (semi-analytical)
956
+ Up (ana., nCa2 + = 1 m
957
+ 3)
958
+ Low (numerical)
959
+ Low (analytical)
960
+ Figure A1. Numerical, analytical and semi-analytical so-
961
+ lutions of the upper and lower limits of ocean pH in the Ca
962
+ system.
963
+ Thus, the lower limit of ocean pH is given by Equation
964
+ 19.
965
+ The analytical solutions of upper and lower limits of
966
+ ocean pH as a function of PCO2 result in a slope of –0.5
967
+ (Fig. A1). Pressure and temperature have a negligible
968
+ effect on ocean pH (Fig. A2).
969
+
970
+ 8
971
+ 100
972
+ 101
973
+ 102
974
+ 103
975
+ P [bar]
976
+ 4
977
+ 5
978
+ 6
979
+ 7
980
+ 8
981
+ 9
982
+ 10
983
+ 11
984
+ Ocean pH
985
+ (a)
986
+ Carbon Cycle
987
+ No Carbon Cycle
988
+ Modern
989
+ Earth pH
990
+ Forbidden
991
+ Forbidden
992
+ Ca
993
+ nCa, tot = fW(PCO2)
994
+ 280
995
+ 300
996
+ 320
997
+ 340
998
+ 360
999
+ T [K]
1000
+ 4
1001
+ 5
1002
+ 6
1003
+ 7
1004
+ 8
1005
+ 9
1006
+ 10
1007
+ 11
1008
+ Ocean pH
1009
+ (b)
1010
+ Carbon Cycle
1011
+ No Carbon Cycle
1012
+ Modern
1013
+ Earth pH
1014
+ Forbidden
1015
+ Forbidden
1016
+ Ca
1017
+ nCa, tot = fW(PCO2)
1018
+ Figure A2. The sensitivity of ocean pH to (a) P and (b) T
1019
+ in the Ca-system.
1020
+ B. CCD WITHOUT SILICATE PRECIPITATION
1021
+ When no silicates are allowed to precipitate, CCDs for
1022
+ the Ca, Mg and Fe systems become deeper for PCO2 <
1023
+ 1 µbar (Fig. B1). This is reflected in the phase stability
1024
+ plots in Fig. B2.
1025
+ 10
1026
+ 8
1027
+ 10
1028
+ 7
1029
+ 10
1030
+ 6
1031
+ 10
1032
+ 5
1033
+ 10
1034
+ 4
1035
+ 10
1036
+ 3
1037
+ 10
1038
+ 2
1039
+ 10
1040
+ 1
1041
+ PCO2 [bar]
1042
+ 280
1043
+ 300
1044
+ 320
1045
+ 340
1046
+ 360
1047
+ T [K]
1048
+ (a)
1049
+ nCa, tot = 100 m
1050
+ 3
1051
+ nCa, tot = 1 m
1052
+ 3
1053
+ Carbon Cycle
1054
+ No Carbon Cycle
1055
+ (too little CO2)
1056
+ No Carbon Cycle
1057
+ (too acidic)
1058
+ nCa, tot = fW(PCO2, T)
1059
+ Ca CCD
1060
+ 1
1061
+ 2
1062
+ 4
1063
+ 10
1064
+ 20
1065
+ 40
1066
+ CCD [km]
1067
+ 10
1068
+ 8
1069
+ 10
1070
+ 7
1071
+ 10
1072
+ 6
1073
+ 10
1074
+ 5
1075
+ 10
1076
+ 4
1077
+ 10
1078
+ 3
1079
+ 10
1080
+ 2
1081
+ 10
1082
+ 1
1083
+ PCO2 [bar]
1084
+ 280
1085
+ 300
1086
+ 320
1087
+ 340
1088
+ 360
1089
+ T [K]
1090
+ (b)
1091
+ nMg, tot = 100 m
1092
+ 3
1093
+ nMg, tot = 1 m
1094
+ 3
1095
+ Carbon Cycle
1096
+ No Carbon Cycle
1097
+ (too little CO2)
1098
+ No Carbon Cycle
1099
+ (too acidic)
1100
+ nMg, tot = fW(PCO2, T)
1101
+ Mg CCD
1102
+ 1
1103
+ 2
1104
+ 4
1105
+ 10
1106
+ 20
1107
+ 40
1108
+ CCD [km]
1109
+ 10
1110
+ 8
1111
+ 10
1112
+ 7
1113
+ 10
1114
+ 6
1115
+ 10
1116
+ 5
1117
+ 10
1118
+ 4
1119
+ 10
1120
+ 3
1121
+ 10
1122
+ 2
1123
+ 10
1124
+ 1
1125
+ PCO2 [bar]
1126
+ 280
1127
+ 300
1128
+ 320
1129
+ 340
1130
+ 360
1131
+ T [K]
1132
+ (c)
1133
+ nFe, tot = 100 m
1134
+ 3
1135
+ nFe, tot = 1 m
1136
+ 3
1137
+ Carbon Cycle
1138
+ nFe, tot = fW(PCO2, T)
1139
+ Fe CCD
1140
+ 1
1141
+ 2
1142
+ 4
1143
+ 10
1144
+ 20
1145
+ 40
1146
+ CCD [km]
1147
+ Figure B1. Same as Fig. 3 but with no silica nSiO2,tot = 0.
1148
+
1149
+ 9
1150
+ 10
1151
+ 8
1152
+ 10
1153
+ 7
1154
+ 10
1155
+ 6
1156
+ 10
1157
+ 5
1158
+ 10
1159
+ 4
1160
+ 10
1161
+ 3
1162
+ 10
1163
+ 2
1164
+ 10
1165
+ 1
1166
+ PCO2 [bar]
1167
+ 10
1168
+ 1
1169
+ 100
1170
+ 101
1171
+ n [m
1172
+ 3]
1173
+ (a)
1174
+ nCa, tot = fW(PCO2)
1175
+ Ca Partitioning
1176
+ Ca++
1177
+ Calcite
1178
+ Silicates
1179
+ 10
1180
+ 8
1181
+ 10
1182
+ 7
1183
+ 10
1184
+ 6
1185
+ 10
1186
+ 5
1187
+ 10
1188
+ 4
1189
+ 10
1190
+ 3
1191
+ 10
1192
+ 2
1193
+ 10
1194
+ 1
1195
+ PCO2 [bar]
1196
+ 10
1197
+ 1
1198
+ 100
1199
+ 101
1200
+ n [m
1201
+ 3]
1202
+ (b)
1203
+ nMg, tot = fW(PCO2)
1204
+ Mg Partitioning
1205
+ Mg++
1206
+ Magnesite
1207
+ Silicates
1208
+ 10
1209
+ 8
1210
+ 10
1211
+ 7
1212
+ 10
1213
+ 6
1214
+ 10
1215
+ 5
1216
+ 10
1217
+ 4
1218
+ 10
1219
+ 3
1220
+ 10
1221
+ 2
1222
+ 10
1223
+ 1
1224
+ PCO2 [bar]
1225
+ 10
1226
+ 1
1227
+ 100
1228
+ 101
1229
+ n [m
1230
+ 3]
1231
+ (c)
1232
+ nFe, tot = fW(PCO2)
1233
+ Fe Partitioning
1234
+ Fe++
1235
+ Siderite
1236
+ Silicates
1237
+ Figure B2. Same as Fig. 4 but with no silica nSiO2,tot = 0.
1238
+
1239
+ 10
1240
+ REFERENCES
1241
+ Astropy Collaboration, Robitaille, T. P., Tollerud, E. J.,
1242
+ et al. 2013, A&A, 558, A33,
1243
+ doi: 10.1051/0004-6361/201322068
1244
+ Astropy Collaboration, Price-Whelan, A. M., Lim, P. L.,
1245
+ et al. 2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74
1246
+ Berger, W. H., Adelseck, C. G., J., & Mayer, L. A. 1976,
1247
+ J. Geophys. Res., 81, 2617,
1248
+ doi: 10.1029/JC081i015p02617
1249
+ Berner, R. A. 2004, The Phanerozoic carbon cycle: CO2
1250
+ and O2 (Oxford University Press on Demand)
1251
+ Bond, J. C., O’Brien, D. P., & Lauretta, D. S. 2010, ApJ,
1252
+ 715, 1050, doi: 10.1088/0004-637X/715/2/1050
1253
+ Catling, D. C., & Kasting, J. F. 2017, Atmospheric
1254
+ Evolution on Inhabited and Lifeless Worlds (Cambridge
1255
+ University Press)
1256
+ Drummond, S. 1981, PhD thesis, Pennsylvania State
1257
+ University
1258
+ Elser, S., Meyer, M. R., & Moore, B. 2012, Icarus, 221, 859,
1259
+ doi: 10.1016/j.icarus.2012.09.016
1260
+ Foley, B. J. 2015, ApJ, 812, 36,
1261
+ doi: 10.1088/0004-637X/812/1/36
1262
+ Hakim, K. 2022, kaustubhhakim/ocra: OCRA, v1.0,
1263
+ Zenodo, doi: 10.5281/zenodo.7380526.
1264
+ https://doi.org/10.5281/zenodo.7380526
1265
+ Hakim, K., Bower, D. J., Tian, M., et al. 2021, PSJ, 2, 49,
1266
+ doi: 10.3847/PSJ/abe1b8
1267
+ Halevy, I., & Bachan, A. 2017, Science, 355, 1069,
1268
+ doi: 10.1126/science.aal4151
1269
+ Harris, C. R., Millman, K. J., van der Walt, S. J., et al.
1270
+ 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2
1271
+ Helgeson, H. C., Kirkham, D. H., & Flowers, G. C. 1981,
1272
+ American Journal of Science, 281, 1249,
1273
+ doi: 10.2475/ajs.281.10.1249
1274
+ Hinkel, N. R., Timmes, F. X., Young, P. A., Pagano, M. D.,
1275
+ & Turnbull, M. C. 2014, AJ, 148, 54,
1276
+ doi: 10.1088/0004-6256/148/3/54
1277
+ Holland, H. D. 1978, The chemistry of the atmosphere and
1278
+ oceans (New York, NY (USA) Wiley-Interscience)
1279
+ Hunter, J. D. 2007, Computing in Science & Engineering, 9,
1280
+ 90, doi: 10.1109/MCSE.2007.55
1281
+ Johnson, J. W., Oelkers, E. H., & Helgeson, H. C. 1992,
1282
+ Computers and Geosciences, 18, 899,
1283
+ doi: 10.1016/0098-3004(92)90029-Q
1284
+ Kasting, J. F., Whitmire, D. P., & Reynolds, R. T. 1993,
1285
+ Icarus, 101, 108, doi: 10.1006/icar.1993.1010
1286
+ Kelemen, P. B., & Manning, C. E. 2015, Proceedings of the
1287
+ National Academy of Science, 112, E3997,
1288
+ doi: 10.1073/pnas.1507889112
1289
+ Krissansen-Totton, J., Arney, G. N., & Catling, D. C. 2018,
1290
+ Proceedings of the National Academy of Science, 115,
1291
+ 4105, doi: 10.1073/pnas.1721296115
1292
+ Leal, A. M. M. 2015, Reaktoro: An open-source unified
1293
+ framework for modeling chemically reactive systems, 2.0.
1294
+ https://reaktoro.org
1295
+ Leal, A. M. M., Kulik, D. A., Smith, W. R., & Saar, M. O.
1296
+ 2017, Pure and Applied Chemistry, 89, 597,
1297
+ doi: doi:10.1515/pac-2016-1107
1298
+ Leroy, C. C., & Parthiot, F. 1998, Acoustical Society of
1299
+ America Journal, 103, 1346, doi: 10.1121/1.421275
1300
+ Lodders, K. 2003, ApJ, 591, 1220, doi: 10.1086/375492
1301
+ Peng, D.-Y., & Robinson, D. B. 1976, Industrial &
1302
+ Engineering Chemistry Fundamentals, 15, 59
1303
+ Ridgwell, A., & Zeebe, R. E. 2005, Earth and Planetary
1304
+ Science Letters, 234, 299, doi: 10.1016/j.epsl.2005.03.006
1305
+ Rye, R., Kuo, P. H., & Holland, H. D. 1995, Nature, 378,
1306
+ 603, doi: 10.1038/378603a0
1307
+ Sleep, N. H., & Zahnle, K. 2001, J. Geophys. Res., 106,
1308
+ 1373, doi: 10.1029/2000JE001247
1309
+ Sverjensky, D., & Lee, N. 2010, Elements, 6, 31,
1310
+ doi: 10.2113/gselements.6.1.31
1311
+ The pandas development team. 2020, pandas-dev/pandas:
1312
+ Pandas, latest, Zenodo, doi: 10.5281/zenodo.3509134.
1313
+ https://doi.org/10.5281/zenodo.3509134
1314
+ Thiabaud, A., Marboeuf, U., Alibert, Y., Leya, I., &
1315
+ Mezger, K. 2015, A&A, 580, A30,
1316
+ doi: 10.1051/0004-6361/201525963
1317
+ Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020,
1318
+ Nature Methods, 17, 261,
1319
+ doi: https://doi.org/10.1038/s41592-019-0686-2
1320
+ Walker, J. C. G., Hays, P. B., & Kasting, J. F. 1981,
1321
+ J. Geophys. Res., 86, 9776,
1322
+ doi: 10.1029/JC086iC10p09776
1323
+ Wolf-Gladrow, D. A., Zeebe, R. E., Klaas, C., K¨ortzinger,
1324
+ A., & Dickson, A. G. 2007, Marine Chemistry, 106, 287,
1325
+ doi: https://doi.org/10.1016/j.marchem.2007.01.006
1326
+ Zeebe, R. E. 2012, Annual Review of Earth and Planetary
1327
+ Sciences, 40, 141,
1328
+ doi: 10.1146/annurev-earth-042711-105521
1329
+ Zeebe, R. E., & Westbroek, P. 2003, Geochemistry,
1330
+ Geophysics, Geosystems, 4, 1104,
1331
+ doi: 10.1029/2003GC000538
1332
+ Zimmer, K., Zhang, Y., Lu, P., et al. 2016, Computers and
1333
+ Geosciences, 90, 97, doi: 10.1016/j.cageo.2016.02.013
1334
+
BdE0T4oBgHgl3EQfyAIA/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
BdE3T4oBgHgl3EQftAtl/content/tmp_files/2301.04672v1.pdf.txt ADDED
@@ -0,0 +1,897 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Astronomy & Astrophysics manuscript no. VMS
2
+ ©ESO 2023
3
+ January 13, 2023
4
+ Clues on the presence and segregation of very massive stars in
5
+ the Sunburst Lyman-continuum cluster at z=2.37⋆
6
+ U. Meštri´c1,⋆⋆, E. Vanzella1, A. Upadhyaya2, F. Martins3, R. Marques-Chaves2, D. Schaerer2, 4,
7
+ J. Guibert2, A. Zanella5, C. Grillo6, 7, P. Rosati8, F. Calura1, G.B. Caminha9, 10, A. Bolamperti5, 11, 12,
8
+ M. Meneghetti1, P. Bergamini1, 6, A. Mercurio13, 14, M. Nonino15, R. Pascale1
9
+ 1 INAF – OAS, Osservatorio di Astrofisica e Scienza dello Spazio di Bologna, via Gobetti 93/3, I-40129 Bologna, Italy
10
+ 2 Geneva Observatory, Department of Astronomy, University of Geneva, Chemin Pegasi 51, CH-1290 Versoix, Switzerland
11
+ 3 LUPM, Université de Montpellier, CNRS, Place Eugène Bataillon, F-34095 Montpellier, France
12
+ 4 CNRS, IRAP, 14 Avenue E. Belin, 31400 Toulouse, France
13
+ 5 INAF – Osservatorio Astronomico di Padova, Vicolo dell’Osservatorio 5, 35122, Padova, Italy
14
+ 6 Dipartimento di Fisica, Università degli Studi di Milano, via Celoria 16, I-20133 Milano, Italy
15
+ 7 INAF – IASF Milano, via A. Corti 12, I-20133 Milano, Italy
16
+ 8 Dipartimento di Fisica e Scienze della Terra, Università degli Studi di Ferrara, via Saragat 1, I-44122 Ferrara, Italy
17
+ 9 Technical University of Munich, TUM School of Natural Sciences, Department of Physics, James-Franck-Str 1, 85748 Garching,
18
+ Germany
19
+ 10 Max-Planck-Institut für Astrophysik, Karl-Schwarzschild-Str. 1, D-85748 Garching, Germany
20
+ 11 Dipartimento di Fisica e Astronomia, Università degli Studi di Padova, Vicolo dell’Osservatorio 3, I-35122 Padova, Italy
21
+ 12 European Southern Observatory, Karl-Schwarzschild-Strasse 2, D-85748 Garching bei München, Germany
22
+ 13 Dipartimento di Fisica “E.R. Caianiello”, Università Degli Studi di Salerno, Via Giovanni Paolo II, I–84084 Fisciano (SA), Italy
23
+ 14 INAF – INAF - Osservatorio Astronomico di Capodimonte, Via Moiariello 16, 80131 Napoli, Italy
24
+ 15 INAF – Osservatorio Astronomico di Trieste, via G. B. Tiepolo 11, I-34143, Trieste, Italy
25
+ ABSTRACT
26
+ We report on the identification of very massive stars (VMS, mass > 100 M⊙) possibly segregated in the center of the young massive
27
+ star cluster at z=2.37 hosted in the Sunburst lensed galaxy. Such a result is based on two pieces of evidence: (1) the VLT/MUSE
28
+ spectra of several multiple images of the same star cluster show key spectral signatures of VMS, like the Heiiλ1640 broad emission,
29
+ Nivλ1486 emission and Nivλ1720 P-Cygni profile. In particular, Heiiλ1640 is broad (∼ 1610 ± 300 km s−1) with an equivalent width
30
+ of 3Å and shows an asymmetric profile. Such features require an extremely young (∼ 2.5 Myr) stellar population component with
31
+ masses of the stars exceeding 100 M⊙. Assuming a Salpeter IMF and BPASS models for normal massive stars, the observed spectral
32
+ features require ∼400 VMS; (2) the same star cluster is detected at S/N ∼ 100 in the LyC domain (λ < 900Å). The LyC emission
33
+ emerges from a region with a radius at least 2 times smaller than what is observed at 1700Å (independently from magnification)
34
+ and is located in the center of the cluster. In absolute scales, after de-lensing, the effective radii are Reff[LyC] ∼ 4.7 ± 1.5 pc and
35
+ Reff[1700] = 7.8 ± 1.4 pc. The LyC radiation is mainly produced by hot and massive stars, implying that their spatial distribution
36
+ (including VMS) is preferentially more confined in the central parts of the cluster. Approximately 400 VMS hosted by a cluster of
37
+ ∼ 107 M⊙ are producing ∼15% of the escaping LyC photons, while the rest is produced from other massive early-type stars.
38
+ Key words. galaxies: high-redshift – galaxies: star formation – galaxies: ISM – galaxies: star clusters: general – gravitational lensing:
39
+ strong – galaxies: individual: Sunburst galaxy.
40
+ 1. Introduction
41
+ For many years the existence and occurrence of very massive
42
+ stars (VMS) was mostly associated with the early Universe and
43
+ metal-free environments in the context of the so-called Popula-
44
+ tion III stars (e.g. Abel et al. 2002). VMS are short-lived stars ∼
45
+ 2 – 3 Myr (e.g. Yusof et al. 2013) with mass M > 100 M⊙ (Vink
46
+ et al. 2015) and predominantly populate the central regions of
47
+ young massive star clusters (within the core radius rc ∼ 0.1−0.2
48
+ pc, Portegies Zwart et al. 2010). Due to their narrow lifetime,
49
+ ⋆ Based on observations collected at the European Southern Observa-
50
+ tory for Astronomical research in the Southern Hemisphere under ESO
51
+ programmes DDT MUSE program ID 107.22SK.001 (PI E. Vanzella),
52
+ X-Shooter program ID 0103.A-0688 (PI E. Vanzella) and DDT MUSE
53
+ program ID 297.A-5012(A) (PI Aghanim).
54
+ ⋆⋆ E-mail: uros.mestric@inaf.it
55
+ studies of VMS in Milky Way star clusters is limited only to
56
+ few targets, for example, the Arches cluster (Martins et al. 2008)
57
+ or NGC3603 (Crowther et al. 2010). Individual VMS have been
58
+ investigated in the local Universe, with high spatial resolution,
59
+ thanks to the Hubble Space Telescope (HST, Cignoni et al. 2015;
60
+ Crowther et al. 2016; Calzetti et al. 2015; Smith et al. 2016,
61
+ 2020; Brands et al. 2022). Very massive stars with masses above
62
+ 100 M⊙ are recognized as objects with significant impact on the
63
+ evolution of early galaxies, influencing their chemical enrich-
64
+ ment and star formation through feedback (e.g. Goswami et al.
65
+ 2021). Therefore, extending upper masses beyond 100 M⊙ of the
66
+ current population synthesis models is essential for investigating
67
+ and understanding young massive star clusters and VMS at dif-
68
+ ferent redshifts (Smith et al. 2016; Crowther et al. 2016).
69
+ Article number, page 1 of 10
70
+ arXiv:2301.04672v1 [astro-ph.GA] 11 Jan 2023
71
+
72
+ A&A proofs: manuscript no. VMS
73
+ Despite some progress, the maximum stellar mass attained
74
+ and the conditions determining the presence of VMS remain
75
+ largely unknown. Recent observations of local star clusters re-
76
+ port initial stellar masses up to ∼ 270 M⊙ (Brands et al. 2022) in
77
+ the star cluster R136, with a cluster age of ∼1.5 Myr (Crowther
78
+ et al. 2016). Furthermore, observations of young stellar clus-
79
+ ters have revealed the presence of peculiar spectroscopic fea-
80
+ tures such as unusually strong broad Heiiλ1640 emission (with
81
+ FWHM > 1000 km s−1), which suggests the presence of VMS
82
+ in these objects (e.g., Wofford et al. 2014; Crowther et al. 2016;
83
+ Senchyna et al. 2021).
84
+ Alongside with the observations, different models are try-
85
+ ing to predict and trace the evolution, through different ages
86
+ and masses, of the various spectroscopic features characteristic
87
+ to VMS (e.g., Köhler et al. 2015; Gräfener 2021). For exam-
88
+ ple, Martins & Palacios (2022) have generated new evolution-
89
+ ary models and synthetic spectra of stars with initial masses in
90
+ the range 150 – 400 M⊙, taking into account the existence of
91
+ stellar winds stronger than typical OB-type stars produce. The
92
+ resulting models predict specific features in the UV and optical
93
+ part of the spectra, which are characteristic signatures of VMS.
94
+ The most robust ultraviolet spectral features associated to VMS
95
+ are Nivλ1486, broad Heiiλ1640 emission, and the Nivλ1720 P-
96
+ Cygni profile (Martins & Palacios 2022). Such lines are expected
97
+ to have equivalent widths spanning the interval 0.1 − 7 Å rest-
98
+ frame. High signal-to-noise spectra with well detected contin-
99
+ uum are therefore required to identify them, as shown, e.g., by
100
+ Crowther et al. (2016) in the R136 stellar cluster in the local
101
+ Universe. At cosmological distance strong gravitational lensing
102
+ is necessary to detect these faint spectral features, allowing us to
103
+ further gain in spatial resolution at tens of parsec scale and depth
104
+ (see also, Vanzella et al. 2016; Johnson et al. 2017; Rigby et al.
105
+ 2017, 2018b,a; Vanzella et al. 2017, 2021; Meštri´c et al. 2022;
106
+ Vanzella et al. 2022b).
107
+ In this paper, we present for the first time convincing spec-
108
+ troscopic evidence for the presence of VMS in a stellar cluster
109
+ at cosmological distance (z=2.37, Vanzella et al. 2020a, 2022a).
110
+ The host galaxy is dubbed Sunburst (Rivera-Thorsen et al.
111
+ 2019, 2017), and Lyman continuum (LyC) radiation is detected
112
+ from the same clumpy regions which are showing the presence
113
+ of VMS. Those massive stars in the center of the stellar cluster
114
+ are significant producers of LyC radiation and hence are the main
115
+ culprits for creating porous interstellar medium (ISM) enabling
116
+ LyC escape. For purpose of this work, we perform a compre-
117
+ hensive analysis of deep VLT/MUSE, X-Shooter and synthetic
118
+ spectra with aim to confirm presence of VMS. Additionally we
119
+ investigate the existence of the segregation of VMS by modeling
120
+ the morphology of the young massive star cluster (YMC) hosted
121
+ in the Sunburst galaxy.
122
+ The paper is organized as follows. In Section 2 we briefly de-
123
+ scribe the Sunburst galaxy and the available observational data.
124
+ In Section 3 we analyze the spectral signatures of very massive
125
+ stars using MUSE/IFU and X-Shooter observations in combina-
126
+ tion with the latest evolutionary models and synthetic spectra. In
127
+ Section 4 we discuss the morphological properties of the YMC
128
+ (dubbed 5.1) and the possible segregation of the (very) massive
129
+ stars in its central parts. We present our conclusion in Section 5.
130
+ We assume a flat cosmology with ΩM= 0.3, ΩΛ= 0.7 and
131
+ H0 = 70 km s−1 Mpc−1. Within this model, one arcsec at z = 2.37
132
+ corresponds to a projected physical scale of 8200 parsec. All
133
+ magnitudes are given in the AB system.
134
+ Fig. 1. Left: The HST F555W band image, showing the six aperture po-
135
+ sitions where a MUSE 1D spectrum is extracted (red contours). White
136
+ arrows point to the multiple images of the young stellar cluster. Right:
137
+ The MUSE IFU image at ∼ 1800Å of the same region shown on the left
138
+ with the same apertures in red.
139
+ 2. The Sunburst lensed galaxy
140
+ The Sunburst is a galaxy at z=2.37, strongly lensed by the
141
+ Planck cluster PSZ1 G311.65-18.48 at z=0.44, initially reported
142
+ by Dahle et al. (2016). The strong gravitational lensing effect
143
+ deflects the light from the background high-z Sunburst galaxy
144
+ into four bright arcs. These bright arcs harbor at least 13 star-
145
+ forming knots, which likely are stellar clusters. There are more
146
+ than 50 multiple images of this system (Pignataro et al. 2021),
147
+ whose physical properties are studied in detail in Vanzella et al.
148
+ (2022a). Among the 13 young stellar clusters, one has been iden-
149
+ tified 12 times (dubbed 5.1) and it is the subject of this work
150
+ (see Figure 1). The source 5.1 shows a multi-peaked Lyα emis-
151
+ sion consistent with an optically thin medium and Lyman con-
152
+ tinuum (LyC) leakage along the line of sight (Rivera-Thorsen
153
+ et al. 2017). Furthermore, the detection of LyC radiation emerg-
154
+ ing from the 12 detected multiple images of the 5.1 young mas-
155
+ sive star cluster is confirmed by HST multi-band observations
156
+ (Rivera-Thorsen et al. 2019). Additional analyses of the 12 LyC
157
+ multiple images of 5.1 have revealed that the star cluster has
158
+ an age younger than 3 Myr and a stellar metallicity of 0.5Z⊙
159
+ (Chisholm et al. 2019), with a physical size of ≃ 10 pc and a
160
+ stellar (and dynamical) mass value of ≃ 107 M⊙ (Vanzella et al.
161
+ 2022a).
162
+ The Sunburst was observed with HST, providing multi-
163
+ band photometry in the F275W, F410M, F555W, F606W,
164
+ F814W, F098M, F105W, F140W and F160W filters, under the
165
+ programs 15101 (PI Dahle), 15949 (PI Gladders), and 15377
166
+ (PI Bayliss). Sunburst has also been targeted with ground-
167
+ based high resolution (R ∼ 5000 − 9000) VLT/X-Shooter spec-
168
+ troscopy covering the spectral range 3000-22000Å in three main
169
+ arms, UVB, VIS abd NIR. The observational strategy and the
170
+ data reduction procedures applied to HST imaging and VLT/X-
171
+ Shooter spectroscopy have been presented in Vanzella et al.
172
+ (2020b, 2022a). VLT/MUSE integral field spectroscopy at res-
173
+ olution R = 3000 and covering the spectral range 4800-9400Å
174
+ was obtained during 2016 (1h integration, DDT, PI. Aghanim)
175
+ and 2021 (1h integration, PI, Vanzella) in the wide field mode
176
+ configuration. The final datacube which combines the two hours
177
+ and the data reduction is described in Vanzella et al. (2022a).
178
+ We also presented a first version of the lens model in Pignataro
179
+ et al. (2021) based on the 62 spectroscopically confirmed mul-
180
+ tiple images in the redshift range 1 < z < 3.5 (see also Sharon
181
+ et al. 2022; Diego et al. 2022). A revised lens model will be com-
182
+ puted once the new VLT/MUSE observations (7h integration)
183
+ planned during 2023 will be performed (prog. 110.249D.001,
184
+ PI. Vanzella).
185
+ Article number, page 2 of 10
186
+
187
+ PSF
188
+ HST F555W
189
+ PSF
190
+ MUSE IFU
191
+ AP1
192
+ AP2
193
+ AP3
194
+ 5.1a
195
+ 5.1b
196
+ 5.1c
197
+ 5.1d'
198
+ 5.1f
199
+ 5.1e
200
+ AP4
201
+ AP5
202
+ AP5
203
+ .5.1h
204
+ AP6
205
+ 5.1iU. Mestric et al.: Very massive, spatially segregated stars at z=2.4
206
+ Here we focus on the ≃ 3 Myr old, UV-bright and Ly-
207
+ man continuum source with MUV = −18.6 (1700Å magnitude
208
+ and ultraviolet slope β = −1.71 ± 0.01, Fλ ∼ λβ), massive
209
+ (M ∼ 107 M⊙) star cluster 5.1, subjected to large magnification
210
+ values (µ ∼ 10 − 70 over 12 multiple images, Pignataro et al.
211
+ 2021; Vanzella et al. 2022a). In the following we perform a new
212
+ analysis focusing on the nature of the ionizing source (Sect. 3)
213
+ and its morphology (Sect. 4).
214
+ 3. Spectral signatures of very massive stars in the
215
+ Sunburst star cluster at z=2.37
216
+ We aim to investigate the UV and optical spectroscopic prop-
217
+ erties of the young stellar cluster 5.1. The VLT/MUSE one-
218
+ dimensional spectra are extracted from six apertures enclosing
219
+ nine multiple images of 5.1 (shown in Fig. 1) and subsequently
220
+ combined to produce a continuum-detected high signal-to-noise
221
+ ratio SNR (> 60) weighted-average spectrum (Figure 2). The
222
+ stacked spectrum shown in Figure 2 is equivalent to an inte-
223
+ gration time of (2 × 9) × 302 > 16, 000 hours without lensing
224
+ amplification, adopting the minimum amplification among the 9
225
+ multiple images (µ = 30).
226
+ 3.1. Observed VMS features with VLT MUSE and X-Shooter
227
+ The very high SNR MUSE spectrum (Fig. 2) allows us to
228
+ identify several emission and absorption lines. Among them
229
+ we have the nebular emission lines associated to the interstel-
230
+ lar medium of the galaxy, like Oiii]λ1661, 1666, Niii]λ1750,
231
+ [Siiii]λ1883, 1892, and Ciii]λλ1907, 1909. The well detected
232
+ continuum allows us to investigate faint line emissions (of a frac-
233
+ tion of an Å rest-frame equivalent width), and to sample the de-
234
+ tails of the line profiles, otherwise not accessible without lens-
235
+ ing amplification. In particular, faint Nivλ1486 emission, the ev-
236
+ ident P-Cygni profile of the Civλ1550, the prominent broad and
237
+ asymmetric Heiiλ1640 line profile, and the P-Cygni signature of
238
+ Nivλ1720 clearly stand out. All these lines are associated with
239
+ young, hot and (very) massive stars.
240
+ We report detection of Heiiλ1640 emission with measured
241
+ rest frame EW=3.0±0.3Å and FWHM=8.8±1.7Å (∼ 1610±300
242
+ km s−1). The broad shape of the Heiiλ1640 emission line ob-
243
+ served in the Sunburst cluster is asymmetric and resembles a
244
+ typical P-Cygni profile. The blue end of the emission line drops
245
+ steeply, while the red end drops more gradually. The P-Cygni
246
+ profile of Heiiλ1640 line is consistent with that predicted by
247
+ models and synthetic spectra (see, Martins & Palacios 2022).
248
+ Broad Heiiλ1640 emission observed in galaxies is usually re-
249
+ lated to non-nebular origin, commonly associated with Wolf-
250
+ Rayet (WR) stars, (e.g. Schaerer & Stasi´nska 1999; Brinchmann
251
+ et al. 2008; Leitherer et al. 2018; Senchyna et al. 2021), though
252
+ the failure of the synthesis models to reproduce some of the
253
+ strong Heiiλ1640 lines might be related to missing ingredients in
254
+ stellar evolution models (see, e.g., Leitherer et al. 2018). How-
255
+ ever, far-UV spectroscopic investigation of ∼57 individual stars
256
+ located within the R136 star cluster reveals that massive stars
257
+ with M>100 M⊙ have a crucial role in producing the Heiiλ1640
258
+ emission line (Gräfener & Vink 2015; Crowther et al. 2016).
259
+ On the other hand, Martins & Palacios (2022) have shown that
260
+ Heiiλ1640 can be produced in significant amount only when stel-
261
+ lar winds are stronger than in normal O stars. VMS develop
262
+ such strong winds because of their proximity to the Eddington
263
+ limit (Vink et al. 2011; Bestenlehner 2020; Gräfener 2021). At
264
+ the same time, these winds peel off the external layers of the
265
+ stars and expose to the surface the products of hydrogen burn-
266
+ ing through the CNO cycle. This results in a strong nitrogen
267
+ (and helium) enrichment that boosts the strength of Nivλ1486
268
+ and Nivλ1720. This typically happens after ∼1.5 Myr. Both
269
+ mentioned emission lines are detected in the spectrum of the
270
+ Sunburst cluster at SNR > 15 (Figure 2), with EW=0.2Å and
271
+ FWHM=2.9Å for Nivλ1486 and EW=0.15Å and FWHM ∼ 2Å
272
+ for Nivλ1720. The helium enrichment also contributes to the
273
+ strength of Heiiλ1640. The same effects (strong winds combined
274
+ with surface chemical enrichment) happen in normal evolved
275
+ massive stars when they are seen as WR stars. The key difference
276
+ compared to VMS is that helium enrichment takes place only af-
277
+ ter the main sequence (>∼ 4 Myr), while the same process takes
278
+ place at younger ages in VMS. Furthermore, VMS are more lu-
279
+ minous than normal WR stars and hence their contribution to
280
+ integrated light is larger.
281
+ The nebular Hα equivalent width provides constraints on the
282
+ cluster age and hence on whether the Heiiλ1640 line is primar-
283
+ ily due to WR stars or VMS. According to the BPASS mod-
284
+ els and results from Eldridge & Stanway (2012) (their Fig. 3)
285
+ they predict that normal and WR stars produce EWHα < 1Å for
286
+ ages <∼ 3Myr. The X-Shooter spectrum reveals a prominent Hα
287
+ line and no continuum detection, which very conservatively im-
288
+ plies an equivalent width larger than 200Å rest-frame at 1-sigma.
289
+ However, if we assume for Hα the same continuum level ob-
290
+ served at λ ∼ 5000Å rest-frame in the photometric spectral en-
291
+ ergy distribution (SED) by Vanzella et al. (2022a) such a limit in-
292
+ creases to ∼ 840Å. This value would be still a lower limit, even in
293
+ the case of leakage of ionizing photons. After correcting the Hα
294
+ flux for the fraction of escaping LyC photons (Hα/(1− f abs
295
+ esc )), the
296
+ resulting EW increases to EWHα ∼ 1231Å. We adopt f abs
297
+ esc values
298
+ from Rivera-Thorsen et al. (2019), where the corresponding ab-
299
+ solute escape fraction of LyC photons along the line of sight is
300
+ f abs
301
+ esc = 32+2
302
+ −4% . Such a large Hα equivalent width is consistent
303
+ with a star-forming burst younger than ∼ 3 Myr (e.g., Leitherer
304
+ et al. 2014). Furthermore as discussed in Chisholm et al. (2019)
305
+ Nvλ1240 stellar wind profile predominantly depends on the stel-
306
+ lar age while variations due to different metallicity are negligible
307
+ and it is related to the young stellar populations (< 5 Myr). From
308
+ the comparison of the observed Nvλ1240 with the models Figure
309
+ 3 and 4 we additionally demonstrate that the age of the cluster is
310
+ < 3 Myr. From our age analysis, we can conclude that properties
311
+ of both Nvλ1240 and Hα fit well with < 3 Myr age of the stel-
312
+ lar cluster which requires other sources than WR stars to explain
313
+ the observed strong Heiiλ1640 EW=3.0±0.3Å. Therefore these
314
+ results strongly suggest that VMS are responsible for the pro-
315
+ duction of the spectral ultraviolet features we observe in such a
316
+ young massive star cluster. Moreover, Wofford et al. (2014) and
317
+ Smith et al. (2016) have argued that the presence of Ovλ1371 in
318
+ integrated light of the clusters was also a key feature of VMS.
319
+ This line is not seen in the Sunburst cluster, see Figure 2. As
320
+ demonstrated by Martins & Palacios (2022), this is not incom-
321
+ patible with the presence of VMS, since Ovλ1371 disappears as
322
+ VMS evolve to lower effective temperature. In their Fig. 4, we
323
+ see that no sign of Ovλ1371 exists after ∼1 Myr. This, together
324
+ with the presence of Nivλ1486, places a rather tight constraint
325
+ on the cluster age.
326
+ 3.2. Comparing observations with models
327
+ To investigate the rest-frame UV spectrum of the cluster, we have
328
+ created an integrated VMS model following Martins & Palacios
329
+ Article number, page 3 of 10
330
+
331
+ A&A proofs: manuscript no. VMS
332
+ Fig. 2. The MUSE IFU spectrum of the 5.1 young massive star cluster extracted from 6 apertures is shown in black (thin line) and the X-Shooter
333
+ long slit spectrum of 5.1l knot is shown in blue (bold line). The key confirmed features indicating the presence of VMS in the stellar cluster
334
+ are marked with shaded light red strips while the dark-orange line shows the best-fit Fλ ∼ λβ, with β = −1.71. The shaded grey strip indicates
335
+ the (absence of) Ovλ1371 line, which usually is an indicator of VMS too. The prominent P-Cygni of Nvλ1240 and strong emission part of the
336
+ Civλ1550 are present and indicate the young age of the stellar cluster (black bold markers). In the bottom, the 1-sigma errors of both spectra are
337
+ shown. Other detected interstellar features and stellar features are marked with dashed red and green lines, respectively.
338
+ (2022) that includes normal mass stars (0.1-100 M⊙) with differ-
339
+ ent VMS (150 M⊙ and 200 M⊙).
340
+ We have used the spectral energy distribution (SEDs) of
341
+ BPASS (Eldridge et al. 2017; Stanway & Eldridge 2018) v2.2.1
342
+ single-star population synthesis model. The model has an up-
343
+ per mass limit of 100 M⊙ with the Salpeter IMF, metallicity of
344
+ Z=0.006 (where 0.02 corresponds to solar metallicity), and in-
345
+ stantaneous star formation history, with a burst of mass 106 M⊙.
346
+ The adopted metallicity of the model is the closest to our mea-
347
+ sured value based on N2 index (Marino et al. 2013), which is
348
+ ≃ 0.4Z⊙ and consistent with the estimate provided by Mainali
349
+ et al. (2022) (see also Chisholm et al. 2019).
350
+ We have extrapolated the Salpeter IMF to 225 M⊙ upper
351
+ mass limit within a few mass bins given by Equation 1 from the
352
+ BPASS manual1. Equation 1 gives the number of massive stars
353
+ in the mass range [Ma; Mb].
354
+ N(Ma; Mb) = C × Mα1
355
+ 1
356
+ � Mb
357
+ Ma
358
+ Mα2 dM
359
+ (1)
360
+ Here, C is a constant and has a value of 1.23×105 for an arbitrary
361
+ burst mass of 106 M⊙. Also, M1 = 0.5 M⊙ α1 = -1.3, α2 = -
362
+ 2.35. The mass bins are selected in a way to add the SEDs of
363
+ appropriate numbers of single VMS stars, which are available
364
+ for discrete sets of VMS with masses including 150 M⊙ and 200
365
+ M⊙. In this manner we compute SEDs including VMS with IMFs
366
+ extending up to 175 and 225 M⊙, respectively, following Martins
367
+ & Palacios (2022).
368
+ 1 https://flexiblelearning.auckland.ac.nz/bpass/9.html
369
+ From Figure 2, we can see that the cluster shows signifi-
370
+ cant Nivλ1486 emission. From the VMS models and synthetic
371
+ spectra, Nivλ1486 emission only appears after 1.5 Myr of VMS
372
+ evolution (see, Martins & Palacios 2022) and VMS last approx-
373
+ imately until 2.5 Myr. Based on this, we created the SEDs of in-
374
+ tegrated VMS models at 1.5 Myr, 2 Myr, and 2.5 Myr. We have
375
+ normalized the spectrum of the cluster and the models by fitting
376
+ a UV power law by using the spectral windows provided by Rix
377
+ et al. (2004).
378
+ We have directly compared the cluster spectrum with the two
379
+ VMS models at 3 different ages. The comparison shows that
380
+ VMS are clearly needed to reproduce the observations (see, Fig-
381
+ ures 3, 4, and A.1). However, the Heiiλ1640 and Nivλ1720 lines
382
+ in the models appear stronger than observed even at the age of
383
+ 1.5 Myr and with a maximum mass of 175 M⊙. To match the
384
+ observed Heiiλ1640 and Nivλ1720 profiles, we have therefore
385
+ reduced the VMS contribution by decreasing their numbers. We
386
+ find good agreement if we reduce the VMS contribution by a fac-
387
+ tor of 6 in the VMS model, which includes only 150 M⊙ VMS
388
+ at 2.5 Myr (Fig. 3). Alternatively, a similar match is also found
389
+ by reducing the VMS contribution by a factor of 8 in models in-
390
+ cluding also the 200 M⊙ VMS (Fig. 4). In short, the observations
391
+ are compatible with an IMF extending up to ∼ 175 or 225 M⊙,
392
+ but with an IMF slope steeper than Salpeter (α2 < −2.35) for
393
+ M > 100 M⊙.
394
+ Article number, page 4 of 10
395
+
396
+ Aobs / A
397
+ 4000
398
+ 4500
399
+ 5000
400
+ 5500
401
+ 6000
402
+ 6500
403
+ 3.0
404
+ 1H13
405
+ IIIS.+IO
406
+ AS:
407
+ {IIIN
408
+ IIIS
409
+ IIIS
410
+ AID
411
+ 全13
412
+ sv
413
+ CIV1550
414
+ 2.5
415
+ NV1240
416
+ y1486
417
+ ux
418
+ fl
419
+ Normalized
420
+ 1.0
421
+ 0.5
422
+ 0.0
423
+ F
424
+ 1100
425
+ 1200
426
+ 1300
427
+ 1400
428
+ 1500
429
+ 1600
430
+ 1700
431
+ 1800
432
+ 1900
433
+ 2000
434
+ Arest / AU. Mestric et al.: Very massive, spatially segregated stars at z=2.4
435
+ 3.3. VMS contribution to the LyC budget of the young stellar
436
+ cluster
437
+ After adopting the results from the previous section, we can now
438
+ estimate the number of O-type stars hosted by the same stellar
439
+ cluster and the percentage of LyC photons emitted by VMS only.
440
+ Measurements are performed in the range of ∼730–900Å
441
+ rest-frame (range covered by HST F275W filter in which LyC
442
+ radiation is detected and fesc later on evaluated). First, we cal-
443
+ culate the mean flux from the model which include both normal
444
+ and VMS stars and, secondly, we calculate the mean flux from
445
+ the model including only VMS. The resulting ratio of those two
446
+ models gives us the fraction of the LyC photons produced by
447
+ VMS, which is ∼15%. Since LyC photons are mainly produced
448
+ by O-type and more massive stars, we can see that ∼15% of the
449
+ LyC production is generated only from ∼1% of the stars capable
450
+ of producing LyC photons. It is worth noting that the fraction
451
+ of LyC ionizing radiation produced from VMS in the Sunburst
452
+ 5.1 stellar cluster is smaller than the predicted LyC fraction pro-
453
+ duced by the VMS located in R136 stellar cluster, which is 25%
454
+ (Doran et al. 2013). However, we also note that in the case of the
455
+ Sunburst stellar cluster the light coming from the host galaxy
456
+ could slightly decreases the inferred equivalent width of the key
457
+ spectral features discussed above. While such dilution is difficult
458
+ to address with the present ground-based spectroscopic data2, its
459
+ effect implies a possible slightly higher contribution of VMS to
460
+ the LyC radiation.
461
+ 4. Spatial segregation of the Lyman continuum
462
+ radiation
463
+ We now address the morphological properties of image 5.1l,
464
+ which is the most magnified among the multiple images of the
465
+ star cluster (µtot ≃ 76, Pignataro et al. 2021). 5.1l is the brightest
466
+ image detected with a large SNR in the F275W (SNR ∼ 90) and
467
+ F555W (SNR ≫100), allowing us to investigate and compare
468
+ the morphology in these two spectral regions: the emitting LyC
469
+ (λ < 900Å, in HST F275W band) and the non-ionizing radiation
470
+ at 1700Å (HST F555W band). We follow two approaches: (1)
471
+ we ran simulations injecting the sources in the F275W band and
472
+ (2) we analyzed the curve of growth of the resulting images.
473
+ Figure 5 shows the F555W image of 5.1l, in which the elon-
474
+ gation is clearly visible in the direction of the tangential stretch
475
+ produced by gravitational lensing. As discussed in Pignataro
476
+ et al. (2021) (see also Vanzella et al. 2022a), the tangential am-
477
+ plification largely dominates along the arc (µtang ≃ 57). We per-
478
+ form here a relative comparison between images, to ensure that
479
+ the the results do not depend on the magnification values.
480
+ As a first step, we compute a realistic model of 5.1l on the
481
+ HST F555W image using Galfit (Peng et al. 2010). The point
482
+ spread function (PSF) has been extracted by combining non-
483
+ saturated stars available in the field of view. While the fit with a
484
+ single component does not produce acceptable residuals (larger
485
+ than 20%), we reproduce quite well the light profile of the ob-
486
+ ject by combining two components: a core with a Gaussian light
487
+ profile and an effective radius (Reff) smaller than 0.5 pixels (in
488
+ practice nearly unresolved) and an extended component with
489
+ Reff = 6 pixels and Sersic index n=1 (similar results are ob-
490
+ tained also with n=0.5). The combination of the two components
491
+ produces an optimal shape which leaves normalized residuals
492
+ smaller than 10% (see Figure 5). It is worth now investigating if
493
+ 2 JWST/NIRSpec-IFU and NIRCAM observations on the same YMC
494
+ are planned during 2023, prog. 2555, PI. Rivera-Thorsen
495
+ such a resolved shape (sampled at 1700Å) is recovered if placed
496
+ in the F275W Lyman continuum image. For this check, we in-
497
+ jected mock images of 5.1l into the F275W image on five dif-
498
+ ferent positions around 5.1l, which are not contaminated by the
499
+ flux coming from other sources. Such images are produced from
500
+ the aforementioned two-component Galfit model constructed
501
+ at 1700Å (F555W), but now accounting for the F275W PSF (in
502
+ other words convolved by the F275W PSF) to allow for a proper
503
+ comparison with the LyC 5.1l source (observed in HST F275W
504
+ band). Such images have been added to F275W after rescaling
505
+ each of them to the observed peak value of the LyC 5.1l ob-
506
+ ject. This step has been performed with IRAF (Tody 1986) task
507
+ IMARITH and IMCOPY. Figure 6 shows the results, in which all
508
+ the injected images show a spatially-resolved morphology along
509
+ the tangential magnification. Conversely, the observed LyC im-
510
+ age (of 5.1l) appears nucleated, suggesting that the emitting LyC
511
+ region is smaller than the one at 1700Å.
512
+ To quantify this result, we calculate the curve of growth
513
+ (CoG) of the images shown in Figure 6. The flux is then mea-
514
+ sured in the F275W band in 34 circular apertures. The small-
515
+ est aperture has a radius of 0.1 pixel. Intermediate apertures are
516
+ drawn with increasing radii, with a step of 0.5 pix, up to largest
517
+ one, which has a radius of 34 pixels. As a reference point-like
518
+ source, we constructed the mean CoG from a selected sample of
519
+ twenty non-saturated and non-contaminated stars. The resulting
520
+ CoG is shown in Figure 7, where the y-axis reports the frac-
521
+ tion of the flux enclosed at the corresponding radius in pixels
522
+ (x-axis).
523
+ The same procedure has been applied to the LyC emitting
524
+ source 5.1l, while another CoG has been constructed by averag-
525
+ ing the five CoG of the injected models resembling the morphol-
526
+ ogy at 1700Å. Figure 7 compares all the CoG after normalizing
527
+ them to the saturation value at the largest radius. The first re-
528
+ sult which emerges from this test is the clear deviation of the
529
+ CoG of the observed 5.1l LyC source from the behavior of a
530
+ point-like source (stars). This was not explored before and sug-
531
+ gests that in the most magnified image of the star cluster the
532
+ LyC appears spatially resolved. This is the first evidence of a re-
533
+ solved stellar LyC emission at cosmological distance. Second,
534
+ such barely resolved LyC emission appears more nucleated than
535
+ the one at 1700Å. Consequently, the sources of ionizing radi-
536
+ ation appears located in the central part of the cluster. Indeed,
537
+ from those curves, it emerges that 50% of the flux of the stars is
538
+ enclosed within a radius of ∼1.9 pixel, while for 5.1l it lies within
539
+ ∼2.2 pixels. Additionally, we perform the Kolmogorov-Smirnov
540
+ two-sample test (KS-test) to check if the CoGs derived from the
541
+ stars and 5.1l source follow the same distribution (null hypothe-
542
+ sis). For this purpose, we used the statistical function ks_2samp
543
+ from scipy.stats. After comparing the average CoG of the
544
+ stars with cyan and violet CoGs, the KS-test gives p << 0.05. It
545
+ means that the null hypothesis is not satisfied with the LyC pro-
546
+ file of 5.1l and it deviates from the CoG of a point-like source.
547
+ Furthermore, we also find that the half-light size of 5.1l at 1700Å
548
+ is larger than the ionizing region, ∼2.6 pixels compared to the
549
+ ∼2.2 pixels. If we correct such radii for the instrumental reso-
550
+ lution (given by the stars) we obtain an effective radius for 5.1l
551
+ at 1700Å ≃ 7.8 ± 1.4 pc after de-lensing3, in agreement with
552
+ Vanzella et al. (2020a), while the LyC image (5.1l) has a smaller
553
+ radius, Reff ≃ 4.7 ± 1.5 pc. We therefore find a LyC emission
554
+ which is more compact than the non-ionizing UV continuum,
555
+ 3 Adopting the pixel scale of 0.03′′/pixel, 8200 pc per arcsecond at
556
+ z=2.37 and µtang ≃ 57, Reff = 0.03∗8200∗((2.62−1.92)0.5)/57, adopting
557
+ the same uncertainty on µtang reported by Vanzella et al. (2022a).
558
+ Article number, page 5 of 10
559
+
560
+ A&A proofs: manuscript no. VMS
561
+ Fig. 3. The MUSE ultraviolet spectra of the young star cluster (blue) is shown in the bottom panel with the X-Shooter spectrum (green). The
562
+ grey-shaded regions show specific UV features closely associated with the presence of the VMS (Nivλ1486, Heiiλ1640, and Nivλ1720) and some
563
+ of them show a P-Cygni profile, characteristic of young and massive stars. Furthermore, the black line shows the single BPASS model including
564
+ only normal stars at 2.5 Myr, while the red line shows the BPASS single-star model augmented by VMS with masses up to 150 M⊙. The upper
565
+ panels show the zoom in VMS characteristic features compared with models and Nvλ1240 P-Cygni line, characteristic due to the presence of very
566
+ young stellar populations.
567
+ which we interpret as a spatial segregation of the most massive
568
+ stars.
569
+ 5. Summary and Conclusions
570
+ In this paper, we have presented a detailed spectroscopic and
571
+ morphological analysis of the massive and young stellar cluster
572
+ hosted in the Sunburst lensed galaxy at z=2.37, for which also
573
+ LyC emission was confirmed in the literature. We used results
574
+ from recent stellar evolutions and atmosphere models including
575
+ VMS (Martins & Palacios 2022) to conduct extensive compar-
576
+ isons with high spectral resolution observations performed with
577
+ VLT/MUSE and X-Shooter. The main results of this work can
578
+ be summarized as follows:
579
+ – In the spectroscopic observations, the high signal-to-noise
580
+ MUSE and X-Shooter spectra reveal features of broad (and
581
+ asymmetric) Heiiλ1640 emission with EW ≃ 3Å rest-frame
582
+ and line width of 1610 km s−1, and Nivλ1486 with EW ≃
583
+ 0.2Å emission. In addition, the P-Cygni profile of Nivλ1720
584
+ (along with NV and CIV) is also observed. All these features
585
+ suggest the presence of very massive (> 100M⊙) stars. The
586
+ absence of Ovλ1371 provides a lower age limit of 1 Myr. On
587
+ the other hand, the large Hα EW (> 1231Å after correcting
588
+ for the escaping LyC radiation) indicates an age younger than
589
+ ∼ 3 Myr. These narrow age constraints strongly favor the
590
+ existence of VMS over WR stars, implying that the strength
591
+ of the Heiiλ1640 emission line is entirely due to VMS.
592
+ – A comparison of the observations with the models reveals
593
+ that the most plausible age of the star cluster is 2.5 Myr, and
594
+ an estimated number of ∼ 370 − 400 VMS for a cluster mass
595
+ of 107 M⊙. The observations are compatible with an IMF
596
+ extending up to ∼ 175 − 225 M⊙, but with a slope which is
597
+ steeper than the Salpeter IMF.
598
+ – The fraction of LyC radiation emerging from the VMS com-
599
+ ponent is not negligible. We estimate that in the 730Å – 900Å
600
+ range (probed by the HST/F275W band) about 360 – 400
601
+ VMS (or roughly 1% of the total population of O-type stars
602
+ in the star cluster) account for 15% of the escaping LyC pho-
603
+ tons, with the rest being produced mostly by the other less
604
+ massive O-type stars.
605
+ – Detailed morphological analysis of the most magnified im-
606
+ age of the star cluster shows that the region emitting LyC
607
+ is not point-like, with a light profile different from the av-
608
+ erage profile of stars present in the same field of view. This
609
+ is the first evidence of a resolved LyC emission at any red-
610
+ shift. Remarkably, the physical scale of the LyC emitting re-
611
+ gion appears also smaller (with a significant K-S probability
612
+ p << 0.05) than the non-ionizing region (1700Å), suggest-
613
+ ing that massive O-type stars responsible for the LyC radi-
614
+ ation, and likely the VMS (significantly contributing to it),
615
+ are segregated in the central part of the star cluster. After de-
616
+ lensing the angular half-light radii, the LyC region appears
617
+ barely resolved with Reff ≃ 4.7 ± 1.5pc, while at 1700Å it
618
+ is Reff ≃ 7.8 ± 1.4 pc. The packaging of such a large num-
619
+ ber of massive O-type stars per parsec cube in the central
620
+ region, ≃ 70 pc−3, is likely a element which allowed to carve
621
+ the ionizing channel and the development of a high-speed
622
+ outflowing gas (Rivera-Thorsen et al. 2017; Vanzella et al.
623
+ 2022a; Mainali et al. 2022).
624
+ Acknowledgements. We acknowledge financial support through grants PRIN-
625
+ MIUR 2017WSCC32, 2020SKSTHZ and the INAF GO Grant 2022 “The rev-
626
+ Article number, page 6 of 10
627
+
628
+ Nv1240A
629
+ NIV1486A
630
+ NIV1720A
631
+ HeI1640A
632
+ 1.6
633
+ 1.4
634
+ Sunburst Cluster Mus
635
+ 1.5
636
+ 1.3 E
637
+ BPASS 100 M。+ 150 M。VMS (39.43) at 2.5 Myr
638
+ 1.4
639
+ 1.2
640
+ 0.7 E
641
+ 0.7
642
+ 0.4E
643
+ 0.61705
644
+ 1220
645
+ 1492
646
+ 1640
647
+ 1650
648
+ 1240
649
+ 1480
650
+ 1484
651
+ 1488
652
+ 1660
653
+ 670
654
+ 1710
655
+ 1715
656
+ 1720
657
+ 1725
658
+ 1730
659
+ Rest Frame Wavelength [A]
660
+ Rest Frame Wavelength [A]
661
+ Rest Frame Wavelength [A]
662
+ Rest Frame Wavelength [A]
663
+ 2.0E
664
+ Sunburst Cluster MUSE
665
+ Sunburst Cluster Xshooter
666
+ BPASS 2.5 Myr
667
+ 1.8E
668
+ Single BPASSup to 100 Mo + 39.43 number of 150 Mo VMS at 2.5 Myr
669
+ 1.6E
670
+ 1.4
671
+ 1.2E
672
+ 1.0M
673
+ W
674
+ 0.8E
675
+ Nor
676
+ 0.6
677
+ 0.4
678
+ 0.2日
679
+ 1250
680
+ 1300
681
+ 1350
682
+ 1400
683
+ 1450
684
+ 1500
685
+ 1550
686
+ 1600
687
+ 1650
688
+ 1700
689
+ 1750
690
+ Rest Frame Wavelength [A]U. Mestric et al.: Very massive, spatially segregated stars at z=2.4
691
+ Fig. 4. All symbols as in Figure 3 except for the red lines, showing the BPASS single-star model augmented by VMS with masses up to 200 M⊙.
692
+ Fig. 5. The results from Galfit modeling after using a two-component
693
+ fit. From the left, the first panel shows the 5.1l source in the F555W
694
+ band (UV1700Å). The second panel shows the model from Galfit. The
695
+ third panel shows the residual and the fourth panel is the normalized
696
+ residual produced after dividing the residual with the original image.
697
+ The white contour encloses the region used to check the quality of the
698
+ produced model.
699
+ olution is around the corner: JWST will probe globular cluster precursors and
700
+ Population III stellar clusters at cosmic dawn” (PI Vanzella). FC and RP ac-
701
+ knowledge funding from PRIN INAF 1.05.01.85.01.
702
+ References
703
+ Abel, T., Bryan, G. L., & Norman, M. L. 2002, Science, 295, 93
704
+ Bestenlehner, J. M. 2020, MNRAS, 493, 3938
705
+ Brands, S. A., de Koter, A., Bestenlehner, J. M., et al. 2022, A&A, 663, A36
706
+ Brinchmann, J., Kunth, D., & Durret, F. 2008, A&A, 485, 657
707
+ Calzetti, D., Johnson, K. E., Adamo, A., et al. 2015, ApJ, 811, 75
708
+ Chisholm, J., Rigby, J. R., Bayliss, M., et al. 2019, ApJ, 882, 182
709
+ Cignoni, M., Sabbi, E., van der Marel, R. P., et al. 2015, ApJ, 811, 76
710
+ Crowther, P. A., Caballero-Nieves, S. M., Bostroem, K. A., et al. 2016, MNRAS,
711
+ 458, 624
712
+ Crowther, P. A., Schnurr, O., Hirschi, R., et al. 2010, MNRAS, 408, 731
713
+ Dahle, H., Aghanim, N., Guennou, L., et al. 2016, A&A, 590, L4
714
+ Diego, J. M., Pascale, M., Kavanagh, B. J., et al. 2022, A&A, 665, A134
715
+ Doran, E. I., Crowther, P. A., de Koter, A., et al. 2013, A&A, 558, A134
716
+ Eldridge, J. J. & Stanway, E. R. 2012, MNRAS, 419, 479
717
+ Eldridge, J. J., Stanway, E. R., Xiao, L., et al. 2017, PASA, 34, e058
718
+ Goswami, S., Slemer, A., Marigo, P., et al. 2021, A&A, 650, A203
719
+ Gräfener, G. 2021, A&A, 647, A13
720
+ Gräfener, G. & Vink, J. S. 2015, A&A, 578, L2
721
+ Fig. 6. The HST F275W image showing 5.1l LyC leaking source in
722
+ its centre; other two multiple images, of the same source, are labeled
723
+ as 5.1i and 5.1h. Around 5.1l, in the region not contaminated by other
724
+ sources, five models are injected (see text for more details), enclosed
725
+ in the white squares. In the bottom right corner, we shown the largest
726
+ (34pix diameter) aperture size used to construct CoGs.
727
+ Johnson, T. L., Rigby, J. R., Sharon, K., et al. 2017, ApJ, 843, L21
728
+ Köhler, K., Langer, N., de Koter, A., et al. 2015, A&A, 573, A71
729
+ Leitherer, C., Byler, N., Lee, J. C., & Levesque, E. M. 2018, ApJ, 865, 55
730
+ Leitherer, C., Ekström, S., Meynet, G., et al. 2014, ApJS, 212, 14
731
+ Mainali, R., Rigby, J. R., Chisholm, J., et al. 2022, ApJ, 940, 160
732
+ Marino, R. A., Rosales-Ortega, F. F., Sánchez, S. F., et al. 2013, A&A, 559, A114
733
+ Article number, page 7 of 10
734
+
735
+ Nv1240A
736
+ NIV1486A
737
+ HeII1640A
738
+ NIV1720A
739
+ - Sunburst Cluster Xshootel
740
+ 1.05
741
+ 1260
742
+ 1480
743
+ 1492
744
+ 1496
745
+ 1630
746
+ 1640
747
+ 1650
748
+ 1670
749
+ 1220
750
+ 1484
751
+ 1488
752
+ 1710
753
+ 1715
754
+ 1725
755
+ 1730
756
+ 1230
757
+ Rest Frame Wavelength [A]
758
+ Rest Frame Wavelength [A]
759
+ Rest Frame Wavelength [A]
760
+ Rest Frame Wavelength [A]
761
+ 2.0
762
+ Sunburst Cluster MUSE
763
+ 1.8E
764
+ Sunburst Cluster Xshooter
765
+ BPASS 2.5 Myr
766
+ 1.6E
767
+ 1.4
768
+ .2
769
+ 1.0
770
+ MA
771
+ 0.8
772
+ LLLLLLLLLLLLL
773
+ LJON
774
+ 0.6
775
+ 0.4E
776
+ 0.2E
777
+
778
+ 0.0
779
+ 1250
780
+ 1300
781
+ 1350
782
+ 1400
783
+ 1450
784
+ 1500
785
+ 1550
786
+ 1600
787
+ 1650
788
+ 1700
789
+ 1750
790
+ Rest Frame Wavelength [A]5.11
791
+ normalised
792
+ model
793
+ residual
794
+ HST
795
+ F555W
796
+ residual5.1h
797
+ 1"
798
+ Model5
799
+ Model
800
+ 5.1i
801
+ Model
802
+ 5.11
803
+ Model
804
+ ModelA&A proofs: manuscript no. VMS
805
+ Fig. 7. Three curves of growth normalized to 1. The cyan CoG cor-
806
+ responds to the 5.1l LyC leaking source located in the Sunburst arc
807
+ (F275W band). The violet CoG is the PSF-convolved, best-fit model of
808
+ the 5.1l source observed in F555W constructed averaging 5 CoGs (see
809
+ text for more details). The orange growth curve is used for comparison
810
+ and it is constructed averaging 20 single CoGs from randomly selected
811
+ stars. In both cases (cyan and violet) error bars are 1σ. Vertical lines in
812
+ the bottom left part of the figure mark the pixel radii at which 50% of
813
+ the light is enclosed. The colors corresponds to the CoG colors.
814
+ Martins, F., Hillier, D. J., Paumard, T., et al. 2008, A&A, 478, 219
815
+ Martins, F. & Palacios, A. 2022, A&A, 659, A163
816
+ Meštri´c, U., Vanzella, E., Zanella, A., et al. 2022, MNRAS, 516, 3532
817
+ Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2010, AJ, 139, 2097
818
+ Pignataro, G. V., Bergamini, P., Meneghetti, M., et al. 2021, A&A, 655, A81
819
+ Portegies Zwart, S. F., McMillan, S. L. W., & Gieles, M. 2010, ARA&A, 48, 431
820
+ Rigby, J. R., Bayliss, M. B., Chisholm, J., et al. 2018a, ApJ, 853, 87
821
+ Rigby, J. R., Bayliss, M. B., Sharon, K., et al. 2018b, AJ, 155, 104
822
+ Rigby, J. R., Johnson, T. L., Sharon, K., et al. 2017, ApJ, 843, 79
823
+ Rivera-Thorsen, T. E., Dahle, H., Chisholm, J., et al. 2019, Science, 366, 738
824
+ Rivera-Thorsen, T. E., Dahle, H., Gronke, M., et al. 2017, A&A, 608, L4
825
+ Rix, S. A., Pettini, M., Leitherer, C., et al. 2004, ApJ, 615, 98
826
+ Schaerer, D. & Stasi´nska, G. 1999, A&A, 345, L17
827
+ Senchyna, P., Stark, D. P., Charlot, S., et al. 2021, MNRAS, 503, 6112
828
+ Sharon, K., Mahler, G., Rivera-Thorsen, T. E., et al. 2022, arXiv e-prints,
829
+ arXiv:2209.03417
830
+ Smith, L. J., Bajaj, V., Ryon, J., & Sabbi, E. 2020, ApJ, 896, 84
831
+ Smith, L. J., Crowther, P. A., Calzetti, D., & Sidoli, F. 2016, ApJ, 823, 38
832
+ Stanway, E. R. & Eldridge, J. J. 2018, MNRAS, 479, 75
833
+ Tody, D. 1986, in Proc. SPIE, Vol. 627, Instrumentation in astronomy VI, ed.
834
+ D. L. Crawford, 733
835
+ Vanzella, E., Calura, F., Meneghetti, M., et al. 2017, MNRAS, 467, 4304
836
+ Vanzella, E., Caminha, G. B., Calura, F., et al. 2020a, MNRAS, 491, 1093
837
+ Vanzella, E., Caminha, G. B., Rosati, P., et al. 2021, A&A, 646, A57
838
+ Vanzella, E., Castellano, M., Bergamini, P., et al. 2022a, A&A, 659, A2
839
+ Vanzella, E., Castellano, M., Bergamini, P., et al. 2022b, ApJ, 940, L53
840
+ Vanzella, E., De Barros, S., Cupani, G., et al. 2016, ApJ, 821, L27
841
+ Vanzella, E., Meneghetti, M., Pastorello, A., et al. 2020b, MNRAS, 499, L67
842
+ Vink, J. S., Heger, A., Krumholz, M. R., et al. 2015, Highlights of Astronomy,
843
+ 16, 51
844
+ Vink, J. S., Muijres, L. E., Anthonisse, B., et al. 2011, A&A, 531, A132
845
+ Wofford, A., Leitherer, C., Chandar, R., & Bouret, J.-C. 2014, ApJ, 781, 122
846
+ Yusof, N., Hirschi, R., Meynet, G., et al. 2013, MNRAS, 433, 1114
847
+ Article number, page 8 of 10
848
+
849
+ 1.0
850
+ 0.8
851
+ iction
852
+ rd
853
+ 0.6
854
+ ux
855
+ 0.4
856
+ 0.2
857
+ observed CoG of 5.1l in F275W
858
+
859
+ avg CoG for 5 observed Stars
860
+ 0.0
861
+ 6AE
862
+ CoG model n=0.5
863
+ 0.0
864
+ 2.5
865
+ 5.0
866
+ 7.5
867
+ 10.0
868
+ 12.5
869
+ 15.0
870
+ 17.5
871
+ Radius
872
+ [pixels]U. Mestric et al.: Very massive, spatially segregated stars at z=2.4
873
+ Appendix A: Initial models
874
+ As described in Section 3.2, we compare the X-Shooter and
875
+ MUSE spectroscopic observations with the BPASS models at
876
+ different ages (1.5 Myr, 2 Myr, and 2.5 Myr). We narrow our
877
+ models to the mentioned age range since the predicted age of the
878
+ cluster is higher than 1.5 Myr (inferred from Nivλ1486 emission
879
+ line) and the lifetime of the VMS is about 2.5 Myr (Martins &
880
+ Palacios 2022). We started with the BPASS which has an upper
881
+ mass limit of 100 M⊙ and added 236.56 stars in the 100 - 175
882
+ M⊙ mass range and 60.30 stars in the mass range 175 – 225 M⊙.
883
+ Resulting models (at different ages) are shown in A.1; all models
884
+ produce significantly strong spectroscopic features characteristic
885
+ of the presence of VMS. To better match models with observa-
886
+ tions, we decreased the numbers of the VMS and the final results
887
+ are presented in Section 3 (Figure. 3 and 4).
888
+ Article number, page 9 of 10
889
+
890
+ A&A proofs: manuscript no. VMS
891
+ Fig. A.1. The six panels are showing the comparison of the observations (MUSE spectrum, red line) with BPASS models (black line). The left
892
+ column shows three BPASS models with an IMF up to 100 M⊙, ages of 1.5 Myr, 2 Myr, 2.5 Myr and an added number of 236.56 150 M⊙ VMS.
893
+ The right column shows BPASS model with an IMF up to 100 M⊙ at same ages (as shown in previous column) but with added nuber of 236.56
894
+ and 60.30 VMS of 150 M⊙ and 200 M⊙, respectively. In all six panels we can see that the strength of feature characteristic to VMS is higher than
895
+ in observed cluster, see the Section 3 for more details.
896
+ Article number, page 10 of 10
897
+
BdE3T4oBgHgl3EQftAtl/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
BtFAT4oBgHgl3EQfsR7g/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8878f216a8ecf8097bc30f5b571f2f56b794180aef570db16c872faf056111a7
3
+ size 2621485
CNFRT4oBgHgl3EQfvzii/content/2301.13636v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dba39f7ed69b2744e3f38595416007033c135d407c43a95e8e5109d2653dc59c
3
+ size 6051264
CNFRT4oBgHgl3EQfvzii/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce27fd9f836e63a6939bc3b4a54fa1ecf436f9ece56403f0534d967f227b47e9
3
+ size 4784173
CNFRT4oBgHgl3EQfvzii/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b06da2428b6e1f8c6dce52e2fcf7faf559a353ee027239dae852d0800be63f5
3
+ size 189144
CtAyT4oBgHgl3EQfR_f1/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f56613ea7b0527205786e26b426afe4a804acecc2226db0a370d0dcc2fa9bfd
3
+ size 11993133
DdAzT4oBgHgl3EQfif0c/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d14d4d211746405171b0420b70d4934a71c78ffd19567b12dd4bdcc9e2da31c
3
+ size 5767213
E9E0T4oBgHgl3EQfQwCg/content/tmp_files/2301.02198v1.pdf.txt ADDED
@@ -0,0 +1,1971 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+ Privacy-Preserving Distributed Energy Resource
3
+ Control with Decentralized Cloud Computing
4
+ Xiang Huo, Graduate Student Member, IEEE, Mingxi Liu, Member, IEEE
5
+ Abstract—The rapidly growing penetration of renewable en-
6
+ ergy resources brings unprecedented challenges to power distri-
7
+ bution networks – management of a large population of grid-
8
+ tied controllable devices encounters control scalability crises and
9
+ potential end-user privacy breaches. Despite the importance,
10
+ research on privacy preservation of distributed energy resource
11
+ (DER) control in a fully scalable manner is lacked. To fill the
12
+ gap, this paper designs a novel decentralized privacy-preserving
13
+ DER control framework that 1) achieves control scalability over
14
+ DER population and heterogeneity; 2) eliminates peer-to-peer
15
+ communications and secures the privacy of all participating
16
+ DERs against various types of adversaries; and 3) enjoys higher
17
+ computation efficiency and accuracy compared to state-of-the-
18
+ art privacy-preserving methods. A strongly coupled optimization
19
+ problem is formulated to control the power consumption and
20
+ output of DERs, including solar photovoltaics and energy storage
21
+ systems, then solved using the projected gradient method. Cloud
22
+ computing and secret sharing are seamlessly integrated into the
23
+ proposed decentralized computing to achieve privacy preserva-
24
+ tion. Simulation results prove the capabilities of the proposed
25
+ approach in DER control applications.
26
+ Index Terms—Decentralized optimization, distributed energy
27
+ resources, privacy preservation, secret sharing
28
+ I. INTRODUCTION
29
+ A. Related Works
30
+ L
31
+ ARGE-scale deployment of distributed energy resources
32
+ (DERs) has proven efficacy in reducing carbon footprint
33
+ and providing grid-edge services such as voltage control, load
34
+ following, and backup power supply [1]. DERs, including
35
+ energy storage systems (ESSs), solar photovoltaic (PV), and
36
+ electric vehicles (EVs), along with other monitoring and
37
+ controllable devices, can offer significant opportunities for
38
+ advancing efficient, reliable, and cost-effective power grids [2],
39
+ [3]. Though integrating DERs into power grids can provide
40
+ multifarious benefits, such as enhanced energy efficiency and
41
+ economic boost, the high penetration of DERs raises surging
42
+ challenges on the scalability of existing control strategies [4].
43
+ To address the aforementioned challenges in large-scale
44
+ DER control problems, distributed and decentralized control
45
+ strategies are drawing increased attention owing to their
46
+ superior scalability. For instance, a distributed coordination
47
+ method based on local droop control and consensus control
48
+ was designed in [5] to deal with the voltage rise problem
49
+ caused by the high penetration of solar PVs. Zhang et al. in [6]
50
+ proposed an asynchronous distributed leader-follower control
51
+ strategy that optimally schedules DERs to lower the voltage
52
+ The authors are with the Department of Electrical and Computer Engineer-
53
+ ing, University of Utah, Salt Lake City, UT 84112 USA (e-mail: xiang.huo,
54
+ mingxi.liu@utah.edu).
55
+ for peak load shaving and long-term energy saving. To reduce
56
+ the communication burden, a distributed low-communication
57
+ algorithm was proposed in [7] to control islanded PV-battery-
58
+ hybrid systems. Though distributed methods can achieve
59
+ scalability, they generically suffer from massive peer-to-peer
60
+ communications. To overcome this issue, Navidi et al. in
61
+ [8] developed a two-layer decentralized DER coordination
62
+ architecture that can scale the solution to large networks, and
63
+ no direct communication is required between local controllers.
64
+ In [9], a decentralized stochastic control strategy was designed
65
+ for radial distribution systems with controllable PVs and ESSs
66
+ to minimize the demand balancing cost. Huo et al. in [10] pro-
67
+ posed a decentralized shrunken primal-multi-dual subgradient
68
+ algorithm with dimension reduction to achieve scalability w.r.t.
69
+ both agent population size and network dimension.
70
+ Despite the superior scalability and communication effi-
71
+ ciency of decentralized methods, their implementation has
72
+ been significantly hampered by the vulnerability to privacy
73
+ breaches. Furthermore, both distributed and decentralized
74
+ strategies rely heavily on mandatory communications which
75
+ can disclose users’ sensitive information and expose system
76
+ vulnerabilities to adversaries. Differential privacy (DP) has re-
77
+ ceived substantial attention in addressing privacy concerns due
78
+ to its rigorous mathematical formulation [11]. DP-based meth-
79
+ ods add persistent randomized perturbations to the datasets,
80
+ constraints, or objective functions for privacy preservation.
81
+ In [12], a DP-based aggregation algorithm is proposed to
82
+ compensate for solar power fluctuations and protect users’
83
+ personal information. Han et al. in [13] developed a distributed
84
+ optimization algorithm based on DP to preserve the privacy
85
+ of the participating agents. Gough et al. in [14] designed an
86
+ innovative DP-compliant algorithm to ensure that the data
87
+ from consumers’ smart meters are protected. Despite the
88
+ success in privacy preservation, DP-based methods inevitably
89
+ suffer from accuracy loss due to the added perturbations.
90
+ In contrast, encryption-based strategies achieve privacy
91
+ preservation with high accuracy by encrypting the original
92
+ data into cyphertexts, and only those holding private keys
93
+ can decrypt the cyphertexts. Lu et al. in [15] proposed an
94
+ efficient and privacy-preserving aggregation scheme for smart
95
+ grid communications, in which the data is encrypted by Paillier
96
+ cryptosystem. In [16], a privacy-preserving and fault-tolerant
97
+ scheme was designed based on homomorphic cryptosystem
98
+ to achieve secure aggregation of metering data. Similarly,
99
+ Cheng et al. in [17] proposed a novel private collaborative
100
+ distributed energy management system based on homomorphic
101
+ encryption to solve the privacy issues in distribution systems
102
+ and microgirds. Despite the high accuracy, the drawback
103
+ arXiv:2301.02198v1 [math.OC] 5 Jan 2023
104
+
105
+ 2
106
+ of encryption-based methods lies in the prevalent comput-
107
+ ing overhead caused by encryption and decryption. Other
108
+ hardware-integrated privacy-preserving methods, e.g., garbled
109
+ circuit [18], [19], are deficient in flexibility and uneconomic
110
+ due to the hardware cost.
111
+ Secret sharing (SS) [20] is a lightweight cryptographic
112
+ method that can securely distribute a secret among a group
113
+ of participants. Each participant will be allocated a share
114
+ of the secret, and only through the collaboration of certain
115
+ participants where the number of participants is greater than a
116
+ threshold can the secret be reconstructed from their shares.
117
+ Adopting SS, Nabil et al. in [21] designed an SS-based
118
+ detection scheme to identify malicious consumers who steal
119
+ electricity, in which system operators only collect masked
120
+ meter readings from the consumers to avoid privacy vio-
121
+ lation. In [22], an SS-based EV charging control protocol
122
+ was developed to achieve privacy-preserving EV charging
123
+ control for overnight valley filling. Compared with encryption-
124
+ based strategies, SS-based methods can preserve privacy while
125
+ avoiding the heavy computational load. Despite the superiority,
126
+ few research studied the integration of SS into DER control
127
+ due to the highly complex distribution network structure, large
128
+ DER population, and lack of theoretical support in privacy
129
+ guarantees. To fill these gaps, this paper designs a novel SS-
130
+ based privacy-preserving algorithm that merits high efficiency,
131
+ security, and accuracy for large-scale DER control problems.
132
+ B. Statement of Contributions
133
+ The contribution of this paper is three-fold: 1) We propose
134
+ a novel decentralized privacy-preserving algorithm that con-
135
+ currently achieves scalability and privacy in large-scale DER
136
+ control. To the best of our knowledge, this is the first paper
137
+ that proposes a decentralized SS-based algorithm for DER
138
+ privacy preservation, in which decentralized solutions, privacy
139
+ guarantees, and rigorous security proofs are provided; 2) The
140
+ proposed method eliminates the frequent peer-to-peer commu-
141
+ nications and secures the privacy of the participating DERs
142
+ against various types of adversaries. The designed framework
143
+ serves as a benchmark for secure and scalable DER control. 3)
144
+ Compared to state-of-the-art approaches, the proposed method
145
+ can achieve lower computational overhead and identically
146
+ accurate solutions as the non-privacy-concerned algorithms.
147
+ The rest of this paper is organized as follows: In Sec-
148
+ tion II, we construct the models of distribution networks,
149
+ PVs, and ESSs, then formulate the DER control problem
150
+ into a constrained optimization problem. Section III derives
151
+ the decentralized solution via the projected gradient method
152
+ and presents the corresponding DER aggregation and control
153
+ strategies. The SS-based privacy-preserving DER control al-
154
+ gorithm and privacy analyses are provided in Section IV. We
155
+ give simulation results and analyses in Section V. Section VI
156
+ concludes this paper.
157
+ II. PROBLEM FORMULATION
158
+ A. Branch Flow Model
159
+ Consider an n-bus radial distribution network where B =
160
+ {0, 1, . . . , n} denotes the set of buses. Let lij denote the line
161
+ segment connecting buses i and j, L = {1, . . . , h} denote
162
+ the set of lines, Cj denote the set of bus j’s child buses, Vj
163
+ denote the voltage magnitude at bus j, Pij and Qij denote
164
+ the active and reactive power flow from bus i to bus j,
165
+ respectively, and rij and xij be the resistance and reactance of
166
+ line lij, respectively. For bus j, let pc
167
+ j and qc
168
+ j denote the active
169
+ and reactive power consumptions, respectively, and pg
170
+ j and qg
171
+ j
172
+ denote its active and reactive power generations, respectively.
173
+ To simplify the network model, a nonlinear DistFlow model
174
+ [23] can be linearized to the LinDistFlow model by omitting
175
+ the higher order terms with negligible error [24]. Therefore,
176
+ this paper adopts the LinDistFlow model, represented as
177
+ Pij −
178
+
179
+ u∈Cj
180
+ Pju = pc
181
+ j − pg
182
+ j
183
+ (1a)
184
+ Qij −
185
+
186
+ u∈Cj
187
+ Qju = qc
188
+ j − qg
189
+ j
190
+ (1b)
191
+ V 2
192
+ i − V 2
193
+ j = 2(rijPij + xijQij).
194
+ (1c)
195
+ A radial 13-bus distribution network connected with rooftop
196
+ solar PVs and ESSs is shown in Fig. 1 and will be used as an
197
+ example throughout this paper.
198
+ 10
199
+ 3
200
+ 2
201
+ 11
202
+ 6
203
+ 7
204
+ 5
205
+ 9
206
+ 8
207
+ 4
208
+ 1
209
+ 0
210
+ P1, Q1
211
+ <latexit sha1_base64="nPdIyijfg7kH2CtCftJt/5gzaLY=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoi4mNXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0XWjbC8MczrmXe+5xY0aFNIyRVlhaXldK6XNja3tnf03
212
+ b2miBKOSQNHLOJtFwnCaEgakpG2jEnKHAZabn9u1xvPRIuaBQ+yEFM7AB1Q+pTjKSiHP3SciPmiUGgvtQKkOxhxNJaljnmGVyo1XPN0ctGxRgXnAfmFJSrh8P6z9PRsObo35YX4SQgocQMCdExjVjaKeKSYkaykpUIEiPcR13SUTBEARF2Or4vgyeK8aAfcfVCcfs34kUBSK3qTpzl2JWy8lFWieR/rWd0jBOJAnxZJGfMCgjmIcFPcoJlmygAMKcKq8Q9xBHWKpISyoEc/bkedA8r5gXlZu6Wa7egkVwQE4BqfABFegCu5BDTQABs/gFbyD+1Fe9M+ta9Ja0GbzuyDf6WNfgEg2af6</latexit>
213
+ P2, Q2
214
+ <latexit sha1_base64="IaoZ/P1zRc4iegI/JD0D2LbqOjw=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkpSxMeu6MZlC/YBTQiTyaQdOnkwMxFKyNK/cONfdO3GhSLirt/gTzhpu9DaC8MczrmXe+5xY0aFNIyxVlhaXldK6XNja3tnf03b2
215
+ WiBKOSRNHLOIdFwnCaEiakpGOjEnKHAZabuD21xvPxAuaBTey2FM7AD1QupTjKSiHP3CciPmiWGgvtQKkOxjxNJ6ljnVM7hQa+Sao5eNijEp+B+YM1CuHY4a349Ho7qjf1lehJOAhBIzJETXNGJp4hLihnJSlYiSIzwAPVIV8EQBUTY6eS+DJ4oxoN+xNULJZywvydSFIjcpurMXYp5LScXad1E+ld2SsM4kSTE0V+wqCMYB4W9CgnWLKhAghzqrxC3EcYakiLakQzPmT/4NWtWKeV64bZrl2A6ZVBAfgGJwCE1yCGrgDdAEGDyBF/AG3rVn7VX70D6nrQVtNrMP/pQ2/gEj/af8</latexit>
216
+ p3, q3
217
+ <latexit sha1_base64="gDgNMRKEQIJW/+4UL0fQMJi6A4=">AC3icbVDLSsNAFJ3UV62vqEtFhbBhZREBXVXdOyBfuANoTJZNIOnUzizEQoUvBjb/ixoVF3PoD7vwGf8J20WtHhjmcM693HuPFzMqlWV9GbmFxaXlfxqYW19Y3PL3N5
218
+ pyCgRmNRxCLR8pAkjHJSV1Qx0oFQaHSNPrX2d+854ISN+qwYxcULU5TSgGCktuWax40XMl4NQf2k8dE+P4axypxXLFlawz4l9hTUqrsj2rfDwejqmt+dvwIJyHhCjMkZdu2YuWkSCiKGRkWOokMcJ91CVtTkKiXTS8S1DeKgVHwaR0I8rOFZnO1IUymw5XRki1ZPzXib+57UTFVw4KeVxogjHk0FBwqCKYBYM9KkgWLGBJgLqneFuIcEwkrHV9Ah2PMn/yWNk7J9Vr6s6TSuwAR5sAeK4AjY4BxUwA2ogjrA4BE8g1cwMp6MF+PNeJ+U5oxpzy74BePjB5D3nxk=</latexit>
219
+ p2, q2
220
+ <latexit sha1_base64="JCjfzJ1/fuXMTynwfU5OWgRT7cw=">AC3icbVDLSsNAFJ34rPUVdanI0CK4kJIUQd0V3bhswT6gDWEymbRDJ5M4MxFK6FJw46+4cWERt/6AO7/Bn3DSdlFbDwxzOde7r3HixmVyrK+jaXldW19dxGfnNre2fX3
221
+ NtvyCgRmNRxCLR8pAkjHJSV1Qx0oFQaHSNPr32R+84EISN+pwYxcULU5TSgGCktuWah40XMl4NQf2k8dMtncFa514prFq2SNQZcJPaUFCtHo9rP4/Go6pfHT/CSUi4wgxJ2batWDkpEopiRob5TiJjHAfdUlbU45CIp10fMsQnmjFh0Ek9OMKjtXZjhSFMltOV4ZI9eS8l4n/e1EBZdOSnmcKMLxZFCQMKgimAUDfSoIVmygCcKC6l0h7iGBsNLx5XUI9vzJi6RLtnpauaTuMaTJADh6AToENLkAF3IqAMnsALeAMj49l4Nd6Nj0npkjHtOQB/YHz+Ao3dnxc=</latexit>
222
+ p1, q1
223
+ <latexit sha1_base64="QFaEiH12z+13LcqCfvsdkGI/A7o=">AC3icbVDLSsNAFJ3UV62vqEtFhbBhZREBHVXdOyBfuANoTJZNIOnUzizEQoUvBjb/ixoVF3PoD7vwGf8J20VtPTDM4Zx7ufceL2ZUKsv6NnJLyura/n1wsbm1vaOu
224
+ bvXkFEiMKnjiEWi5SFJGOWkrqhipBULgkKPkabXv8n85gMRkb8Tg1i4oSoy2lAMVJacs1ix4uYLweh/tJ46NqncFa514prlqyNQZcJPaUlCqHo9rP49Go6pfHT/CSUi4wgxJ2batWDkpEopiRoaFTiJjHAfdUlbU45CIp10fMsQHmvFh0Ek9OMKjtXZjhSFMltOV4ZI9eS8l4n/e1EBZdOSnmcKMLxZFCQMKgimAUDfSoIVmygCcKC6l0h7iGBsNLxFXQI9vzJi6RxVrbPy1c1u1S5BhPkwQEoghNgwtQAbegCuoAgyfwAt7AyHg2Xo1342NSmjOmPfvgD4zPX4sTnxY=</latexit>
225
+ p6, q6
226
+ <latexit sha1_base64="KwG4fbN+IAWqw0yIdyx7ph0j6fs=">AC3icbVDLSsNAFJ3UV62vqEtFhbBhZRExMeu6MZlC/YBbQiTyaQdOsnEmYlQpeCG3/FjQuLuPUH3PkN/oSTtotaPTDM4Zx7ufceL2ZUKsv6MnILi0vLK/nVwtr6xuaWub3
227
+ TkDwRmNQxZ1y0PCQJoxGpK6oYacWCoNBjpOn1rzO/eU+EpDy6VYOYOCHqRjSgGCktuWax43Hmy0GovzQeumfHcFa504prlqyNQb8S+wpKVX2R7Xvh4NR1TU/Oz7HSUgihRmSsm1bsXJSJBTFjAwLnUSGOE+6pK2phEKiXTS8S1DeKgVHwZc6BcpOFZnO1IUymw5XRki1ZPzXib+57UTFVw4KY3iRJEITwYFCYOKwywY6FNBsGIDTRAWVO8KcQ8JhJWOr6BDsOdP/ksaJ2X7tHxZs0uVKzBHuyBIjgCNjgHFXADqAOMHgEz+AVjIwn48V4M94npTlj2rMLfsH4+AGalZ8g</latexit>
228
+ P4, Q4
229
+ <latexit sha1_base64="NGcTgvPm34Z0bQRxgvkRAHF23E4=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoixceu6MZlC/YBTQiTyaQdOnkwMxFKyNK/cONfdO3GhSLirt/gTzhpu9DaC8MczrmXe+5xY0aFNIyxVlhaXldK6XNja3tnf03b2
230
+ WiBKOSRNHLOIdFwnCaEiakpGOjEnKHAZabuD21xvPxAuaBTey2FM7AD1QupTjKSiHP3CciPmiWGgvtQKkOxjxNJ6ljnVM7hQa+Sao5eNijEp+B+YM1CuHY4a349Ho7qjf1lehJOAhBIzJETXNGJp4hLihnJSlYiSIzwAPVIV8EQBUTY6eS+DJ4oxoN+xNULJZywvydSFIjcpurMXYp5LScXad1E+ld2SsM4kSTE0V+wqCMYB4W9CgnWLKhAghzqrxC3EcYakiLakQzPmT/4PWecWsVq4bZrl2A6ZVBAfgGJwCE1yCGrgDdAEGDyBF/AG3rVn7VX70D6nrQVtNrMP/pQ2/gEqRagA</latexit>
231
+ P5, Q5
232
+ <latexit sha1_base64="I1WyzaQL94IsEQIjMxbBVPfB/oY=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoiPndFNy5bsA9oQphMJu3QyYOZiVBClv6FG/+iazcuFBF3/QZ/wknrQteGOZwzr3c48bMyqkYy0wsLi0vJKcbW0tr6xuaVv7
233
+ zRFlHBMGjhiEW+7SBGQ9KQVDLSjlBgctIy+3f5nrgXBo/BeDmJiB6gbUp9iJBXl6BeWGzFPDAL1pVaAZA8jltayzDk/gXO1eq45etmoGOCs8D8BeXq/rD+/XgwrDn6l+VFOAlIKDFDQnRMI5Z2irikmJGsZCWCxAj3UZd0FAxRQISdju/L4JFiPOhHXL1QwjH7dyJFgchtqs7cpZjWcnKe1kmkf2WnNIwTSUI8WeQnDMoI5mFBj3KCJRsogDCnyivEPcQRlirSkgrBnD5FjRPK+Z5bpulqs3YFJFsAcOwTEwSWogjtQAw2AwRN4AW/gXvWXrUP7XPSWtB+Z3bBv9JGPy1pqAI=</latexit>
234
+ P6, Q6
235
+ <latexit sha1_base64="tGy7FfHzImGqZJ+0hCm6mVtXxk=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoiUnVXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0XWjthWEO59zLPfe4MaNCGsZYKywtr6yuFdLG5tb2zv67l5
236
+ LRAnHpIkjFvGOiwRhNCRNSUjnZgTFLiMtN3Bba63HwgXNArv5TAmdoB6IfUpRlJRjl613Ih5YhioL7UCJPsYsbSeZU71DC7UGrnm6GWjYkwK/gfmDJRrh6PG9+PRqO7oX5YX4SQgocQMCdE1jVjaKeKSYkaykpUIEiM8QD3SVTBEARF2OrkvgyeK8aAfcfVCSfs74kUBSK3qTpzl2Jey8lFWjeR/pWd0jBOJAnxdJGfMCgjmIcFPcoJlmyoAMKcKq8Q9xFHWKpISyoEc/7k/6B1XjEvKtcNlcYNmFYRHIBjcApMcAlq4A7UQRNg8ARewBt41561V+1D+5y2FrTZzD74U9r4BzA9qAM=</latexit>
237
+ P7, Q7
238
+ <latexit sha1_base64="JI1vorLiGfMRP0ugcJWOqAPkm+M=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoiYnVXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0XWjthWEO59zLPfe4MaNCGsZYKywtr6yuFdLG5tb2zv67l5L
239
+ RAnHpIkjFvGOiwRhNCRNSUjnZgTFLiMtN3Bba63HwgXNArv5TAmdoB6IfUpRlJRjn5puRHzxDBQX2oFSPYxYmk9y5zqGVyoNXLN0ctGxZgU/A/MGSjXDkeN78ejUd3RvywvwklAQokZEqJrGrG0U8QlxYxkJSsRJEZ4gHqkq2CIAiLsdHJfBk8U40E/4uqFEk7Y3xMpCkRuU3XmLsW8lpOLtG4i/Ss7pWGcSBLi6SI/YVBGMA8LepQTLNlQAYQ5V4h7iOsFSRlQI5vzJ/0HrvGJeVK4bZrl2A6ZVBAfgGJwCE1RBDdyBOmgCDJ7AC3gD79qz9qp9aJ/T1oI2m9kHf0ob/wAzsagG</latexit>
240
+ P8, Q8
241
+ <latexit sha1_base64="8wbopYeLFJZVZxEdN6vwTq36xvM=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoionVXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0XWjthWEO59zLPfe4MaNCGsZYKywtr6yuFdLG5tb2zv67l5
242
+ LRAnHpIkjFvGOiwRhNCRNSUjnZgTFLiMtN3Bba63HwgXNArv5TAmdoB6IfUpRlJRjn5puRHzxDBQX2oFSPYxYmk9y5zqGVyoNXLN0ctGxZgU/A/MGSjXDkeN78ejUd3RvywvwklAQokZEqJrGrG0U8QlxYxkJSsRJEZ4gHqkq2CIAiLsdHJfBk8U40E/4uqFEk7Y3xMpCkRuU3XmLsW8lpOLtG4i/aqd0jBOJAnxdJGfMCgjmIcFPcoJlmyoAMKcKq8Q9xFHWKpISyoEc/7k/6B1XjEvKtcNs1y7AdMqgNwDE6BCa5ADdyBOmgCDJ7AC3gD79qz9qp9aJ/T1oI2m9kHf0ob/wA21agI</latexit>
243
+ P9, Q9
244
+ <latexit sha1_base64="CiOXOvc/hpLGB3YzgzHfSe57YrA=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkoionZXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0XWjthWEO59zLPfe4MaNCGsZYKywtr6yuFdLG5tb2zv67l5
245
+ LRAnHpIkjFvGOiwRhNCRNSUjnZgTFLiMtN3Bba63HwgXNArv5TAmdoB6IfUpRlJRjn5puRHzxDBQX2oFSPYxYmk9y5zqGVyoNXLN0ctGxZgU/A/MGSjXDkeN78ejUd3RvywvwklAQokZEqJrGrG0U8QlxYxkJSsRJEZ4gHqkq2CIAiLsdHJfBk8U40E/4uqFEk7Y3xMpCkRuU3XmLsW8lpOLtG4i/Ws7pWGcSBLi6SI/YVBGMA8LepQTLNlQAYQ5V4h7iOsFSRlQI5vzJ/0HrvGJeVKoNs1y7AdMqgNwDE6BCa5ADdyBOmgCDJ7AC3gD79qz9qp9aJ/T1oI2m9kHf0ob/wA5+agK</latexit>
246
+ P10, Q10
247
+ <latexit sha1_base64="JIifQ6gP2q/bqPt/JVjzE8cWt0Y=">ACJXicbVDLSsNAFJ3UV62vqEtFBovgQkoigouim5ctmAf0IQwmUzaoZMHMxOhCz9ETf+Qj/BjQuLCK78AH/CSduFtr0wzOGce7nHjdmVEjD+NIKS8srq2vF9dLG5tb2jr671xR
248
+ RwjFp4IhFvO0iQRgNSUNSyUg75gQFLiMt3+X61HwgWNwgc5iIkdoG5IfYqRVJSj31huxDwxCNSXWgGSPYxYWsyJzWN7AwulOtT2dHLRsUYF5wH5hSUq4fD+s/T0bDm6CPLi3ASkFBihoTomEYs7RxSTEjWclKBIkR7qMu6SgYoAIOx1fmcETxXjQj7h6oYRj9u9EigKRO1WduVExq+XkIq2TSP/KTmkYJ5KEeLITxiUEcwjgx7lBEs2UABhTpVXiHuIyxVsCUVgjl78jxonlfMi8p13SxXb8GkiuAHINTYIJLUAX3oAYaAINn8ArewUh70d60D+1z0lrQpjP74F9p378KEKqG</latexit>
249
+ P11, Q11
250
+ <latexit sha1_base64="AhxFdRoafdJMlBFtfVTcJh7VuMA=">ACJXicbVDLSsNAFJ3UV62vqEtFBovgQkoigouim5ctmAf0IQwmUzaoZMHMxOhCz9ETf+Qj/BjQuLCK78AH/CSduFtr0wzOGce7nHjdmVEjD+NIKS8srq2vF9dLG5tb2jr67
251
+ 1xRwjFp4IhFvO0iQRgNSUNSyUg75gQFLiMt3+X61HwgWNwgc5iIkdoG5IfYqRVJSj31huxDwxCNSXWgGSPYxYWsyJzXN7AwulOtT2dHLRsUYF5wH5hSUq4fD+s/T0bDm6CPLi3ASkFBihoTomEYs7RxSTEjWclKBIkR7qMu6SgYoAIOx1fmcETxXjQj7h6oYRj9u9EigKRO1WduVExq+XkIq2TSP/KTmkYJ5KEeLITxiUEcwjgx7lBEs2UABhTpVXiHuIyxVsCUVgjl78jxonlfMi8p13SxXb8GkiuAHINTYIJLUAX3oAYaAINn8ArewUh70d60D+1z0lrQpjP74F9p378NOaqI</latexit>
252
+ P12, Q12
253
+ <latexit sha1_base64="46XsXM3IJrv4KoylzA9I+4y9lFw=">ACJXicbVDLSsNAFJ3UV62vqEtFBovgQkpSBVcFN24bME+oAlhMpm0QycPZiZCVn6I278hX6CGxcWEVz5Af6Ek7YLrb0wzOGce7nHjdmVEjD+NQKS8srq2vF9dLG5tb2jr671x
254
+ JRwjFp4ohFvOMiQRgNSVNSyUgn5gQFLiNtd3Cb6+0HwgWNwns5jIkdoF5IfYqRVJSjX1tuxDwxDNSXWgGSfYxYWs8yJzWr2RlcKDdmsqOXjYoxKfgfmDNQrh2OGt+PR6O6o48tL8JQEKJGRKiaxqxtFPEJcWMZCUrESRGeIB6pKtgiAIi7HRyZQZPFONBP+LqhRJO2N8TKQpE7lR15kbFvJaTi7RuIv1LO6VhnEgS4ukiP2FQRjCPDHqUEyzZUAGEOVeIe4jrBUwZUCOb8yf9Bq1oxzytXDbNcuwHTKoIDcAxOgQkuQA3cgTpoAgyewAt4A2PtWXvV3rWPaWtBm83sgz+lf0AEGKqig=</latexit>
255
+ p4, q4
256
+ <latexit sha1_base64="PfDf0azKs219i32pWKcsBWEfPAo=">AC3icbVDLSsNAFJ34rPUVdanI0CK4kJIQd0V3bhswT6gDWEymbRDJ5k4MxFK6FJw46+4cWERt/6AO7/Bn3DSdlFbDwxzOde7r3HixmVyrK+jaXldW19dxGfnNre2fX3NtvS
257
+ J4ITOqYMy5aHpKE0YjUFVWMtGJBUOgx0vT6N5nfCBCUh7dqUFMnB1IxpQjJSWXLPQ8Tjz5SDUXxoP3fIZnFXuteKaRatkjQEXiT0lxcrRqPbzeDyquZXx+c4CUmkMENStm0rVk6KhKYkWG+k0gSI9xHXdLWNEIhkU46vmUIT7Tiw4AL/SIFx+psR4pCmS2nK0OkenLey8T/vHaigksnpVGcKBLhyaAgYVBxmAUDfSoIVmygCcKC6l0h7iGBsNLx5XUI9vzJi6RxXrLpauaXaxcgwly4BAUwCmwQWogFtQBXWAwRN4AW9gZDwbr8a78TEpXTKmPQfgD4zPX5Rhnxw=</latexit>
258
+ p5, q5
259
+ <latexit sha1_base64="pFcmK4pQavogwo5Qgia2L9TsCjY=">AC3icbVDLSsNAFJ3UV62vqEtFhbBhZREFHVXdOyBfuANoTJZNIOnWTizEQoUvBjb/ixoVF3PoD7vwGf8J20WtHhjmcM693HuPFzMqlWV9GbmFxaXlfxqYW19Y3PL3N5pSJ4ITOqYMy5aHpKE0YjUFVWMtGJBUOgx0vT615nfvCdCUh7dqkFMnB1IxpQjJSWXLPY8Tjz5SDUXxoP3bNjOKvcacU1S1bZGgP+JfaUlCr7o9r3w8Go6pqfHZ/jJCSRwgxJ2batWDkpEopiRoaFTiJjHAfdUlb0wiFRDrp+JYhPNSKDwMu9I
260
+ sUHKuzHSkKZbacrgyR6sl5LxP/89qJCi6clEZxokiEJ4OChEHFYRYM9KkgWLGBJgLqneFuIcEwkrHV9Ah2PMn/yWNk7J9Wr6s2aXKFZgD/ZAERwBG5yDCrgBVAHGDyCZ/AKRsaT8WK8Ge+T0pwx7dkFv2B8/ACXe58e</latexit>
261
+ p7, q7
262
+ <latexit sha1_base64="tnHfbmSvg0HeAm+CKukTGPaJaY=">AC3icbVDLSsNAFJ34rPUVdanI0CK4kJKIUN0V3bhswT6gDWEymbRDJ5k4MxFK6FJw46+4cWERt/6AO7/Bn3DSdlFbDwxzOde7r3HixmVyrK+jaXldW19dxGfnNre2fX3Nt
263
+ vSJ4ITOqYMy5aHpKE0YjUFVWMtGJBUOgx0vT6N5nfCBCUh7dqUFMnB1IxpQjJSWXLPQ8Tjz5SDUXxoP3fIZnFXuteKaRatkjQEXiT0lxcrRqPbzeDyquZXx+c4CUmkMENStm0rVk6KhKYkWG+k0gSI9xHXdLWNEIhkU46vmUIT7Tiw4AL/SIFx+psR4pCmS2nK0OkenLey8T/vHaigksnpVGcKBLhyaAgYVBxmAUDfSoIVmygCcKC6l0h7iGBsNLx5XUI9vzJi6RxXrIvSlc1u1i5BhPkwCEogFNgzKogFtQBXWAwRN4AW9gZDwbr8a78TEpXTKmPQfgD4zPX52vnyI=</latexit>
264
+ p8, q8
265
+ <latexit sha1_base64="epo4Ir38SV16TbTwKNAhVqXnxk=">AC3icbVDLSsNAFJ34rPUVdanI0CK4kJKIYN0V3bhswT6gDWEymbRDJ5M4MxFK6FJw46+4cWERt/6AO7/Bn3DSdlFbDwxzOde7r3HixmVyrK+jaXldW19dxGfnNre2fX3NtvyCgRmNRxCLR8pAkjHJSV1Qx0oFQaHSNPr32R+84EISN+pwYxcULU5TSgGCktuWah40XMl4NQf2k8dMtncFa514prFq2SNQZcJPaUFCtHo9rP4/Go6pfHT/CSUi4wgxJ2batWDkpEopiRob5TiJjHAfdUlbU45CIp10fMsQnmjFh0Ek9OMKjt
266
+ XZjhSFMltOV4ZI9eS8l4n/e1EBWUnpTxOFOF4MihIGFQRzIKBPhUEKzbQBGFB9a4Q95BAWOn48joEe/7kRdI4L9kXpauaXaxcgwly4BAUwCmwSWogFtQBXWAwRN4AW9gZDwbr8a78TEpXTKmPQfgD4zPX6DJnyQ=</latexit>
267
+ p9, q9
268
+ <latexit sha1_base64="Lj1dGxs0OP1UCHefGjQ0m0AFVjs=">AC3icbVDLSsNAFJ34rPUVdanI0CK4kJKIoN0V3bhswT6gDWEymbRDJ5M4MxFK6FJw46+4cWERt/6AO7/Bn3DSdlFbDwxzOde7r3HixmVyrK+jaXldW19dxGfnNre2fX3Nt
269
+ vyCgRmNRxCLR8pAkjHJSV1Qx0oFQaHSNPr32R+84EISN+pwYxcULU5TSgGCktuWah40XMl4NQf2k8dMtncFa514prFq2SNQZcJPaUFCtHo9rP4/Go6pfHT/CSUi4wgxJ2batWDkpEopiRob5TiJjHAfdUlbU45CIp10fMsQnmjFh0Ek9OMKjtXZjhSFMltOV4ZI9eS8l4n/e1EBVdOSnmcKMLxZFCQMKgimAUDfSoIVmygCcKC6l0h7iGBsNLx5XUI9vzJi6RxXrIvSuWaXaxcgwly4BAUwCmwSWogFtQBXWAwRN4AW9gZDwbr8a78TEpXTKmPQfgD4zPX6PjnyY=</latexit>
270
+ p10, q10
271
+ <latexit sha1_base64="4Qbrgr2riCekHI8Ziewjbr6x+8w=">ACEXicbVDLSsNAFJ34rPUVdanIYBG6kJKIoO6Kbly2YB/QhjCZTNqhk4czE6GELN268Vdc1IUibt258xv8CSdNF7X1wDCHc+7l3nuciFEhDeNbW1hcWl5ZLawV1zc2t7b1nd2mCGOSQOHL
272
+ ORtBwnCaEAakpG2hEnyHcYaTmD68xv3RMuaBjcymFELB/1AupRjKSbL3cdULmiqGviRK7cQ0hM4Ld7loq2XjIoxBpwn5oSUqgej+s/D4ahm619dN8SxTwKJGRKiYxqRtBLEJcWMpMVuLEiE8AD1SEfRAPlEWMn4ohQeK8WFXsjVCyQcq9MdCfJFtp+q9JHsi1kvE/zOrH0LqyEBlEsSYDzQV7MoAxhFg90KSdYsqEiCHOqdoW4jzjCUoVYVCGYsyfPk+ZpxTyrXNbNUvUK5CiAfXAEysAE56AKbkANAGj+AZvI37Ul70d61j7x0QZv07IE/0D5/AUNMoaI=</latexit>
273
+ p11, q11
274
+ <latexit sha1_base64="KBzJnFLiRLRwcByMfRML5Asryo=">ACEXicbVDLSsNAFJ34rPUVdanIYBG6kJKIoO6Kbly2YB/QhjCZTNqhk4czE6GELN268Vdc1IUibt258xv8CSdNF7X1wDCHc+7l3nuciFEhDeNbW1hcWl5ZLawV1zc2t7b1nd2mCGOSQO
275
+ HLORtBwnCaEAakpG2hEnyHcYaTmD68xv3RMuaBjcymFELB/1AupRjKSbL3cdULmiqGviRK7cQ0xM4Ld7loq2XjIoxBpwn5oSUqgej+s/D4ahm619dN8SxTwKJGRKiYxqRtBLEJcWMpMVuLEiE8AD1SEfRAPlEWMn4ohQeK8WFXsjVCyQcq9MdCfJFtp+q9JHsi1kvE/zOrH0LqyEBlEsSYDzQV7MoAxhFg90KSdYsqEiCHOqdoW4jzjCUoVYVCGYsyfPk+ZpxTyrXNbNUvUK5CiAfXAEysAE56AKbkANAGj+AZvI37Ul70d61j7x0QZv07IE/0D5/AUZroaQ=</latexit>
276
+ p12, q12
277
+ <latexit sha1_base64="dWxgl5imBz+iP1N8FA6fR6Jc7yY=">ACEXicbVC7SgNBFJ2Nrxhfq5aKDAYhYTdIKhd0MYyAfOAZFlmZyfJkNmHM7NCWLa0tfFXLGKhiK2dnd/gTzibTRETDwxzOde7r3HCRkV0jC+tdzS8srqWn69sLG5tb2j7+41RBxTBo4Y
278
+ AFvO0gQRn3SkFQy0g45QZ7DSMsZXqd+65wQP/Vo5CYnmo79MexUgqydZLXSdgrh56ovDxI7NSnIKZ8W7TLT1olE2JoCLxJySYvVwXP95OBrXbP2r6wY48ogvMUNCdEwjlFaMuKSYkaTQjQJER6iPuko6iOPCueXJTAE6W4sBdw9XwJ+psR4w8ke6nKj0kB2LeS8X/vE4kexdWTP0wksTH2aBexKAMYBoPdCknWLKRIghzqnaFeIA4wlKFWFAhmPMnL5JmpWyelS/rZrF6BTLkwQE4BiVgnNQBTegBhoAg0fwDF7Bm/akvWjv2kdWmtOmPfvgD7TPX0mKoaY=</latexit>
279
+ P3, Q3
280
+ <latexit sha1_base64="6EOuKvteBQqnMgvSnp+f414+4jo=">ACH3icbVDLSsNAFJ3UV62vqEtFBovgQkqi4mNXdOyBfuAJoTJZNIOnTyYmQglZOlfuPEvunbjQhFx12/wJ5y0LrTthWEO59zLPfe4MaNCGsZIKywsLi2vFdLa+sbm1v69k5
281
+ TRAnHpIEjFvG2iwRhNCQNSUj7ZgTFLiMtNz+ba63HgXNArv5SAmdoC6IfUpRlJRjn5huRHzxCBQX2oFSPYwYmkty5yzEzhXq+eao5eNijEuOAvMX1Cu7g/r348Hw5qjf1lehJOAhBIzJETHNGJp4hLihnJSlYiSIxwH3VJR8EQBUTY6fi+DB4pxoN+xNULJRyzfydSFIjcpurMXYpLSfnaZ1E+ld2SsM4kSTEk0V+wqCMYB4W9CgnWLKBAghzqrxC3EMcYakiLakQzOmTZ0HztGKeV67rZrl6AyZVBHvgEBwDE1yCKrgDNdAGDyBF/AG3rVn7VX70D4nrQXtd2YX/Ct9AMnIaf+</latexit>
282
+ 12
283
+ Fig. 1.
284
+ A radial 13-bus distribution network connected with rooftop solar
285
+ PVs and ESSs.
286
+ In this paper, one control objective is to minimize the
287
+ total power loss of the distribution network by controlling the
288
+ dynamics of PVs and ESSs, which is approximated by
289
+ f1(pg
290
+ 1, . . . , pg
291
+ n) =
292
+
293
+ lij∈L
294
+ rij
295
+ �∥Pij∥2
296
+ 2 + ∥Qij∥2
297
+ 2
298
+ V 2
299
+ 0
300
+
301
+ (2)
302
+ where V0 denotes the nominal voltage magnitude, pg
303
+ j, Pij, and
304
+ Qij ∈ RT are augmented vectors of pg
305
+ j, Pij, and Qij across T
306
+ time intervals, respectively. Note that we only consider active
307
+ power loss and assume reactive power flows Qij to be constant
308
+ vectors. Though the reactive power loss is not included here for
309
+ simplicity, it can be added without affecting algorithm design.
310
+ The active power flows are constrained by
311
+ 0 ≤ Pij ≤ Pij
312
+ (3)
313
+ where Pij denotes the maximum active power flow limit.
314
+ B. Solar Photovoltaic
315
+ Let V denote the set of in total V solar PVs. During T time
316
+ intervals of a day, the active power injection ˜pν ∈ RT from
317
+ the νth PV inverter should satisfy
318
+ 0 ≤ ˜pν ≤ pv
319
+ ν
320
+ (4)
321
+
322
+ 田田3
323
+ where pv
324
+ ν denotes the maximum active power injection and is
325
+ assumed to be known by the forecast. Herein, the curtailment
326
+ cost can be calculated by [25]
327
+ f2(˜pν) = ∥˜pν − pv
328
+ ν∥2
329
+ 2.
330
+ (5)
331
+ C. Energy Storage System
332
+ Let S denote the set of E ESSs. The charging/discharging
333
+ power ˆpσ ∈ RT of the σth ESS is constrained by
334
+ − ps
335
+ σ ≤ ˆpσ ≤ ps
336
+ σ
337
+ (6)
338
+ where ps
339
+ σ and ps
340
+ σ denote the maximum discharging and
341
+ charging power, respectively. Let s0
342
+ σ denote the initial state of
343
+ charge (SoC) of the σth ESS and Hσ ≜ [s0
344
+ σ, . . . , s0
345
+ σ]T ∈ RT .
346
+ Aggregate the charging/discharging power across T time in-
347
+ tervals, then the capacity of the σth ESS is constrained by
348
+ pa
349
+ σ ≤ Hσ + Aˆpσ∆T ≤ pa
350
+ σ
351
+ (7)
352
+ where pa
353
+ σ and pa
354
+ σ denote its lower and upper capacity
355
+ bounds, respectively, ∆T denotes the sampling time, and
356
+ the aggregation matrix A is lower triangular consisting of
357
+ ones and zeros, i.e., element Aˆı,ˆȷ = 1 if ˆı ≥ ˆȷ, element
358
+ Aˆı,ˆȷ = 0 if ˆı < ˆȷ, ∀ˆı, ˆȷ = 1, . . . , T. Therefore, the SoCs of
359
+ ESS σ during T time slots are obtained by aggregating the
360
+ charging/discharging power using A.
361
+ Furthermore, the σth ESS’s degradation cost is calculated in
362
+ terms of the smoothness of charging and discharging by [26]
363
+ f3(ˆpσ) = ∥B ˆpσ∥2
364
+ 2.
365
+ (8)
366
+ where B calculates discharging/charging differences between
367
+ adjacent times, i.e., Bˆı,ˆı = 1, ∀ˆı = 1, . . . , T, Bˆı,ˆı+1 =
368
+ −1, ∀ˆı = 1, . . . , T − 1, and all other elements are zeros.
369
+ D. Problem Formulation
370
+ The optimization problem is then formulated to minimize
371
+ the summation of total active power loss, PV curtailment cost,
372
+ and ESS degradation cost within the distribution network as
373
+ min
374
+ ˜p, ˆp
375
+ δ1f1(pg) +
376
+ V
377
+
378
+ ν=1
379
+ δ2f2(˜pν) +
380
+ E
381
+
382
+ σ=1
383
+ δ3f3(ˆpσ)
384
+ s.t.
385
+ (1a), (3), (4), (6), (7)
386
+ (P1)
387
+ where
388
+ ˜p
389
+ =
390
+ [˜pT
391
+ 1 , . . . , ˜pT
392
+ n]T,
393
+ ˆp
394
+ =
395
+ [ˆpT
396
+ 1 , . . . , ˆpT
397
+ n]T, pg
398
+ =
399
+ [pg
400
+ 1
401
+ T, . . . , pg
402
+ n
403
+ T]T, and δα denotes the cost coefficient asso-
404
+ ciated with the objective function fα(·). Note that the cost
405
+ coefficients are constants that allow flexible adjustments on
406
+ the weights of the global and local objective functions and
407
+ regulate different units.
408
+ III. DECENTRALIZED OPTIMIZATION
409
+ A. Projected Gradient Method
410
+ This paper achieves scalability in solving (P1) via projected
411
+ gradient method (PGM). PGM decomposes a centralized opti-
412
+ mization problem into local optimizations at agents, resulting
413
+ in a paralleled computing structure. Let M = {1, . . . , m}
414
+ denote the set of agents, e.g., buses or DERs, who work
415
+ cooperatively in solving (P1). In this setting, the κth agent
416
+ updates its decision variable xκ using PGM by
417
+ x(ℓ+1)
418
+ κ
419
+ = PXκ[x(ℓ)
420
+ κ − γ(ℓ)
421
+ κ Φκ(x(ℓ))]
422
+ (9)
423
+ where
424
+
425
+ denotes
426
+ the
427
+ iteration
428
+ number,
429
+ x(ℓ)
430
+ =
431
+ [x(ℓ)
432
+ 1
433
+ T, . . . , x(ℓ)
434
+ m
435
+ T]T
436
+ includes
437
+ all
438
+ decision
439
+ variables,
440
+ i.e.,
441
+ ˜pν and ˆpσ in problem (P1), γ(ℓ)
442
+ k
443
+ denotes the step size, Φκ(·)
444
+ denotes the gradient of the Lagrangian w.r.t. x(ℓ)
445
+ κ , and PXκ[·]
446
+ denotes the projection operation onto set Xκ.
447
+ In (P1), the local constraint of the νth PV in (4) and local
448
+ constraints of the σth ESS in (6) and (7) can be represented
449
+ by two feasible sets Pv
450
+ ν and Pe
451
+ σ as
452
+ Pv
453
+ ν ≜ {˜pν| 0 ≤ ˜pν ≤ pv
454
+ ν}
455
+ (10a)
456
+ Pe
457
+ σ ≜ {ˆpσ| − ps
458
+ σ≤ˆpσ≤ps
459
+ σ, pa
460
+ σ≤H+Aˆpσ∆T ≤ pa
461
+ σ}. (10b)
462
+ In what follows, aiming at reducing the number of coupling
463
+ terms, we rewrite the networked constraints in (1a) and (3) to
464
+ a single inequality constraint based on the network topology.
465
+ To this end, we first represent the active power flows in (1a)
466
+ through active power generations of each bus using
467
+ pi = ˜pi − ˆpi − pc
468
+ i
469
+ (11)
470
+ where pi denotes the aggregated active power generation at
471
+ bus i, ˜pi = �Vi
472
+ ν=1 ˜pν and ˆpi = �Ei
473
+ σ=1 ˆpσ denote the aggre-
474
+ gated active power of all PVs and ESSs that are connected at
475
+ bus i, respectively. Vi and Ei denote the total number of PVs
476
+ and ESSs connected at bus i, respectively.
477
+ For the ιth line flow Pι in the distribution network, the
478
+ from-bus is defined by the bus where the flow begins, and the
479
+ to-bus set is defined by the set of buses that the ιth line flow
480
+ travels to till reaching the edge of the distribution network.
481
+ Let Z ∈ Rn×n denote the adjacency matrix of the distribution
482
+ network and Zι denote the ιth row of Z that represents the
483
+ adjacency vector of the ιth line flow. Let Zι(i) denote the ith
484
+ element of Zι, and Zι(i) = 1 if the ιth power flow has bus i
485
+ as a to-bus, e.g., Z9 = [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0]. Then, the
486
+ power flows in the distribution network can be represented by
487
+ Z. Expand Z across T time slots, we have
488
+ ˜Z =
489
+
490
+ ����
491
+ Z1(1)I Z1(2)I · · · Z1(n)I
492
+ ...
493
+ ...
494
+ ...
495
+ Zn(1)I Zn(2)I · · · Zn(n)I
496
+
497
+ ����
498
+ (12)
499
+ where I∈RT ×T denotes the identity matrix and ˜Z ∈ RnT ×nT .
500
+ In what follows, let ˜P ∈ RnT denote the aggregated active
501
+ power generations defined in (11) from all buses, we have
502
+ ˜P =
503
+
504
+ ��
505
+ p1
506
+ ...
507
+ pn
508
+
509
+ �� =
510
+
511
+ ����
512
+ �V1
513
+ ν=1 ˜pν − �E1
514
+ σ=1 ˆpσ − pc
515
+ 1
516
+ ...
517
+ �Vn
518
+ ν=Vn−1+1 ˜pν − �En
519
+ σ=En−1+1 ˆpσ − pc
520
+ n
521
+
522
+ ���� .
523
+ (13)
524
+ Furthermore, ˜P can be rewritten compactly as
525
+ ˜P =
526
+ n
527
+
528
+ i=1
529
+ ∆i (˜pi − ˆpi − pc
530
+ i)
531
+ (14)
532
+
533
+ 4
534
+ where ∆i denotes the aggregation matrix whose ith block is
535
+ represented by the identity matrix I, and all other blocks are
536
+ zeros, e.g., ∆1 = [I, 0, . . . , 0]T ∈ RnT ×T . Then, the active
537
+ power flow of the ιth line can be calculated by
538
+ Pι = ˜Zι ˜P .
539
+ (15)
540
+ Consequently, the power flow limit constraint in (3) becomes
541
+ 0 ≤ ˜Zι ˜P ≤ Pι.
542
+ (16)
543
+ Therefore, problem (P1) can be written into
544
+ min
545
+ ˜p, ˆp
546
+ δ1f1(pg) +
547
+ V
548
+
549
+ ν=1
550
+ δ2f2(˜pν) +
551
+ E
552
+
553
+ σ=1
554
+ δ3f3(ˆpσ)
555
+ s.t.
556
+ pν ∈ Pv
557
+ ν, ∀ν ∈ V
558
+ pσ ∈ Pe
559
+ σ, ∀σ ∈ S
560
+ 0 ≤ ˜Zι ˜P ≤ Pι, ∀ι ∈ L
561
+ (P2)
562
+ The optimization problem in (P2) seeks to find the optimal
563
+ decision variables, i.e., charging and discharging power ˜pσ’s
564
+ of the ESSs and the active power injection ˆpν’s of the PVs. In
565
+ what follows, we focus on solving (P2) through a decentralized
566
+ fashion based on PGM defined in (9). To solve (P2) via PGM,
567
+ we firstly derive its relaxed Lagrangian as
568
+ L(˜p, ˆp, µl, µu) = δ1f1(pg) +
569
+ V
570
+
571
+ ν=1
572
+ δ2f2(˜pν) +
573
+ E
574
+
575
+ σ=1
576
+ δ3f3(ˆpσ)
577
+ +
578
+ L
579
+
580
+ ι=1
581
+ µT
582
+ uι( ˜Zι ˜P − Pι)−
583
+ L
584
+
585
+ ι=1
586
+ µT
587
+ lι ˜Zι ˜P (17)
588
+ where µl = [µT
589
+ l1, . . . , µT
590
+ lL]T and µu = [µT
591
+ u1, . . . , µT
592
+ uL]T, µlι
593
+ and µuι denote the dual variables associated with lower and
594
+ upper power flow limits of the line ι, respectively.
595
+ Suppose ˜pν and ˆpσ are decision variables of the νth PV and
596
+ σth ESS connected at bus i, respectively. Take the subgradients
597
+ of (17) w.r.t. the primal variables ˜pν and ˆpσ, we have
598
+ ∇ ˜pνL(·) = 2δ2(˜pν − pv
599
+ ν) + 2δ1
600
+ V 2
601
+ 0
602
+ L
603
+
604
+ ι=1
605
+ rι( ˜Zι∆i)T( ˜Zι ˜P )
606
+ +
607
+ L
608
+
609
+ ι=1
610
+ ( ˜Zι∆i)T(µuι − µlι)
611
+ (18a)
612
+ ∇ ˆpσL(·) = 2δ3 ˆpσ − 2δ1
613
+ V 2
614
+ 0
615
+ L
616
+
617
+ ι=1
618
+ rι( ˜Zι∆i)T( ˜Zι ˜P )
619
+
620
+ L
621
+
622
+ ι=1
623
+ ( ˜Zι∆i)T(µuι − µlι).
624
+ (18b)
625
+ Without affecting the efficacy of the algorithm design, we
626
+ assume all power lines have the same resistance ¯r for the
627
+ simplicity of presentation, herein (18) becomes
628
+ ∇ ˜pνL(·) = 2δ2(˜pν − pv
629
+ ν) + ¯δ1πi ˜P + ψi(µu − µl)
630
+ (19a)
631
+ ∇ ˆpσL(·) = 2δ3 ˆpσ − ¯δ1πi ˜P − ψi(µu − µl)
632
+ (19b)
633
+ where ¯δ1 = 2δ1
634
+ V 2
635
+ 0 ¯r, πi = �L
636
+ ι=1( ˜Zι∆i)T ˜Zι, and ψi denotes the
637
+ ith column block of ˜Z.
638
+ The detailed derivation of the Lagrangian subgradients in
639
+ (19) can be found in APPENDIX A.
640
+ Therefore, based on the calculated subgradients in (18), at
641
+ the ℓth iteration, the νth PV and the σth ESS can update their
642
+ decision variables using PGM by
643
+ ˜p(ℓ+1)
644
+ ν
645
+ = ΠPvν
646
+
647
+ ˜p(ℓ)
648
+ ν
649
+ − αv
650
+ ν,ℓ∇ ˜pνL(ℓ) (·)
651
+
652
+ (20a)
653
+ ˆp(ℓ+1)
654
+ σ
655
+ = ΠPeσ
656
+
657
+ ˆp(ℓ)
658
+ σ − αe
659
+ σ,ℓ∇ ˆpσL(ℓ) (·)
660
+
661
+ (20b)
662
+ where αv
663
+ ν,ℓ and αe
664
+ σ,ℓ denote the primal step sizes of the νth PV
665
+ and the σth ESS, respectively, L(ℓ) (·) denotes the calculated
666
+ Lagrangian in (17) at the ℓth iteration. The dual variables can
667
+ be updated similarly using PGM.
668
+ B. DER Aggregation and Control
669
+ In PGM iterations, the ith agent needs to calculate Φi(xℓ)
670
+ in (9) where the decision variables xi’s from all other agents
671
+ are required. As indicated in (19), calculating subgradients
672
+ ∇ ˜pνL(·) and ∇ ˆpσL(·) indeed requires the decision variables
673
+ ˜P from all the agents. Specifically, the calculation of subgra-
674
+ dients in (19a) and (19b) are coupled through
675
+ C = Cp + Cd = ¯δ1πi ˜P + ψi(µu − µl)
676
+ (21)
677
+ where Cp and Cd denote the coupling terms associated with
678
+ the primal and dual variables, respectively.
679
+ To clearly demonstrate the information exchange needs in
680
+ subgradient calculation, we exemplify the primal update of the
681
+ ˆνth PV connected at bus 2. The ˆνth PV can update its decision
682
+ variable ˜pˆν using the subgradient in (19a) which is
683
+ ∇ ˜pˆνL(·) = 2δ2(˜pˆν − pv
684
+ ˆν) +
685
+ 2
686
+
687
+ ι=1
688
+
689
+ ¯δ1πι ˜P + µuι + µlι
690
+
691
+ (22)
692
+ where π1 ˜P = �n
693
+ i=1 pi and π2 ˜P = p2 + p3. Therefore, the
694
+ ˆνth PV requires the active power generations pi, ∀i = 1, . . . , n
695
+ from all buses to conduct the update in (20a).
696
+ Based on the above observations, two different aggregation
697
+ and control strategies, i.e., Bus-level aggregation and control
698
+ and DER-level aggregation and control, can be applied as
699
+ shown in Fig. 2. In bus-level aggregation and control, the ith
700
+ Each bus aggregates the decision
701
+ variables and
702
+ DERs exchange decision variables
703
+ with others to obtain
704
+ and
705
+ ˜pi=
706
+ XVi
707
+ ⌫=1 ˜p⌫
708
+ <latexit sha1_base64="/FNjpGVvIa+B8d61OziIhaF3i9I=">ACR3icdVDLSgMxFM3Ud31VXboJFsFVmRFBXRENy4VbBU6dchkUg3mMSR3hBLm79y4decvuHGhiEsztQufF0IO59yT3HvSXHALYfgY1CYmp6ZnZufq8wu
709
+ LS8uNldWu1YWhrEO10OYiJZYJrlgHOAh2kRtGZCrYeXpzVOnt8xYrtUZDHPWl+RK8QGnBDyVNC5j4CJjLk61yOxQ+svlZlw7Noljm0hExeroh2VsdKCSw720sWSwDUlwnV9Y/nPC5WtTBrNsBWOCv8G0Rg0bhOksZDnGlaSKaACmJtLwpz6DtigFPBynpcWJYTekOuWM9DRSzfTfKocSbnsnwQBt/FOAR+9XhiLTVhL6z2sD+1CryL61XwGCv7jKC2CKfn40KAQGjatQcYNoyCGHhBquJ8V02tiCAUfd2HEP1c+Tfobreindb+6U7z4HAcxyxaRxtoC0VoF
710
+ x2gY3SCOoiO/SEXtBrcB8B2/B+2drLRh71tC3qgUf3V+2Gw=</latexit>
711
+ ˆpi=
712
+ XEi
713
+ �=1 ˆp�
714
+ <latexit sha1_base64="J1PqXdAVQ0WCWiXxcnVQymKUhI=">ACSXicbVDPSxtBFJ6N2qaprWl79DIYCp7CbhG0B0EshR4jGBWycX07mSD82OZeSuEYf+9Xnrf9DLx4s4snZuIf648EwH9/3vpn3vryQwmEc/4laK6tr163Ter97v9H98PHEmdIyPmRGnuWg+NSaD5EgZKfFZaDyiU/zS+/1frpFbdOGH2Mi4KPFcy0mAoGKise5HOAX2aGzlxCxUuX1RVJqjfr2jqSpX51ImZgv2kSrWRQgl05z5VgHMG0n8PvdWLTzS+Kuv24n68LPocJA3okaYGWfd
715
+ 3OjGsVFwjk+DcKIkLHuwKJjkVSctHS+AXcKMjwLUoLgb+2USFf0cmAmdGhuORrpk/3d4UK4eMnTWK7inWk2+pI1KnO6NvdBFiVyzh4+mpaRoaB0rnQjLGcpFAMCsCLNSNgcLDEP4nRBC8nTl5+DkSz/Z6X892ukdHDZxtMkm2SLbJCG75ID8IAMyJIz8JH/JDfkX/Yquo9vo7qG1FTWeT+RtVbuAWP/tA=</latexit>
716
+ Each individual PV or ESS owns
717
+ decision variable or to itself
718
+ ˜p⌫
719
+ <latexit sha1_base64="SM/x7D2mHQVQXsiJ9CAKr6xkWEY=">ACBXicbVC7TsMwFHXKq5RXgBEGiwqJqUpQJWCrYGEsEn1ITRQ5jtadezIdpCqKAsLv8LCAEKs/AMbf4PTZoCWI1k+
720
+ Oude3XtPmDCqtON8W5WV1bX1jepmbWt7Z3fP3j/oKpFKTDpYMCH7IVKEU46mpG+okKA4Z6YWTm8LvPRCpqOD3epoQP0YjTocUI2kwD72NGURybxQsEhNY/NlSZ4HmcfTPLDrTsOZAS4TtyR1UKId2F9eJHAaE64xQ0oNXCfRfoakpiRvOaliQIT9CIDAzlKCbKz2ZX5PDUKBEcCmke13Cm/u7IUKyKDU1ljPRYLXqF+J83SPXw0s8oT1JNOJ4PGqYMagGLSGBEJcGaTQ1BWFKzK8
721
+ RjJBHWJriaCcFdPHmZdM8brNxdest67LOKrgCJyAM+C9ACt6ANOgCDR/AMXsGb9WS9WO/Wx7y0YpU9h+APrM8fVh+Zxg=</latexit>
722
+ ˆp�
723
+ <latexit sha1_base64="yRsJzl/IUYDZq3zl9fswdZn4Y=">ACBnicbVDLSsNAFJ3UV62vqEsRBovgqiRSUHdFNy4r2Ac0IUwm03boTCbMTIQSsnLjr7hxoYhbv8Gdf+OkzUJbDwxzOde
724
+ 7r0nTBhV2nG+rcrK6tr6RnWztrW9s7tn7x90lUglJh0smJD9ECnCaEw6mpG+okiIeM9MLJTeH3HohUVMT3epoQn6NRTIcUI2kwD72xkhnXihYpKbcfFmS50HmKTriKA/sutNwZoDLxC1JHZRoB/aXFwmchJrzJBSA9dJtJ8hqSlmJK95qSIJwhM0IgNDY8SJ8rPZGTk8NUoEh0KaF2s4U393ZIirYklTyZEeq0WvEP/zBqkeXvoZjZNUkxjPBw1TBrWARSYwopJgzaGICyp2RXiMZIa5NczY
725
+ TgLp68TLrnDbfZuLpr1lvXZRxVcAROwBlwQVogVvQBh2AwSN4Bq/gzXqyXqx362NeWrHKnkPwB9bnDw53mik=</latexit>
726
+ The ith bus performs the primal-
727
+ dual updates for all the DERs
728
+ Each DER performs the primal-
729
+ dual update in Eqs. (22) and (23)
730
+ End iteration if : DERs’ decision variables achieve convergence
731
+ Aim: Calculate subgradients and for the updates in PGM
732
+ r˜p⌫L(·)
733
+ <latexit sha1_base64="NCdmlfZtlskHCRoWxGd7tDpT0mQ=">ACIHicbVBNS8NAEN34WetX1aOXYBH0UhIpqDfRiwcPFWwVmlIm61dutkNuxOhPwUL/4VLx
734
+ 4U0Zv+GjdtDn4NLPt4b4Z58JEcIOe9+HMzM7NLyxWlqrLK6tr67WNzY5RqasTZVQ+iYEwSXrI0cBbtJNIM4FOw6HJ0V+vUd04YreYXjhPViuJV8wCmgpfq1w0BCKCfBchFxLIgVCIy49h+WZLnlpdpngcx4JCyC7yvYBGCvf7tbrX8Cbl/gV+CeqkrFa/9h5EiqYxk0gFGNP1vQR7GWjkVLC8GqSGJUBHcMu6FkqImelkwNzd9cykT
735
+ tQ2j6J7oT9PpFBbArTtrNwan5rBfmf1k1xcNTLuExSZJOFw1S4aJyi7TciGtGUYwtAKq59erSIWigaDOt2hD83yf/BZ2Dht9sHF826yenZRwVsk12yB7xySE5IekRdqEknvySJ7Ji/PgPDmvztu0dcYpZ7bIj3I+vwD93qVW</latexit>
736
+ rˆp�L(·)
737
+ <latexit sha1_base64="WTcG5eEwOkl6SUea4MtPTARM=">ACIXicbVBNS8NAEN34bf2qevQSLEK9lEQE9SZ68eBwX5AU8pks2XbnbD7kQoIX/Fi3/FiwdF
738
+ vIl/xk3bg7YOLPt4b4Z58JEcIOe9+UsLC4tr6yurZc2Nre2d8q7ew2jUk1ZnSqhdCsEwSXrI4cBWslmkEcCtYMh9eF3nxk2nAlH3CUsE4Mfcl7nAJaqls+DySEArpZMADMglCJyIxi+2VJnlvW8H4MeR7EgAMKIrvNqwGNFB53yxWv5o3LnQf+FTItO65c8gUjSNmUQqwJi27yXYyUAjp4LlpSA1LAE6hD5rWyghZqaTjS/M3SPLRG5Pafsk
739
+ umP290QGsSl8287CqZnVCvI/rZ1i7yTcZmkyCSdLOqlwkXlFnG5EdeMohZAFRz69WlA9BA0YZasiH4syfPg8ZJzT+tXdyfVi6vpnGskQNySKrEJ2fktyQO1InlDyRF/JG3p1n59X5cD4nrQvOdGaf/Cn+wfAO6W5</latexit>
740
+ ˜pi=
741
+ XV
742
+ ⌫=1 ˜p⌫
743
+ <latexit sha1_base64="sOXuDwMIMEucHRfVxl8qnlY3zU=">ACRHicdVDLSiQxFE05PtXO7N0E2wEV02VCOqiQZyNSwemW6GrLVKptAbzKJbQhPycW78AHfzBW5cKOJWJtX2wueFkM595CTk5eCW4jf9
744
+ HUj+mZ2bn5hcbi0vLKanPtZ8/qylDWpVpoc5oTywRXrAscBDstDSMyF+wkv/xd6ydXzFiu1V8YlWwgybniQ04JBCpr9lPgomAuzbUo7EiGy5XeZ9x1PE5tJTOXqT+FRpwSUHe+ZSeCEuF63n9jr0+a7bidjwe/BkE9BCkznOmrdpoWklmQIqiLX9JC5h4IgBTgXzjbSyrCT0kpyzfoCKSGYHblyCx5uBKfBQm3AU4DH71uGItHXCsFntx+1mvxK61cw3Bs4rsoKmKvDw0rgUHjulFcMoiFEAhBoesmJ6QyhEHpvhBKSj1/+
745
+ DHrb7WSnvf9np3VwOKljHq2jDbSFErSLDtAROkZdRNE1ukMP6DG6ie6jp+j5dXUqmnh+oXcTvfwHsO62FA=</latexit>
746
+ Buses exchange decision variables
747
+ with others to obtain
748
+ 8i=1 . . . , n
749
+ <latexit sha1_base64="eYA2cw+pfmQ4bl+JGXOd6TBo2Y=">AB/nicbVDLSgMxFL3js9ZXVy5CRbBhZQZKagLoejGZQX7gM5QMpm0Dc0kQ5IRylDwV9y4UMSt3+HOvzFtZ6GtBwKHc+7lnpw4Uwb1/12lpZXVt
750
+ fWCxvFza3tnd3S3n5Ty1QR2iCS9UOsacCdowzHDaThTFchpKxzeTvzWI1WaSfFgRgkNYtwXrMcINlbqlg79nlSYc8Sy67Hn80gafWb1sltxp0CLxMtJGXLUu6UvP5IkjakwhGOtO56bmCDyjDC6bjop5omAxn3YsFTimOsim8cfoxCoRsjnsEwZN1d8bGY61HsWhnYyxGeh5byL+53VS07sMiaS1FBZod6KUdGokXKGKEsNHlmCimM2KyArTIxtrGhL8Oa/vEia5xWvWrm6r5ZrN3kdBTiCYzgFDy6gBndQhwYQyOAZXuHNeXJenHfnYz
751
+ a65OQ7B/AHzucP2q2VcA=</latexit>
752
+ ˆpi=
753
+ XE
754
+ �=1 ˆp�,
755
+ <latexit sha1_base64="GuE1wgouyMWpJt3pSfGs7Vt7lQo=">ACSHicbVBNaxsxENW6zZeTpm57zEXEFHoZrcY2h4CoaXQYwJxEvA6y6ws2yL6WKTZgBH6eb302Ft/Qy89tJTeonX2kK8Bocd786SZV1ZSOEzTn0n
756
+ ydO19Y3Nre72zrPd570XL0+dqS3jI2akseclOC6F5iMUKPl5ZTmoUvKz8vJzo59dceuE0Se4rPhEwVyLmWCAkSp6Rb4A9Hlp5NQtVbx8FUIhqD8INHe1KnzuxFzBQRZybaRQAt2FzxXgoH0X0J49IHWFd4WvX46SFdFH4KsBX3S1lHR+5FPDasV18gkODfO0gonHiwKJno5rXjFbBLmPNxhBoUdxO/CiLQ15GZ0pmx8WikK/a2w4NyzZSxs9nA3dca8jFtXOPsw8QLXdXINbv5aFZLioY2qdKpsJyhXEYAzIo4K2ULsMAwZt+NIWT3V34ITt8NsuHg4/
757
+ Gwf/ipjWOT7JF98oZk5D05JF/JERkRr6RX+QP+Zt8T34n/5L/N62dpPW8Ineq07kGLPy2Kg=</latexit>
758
+ ˆpi, ˜pi, 8i = 1 . . . , n
759
+ <latexit sha1_base64="ChPanXivTdMBX2Dsthv7AEM2Hyg=">ACLnicbVDLSgMxFM3UV62vqks3wSK4KGVGCupCKIrgsoJ9QGcomUymDc1MhuSOUIZ+kRt/ReCirj1M0wfC217IeRwzr3JucdP
760
+ BNdg2+9WbmV1bX0jv1nY2t7Z3SvuHzS1TBVlDSqFVG2faCZ4zBrAQbB2ohiJfMFa/uBmrLcemdJcxg8wTJgXkV7MQ04JGKpbvHX7BDLXlyLQw8hcWTIadXkZu8BFwJYqoVRECMyvHFcEnTZvFOyK/ak8CJwZqCEZlXvFl/dQNI0YjFQbTuOHYCXkYUcCrYqOCmiWEDkiPdQyMScS0l03WHeETwTYuDAnBjxh/05kJNJjy6YzItDX89qYXKZ1UgvIzHSQosptOPwlRgkHicHQ64YhTE0ABCFTdeMe0TRSi
761
+ YhAsmBGd+5UXQPKs41crlfbVUu57FkUdH6BidIgedoxq6Q3XUQBQ9oRf0gT6tZ+vN+rK+p605azZziP6V9fMLNxmqbQ=</latexit>
762
+ r˜p⌫L(·)
763
+ <latexit sha1_base64="NCdmlfZtlskHCRoWxGd7tDpT0mQ=">ACIHicbVBNS8NAEN34WetX1aOXYBH0UhIpqDfRiwcPFWwVmlIm61dutkNuxOhPwUL/4VLx4U0Z
764
+ v+GjdtDn4NLPt4b4Z58JEcIOe9+HMzM7NLyxWlqrLK6tr67WNzY5RqasTZVQ+iYEwSXrI0cBbtJNIM4FOw6HJ0V+vUd04YreYXjhPViuJV8wCmgpfq1w0BCKCfBchFxLIgVCIy49h+WZLnlpdpngcx4JCyC7yvYBGCvf7tbrX8Cbl/gV+CeqkrFa/9h5EiqYxk0gFGNP1vQR7GWjkVLC8GqSGJUBHcMu6FkqImelkwNzd9cykTtQ2j6J7oT9
765
+ PpFBbArTtrNwan5rBfmf1k1xcNTLuExSZJOFw1S4aJyi7TciGtGUYwtAKq59erSIWigaDOt2hD83yf/BZ2Dht9sHF826yenZRwVsk12yB7xySE5IekRdqEknvySJ7Ji/PgPDmvztu0dcYpZ7bIj3I+vwD93qVW</latexit>
766
+ rˆp�L(·)
767
+ <latexit sha1_base64="WTcG5eEwOkl6SUea4MtPTARM=">ACIXicbVBNS8NAEN34bf2qevQSLEK9lEQE9SZ68eBwX5AU8pks2XbnbD7kQoIX/Fi3/FiwdFvIl/
768
+ xk3bg7YOLPt4b4Z58JEcIOe9+UsLC4tr6yurZc2Nre2d8q7ew2jUk1ZnSqhdCsEwSXrI4cBWslmkEcCtYMh9eF3nxk2nAlH3CUsE4Mfcl7nAJaqls+DySEArpZMADMglCJyIxi+2VJnlvW8H4MeR7EgAMKIrvNqwGNFB53yxWv5o3LnQf+FTItO65c8gUjSNmUQqwJi27yXYyUAjp4LlpSA1LAE6hD5rWyghZqaTjS/M3SPLRG5PafskumP290QGsS
769
+ l8287CqZnVCvI/rZ1i7yTcZmkyCSdLOqlwkXlFnG5EdeMohZAFRz69WlA9BA0YZasiH4syfPg8ZJzT+tXdyfVi6vpnGskQNySKrEJ2fktyQO1InlDyRF/JG3p1n59X5cD4nrQvOdGaf/Cn+wfAO6W5</latexit>
770
+ Individual DER acts an agent to
771
+ calculate subgradients
772
+ and
773
+ r˜p⌫L(·)
774
+ <latexit sha1_base64="NCdmlfZtlskHCRoWxGd7tDpT0mQ=">ACIHicbVBNS8NAEN34WetX1aOXYBH0UhIpqDfRiwcPFWwVmlIm61dutkNuxOhPwUL/4VLx4U0Z
775
+ v+GjdtDn4NLPt4b4Z58JEcIOe9+HMzM7NLyxWlqrLK6tr67WNzY5RqasTZVQ+iYEwSXrI0cBbtJNIM4FOw6HJ0V+vUd04YreYXjhPViuJV8wCmgpfq1w0BCKCfBchFxLIgVCIy49h+WZLnlpdpngcx4JCyC7yvYBGCvf7tbrX8Cbl/gV+CeqkrFa/9h5EiqYxk0gFGNP1vQR7GWjkVLC8GqSGJUBHcMu6FkqImelkwNzd9cykTtQ2j6J7oT9
776
+ PpFBbArTtrNwan5rBfmf1k1xcNTLuExSZJOFw1S4aJyi7TciGtGUYwtAKq59erSIWigaDOt2hD83yf/BZ2Dht9sHF826yenZRwVsk12yB7xySE5IekRdqEknvySJ7Ji/PgPDmvztu0dcYpZ7bIj3I+vwD93qVW</latexit>
777
+ rˆp�L(·)
778
+ <latexit sha1_base64="WTcG5eEwOkl6SUea4MtPTARM=">ACIXicbVBNS8NAEN34bf2qevQSLEK9lEQE9SZ68eBwX5AU8pks2XbnbD7kQoIX/Fi3/FiwdFvIl/
779
+ xk3bg7YOLPt4b4Z58JEcIOe9+UsLC4tr6yurZc2Nre2d8q7ew2jUk1ZnSqhdCsEwSXrI4cBWslmkEcCtYMh9eF3nxk2nAlH3CUsE4Mfcl7nAJaqls+DySEArpZMADMglCJyIxi+2VJnlvW8H4MeR7EgAMKIrvNqwGNFB53yxWv5o3LnQf+FTItO65c8gUjSNmUQqwJi27yXYyUAjp4LlpSA1LAE6hD5rWyghZqaTjS/M3SPLRG5PafskumP290QGsS
780
+ l8287CqZnVCvI/rZ1i7yTcZmkyCSdLOqlwkXlFnG5EdeMohZAFRz69WlA9BA0YZasiH4syfPg8ZJzT+tXdyfVi6vpnGskQNySKrEJ2fktyQO1InlDyRF/JG3p1n59X5cD4nrQvOdGaf/Cn+wfAO6W5</latexit>
781
+ Individual bus acts an agent to
782
+ calculate subgradients
783
+ and
784
+ DER-level aggregation and control
785
+ Bus-level aggregation and control
786
+ Fig. 2.
787
+ Aggregation and control of DERs via bus-level and DER-level
788
+ architectures.
789
+ bus (agent) aggregates the decision variables ˜pi = �Vi
790
+ ν=1 ˜pν
791
+
792
+ 5
793
+ and ˆpi = �Ei
794
+ σ=1 ˆpσ where only aggregated decision variables
795
+ are transmitted and used for the primal updates. In contrast,
796
+ DER-level control strategies require each DER to act as an
797
+ agent and receive all data of others that is demanded for
798
+ updates in (20). However, due to the large number of DERs
799
+ connected to the distribution network, DER-level control can
800
+ suffer from massive data exchange and heavy local computa-
801
+ tion. Therefore, we adopt the bus-level aggregation and control
802
+ scheme which is more computing and communicating effi-
803
+ cient. We will later show that the proposed privacy-preserving
804
+ algorithm can be readily extended to the DER-level control
805
+ (See Remark 1 for details).
806
+ Apart from scalability and efficiency, the inevitable private
807
+ information exposure in both bus-level and DER-level methods
808
+ raises fundamental privacy concerns, e.g., the electrical load
809
+ can reveal sensitive business activities and/or customer’s daily
810
+ routines. To address the privacy concerns, we will develop
811
+ a novel SS-based algorithm to achieve secure information
812
+ exchange in executing (20).
813
+ IV. SS-BASED PRIVACY-PRESERVING DER CONTROL
814
+ A. Real Number to Integer Quantization
815
+ Note that the SS scheme requires modular arithmetic instead
816
+ of real arithmetic. However, decentralized optimization ge-
817
+ netically requires real number calculations, e.g., real decision
818
+ variables and parameters. Therefore, a real number to integer
819
+ transformation is needed to integrate SS into decentralized
820
+ optimization. We adopt the fixed-point number quantization
821
+ [27] to map the real numbers onto the integer space and the
822
+ fixed-point real-number set is defined by
823
+ Qθ,γ,ζ≜
824
+
825
+ −θγ, −θγ + θ−ζ, . . . , θγ − 2θ−ζ, θγ − θ−ζ�
826
+ (23)
827
+ where θ ∈ N1+ denotes the basis, γ ∈ N denotes the
828
+ magnitude, and ζ ∈ N denotes the resolution. Therefore, by
829
+ defining a surjective mapping m(·) : R �→ Qθ,γ,ζ, a real
830
+ number can be mapped to the closest point in Qθ,γ,ζ. To limit
831
+ the quantization error, the mapping m(·) needs to satisfy
832
+ |m(ϕ) − ϕ| ≤ θ−ζ, ∀ϕ ∈ [−θγ, θγ]
833
+ (24)
834
+ where the quantization error is restricted by the resolution
835
+ within the range of Qθ,γ,ζ. To map the real-number set onto
836
+ the integer set Z, we simply scale Qθ,γ,ζ by θζ as
837
+ Zθ,γ,ζ = θζQθ,γ,ζ=
838
+
839
+ −θγ+ζ, −θγ+ζ+1, . . . , θγ+ζ−1
840
+
841
+ (25)
842
+ where Zθ,γ,ζ ⊆ Z denotes the fixed-point set in the integer
843
+ field. Moreover, the SS requires the inputs to be within the
844
+ field E. Therefore, we further map each element in z ∈ Zθ,γ,ζ
845
+ onto E with the modular operation as
846
+ g(z) = z mod e.
847
+ (26)
848
+ Note that z ∈ Zθ,γ,ζ can be any negative integer, and the
849
+ modular operation in (26) will change the sign of a negative
850
+ input, i.e., g(ˆz) = ˆz + e for ˆz < 0. To address the negative
851
+ integer operation, we introduce the partial inverse of g(·) as
852
+ ψ(z) =
853
+ � z − e
854
+ if z ≥ e
855
+ 2,
856
+ z
857
+ otherwise.
858
+ (27)
859
+ Therefore, we can readily obtain z = ψ(g(z)), ∀z ∈ E.
860
+ B. SS-based Privacy-Preserving Algorithm
861
+ 1) Shamir’s secret sharing scheme: Before introducing the
862
+ privacy-preserving algorithm design, we first briefly intro-
863
+ duce Shamir’s SS scheme [20] which merits an efficient and
864
+ lightweight private information distribution structure. Suppose
865
+ a manager (secret holder) seeks to distribute a secret ω to
866
+ specific agents and mandates the cooperation of at least d
867
+ agents to retrieve the secret. In such needs, Shamir’s SS is
868
+ grounded on the following idea of Lagrange interpolation for
869
+ secret distribution and recovery.
870
+ Theorem 1 (Polynomial interpolation [28]). Let {(ς1, y1), . . . ,
871
+ (ςd, yd)} ⊆ R2 be a set of points whose values of ςı are all
872
+ distinct. Then there exists a unique polynomial Y of degree
873
+ d − 1 that satisfies yı = Y(ςı), ∀ı = 1, . . . , d.
874
+
875
+ In SS-based schemes, the manager first constructs a random
876
+ polynomial of degree d − 1 as
877
+ y(z) = ω + a1z + · · · + ad−1zd−1
878
+ (28)
879
+ where ω denotes an integer secret, a1, . . . , ad−1 are random
880
+ coefficients that are uniformly distributed in the field E ≜
881
+ [0, e), and e denotes a prime number that is larger than ω.
882
+ Secondly, the manager calculates the outputs of (28) with
883
+ non-zero integer inputs, e.g., setting τ = 1, . . . , n to retrieve
884
+ (τ, y(τ)) where yΠ
885
+ τ
886
+ = y(τ) mod e. Then, the share yΠ
887
+ τ
888
+ is
889
+ distributed to agent τ. Lastly, at least d agents with shares
890
+ are required to reconstruct the polynomial based on Theorem
891
+ 1 and hence recover the secret ω by
892
+ ω =
893
+ d
894
+
895
+ τ=1
896
+
897
+ τ
898
+ d
899
+
900
+ υ=0
901
+ υ̸=τ
902
+ υ
903
+ υ − τ .
904
+ (29)
905
+ 2) Proposed privacy-preserving algorithm: We next present
906
+ the proposed two-layer decentralized privacy-preserving al-
907
+ gorithm based on SS in a bus-level aggregation and control
908
+ architecture, to achieve privacy preservation and scalability
909
+ concurrently. In the distribution network layer, all DERs’ deci-
910
+ sion variables are updated in parallel, and only masked data are
911
+ sent from each bus to the servers. In the cloud computing layer,
912
+ the servers calculate the aggregated messages and distribute
913
+ them to the related buses. The computing structure of the
914
+ proposed privacy-preserving algorithm is shown in Fig. 3.
915
+ Cloud Computing
916
+ Distribution
917
+ Network
918
+ ESS
919
+ Solar
920
+ PV
921
+ Server
922
+ Secure
923
+ Data Flow
924
+ Secure
925
+ Data Flow
926
+ Bus
927
+ Fig. 3. Two-layer privacy-preserving computing structure for DER control in
928
+ distribution networks.
929
+
930
+ 田田Compute20066
931
+ Let C denote the set of clouds and c ≥ 2 denotes the total
932
+ number of clouds. The ith bus generates a random polynomial
933
+ of order d − 1 using (28) to obtain
934
+ y(ℓ)
935
+ i (z) = ω(ℓ)
936
+ i
937
+ + a(ℓ)
938
+ i,1z + · · · + a(ℓ)
939
+ i,d−1zd−1
940
+ (30)
941
+ where 2 ≤ d ≤ c, ω(ℓ)
942
+ i
943
+ denotes the secret of bus i at the ℓth
944
+ iteration, ℓ denotes the iteration number, and a(ℓ)
945
+ i,1, . . . , a(ℓ)
946
+ i,d−1
947
+ denote random coefficients that are uniformly distributed in the
948
+ field E. Note that for a vector secret such as pi, we refer to an
949
+ elementwise calculation of the vector using (30) by default.
950
+ At the ℓth iteration, the uth cloud firstly generates a random
951
+ integer α(ℓ)
952
+ u , then it broadcasts α(ℓ)
953
+ u
954
+ to all the buses. Subse-
955
+ quently, the ith bus can calculate y(ℓ)
956
+ i (α(ℓ)
957
+ u ), ∀u = 1, . . . , c
958
+ using the received inputs based on (30). Finally, the ith bus
959
+ sends y(ℓ)
960
+ i (α(ℓ)
961
+ u ) back to the uth cloud. Note that the coupling
962
+ term πi ˜P in (21) is a linear combination of all pi’s that
963
+ requires the private generation/consumption details from the
964
+ buses. Therefore, a secure computation framework of πi ˜P is
965
+ required to preserve the privacy of buses and DER owners.
966
+ Suppose the clouds are aware of the network topology
967
+ matrix Z which contains no private information of the buses
968
+ or DERs. In order to calculate the aggregated information πi ˜P
969
+ for bus i, the uth cloud firstly multiplies the received outputs
970
+ y1(α(ℓ)
971
+ u ), . . . , yn(α(ℓ)
972
+ u ) utilizing the coefficients of πi to obtain
973
+ {α(ℓ)
974
+ u , πi(1)y(ℓ)
975
+ 1 (α(ℓ)
976
+ u ), . . . , πi(n)y(ℓ)
977
+ n (α(ℓ)
978
+ u )}
979
+ (31)
980
+ Then, the uth cloud sums the outputs in (31) to obtain a new
981
+ pair of input and output as
982
+ ¯
983
+ Au,i = {α(ℓ)
984
+ u ,
985
+ n
986
+
987
+ ˆı=1
988
+ πi(ˆı) y(ℓ)
989
+ ˆı (α(ℓ)
990
+ u )}.
991
+ (32)
992
+ Finally, the uth cloud calculates
993
+ ¯
994
+ Au,i, ∀i = 1, . . . , n and
995
+ broadcasts the new input-output share ¯
996
+ Au,i to the ith bus.
997
+ Therefore, after receiving new shares from in total c clouds
998
+ servers, the ith bus now has access to
999
+ ˜
1000
+ Ai =
1001
+
1002
+ α(ℓ)
1003
+ ˆȷ ,
1004
+ n
1005
+
1006
+ ˆı=1
1007
+ πi(ˆı) y(ℓ)
1008
+ ��ı (α(ℓ)
1009
+ ˆȷ ), ∀ˆȷ = 1, . . . , c
1010
+
1011
+ .
1012
+ (33)
1013
+ Note that ˜
1014
+ Ai contains in total c shares that can construct a
1015
+ new polynomial of the form
1016
+ ˜y(ℓ)
1017
+ i (z) = πi ˜P + ˜a(ℓ)
1018
+ i,1z + · · · + ˜a(ℓ)
1019
+ i,d−1zd−1
1020
+ (34)
1021
+ whose constant term is exactly πi ˜P .
1022
+ During this information exchange process, each bus only
1023
+ sends a single share to each server so that a single cloud server
1024
+ is incapable of reconstructing the secret based on the received
1025
+ shares, and herein cannot infer agents’ true decision variables.
1026
+ The cloud servers only need to calculate aggregated messages
1027
+ using outputs of randomized polynomials. The details of the
1028
+ proposed method are presented via Algorithm 1.
1029
+ Algorithm 1 can achieve privacy preservation while main-
1030
+ taining exact solutions as non-privacy PGM-based methods.
1031
+ The decision variables will be continuously updated till the
1032
+ convergence errors ϵ(ℓ)
1033
+ ν
1034
+ ≜ ∥˜p(ℓ)
1035
+ ν − ˜p(ℓ−1)
1036
+ ν
1037
+ ∥2
1038
+ 2 and ϵ(ℓ)
1039
+ σ
1040
+ ≜ ∥ˆp(ℓ)
1041
+ σ −
1042
+ ˆp(ℓ−1)
1043
+ σ
1044
+ ∥2
1045
+ 2 are smaller than the threshold ϵ0. The correctness of
1046
+ Algorithm 1 is presented via Theorem 2.
1047
+ IEEE 13 bus network
1048
+ Decentralized
1049
+ updates
1050
+ Cloud
1051
+ Aggregation
1052
+ y(`)
1053
+ 3 (↵(`)
1054
+ 1 )
1055
+ <latexit sha1_base64="jnzoug/Af0ACPER59rq8DBO7js8=">ACnicbVDLSsNAFJ3UV42vqEs3o0VoNyXRgrorunFZwT6giWEynbRDJ5MwMxFK6NqNv+LGhSJu/QJ3/o3
1056
+ TNqC2HrhwOde7r0nSBiVyra/jMLS8srqWnHd3Nj
1057
+ c2t6xdvdaMk4FJk0cs1h0AiQJo5w0FVWMdBJBUBQw0g6GVxO/fU+EpDG/VaOEeBHqcxpSjJSWfOtw5J/emVnZJYxVxmUXsWSAfOdHqvhWya7aU8BF4uSkBHI0fOvT7cU4jQhXmCEpu46dKC9DQlHMyNh0U0kShIeoT7qachQR6WXTV8bwWCs9GMZCF1dwqv6eyFAk5SgKdGeE1EDOexPxP6+bqvDcyhPUkU
1058
+ 4ni0KUwZVDCe5wB4VBCs20gRhQfWtEA+QFjp9EwdgjP/8iJpnVSdWvXiplaqX+ZxFMEBOAJl4IAzUAfXoAGaAIMH8ARewKvxaDwb8b7rLVg5DP74A+Mj29
1059
+ 9eJj</latexit>
1060
+ y(`)
1061
+ 3 (↵(`)
1062
+ 2 )
1063
+ <latexit sha1_base64="dxv9zg96EAsVj0zT8hPgd5wTIl4="
1064
+ >ACnicbVDLSsNAFJ34rPEVdelmtAjtpiS1oO6KblxWsA9oYphMJ+3QySTMTIQSunbjr7hxoYhbv8Cdf+O0DaitBy4czrmXe+8
1065
+ JEkalsu0vY2l5ZXVtvbBhbm5t7+xae/stGacCkyaOWSw6AZKEU6aipGOokgKAoYaQfDq4nfvidC0pjfqlFCvAj1OQ0pRkpLv
1066
+ nU0
1067
+ 8k/vzKzkEsbK45KLWDJAfvVHKvtW0a7YU8BF4uSkCHI0fOvT7cU4jQhXmCEpu46dKC9DQlHMyNh0U0kShIeoT7qachQR6WXTV8bwR
1068
+ Cs9GMZCF1dwqv6eyFAk5SgKdGeE1EDOexPxP6+bqvDcyhPUkU4ni0KUwZVDCe5wB4VBCs20gRhQfWtEA+QFjp9EwdgjP/8iJpV
1069
+ StOrXJxUyvWL/M4CuAQHIMScMAZqINr0ABNgMEDeAIv4NV4NJ6N+N91rpk5DMH4A+Mj29/B5jk</latexit>
1070
+ y(`)
1071
+ 3 (↵(`)
1072
+ c )
1073
+ <latexit sha1_base64="AT8U+757zt958DYmNWETF
1074
+ qRUVhM=">ACnicbVDLSsNAFJ3UV42vqEs3o0VoNyXRgrorunFZwT6giWEynbRDJ5MwMxFK6NqNv+LGhSJ
1075
+ u/QJ3/o3TNqC2HrhwOde7r0nSBiVyra/jMLS8srqWnHd3Njc2t6xdvdaMk4FJk0cs1h0AiQJo5w0FVWMdBJ
1076
+ BUBQw0g6GVxO/fU+EpDG/VaOEeBHqcxpSjJSWfOtw5J/emVnZJYxVxmUXsWSAfPwjVXyrZFftKeAicXJSAjka
1077
+ vXp9mKcRoQrzJCUXcdOlJchoShmZGy6qSQJwkPUJ1NOYqI9LpK2N4rJUeDGOhiys4VX9PZCiSchQFujNC
1078
+ aiDnvYn4n9dNVXjuZQnqSIczxaFKYMqhpNcYI
1079
+ 8KghUbaYKwoPpWiAdIKx0eqYOwZl/eZG0TqpOrXpxUyvV
1080
+ L/M4iuAHIEycMAZqINr0ABNgMEDeAIv4NV4NJ6N+N91low8pl98AfGxzfLZpkV</latexit>
1081
+ Cloud1
1082
+ Cloud 2
1083
+ Cloud c
1084
+ DERs
1085
+ DERs
1086
+ DERs
1087
+ Bus 3
1088
+ Bus 2
1089
+ Bus 6
1090
+ Bus 1
1091
+ Bus 4
1092
+ Bus 5
1093
+ Bus 7
1094
+ Bus 9
1095
+ Bus 10
1096
+ Bus 12
1097
+ Bus 11
1098
+ Bus 8
1099
+ Bus 2
1100
+ Bus 1
1101
+ Bus 12
1102
+ ¯
1103
+ A(`)
1104
+ 1,1
1105
+ <latexit sha1_base64="OAQemdne2zbAeumZuA9yhroVF64=">ACnicbVDLSsNAFJ3UV42vqEs30SJUkJIQd3VunFZwT6giWEynbRDJw9mJkIZsnbjr7hxoYhbv8Cdf+OkzUJbD1w4nHMv97jJ5RwYVnfWmlpeWV1rbyub2xube8Yu3
1106
+ sdHqcM4TaKacx6PuSYkgi3BREU9xKGYehT3PXH17nfcCMkzi6E5MEuyEcRiQgCAolecah04RMOiEUIwSpvMoyT9qndnavy6qDKT3JPKNi1awpzEViF6QCrQ848sZxCgNcSQhZz3bSsRroRMERxpjspxwlEYzjEfUjGLuyukrmXmslIEZxExVJMyp+ntCwpDzSeirzvxmPu/l4n9ePxXBhStJlKQCR2i2KEipKWIz8UcEIaRoBNFIGJE3WqiEWQCZWerkKw519eJ2zml2vXd7WK41mEUcZHIAjUAU2OAcNcANaoA0QeATP4BW8aU/ai/aufcxaS1oxsw/+QPv8AVJNmg=</latexit
1107
+ >
1108
+ ¯
1109
+ A(`)
1110
+ 1,2
1111
+ <latexit sha1_base64="BX3MrG24Z53ODrxs+/8snufJ/7o=">AC
1112
+ CXicbVDLSsNAFJ3UV62vqEs3g0WoICUpBXVX68ZlBfuAJobJdNoOnUzCzEQoIVs3/obF4q49Q/c+TdO2iy09cCFwzn3cu89fsSoVJb1bRWVtf
1113
+ WN4qbpa3tnd09c/+gI8NYNLGIQtFz0eSMpJW1HFSC8SBAU+I1/cp353QciJA35nZpGxA3QiNMhxUhpyTOh0QicQKkxhix5CpNvcQ+q6X3Sc
1114
+ UhjJ2mnlm2qtYMcJnYOSmDHC3P/HIGIY4DwhVmS
1115
+ Mq+bUXKTZBQFDOSlpxYkgjhCRqRvqYcBUS6yeyTFJ5oZQCHodDFZypvycSFEg5DXzdmZ0s
1116
+ F71M/M/rx2p4SaUR7EiHM8XDWMGVQizWOCACoIVm2qCsKD6VojHSCsdHglHYK9+PIy6dSqdr16eVsvN5p5HEVwBI5BdjgHDTADWiBNsDgETy
1117
+ DV/BmPBkvxrvxMW8tGPnMIfgD4/MHFgeZ9Q=</latexit>
1118
+ ¯
1119
+ A(`)
1120
+ 1,12
1121
+ <latexit sha1_base64="wbvr5FCeuxjCnzM3Uiy/T
1122
+ 6iGK5w=">ACnicbVDLSsNAFJ3UV62vqEs3o0WoICUpBXVX68ZlBfuAJobJdNoOnTyYmQhlyNqNv+LGhSJu
1123
+ /QJ3/o2TNgutHrhwOde7r3HjxkV0rK+jMLS8srqWnG9tLG5tb1j7u51RJRwTNo4YhHv+UgQRkPSlQy0os5Q
1124
+ YHPSNefXGV+95wQaPwVk5j4gZoFNIhxUhqyTMPnSbiygmQHGPE1GWaeso+tWvpnao4hLGT1DPLVtWaAf4ldk7
1125
+ KIEfLMz+dQYSTgIQSMyRE37Zi6SrEJcWMpCUnESRGeIJGpK9piAIiXDV7JYXHWhnAYcR1hRLO1J8TCgVCTANf
1126
+ d2Y3i0UvE/z+okcnruKhnEiSYjni4YJgzKCWS5wQDnBk01QZhTfSvEY8QRljq9kg7BXnz5L+nUqna9enFTL
1127
+ zeaeRxFcACOQAXY4Aw0wDVogTbA4AE8gRfwajwaz8ab8T5vLRj5zD74BePjG4+2mjA=</latexit>
1128
+ Bus 0
1129
+ Fig. 4. Information exchange structure between the distribution network and
1130
+ cloud servers (only the messages sent from bus 3 and cloud 1 are labeled).
1131
+ Algorithm 1 Decentralized SS-based privacy-preserving DER
1132
+ control strategy
1133
+ 1: Agents initialize decision variables, tolerance ϵ0, basis θ,
1134
+ magnitude γ, resolution ζ, iteration counter ℓ = 0, and
1135
+ maximum iteration ℓmax.
1136
+ 2: while ϵ(ℓ)
1137
+ ν(σ) > ϵ0 and ℓ < ℓmax do
1138
+ 3:
1139
+ Each bus performs real number to integer transforma-
1140
+ tion using (23)-(26), then obtains the integer secret ω(ℓ)
1141
+ i .
1142
+ 4:
1143
+ The uth cloud generates a random integer α(ℓ)
1144
+ u , then
1145
+ broadcasts α(ℓ)
1146
+ u
1147
+ to all the buses.
1148
+ 5:
1149
+ The ith bus generates a random polynomial y(ℓ)
1150
+ i (z)
1151
+ using (30) with ω(ℓ)
1152
+ i
1153
+ as the constant term, calculates the
1154
+ outputs using α(ℓ)
1155
+ 1 , . . . , α(ℓ)
1156
+ c
1157
+ to obtain y(ℓ)
1158
+ i (α(ℓ)
1159
+ 1 ), . . . ,
1160
+ y(ℓ)
1161
+ i (α(ℓ)
1162
+ c ), then sends y(ℓ)
1163
+ i (α(ℓ)
1164
+ u ) to the uth cloud.
1165
+ 6:
1166
+ The uth cloud formulates ¯
1167
+ Au,i in (32), then broadcasts
1168
+ ¯
1169
+ Au,i to the ith bus.
1170
+ 7:
1171
+ The ith bus formulates
1172
+ ˜
1173
+ Ai in (33), reconstructs the
1174
+ aggregated secrets using c shares to obtain πi ˜P , then
1175
+ calculates Cp in (21).
1176
+ 8:
1177
+ The ith bus transforms Cp back to real numbers
1178
+ using (27), then decision variables ˜p(ℓ)
1179
+ ν
1180
+ or ˆp(ℓ)
1181
+ σ
1182
+ of DERs
1183
+ connected at bus i are updated by PGM using (9). The ith
1184
+ bus calculates the error ϵ(ℓ)
1185
+ ν
1186
+ or ϵ(ℓ)
1187
+ σ .
1188
+ 9:
1189
+ ℓ = ℓ + 1.
1190
+ 10: end while
1191
+ Theorem 2 (Correctness). Let E denote the domain of the
1192
+ input secrets ω1, . . . , ωn, and Cp denote the desired outputs.
1193
+ Then, Algorithm 1 satisfies:
1194
+ Pr
1195
+
1196
+ ∀c ≥ d, Rec
1197
+
1198
+ A, E, Z, ¯δ1, θ, γ, ζ
1199
+
1200
+ = Cp
1201
+
1202
+ = 1
1203
+ (35)
1204
+ where A = { ˜
1205
+ A1, . . . , ˜
1206
+ Ac} denotes the set of shares from
1207
+ agents, Pr[·] denotes probability, and Rec(·) denotes the secret
1208
+ reconstruction operation.
1209
+
1210
+ Theorem 2 states that Algorithm 1 can correctly retrieve
1211
+ the aggregated information Cp which would be further used to
1212
+ achieve exact primal and dual updates.
1213
+
1214
+ Compute田田7
1215
+ The detailed proof of Theorem 2 can be found in AP-
1216
+ PENDIX B.
1217
+ Remark 1 : Though Algorithm 1 is developed based on bus-
1218
+ level aggregation and control, it can also be extended to the
1219
+ DER-level aggregation and control. In DER-level aggregation
1220
+ and control, each DER is required to generate a polynomial in
1221
+ (30) and act as an independent agent in secret reconstruction
1222
+ using (33). Besides, depending on the practical applications,
1223
+ DERs can also be clustered and controlled by the household
1224
+ or district where the new clusters act as agents, following the
1225
+ similar design of Algorithm 1.
1226
+
1227
+ Remark 2: The multi-server architecture seamlessly integrates
1228
+ the SS scheme into DER aggregation and control. Shares
1229
+ generated from buses were aggregated and broadcasted to the
1230
+ buses by a group of servers for the purpose of secret retrieval.
1231
+ The aggregation task is distributed to multiple servers to ensure
1232
+ that a single server cannot retrieve any secrets.
1233
+
1234
+ C. Privacy Analysis
1235
+ The proposed approach aims at protecting the decision
1236
+ variables of the DERs whose disclosure can lead to the leakage
1237
+ of customers’ sensitive information. To resolve this issue, Al-
1238
+ gorithm 1 achieves privacy preservation against two types of
1239
+ adversaries, including honest-but-curious-agent who follows
1240
+ the algorithm but may utilize the possessed and received data
1241
+ to infer the private information of other agents, and external
1242
+ eavesdroppers who wiretap and intercept exchanged messages
1243
+ from communication channels.
1244
+ Proposition 1: (Secure cloud computing). In Algorithm 1, any
1245
+ cloud number less than d − 1 cannot infer any information of
1246
+ the aggregated decision variables Cp.
1247
+
1248
+ Proposition 1 presents the security of the proposed al-
1249
+ gorithm against corrupted clouds. Based on the polynomial
1250
+ interpolation in Theorem 1, at least d clouds are required to
1251
+ retrieve any secret through collusion.
1252
+ Proposition 1 is proved based on the correctness analysis.
1253
+ Please refer to APPENDIX C for the detailed proof.
1254
+ Assumption 1. At least one communication link of an indi-
1255
+ vidual agent is secure against external eavesdroppers.
1256
+
1257
+ Assumption 1 is essential and generically used in SS-
1258
+ based schemes. Given d pairs of shares sent via different
1259
+ communication links, i.e., {(ς1, y1), . . . , (ςd, yd)} ⊆ R2, if
1260
+ an external eavesdropper wiretap all communication links to
1261
+ gain access to the shares, then it can simply deduce the secret
1262
+ by Lagrangian interpolation using Theorem 1.
1263
+ Theorem 3 (Privacy preservation against adversaries). By
1264
+ using Algorithm 1, the following two statements stand:
1265
+ 1) Algorithm 1 securely computes and updates the deci-
1266
+ sion variables between agents in the presence of honest-
1267
+ but-curious agents.
1268
+ 2) External eavesdroppers learn no private information of
1269
+ the agents.
1270
+
1271
+ Theorem 3 gives privacy preservation guarantees in the
1272
+ presence of honest-but-curious agents and external eavesdrop-
1273
+ pers. The privacy preservation of Algorithm 1 can be proved
1274
+ from secure multi-party computation (SMC) perspective. Be-
1275
+ fore giving detailed privacy analyses and proofs, we first
1276
+ introduce some concepts of SMC.
1277
+ Definition 1 (Computational indistinguishability [29]). Let
1278
+ {Dκ}κ∈N and {Eκ}κ∈N be two distribution ensembles with
1279
+ security parameter κ; If for any non-uniform probabilistic
1280
+ polynomial-time algorithm G, δ(κ) is negligible, where
1281
+ δ(κ) =
1282
+ ����
1283
+ Pr
1284
+ x1←Dκ[G(x1) = 1] −
1285
+ Pr
1286
+ x2←Eκ[G(x2) = 1]
1287
+ ����
1288
+ (36)
1289
+ we say that {Dκ}κ∈N and {Eκ}κ∈N are computationally
1290
+ indistinguishable, denoted as Dκ
1291
+ c≡ Eκ.
1292
+
1293
+ Therefore, Definition 1 states that any polynomial-time
1294
+ algorithm cannot distinguish two computationally indistin-
1295
+ guishable ensembles because the outputs of those algorithms
1296
+ do not significantly differ. In what follows, Definition 2
1297
+ presents the standard privacy notion in SMC.
1298
+ Definition 2 ([30], [31]). Let Π be an m-party protocol
1299
+ for computing the outputs of function F(¯x) where ¯x =
1300
+ {x1, . . . , xm} and Fρ(¯x) denotes the ρth output of F(¯x). Let
1301
+ M = {M1, . . . , Mm} denote the set of parties. The view
1302
+ of the ρth party during the execution of Π is denoted by
1303
+ VIEWΠ
1304
+ ρ (¯x). We say that Π privately computes F(¯x) if there
1305
+ exists a polynomial-time algorithm S, such that for every party
1306
+ Mρ in M, we have
1307
+ S(ρ, xρ, Fρ(¯x))
1308
+ c≡ VIEWΠ
1309
+ ρ (¯x).
1310
+ (37)
1311
+
1312
+ Definition 2 states that the security of an m-party protocol
1313
+ can be evaluated based on computational indistinguishability,
1314
+ i.e., the view of the parties can be efficiently simulated based
1315
+ solely on their inputs and outputs. In other words, SMC allows
1316
+ a group of participants to learn the correct outputs of some
1317
+ agreed-upon function applied to their private inputs without
1318
+ revealing anything else. The theoretical underpinnings of Def-
1319
+ inition 1 and Definition 2 can help prove that Algorithm 1
1320
+ securely computes π1 ˜P , . . . , πn ˜P between the agents.
1321
+ The detailed proofs of Theorem 3 can be found in AP-
1322
+ PENDIX D.
1323
+ V. SIMULATION RESULTS
1324
+ A simplified single-phase IEEE 13-bus test feeder [32] is
1325
+ used to verify the proposed decentralized privacy-preserving
1326
+ DER control strategy. In specific, each bus, except the feeder
1327
+ head, is assumed to be connected with 2 houses and each house
1328
+ is equipped with an ESS and 5 solar panels that can generate
1329
+ maximum 2.5 kW solar output. The maximum capacity of all
1330
+ residential ESSs are 10 kWh, the initial SoCs of all ESSs are
1331
+ uniformly set to be 4 kWh, and the maximum charging and
1332
+ discharging rates are ±3 kW, respectively [33]. The forecasted
1333
+ solar PV generation is chosen from 01/01/2021 with ∆T = 15
1334
+ mins in California from CAISO [34].
1335
+ In total c = 4 clouds are responsible for message aggrega-
1336
+ tion and distribution. The degree of all polynomials is set to
1337
+ be d−1 = 3 and the integer field is chosen as E = [0, 231−1).
1338
+ For the fixed-point number quantization, the basis, magnitude,
1339
+ and resolution are uniformly set to be θ = 2, γ = 27, and
1340
+ ζ = 4, respectively. For the distribution network shown in
1341
+ Fig. 1, all 24 houses are assumed to be located in the same
1342
+ area with identical solar radiation. The baseline load profiles
1343
+
1344
+ 8
1345
+ 00:00
1346
+ 04:00
1347
+ 08:00
1348
+ 12:00
1349
+ 16:00
1350
+ 20:00
1351
+ 24:00
1352
+ Time
1353
+ 0.6
1354
+ 0.8
1355
+ 1.0
1356
+ 1.2
1357
+ 1.4
1358
+ 1.6
1359
+ Power (kW)
1360
+ (a) Heterogeneous baseline loads of 24
1361
+ houses
1362
+ 00:00
1363
+ 04:00
1364
+ 08:00
1365
+ 12:00
1366
+ 16:00
1367
+ 20:00
1368
+ 24:00
1369
+ Time
1370
+ 0.0
1371
+ 0.5
1372
+ 1.0
1373
+ 1.5
1374
+ 2.0
1375
+ Solar PV generations (kW)
1376
+ (b) Solar power injection of 24 houses
1377
+ 00:00
1378
+ 04:00
1379
+ 08:00
1380
+ 12:00
1381
+ 16:00
1382
+ 20:00
1383
+ 24:00
1384
+ Time
1385
+ −1.00
1386
+ −0.75
1387
+ −0.50
1388
+ −0.25
1389
+ 0.00
1390
+ 0.25
1391
+ 0.50
1392
+ 0.75
1393
+ 1.00
1394
+ Charging/discharging (kW)
1395
+ (c) Charging and discharging power from
1396
+ 24 ESSs
1397
+ 00:00
1398
+ 04:00
1399
+ 08:00
1400
+ 12:00
1401
+ 16:00
1402
+ 20:00
1403
+ 24:00
1404
+ Time
1405
+ 0
1406
+ 5
1407
+ 10
1408
+ 15
1409
+ 20
1410
+ 25
1411
+ Power (kW)
1412
+ (d) Power flows of 12 lines in the
1413
+ distribution network
1414
+ Fig. 5. The optimal solutions of (P2) by controlling DERs in the distribution network.
1415
+ of all houses are shown in Fig. 5(a) [34]. The primal and dual
1416
+ step sizes are chosen based on experience to be αv
1417
+ ν,ℓ = 2.3,
1418
+ αe
1419
+ σ,ℓ = 1.8, and βµlι,ℓ = 5×10−4, respectively. Note that only
1420
+ the lower bound of power flow limits in (16) is active, herein,
1421
+ only the results related to µlι are presented.
1422
+ Fig. 5(b) and Fig. 5(c) show the active power generations
1423
+ and the charging/discharging power from the solar PVs and
1424
+ ESSs, respectively. At around 12:00, the solar PVs generate
1425
+ the maximum amount of energy, and the ESSs charge at
1426
+ peak rates. After 16:00, energy stored in ESSs is extracted to
1427
+ supply in-home use and compensate for the power loss in the
1428
+ distribution network. The power flows of 12 lines are shown
1429
+ in Fig. 5(d) where no inverse flows occur. Moreover, accurate
1430
+ primal and dual solutions are achieved without affecting the
1431
+ anticipated primal-dual convergence. The iterative solutions
1432
+ of the primal and dual variables are shown in Fig. 6.
1433
+ Fig.
1434
+ 0
1435
+ 20
1436
+ 40
1437
+ 60
1438
+ 80
1439
+ 100
1440
+ Iterations
1441
+ 0.0
1442
+ 0.5
1443
+ 1.0
1444
+ 1.5
1445
+ 2.0
1446
+ ˜pν
1447
+ (a) Convergence of solar PVs’ decision
1448
+ variables ˜pν
1449
+ 0
1450
+ 100
1451
+ 200
1452
+ 300
1453
+ 400
1454
+ 500
1455
+ 600
1456
+ 700
1457
+ Iterations
1458
+ 0.00
1459
+ 0.01
1460
+ 0.02
1461
+ 0.03
1462
+ 0.04
1463
+ 0.05
1464
+ µlι
1465
+ (b) Convergence of the dual variable µlι
1466
+ Fig. 6. Convergence of the primal and dual variables
1467
+ Fig. 7. Random shares generated by Bus 6 at different iterations
1468
+ 7 presents normalized shares generated by Bus 6 using the
1469
+ random polynomial y(ℓ)
1470
+ 6 (z) = ω(ℓ)
1471
+ 6
1472
+ + a(ℓ)
1473
+ 1 z + a2z2 + a(ℓ)
1474
+ 3 z3
1475
+ where the coefficients a(ℓ)
1476
+ i , i = 1, 2, 3 are randomized at each
1477
+ iteration and different time slots. The privacy preservation of
1478
+ Algorithm 1 against external eavesdroppers are guaranteed
1479
+ because external eavesdroppers have insufficient information
1480
+ in polynomial reconstruction by wiretapping the transmitted
1481
+ shares. Without loss of generality, suppose bus 6 is honest-
1482
+ but-curious. Fig. 8 shows the existence of a simulator that
1483
+ −1
1484
+ 0
1485
+ 1
1486
+ ×1015
1487
+ True polynomial y6(z)
1488
+ −100
1489
+ −75
1490
+ −50
1491
+ −25
1492
+ 0
1493
+ 25
1494
+ 50
1495
+ 75
1496
+ 100
1497
+ −1
1498
+ 0
1499
+ 1
1500
+ ×1015
1501
+ Simulated polynomial ˜y′
1502
+ 6(z)
1503
+ True constructed polynomial ˜y6(z)
1504
+ Fig. 8.
1505
+ Polynomials simulated by a simulator to achieve computational
1506
+ indistinguishability among agents
1507
+ can generate true polynomial y6(z) and simulated polyno-
1508
+ mials y′
1509
+ i(z) (dashed lines), ∀i = 1, . . . , n, i ̸= 6, such that
1510
+ (π6 ˜P )′ = π6 ˜P . Therefore, the computational indistinguisha-
1511
+ bility ˜y′
1512
+ 6(αj)
1513
+ c≡ ˜y6(αj), ∀j = 1, . . . , c is satisfied at any
1514
+ iteration and any time slot, and herein π1 ˜P , . . . , πn ˜P can
1515
+ be securely computed among buses and the ith bus can only
1516
+ know the information contained in its own view VIEWi.
1517
+ VI. CONCLUSION
1518
+ This
1519
+ paper
1520
+ proposed
1521
+ a
1522
+ novel
1523
+ decentralized
1524
+ privacy-
1525
+ preserving algorithm with cloud computing architecture for
1526
+ DER control in distribution networks. The DER control prob-
1527
+ lem was formulated into a constrained optimization problem
1528
+ with the objectives of minimizing the line loss, PV curtailment
1529
+ cost, and ESS degradation cost. By integrating SS into the
1530
+ decentralized PGM, the proposed approach achieved privacy
1531
+ preservation for DER owners’ private data, including the
1532
+ DERs’ generation, consumption and daily electricity usage.
1533
+ The security of the proposed approach was proved rigor-
1534
+ ously with privacy guarantees and analyses against honest-but-
1535
+ curious agents and external eavesdroppers. Simulation results
1536
+ verified the applicability of the proposed approach on the
1537
+ modified IEEE 13-bus test feeder with controllable ESSs
1538
+ and solar PVs. Moreover, the designed methodology can be
1539
+ readily used in general large-scale decentralized optimization
1540
+ problems in the context of privacy preservation provisions.
1541
+ APPENDIX A
1542
+ DERIVATION OF THE PGM UPDATES
1543
+ We take the IEEE 13-bus test feeder in Fig. 1 for example
1544
+ to illustrate the derivation of subgradients in (18). To prove
1545
+
1546
+ 0.5
1547
+ 0.4
1548
+ 0.3
1549
+ 0.2
1550
+ 0.1
1551
+ 0.0
1552
+ 00:00
1553
+ 04:00
1554
+ 08:00
1555
+ 104
1556
+ 12:00
1557
+ 103
1558
+ 16:00
1559
+ 102
1560
+ Time
1561
+ 20:00
1562
+ 101
1563
+ 24:009
1564
+ (18a), we firstly consider the subgradient of the power loss
1565
+ minimization objective, the active power loss is
1566
+ f1(pg
1567
+ 1, . . . , pg
1568
+ n) = δ1
1569
+
1570
+ lij∈L
1571
+ rij
1572
+ �∥Pij∥2
1573
+ 2
1574
+ V 2
1575
+ 0
1576
+
1577
+ = δ1¯r
1578
+ V 2
1579
+ 0
1580
+
1581
+ ι∈L
1582
+ ∥Pι∥2
1583
+ 2
1584
+ =
1585
+ ¯δ1
1586
+ 2
1587
+
1588
+ ι∈L
1589
+ ∥Pι∥2
1590
+ 2.
1591
+ (38)
1592
+ Take (15) into (38), we have
1593
+ f1(pg
1594
+ 1, . . . , pg
1595
+ n) =
1596
+ ¯δ1
1597
+ 2
1598
+
1599
+ ι∈L
1600
+ ∥ ˜Zι ˜P ∥2
1601
+ 2.
1602
+ (39)
1603
+ Without loss of generality, assume the νth PV with decision
1604
+ variable ˜pν is connected at bus i, we have
1605
+ ∇ ˜pνL(·) = δ1∇ ˜pνf1(pg
1606
+ 1, . . . , pg
1607
+ n) + δ2∇ ˜pνf2(˜pν)
1608
+ +
1609
+ L
1610
+
1611
+ ι=1
1612
+ ∇ ˜pνµT
1613
+ uι( ˜Zι ˜P −Pι)−
1614
+ L
1615
+
1616
+ ι=1
1617
+ ∇ ˜pνµT
1618
+ lι ˜Zι ˜P . (40)
1619
+ Substitute (14) and (38) into the first term of (40), we have
1620
+ δ1∇ ˜pνf1(·) =
1621
+ ¯δ1
1622
+ 2 ∇ ˜pν
1623
+
1624
+ ι∈L
1625
+ ∥ ˜Zι ˜P ∥2
1626
+ 2
1627
+ = ¯δ1
1628
+
1629
+ ι∈L
1630
+
1631
+ ∇ ˜pν ˜Zι
1632
+ n
1633
+
1634
+ ˆı=1
1635
+ ∆ˆı ˜pˆı
1636
+ � �
1637
+ ˜Zι ˜P
1638
+
1639
+ = ¯δ1
1640
+
1641
+ ι∈L
1642
+
1643
+ ˜Zι∆i
1644
+ �T �
1645
+ ˜Zι ˜P
1646
+
1647
+ .
1648
+ (41)
1649
+ Take the subgradient of (5), the second term in (40) becomes
1650
+ δ2∇ ˜pνf2(˜pν) = δ2∇ ˜pν∥˜pν − pv
1651
+ ν∥2
1652
+ 2 = 2δ2 (˜pν − pv
1653
+ ν) . (42)
1654
+ Then, substitute (14) into the third term of (40) on the right
1655
+ hand side, we have
1656
+ L
1657
+
1658
+ ι=1
1659
+ ∇ ˜pνµT
1660
+ uι( ˜Zι ˜P − Pι) =
1661
+ L
1662
+
1663
+ ι=1
1664
+ ∇ ˜pνµT
1665
+ uι ˜Zι(
1666
+ n
1667
+
1668
+ ˆı=1
1669
+ ∆ˆı ˜pˆı)
1670
+ =
1671
+ L
1672
+
1673
+ ι=1
1674
+ ( ˜Zι∆i)
1675
+ Tµuι.
1676
+ (43)
1677
+ Similarly, the last term of (40) can be readily obtained as
1678
+
1679
+ L
1680
+
1681
+ ι=1
1682
+ ∇ ˜pνµT
1683
+ lι( ˜Zι ˜P ) = −
1684
+ L
1685
+
1686
+ ι=1
1687
+ ( ˜Zι∆i)
1688
+ Tµlι.
1689
+ (44)
1690
+ Finally, by substituting (41), (42), (43), (44) into (40), (18a)
1691
+ is readily proved. Following similar lines, subgradients of the
1692
+ primal variable ˆpσ in (18b) can be readily proved.
1693
+ APPENDIX B
1694
+ PROOF OF THEOREM 2
1695
+ Proof: To prove the correctness of Algorithm 1, we show
1696
+ that the proposed method has the same primal and dual
1697
+ solutions as the non-privacy PGM. Recall that the uth cloud
1698
+ multiplies the received n outputs by the elements of πi
1699
+ according to (31), it yields
1700
+
1701
+
1702
+
1703
+
1704
+
1705
+ πi(1)y1(αu) = πi(1)
1706
+
1707
+ ω1 + a1,1αu + · · · + a1,d−1αd−1
1708
+ u
1709
+
1710
+ ...
1711
+ πi(n)yn(αu) = πi(n)
1712
+
1713
+ ωn + an,1αu + · · · + an,d−1αd−1
1714
+ u
1715
+
1716
+ (45)
1717
+ Then, the aggregated outputs �n
1718
+ ˆı=1 πi(ˆı)yˆı(αu) in (31) can be
1719
+ obtained by summing the left hand side of (45). Therefore, in
1720
+ total c pairs of shares from all clouds as in (32) can be seen
1721
+ as the inputs and outputs of a polynomial
1722
+ ˜y(z) =
1723
+ n
1724
+
1725
+ ˆı=1
1726
+ πi(ˆı)ωˆı + ˜a1z + · · · + ˜ad−1zd−1
1727
+ (46)
1728
+ where ˜aˆȷ = �n
1729
+ ˆı=1 πi(ˆı)aˆı,ˆȷ, ˆȷ = 1, . . . , d���1 and �n
1730
+ ˆı=1 πi(ˆı)ωˆı
1731
+ is exactly πi ˜P . Then, the aggregated secret πi ˜P can be
1732
+ readily retrieved by using c pairs of shares in (33) since d ≤ c,
1733
+ as stated by Theorem 1.
1734
+ APPENDIX C
1735
+ PROOF OF PROPOSITION 1
1736
+ Proof: Under the collusion of d − 1 clouds, they can
1737
+ construct the following set of equations
1738
+
1739
+
1740
+
1741
+
1742
+
1743
+ ˜yi(α1) = ˜ω + ˜ai,1α1 + · · · + ˜ai,d−1αd−1
1744
+ 1
1745
+ ...
1746
+ ˜yi(αd−1) = ˜ω + ˜ai,1αd−1 + · · · + ˜ai,d−1αd−1
1747
+ d−1
1748
+ (47)
1749
+ where ˜yi(z) is defined in (34) and ˜ω = πi ˜P . In (47), ˜ai,ı,
1750
+ ∀ı = 1, . . . , d − 1 and ˜ω are unknown, therefore the d − 1
1751
+ clouds can yield in total d − 1 equations yet d unknowns that
1752
+ leads to underdetermined solutions.
1753
+ APPENDIX D
1754
+ PROOF OF THEOREM 3
1755
+ Proof: To prove the privacy preservation of Algorithm
1756
+ 1 against honest-but-curious agents, we aim at verifying
1757
+ that whatever an honest-but-curious agent receives can be
1758
+ efficiently simulated. That being said, the honest-but-curious
1759
+ agent cannot retrieve useful information from others using the
1760
+ received data because it cannot distinguish the received data
1761
+ from its own. During the ℓth iteration of executing Algorithm
1762
+ 1, the view of bus i can be described via
1763
+ VIEWi = {α1, . . . , αc, θ, γ, ζ, πi ˜P , yi(z), ωi, ¯
1764
+ Ai,
1765
+ ˜yi(αj), ∀j = 1, . . . , c, Cp, Cd}.
1766
+ (48)
1767
+ Based on Definition 2, we need to prove the existence of a
1768
+ polynomial-time algorithm, denoted as simulator S, that can
1769
+ simulate VIEWi using the data of agent i, i.e.,
1770
+ S(Ξi)
1771
+ c≡ VIEWi
1772
+ (49)
1773
+ where Ξi ≜ {α1, . . . , αc, θ, γ, ζ, πi ˜P , yi(z), ωi, ¯
1774
+ Ai, ˜yi(αj),
1775
+ ∀j = 1, . . . , c, Cp, Cd} denotes the set of data that agent
1776
+ i has access to. Manifesting (49) indicates that whatever
1777
+ agent i receives can be efficiently reconstructed based on its
1778
+ own knowledge Ξi. To this end, the simulator is required to
1779
+ generate ˜y′
1780
+ i(αj),∀j = 1, . . . , c that satisfy
1781
+ ˜y′
1782
+ i(αj)
1783
+ c≡ ˜yi(αj), ∀j = 1, . . . , c.
1784
+ (50)
1785
+ To achieve this goal, the simulator firstly generates secrets
1786
+ w′
1787
+ j̸=i ∈ E of other agents such that
1788
+ πi ˜P = wi +
1789
+
1790
+ j̸=i
1791
+ w′
1792
+ j.
1793
+ (51)
1794
+
1795
+ 10
1796
+ Then it generates a set of random polynomials as in (30) to
1797
+ obtain y′
1798
+ j(z), ∀j ̸= i with w′
1799
+ j, ∀j ̸= i as the corresponding
1800
+ constant terms, i.e.,
1801
+
1802
+ yi(z) = wi + ai,1z + · · · + ai,d−1zd−1
1803
+ (52a)
1804
+ y′
1805
+ j(z) = w′
1806
+ j + a′
1807
+ i,1z + · · · + a′
1808
+ i,d−1zd−1, ∀j ̸= i.
1809
+ (52b)
1810
+ Consequently, the simulator can use {α1, . . . , αc} as inputs
1811
+ for (52) and obtain
1812
+ ˜
1813
+ A′
1814
+ i =
1815
+
1816
+
1817
+ �αˆȷ, yi(αˆȷ) +
1818
+
1819
+ j̸=i
1820
+ y′
1821
+ j(αˆȷ), ∀, ˆȷ = 1, . . . , c
1822
+
1823
+
1824
+ � .
1825
+ (53)
1826
+ By Theorem 1 and Theorem 2, the shares in (53) can be
1827
+ used to construct a new polynomial in the form of
1828
+ ˜y′
1829
+ i(x) = (πi ˜P )′ + ˜a′
1830
+ i,1z + · · · + ˜a′
1831
+ i,d−1zd−1
1832
+ (54)
1833
+ where (πi ˜P )′ = πi ˜P . Therefore, (50) and (49) hold, by
1834
+ Definition 2, Algorithm 1 securely computes π1 ˜P , . . . , πn ˜P
1835
+ between the agents.
1836
+ In what follows, we prove the privacy preservation of Al-
1837
+ gorithm 1 against external eavesdroppers. Under Assumption
1838
+ 1, assume agent 1 is safe from external eavesdroppers, by
1839
+ wiretapping any other agents’ communication channels, an
1840
+ external eavesdropper can at most have access to
1841
+ Ξe=
1842
+
1843
+ α1,. . ., αc,yi(αu), ¯
1844
+ Au,i, ∀i=2,. . ., n,u=1, . . ., c
1845
+
1846
+ .
1847
+ (55)
1848
+ Since (55) is insufficient to formulate (33), the external eaves-
1849
+ dropper is incapable of inferring either yi(z)’s or ˜y′
1850
+ i(z)’s,
1851
+ i.e., unable to infer agents’ private information pi’s or the
1852
+ aggregated message πi ˜P ’s.
1853
+ REFERENCES
1854
+ [1] J. Campbell, “Ancillary services provided from DER,” Oak Ridge
1855
+ National Lab, Oak Ridge, TN, United States, Tech. Rep., 2005.
1856
+ [2] J. R. Aguero, E. Takayesu, D. Novosel, and R. Masiello, “Modernizing
1857
+ the grid: Challenges and opportunities for a sustainable future,” IEEE
1858
+ Power Energy Mag., vol. 15, no. 3, pp. 74–83, 2017.
1859
+ [3] I.-K. Song, W.-W. Jung, J.-Y. Kim, S.-Y. Yun, J.-H. Choi, and S.-J.
1860
+ Ahn, “Operation schemes of smart distribution networks with distributed
1861
+ energy resources for loss reduction and service restoration,” IEEE Trans.
1862
+ Smart Grid, vol. 4, no. 1, pp. 367–374, 2012.
1863
+ [4] D. K. Molzahn, F. D¨orfler, H. Sandberg, S. H. Low, S. Chakrabarti,
1864
+ R. Baldick, and J. Lavaei, “A survey of distributed optimization and
1865
+ control algorithms for electric power systems,” IEEE Trans. Smart Grid,
1866
+ vol. 8, no. 6, pp. 2941–2962, 2017.
1867
+ [5] M. Zeraati, M. E. H. Golshan, and J. M. Guerrero, “Distributed control
1868
+ of battery energy storage systems for voltage regulation in distribution
1869
+ networks with high PV penetration,” IEEE Trans. Smart Grid, vol. 9,
1870
+ no. 4, pp. 3582–3593, 2016.
1871
+ [6] Q. Zhang, Y. Guo, Z. Wang, and F. Bu, “Distributed optimal conservation
1872
+ voltage reduction in integrated primary-secondary distribution systems,”
1873
+ IEEE Trans. Smart Grid, vol. 12, no. 5, pp. 3889–3900, 2021.
1874
+ [7] Y. Pan, A. Sangwongwanich, Y. Yang, and F. Blaabjerg, “Distributed
1875
+ control of islanded series PV-battery-hybrid systems with low communi-
1876
+ cation burden,” IEEE Trans. Power Electron., vol. 36, no. 9, pp. 10 199–
1877
+ 10 213, 2021.
1878
+ [8] T. Navidi, A. El Gamal, and R. Rajagopal, “A two-layer decentralized
1879
+ control architecture for DER coordination,” in Proc. IEEE Conf. Decis.
1880
+ Control, Miami, FL, USA, Dec. 17-29 2018, pp. 6019–6024.
1881
+ [9] W. Lin and E. Bitar, “Decentralized stochastic control of distributed
1882
+ energy resources,” IEEE Trans. Power Syst., vol. 33, no. 1, pp. 888–
1883
+ 900, 2017.
1884
+ [10] X. Huo and M. Liu, “Two-facet scalable cooperative optimization of
1885
+ multi-agent systems in the networked environment,” IEEE Trans. Control
1886
+ Syst. Technol., vol. 30, no. 6, pp. 2317–2332, 2022.
1887
+ [11] C. Dwork, F. McSherry, K. Nissim, and A. Smith, “Calibrating noise
1888
+ to sensitivity in private data analysis,” in Proc. Theory Cryptogr. Conf.,
1889
+ New York, NY, USA, Mar. 4-7 2006, pp. 265–284.
1890
+ [12] J. Dong, T. Kuruganti, S. Djouadi, M. Olama, and Y. Xue, “Privacy-
1891
+ preserving aggregation of controllable loads to compensate fluctuations
1892
+ in solar power,” in Proc. IEEE Electron. Power Grid, Charleston, SC,
1893
+ USA, Nov. 12-14 2018, pp. 1–5.
1894
+ [13] S. Han, U. Topcu, and G. J. Pappas, “Differentially private distributed
1895
+ constrained optimization,” IEEE Trans. Autom. Control, vol. 62, no. 1,
1896
+ pp. 50–64, 2016.
1897
+ [14] M. B. Gough, S. F. Santos, T. AlSkaif, M. S. Javadi, R. Castro, and
1898
+ J. P. Catal˜ao, “Preserving privacy of smart meter data in a smart grid
1899
+ environment,” IEEE Trans. Ind. Inform., vol. 18, no. 1, pp. 707–718,
1900
+ 2021.
1901
+ [15] R. Lu, X. Liang, X. Li, X. Lin, and X. Shen, “EPPA: An efficient
1902
+ and privacy-preserving aggregation scheme for secure smart grid com-
1903
+ munications,” IEEE Trans. Parallel Distrib. Syst., vol. 23, no. 9, pp.
1904
+ 1621–1631, 2012.
1905
+ [16] A. Mohammadali and M. S. Haghighi, “A privacy-preserving homomor-
1906
+ phic scheme with multiple dimensions and fault tolerance for metering
1907
+ data aggregation in smart grid,” IEEE Trans. Smart Grid, vol. 12, no. 6,
1908
+ pp. 5212–5220, 2021.
1909
+ [17] Z. Cheng, F. Ye, X. Cao, and M.-Y. Chow, “A homomorphic encryption-
1910
+ based private collaborative distributed energy management system,”
1911
+ IEEE Trans. Smart Grid, vol. 12, no. 6, pp. 5233–5243, 2021.
1912
+ [18] S. Wang, Q. Hu, Y. Sun, and J. Huang, “Privacy preservation in location-
1913
+ based services,” IEEE Commun. Mag., vol. 56, no. 3, pp. 134–140, 2018.
1914
+ [19] R. Gilad-Bachrach, K. Laine, K. Lauter, P. Rindal, and M. Rosulek,
1915
+ “Secure data exchange: A marketplace in the cloud,” in Proc. ACM
1916
+ SIGSAC Conf. Cloud Comput. Security Workshop, London, UK, Nov.
1917
+ 11 2019, pp. 117–128.
1918
+ [20] A. Shamir, “How to share a secret,” Commun. ACM, vol. 22, no. 11,
1919
+ pp. 612–613, 1979.
1920
+ [21] M. Nabil, M. Ismail, M. M. Mahmoud, W. Alasmary, and E. Serpedin,
1921
+ “PPETD: Privacy-preserving electricity theft detection scheme with load
1922
+ monitoring and billing for AMI networks,” IEEE Access, vol. 7, pp.
1923
+ 96 334–96 348, 2019.
1924
+ [22] X. Huo and M. Liu, “Distributed privacy-preserving electric vehicle
1925
+ charging control based on secret sharing,” Electr. Power Syst. Res., vol.
1926
+ 211, p. 108357, 2022.
1927
+ [23] M. Baran and F. F. Wu, “Optimal sizing of capacitors placed on a radial
1928
+ distribution system,” IEEE Trans. Power Deliv., vol. 4, no. 1, pp. 735–
1929
+ 743, 1989.
1930
+ [24] M. Farivar, L. Chen, and S. Low, “Equilibrium and dynamics of local
1931
+ voltage control in distribution systems,” in Proc. IEEE Conf. Decis.
1932
+ Control, Florence, Italy, Dec. 10-13 2013, pp. 4329–4334.
1933
+ [25] J. Li, Z. Xu, J. Zhao, and C. Zhang, “Distributed online voltage control
1934
+ in active distribution networks considering PV curtailment,” IEEE Trans.
1935
+ Ind. Inform., vol. 15, no. 10, pp. 5519–5530, 2019.
1936
+ [26] J. Forman, J. Stein, and H. Fathy, “Optimization of dynamic battery
1937
+ parameter characterization experiments via differential evolution,” in
1938
+ Proc. Am. Control Conf., Washington, DC, USA, Jun. 17-19 2013, pp.
1939
+ 867–874.
1940
+ [27] M. S. Daru and T. Jager, “Encrypted cloud-based control using secret
1941
+ sharing with one-time pads,” in Proc. IEEE Conf. Decis. Control, Nice,
1942
+ France, Dec. 11-13 2019, pp. 7215–7221.
1943
+ [28] J. Humpherys and T. J. Jarvis, Foundations of Applied Mathematics,
1944
+ Volume I: Mathematical Analysis.
1945
+ Soc. Ind. Appl. Math, 2020.
1946
+ [29] O. Goldreich, Foundations of Cryptography: Volume 2, Basic Applica-
1947
+ tions.
1948
+ Cambridge Univ. Press, 2009.
1949
+ [30] D. Evans, V. Kolesnikov, and M. Rosulek, “A pragmatic introduction to
1950
+ secure multi-party computation,” Found. Trends Privacy Security, vol. 2,
1951
+ no. 2-3, pp. 70–246, 2018.
1952
+ [31] O. Goldreich, “Secure multi-party computation,” Manuscript. Prelimi-
1953
+ nary Version, vol. 78, p. 110, 1998.
1954
+ [32] M. Liu, P. K. Phanivong, Y. Shi, and D. S. Callaway, “Decentralized
1955
+ charging control of electric vehicles in residential distribution networks,”
1956
+ IEEE Trans. Control Syst. Technol., vol. 27, no. 1, pp. 266–281, 2019.
1957
+ [33] National Renewable Energy Laboratory. Residential battery storage.
1958
+ [Online].
1959
+ Available:
1960
+ https://atb.nrel.gov/electricity/2021/residential
1961
+ battery storage
1962
+ [34] U.S.
1963
+ Energy
1964
+ Information
1965
+ Administration.
1966
+ Electric
1967
+ power
1968
+ annual.
1969
+ [Online]. Available: https://www.eia.gov/todayinenergy/detail.php?id=
1970
+ 49276
1971
+
E9E0T4oBgHgl3EQfQwCg/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
EtAzT4oBgHgl3EQfUPx6/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c123615b85a30d0ea3672d5747db8a117755bbd3ffda424a91aeb33692a432a
3
+ size 5767213
F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf ADDED
Binary file (86.7 kB). View file
 
F9E3T4oBgHgl3EQfWAol/content/tmp_files/2301.04464v1.pdf.txt ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.04464v1 [math.NT] 8 Jan 2023
2
+ Runs of Consecutive Integers Having the
3
+ Same Number of Divisors
4
+ By Vlad-Titus Sp˘ataru
5
+ Abstract
6
+ Our principal objective is to provide an upper bound for the length ℓN of the
7
+ longest run of consecutive integers smaller than N which have the same number of
8
+ divisors. We prove that ℓN ⩽ exp �C√log N log log N� in an elementary manner.
9
+ 1. Introduction
10
+ The equation d(n) = d(n + k) has been studied extensively.
11
+ In 1981, Spiro [Spi81]
12
+ showed that it has infinitely many solutions for k = 5040.
13
+ Subsequently, Heath-Brown
14
+ [HB84] extended Spiro’s work to deal with the case k = 1, and Pinner [Pin97] ultimately
15
+ proved that, in fact, all values of k yield infinitely many solutions.
16
+ As d(n) = d(n+1) infinitely often, one naturally wonders how many consecutive integers
17
+ can there be, having the same number of divisors. Erd˝os and Mirsky [EM52] conjectured
18
+ that there are arbitrarily long such runs of integers. They were not able to provide any
19
+ estimates for the length of such sequences: “A related problem consists in the estimation of
20
+ the longest run of consecutive integers ⩽ x all of which have the same number of divisors.
21
+ This problem seems to be one of exceptional difficulty, and we [Erd˝os & Mirsky] have not
22
+ been able to make any progress with it.”
23
+ Our principal objective is to provide an upper bound for the length of the runs in
24
+ question. We shall obtain the following result, in an elementary manner:
25
+ Theorem 1. Let ℓN denote the length of the longest run of consecutive integers smaller
26
+ than N, having the same number of divisors. Then,
27
+ ℓN ⩽ exp
28
+ Ä
29
+ C
30
+
31
+ log N · log log N
32
+ ä
33
+ ,
34
+ for an absolute constant C.
35
+ Keywords: divisor counting function, consecutive equidivisible integers.
36
+ 2020 Mathematics Subject Classification: Primary: 11A25, 11N37.
37
+ 1
38
+
39
+ 2
40
+ VLAD-TITUS SP˘ATARU
41
+ 2. The main result
42
+ In proving theorem 1, we will make use of the following lemmas, the first being proven
43
+ in an elementary manner in [Far09] and the second being Mertens’ bound.
44
+ Lemma 1. Let n be a positive integer. Then, lcm(1, 2, . . . , n + 1) ⩾ 2n.
45
+ Lemma 2. There exists an absolute constant C1 such that for any positive integer n ⩾ 2,
46
+
47
+ p⩽n
48
+ 1
49
+ p ⩽ C1 · log log n,
50
+ the sum being over all prime numbers p not exceeding n.
51
+ Note that it suffices to prove that theorem 1 holds for large enough N. Assume that
52
+ there exist k > 2 consecutive numbers smaller than N, having the same number of divisors.
53
+ Let them be n + 1, n + 2, . . . , n + k and write
54
+ d(n + 1) = d(n + 2) = · · · = d(n + k) = D.
55
+ We will firstly provide an estimate for D, in terms of k. For simplicity, let K = ⌊log2 k⌋.
56
+ As k ⩾ 2K, all residues modulo 2K are among n + 1, n + 2, . . . , n + k. Therefore, for all
57
+ 1 ⩽ i ⩽ K − 1, there exists some 1 ⩽ ti ⩽ k such that n + ti ≡ 2i mod 2K. Consequently,
58
+ ν2(n + ti) = i, so (i + 1) divides d(n + ti) = D.
59
+ Hence, D is divisible by lcm(1, 2, . . . , K). Using lemma 1, we infer that
60
+ D ⩾ lcm(1, 2, . . . , K) ⩾ 2K−1.
61
+ Recall that K = ⌊log2 k⌋ ⩾ log2 k − 1, so D ⩾ k/4.
62
+ Next, we will bound ω((n + 1) · · ·(n + k)). Choose 1 ⩽ l ⩽ k arbitrarily. As n + l ⩽ N,
63
+ it follows that νp(n + l) ⩽ logp N ⩽ log2 N for all prime numbers p. Therefore,
64
+ D = d(n + l) =
65
+
66
+ p
67
+ (νp(n + l) + 1) ⩽
68
+
69
+ p|n+l
70
+ (log2 N + 1) = (log2 N + 1)ω(n+l),
71
+ where p always represents a prime number. Thus, ω(n+l) ⩾ log D/ log(log2 N +1). A prime
72
+ number p can divide at most ⌈k/p⌉ ⩽ k/p + 1 numbers among n + 1, . . . , n + k, so
73
+ ω((n + 1) · · ·(n + k)) ⩾
74
+ k
75
+
76
+ i=1
77
+ ω(n + i) −
78
+
79
+ p⩽k
80
+ k
81
+ p,
82
+
83
+ RUNS OF CONSECUTIVE INTEGERS HAVING THE SAME NUMBER OF DIVISORS
84
+ 3
85
+ the second sum being taken over all prime numbers p not exceeding k. Using lemma 2 and
86
+ the inequality we have previously deduced for ω(n + l), we may finally infer that
87
+ ω((n + 1) · · ·(n + k)) ⩾
88
+ k · log D
89
+ log(log2 N + 1) − C1k log log k.
90
+ Further, we will write log(log2 N +1) ⩽ C2 log log N, for an absolute constant C2. Recall
91
+ that D ⩾ k/4, so we have
92
+ ω((n + 1) · · ·(n + k)) ⩾ k · log(k/4)
93
+ C2 log log N − C1k log log k.
94
+ (1)
95
+ Write the right-hand side of equation 1 as k · fN(k). Clearly, if ω(a) ⩾ b then a ⩾ b!.
96
+ Using this remark on equation 1, we get (n+1) · · · (n+k) ⩾ ⌈k ·fN(k)⌉!. Moreover, because
97
+ Nk ⩾ (n + 1) · · ·(n + k), by taking logarithms and applying the well-known inequality
98
+ log t! ⩾ t log t − t, we have
99
+ k log N ⩾ log ((n + 1) · · · (n + k)) ⩾ log (⌈k · fN(k)⌉!)
100
+ ⩾ k · fN(k) · log(k · fN(k)) − k · fN(k).
101
+ (2)
102
+ Finally, dividing equation 2 by k we obtain
103
+ log N ⩾ fN(k) · log(k · fN(k)) − fN(k).
104
+ (3)
105
+ Define the interval IN = [exp (C1 · C2 · log log N) , ∞). Using standard arguments, one
106
+ may infer that fN is increasing on IN.
107
+ Let us suppose, for the sake of contradiction, that k > exp �C√log N log log N�, where
108
+ C > max(√C2, C1 · C2). Firstly, note that since log N > log log N and C > C1 · C2 then
109
+ exp �C√log N log log N� and k are in IN. Therefore, we have
110
+ fN(k) > fN
111
+ Ä
112
+ exp
113
+ Ä
114
+ C
115
+
116
+ log N · log log N
117
+ ää
118
+ = C
119
+ C2
120
+  
121
+ log N
122
+ log log N −
123
+ log 4
124
+ C2 log log N − C1 log
125
+ Ä
126
+ C
127
+
128
+ log N · log log N
129
+ ä
130
+ .
131
+ (4)
132
+ Viewing equation 4 as a function in N, it is evident that for large enough N (greater than
133
+ some N1) we also have fN(k) > e. In what follows, we will assume that N > N1.
134
+ As fN(k) > e, it follows from equation 3 that log N ⩾ fN(k) · log k. Further, applying
135
+ equation 4 and the estimate for k and isolating the term log N, we get
136
+ C log 4
137
+ C2
138
+  
139
+ log N
140
+ log log N + C1C
141
+
142
+ log N log log N log
143
+ Ä
144
+ C
145
+
146
+ log N log log N
147
+ ä
148
+
149
+ ÅC2
150
+ C2
151
+ − 1
152
+ ã
153
+ log N.
154
+ Recall that C > √C2, so the latter inequality is absurd for large enough N (greater than some
155
+ N2), as the left-hand side is asymptotically much smaller than log N. Therefore, theorem 1
156
+ holds for N > max(N1, N2) and C > max(√C2, C1 · C2).
157
+
158
+ 4
159
+ VLAD-TITUS SP˘ATARU
160
+ 3. Acknowledgments
161
+ The author thanks Alexandru Gica for his proofreading and valuable comments.
162
+ References
163
+ [EM52] P. Erd˝os and L. Mirsky, The distribution of values of the divisor function d(n), Pro-
164
+ ceedings of the London Mathematical Society no. 1 (1952), 257–271.
165
+ [Far09] B. Farhi, An identity involving the least common multiple of binomial coefficients and its
166
+ application, The American Mathematical Monthly 116 no. 9 (2009), 836–839.
167
+ [HB84] D. R. Heath-Brown, The divisor function at consecutive integers, Mathematika 31 no. 1
168
+ (1984), 141–149.
169
+ [Pin97] C. G. Pinner, Repeated values of the divisor function, The Quarterly Journal of Mathe-
170
+ matics 48 no. 4 (1997), 499–502.
171
+ [Spi81] C. A. Spiro, The Frequency with Which an Integral-Valued, Prime-Independent, Mul-
172
+ tiplicative or Additive Function of n Divides a Polynomial Function of n, Ph.D. thesis,
173
+ University of Illinois at Urbana-Champaign, 1981.
174
+ V. T. Sp˘ataru, Bucharest, Romania
175
+ E-mail : vtspataru@gmail.com
176
+
F9E3T4oBgHgl3EQfWAol/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf,len=106
2
+ page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
3
+ page_content='04464v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
4
+ page_content='NT] 8 Jan 2023 Runs of Consecutive Integers Having the Same Number of Divisors By Vlad-Titus Sp˘ataru Abstract Our principal objective is to provide an upper bound for the length ℓN of the longest run of consecutive integers smaller than N which have the same number of divisors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
5
+ page_content=' We prove that ℓN ⩽ exp �C√log N log log N� in an elementary manner.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
6
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
7
+ page_content=' Introduction The equation d(n) = d(n + k) has been studied extensively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
8
+ page_content=' In 1981, Spiro [Spi81] showed that it has infinitely many solutions for k = 5040.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
9
+ page_content=' Subsequently, Heath-Brown [HB84] extended Spiro’s work to deal with the case k = 1, and Pinner [Pin97] ultimately proved that, in fact, all values of k yield infinitely many solutions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
10
+ page_content=' As d(n) = d(n+1) infinitely often, one naturally wonders how many consecutive integers can there be, having the same number of divisors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
11
+ page_content=' Erd˝os and Mirsky [EM52] conjectured that there are arbitrarily long such runs of integers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
12
+ page_content=' They were not able to provide any estimates for the length of such sequences: “A related problem consists in the estimation of the longest run of consecutive integers ⩽ x all of which have the same number of divisors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
13
+ page_content=' This problem seems to be one of exceptional difficulty, and we [Erd˝os & Mirsky] have not been able to make any progress with it.” Our principal objective is to provide an upper bound for the length of the runs in question.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
14
+ page_content=' We shall obtain the following result, in an elementary manner: Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
15
+ page_content=' Let ℓN denote the length of the longest run of consecutive integers smaller than N, having the same number of divisors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
16
+ page_content=' Then, ℓN ⩽ exp Ä C � log N · log log N ä , for an absolute constant C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
17
+ page_content=' Keywords: divisor counting function, consecutive equidivisible integers.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
18
+ page_content=' 2020 Mathematics Subject Classification: Primary: 11A25, 11N37.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
19
+ page_content=' 1 2 VLAD-TITUS SP˘ATARU 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
20
+ page_content=' The main result In proving theorem 1, we will make use of the following lemmas, the first being proven in an elementary manner in [Far09] and the second being Mertens’ bound.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
21
+ page_content=' Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
22
+ page_content=' Let n be a positive integer.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
23
+ page_content=' Then, lcm(1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
24
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
25
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
26
+ page_content=' , n + 1) ⩾ 2n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
27
+ page_content=' Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
28
+ page_content=' There exists an absolute constant C1 such that for any positive integer n ⩾ 2, � p⩽n 1 p ⩽ C1 · log log n, the sum being over all prime numbers p not exceeding n.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
29
+ page_content=' Note that it suffices to prove that theorem 1 holds for large enough N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
30
+ page_content=' Assume that there exist k > 2 consecutive numbers smaller than N, having the same number of divisors.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
31
+ page_content=' Let them be n + 1, n + 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
32
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
33
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
34
+ page_content=' , n + k and write d(n + 1) = d(n + 2) = · · · = d(n + k) = D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
35
+ page_content=' We will firstly provide an estimate for D, in terms of k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
36
+ page_content=' For simplicity, let K = ⌊log2 k⌋.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
37
+ page_content=' As k ⩾ 2K, all residues modulo 2K are among n + 1, n + 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
38
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
39
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
40
+ page_content=' , n + k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
41
+ page_content=' Therefore, for all 1 ⩽ i ⩽ K − 1, there exists some 1 ⩽ ti ⩽ k such that n + ti ≡ 2i mod 2K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
42
+ page_content=' Consequently, ν2(n + ti) = i, so (i + 1) divides d(n + ti) = D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
43
+ page_content=' Hence, D is divisible by lcm(1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
44
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
45
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
46
+ page_content=' , K).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
47
+ page_content=' Using lemma 1, we infer that D ⩾ lcm(1, 2, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
48
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
49
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
50
+ page_content=' , K) ⩾ 2K−1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
51
+ page_content=' Recall that K = ⌊log2 k⌋ ⩾ log2 k − 1, so D ⩾ k/4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
52
+ page_content=' Next, we will bound ω((n + 1) · · ·(n + k)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
53
+ page_content=' Choose 1 ⩽ l ⩽ k arbitrarily.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
54
+ page_content=' As n + l ⩽ N, it follows that νp(n + l) ⩽ logp N ⩽ log2 N for all prime numbers p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
55
+ page_content=' Therefore, D = d(n + l) = � p (νp(n + l) + 1) ⩽ � p|n+l (log2 N + 1) = (log2 N + 1)ω(n+l), where p always represents a prime number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
56
+ page_content=' Thus, ω(n+l) ⩾ log D/ log(log2 N +1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
57
+ page_content=' A prime number p can divide at most ⌈k/p⌉ ⩽ k/p + 1 numbers among n + 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
58
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
59
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
60
+ page_content=' , n + k, so ω((n + 1) · · ·(n + k)) ⩾ k � i=1 ω(n + i) − � p⩽k k p, RUNS OF CONSECUTIVE INTEGERS HAVING THE SAME NUMBER OF DIVISORS 3 the second sum being taken over all prime numbers p not exceeding k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
61
+ page_content=' Using lemma 2 and the inequality we have previously deduced for ω(n + l), we may finally infer that ω((n + 1) · · ·(n + k)) ⩾ k · log D log(log2 N + 1) − C1k log log k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
62
+ page_content=' Further, we will write log(log2 N +1) ⩽ C2 log log N, for an absolute constant C2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
63
+ page_content=' Recall that D ⩾ k/4, so we have ω((n + 1) · · ·(n + k)) ⩾ k · log(k/4) C2 log log N − C1k log log k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
64
+ page_content=' (1) Write the right-hand side of equation 1 as k · fN(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
65
+ page_content=' Clearly, if ω(a) ⩾ b then a ⩾ b!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
66
+ page_content='. Using this remark on equation 1, we get (n+1) · · · (n+k) ⩾ ⌈k ·fN(k)⌉!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
67
+ page_content='. Moreover, because Nk ⩾ (n + 1) · · ·(n + k), by taking logarithms and applying the well-known inequality log t!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
68
+ page_content=' ⩾ t log t − t, we have k log N ⩾ log ((n + 1) · · · (n + k)) ⩾ log (⌈k · fN(k)⌉!' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
69
+ page_content=') ⩾ k · fN(k) · log(k · fN(k)) − k · fN(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
70
+ page_content=' (2) Finally, dividing equation 2 by k we obtain log N ⩾ fN(k) · log(k · fN(k)) − fN(k).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
71
+ page_content=' (3) Define the interval IN = [exp (C1 · C2 · log log N) , ∞).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
72
+ page_content=' Using standard arguments, one may infer that fN is increasing on IN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
73
+ page_content=' Let us suppose, for the sake of contradiction, that k > exp �C√log N log log N�, where C > max(√C2, C1 · C2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
74
+ page_content=' Firstly, note that since log N > log log N and C > C1 · C2 then exp �C√log N log log N� and k are in IN.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
75
+ page_content=' Therefore, we have fN(k) > fN Ä exp Ä C � log N · log log N ää = C C2 log N log log N − log 4 C2 log log N − C1 log Ä C � log N · log log N ä .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
76
+ page_content=' (4) Viewing equation 4 as a function in N, it is evident that for large enough N (greater than some N1) we also have fN(k) > e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
77
+ page_content=' In what follows, we will assume that N > N1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
78
+ page_content=' As fN(k) > e, it follows from equation 3 that log N ⩾ fN(k) · log k.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
79
+ page_content=' Further, applying equation 4 and the estimate for k and isolating the term log N, we get C log 4 C2 log N log log N + C1C � log N log log N log Ä C � log N log log N ä ⩾ ÅC2 C2 − 1 ã log N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
80
+ page_content=' Recall that C > √C2, so the latter inequality is absurd for large enough N (greater than some N2), as the left-hand side is asymptotically much smaller than log N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
81
+ page_content=' Therefore, theorem 1 holds for N > max(N1, N2) and C > max(√C2, C1 · C2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
82
+ page_content=' 4 VLAD-TITUS SP˘ATARU 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
83
+ page_content=' Acknowledgments The author thanks Alexandru Gica for his proofreading and valuable comments.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
84
+ page_content=' References [EM52] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
85
+ page_content=' Erd˝os and L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
86
+ page_content=' Mirsky, The distribution of values of the divisor function d(n), Pro- ceedings of the London Mathematical Society no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
87
+ page_content=' 1 (1952), 257–271.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
88
+ page_content=' [Far09] B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
89
+ page_content=' Farhi, An identity involving the least common multiple of binomial coefficients and its application, The American Mathematical Monthly 116 no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
90
+ page_content=' 9 (2009), 836–839.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
91
+ page_content=' [HB84] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
92
+ page_content=' R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
93
+ page_content=' Heath-Brown, The divisor function at consecutive integers, Mathematika 31 no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
94
+ page_content=' 1 (1984), 141–149.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
95
+ page_content=' [Pin97] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
96
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
97
+ page_content=' Pinner, Repeated values of the divisor function, The Quarterly Journal of Mathe- matics 48 no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
98
+ page_content=' 4 (1997), 499–502.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
99
+ page_content=' [Spi81] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
100
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
101
+ page_content=' Spiro, The Frequency with Which an Integral-Valued, Prime-Independent, Mul- tiplicative or Additive Function of n Divides a Polynomial Function of n, Ph.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
102
+ page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
103
+ page_content=' thesis, University of Illinois at Urbana-Champaign, 1981.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
104
+ page_content=' V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
105
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
106
+ page_content=' Sp˘ataru, Bucharest, Romania E-mail : vtspataru@gmail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
107
+ page_content='com' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/F9E3T4oBgHgl3EQfWAol/content/2301.04464v1.pdf'}
FNAyT4oBgHgl3EQf4_q8/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:666a81a47f06234618ac90e3cddccdb82bda00ef08f3e82cc951762b5bb05d0d
3
+ size 5439533
GNE1T4oBgHgl3EQfFANy/content/tmp_files/2301.02897v1.pdf.txt ADDED
@@ -0,0 +1,1197 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+
3
+ Catalytic action of two-dimensional layered materials (WS2, and MoS2) on hydrogen
4
+ sorption properties of MgH2
5
+ Satish Kumar Verma1, Mohammad Abu Shaz1, Thakur Prasad Yadav1,2*
6
+ 1Hydrogen Energy Centre, Department of Physics, Banaras Hindu University, Varanasi-
7
+ 221005, India.
8
+ 2Department of Physics, Faculty of Science, University of Allahabad, Prayagraj-211002,
9
+ India.
10
+
11
+ Abstract:
12
+ The present study reports the catalytic action of two-dimensional (2D) layered materials
13
+ (MoS2 and WS2) for improving the de/re-hydrogenation kinetics of MgH2. The MgH2
14
+ start desorbing at 277 ºC with a hydrogen storage capacity of 5.95 wt% in the presence of
15
+ WS2 catalyst whereas onset desorption temperature of MgH2 catalyzed by MoS2 is 330
16
+ ºC. The MgH2-WS2 absorbed hydrogen ~ 3.72 wt% within 1.3 minutes at 300 ºC under 13
17
+ atm hydrogen pressure and it desorbed ~5.57 wt% within 20 minutes at 300 ºC under 1
18
+ atm hydrogen pressure. We have performed 25 cycles of dehydrogenation (under 1 atm
19
+ hydrogen pressure at 300 ºC) and re-hydrogenation (under 13 atm hydrogen pressure at
20
+ 300 °C) to ensure cyclic stability of catalyzed version of MgH2 where MgH2-WS2 shows
21
+ better cyclic stability than MgH2-MoS2. MgH2-WS2 also shows the lower reaction
22
+ activation energy ~117 kJ/mol as compare to other catalyzed and uncatalyzed samples.
23
+ On the other hand, these catalysts (WS2 and MoS2) do not have any impact on the
24
+ thermodynamical parameters that is change in enthalpy.
25
+
26
+ Key words: 2D layered materials, De/re-hydrogenation kinetics, Activation energy,
27
+ MgH2.
28
+ *Corresponding author Email: yadavtp@gmail.com
29
+
30
+
31
+ 2
32
+
33
+ 1. Introduction
34
+ A crucial and promising area of research for onboard hydrogen applications is the
35
+ development of safe and efficient hydrogen storage. The solid-state approach is one of
36
+ the most appropriate, secure, and effective ways to store hydrogen among the several
37
+ methods that can be used, including gaseous, liquid, and solid-state storage [1,2]. Due to
38
+ its high hydrogen storage capacity (110 g/L volumetric and 7.6 wt% gravimetric), low
39
+ cost, light weight, and large abundance (in the form of Mg) in earth crust (8th most) and
40
+ seawater (3rd most), MgH2 is a leading choice for hydrogen storage in the solid-state
41
+ mode [3–6]. According to the United States Department of Energy (US DOE) technical
42
+ targets for hydrogen storage systems [7], MgH2 has certain advantages that make it a
43
+ viable option. The high dehydrogenation temperature (above 400 ºC), slow kinetics
44
+ (hydrogen de/re-hydrogenation kinetics 0.4 kg-H2/min), and high thermodynamic
45
+ properties (high reaction enthalpy 74 kJ/mol) of MgH2 prevent it from being a suitable
46
+ material for onboard applications even with these advantages [8–10]. In recent years, the
47
+ creation of suitable catalyst(s), alloys, composite materials with complicated hydrides,
48
+ and scaffolding have all been used as feasible methods to improve the hydrogen storage
49
+ performance of MgH2 [11,12]. The use of various types of catalysts and additives to
50
+ enhance the performance of Mg/MgH2 has been the subject of several studies by various
51
+ research organizations [13–17].
52
+ Another application for the 2D materials is as a catalyst for improving the hydrogen
53
+ characteristics of MgH2 [18–22]. Due to its enormous surface area, ballistic conduction,
54
+ thermal conductivity, mechanical stability, and light weight, graphene, which has a 2D
55
+ planer structure with sp2 carbon atoms arranged in a hexagonal framework, has attracted
56
+ a lot of attention as a catalyst and as a template material for hydrogen storage application
57
+ in MgH2 [23,24]. MgH2's de/rehydrogenation kinetics exhibit effective catalytic behavior
58
+ in the graphene layer, which also inhibits MgH2's agglomeration and grain growth
59
+ [3,5,25]. Liu et al., for instance, have created MgH2-5% Gr nanosheets [26]. They have
60
+ demonstrated that graphene nanosheets offer a significant hydrogen diffusion pathway
61
+ and prevent MgH2 from aggregating. According to Huang et al., [27] report's MgH2
62
+ nanoparticles supported by graphene exhibit remarkable hydrogen sorption kinetics and
63
+
64
+ 3
65
+
66
+ cyclic stability. Due to the strong interaction between graphene and MgH2 nanoparticles
67
+ and the prevention of nanoparticle agglomeration, the MgH2 nanoparticles demonstrated
68
+ excellent hydrogen storage performance. Additionally, grapheme prevents the
69
+ aggregation of nanoparticles during the rehydrogenation of MgH2, according to a
70
+ theoretical study using molecular dynamics simulation [28]. Rough studies are still
71
+ required to determine the impact of graphene and other 2D layered materials on MgH2,
72
+ even though some prior studies have shown the remarkable catalytic/co-catalytic and
73
+ agglomeration blocking properties of Gr on MgH2.
74
+ We have examined a comparison between WS2 and MoS2 as a catalyst for enhancing
75
+ hydrogen sorption properties of MgH2. WS2 and MoS2 are suitable alternatives to
76
+ graphene for the catalytic action on MgH2 due to their high conductivity (metallic
77
+ nature), thermal stability, and strong catalytic behavior [29,30]. Tungsten (W) and
78
+ Molybdenum (Mo) are sandwiched between two Sulphur layers with weak Van der
79
+ Waals interactions in the family of layered transition-metal dichalcogenides (TMDs)
80
+ materials that include WS2 and MoS2. The re/de-hydrogenation kinetics, and catalytic
81
+ behavior of WS2 and MoS2 on MgH2 has been investigated in details.
82
+ 2. Experimental section
83
+ 2.1. Synthesis of a few layered WS2
84
+ The bulk tungsten sulfide (WS2) (99.80 %) powder was procured from the Alfa Aesar for
85
+ the present investigation. For the preparation of few layered WS2, WS2 powder was
86
+ dispersed in de-ionized water and sonicated it for 74 hours using ultrasonicator at 20 kHz
87
+ frequency. The sonicated sample was then dried at 50 °C under a dynamic vacuum of
88
+ order 10-2 torr to form the few layered WS2 powder. This preparation method can also be
89
+ understood by the schematic given in Fig. 1.
90
+
91
+ 4
92
+
93
+
94
+
95
+
96
+
97
+
98
+
99
+
100
+
101
+
102
+
103
+
104
+
105
+
106
+ Fig.1: Schematic diagram for the synthesis of a few layers WS2.
107
+
108
+ 2.2. Synthesis of few-layer MoS2
109
+ The Otto Chemica bulk molybdenum disulfide (MoS2) (99 %) powder was used for the
110
+ present investigation. MoS2 powder was dispersed in de-ionized water and sonicate it for
111
+ 74 hours using ultrasonicator at 20 kHz frequency to obtain the few-layered MoS2. The
112
+ sonicated sample was then dried at 50 °C under dynamic vacuum of order 10-2 torr to
113
+ form the few layered MoS2 powder. Fig. 2, shows the schematic diagram for preparation
114
+ of few-layered MoS2.
115
+
116
+
117
+
118
+
119
+
120
+
121
+
122
+
123
+
124
+
125
+
126
+ BulkMoS2
127
+ BulkMoS,
128
+ FewlayeredMoS2
129
+ DoublelayeredMoS2
130
+ UltrasonicationofbulkMoS,BulkWS2
131
+ BulkWs,
132
+ FewlayeredwS2
133
+ DoublelayeredWS2
134
+ Ultrasonicationof bulkWS25
135
+
136
+
137
+ Fig.2: Schematic diagram for the synthesis of a few layers MoS2.
138
+
139
+ 2.3. Synthesis of MgH2 catalyzed by WS2, and MoS2
140
+ The pure MgH2 was procured from Fujifilm (Japan) (99.9%) for the present investigation.
141
+ Mechanical ball-milling of MgH2 with graphene at 180 rpm for 24 hours with a ball-to-
142
+ powder ratio of 50:1 (by weight) using a planetary ball-miller (Retsch PM 400) was used
143
+ to synthesize MgH2 catalyzed by WS2 (MgH2-WS2). To explore the optimum catalyst
144
+ concentration for hydrogen sorption kinetics of Mg/MgH2, we have synthesized a set of
145
+ different catalyst concentrations (5, 10, 12 wt%) to catalyze MgH2. For hydrogen
146
+ sorption in Mg/MgH2, 10 wt% catalysts were found to be optimal (in terms of desorption
147
+ temperature and hydrogen storage capacity). The ball-miller vials were filled with 5 atm
148
+ H2 pressure to compensate for the loss of hydrogen from MgH2 during milling. All the
149
+ loading and unloading of the samples was done inside the N2-filled glove box
150
+ (MBRAUM MB10 compact) with O2 and H2O levels < 1 ppm. The synthesis of MgH2
151
+ catalyzed by MoS2 (MgH2-MoS2) was done using the same synthesis route as MgH2-
152
+ WS2.
153
+ 2.4. Characterization techniques
154
+ The structural characterization of prepared samples was carried out by XRD technique
155
+ using Empyrean PANalytical X-ray diffractometer equipped with 2D detector with a Cu
156
+ Kα beam (λ = 1.5415 Å) operated at 40 kV and 40 mA. The microstructural and selected
157
+ area electron diffraction (SAED) analysis of as-prepared samples was carried out by
158
+ TEM (Technai-20G2) operating at the accelerating voltage of 200 kV. Perkin Elmer
159
+ (Spectrum 100) spectrometer in transmission mode with attenuated total reflectance
160
+ (ATR) sampling mode (wavenumber range 500–4000 cm-1) was used to carry out FTIR
161
+ spectroscopy. The Raman spectra have been acquired at -60 ºC using Horiba-Jobin-Yvon
162
+ LABRAM-HR800 spectrometer with diode LASER (532 nm). The desired thickness and
163
+ surface topography of the prepared samples were examined by using solver next AFM in
164
+ non-contact mode. The characterized samples then proceed for the hydrogen desorption
165
+ and absorption using automated two-channel volumetric sieverts type apparatus. The
166
+ temperature programmed desorption (TPD) was carried out with a heating rate of 5
167
+ oC-
168
+
169
+ 6
170
+
171
+ min-1. The activation energy (Ea) study of prepared catalyzed samples has been done by
172
+ using DSC (Perkin Elmer DSC 8000) with a heating rate of 15
173
+ oC/min, 18
174
+ oC/min, 21
175
+ oC/min, and 24
176
+ oC/min under nitrogen atmosphere (20 ml/min).
177
+
178
+ 3. Results and discussion
179
+ 3.1. Structural, microstructural, and spectroscopic characterization analysis
180
+ The structural characteristics of as-prepared samples have been examined using the XRD
181
+ characterization. Fig. 3(a) shows the XRD pattern of pristine MgH2, which matches well
182
+ with the tetragonal MgH2 with space group P42/mnm (136) and a=b= 4.516 Å, c = 3.020
183
+ Å (JCPDS no. 740934). Fig. 3(b) shows the XRD pattern of MoS2, which matches well
184
+ with the hexagonal structure of MoS2 with space group P63/mmc(194) and a=b= 3.1602
185
+ Å, c = 12.294 Å (Joint Committee on Powder Diffraction Standards (JCPDS) no.
186
+ 651951). The XRD pattern of as-prepared WS2 is shown in Fig. 3(c), that matches well
187
+ with the hexagonal structure of WS2 with space group P63/mmc(194) and a=b= 3.1532
188
+ Å, c = 12.323 Å (JCPDS no. 841398). The usual diffraction pattern of MgH2-MoS2, and
189
+ MgH2-WS2 are shown in Fig. 3(d-e), respectively, where besides the tetragonal phase of
190
+ MgH2, some peaks of WS2 and MoS2 are either suppressed or masked by the peaks of
191
+ MgH2. The diffraction peaks of WS2 and MoS2 are identified and labeled in the Fig. 3(d-
192
+ e), respectively.
193
+ The different bands position, shapes, and relative intensities of Raman spectra give us
194
+ essential information about the materials and stacking of layers, i.e., Raman spectroscopy
195
+ can determine the layer thickness at the atomic level. The Raman spectra of as-prepared
196
+ WS2, and MoS2 have shown in Fig. 4. In the case of MoS2, the two Raman modes are
197
+ appeared at ~ 345 cm-1 and ~ 370 cm-1 corresponds to E12g and A1g modes of vibrations
198
+ (labeled in Fig. 4(b)). The indicated modes of MoS2 have frequency difference of ~ 25
199
+ cm-1, that means the MoS2 as layered material with few layers of stacking (3-5 layers)
200
+ [31,32]. The FWHM of A1g mode is ~ 7 cm-1, which can also be referred to stacking a
201
+ few layers of MoS2 [33]. The Raman shifts at ~316 cm-1 and 384 cm-1 (shown in Fig.
202
+ 4(a)) corresponds to the presence of E12g and A1g modes respectively in WS2 sample. The
203
+
204
+ 7
205
+
206
+ intensity ratio of E12g and A1g modes was estimated E12g/A1g i.e. = 1.26, which is higher
207
+ than the intensity ratio of bulk WS2 (E12g/A1g = 0.47) and lower than the monolayer WS2
208
+ (E12g/A1g = 2.2) [34,35]. This calculated intensity ratio (E12g/A1g = 1.26) is compatible
209
+ with the range of 2-3 layers of WS2.
210
+
211
+
212
+
213
+
214
+
215
+
216
+
217
+
218
+
219
+
220
+
221
+
222
+
223
+
224
+
225
+ Fig. 3: XRD patterns of (a) Pristine MgH2, (b) MoS2, (c) WS2, (d) MgH2-MoS2, and (e)
226
+ MgH2-WS2.
227
+
228
+
229
+ o-Parafilm,
230
+ *-MgH2,
231
+ t-Ws2, u -Mos2
232
+ (e)MgH2-Ws
233
+ 52
234
+ T
235
+ *
236
+ *
237
+ *
238
+ *
239
+ (d) mgh2-Mos2
240
+ *
241
+ Intensity (wt%)
242
+ (c) WS 2
243
+ (002)
244
+ -(004)
245
+ (100)
246
+ (101)
247
+ (900)
248
+ (105)
249
+ (110)
250
+ (112)
251
+ -(114)
252
+ (203)
253
+ (116)
254
+ 8
255
+ tt
256
+ T
257
+ T
258
+ .1
259
+ 1
260
+ T
261
+ (b)Mos.
262
+ (00L)2
263
+ 2
264
+ (002)
265
+ U
266
+ C(105)
267
+ (102)
268
+ (103)
269
+ C(110)
270
+ (112)
271
+ c(108)
272
+ (203)
273
+ U
274
+ (a) Pristine
275
+ MgH2
276
+ *
277
+ (200)
278
+ (110)
279
+ (220)
280
+ *(002)
281
+ (310)
282
+ (112)
283
+ (301)
284
+ (202)
285
+ (211)
286
+ *
287
+ 8
288
+ 8
289
+ *
290
+ *
291
+
292
+ 10
293
+ 20
294
+ 30
295
+ 40
296
+ 50
297
+ 60
298
+ 70
299
+ 80
300
+ 2e(degree8
301
+
302
+
303
+
304
+
305
+
306
+
307
+
308
+
309
+
310
+
311
+
312
+ Fig. 4: Raman spectra of (a) WS2 and (b) MoS2.
313
+
314
+
315
+
316
+ The information about stacking layers in 2D layered materials (like WS2 and
317
+ MoS2) can also be verified by AFM analysis. The surface topography and height profile
318
+ of prepared MoS2, and WS2 were examined along the blue dotted line as shown in Fig.
319
+ S1(a-b) (given in supporting information). The layered surface morphology along with
320
+ height profile (shown in Fig. S1(a-a1)) shown the average thickness of MoS2 is ~1.3 nm,
321
+ that indicates the presence of ~2 layers of stacking in the MoS2 sample [36,37]. The ~7-8
322
+ layers of stacking were present in the case of WS2 (shown in Fig. S1(b-b1)) with a
323
+ monolayer height of ~0.7 nm [38].
324
+
325
+
326
+
327
+
328
+ (b) Raman spectra of MoS2
329
+ (370)
330
+ A1g
331
+ (a) Raman spectra of WS2
332
+ (345)
333
+ 2g
334
+ Intensity (a.u.)
335
+ (384)
336
+ 2g
337
+ A1
338
+ (316)
339
+ 175200225250275300325
340
+ 350
341
+ 375400
342
+ 425
343
+ 450475500
344
+ Raman shift (cm9
345
+
346
+ .
347
+
348
+ 3.2 De/Re-hydrogenation kinetics of catalyzed MgH2
349
+ To identify the optimal percentage of catalyst in MgH2 with optimum temperature range
350
+ where material performed promptly, we have characterized as-prepared samples for the
351
+ temperature programmed desorption (TPD) analysis. The TPD curves of MgH2-MoS2
352
+ have seen in Fig. S2 (given in supporting information). The MgH2-5%MoS2, MgH2-
353
+ 10%MoS2, and MgH2-12%MoS2, starts releasing hydrogen at ~ 357 °C, ~ 330 °C, ~ 302
354
+ °C with ~ 6.41 wt%, ~ 6.00 wt%, ~ 4.88 wt% of hydrogen storage capacity respectively.
355
+ On the other hand, MgH2-5%WS2, MgH2-10%WS2, and MgH2-12%WS2, starts releasing
356
+ hydrogen at ~ 339 °C, ~ 277 °C, ~ 258 °C with ~ 6.54 wt%, ~ 5.95 wt%, ~ 5.14 wt% of
357
+ hydrogen storage capacity respectively (shown in Fig. S3 in supporting information).
358
+ Based on TPD analysis, the optimum catalyst concentration for catalyzing MgH2 is 10
359
+ wt% for all catalysts.
360
+ After getting information about the optimum catalyst for MgH2, we compared the TPD
361
+ analysis of all optimum catalyzed samples with pristine MgH2, as shown in Fig. 5. The
362
+ TPD of pristine MgH2 (shown in Fig. 5(a)) was then carried out to compare hydrogen
363
+ storage properties with catalyzed samples. The pristine MgH2 has an onset desorption
364
+ temperature of 376
365
+ oC with a total release of ~7.45 wt% storage capacity. The onset
366
+ desorption temperature of MgH2-MoS2 (MgH2-10%MoS2) is ~ 330
367
+ oC, and it desorbs ~
368
+ 6.00 wt% hydrogen while the desorption gets completed at 396
369
+ oC (Fig. 5(b)). In the case
370
+ of MgH2-WS2 (MgH2-10%WS2), it starts desorbing hydrogen at ~ 277
371
+ oC with a storage
372
+ capacity of 5.95 wt% (Fig. 5(c)).
373
+
374
+ 10
375
+
376
+
377
+
378
+
379
+
380
+
381
+
382
+
383
+
384
+
385
+
386
+
387
+
388
+ Fig. 5: Comparative TPD analysis of (a) Pristine MgH2, (b) MgH2-MoS2 and (c) MgH2-
389
+ WS2.
390
+
391
+ The desorbed samples then proceed for re/de-hydrogenation to check the cyclic stability
392
+ and reversibility of catalyzed and pristine MgH2. The re-hydrogenation kinetics was
393
+ carried out at 300
394
+ oC under 13 atm hydrogen pressures, as shown in Fig. 6. It can be seen,
395
+ the pristine MgH2 absorbed ~1.16 wt% hydrogen in 1.2 minutes whereas MgH2-MoS2,
396
+ MgH2-WS2 absorbed 4.60 wt%, 3.72 wt%, hydrogen, respectively, under similar
397
+ conditions of temperature and pressure.
398
+
399
+
400
+
401
+
402
+ 0
403
+ Hydrogen desorbed (wt%)
404
+ (a)
405
+ (b)
406
+ (c)
407
+ (a)PristineMgH
408
+ 5
409
+ (b) MgH,-Mos,
410
+ 6
411
+ (c) MgH,-WS,
412
+ 7
413
+ 8
414
+ 200
415
+ 225
416
+ 250
417
+ 275
418
+ 300
419
+ 325
420
+ 350
421
+ 375
422
+ 400
423
+ 425
424
+ Temperature (C)11
425
+
426
+
427
+ Fig. 6: Rehydrogenation kinetics curves at 300 °C under 13 atm H2 pressure of (a) b)
428
+ MgH2-WS2, (c) MgH2-MoS2 and (e) Pristine MgH2.
429
+
430
+ The rehydrogenated samples were then dehydrogenated at 300
431
+ oC under 1 atm
432
+ hydrogen pressure. It can be seen clearly in Fig. 7, that the MgH2-WS2 sample releases
433
+ 5.57 wt% hydrogen within 20 minutes while MgH2-MoS2 and pristine MgH2 releasees
434
+ 2.25 wt%, and 0.23 wt% of hydrogen under similar temperature and pressure conditions,
435
+ which is 3.32 wt%, and 4.48 wt% more than pristine MgH2, MgH2-MoS2, respectively.
436
+ Based on the above re/de-hydrogenation kinetics study, it is clearly shown that WS2
437
+ works as a superior catalyst to MoS2 for catalyzing MgH2. Therefore, in present study
438
+ WS2 is a prominent catalyst to catalyze MgH2.
439
+
440
+ (b)
441
+ 5.
442
+ Hydrogen absorbed (wt%)
443
+ (a)
444
+ (c)
445
+ - (a) MgH,-WS
446
+ (b) MgH,-Mos
447
+ (c) Pristine MgH,
448
+ 0
449
+ 0
450
+ 2
451
+ 4
452
+ 6
453
+ 8
454
+ 10
455
+ 12
456
+ 14
457
+ 16
458
+ 18
459
+ 20
460
+ 22
461
+ 24
462
+ Time (Min.)12
463
+
464
+
465
+
466
+
467
+
468
+
469
+
470
+
471
+
472
+
473
+
474
+
475
+
476
+
477
+
478
+ Fig. 7: Dehydrogenation kinetics curves at 300 °C under 1 atm H2 pressure of (a) MgH2-
479
+ WS2, (b) MgH2-MoS2, and (c) Pristine MgH2.
480
+
481
+ 3.3. Study of kinetics: Estimation of activation energy
482
+ The DSC was carried out to determine the hydrogen desorption activation energy barrier
483
+ to convert MgH2 into Mg. The DSC profile of MgH2-MoS2, MgH2-WS2, are shown in
484
+ Figs. 8-9. In the case of MgH2-WS2, the peak desorption temperature found from DSC is
485
+ ~ 380
486
+ oC, while the onset desorption temperature found from TPD is ~ 277
487
+ oC. There is a
488
+ difference in desorption temperature in TPD (Fig. 5(c)) and DSC (Fig. 9(a)) curves due to
489
+ the TPD being performed under vacuum with a temperature ramping rate of 5
490
+ oC/min
491
+ while DSC was performed under N2 atmosphere with a temperature ramping rate of 15
492
+
493
+ oC/min. For calculating the desorption activation energy, we have performed DSC with a
494
+ set of the various rate of heating (15, 18, 21, 24
495
+ oC/min) and plotted the Kissinger curve
496
+ by using the Kissinger equation[39] as given:
497
+
498
+ (a) Pristine MgH,
499
+ - (b) MgH,-MoS,
500
+ Hydrogen desorbed (wt%)
501
+ 5
502
+ (c) MgH,-WS
503
+ (a)
504
+ (b)
505
+ 2
506
+ (c)
507
+ 0
508
+ 0
509
+ 5
510
+ 10
511
+ 15
512
+ 20
513
+ 25
514
+ 30
515
+ 35
516
+ 40
517
+ 45
518
+ 50
519
+ 55
520
+ 60
521
+ Time (min.)13
522
+
523
+
524
+
525
+
526
+
527
+
528
+
529
+ (1)
530
+ Where β, Tp, and Ea are the heating rate, corresponding peak desorption temperature, and
531
+ activation energy, respectively. The slope of Kissinger plot (ln(β/Tp
532
+ 2) vs. 1000/Tp
533
+ 2 plot)
534
+ (Figs. 8-9) is used to calculate the desorption activation energy. The calculated activation
535
+ energy for MgH2-MoS2, and MgH2-WS2 is 117.09 kJ/mol (± 1.60 kJ/mol), and 104.00
536
+ kJ/mol (± 2.74 kJ/mol) respectively. This activation energy indicates that ~104 kJ/mol
537
+ energy is required to overcome the barrier to convert MgH2 into Mg in the presence of a
538
+ WS2 catalyst. These calculated activation energies are significantly lower than the
539
+ activation energy of pristine MgH2 [3,40].
540
+ Table 1: Table for plateau pressures at corresponding temperatures, change in enthalpy,
541
+ and activation energy of MgH2-MoS2, and MgH2-WS2.
542
+
543
+
544
+
545
+ S.No.
546
+ Sample
547
+ name
548
+ Plateaus
549
+ pressure
550
+ (atm)
551
+ Temperature
552
+ (
553
+ ºC)
554
+ Change
555
+ in
556
+ enthalpy
557
+ (kJ/mol)
558
+ Activation
559
+ energy
560
+ (kJ/mol)
561
+ 1.
562
+ MgH2-
563
+ MoS2
564
+ 1.03
565
+ 272.62
566
+
567
+ -78.33
568
+
569
+ 117.09
570
+ 2.03
571
+ 292.28
572
+ 3.77
573
+ 313.28
574
+ 2.
575
+ MgH2-
576
+ WS2
577
+ 1.52
578
+ 281.26
579
+
580
+ -77.44
581
+
582
+ 104.66
583
+ 2.92
584
+ 300.60
585
+ 3.45
586
+ 316.29
587
+
588
+ 14
589
+
590
+
591
+
592
+
593
+ Fig. 8: (i) DSC profile for desorption of MgH2-MoS2 with the heating rate (a) 15 ºC/min,
594
+ (b) 18 ºC/min, (c) 21 ºC/min, (d) 24 ºC/min, and (ii) corresponding Kissinger plot for
595
+ evaluating the desorption activation energy of MgH2-MoS2.
596
+
597
+
598
+ DSCprofilefor MgH2-MoS2
599
+ -9.8
600
+ Kissinger plot for MgHb-MoS2
601
+ Linear fit
602
+ (d) 24°C/min
603
+ -9.9
604
+ (a.u.)
605
+ -10.0
606
+ (c)21cC/min
607
+ Heatflow(
608
+ P
609
+ -10.1
610
+ Endo up
611
+ (b) 18 C/min
612
+ Eguation
613
+ y=a+b
614
+ -10.2
615
+ Adj. R-Squ
616
+ 0.99937
617
+ Value
618
+ Standard Er
619
+ -10.3
620
+ In(beta/Tp2) Irtercept
621
+ 11.212
622
+ 0.30766
623
+ (a) 15 C/min
624
+ In(beta/Tp2) Slope
625
+ -14.083
626
+ 0.20379
627
+ 1.4881.4941.5001.5061.5121.5181.5241.530
628
+ 250
629
+ 275
630
+ 300
631
+ 325
632
+ 350
633
+ 375
634
+ 400
635
+ 425
636
+ 450
637
+ 1000/T,(K1)
638
+ Temperature (cC)DSCprofileforMgH2-WS2
639
+ Kissinger plot for MgH2-WS2
640
+ -9.8
641
+ (d) 24 C/min
642
+ Linear fit
643
+ -9.9
644
+ Heat flow (a.u.)
645
+ (c) 21 °C/min
646
+ -10.0
647
+ [β/T,
648
+ (b) 18 °C/min
649
+ -10.1
650
+ Endo
651
+ Equation
652
+ =a+
653
+ Adj. R-Sq
654
+ 0.9979
655
+ -10.2
656
+ Value
657
+ Standard
658
+ (a) 15 °C/min
659
+ In(beta/Tp Interce
660
+ 9.3313
661
+ 0.50735
662
+ In(beta/Tp Slope
663
+ -12.58
664
+ 0.33004
665
+ -10.3
666
+ 1.520
667
+ 1.525
668
+ 1.530
669
+ 1.535
670
+ 1.540
671
+ 1.545
672
+ 1.550
673
+ 1.555
674
+ 1000/T(K
675
+ 320330340350360370380390400410420430440450
676
+ Temperature (C)15
677
+
678
+ Fig. 9: (i) DSC profile for desorption of MgH2-WS2 with the heating rate (a) 15 ºC/min,
679
+ (b) 18 ºC/min, (c) 21 ºC/min, (d) 24 ºC/min, and (ii) corresponding Kissinger plot for
680
+ evaluating the desorption activation energy of MgH2-WS2.
681
+
682
+ 3.4. Study of thermodynamics
683
+ After the kinetics and reversibility study, we have proceeded with the thermodynamic
684
+ analysis of catalyzed MgH2 for comparing the change in enthalpy and entropy of the
685
+ system using well known Van’t Hoff equation [41].
686
+
687
+
688
+
689
+ lnP = (ΔH/RT) - (ΔS/R)
690
+
691
+
692
+ (2)
693
+ Where P, ∆H, R, T, and ∆S are the pressure, change in enthalpy, gas constant, absolute
694
+ temperature, and change in entropy, respectively. The PCI isotherms (Figs. 10(i)-11(i)
695
+ and Van’t Hoff plots (Figs. 10(ii)-11(ii)) were used for the calculation of change in
696
+ enthalpy of MgH2-MoS2 and MgH2-WS2, respectively. The calculated change in
697
+ desorption enthalpy was found to be 78.33 kJ/mol (± 1.40 kJ/mol), and 77.44 kJ/mol (±
698
+ 1.13 kJ/mol), for MgH2-MoS2 and MgH2-WS2 respectively. It is clear from the above
699
+ estimation of change in enthalpy, that there is no significant enthalpy change in the
700
+ presence of a catalyst. Thus MoS2, WS2 have not positively impacted the thermodynamic
701
+ barrier of the MgH2. The plateau pressure at corresponding temperatures, change in
702
+ enthalpy, and activation energy has been tabulated in Table 1.
703
+
704
+ (i) PCl desorption for MgH2-MoS2
705
+ 1.4.
706
+ (ii) Vant's Hoff plot for MgH2-MoS2
707
+ 6.
708
+ - Linear fit
709
+ 1.2
710
+ 5.
711
+ 1.0
712
+ (atm)
713
+ 320°C
714
+ 4
715
+ 0.8.
716
+ Pressure (
717
+ P
718
+ 三 0.6
719
+ 3.
720
+ 300°C
721
+ 0.4
722
+ 2
723
+ Equation
724
+ y=a+b*
725
+ 0.2-
726
+ Adj. R-Squar0.99936
727
+ 280 °C
728
+ Value
729
+ Standard Err
730
+ InP
731
+ Intercept
732
+ 17.3878
733
+ 0.298
734
+ 0.0
735
+ InP
736
+ Slope
737
+ 9.4207
738
+ 0.16804
739
+ 0
740
+ +
741
+ 1.70 1.72 1.74 1.76 1.78 1.80 1.82 1.84 1.86 1.88
742
+ 0
743
+ 1
744
+ 2
745
+ 3
746
+ 4
747
+ 5
748
+ 1000/T (K
749
+ Hydropgen capacity (wt%)16
750
+
751
+ Fig. 10: (i) PCI desorption plots for MgH2-MoS2 at different temperatures and (ii)
752
+ corresponding Van't Hoff plot for calculating the change in enthalpy
753
+
754
+ Fig. 11: (i) PCI desorption plots for MgH2-WS2 at different temperatures and (ii)
755
+ corresponding Van't Hoff plot for calculating the change in enthalpy
756
+ 3.5 Cyclic stability of catalyzed MgH2
757
+ The WS2 (optimum catalyst) plays a significant role in improving the kinetics of MgH2.
758
+ The cyclic stability is an essential characteristic of the hydride material (MgH2) besides
759
+ kinetic and thermodynamics, making it a worthy hydrogen storage material. Therefore, it
760
+ is crucial to look at the cyclic stability of the catalyzed MgH2 samples. We have
761
+ performed 25 cycles of dehydrogenation (under 1 atm hydrogen pressure at 300 °C) and
762
+ re-hydrogenation (under 13 atm hydrogen pressure at 300 °C) to ensure cyclic stability of
763
+ catalyzed MgH2. The cyclic stability curve of MgH2-MoS2 and MgH2-WS2 are shown in
764
+ Fig. 12. From Fig. 12(a) MgH2-MoS2 shows the ~ 0.42 wt% (from 5.77 wt% to 5.35
765
+ wt%) degradation in hydrogen storage capacity during rehydrogenation and ~ 0.38 wt%
766
+ (from 5.69 wt% to 5.31 wt%) in dehydrogenation. The MgH2-WS2 has the loss of
767
+ hydrogen storage capacity ~ 0.3 wt% (from 5.80 wt% to 5.50 wt%) during re-
768
+ hydrogenation and ~ 0.36 wt% (from 5.76 wt% to 5.40 wt%) during dehydrogenation.
769
+ Thus, MgH2-WS2 has more substantial cyclic stability than MgH2-MoS2 under similar
770
+
771
+ 8
772
+ (i) PCI desorption for MgH2-WS2
773
+ 1.4
774
+ (ii) Vant's Hoff plot for MgH2-WS2
775
+ 1.2
776
+ Linear fit
777
+ 6
778
+ (atm)
779
+ 1.0
780
+ 5
781
+ Pressure
782
+ 320 °C
783
+ InP
784
+ 0.8
785
+ 300°℃
786
+ 3
787
+ 0.6
788
+ Equation
789
+ y=a+
790
+ Adj. R-Squ0.9995
791
+ 2
792
+ 280°C
793
+ Value
794
+ Standard E
795
+ 0.4 .
796
+ InP
797
+ Intercep
798
+ 17.038
799
+ 0.23617
800
+ Inp
801
+ Slope
802
+ -9.314
803
+ 0.1362
804
+ 0.2
805
+ 1.68
806
+ 1.70
807
+ 1.72
808
+ 1.74
809
+ 1.76
810
+ 1.78
811
+ 0
812
+ 1.80
813
+ 1000/T (K-1)
814
+ 0
815
+ 1
816
+ 2
817
+ 3
818
+ 4
819
+ 5
820
+ 6
821
+ Hydrogen capacity (wt%)17
822
+
823
+ temperature and pressure conditions. The comparative study for hydrogen storage
824
+ properties of different recently used 2D materials as the catalyst for MgH2 is explored in
825
+ Table 2.
826
+
827
+ Fig. 12: Cyclic stability of (a) MgH2-MoS2 and (b) MgH2-WS2.
828
+
829
+ Table 2: Table for different 2D materials as the catalyst for hydrogen storage application.
830
+ S.
831
+ No.
832
+ Material
833
+ 2D- based
834
+ catalyst
835
+ Hydrogen
836
+ storage
837
+ capacity
838
+ (wt%)
839
+ Onset
840
+ dehydrogen
841
+ ation
842
+ temperature
843
+ (ºC)
844
+
845
+ Activation
846
+ energy
847
+ (kJ/mol)
848
+ Change
849
+ in
850
+ enthalpy
851
+ (kJ/mol)
852
+
853
+ Ref.
854
+ 1.
855
+ Mg6C2N
856
+ C2N
857
+ 6.79
858
+ --
859
+ --
860
+ --
861
+ [20]
862
+ 2.
863
+ MgH2-LiAlH4-
864
+ Ti3C2
865
+ Ti3C2
866
+ 6.50
867
+ 63.0
868
+ 128.4
869
+ 74.3
870
+ [22]
871
+ 3.
872
+ MgH2-
873
+ Nb4C3Tx
874
+ Nb4C3Tx
875
+ 3.50
876
+ 150.6
877
+ 81.2
878
+ --
879
+ [21]
880
+ 4.
881
+ 1T’-MoS2
882
+
883
+ 3.90
884
+ --
885
+ --
886
+ --
887
+ [42]
888
+ 5.
889
+ MgH2-Gr
890
+ Graphene
891
+ 5.80
892
+ 300.0
893
+ --
894
+ --
895
+ [43]
896
+
897
+ Cyclic stability for MgH,-MoS
898
+ capacity (wt%)
899
+ 6
900
+ 00300
901
+ =0=0-0:
902
+ 5
903
+ -I- Rehydrogenation
904
+ - Dehydrogenation
905
+ 4
906
+ 3
907
+ Degradation during rehydrogenation=0.42 wt%
908
+ Degradationduring dehydrogenation=0.38 wt%
909
+ Hydrogen :
910
+ 2
911
+ 1
912
+ 0
913
+ 6
914
+ 8
915
+ 10
916
+ 16
917
+ 18
918
+ 222426
919
+ No.of cycleCyclic stability for MgH,-WS
920
+ 6
921
+ 5.
922
+ Rehydrogenation
923
+ 4.
924
+ +- Dehydrogenation
925
+ 3
926
+ Degradation during rehydrogenation=0.30 wt%
927
+ Degradation during dehydrogenation=0.36 wt%
928
+ Hydrogen :
929
+ 2
930
+ 0
931
+ 10
932
+ 12
933
+ 16
934
+ 18.20
935
+ 222426
936
+ No. of cycle18
937
+
938
+ 6.
939
+ MgH2-
940
+ TiH2@Gr
941
+ Graphene
942
+ 6.77
943
+ 204.0
944
+ 88.89
945
+ 74.54
946
+ [3]
947
+ MgH2-
948
+ TiO2@Gr
949
+ Graphene
950
+ 5.98
951
+ 240.0
952
+ 98.00
953
+ 76.87
954
+ MgH2-Ti@Gr
955
+ Graphene
956
+ 5.70
957
+ 235.0
958
+ 103.03
959
+ 75.65
960
+ 7.
961
+ MgH2-Gr
962
+ Graphene
963
+ 6.14
964
+ 300.0
965
+ 134.95
966
+ 77.90
967
+ [13]
968
+ 8.
969
+ MgH2-VS2
970
+ VS2
971
+ 6.51
972
+ 242.0
973
+ 98.10
974
+ 76.83
975
+ 9.
976
+ MgH2-WS2
977
+ WS2
978
+ 5.95
979
+ 277.0
980
+ 104.66
981
+ 77.44
982
+ Pres
983
+ ent
984
+ stud
985
+ y
986
+ 10.
987
+ MgH2-MoS2
988
+ MoS2
989
+ 6.00
990
+ 330.0
991
+ 117.09
992
+ 78.33
993
+ Pres
994
+ ent
995
+ stud
996
+ y
997
+
998
+ 4. Conclusions
999
+
1000
+ The catalytic effect of MoS2, and WS2 on MgH2 was evaluated and compared.
1001
+ Based on the de/re-hydrogenation study, it is found that WS2 works as an optimum
1002
+ catalyst over MoS2 for MgH2. The MgH2-WS2 has an onset de-hydrogenation ~277 oC
1003
+ with a hydrogen storage capacity of 5.95 wt%. The MgH2-WS2 absorbed hydrogen ~ 3.72
1004
+ wt% within 1.3 minutes at 300 oC under 13 atm hydrogen pressure and it desorbed ~5.57
1005
+ wt% within 20 minutes at 300 oC under 1 atm hydrogen pressure. The MgH2-WS2 shows
1006
+ a minimum degradation of hydrogen storage capacity ~ 0.3 wt% upto 25 cycles which
1007
+ shows a better cyclic stability than cyclic stability of MgH2-MoS2 (~ 0.4 wt% loss in
1008
+ hydrogen storage capacity). MgH2-WS2 also shows the lower reaction activation energy
1009
+ ~117 kJ/mol as compare to other catalyzed and uncatalyzed samples. On the other hand,
1010
+ these catalysts (WS2 and MoS2) do not have any impact on the thermodynamical
1011
+ parameters that is change in enthalpy. This study opens a new era to further applications
1012
+ of 2D layered materials for various applications like template materials.
1013
+
1014
+ 19
1015
+
1016
+ Acknowledgments
1017
+ We gratefully accept funding assistance from the Department of Science and Technology
1018
+ (DST), New Delhi, India. The Council of Scientific and Industrial Research (CSIR), New
1019
+ Delhi, India, has awarded the author (S.K.V.) a CSIR-Senior Research Fellowship
1020
+ (Award No. 09/013(0872)/2019-EMR-I), for which the author is grateful.
1021
+ Conflict of Interest Declaration
1022
+ There are no conflicts of interest among the authors.
1023
+
1024
+
1025
+ References
1026
+ [1]
1027
+ TP Yadav, A Kumar, SK Verma, NK Mukhopadhyay, High-Entropy Alloys for
1028
+ Solid Hydrogen Storage: Potentials and Prospects, Transactions of the Indian
1029
+ National Academy of Engineering, 7 (2022) 147–156.
1030
+ https://doi.org/10.1007/s41403-021-00316-w.
1031
+ [2]
1032
+ A. Kumar, T.P. Yadav, N. K. Mukhopadhyay, Notable hydrogen storage in Ti–Zr–
1033
+ V–Cr–Ni high entropy alloy, International Journal of Hydrogen Energy 47 (2022)
1034
+ 22893-22900. https://doi.org/10.1016/j.ijhydene.2022.05.107
1035
+ [3]
1036
+ Verma SK, Bhatnagar A, Shukla V, Soni PK, Pandey AP, Yadav TP, et al.
1037
+ Multiple improvements of hydrogen sorption and their mechanism for MgH2
1038
+ catalyzed through TiH2@Gr. Int J Hydrogen Energy 2020;45:19516–30.
1039
+ https://doi.org/10.1016/j.ijhydene.2020.05.031.
1040
+ [4]
1041
+ SK Pandey, SK Verma, A Bhatnagar, TP Yadav, Catalytic characteristics of
1042
+ titanium‐(IV)‐isopropoxide (TTIP) on de/re‐hydrogenation of wet ball‐milled
1043
+ MgH2/Mg, International Journal of Energy Research 46 (12) (2022) 17602-17615.
1044
+ https://doi.org/10.1002/er.8427.
1045
+ [5]
1046
+ Shukla V, Bhatnagar A, Verma SK, Pandey AP, Vishwakarma AK, Srivastava P,
1047
+ et al. Simultaneous improvement of kinetics and thermodynamics based on SrF2
1048
+ and SrF2@Gr additives on hydrogen sorption in MgH2. Mater Adv 2021;2:4277–
1049
+ 90. https://doi.org/10.1039/D1MA00012H.
1050
+ [6]
1051
+ Srivastava ON, Yadav TP, Shahi RR, Pandey SK, Shaz MA, Bhatnagar A.
1052
+
1053
+ 20
1054
+
1055
+ Hydrogen energy in India: Storage to application. Proc Indian Natl Sci Acad
1056
+ 2015;81:915–37. https://doi.org/10.16943/ptinsa/2015/v81i4/48303.
1057
+ [7]
1058
+ US DoE. Target Explanation Document: Onboard Hydrogen Storage for Light-
1059
+ Duty Fuel Cell Vehicles. US Drive 2017:1–29.
1060
+ [8]
1061
+ Bhatnagar A, Pandey SK, Vishwakarma AK, Singh S, Shukla V, Soni PK, et al.
1062
+ Fe3O4 @graphene as a superior catalyst for hydrogen de/absorption from/in MgH 2
1063
+ /Mg. J Mater Chem A 2016;4:14761–72. https://doi.org/10.1039/c6ta05998h.
1064
+ [9]
1065
+ Sadhasivam T, Kim H-T, Jung S, Roh S-H, Park J-H, Jung H-Y. Dimensional
1066
+ effects of nanostructured Mg/MgH2 for hydrogen storage applications: A review.
1067
+ Renew Sustain Energy Rev 2017;72:523–34.
1068
+ https://doi.org/https://doi.org/10.1016/j.rser.2017.01.107.
1069
+ [10] Yartys VA, Lototskyy M V, Akiba E, Albert R, Antonov VE, Ares JR, et al.
1070
+ Magnesium based materials for hydrogen based energy storage: Past, present and
1071
+ future. Int J Hydrogen Energy 2019;44:7809–59.
1072
+ https://doi.org/https://doi.org/10.1016/j.ijhydene.2018.12.212.
1073
+ [11] SK Pandey, A Bhatnagar, SS Mishra, TP Yadav, MA Shaz, ON Srivastava,
1074
+ Curious Catalytic Characteristics of Al–Cu–Fe Quasicrystal for
1075
+ e/Rehydrogenation of MgH2, The Journal of Physical Chemistry C 121 (45) (2017)
1076
+ 24936-24944. https://doi.org/10.1021/acs.jpcc.7b07336
1077
+ [12] V Shukla, TP Yadav, Notable catalytic activity of CuO nanoparticles derived from
1078
+ metal‐organic frameworks for improving the hydrogen sorption properties of
1079
+ MgH2, International Journal of Energy Research, 46 (9) (2022) 12804-12819.
1080
+ https://doi.org/10.1002/er.8050.
1081
+ [13] Verma SK, Shaz MA, Yadav TP. Enhanced hydrogen absorption and desorption
1082
+ properties of MgH2 with graphene and vanadium disulfide. Int J Hydrogen Energy
1083
+ 2022. https://doi.org/https://doi.org/10.1016/j.ijhydene.2021.12.269.
1084
+ [14] Patelli N, Calizzi M, Migliori A, Morandi V, Pasquini L. Hydrogen Desorption
1085
+ Below 150 °C in MgH2–TiH2 Composite Nanoparticles: Equilibrium and Kinetic
1086
+ Properties. J Phys Chem C 2017;121:11166–77.
1087
+ https://doi.org/10.1021/acs.jpcc.7b03169.
1088
+ [15] Korablov D, Besenbacher F, Jensen TR. Kinetics and thermodynamics of
1089
+
1090
+ 21
1091
+
1092
+ hydrogenation-dehydrogenation for Mg-25%TM (TM = Ti, Nb or V) composites
1093
+ synthesized by reactive ball milling in hydrogen. Int J Hydrogen Energy
1094
+ 2018;43:16804–14. https://doi.org/10.1016/j.ijhydene.2018.05.091.
1095
+ [16] Bhatnagar A, Johnson JK, Shaz MA, Srivastava ON. TiH 2 as a Dynamic Additive
1096
+ for Improving the De/Rehydrogenation Properties of MgH 2 : A Combined
1097
+ Experimental and Theoretical Mechanistic Investigation. J Phys Chem C
1098
+ 2018;122:21248–61. https://doi.org/10.1021/acs.jpcc.8b07640.
1099
+ [17] Zhang J, Xia G, Guo Z, Zhou D. Synergetic Effects toward Catalysis and
1100
+ Confinement of Magnesium Hydride on Modified Graphene: A First-Principles
1101
+ Study. J Phys Chem C 2017;121:18401–11.
1102
+ https://doi.org/10.1021/acs.jpcc.7b05848.
1103
+ [18] Singh MK, Bhatnagar A, Pandey SK, Mishra PC, Srivastava ON. Experimental
1104
+ and first principle studies on hydrogen desorption behavior of graphene nanofibre
1105
+ catalyzed MgH2. Int J Hydrogen Energy 2017;42:960–8.
1106
+ https://doi.org/10.1016/J.IJHYDENE.2016.09.210.
1107
+ [19] Liu G, Wang Y, Jiao L, Yuan H. Understanding the role of few-layer graphene
1108
+ nanosheets in enhancing the hydrogen sorption kinetics of magnesium hydride.
1109
+ ACS Appl Mater Interfaces 2014;6:11038–46. https://doi.org/10.1021/am502755s.
1110
+ [20] Varunaa R, Ravindran P. Potential hydrogen storage materials from metal
1111
+ decorated 2D-C2N: An: ab initio study. Phys Chem Chem Phys 2019;21:25311–
1112
+ 22. https://doi.org/10.1039/c9cp05105h.
1113
+ [21] Liu Y, Gao H, Zhu Y, Li S, Zhang J, Li L. Excellent catalytic activity of a two-
1114
+ dimensional Nb4C3Tx (MXene) on hydrogen storage of MgH2. Appl Surf Sci
1115
+ 2019;493:431–40. https://doi.org/https://doi.org/10.1016/j.apsusc.2019.07.037.
1116
+ [22] Chen G, Zhang Y, Cheng H, Zhu Y, Li L, Lin H. Effects of two-dimension MXene
1117
+ Ti3C2 on hydrogen storage performances of MgH2-LiAlH4 composite. Chem
1118
+ Phys 2019;522:178–87.
1119
+ https://doi.org/https://doi.org/10.1016/j.chemphys.2019.03.001.
1120
+ [23] Raccichini R, Varzi A, Passerini S, Scrosati B. The role of graphene for
1121
+ electrochemical energy storage. Nat Mater 2015;14:271–9.
1122
+ https://doi.org/10.1038/nmat4170.
1123
+
1124
+ 22
1125
+
1126
+ [24] Pumera M. Graphene-based nanomaterials for energy storage. Energy Environ Sci
1127
+ 2011;4:668–74. https://doi.org/10.1039/C0EE00295J.
1128
+ [25] Singh S, Bhatnagar A, Shukla V, Vishwakarma AK, Soni PK, Verma SK, et al.
1129
+ Ternary transition metal alloy FeCoNi nanoparticles on graphene as new catalyst
1130
+ for hydrogen sorption in MgH2. Int J Hydrogen Energy 2020;45:774–86.
1131
+ https://doi.org/10.1016/j.ijhydene.2019.10.204.
1132
+ [26] Liu G, Wang Y, Xu C, Qiu F, An C, Li L, et al. Excellent catalytic effects of
1133
+ highly crumpled graphene nanosheets on hydrogenation/dehydrogenation of
1134
+ magnesium hydride. Nanoscale 2013;5:1074–81.
1135
+ https://doi.org/10.1039/c2nr33347c.
1136
+ [27] Huang Y, Xia G, Chen J, Zhang B, Li Q, Yu X. One-step uniform growth of
1137
+ magnesium hydride nanoparticles on graphene. Prog Nat Sci Mater Int
1138
+ 2017;27:81–7. https://doi.org/https://doi.org/10.1016/j.pnsc.2016.12.015.
1139
+ [28] Liu R, Zhao Y, Chu T. Theoretical exploration of MgH2 and graphene nano-flakes
1140
+ in cyclohexane: proposing a new perspective toward functional hydrogen storage
1141
+ material. Chem Commun 2015;51:2429–32.
1142
+ https://doi.org/10.1039/C4CC09424G.
1143
+ [29] Li W, Huang J, Feng L, Cao L, Feng Y, Wang H, et al. Facile: In situ synthesis of
1144
+ crystalline VOOH-coated VS2 microflowers with superior sodium storage
1145
+ performance. J Mater Chem A 2017;5:20217–27.
1146
+ https://doi.org/10.1039/c7ta05205g.
1147
+ [30] Fang W, Zhao H, Xie Y, Fang J, Xu J, Chen Z. Facile Hydrothermal Synthesis of
1148
+ VS2/Graphene Nanocomposites with Superior High-Rate Capability as Lithium-
1149
+ Ion Battery Cathodes. ACS Appl Mater Interfaces 2015;7:13044–52.
1150
+ https://doi.org/10.1021/acsami.5b03124.
1151
+ [31] Ma H, Shen Z, Ben S. Understanding the exfoliation and dispersion of MoS2
1152
+ nanosheets in pure water. J Colloid Interface Sci 2018;517:204–12.
1153
+ https://doi.org/10.1016/j.jcis.2017.11.013.
1154
+ [32] Lu X, Utama MIB, Zhang J, Zhao Y, Xiong Q. Layer-by-layer thinning of MoS2
1155
+ by thermal annealing. Nanoscale 2013;5:8904–8.
1156
+ https://doi.org/10.1039/C3NR03101B.
1157
+
1158
+ 23
1159
+
1160
+ [33] Chakraborty B, Matte HSSR, Sood AK, Rao CNR. Layer-dependent resonant
1161
+ Raman scattering of a few layer MoS2. J Raman Spectrosc 2013;44:92–6.
1162
+ https://doi.org/https://doi.org/10.1002/jrs.4147.
1163
+ [34] Hazarika SJ, Mohanta D. Inorganic fullerene-type WS2 nanoparticles: processing,
1164
+ characterization and its photocatalytic performance on malachite green. Appl Phys
1165
+ A 2017;123:381. https://doi.org/10.1007/s00339-017-0965-7.
1166
+ [35] Berkdemir A, Gutiérrez HR, Botello-Méndez AR, Perea-López N, Elías AL, Chia
1167
+ C-I, et al. Identification of individual and few layers of WS2 using Raman
1168
+ Spectroscopy. Sci Rep 2013;3:1755. https://doi.org/10.1038/srep01755.
1169
+ [36] Frindt RF. Single Crystals of MoS2 Several Molecular Layers Thick. J Appl Phys
1170
+ 1966;37:1928–9. https://doi.org/10.1063/1.1708627.
1171
+ [37] Radisavljevic B, Radenovic A, Brivio J, Giacometti V, Kis A. Single-layer MoS2
1172
+ transistors. Nat Nanotechnol 2011;6:147–50.
1173
+ https://doi.org/10.1038/nnano.2010.279.
1174
+ [38] Ovchinnikov D, Allain A, Huang Y-S, Dumcenco D, Kis A. Electrical Transport
1175
+ Properties of Single-Layer WS2. ACS Nano 2014;8:8174–81.
1176
+ https://doi.org/10.1021/nn502362b.
1177
+ [39] Kissinger HE. Reaction Kinetics in Differential Thermal Analysis. Anal Chem
1178
+ 1957;29:1702–6. https://doi.org/10.1021/ac60131a045.
1179
+ [40] Wu Z, Fang J, Liu N, Wu J, Kong L. The Improvement in Hydrogen Storage
1180
+ Performance of MgH2 Enabled by Multilayer Ti3C2. Micromachines 2021;12.
1181
+ https://doi.org/10.3390/mi12101190.
1182
+ [41] Lima EC, Hosseini-Bandegharaei A, Moreno-Piraján JC, Anastopoulos I. A
1183
+ critical review of the estimation of the thermodynamic parameters on adsorption
1184
+ equilibria. Wrong use of equilibrium constant in the Van’t Hoof equation for
1185
+ calculation of thermodynamic parameters of adsorption. J Mol Liq 2019;273:425–
1186
+ 34. https://doi.org/https://doi.org/10.1016/j.molliq.2018.10.048.
1187
+ [42] Chen J, Cao J, Zhou J, Zhang Y, Li M, Wang W, et al. Mechanism of highly
1188
+ enhanced hydrogen storage by two-dimensional 1T′ MoS2. Phys Chem Chem
1189
+ Phys 2020;22:430–6. https://doi.org/10.1039/c9cp04402g.
1190
+ [43] Singh MK, Bhatnagar A, Pandey SK, Mishra PC, Srivastava ON. Experimental
1191
+
1192
+ 24
1193
+
1194
+ and first principle studies on hydrogen desorption behavior of graphene nanofibre
1195
+ catalyzed MgH 2. Int J Hydrogen Energy 2017;42:960–8.
1196
+ https://doi.org/10.1016/j.ijhydene.2016.09.210.
1197
+
GNE1T4oBgHgl3EQfFANy/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
GtE5T4oBgHgl3EQfWA9Z/content/tmp_files/2301.05555v1.pdf.txt ADDED
@@ -0,0 +1,1575 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Magnetic phase diagram of the breathing-kagome antiferromagnet Nd3BWO9
2
+ D. Flavi´an,1, ∗ J. Nagl,1 S. Hayashida,1, 2 M. Yan,1 O. Zaharko,3
3
+ T. Fennell,3 D. Khalyavin,4 Z. Yan,1 S. Gvasaliya,1 and A. Zheludev1, †
4
+ 1Laboratory for Solid State Physics, ETH Z¨urich, 8093 Z¨urich, Switzerland
5
+ 2Max-Planck-Institut f¨ur Festk¨orperforschung, Heisenbergstraße 1, 70569 Stuttgart, Germany
6
+ 3Laboratory for Neutron Scattering and Imaging,
7
+ Paul Scherrer Institut, 5232 Villigen, Switzerland
8
+ 4ISIS Facility, Rutherford Appleton Laboratory, Chilton, Didcot, Oxon OX11 0QX, United Kingdom
9
+ (Dated: January 16, 2023)
10
+ The highly-frustrated rare-earth based magnet Nd3BWO9 is a promising candidate in the search
11
+ for proximate spin liquid physics. We present a thorough investigation on single crystals of this ma-
12
+ terial using bulk and microscopic techniques. Magnetization data reveal a fractional magnetization
13
+ plateau for three different investigated field directions. The magnetic phase diagram is mapped out
14
+ from calorimetric data and exhibits several domes of magnetic order below 0.3 K. Propagation vec-
15
+ tors for all ordered phases are presented. The results suggest complex ordering in this material, and
16
+ unveil the existence of a commensuration transition of the propagation vector at zero magnetic field.
17
+ A scenario where interplane exchange interactions are essential to a magnetic model of Nd3BWO9
18
+ is discussed.
19
+ I.
20
+ INTRODUCTION
21
+ Strongly frustrated quantum antiferromagnets (AFM)
22
+ are known to realize a panoply of magnetic states due
23
+ to the delicate equilibrium between the magnetic in-
24
+ teractions.
25
+ In the presence of magnetic fields, the
26
+ large ground state degeneracy is lifted in subtle and
27
+ diverse ways, which leads to extremely rich phase di-
28
+ agrams. Realization of spin-density waves [1], magne-
29
+ tization plateaus [2, 3] commensurate-incommensurate
30
+ transitions [4], and even more exotic order like spin
31
+ nematicity [5, 6] is not rare, particularly in quasi-low-
32
+ dimensional systems.
33
+ The archetypal model in 2D frustrated magnetism is
34
+ the kagome lattice Heisenberg S = 1/2 AFM (KHAF).
35
+ The impossibility of satisfying all magnetic interactions
36
+ in this lattice results in a macroscopic degeneracy of the
37
+ ground state already at a classical level [7]. Turning to
38
+ S = 1/2 spins promotes the quantum fluctuations on the
39
+ ground state giving rise to highly non-trivial phases [8].
40
+ Arguably, the most intriguing state is the hypothesized
41
+ Quantum Spin Liquid (QSL) [9] ground state. The pre-
42
+ diction of fractionalization of quasiparticles in a 2D sys-
43
+ tem triggered extensive effort from both theory and ex-
44
+ perimental perspectives [10, 11]. Nevertheless, the QSL
45
+ phase remains elusive [12] as it constitutes a very frag-
46
+ ile state. One of the main causes of the instability of
47
+ the QSL states is the presence of terms in the Hamilto-
48
+ nian that lift the ground state degeneracy [13, 14]. The
49
+ many different ways to lift this degeneracy have led to
50
+ a flurry of new magnetic structures [15–18]. However,
51
+ occasionally deviations from a putative KHAF tend to
52
+ stabilize QSL phases. In particular, the so called breath-
53
+ ing anisotropy has been predicted to favor a resonance
54
+ ∗ daniefla@ethz.ch
55
+ † zhelud@ethz.ch; http://www.neutron.ethz.ch/
56
+ valence bond solid ground state for a wide range of cou-
57
+ pling parameters [19, 20].
58
+ In
59
+ this
60
+ context,
61
+ the
62
+ recently
63
+ discovered
64
+ family
65
+ R3BWO9 of rare-earth antiferromagnets is an optimal
66
+ platform for the search of spin-liquid candidates [21].
67
+ Here R is a trivalent rare-earth element and the large
68
+ difference in size of the constituent atoms prevents anti-
69
+ site chemical disorder. All of the members of the family
70
+ realize a breathing kagome lattice in their basal plane
71
+ and show no sign of magnetic ordering down to 2 K.
72
+ The strong spin-orbit coupling in combination with crys-
73
+ tal electric field effects opens the possibility of realizing
74
+ effective Jeff = 1/2 magnetic moments.
75
+ Among all compounds in the family, the most promis-
76
+ ing is Nd3BWO9. A large Weiss temperature [21] has
77
+ been reported and the total angular momentum of Nd3+
78
+ (J = 9/2) makes it a Kramers-doublet system. No mag-
79
+ netic long-range order has been found in previous studies
80
+ down to 1.8 K. However, little is known so far about its
81
+ magnetism. In this study we report on the low tempera-
82
+ ture properties of single crystals of Nd3BWO9. We found
83
+ static magnetic long-range order below 0.3 K. The ob-
84
+ served magnetism suggests a three dimensional network
85
+ of exchange interactions. Nonetheless, due to the highly
86
+ frustrated interaction a complex phase diagram is real-
87
+ ized.
88
+ The paper is structured as follows. First, a summary
89
+ of the various methods used is provided. Then, we out-
90
+ line the main results of the experiments. Subsequently,
91
+ a detailed discussion of the main outcome is provided,
92
+ including a thorough description of the magnetic struc-
93
+ ture and a detailed picture of the magnetic phase dia-
94
+ gram under applied fields. Finally, the main conclusions
95
+ are drawn and further steps in the search of QSL physics
96
+ are examined.
97
+ arXiv:2301.05555v1 [cond-mat.str-el] 13 Jan 2023
98
+
99
+ (b)
100
+ (e)
101
+ 2 mm
102
+ a
103
+ b
104
+ c
105
+ 2.66 Å
106
+ 2.57 Å
107
+ 2.25 Å
108
+ 2.49 Å
109
+ 2.38 Å
110
+ 2.36 Å
111
+ Nd
112
+ 2.55 Å
113
+ 2.42 Å
114
+ c
115
+ WO6
116
+ B
117
+ NdO8
118
+ a
119
+ b*a*
120
+ b
121
+ c
122
+ (a)
123
+ (d)
124
+ Nd
125
+ a
126
+ b
127
+ c
128
+ l = 16.44 Å
129
+ (c)
130
+ 3.95 Å
131
+ 4.92 Å
132
+ 4.25 Å
133
+ FIG. 1.
134
+ Crystal structure and superexchange topology in
135
+ Nd3BWO9. (a) Schematic structure reflecting the purported
136
+ kagome interaction in the crystallographic ab plane.
137
+ Only
138
+ atoms with 0 ≤ z ≤ 0.5 are shown here. There is an addi-
139
+ tional kagome plane displaced by half lattice parameter along
140
+ the c crystallographic direction. (b) The shortest superex-
141
+ change Nd-O-Nd bond links neodymium atoms in different
142
+ kagome planes, forming isolated spin tubes along the c axis
143
+ arranged in a triangular lattice. The kagome bonds are shown
144
+ for reference along with bond distances. (c) A typical single
145
+ crystal sample of Nd3BWO9. (d) A single spin tube is unfrus-
146
+ trated. However, further-neighbor interactions frustrate the
147
+ system. An arrow indicates the size of the magnetic supercell
148
+ at zero field. (e) The environment of neodymium has very
149
+ low symmetry, resulting in a C1 point group for the magnetic
150
+ ion. Nd-O distances are indicated.
151
+ II.
152
+ METHODS
153
+ Nd3BWO9 crystallizes in a hexagonal structure, with
154
+ space group P63 (No. 173), where the magnetism stems
155
+ from the effective magnetic moment of the Nd3+ ions.
156
+ Single crystal samples were grown by spontaneous crys-
157
+ tallization using a flux method as described in [22]. Pur-
158
+ ple transparent single crystals with well defined facets
159
+ were obtained [Fig. 1(c)].
160
+ Typical masses range from
161
+ a few micrograms to 40 mg and different samples were
162
+ used in this study, depending on the technique.
163
+ The
164
+ chemical structure of the different single-crystal samples
165
+ used in this study was validated using single-crystal X-
166
+ ray diffraction on a Bruker APEX-II instrument, and was
167
+ found to be in agreement with previous reports [21]. The
168
+ structure is schematically depicted in Fig. 1, where the
169
+ kagome-lattice bonds can be readily identified. Powder
170
+ samples of Nd3BWO9, as well as of the non-magnetic
171
+ La3BWO9, were synthesized by a solid state reaction.
172
+ The correct chemical structure and the quality of the
173
+ powders was checked with powder X-ray diffraction in
174
+ a Rigaku MiniFlex diffractometer.
175
+ Boron-11 enriched
176
+ samples (both powder and single crystals) were also pre-
177
+ pared for their use in neutron scattering experiments.
178
+ Measurements of heat capacity, magnetocaloric effect
179
+ (MCE), magnetization and magnetic torque were carried
180
+ out using a 3He-4He dilution refrigerator insert for the
181
+ Quantum Design Physical Property Measurement Sys-
182
+ tem (PPMS). A sample of mass 0.131 mg was used for
183
+ both heat capacity and MCE measurements. Heat ca-
184
+ pacity data were collected using a standard relaxation
185
+ method from Quantum Design for temperatures 100 mK
186
+ < T < 4 K in applied fields of 0 T < µ0H < 3 T. The
187
+ magnetic field was applied along the crystallographic a∗,
188
+ and c directions. In zero field, data were collected from
189
+ 100 mK to 300 K. Heat capacity data of La3BWO9 were
190
+ measured down to 2 K and extrapolated to lower tem-
191
+ peratures from an empirical fit to a T 3-power law. MCE
192
+ data were measured using the same puck as for heat ca-
193
+ pacity. The change of temperature of the sample was
194
+ recorded as the magnetic field was swept up and down
195
+ at a constant rate. In order to avoid self heating of the
196
+ puck, the field change rate was optimized and a value of
197
+ 0.5 mT/s was selected. In the terminology of MCE mea-
198
+ surements, our experiment was conducted under equilib-
199
+ rium conditions.
200
+ Magnetization was measured using an in house made
201
+ Faraday-balance capacitive magnetometer [23] at 120
202
+ mK and 2 K and magnetic fields applied along three ori-
203
+ entations: a∗, and b, and c. Additional measurements
204
+ of magnetization carried out in the MPMS system at 2
205
+ K were used to calibrate the low temperature data and
206
+ obtain absolute units (not shown here). Using the same
207
+ setup, magnetic torque was measured up to 3 T and
208
+ for temperature from 120 mK to 600 mK. The torque
209
+ data correspond to the deflection of a small cantilever
210
+ on which the sample is mounted.
211
+ The magnetic field
212
+ sweeping rate was also optimized to minimize heating
213
+ due to eddy currents.
214
+ Magnetic susceptibility was measured using the Quan-
215
+ tum Design Magnetic Property Measurement System
216
+ (MPMS) SQUID Magnetometer.
217
+ The temperature
218
+ range from 1.8 K to 300 K was probed using a small po-
219
+ larizing field applied along three crystal directions: a∗,
220
+ and b, and c. The probing field was µ0H = 0.1 T, where
221
+ µ0 denotes the permeability of vacuum.
222
+ Inelastic neutron scattering on powder samples of
223
+ Nd3BWO9 was measured to investigate the crystal elec-
224
+ tric field induced scheme of total angular momentum
225
+ states. The instrument of choice was the thermal neu-
226
+ tron triple-axis-spectrometer EIGER at PSI. 11.1 g of
227
+ Nd3 11BWO9 was sealed in an aluminum can and in-
228
+ stalled in a standard 4He orange cryostat. A final wave-
229
+ length of kf= 2.66 ˚A−1 (λ = 2.36 ˚A) was chosen, us-
230
+ ing a pyrolytic graphite filter to eliminate higher-order
231
+ neutrons without further collimation. Data were mea-
232
+ sured at constant scattering angle, 2θ. The background
233
+ was investigated to select the optimal value for the scat-
234
+ tering angle, sufficiently far from the direct beam and
235
+ low enough to have good counting and small decay in
236
+ the signals due to magnetic structure factors. A value
237
+ of 2θ = 10◦ was chosen, and the incident energy was
238
+ scanned at three different temperatures: 1.5 K, 100 K
239
+ and 300 K.
240
+ Neutron single crystal diffraction was used to investi-
241
+ 2
242
+
243
+ J0
244
+ 100
245
+ 200
246
+ 300
247
+ T (K)
248
+ 0
249
+ 50
250
+ 100
251
+ 150
252
+ 200
253
+ -1 (mol T μB )
254
+ H || a*
255
+ θCW = -3.76 K
256
+ μ0H = 0.1 T
257
+ H || b
258
+ H || c
259
+ -1
260
+ FIG. 2.
261
+ Inverse magnetic susceptibility on single crystals.
262
+ Data show measurements for three field orientations. A small
263
+ probing field of 0.1 T was used for all measurements. The
264
+ black solid line represents a Curie-Weiss model with the av-
265
+ erage Weiss temperature and effective moment parameters,
266
+ given in Table. I.
267
+ gate the magnetic structures in the ordered phases. A
268
+ single crystal sample of 18 mg in mass of Nd3 11BWO9
269
+ and 5.5×1.4×0.8 mm3 was studied using two different
270
+ instruments. Measurements with H ∥ a∗ were carried
271
+ out at the Thermal Single Crystal Diffractometer ZE-
272
+ BRA at the Swiss Spallation Neutron Source, SINQ,
273
+ in the Paul Scherrer Institut (PSI, Switzerland).
274
+ The
275
+ diffractometer was used in conjunction with a 3He-4He
276
+ dilution refrigerator and a 6-T magnet.
277
+ The crystal
278
+ was aligned with its a∗ axis vertical, the same direc-
279
+ tion as the applied magnetic field. Neutron wavelengths
280
+ of λ = 2.314 ˚A and 1.383 ˚A were selected, provided by
281
+ the PG(200) and Ge(220) monochromators. Additional
282
+ measurements with H ∥ c were carried out in the time-
283
+ of-flight diffractometer WISH at the ISIS facility in the
284
+ Rutherford Appleton Laboratory, in the United King-
285
+ dom. The sample was mounted with its c axis vertical
286
+ and parallel to the magnetic field. A 3He-4He dilution
287
+ refrigerator and a 10-T magnet were used to access the
288
+ ordered states in Nd3BWO9.
289
+ III.
290
+ EXPERIMENTAL RESULTS
291
+ A.
292
+ Magnetic susceptibility
293
+ Figure 2 shows inverse susceptibility measurements for
294
+ probing fields applied along the crystallographic direc-
295
+ tions a∗b, and c.
296
+ Down to the lowest accessible tem-
297
+ perature of 1.8 K, these data show no sign of magnetic
298
+ ordering.
299
+ A fit of the experimental data to a Curie-Weiss model
300
+ is shown overlaid on the experimental results. A good
301
+ TABLE I. Fitting parameters from the Curie-Weiss model for
302
+ data shown in Fig.
303
+ 2.
304
+ 200 K ≤ T ≤ 300 K 20 K ≤ T ≤ 60 K
305
+ θW (K)
306
+ µeff (µB)
307
+ θW (K) µeff (µB)
308
+ H ∥ a∗
309
+ -54.3
310
+ 3.76
311
+ -3.78
312
+ 2.94
313
+ H ∥ b
314
+ -54.7
315
+ 3.79
316
+ -3.82
317
+ 2.90
318
+ H ∥ c
319
+ -59.2
320
+ 3.77
321
+ -3.68
322
+ 2.91
323
+ agreement is found for data above 130 K, with a large,
324
+ negative Weiss temperature. The resulting Weiss tem-
325
+ peratures, θW are given in Table. I, as well as the cor-
326
+ responding effective magnetic moments extracted from
327
+ the Curie constants as C = NAµ0µ2
328
+ eff/(3kB). The ob-
329
+ tained effective magnetic moments are close to the value
330
+ expected for a free Nd3+ ion: µeff = gJ
331
+
332
+ J(J + 1)µB =
333
+ 3.6µB. Importantly, the susceptibility data show little
334
+ dependence on the direction of the magnetic field, which
335
+ suggests that, the resulting magnetic anisotropy remains
336
+ quite small.
337
+ Our results are consistent with those re-
338
+ ported in Ref.[21] on polycrystal samples.
339
+ Below 130 K a clear deviation from the high temper-
340
+ ature fit is observed.
341
+ This is roughly consistent with
342
+ the existence of a crystal electric field (CEF) level at
343
+ 15.9 meV (see below), signaling the total depletion of
344
+ the population of the first excited state. A Curie-Weiss
345
+ analysis is heavily affected by the partial population of
346
+ excited multiplets and lead to an overestimation of ex-
347
+ change parameters and exchange couplings. Therefore,
348
+ an additional fit to a Curie-Weiss law for a tempera-
349
+ ture range far enough from the CEF resonance has been
350
+ performed.
351
+ The results are also summarized in Table
352
+ I. Temperatures in the range between 20 K and 60 K
353
+ were considered for this fit. The resulting Weiss temper-
354
+ atures are much reduced compared to the high temper-
355
+ ature fit. However, they still reflect a predominant an-
356
+ tiferromagnetic interaction in Nd3BWO9. The effective
357
+ magnetic moments are also reduced with respect to their
358
+ high temperature value, yielding an average moment of
359
+ µeff = 2.92µB.
360
+ B.
361
+ CEF level scheme
362
+ The inelastic neutron scattering spectra are shown in
363
+ Fig. 3.
364
+ Large intensity at zero energy transfer corre-
365
+ sponds to quasielastic scattering. Three resonances are
366
+ identified at 15.9, 32.8, and 43.7 meV, which we ascribe
367
+ to CEF induced levels due to their temperature depen-
368
+ dence. Importantly, no resonance is found below 15.9
369
+ meV. Since the total angular momentum J = 9/2 of the
370
+ free Nd3+ is expected to be fully split into five Kramers
371
+ doublets, this suggests that the low temperature physics
372
+ of Nd3BWO9 can indeed be described in terms of the
373
+ lowest laying doublet, giving rise to an effective two-level
374
+ system well below ∆ = 15.9 meV ≈ 180 K.
375
+ 3
376
+
377
+ 0
378
+ 10
379
+ 20
380
+ 30
381
+ 40
382
+ 0
383
+ 0
384
+ 0
385
+ ħω (meV)
386
+ T = 1.5 K
387
+ T = 100 K
388
+ T = 300 K
389
+ 0
390
+ 0.1
391
+ 0.2
392
+ 0.3
393
+ 0.4
394
+ 0.5
395
+ 0.6
396
+ 0.7
397
+ 0.8
398
+ 0.9
399
+ 1
400
+ Ef = 14.7 meV
401
+ 2θ = 10°
402
+ Intensity (arb. units)
403
+ FIG. 3. Inelastic neutron scattering intensity at a constant
404
+ scattering angle for three different temperatures. The final
405
+ energy of Ef= 14.7 meV was fixed and incident energy var-
406
+ ied, fixing a 10 degree scattering angle. CEF resonances are
407
+ indicated by black arrows. An offset of 0.25 and 0.50 units
408
+ was added for visibility, a dashed line indicates the reference
409
+ zero for those data sets.
410
+ C.
411
+ Specific heat
412
+ Specific heat as a function of temperature and mag-
413
+ netic field is used to unveil the magnetic phase diagram
414
+ of Nd3BWO9 at ultra-low temperatures. Data obtained
415
+ at zero field are shown in Fig. 4. Nd3BWO9 shows an up-
416
+ turn in specific heat below 4 K with two clearly distinct
417
+ features [Fig. 4(a)]. Around 1 K, a hump in specific heat
418
+ suggests the onset of short-range magnetic correlations
419
+ [24]. At TN = 300 mK we found a sharp lambda anomaly
420
+ representing the transition into magnetic long range or-
421
+ der.
422
+ Below TN the specific heat signal remains large
423
+ down to the lowest accessible temperatures in our setup,
424
+ likely due to nuclear specific heat from the rare-earth
425
+ ions. In order to understand exactly the nature of the
426
+ magnetic specific heat, we have examined the different
427
+ contributions and subtracted them from the measured
428
+ total specific heat.
429
+ To estimate the phononic contribution, we synthesized
430
+ the non-magnetic isostructural material La3BWO9 and
431
+ measured its specific heat in the same range of temper-
432
+ atures.
433
+ This is shown in Fig. 4(a) and represents the
434
+ lattice contribution, CL, in Fig. 4(b).
435
+ An accurate estimation of the nuclear contribution to
436
+ specific heat is usually much more complicated, as a
437
+ 1
438
+ 10
439
+ 100
440
+ T (K)
441
+ 0
442
+ 10
443
+ 20
444
+ 30
445
+ 40
446
+ 50
447
+ Cp (J mol-1 K-1)
448
+ Nd3BWO9
449
+ Nd3BWO9
450
+ TN
451
+ La3BWO9
452
+ 0.1
453
+ 0.4
454
+ 1
455
+ 4
456
+ 10
457
+ T (K)
458
+ Rln(2)
459
+ 0
460
+ 10
461
+ 20
462
+ Cp/T (J mol-1 K-2)
463
+ CL
464
+ CN
465
+ Ctot
466
+ Cmag
467
+ 0
468
+ 1
469
+ 2
470
+ 3
471
+ 4
472
+ T (K)
473
+ 0
474
+ 2
475
+ 4
476
+ 6
477
+ 8
478
+ Smag (J mol-1
479
+ NdK-1)
480
+ (a) μ0H = 0 T
481
+ (b)
482
+ (c)
483
+ TN
484
+ FIG. 4.
485
+ (a) Total specific heat at zero magnetic field for
486
+ Nd3BWO9 and the nonmagnetic isostructural compound
487
+ La3BWO9.
488
+ Nd3BWO9 shows a substantial magnetic con-
489
+ tribution to specific heat below 3 K. (b) Total specific heat
490
+ (open circles) and magnetic specific heat (filled circles) after
491
+ subtraction of lattice and nuclear degrees of freedom. Lat-
492
+ tice (CL) and nuclear (CN) contribution are estimated as
493
+ discussed in the text. A lambda anomaly can be found at
494
+ TN = 0.30 K, signaling the onset of long-range magnetic or-
495
+ der. (c) The magnetic entropy per Nd3+ ion saturates above
496
+ 3 K. A dashed line represents the expected value for a two-
497
+ level system at infinite temperature.
498
+ variety of effects has to be considered.
499
+ These include
500
+ dipole and quadrupolar splitting, or hyperfine coupling
501
+ between nuclei and electrons (which can be quite signif-
502
+ icant in magnetically ordered materials).
503
+ Neodymium
504
+ has two isotopes with nonzero dipolar and quadrupolar
505
+ momenta, out of its 7 stable isotopes. Following the rea-
506
+ soning in Ref. [25], the effect of quadrupolar splitting is
507
+ assumed to be small compared to that of hyperfine cou-
508
+ pling, and we neglect it here. In a magnetized phase,
509
+ local fields are expected to be sizable and therefore hy-
510
+ perfine coupling may significantly contribute to specific
511
+ heat. The contribution from dipole field splitting from a
512
+ single isotopic species is given by
513
+ 4
514
+
515
+ Cp/T (J/molK-2)
516
+ (a) H || a*
517
+ 0
518
+ 0.5
519
+ 1
520
+ 1.5
521
+ T (K)
522
+ 0
523
+ 20
524
+ 40
525
+ 60
526
+ 80
527
+ Cp/T (J/molK-2)
528
+ 0 T
529
+ 0.85 T
530
+ 0.55 T
531
+ 1.2 T
532
+ 0
533
+ 10
534
+ 20
535
+ 30
536
+ 40
537
+ 50
538
+ 60
539
+ 70
540
+ 0 T
541
+ 0.85 T
542
+ 0.975 T
543
+ (b) H || c
544
+ FIG. 5. Typical temperature scans of specific heat for differ-
545
+ ent fixed values of magnetic field applied along (a) H ∥ c and
546
+ (b) H ∥ a∗. An offset of 15 J/mol/K2 has been added for
547
+ visibility. Solid filled triangles show features associated with
548
+ the phase transitions discussed in the main text.
549
+ CH,i =
550
+ NAkB
551
+ α2
552
+ i
553
+ 4I2
554
+ i
555
+
556
+
557
+ 1
558
+ sinh2 �
559
+ αi
560
+ 2Ii
561
+ � −
562
+ (2Ii + 1)2
563
+ sinh2 �
564
+ (2Ii+1)αi
565
+ 2Ii
566
+
567
+
568
+
569
+ (1)
570
+ where αi = AH(µNd
571
+ Hyp/gJ)Ii/kBT, and Ii is the nuclear
572
+ spin, gJ = 8/11 (Land´e factor for Nd), NA is the Avo-
573
+ gadro constant, and kB the Boltzmann constant. AHyp
574
+ represents the strength of the hyperfine coupling and
575
+ here we made a second approximation. We assume all
576
+ the nuclei couple equally to the electron density and the
577
+ value of AHyp is approximated as that of Nd metal [26].
578
+ µNd
579
+ Hyp denotes the static dipole moment of the Nd3+ ions.
580
+ This is precisely the origin of the local field and for its
581
+ value we chose the averaged effective magnetic moment
582
+ from the magnetic susceptibility data at low tempera-
583
+ tures µNd
584
+ Hyp = 2.914µB. Finally, the different species are
585
+ summed, weighted by their isotopical abundance to ob-
586
+ tain the temperature dependence of nuclear specific heat.
587
+ This model with no free parameters is in excellent
588
+ agreement with the lowest temperature data, as shown in
589
+ Fig. 4(b). Having modeled the nuclear specific heat, the
590
+ magnetic specific heat can be extracted by subtraction.
591
+ The magnetic specific heat was subsequently integrated
592
+ to obtain the temperature dependence of magnetic en-
593
+ Cp/T (J/mol K-2)
594
+ 150 mK
595
+ 250 mK
596
+ 350 mK
597
+ 500 mK
598
+ 800 mK
599
+ 0
600
+ 0.5
601
+ 1
602
+ 1.5
603
+ 2
604
+ 2.5
605
+ 3
606
+ 0H (T)
607
+ 0
608
+ 20
609
+ 40
610
+ 60
611
+ Cp/T (J/mol K-2)
612
+ (a) H || a*
613
+ 0
614
+ 20
615
+ 40
616
+ 60
617
+ 80
618
+ 150 mK
619
+ 250 mK
620
+ 350 mK
621
+ 500 mK
622
+ 800 mK
623
+ (b) H || c
624
+ *
625
+ FIG. 6. Typical field scans of specific heat measured at con-
626
+ stant temperature in Nd3BWO9 for (a) H ∥ c and (b) H ∥ a∗.
627
+ An offset of 10 or 15 J/mol/K2 is added for visibility be-
628
+ tween the scans for (a) and (b), respectively.
629
+ Solid filled
630
+ triangles show features associated with the phase transitions
631
+ discussed in the main text. Black arrows signal the existence
632
+ of broad double-hump features, described in the text.
633
+ An
634
+ asterisk shows a feature above the saturation transition.
635
+ tropy, depicted in Fig. 4(c). The high temperature trend
636
+ of this quantity approaches the value of R ln(2), the ex-
637
+ pected value of a two-level system.
638
+ In a magnetic field, a simple estimation of the contri-
639
+ bution of the nuclear spin due to Zeeman splitting could
640
+ not account for the effects observed here. Low tempera-
641
+ ture data in Fig. 5 show that the effect of nuclear specific
642
+ heat is of the same order of magnitude up to 1.2 T and it
643
+ is not strongly field dependent. This suggests that also
644
+ in a field the main contribution comes from hyperfine
645
+ coupling. However, a quantitative determination of this
646
+ effect under magnetic fields becomes paramount.
647
+ The evolution of the specific heat of Nd3BWO9 under
648
+ magnetic fields is shown in Fig. 6 for fields along two
649
+ different crystallographic directions. The total heat ca-
650
+ pacity is displayed here, without subtraction of lattice
651
+ or nuclear degrees of freedom. Typical-field scans show
652
+ a number of anomalies that are consistent with the exis-
653
+ tence of three different phases with static magnetic order
654
+ at low temperatures.
655
+ Up to three distinct features can be observed for
656
+ H ∥ a∗ at the lowest temperature, at 0.45, 0.62 and 1.05
657
+ T and are marked with triangles in Fig. 6(a).These fea-
658
+ 5
659
+
660
+ tures are rather spread in fields, specially at saturation.
661
+ However, the existence of thermodynamic transitions has
662
+ been confirmed by neutron diffraction (as discussed be-
663
+ low). The two lower field anomalies move apart as the
664
+ temperature is increased. The two higher field anoma-
665
+ lies merge at 0.25 K, denoting the highest temperature
666
+ of the ordered phase. Though the specific heat anoma-
667
+ lies in Fig. 6(a) are too broad for a precise estimation
668
+ of the upper critical field, this quantity can be deduced
669
+ from magnetocaloric effect measurements (see below).
670
+ For fields orthogonal to the hexagonal plane (H ∥ c)
671
+ at the lowest temperature one finds two anomalies at 0.5
672
+ T, 0.8 T and a sharper one at 0.95 T. [Fig. 6(b)] Notably,
673
+ in this configuration the different anomalies appear nar-
674
+ rower than for H ∥ a∗, especially at the saturation field.
675
+ The first two anomalies move apart as the temperature
676
+ is increased, while the higher field anomaly barely shifts
677
+ in position up to 0.2 K. The low field anomaly shifts to-
678
+ wards zero field and disappears as TN is reached. The
679
+ two high-field anomalies merge at T = 0.2 K. From the
680
+ high field anomaly we extract an estimate of the satura-
681
+ tion field of µ0Hc = 0.975(3) T. Interestingly, an extra
682
+ feature can be identified above saturation (asterisk in
683
+ Fig. 6(a)).
684
+ This feature shifts to higher fields as the
685
+ temperature is increased and decreases rapidly in mag-
686
+ nitude. Above 0.2 K it is hardly identifiable.
687
+ Finally, double-hump features can be observed above
688
+ 0.3 K for both magnetic field configurations. These are
689
+ significant up to the highest measured temperatures and
690
+ particularly prominent around the saturation field (black
691
+ arrows in Fig. 6).
692
+ For H ∥ c the amplitude of these
693
+ modulations is larger than in H ∥ a∗. Such features are
694
+ often associated with a low-dimensional crossover from
695
+ the zero field disordered phase to the fully polarized state
696
+ without the occurrence of a phase transition [27–30].
697
+ D.
698
+ Magnetocaloric effect
699
+ Magnetocaloric
700
+ effect
701
+ (MCE)
702
+ measurements
703
+ in
704
+ Nd3BWO9 provide key information on the nature of the
705
+ various phase transitions found with other techniques
706
+ [31–33].
707
+ Representative temperature profiles are sum-
708
+ marized in Fig.
709
+ 7. Several crossings can be observed
710
+ for both configurations.
711
+ The observed anomalies are
712
+ too broad to assign exactly a transition point.
713
+ Due
714
+ to the proximity of the thermodynamic transitions
715
+ in the phase diagram, features corresponding to both
716
+ transitions merge and overlap.
717
+ In our measurements
718
+ the field is swept slow enough as to ensure equilibrium
719
+ conditions.
720
+ Data measured with H ∥ a∗ show mostly symmetric
721
+ features around the crossing points.
722
+ Particularly, this
723
+ suggests that the measured phase transitions are of sec-
724
+ ond order. In contrast, the low temperature profiles for
725
+ H ∥ c show two distinct behaviors. At 0.6 T one finds a
726
+ roughly symmetric feature, suggesting again a second or-
727
+ der phase transition. This is different at 0.975 T, where
728
+ a very asymmetric feature appears, pointing to a first
729
+ 0
730
+ μ
731
+ μ
732
+ 1
733
+ 2
734
+ 0H (T)
735
+ 0.1
736
+ 0.2
737
+ 0.3
738
+ 0.5 mT/s
739
+ 0.5 mT/s
740
+ 0.4
741
+ 0.5
742
+ 0.6
743
+ 0.7
744
+ T (K)
745
+ 0
746
+ 1
747
+ 2
748
+ 0H (T)
749
+ (b) H || c
750
+ (a) H || a*
751
+ FIG. 7. Plots of the magnetocaloric effect in Nd3BWO9 for
752
+ different base temperatures and fields applied along (a) H ∥
753
+ a∗ and (b) H ∥ c. For all the scans, red (blue) color represents
754
+ data measured while driving the magnetic field up (down).
755
+ A ramping rate of 0.5 mT/s was used throughout all the
756
+ measurements.
757
+ Small prominent features (specially at low
758
+ fields) are spurious and the result of an unstable platform.
759
+ order or discontinuous transition.
760
+ Finally, the absence of anomalies above the saturation
761
+ field for H ∥ c must be noted. The features observed in
762
+ the fully polarized phase in Fig. 6(b) leave no trace in
763
+ the MCE data in the same configuration.
764
+ The MCE technique is based on the change of entropy
765
+ in a magnetic system as it is driven through a phase
766
+ transition, crossover, level crossing, etc. Consequently,
767
+ one can retrieve the change in entropy in a system from
768
+ the change in temperature against magnetic field [31].
769
+ Under equilibrium conditions, we obtain the entropy as
770
+ ∆S = S(H) − S0 = −
771
+
772
+ κT − Tbath
773
+ T
774
+ dt
775
+ (2)
776
+ where κ is the thermal conductivity of the thermal
777
+ link in the calorimeter, T is the sample temperature, and
778
+ Tbath is the thermal bath temperature. Integration of the
779
+ data in Fig. 7 gives rise to the entropy maps displayed in
780
+ Fig. 8. The data above 0.2 K are a good picture of the
781
+ entropy stored in the magnetic subsystem. However, for
782
+ temperatures below 0.15 K imperfect equilibrium condi-
783
+ tions prevent a reliable estimation of entropy. A strong
784
+ accumulation of entropy is observed above the saturation
785
+ transitions for both field configurations. The position of
786
+ the peaks in entropy match the estimated position of the
787
+ critical fields from specific heat. For H ∥ a∗, the maxima
788
+ in entropy at different temperatures were used to obtain
789
+ an accurate estimate of the upper critical field.
790
+ A fit
791
+ to the data provides Hc,a∗ = 1.187(13) T. This value is
792
+ consistent with the various probes used in this study.
793
+ 6
794
+
795
+ 0
796
+ 0.5
797
+ 1
798
+ 1.5
799
+ 2
800
+ 0
801
+ 0.2
802
+ 0.4
803
+ 0.6
804
+ 0.8
805
+ T (K)
806
+ μ0H (T)
807
+ 0
808
+ 0.2
809
+ 0.4
810
+ 0.6
811
+ 0.8
812
+ T (K)
813
+ (a) H || a*
814
+ A
815
+ B
816
+ S (J molNd K-1)
817
+ -1
818
+ 0
819
+ 1
820
+ 2
821
+ 3
822
+ (b) H || c
823
+ A
824
+ C
825
+ FIG. 8. Entropy maps in false color for two magnetic field
826
+ orientations. In false color plots, the change in entropy ex-
827
+ tracted from magnetocaloric data from Fig. 7. Filled circles
828
+ (diamonds) denote phase anomalies associated with phase
829
+ transitions from specific heat field (temperature) scans for
830
+ (a) H ∥ a∗ and (b) H ∥ c.
831
+ In (a) red squares show the
832
+ maxima of entropy at the measured temperatures. A white
833
+ dashed line is a power law fit to the data showing the best
834
+ estimate for the upper critical field.
835
+ E.
836
+ Magnetization
837
+ The evolution of magnetization under a magnetic field
838
+ provides insight on the type of order in Nd3BWO9.
839
+ Strikingly, a fractional magnetization plateau is observed
840
+ for all measured configurations, as displayed in Fig. 9.
841
+ The value of magnetization is consistent with a fractional
842
+ m=1/3 plateau and spans a range of fields of 0.2-0.3 T.
843
+ In addition, the zero field phase shows zero magnetiza-
844
+ tion for all applied fields, which indicates the realization
845
+ of a gapped phase at T = 0. Magnetization data for in-
846
+ equivalent directions in the hexagonal plane show very
847
+ similar behavior, but differ from the results perpendicu-
848
+ lar to the plane.
849
+ For H ∥ a∗ and H ∥ b the zero magnetization phase
850
+ extends up to 0.4 T. Above 0.5 T the system transitions
851
+ into the factional magnetization plateau state up to a
852
+ 0
853
+ 0.5
854
+ 1
855
+ 1.5
856
+ 2
857
+ 0H (T)
858
+ 0
859
+ 0.5
860
+ 1
861
+ 1.5
862
+ 2
863
+ 2.5
864
+ M ( B/Nd3+)
865
+ 0
866
+ 2
867
+ 4
868
+ 6
869
+ 8
870
+ 10
871
+ 12
872
+ Ms,c/3
873
+ (a.u.)
874
+ H || a*
875
+ H || b
876
+ H || c
877
+ T = 120 mK
878
+ Magnetometer
879
+ H||c
880
+ (0,2,0), 55 mK
881
+ H||a*
882
+ (0,0,2), 130 mK
883
+ Ms,a*/3
884
+ Ms,b/3
885
+ FIG. 9. Magnetization per Nd3+ ion measured at 120 mK
886
+ in Nd3BWO9 for magnetic fields along the crystallographic
887
+ directions a∗, b, and c from bulk measurements (left axis).
888
+ Magnetization extracted from neutron diffraction intensity
889
+ of nuclear reflections is superimposed to the corresponding
890
+ bulk data. Plotted is the rescaled square root of the static,
891
+ magnetic structure factor S∞
892
+ z,z(q) (right axis). The measured
893
+ reflections (Q) are indicated in the figure. Two of the data
894
+ sets have been offset vertically by 0.5 and 1.0 units to improve
895
+ visibility (a dashed line indicates their respective zero). The
896
+ magnetization value at 1/3 of saturation is indicated for each
897
+ individual data set by an arrow next to the plateau state.
898
+ marked limit at 1 T. The transition into the fully satu-
899
+ rated phase is gradual between 1 T and 1.3 T.
900
+ In contrast, much sharper features are found when
901
+ fields H ∥ c are applied. A non-magnetizable phase ap-
902
+ pears up to 0.5 T, above which the system jumps rapidly
903
+ into the plateau state at 0.65 T. The plateau terminates
904
+ in a first-order jump to saturation around 1 T. Notably,
905
+ despite the presence of a first-order transition, our mea-
906
+ surements did not show signatures of hysteresis across
907
+ the saturation transition for H ∥ c.
908
+ Saturation fields extracted from magnetization data
909
+ are consistent with those found in the specific heat
910
+ data. The values for saturation magnetization show little
911
+ anisotropy, finding 1.34(6) µB, 1.31(3) µB, and 1.35(4)
912
+ µB per magnetic ion for configurations a∗, b, and c re-
913
+ spectively. It suggests a nearly isotropic g-tensor in the
914
+ material.
915
+ F.
916
+ Magnetic torque
917
+ Magnetic torque is arguably the most sensitive tech-
918
+ nique to magnetic phase transitions.
919
+ Raw data are
920
+ presented as the change in the measured capacitance
921
+ 7
922
+
923
+ -0.2
924
+ -0.15
925
+ -0.1
926
+ -0.05
927
+ 0
928
+ -0.5
929
+ 0
930
+ 0.5
931
+ -0.4
932
+ -0.2
933
+ 0
934
+ 0.2
935
+ C (fF)
936
+ -2
937
+ -1
938
+ 0
939
+ 1
940
+ dC/dH (fF/T)
941
+ 0
942
+ 1
943
+ 2
944
+ 0
945
+ 0.2
946
+ 0.4
947
+ 0.6
948
+ 0
949
+ 1
950
+ 2
951
+ 0H (T)
952
+ -0.5
953
+ 0
954
+ 0.5
955
+ 1
956
+ 1.5
957
+ 2
958
+ 2.5
959
+ 175 mK
960
+ 200 mK
961
+ 225 mK
962
+ 250 mK
963
+ 275 mK
964
+ 300 mK
965
+ 350 mK
966
+ 400 mK
967
+ 500 mK
968
+ 600 mK
969
+ (a) H || a*
970
+ (b) H || a*
971
+ (c) H || b
972
+ (d) H || b
973
+ (e) H || c
974
+ (f) H || c
975
+ FIG. 10. Magnetic torque (∆C) and its field derivative mea-
976
+ sured at constant temperatures against magnetic field, for
977
+ three field orientations: (a,b) a∗, (c,d) b, and (d,e) c. For all
978
+ data sets, a reference value of capacitance at zero field has
979
+ been chosen and subtracted. Black arrows indicate features
980
+ that may be identified with phase transitions.
981
+ ∆C = C(H) − C(H = 0 T) as a function of magnetic
982
+ field for each temperature (Fig. 10). The torque data
983
+ show strong differences between the measurements in the
984
+ basal plane and perpendicular to it, but the obtained re-
985
+ sults are very similar for both measurements within the
986
+ plane. The raw data show some structure, but not sharp
987
+ features as is customary in such measurements. Phase
988
+ transitions are best captured in the first derivative of the
989
+ raw data 10.
990
+ Field derivative data show features that correspond
991
+ with transitions observed in the other techniques re-
992
+ ported in this study. Direct comparison with the other
993
+ data sets is necessary to pinpoint what anomalies repre-
994
+ sent real phase transitions. These features are indicated
995
+ with arrows in the field derivative data [Fig. 10(b), 10(d),
996
+ and 10(f)]. For H ∥ a∗ two distinct anomalies can be ob-
997
+ served in the scan at 175 mK, at 0.68 T and at 0.99 T.
998
+ These correspond to the lower and upper boundaries of
999
+ the plateau phase. Data for H ∥ b show three anomalies
1000
+ at 0.68 T, 1.02 T and 1.28 T. The lower fields correspond
1001
+ again to the boundaries of the plateau phase. Notably,
1002
+ these two anomalies come together as the temperature is
1003
+ increased and disappear above 300 mK. The higher field
1004
+ anomaly, which is broader and less sharp, corresponds
1005
+ to the crossover into the fully saturated state. Finally,
1006
+ fields applied along the c direction reveal a completely
1007
+ different structure. Three anomalies can be identified at
1008
+ 0.48 T, 0.71 T, and 0.95 T. The associated transitions in
1009
+ this case are the boundaries of the plateau for the high
1010
+ field features and the transition from the low field phase
1011
+ to paramagnet for the low field anomaly. The low field
1012
+ features, though weak, fade away as the transition tem-
1013
+ perature is overcome. The high field anomaly remains up
1014
+ to the highest temperatures representing the crossover of
1015
+ the system into the fully polarized pseudospin.
1016
+ G.
1017
+ Neutron diffraction
1018
+ We resorted to single-crystal neutron diffraction to in-
1019
+ vestigate the magnetic structures realized in the low field
1020
+ and the plateau phases. Figure 11 summarizes the re-
1021
+ sults obtained from the different instruments. The field
1022
+ dependence of the order parameter is depicted for both
1023
+ field configurations, which is in perfect agreement with
1024
+ our thermodynamic measurement data.
1025
+ Zero field data from both experiments unveil a com-
1026
+ mensurate phase with propagation vector Q = (0, 0,1/3).
1027
+ Fig.
1028
+ 11(a) and Fig.
1029
+ 11(b) show that magnetic reflec-
1030
+ tion (1,1,-1/3) is present throughout phase A for both
1031
+ field orientations.
1032
+ The phase is consistent with fully
1033
+ commensurate order, which leads to the appearance of a
1034
+ magnetic supercell, as is shown in Fig. 1(d). Integrated
1035
+ intensity of reflection (1,1,-1/3) drops at the intermedi-
1036
+ ate transition field, above which a different type of or-
1037
+ der is found depending on the direction of the magnetic
1038
+ field. For phase B (H ∥ a∗) we found magnetic reflec-
1039
+ tions (0,1/2,1/2) and (1/2,0,1/2). These reflections van-
1040
+ ish at fields slightly below saturation. Finally, phase C
1041
+ (H ∥ c) has been found to realize order with propaga-
1042
+ tion vector (1/3,1/3,1/3). Magnetic reflection (1/3,1/3,-
1043
+ 1/3), which is inequivalent to the former, has also been
1044
+ found. Fig. 11(b) shows an abrupt drop in the intensity
1045
+ of reflection (1/3,1/3,1/3), consistent with a first order
1046
+ transition to saturation.
1047
+ An external magnetic field induces a ferromagnetic
1048
+ component in every lattice site that gives rise to the
1049
+ bulk magnetization. This produces extra scattering pro-
1050
+ portional to the square of the induced magnetic mo-
1051
+ mentum at the position of each nuclear peak . Fig. 9
1052
+ shows the uniform magnetization density extracted from
1053
+ two nuclear reflections: (020) for H ∥ a∗ and (200) for
1054
+ H ∥ c. We selected reflections where nuclear contribu-
1055
+ tion is minimal while a measurable magnetic intensity
1056
+ can be observed. The zero-field integrated intensity is
1057
+ subtracted from the data in a field to obtain the cor-
1058
+ 8
1059
+
1060
+ 0
1061
+ 2
1062
+ 4
1063
+ 6
1064
+ 0.1
1065
+ 0
1066
+ 0.5
1067
+ 1
1068
+ 1.5
1069
+ 0H (T)
1070
+ 0
1071
+ 5
1072
+ 10
1073
+ Integrated Intensity (arb. units)
1074
+ 0.2
1075
+ 0.3
1076
+ 0.1
1077
+ 0.3
1078
+ 0.1
1079
+ 0.4
1080
+ T (K)
1081
+ 0
1082
+ 1
1083
+ 2
1084
+ 3
1085
+ 4
1086
+ 5
1087
+ Int.(arb. units)
1088
+ 0.2
1089
+ 0.4
1090
+ T (K)
1091
+ 0.36
1092
+ 0.35
1093
+ 0.34
1094
+ 0.33
1095
+ 0.32
1096
+ (1,1,-l) (r.l.u.)
1097
+ Q = (1, 1, -1/3)
1098
+ Q = (0, 1/2, 1/2)
1099
+ Q = (1, 1, -1/3)
1100
+ Q = (1/3, 1/3, 1/3)
1101
+ T = 120 mK
1102
+ A
1103
+ B
1104
+ C
1105
+ A
1106
+ (a)
1107
+ (b)
1108
+ μ0H || a*
1109
+ ZEBRA
1110
+ ZEBRA
1111
+ T = 55 mK
1112
+ μ0H || c
1113
+ WISH
1114
+ (d)
1115
+ µ0H = 0 T
1116
+ 0.1
1117
+ 1.0
1118
+ Int.
1119
+ (a. u.)
1120
+ (c)
1121
+ Q = (1, 1,-⅓-δ)
1122
+ µ0H = 0 T
1123
+ FIG. 11. Results from single crystal magnetic neutron diffrac-
1124
+ tion. (a,b) Field dependence of the integrated neutron inten-
1125
+ sity at the magnetic propagation vectors for: (a) H ∥ a∗
1126
+ (ZEBRA, PSI) and (b) H ∥ c (WISH, ISIS). Note that in (a)
1127
+ the intensity of the (0,1/2,1/2) reflection has been rescaled by
1128
+ ×0.1. In (b) the limits of the ordered phases are highlighted
1129
+ and shown with arrows. (c) Evolution of the integrated in-
1130
+ tensity of the reflection (1,1,-1/3) with temperature at zero
1131
+ magnetic field. (d) Incommensuration of the propagation vec-
1132
+ tor at zero field against temperature, shown as a shift in the
1133
+ peak position of the (1,1,l) reflection.
1134
+ responding magnetic scattering. Longitudinal magneti-
1135
+ zation is then plotted as the square root of mangetic
1136
+ intensity. The agreement with bulk measurements is re-
1137
+ markable and further highlights the existence of magne-
1138
+ tization plateaus regardless of field orientation.
1139
+ Finally, zero field neutron diffraction reveals an incom-
1140
+ mensurate state between the low temperature ordered
1141
+ phase and the paramagnetic phase.
1142
+ The onset of in-
1143
+ commensurate magnetic order appears around 0.34 K at
1144
+ the wavevector Q = (0, 0, 1/3 + δ).
1145
+ Temperature de-
1146
+ pendence of the intensity around the (1,1,l) reflection
1147
+ in Fig. 11(d), where the peak position is superimposed,
1148
+ shows this incommensuration.
1149
+ Reduction of the tem-
1150
+ perature leads to a change in the incommensurate prop-
1151
+ agation vector roughly linearly with temperature.
1152
+ At
1153
+ 0.26 K the propagation vector locks into the commensu-
1154
+ rate Q = (0, 0, 1/3), as observed for the low temperature
1155
+ structure. The robustness of this evolution to commen-
1156
+ suration has been verified for several additional magnetic
1157
+ reflections.
1158
+ IV.
1159
+ DISCUSSION
1160
+ The purported breathing kagome structure is shown in
1161
+ Fig. 1. Unequal Nd-Nd distances and Nd-O-Nd angles
1162
+ result in inequivalent exchange parameters for neighbor-
1163
+ ing corner-sharing triangles [21]. This is represented by
1164
+ the exchange constants J△ and J▽, respectively. How-
1165
+ ever, a crystallographic analysis cannot rule out the ex-
1166
+ istence of interaction between adjacent kagome planes.
1167
+ Due to the short distances between kagome planes, the
1168
+ topology of the exchange interaction in Nd3BWO9 is
1169
+ likely three dimensional. In fact, the shortest superex-
1170
+ change Nd-O-Nd pathway (nearest neighbors, J1) links
1171
+ rare-earth ions belonging to different kagome planes [Fig.
1172
+ 1(b)]. These couplings are arranged into isolated twisted
1173
+ 3-legged spin tubes, one-dimensional structures that ex-
1174
+ tend perpendicular to the kagome planes [see Fig. 1(c)].
1175
+ Noteworthy, the resulting structure considering only
1176
+ nearest neighbor coupling is bipartite.
1177
+ A single tube
1178
+ would show no frustration, highlighting the relevance
1179
+ of further neighbor interactions.
1180
+ A three-dimensional
1181
+ structure with several exchange parameters may have
1182
+ to be regarded, as opposed to the originally suggested
1183
+ kagome structure. Yet, the onset of static magnetic or-
1184
+ der is extremely suppressed by the strong magnetic frus-
1185
+ tration f = −θW /TN ≈ 12.6, confirmed from magnetic
1186
+ susceptibility.
1187
+ Six magnetic rare-earth Nd3+ ions occupy general
1188
+ Wyckoff positions in the unit cell. The reduced point
1189
+ symmetry around the Nd3+ ions [Fig.
1190
+ 1(e)] fully lifts
1191
+ the degeneracy of the total angular momentum levels (J
1192
+ = 9/2) into five Kramers doublets. The strong CEF iso-
1193
+ lates a single Kramers doublet with a large gap to excited
1194
+ multiplets. The obtained zero-field entropy is consistent
1195
+ with a value of S = R ln(2).
1196
+ These two observations
1197
+ show that Nd3BWO9 can be described as an effective
1198
+ spin S = 1/2 system below 100 K. However, the low
1199
+ symmetry precludes attempts to identify unequivocally
1200
+ a CEF-Hamiltonian and to extract the eigenstates of the
1201
+ lowest energy multiplet.
1202
+ Both magnetization and susceptibility suggest very lit-
1203
+ tle magneto-crystalline anisotropy. Susceptibility mea-
1204
+ surements suggest no preferential direction in the high
1205
+ temperature paramagnetic state.
1206
+ In addition, low-
1207
+ temperature magnetization in the fully saturated pseu-
1208
+ dospin phase shows no increase up to the highest probed
1209
+ fields. The increase of magnetization may be a rough
1210
+ estimator of the eigenstate admixing due to anisotropies
1211
+ (via Van-Vleck terms). No appreciable change in mag-
1212
+ netization is observed up to 2 T, indicating the total
1213
+ magnetization in the restricted pseudospin subspace is
1214
+ likely to be an approximately good quantum number. It
1215
+ 9
1216
+
1217
+ is, thus, likely that the low energy physics in Nd3BWO9
1218
+ can be described in terms of a highly symmetric spin
1219
+ Hamiltonian. A small axial anisotropy may be needed
1220
+ to account for the sharp features found for H ∥ c.
1221
+ To map out the phase diagram in the low tempera-
1222
+ ture regime for Nd3BWO9 we use specific heat measure-
1223
+ ments. Using a combination of all outlined techniques,we
1224
+ identify several regions of magnetic order.
1225
+ As shown
1226
+ in Fig. 12, the system reveals complex behaviour, with
1227
+ two different domes of long-range order observed for each
1228
+ configuration.
1229
+ A low field phase (A) extends roughly up to 0.6 T
1230
+ for both studied orientations. This phase possesses com-
1231
+ mensurate order with propagation vector Q =(0,0,1/3).
1232
+ Magnetization measurements show that this phase is
1233
+ hardly magnetizable, suggesting a gapped state in this
1234
+ field range. Although further analysis is needed to un-
1235
+ derstand the magnetic structures of the different phases
1236
+ in detail, a series of general remarks can be deduced
1237
+ from the data. For phase A, the presence of reflections
1238
+ (0,0,±2/3) forbids the existence of a collinear structure
1239
+ with spins parallel to c. Thus, a coplanar structure in
1240
+ the ab plane is likely realized.
1241
+ By increasing the magnetic field the system transitions
1242
+ into a field-induced ordered phase. A field H ∥ a∗ leads
1243
+ to the fractional m = 1/3 plateau phase B, characterized
1244
+ by a propagation vector Q =(0,1/2,1/2). The additional
1245
+ presence of wavevectors (1/2,0,1/2) and equivalent sug-
1246
+ gests a multi-Q structure or the presence of domains in
1247
+ the B phase.
1248
+ Strikingly, the order realized in the plateau is com-
1249
+ pletely different when fields are applied in the basal ab
1250
+ plane or perpendicular to it. In a field H ∥ c, phase C
1251
+ is found with propagation vector Q =(1/3,1/3,1/3). In
1252
+ contrast, saturation H ∥ c occurs through a sharp first
1253
+ order phase transition. Magnetocaloric effect supports
1254
+ this claim. A tricritical termination point appears where
1255
+ first and second order transition lines converge as shown
1256
+ in Fig. 12(b), at 0.20 K and 0.975 T. The presence of
1257
+ magnetic reflections (1/3,1/3,-1/3) and equivalent also
1258
+ indicates a complex spin texture, with either a multi-Q
1259
+ structure or the presence of domains.While here domains
1260
+ may be consistent with the observed first order transition
1261
+ to saturation, it is not possible at this stage to exclude
1262
+ either possibility.
1263
+ The existence of a tricritical point only for one ori-
1264
+ entation may be related to the large spin-lattice inter-
1265
+ action stemming from strong spin-orbit coupling. The
1266
+ transition to saturation for H ∥ c can be prematurely
1267
+ precipitated via an ’order by distortion’ [34] mechanism.
1268
+ A gain in magnetic energy compensates a small loss in
1269
+ elastic energy, leading to a first order transition to sat-
1270
+ uration. Though our neutron diffraction data show no
1271
+ evident change in the space group or lattice parameters
1272
+ in the high field phase, a detailed study would be neces-
1273
+ sary to discard this possibility.
1274
+ Phases A and B appear to merge below 100 mK at
1275
+ 0.56 T. A first order phase transition is speculated be-
1276
+ tween A and B, with a termination bicritical point where
1277
+ 0
1278
+ 0.2
1279
+ 0.4
1280
+ 0.6
1281
+ 0.8
1282
+ T (K)
1283
+ 0
1284
+ 5
1285
+ 10
1286
+ 15
1287
+ Cp/T (J mol-1K-2)
1288
+ (b) µ0H || c
1289
+ (a) µ0H || a*
1290
+ A
1291
+ A
1292
+ C
1293
+ B
1294
+ 0
1295
+ 0.5 ?
1296
+ (0,0,⅓)
1297
+ (0,0,⅓)
1298
+ (0,½,½)
1299
+ (⅓,⅓,⅓)
1300
+ 1
1301
+ μ0H (T)
1302
+ 0
1303
+ 0.2
1304
+ 0.4
1305
+ 0.6
1306
+ FPP
1307
+ 20
1308
+ 25
1309
+ 30
1310
+ FPP
1311
+ 1.5
1312
+ 2
1313
+ FIG. 12. Magnetic phase diagram of Nd3BWO9 in a mag-
1314
+ netic field applied along the principal directions: (a) a∗ and
1315
+ (b) c. The background depicts false color maps of Cp(H, T),
1316
+ with a shared color scale. Symbols: white circles and dia-
1317
+ monds represent transitions obtained from field and tempera-
1318
+ ture scans of specific heat, respectively. Green squares repre-
1319
+ sent the phase boundaries extracted from neutron diffraction
1320
+ data in Fig. 11. Upward-facing blue triangles show transi-
1321
+ tions extracted from bulk magnetization, downward facing
1322
+ pink triangles transitions from magnetic torque. A red dia-
1323
+ mond denotes the estimated position of the tricritical point
1324
+ for H ∥ c. An orange star shows the upper critical field es-
1325
+ timated in Fig. 8. Solid and dashed lines are a guide to the
1326
+ eye, representing second and first order transitions, respec-
1327
+ tively. The different phases are labeled as: A, B, C and Fully
1328
+ Polarized Pseudospin (FPP). The ordered phases show their
1329
+ corresponding magnetic propagation vector, as discussed in
1330
+ the text.
1331
+ all phase boundaries meet. Neutron diffraction data in
1332
+ Fig. 11(a) indicate the phases will likely merge slightly
1333
+ below 120 mK. Interestingly, between A and C the phase
1334
+ boundaries seem to develop smoothly down to the lowest
1335
+ measured temperatures and converge at T = 0. Neutron
1336
+ data at 55 mK show the phases are still separated by
1337
+ paramagnetism at this temperature Fig. 11(b). A highly
1338
+ non-trivial order-to-order quantum phase transition may
1339
+ take place between A and C at zero temperature (indi-
1340
+ cated with a question mark). Precise measurements in
1341
+ 10
1342
+
1343
+ the vicinity of these phase transitions would provide im-
1344
+ portant insight on their nature. However, the strong sig-
1345
+ nal from nuclear degrees of freedom and the extremely
1346
+ low temperatures involved prevent further investigation.
1347
+ The double hump features in specific above the transi-
1348
+ tion temperature represent a crossover from the low field
1349
+ disordered phase to the high field polarized phase. Such
1350
+ features can be understood in terms of models of hard-
1351
+ core bosons and are usually associated with quantum
1352
+ critical behaviour in one dimensional magnets [35, 36].
1353
+ They can be observed in several quasi-1D antiferromag-
1354
+ nets [27, 28], and therefore suggest the relevance of one-
1355
+ dimensional correlations for the physics of Nd3BWO9.
1356
+ These modulations are accentuated when the field is
1357
+ applied along the direction of the spin tubes (H ∥ c).
1358
+ Notably, despite the first-order nature of the transition
1359
+ these modulations are still present and seem to be most
1360
+ prominent around the tricritical termination point.
1361
+ Plateaux in the magnetically ordered sector are a hall-
1362
+ mark of frustrated magnets. The existence of magneti-
1363
+ zation plateaus (and particularly at 1/3 of saturation)
1364
+ has been predicted for both kagome antiferromagnets
1365
+ [37, 38], as well as for a model of isolated spin tubes with
1366
+ a weak triangular rung interaction (see Fig. 1(c)) [39–41].
1367
+ The presence of magnetization plateaux independent of
1368
+ the orientation of the applied magnetic field suggests an
1369
+ stabilizing interplay between frustration mechanisms.
1370
+ Finally,
1371
+ we comment on the origin of the ob-
1372
+ served incommensurate-commensurate (IC-C) transi-
1373
+ tion.
1374
+ Dipolar interactions are not uncommon in the
1375
+ study of rare-earth based magnets due to their large
1376
+ magnetic moments (µ(Nd3+) = 3.6µB) [42]. Their sta-
1377
+ bilizing role on incommensurate structures at temper-
1378
+ atures above commensurate order has been argued in
1379
+ several systems with hexagonal structure [43–45]. The
1380
+ realization of a IC-C transition at zero field opens the
1381
+ question to the importance of dipolar coupling for the
1382
+ low temperature properties of Nd3BWO9.
1383
+ We conclude the discussion by comparing Nd3BWO9
1384
+ to its isostructural compounds. To this point, only two
1385
+ other systems in the R3BWO9 family have been studied
1386
+ at low temperatures. NMR spectra reveal an inconm-
1387
+ mensurate magnetic structure in Sm3BWO9[46], while a
1388
+ dynamical state has been proposed for Pr3BWO9 at tem-
1389
+ peratures as low as 90 mK [47]. These two systems have
1390
+ been analyzed in terms of 2D Hamiltonians based on
1391
+ the existence of the kagome planes. However, our work
1392
+ highlights the presence of three-dimensional couplings
1393
+ and the potential dominance of the one-dimensional spin
1394
+ tubes. The discussion outlined here is inevitably rele-
1395
+ vant for investigations on other members of the family
1396
+ of R3BWO9.
1397
+ V.
1398
+ CONCLUSION
1399
+ We have presented a comprehensive study of the low
1400
+ temperature physics of the highly frustrated quantum
1401
+ antiferromagnet Nd3BWO9. Calorimetric and neutron
1402
+ scattering data support the realization of strongly in-
1403
+ teracting effective spin-1/2 moments below 100 K. Our
1404
+ measurements reveal a complex magnetic phase diagram
1405
+ below 300 mK, featuring magnetization plateaux for all
1406
+ field orientations. The ordering brings about important
1407
+ insight about the relevant magnetic interactions. Differ-
1408
+ ent magnetic structures are realized in the plateau states,
1409
+ depending on the direction of the magnetic field. Even
1410
+ though the phase diagram is considerably anisotropic, it
1411
+ can be described in terms of an effective S = 1/2 pseu-
1412
+ dospin.
1413
+ The experimental framework provided here is key for
1414
+ future studies on Nd3BWO9 and in the remaining mem-
1415
+ bers of the R3BWO9.
1416
+ The presence of the spin-tube
1417
+ structures perpendicular to the kagome planes is indi-
1418
+ cates that the magnetic properties of these highly frus-
1419
+ trated systems cannot be understood in terms of kagome-
1420
+ lattice physics. Further work is needed to fathom the
1421
+ effective dimensionality of the magnetic lattice.
1422
+ VI.
1423
+ ACKNOWLEDGEMENTS
1424
+ This work is supported by a MINT grant of the Swiss
1425
+ National Science Foundation.
1426
+ [1] L. Facheris, K. Y. Povarov, S. D. Nabi, D. G. Mazzone,
1427
+ J. Lass, B. Roessli, E. Ressouche, Z. Yan, S. Gvasaliya,
1428
+ and A. Zheludev, Phys. Rev. Lett. 129, 087201 (2022).
1429
+ [2] A. I. Smirnov, H. Yashiro, S. Kimura, M. Hagiwara,
1430
+ Y. Narumi, K. Kindo, A. Kikkawa, K. Katsumata, A. Y.
1431
+ Shapiro,
1432
+ and a. L. N. Demianets, Phys. Rev. B 75,
1433
+ 134412 (2007).
1434
+ [3] F. Levy, I. Sheikin, C. Berthier, M. Horvati´c, M. Taki-
1435
+ gawa, H. Kageyama, T. Waki, and Y. Ueda, EPL (Eu-
1436
+ rophysics Letters) 81, 67004 (2008).
1437
+ [4] N. Wada, K. Ubukoshi, and K. Hirakawa, J. Phys. Soc.
1438
+ Jpn. 51, 2833 (1982).
1439
+ [5] M. Mourigal, M. Enderle, B. F˚ak, R. K. Kremer, J. M.
1440
+ Law, A. Schneidewind, A. Hiess, and A. Prokofiev, Phys.
1441
+ Rev. Lett. 109, 027203 (2012).
1442
+ [6] V. K. Bhartiya, K. Y. Povarov, D. Blosser, S. Bet-
1443
+ tler, Z. Yan, S. Gvasaliya, S. Raymond, E. Ressouche,
1444
+ K. Beauvois, J. Xu, et al., Phys. Rev. Research 1, 033078
1445
+ (2019).
1446
+ [7] E. Kermarrec,
1447
+ R. Kumar,
1448
+ G. Bernard,
1449
+ R. Henaff,
1450
+ P. Mendels, F. Bert, P. L. Paulose, B. K. Hazra,
1451
+ and
1452
+ B. Koteswararao, Phys. Rev. Lett. 127, 157202 (2021).
1453
+ [8] P. W. Anderson, Mater. Res. Bull. 8, 153 (1973).
1454
+ [9] Y. Zhou, K. Kanoda,
1455
+ and T.-K. Ng, Rev. Mod. Phys.
1456
+ 89, 025003 (2017).
1457
+ [10] L. Savary and L. Balents, Rep. Prog. Phys. 80, 016502
1458
+ (2016).
1459
+ 11
1460
+
1461
+ [11] J. R. Chamorro, T. M. McQueen,
1462
+ and T. T. Tran,
1463
+ Chem. Rev. 121, 2898 (2020).
1464
+ [12] C. Broholm, R. J. Cava, S. A. Kivelson, D. G. Nocera,
1465
+ M. R. Norman, and T. Senthil, Science 367 (2020).
1466
+ [13] B.
1467
+ Bernu,
1468
+ C.
1469
+ Lhuillier,
1470
+ E.
1471
+ Kermarrec,
1472
+ F.
1473
+ Bert,
1474
+ P. Mendels, R. H. Colman, and A. S. Wills, Phys. Rev.
1475
+ B 87, 155107 (2013).
1476
+ [14] H. K. Yoshida, J. Phys. Soc. Jpn. 91, 101003 (2022).
1477
+ [15] M. D. LeBlanc, M. L. Plumer, J. P. Whitehead,
1478
+ and
1479
+ B. W. Southern, Phys. Rev. B 88, 094406 (2013).
1480
+ [16] H. Yoshida,
1481
+ N. Noguchi,
1482
+ Y. Matsushita,
1483
+ Y. Ishii,
1484
+ Y.
1485
+ Ihara,
1486
+ M.
1487
+ Oda,
1488
+ H.
1489
+ Okabe,
1490
+ S.
1491
+ Yamashita,
1492
+ Y. Nakazawa, A. Takata, et al., J. Phys. Soc. Jpn. 86,
1493
+ 033704 (2017).
1494
+ [17] R. Okuma,
1495
+ D. Nakamura,
1496
+ T. Okubo,
1497
+ A. Miyake,
1498
+ A. Matsuo, K. Kindo, M. Tokunaga, N. Kawashima,
1499
+ S. Takeyama, and Z. Hiroi, Nat. Comm. 10, 1 (2019).
1500
+ [18] H. O. Jeschke, H. Nakano, and T. Sakai, Phys. Rev. B
1501
+ 99, 140410(R) (2019).
1502
+ [19] M. Iqbal, D. Poilblanc,
1503
+ and N. Schuch, Phys. Rev. B
1504
+ 101, 155141 (2020).
1505
+ [20] S. Jahromi, R. Orus, D. Poilblanc, and F. Mila, SciPost
1506
+ Phys. 9, 092 (2020).
1507
+ [21] M. Ashtar, J. Guo, Z. Wan, Y. Wang, G. Gong, Y. Liu,
1508
+ Y. Su, and Z. Tian, Inor. Chem. 59, 5368 (2020).
1509
+ [22] A. Majchrowski, E. Michalski, and A. Brenier, J. Cryst.
1510
+ Growth 247, 467 (2003).
1511
+ [23] D. Blosser, L. Facheris,
1512
+ and A. Zheludev, Rev. of Sci.
1513
+ Instr. 91, 073905 (2020).
1514
+ [24] S. E. Dutton, M. Kumar, M. Mourigal, Z. G. Soos, J.-
1515
+ J. Wen, C. L. Broholm, N. H. Andersen, Q. Huang,
1516
+ M. Zbiri, R. Toft-Petersen, and R. J. Cava, Phys. Rev.
1517
+ Lett. 108, 187206 (2012).
1518
+ [25] K. Kimura, S. Nakatsuji, J. J. Wen, C. Broholm, M. B.
1519
+ Stone, E. Nishibori,
1520
+ and H. Sawa, Nat. Comm. 4, 1
1521
+ (2013).
1522
+ [26] B. Bleaney, J. Appl. Phys. 34, 1024 (1963).
1523
+ [27] C. R¨uegg, K. Kiefer, B. Thielemann, D. F. McMor-
1524
+ row, V. Zapf, B. Normand, M. B. Zvonarev, P. Bouil-
1525
+ lot, C. Kollath, T. Giamarchi, S. Capponi, D. Poilblanc,
1526
+ D. Biner,
1527
+ and K. W. Kr¨amer, Phys. Rev. Lett. 101,
1528
+ 247202 (2008).
1529
+ [28] D. Blosser, V. K. Bhartiya, D. J. Voneshen, and A. Zhe-
1530
+ ludev, Phys. Rev. Lett. 121, 247201 (2018).
1531
+ [29] S. Hayashida, D. Blosser, K. Y. Povarov, Z. Yan,
1532
+ S. Gvasaliya, A. N. Ponomaryov, S. A. Zvyagin,
1533
+ and
1534
+ A. Zheludev, Phys. Rev. B 100, 134427 (2019).
1535
+ [30] D. Flavi´an, S. Hayashida, L. Huberich, D. Blosser, K. Y.
1536
+ Povarov, Z. Yan, S. Gvasaliya, and A. Zheludev, Phys.
1537
+ Rev. B 101, 224408 (2020).
1538
+ [31] V. Zapf, M. Jaime, and C. D. Batista, Rev. Mod. Phys.
1539
+ 86, 563 (2014).
1540
+ [32] Y. Kohama, C. Marcenat, T. Klein, and M. Jaime, Rev.
1541
+ Sci. Instrum. 81, 104902 (2010).
1542
+ [33] Y. Kohama, J. Phys. Soc. Jpn. 91, 101004 (2022).
1543
+ [34] O. Tchernyshyov, R. Moessner, and S. L. Sondhi, Phys.
1544
+ Rev. Lett. 88, 067203 (2002).
1545
+ [35] V. E. Korepin and N. A. Slavnov, Commun. Math. Phys.
1546
+ 129, 103 (1990).
1547
+ [36] S. Sachdev, T. Senthil,
1548
+ and R. Shankar, Phys. Rev. B
1549
+ 50, 258 (1994).
1550
+ [37] T. Sakai and H. Nakano, Phys. Rev. B 83, 100405(R)
1551
+ (2011).
1552
+ [38] S. Capponi, O. Derzhko, A. Honecker, A. M. L¨auchli,
1553
+ and J. Richter, Phys. Rev. B 88, 144416 (2013).
1554
+ [39] J.-B. Fouet, A. L¨auchli, S. Pilgram, R. M. Noack, and
1555
+ F. Mila, Phys. Rev. B 73, 014409 (2006).
1556
+ [40] K. Yonaga and N. Shibata, J. Phys. Soc. Jpn. 84, 094706
1557
+ (2015).
1558
+ [41] T. Sakai, M. Sato, K. Okunishi, K. Okamoto,
1559
+ and
1560
+ C. Itoi, in J. Phys., Vol. 400 (IOP Publishing, 2012).
1561
+ [42] S. E. Palmer and J. T. Chalker, Phys. Rev. B 62, 488
1562
+ (2000).
1563
+ [43] H. Shiba and N. Suzuki, J. Phys. Soc. Jpn. 51, 3488
1564
+ (1982).
1565
+ [44] N. Suzuki, J. Phys. Soc. Jpn. 52, 3199 (1983).
1566
+ [45] R. S. Gekht and V. I. Ponomarev, Phase Transitions 20,
1567
+ 27 (1990).
1568
+ [46] K.-Y. Zeng, F.-Y. Song, L.-S. Ling, W. Tong, S.-L. Li,
1569
+ Z.-M. Tian, L. Ma, and L. Pi, Chinese Phys. Lett. 39,
1570
+ 107501 (2022).
1571
+ [47] K. Y. Zeng, F. Y. Song, Z. M. Tian, Q. Chen, S. Wang,
1572
+ B. Liu, S. Li, L. S. Ling, W. Tong, L. Ma, et al., Phys.
1573
+ Rev. B 104, 155150 (2021).
1574
+ 12
1575
+
GtE5T4oBgHgl3EQfWA9Z/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
HtAzT4oBgHgl3EQfjf0-/content/2301.01516v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27221544b4f03a75c2900d5415eeda534149a0e08f3754d5cc561f63cea10b8f
3
+ size 2955122
ItFLT4oBgHgl3EQfJS9D/content/2301.12003v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5e3e66e2cfa42b21632d9eac84b63956f7eec4cb18b80291495d1742c78a235
3
+ size 3394323
ItFLT4oBgHgl3EQfJS9D/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:538e24c26308d8f9214e59521ef851abba9a18152ace59abe6f821a46ea0c0d3
3
+ size 133300
JNFAT4oBgHgl3EQfvB44/content/tmp_files/2301.08673v1.pdf.txt ADDED
@@ -0,0 +1,887 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Disintegration of Long-Period Comet C/2021 A1 (Leonard)
2
+ David Jewitt1, Yoonyoung Kim2, Michael Mattiazzo3, Max Mutchler4, Jing Li1
3
+ and Jessica Agarwal2
4
+ 1Department of Earth, Planetary and Space Sciences, UCLA
5
+ 2Institute for Geophysics and Extraterrestrial Physics, TU Braunschweig, D-38106
6
+ Braunschweig, Germany
7
+ 3Swan Hill Observatory, Australia
8
+ 4 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218
9
+ jewitt@ucla.edu
10
+ Received
11
+ ;
12
+ accepted
13
+ Revised 2022 January 16
14
+ arXiv:2301.08673v1 [astro-ph.EP] 20 Jan 2023
15
+
16
+ – 2 –
17
+ ABSTRACT
18
+ We present imaging observations of the disintegrating long-period comet
19
+ C/2021 A1 (Leonard). High resolution observations with Hubble Space Tele-
20
+ scope show no evidence for surviving fragments, and place a 3σ upper limit to
21
+ their possible radius ∼60 m (albedo 0.1 assumed). In contrast, wide field ob-
22
+ servations from the Swan Hill Observatory, Australia, show an extensive debris
23
+ cloud, the cross-section and estimated mass of which are consistent with com-
24
+ plete disintegration of the nucleus near mid- December 2021 (at about 0.8 au).
25
+ Two methods give the pre-disruption nucleus radius, rn = 0.6 ± 0.2 km. Tidal,
26
+ collisional, sublimation and pressure-confined explosion models provide implau-
27
+ sible explanations of the disintegration. However, rotational instability driven
28
+ by outgassing torques has a very short timescale (∼0.1 year) given the orbit and
29
+ size of the C/2021 A1 nucleus, and offers the most plausible mechanism for the
30
+ disruption. Initial rotational breakup is accelerated by the exposure and strong
31
+ sublimation of previously buried volatiles, leading to catastrophic destruction of
32
+ the nucleus.
33
+ Subject headings: comets: general—comets: individual C/2021 A1
34
+
35
+ – 3 –
36
+ 1.
37
+ INTRODUCTION
38
+ Comet C/2021 A1 (Leonard), hereafter “A1”, was discovered on UT 2021 January 3 as
39
+ a diffuse V ∼ 19 magnitude object inbound to the Sun at heliocentric distance rH = 5 au
40
+ (Leonard 2021). A1 is a long-period comet, with heliocentric osculating semimajor axis a
41
+ = -6124 au, eccentricity e = 1.0001 and inclination i = 132.6◦, reaching perihelion (at rH
42
+ = 0.615 au) on UT 2022 January 03.3, about a year after discovery. Although presently
43
+ following a weakly hyperbolic orbit, the pre-entry orbital elements (corrected for planetary
44
+ perturbations to 1900 January 1, when the heliocentric distance was 137 au) are those of a
45
+ bound object, a = 2020 au, e = 0.999696 and i = 132.7◦. A1 is thus not a dynamically new
46
+ comet, having passed through the planetary system ∼ 105 years ago.
47
+ Comet A1 attained naked eye visibility in late 2021 and then displayed spectacular
48
+ gas and dust tails. However, images and commentary recorded in public on-line archives1
49
+ indicate that A1 became photometrically unstable in 2021 December and 2022 January.
50
+ Measurements of the OH production rate from the Nancay radio telescope were steady near
51
+ QOH = 2.6×1028 s−1 between UT 2021 December 9 and 12, but jumped by a factor of ∼8
52
+ to QOH = 22×1028 s−1 on December 15, even as the heliocentric distance barely decreased
53
+ from 0.80 au to 0.74 au (Crovisier et al. 2021). The morphology also changed, becoming
54
+ more diffuse and with “the tail being more prominent than the head” on UT 2022 January
55
+ 222 at rH ∼ 0.74 au outbound. Based on these early observational reports we requested
56
+ Director’s Discretionary Time on the Hubble Space Telescope (HST), with the science
57
+ objective being to study the presumed breakup of this long-period comet at the highest
58
+ angular resolution. Independently, coauthor Mattiazzo also obtained wide-field imaging
59
+ data using a private telescope at the Swan Hill Observatory in Australia. The wide-field and
60
+ 1e.g. https://britastro.org/cometobs/2021a1/thumbnails.html
61
+ 2https://groups.io/g/comets-ml/message/30541
62
+
63
+ – 4 –
64
+ HST data are highly complementary, with the former providing sensitivity to low surface
65
+ brightness debris over a wide angle and the latter providing ultra-high resolution and very
66
+ deep imaging of the near-nucleus region.
67
+ While the phenomenon of cometary breakup has been known for over a century, very
68
+ few physical observations of disintegrating comets are to be found in the refereed literature.
69
+ In this paper, we present the observations and consider possible causes of the breakup of
70
+ comet A1.
71
+ 2.
72
+ OBSERVATIONS
73
+ 2.1.
74
+ Hubble Space Telescope
75
+ The 2.4 m diameter Hubble Space Telescope was used to observe disintegrating A1
76
+ under program GO 16929. We used the WFC3 camera, which houses two 2015×4096 pixel
77
+ charge coupled devices separated by a 1.2′′ wide gap. The 0.04′′ pixel−1 image scale gives a
78
+ full-frame 162′′×162′′ field of view. HST images were taken using the F350 LP filter in order
79
+ to maximize throughput. This filter has an effective central wavelength λc = 6230˚A when
80
+ observing a Sun-like (G2V) source and a FWHM ∆λ = 4758˚A. We secured four images
81
+ each of 450 s duration in each of the first three orbits and five frames of 285 s, with a
82
+ sub-frame readout, in the fourth. The first three orbits were obtained in 2022 April with
83
+ spacings of one and four days, with the intention being to measure the sky-plane motions
84
+ of fragments produced by the break-up of A1. The fourth orbit was scheduled on UT 2022
85
+ June 7 to coincide with the passage of the Earth through the projected orbit plane of the
86
+ comet. Observations from this vantage point provide a model-free measure of the thickness
87
+ of the dust distribution perpendicular to the plane. Unfortunately, the images from the
88
+ fourth orbit suffered from extreme field star contamination, as a result of the low (-6◦)
89
+
90
+ – 5 –
91
+ galactic latitude of the comet, and were not useful.
92
+ 2.2.
93
+ Swan Hill Observatory
94
+ Wide-field observations were taken by co-author Michael Mattiazzo using a 0.28 m
95
+ diameter, f/2.2 wide-field telescope at the Swan Hill Observatory (observatory code Q38),
96
+ located in Victoria, Australia. A 4655×3522 pixel CMOS imaging device (Panasonic model
97
+ QHY163M) provided an image scale of 1.27′′ pixel−1, and a field of view approximately
98
+ 1.6◦×1.2◦. Each pixel of the 0.28 m telescope subtends a solid angle equal to 103 HST
99
+ pixels. Ten images each of 30 s duration were obtained, during which time the comet moved
100
+ relative to field stars by about 2.7′′, which is small compared to the 5.1′′ full width at half
101
+ maximum of point source objects in the data. The wide field image shows evidence for loss
102
+ of sensitivity due to vignetting, especially near the corners of the device. We removed this
103
+ by fitting a cubic spline surface to the image, using the median signal within 50×50 pixel
104
+ boxes (after checking that the procedure did not self-subtract the comet).
105
+ No filter was employed in order to maximize the throughput of the system. The
106
+ quantum efficiency of the detector peaks near a central wavelength 5500˚A, and has a
107
+ FWHM estimated at ∼4000˚A. The central wavelength is close to that of Johnson V (see
108
+ the discussion in Bessel 1990), but the response is so broad that it captures the same light
109
+ as the Johnson B, V and R filters (or, equivalently, the Sloan g and r filters) combined.
110
+ The large bandwidth and lack of a standard filter together limit the accuracy with which
111
+ the measured magnitudes can be related to, for example, the V band magnitudes. We
112
+ calibrated the data using measurements of field stars on the Sloan filter system, provided
113
+ by the Skymapper southern survey (Wolf et al. 2018). For this purpose we extracted
114
+ measurements using circular apertures of projected radius 12.7′′, with sky subtraction from
115
+ the median signal within a concentric annulus having inner and outer radii 19.1′′ and 38.1′′,
116
+
117
+ – 6 –
118
+ respectively. In order to minimize the color term in our photometry, we selected stars with
119
+ optical color g-r ∼0.4 to 0.5, so as to approximately match the color of the Sun (given as g-r
120
+ = 0.45±0.02 by Holmberg et al. 2006). We further selected these stars to lie within ∼1′ of
121
+ the comet in order to minimize spatial variations in the photometry caused by imperfect
122
+ flatness of the data.
123
+ The geometrical circumstances of observation are given in Table 1.
124
+ 3.
125
+ RESULTS
126
+ 3.1.
127
+ High Resolution Data
128
+ We combined the four images from each orbit in order to reject cosmic rays, suppress
129
+ trailed field objects, and reach a fainter limiting magnitude. The composite from UT 2022
130
+ April 5 is shown in Figure 1; composites from April 6 and 10 look the same. The predicted
131
+ location of the nucleus is indicated in the Figure. The JPL Horizons ephemeris for April
132
+ 5 gives 3σ positional uncertainties of ±1.3′′ in right ascension and ±1.0′′ in declination,
133
+ both of which are negligible compared to the 160′′ field of view of WFC3. We searched
134
+ for the principal nucleus and discrete fragments in the data by comparing image subsets
135
+ to identify correlated motion, but found none. Instead, the images show evidence for
136
+ diffuse light scattered from cometary dust, evident in Figure 1 as a region of slightly higher
137
+ surface brightness in the south east quadrant of the image (marked by a dashed white
138
+ line in the right-hand panel of the figure). Although it at first resembles a flat-field defect
139
+ or a smudge of internally scattered light, two lines of evidence show that this region of
140
+ diffuse brightness is neither. First, the enhanced region is fixed with respect to the daily
141
+ predicted ephemeris position of A1. Second, the enhanced region moves on the detector as
142
+ the telescope orientation angle changes. The enhancement appears at the same position in
143
+
144
+ – 7 –
145
+ image composites from all three dates in April, whereas scattered light from bright stars
146
+ outside the WFC3 field of view would vary as the background stars are completely different
147
+ from day to day. A flat-field defect would not rotate as the telescope orientation changes.
148
+ We conclude that the diffuse light is sunlight scattered from cometary debris released from
149
+ the now invisible nucleus of A1.
150
+ The on-line WFC3 Exposure Time Calculator3 gives a 3σ limit for detection of point
151
+ source objects at V = 26.7, in each of our orbits. This limiting magnitude is consistent
152
+ with the measured sky noise in the data. Corrected to absolute magnitude using phase
153
+ coefficient β = 0.04 magnitude degree−1, we find H ≥ 22.81. For a nominal albedo, pV =
154
+ 0.1, this corresponds to a 3σ limit to the fragment radius, r ≤ 60 m.
155
+ 3.2.
156
+ Wide Field Data
157
+ The composite wide field image is shown in Figure 2. A low surface brightness dust
158
+ structure extends over at least 0.4◦ (2×106 km in the plane of the sky), with a position
159
+ angle 120◦±2◦ and no indication of a brightness peak at the expected location of the
160
+ nucleus. The latter was determined from the JPL Horizons ephemeris for the mid-time of
161
+ the image, and is marked in the figure. Overall, the morphology is similar to that of C/2010
162
+ X1 (Elenin), a long period comet which disintegrated when inbound near rH = 0.6 AU (Li
163
+ and Jewitt 2015), and C/2019 J2 (Palomar), which disintegrated pre-perihelion near rH
164
+ = 1.9 au (Jewitt and Luu 2019). Comparison with Figure 1 shows that the HST, which
165
+ was pointed at the expected location of the nucleus, indeed recorded diffuse light from the
166
+ western tip of this dust structure.
167
+ We estimated the total light from the dust as follows. First, we rotated the image to
168
+ 3https://etc.stsci.edu/etc/input/wfc3uvis/imaging/
169
+
170
+ – 8 –
171
+ bring the long axis of the dust tail to the horizontal (upper panel in Figure 3). Next, we
172
+ manually replaced field stars with the average of surrounding pixels. The median signal from
173
+ the comet was then computed within a rectangular box, “A” in the lower panel of Figure
174
+ 3) 1105′′ long by 380′′ tall, and the background sky estimated from equal-sized photometry
175
+ boxes contiguous with the comet box but displaced above and below it (“B” and “C” in
176
+ Figure 3). Figure 3 shows that the tail extends beyond the left edge of the photometry box
177
+ “A” but the increased uncertainty imposed by the sky rendered measurements of this very
178
+ faint material impractical. The light from the tail was calculated from fT = fA−(fB+fC)/2,
179
+ where fx is the flux in box “x”. Then, applying the calibration obtained from field stars,
180
+ we find VT = 10.9±0.5, where the quoted error is our best estimate of the uncertainty
181
+ resulting from non-flatness of the data, the transformation from the wide response of the
182
+ camera and the effective V magnitude. With assumed phase function 0.02±0.02 magnitude
183
+ degree−1 and the geometry given in Table 1, the corresponding absolute magnitude is H =
184
+ 7.6±0.6, where the larger uncertainty is introduced by the phase correction. The scattering
185
+ cross-section needed to give this absolute magnitude is C = 1.4+1.0
186
+ −0.8 × 1010 m2, assuming
187
+ geometric albedo pV = 0.1 (appropriate for cometary dust; Zubko et al. 2017).
188
+ Figure 4 shows the averaged surface brightness profile from the March 31 image,
189
+ measured parallel to the long axis of region A in Figure 3. Most of the scatter in the
190
+ surface brightness profile is statistical noise in the data, but larger oscillations (for example
191
+ at ∼480′′ and 750′′) result from spatial background variations caused by the digital removal
192
+ of field stars. In this plot, the peak of 1000 units corresponds to a surface brightness Σ =
193
+ 24.4 magnitudes arcsec−1, about 5% of the surface brightness of the night sky. The surface
194
+ brightness shows a steep increase, reaching a maximum at about 100′′ from the ephemeris
195
+ nucleus location, followed by a steady decline at larger projected angles. This profile shape
196
+ is indicative of a suddenly terminated dust mass release, with the peak of the profile giving
197
+ the distance traveled by the largest, slowest particles.
198
+
199
+ – 9 –
200
+ 4.
201
+ DISCUSSION
202
+ 4.1.
203
+ Radius and Mass of the Nucleus
204
+ We use the effective spherical nucleus radius of A1 ¯r = 0.6±0.2 km from Jewitt (2022).
205
+ This estimate is based on independent measurements of QH2O(1), the gas production rate
206
+ at 1 au, and of α1, the non-gravitational acceleration at 1 au. Comet A1 has QH2O(1) =
207
+ 1.9×1028 s−1 (only pre-perihelion observations are used because post-perihelion rates are
208
+ clearly affected by the breakup) and α1 = 1.3×10−6 m s−2, provided by JPL Horizons. A
209
+ substantially smaller nucleus would have a surface area insufficient to supply the QH2O(1),
210
+ while a substantially larger nucleus would have too much mass to be accelerated at α1
211
+ given the known gas production rate. Using ¯r and nominal nucleus density ρn = 500 kg
212
+ m−3 (Groussin et al. 2019), we estimate the nucleus mass Mn = (4.5+6.5
213
+ −3.2) × 1011 kg. The
214
+ largest surviving fragments, with radii <60 m, individually contain < 10−3 of the mass of
215
+ the primary.
216
+ 4.2.
217
+ Time of Disruption
218
+ Syndynes (the loci of particles having one size, released with zero initial relative
219
+ velocity over a range of times; Finson & Probstein (1968)) are curved and do not match
220
+ the linear shape of the debris cloud in A1. Instead, the morphology more resembles a
221
+ set of synchrones as shown in Figure 5. Synchrones trace the loci of particles in the sky
222
+ plane having a range of sizes (hence, radiation pressure accelerations) but released from the
223
+ nucleus simultaneously. The position angle of the debris trail in A1 is most compatible with
224
+ ejection 110±10 days before the image was taken, i.e. on UT 2021 December 11±10. This is
225
+ about a month before reports of distinct morphological change appeared but coincides with
226
+ a dramatic increase in the OH production rate from 4.4×1028 s−1 on UT 2021 December
227
+
228
+ – 10 –
229
+ 19 to 14×1028 s−1 on UT 2021 December 21, in unpublished SOHO/SWAN data (personal
230
+ communication M. Combi). It is also close to a reported OH outburst on UT 2021 December
231
+ 15 (Crovisier et al. 2021). While we lack continuous coverage of the gas production from
232
+ A1, it is likely that the sublimation rate became highly unstable as a result of the breakup
233
+ of the nucleus when close to perihelion.
234
+ We assume that the disintegration began on UT 2021 December 11±10. To reach the
235
+ far end of the measured debris cloud (an angular distance ∼1500′′, corresponding to linear
236
+ distance L = 2.2 × 106 km) under the action of radiation pressure requires an average
237
+ acceleration 2L/∆T 2, where ∆T = 111 days (9.6×106 s) is the interval between the time
238
+ of disintegration and the Swan Hill image from UT 2022 March 31. In units of the solar
239
+ gravitational acceleration at the average rH = 1.3 au heliocentric distance in this period,
240
+ β =
241
+ 2Lr2
242
+ H
243
+ g⊙(1)∆T 2
244
+ (1)
245
+ where g⊙(1) = 0.006 m s−2 is the solar gravity at 1 au and rH is expressed in au.
246
+ Substituting, we obtain β = 0.01. With β ∼ 1/aµm, where aµm is the particle radius
247
+ expressed in microns (c.f. Bohren & Huffman (1983)), we infer that the particles at the far
248
+ end of the tail in the March 31 image had aµm ∼ 75 µm. All particles in the visible debris
249
+ cloud on UT 2022 March 31 must be larger, while smaller particles were presumably ejected
250
+ but have been swept by radiation pressure beyond the visible extent of the tail. Particles
251
+ near the peak of the surface brightness profile (angular distance ∼100′′, corresponding to
252
+ L = 1.4 × 105 km) have β ∼ 10−3 by Equation 1 and, therefore, radii ∼1 mm.
253
+
254
+ – 11 –
255
+ 4.3.
256
+ Mass of the Optical Debris
257
+ How does the mass of the debris compare with the mass of the nucleus prior to its
258
+ disappearance? To answer this question, we treat the debris as consisting of a distribution
259
+ of spherical particles with radii between a and a+da written as n(a)da. Then, the combined
260
+ mass of the particles between minimum radius a1 and maximum radius a2 is
261
+ Md =
262
+ � a2
263
+ a1
264
+ 4
265
+ 3πρa3n(a)da
266
+ (2)
267
+ while their combined cross-section is
268
+ C =
269
+ � a2
270
+ a1
271
+ πa2n(a)da
272
+ (3)
273
+ It is useful to represent the size distribution as a power law
274
+ n(a)da = Γa−γda
275
+ (4)
276
+ where γ is the differential size distribution index and Γ is a normalizing constant.
277
+ Substituting equation 4 into equations 2 and 3 and eliminating Γ, we obtain
278
+ Md = 4
279
+ 3ρC
280
+ � a2
281
+ a1 a3−γda
282
+ � a2
283
+ a1 a2−γda
284
+ (5)
285
+ The minimum particle radius is selected as a1 = 75 µm, since all smaller particles
286
+ would have been swept out of the image field in the time since ejection. The maximum
287
+ radius, a2 = 60 m, is set by the non-detection of larger bodies in our deep HST imaging
288
+ data. With these values for a1 and a2, we plot Equation 5 as a function of γ in the range 2.5
289
+ ≤ γ ≤ 4.0 (Figure 6). The particle mass required to account for the measured cross-section,
290
+
291
+ – 12 –
292
+ C, is seen to vary by orders of magnitude for modest changes in the index, γ, with smaller
293
+ values (flatter distributions) hiding a larger fraction of the total mass in big bodies.
294
+ Also plotted in the figure is the nucleus mass, Mn = (4.5+6.5
295
+ −3.2) × 1011 kg, computed from
296
+ the effective radius, rn = 0.6±0.2 km, (Section 4.1), and density, ρn = 500 kg m−3, with the
297
+ mass uncertainty marked as a horizontal yellow band. The red point marks the intersection
298
+ of the two curves where Md = Mn and shows that, for index γ = 3.5±0.1, the debris mass
299
+ and nucleus mass are equal. The upper limit to the size distribution could be substantially
300
+ smaller than the 0.6 km limit set by the Hubble data, in which case a smaller value of
301
+ the index would be needed for the mass of the debris to equal the mass of the nucleus.
302
+ A relevant comparison can be made with the size distribution of the Kreutz sungrazing
303
+ comets, which are themselves produced by the fragmentation of a precursor body. The
304
+ Kreutz objects have γ = 3.2 in the 5 m to 35 m radius range (Knight et al. 2010), plotted
305
+ as a blue square in Figure 6. The uncertainty on γ for the Kreutz objects is not stated;
306
+ we have plotted a nominal ±0.1 error bar for reference and note reasonable agreement
307
+ with the index deduced for A1 within the uncertainties. Perhaps less relevant are radar
308
+ measurements of the debris size distributions in six meteoroid streams, most associated
309
+ with decaying comets. These are plotted for comparison using green triangular symbols
310
+ (Blaauw et al. (2011)). The formal meteoroid stream index uncertainties are comparable to
311
+ the size of the symbols in the figure. The measured indices span the range γ = 3.2 to 3.7,
312
+ encompassing the values found for A1 and the Kreutz comets.
313
+ We conclude that the optical cross-section presented by the debris in 2022 March is
314
+ consistent with the complete disintegration of the original ∼0.6 km scale nucleus into a
315
+ power law distribution (index γ = 3.5±0.1) of particle sizes. We emphasize that we possess
316
+ no independent evidence that the debris mass and original nucleus mass are equal, although
317
+ a consideration of the particle properties using more detailed considerations (section 4.4)
318
+
319
+ – 13 –
320
+ supports this result. It should also be noted that 60 m is an upper limit to the size of
321
+ the largest post-disruption “particles” and our result would be changed if a2 ≪ 60 m, as
322
+ it would if the size distribution of particles is not well represented by a single power law
323
+ across the full range of sizes. It is also not obvious that the density of the particles should
324
+ necessarily be the same as the bulk density of the nucleus, as we have assumed. These and
325
+ other physically plausible possibilities lie beyond the observational constraints obtained
326
+ from the data.
327
+ 4.4.
328
+ Monte Carlo Simulation
329
+ We next used a Monte Carlo simulation as developed by Ishiguro et al. (2007) (see
330
+ also Kim et al. (2017)) to model the cometary debris in more detail. The model is
331
+ under-constrained and cannot provide unique solutions for the particle properties. It
332
+ is nevertheless valuable in allowing us to test the deductions made based on order of
333
+ magnitude considerations, and also to more fully explore the range of plausible solutions.
334
+ We particularly examined the effect of the particle size distribution index and the minimum
335
+ and maximum particles sizes in the distribution.
336
+ Figure 7 shows the data with results of simulations for γ = 3.3, 3.4 and 3.5 and size
337
+ parameter in the range 7 × 10−4 ≤ β ≤ 0.07, with ejection on 2021 December 11. The
338
+ upper limit to β (lower limit to particle radius) is set by the field of view, with smaller
339
+ particles have already been pushed out of the field by radiation pressure. We obtain a ≥
340
+ 14 µm, different by a factor of five from the limit a ≥ 75 µm estimated by the order of
341
+ magnitude procedure, above. The lower limit to β (upper limit to the particle size of ∼1.4
342
+ mm) is determined from the location of the surface brightness peak in Figure 7. This is very
343
+ small compared to the 60 m upper limit to the radius of the largest possible fragment, set
344
+ by non-detection in the HST images. However, this difference is understandable since, for
345
+
346
+ – 14 –
347
+ commonly measured cometary size distributions, the scattering cross-section is dominated
348
+ by the smallest particles; large particles contribute little to the cross-section and thus are
349
+ poorly constrained by scattered light observations. In order to fit the data, we assumed
350
+ that the particle ejection speed varies with size parameter as V = V0β1/2, with V0 = 550
351
+ m s−1 being the gas thermal speed. Unlike the particle trails of weakly active comets and
352
+ asteroids, a high ejection speed is required in order to fit the large width of the debris cloud
353
+ in A1.
354
+ As is evident in Figure 7, the plotted models do not perfectly reproduce the measured
355
+ surface brightness profile, with larger γ models being 25% to 30% brighter than the data at
356
+ large distances from the nucleus and smaller γ models being too sharply peaked compared
357
+ to the measurements. If they are real, these differences could result from physical effects not
358
+ included in the model. For example, we have ignored dust released before disintegration,
359
+ reasoning that the dramatic outbursts and brightening starting in mid-December would
360
+ swamp any signal from older material. As another example, large aggregate grains in the
361
+ tail might break up into smaller particles which would be quickly swept from the field of
362
+ view by radiation pressure, perhaps explaining the lower brightness of the tail ≳1000′′ from
363
+ the nucleus. On the other hand, the differences between the models and the measured
364
+ profile are certainly affected by systematic uncertainties intrinsic to the wide field data,
365
+ particularly by imperfect flatness of the data and by the presence of scattered light from
366
+ bright background sources. Rather than over-interpret the data, we conclude from the
367
+ Monte Carlo simulation only that γ ∼ 3.4 ± 0.1 provides a broad match to the profile, while
368
+ much steeper and much less steep distributions do not. The range of allowable indices
369
+ deduced from Monte Carlo models is consistent with γ = 3.5 ± 0.1 as inferred from the
370
+ debris mass in Section 4.3 (c.f. Figure 6).
371
+ Lastly, we used the Monte Carlo model to test the possibility that the debris observed
372
+
373
+ – 15 –
374
+ in 2022 March could be long-lived material released before perihelion, in the form of a
375
+ so-called “neck-line” structure (e.g. Pansecchi and Fulle 1990). We find that material
376
+ ejected in the period 2021 November 15 to December 15 would produce a tail structure
377
+ in March having position angle (113◦) distinctly different from that measured (120◦) or
378
+ calculated from the impulsive ejection model (119◦). In addition, neck-line structures in
379
+ other comets are most prominent when observed from near the projected orbital plane,
380
+ whereas our observations were taken ∼20◦ from the orbital plane of C/2021 A1 (c.f. Table
381
+ 1). The combination of the unfavorable observing geometry, the failure to reproduce the
382
+ measured position angle of the dust in 2022 March, and the obvious importance of the
383
+ outbursts reported in 2021 December together show that pre-perihelion dust is a negligible
384
+ contributor to the post-perihelion appearance.
385
+ 4.5.
386
+ Disintegration Mechanism
387
+ The preceding discussion shows that a ∼0.6 km scale nucleus disintegrated into
388
+ fragments, the largest of which were no more than about 10% of the radius of the original
389
+ body. What process could lead to such a dramatic outcome?
390
+ Tidal Breakup: The 0.615 au perihelion distance of A1 far exceeds the Roche radius
391
+ of the Sun (∼10−2 au), negating the possibility of a tidal breakup. Comet A1 did pass
392
+ within a distance rV = 0.029 au from Venus on UT 2021 December 18 (Zhang et al. 2021)
393
+ but this is still ∼300 times the Roche radius (∼10−4 au) of the planet. To within a numerical
394
+ multiplier, the differential of the gravitational force on opposite sides of the nucleus is
395
+ ∆F ∼ (GMV ρnr3
396
+ n/r2
397
+ V )(rn/rV ) giving an order of magnitude tidal stress S ∼ ∆F/r2
398
+ n or
399
+ S ∼ GMV ρnr2
400
+ n
401
+ r3
402
+ V
403
+ ,
404
+ (6)
405
+
406
+ – 16 –
407
+ where G = 6.67 × 10−11 N kg−2 m2 is the gravitational constant, MV = 5 × 1024 kg is
408
+ the mass of Venus and the other quantities are already defined. Substituting ρn = 500 kg
409
+ m−3, rn = 600 m, and rV = 0.029 au, we estimate S ∼ 10−6 N m−2 at closest approach,
410
+ which is orders of magnitude smaller even than the cohesive strengths of fine, unconfined
411
+ powders (S ≳ 100 N m−2) measured in the laboratory (Garcia-Trinanes et al. 2019). The
412
+ disintegration of A1 is very unlikely to be a consequence of tidally induced stresses.
413
+ Equilibrium Sublimation: The rate of loss of surface material is drn/dt ∼ −fs/ρ,
414
+ where fs ∼ 2 × 10−4 kg m−2 s−1, at 1 au. Substitution gives drn/dt ∼ -3 cm day−1. At this
415
+ rate, the timescale for eroding the whole nucleus would be |rn/(drn/dt)| ∼ 40 years, which
416
+ is very large compared to the ∼1 year spent by A1 in the vicinity of the Sun. In any case,
417
+ sublimation would produce steady erosion of the comet not a catastrophic disintegration
418
+ like that observed. Equilibrium sublimation cannot account for the sudden disintegration
419
+ of A1.
420
+ Collisional Disruption: Collisional disruption timescales for 0.6 km scale objects,
421
+ even in the dense parts of the asteroid belt, are measured in hundreds of millions of years
422
+ (Bottke et al. 2005). Comet A1 arrived from a high inclination orbit and disintegrated ∼0.5
423
+ au from the ecliptic plane where there are no known objects with which to collide. We
424
+ confidently dismiss the possibility that A1 was collisionally disrupted.
425
+ Internal Pressure: Could internal pressure build-up from sublimated gases cause
426
+ the nucleus to explode (Samarasinha 2001)? The core temperature of the nucleus of A1
427
+ is comparable to the Oort cloud equilibrium temperature of just a few degrees above
428
+ absolute zero. Heat transport from the surface to the interior by conduction is controlled
429
+ by the thermal diffusivity, which is proportional to the conductivity and which, in turn,
430
+ is strongly affected by the particulate nature and porosity of the cometary material.
431
+ Laboratory measurements of porous, dielectric powders yield conductivities ∼ 102 to 103
432
+
433
+ – 17 –
434
+ times smaller than the solid material (Henke et al. 2012). The expected high porosities
435
+ and low thermal diffusivities of cometary material lead to small thermal skin depths that
436
+ make deep conduction impossible. Heat applied for a time τ will conduct over a distance
437
+ d ∼ (κτ)1/2, where κ is the diffusivity. For example, with κ = 10−8 to 10−9 m2 s−1, even in
438
+ the year between discovery at rH = 5 au and perihelion at 0.6 au, conducted heat travelled
439
+ into the nucleus by a characteristic distance only d ∼ 0.2 m to 0.5 m. This distance is so
440
+ small compared to the nucleus radius that it is difficult to see how subsurface gas produced
441
+ by surface heating could have any relevance to the complete disintegration of the nucleus.
442
+ Rotational Instability: The remaining possibility for nucleus break-up is also the
443
+ most plausible. The timescale for changing the spin angular momentum of a spherical
444
+ nucleus through outgassing torques is (Jewitt 2021)
445
+ τs =
446
+ �16π2
447
+ 15
448
+ � � ρnr4
449
+ n
450
+ kTVthP
451
+ � � 1
452
+ ˙M
453
+
454
+ ,
455
+ (7)
456
+ where P is the instantaneous spin-period and kT is the dimensionless moment arm, equal
457
+ to the fraction of the outflow momentum that exerts a torque on the nucleus. The median
458
+ values in a sample of short-period comet nuclei with perihelia in the range 1 ≤ q ≤ 2 au
459
+ are kT = 0.007 and P = 15 hours (5×104 s) (Jewitt 2021). We substitute
460
+ ˙M = 800 kg
461
+ s−1, equal to the sublimation rate at 1 au as measured by Combi’s Lyman-α data, on the
462
+ understanding that this sets a lower bound to the mass loss rate at smaller distances and
463
+ therefore sets an upper limit to τs. With ρn = 500 kg m−3, rn = 600 m, and Vth = 500
464
+ m s−1, substitution into Equation 7 gives τs < 5 × 106 s (0.16 year, or 2 months), which
465
+ compares to the 6 weeks (0.12 year) spent by A1 with rH < 1 au. While this is not proof
466
+ that A1 disintegrated through a rotational instability, given the nominal nucleus parameters
467
+ and measured mass loss rate, rotational instability does offer a plausible mechanism for
468
+ nucleus disintegration.
469
+
470
+ – 18 –
471
+ Rotational breakup is expected to launch fragments with a velocity dispersion
472
+ comparable to the tangential speed of the nucleus due to its rotation. For a strengthless
473
+ nucleus, this equals the gravitational escape speed from the primary, in this case ∼0.3
474
+ m s−1. In contrast, the Monte Carlo models show that larger speeds are required to fit
475
+ the head width of the debris trail. For example, with V = V0β1/2 and V0 = 550 m s−1,
476
+ millimeter sized particles (β = 0.001) would have V ∼ 17 m s−1, about 60 times the escape
477
+ speed. We conjecture that these higher speeds result from gas drag acceleration following
478
+ the exposure and intense sublimation of previously buried ices caused by rotational breakup
479
+ at rH ∼ 0.8 au.
480
+ Very large particles and boulders would not be substantially accelerated by gas drag
481
+ and should leave the disintegrating nucleus at about the escape velocity of the primary.
482
+ In the ∼3 months elapsed between the first signs of breakup and the HST observations,
483
+ such slow-moving fragments would travel ∼2000 km, a distance subtending 1′′ to 2′′ in the
484
+ plane of the sky (c.f. Table 1). Large fragments should therefore be resolvable in the HST
485
+ data (the resolution is ∼0.08′′) but, nevertheless, remain unseen. This might reflect the
486
+ continued disintegration of the fragments, again aided by the new exposure to the heat of
487
+ the Sun of previously buried volatiles. The breakup process would then be catastrophic.
488
+ Smaller fragments produced by breakup of the primary nucleus would have progressively
489
+ shorter and shorter spin-up times, owing to their smaller size (c.f. Equation 7) and to the
490
+ sudden exposure of large areas of previously buried ice which could amplify the moment
491
+ arm, kT, by orders of magnitude. The expected result is a runaway fragmentation cascade.
492
+ 4.6.
493
+ Gas Production Resulting from Nucleus Disintegration
494
+ Disintegration of the nucleus must suddenly expose previously buried ices to the heat
495
+ of the Sun, leading to a burst in the gas production rate caused by sublimation. Indeed,
496
+
497
+ – 19 –
498
+ measurements of the gas production rate in the mid-December to January period are highly
499
+ variable, peaking near QH2O = 2.4 × 1029 s−1 in radio (Crovisier et al. 2021), Lyman-α (M.
500
+ Combi, (private communication)), and near-ultraviolet (Jehin et al. 2021, 2022a, 2022b)
501
+ observations. At break up, A1 was about rH = 0.8 AU from the Sun and ∆ = 0.2 AU from
502
+ the SWAN/SOHO observatory used to take the Lyman-α data. The latter has 1◦ wide
503
+ pixels, corresponding to about w ∼ 6 × 105 km per pixel at the comet and 1.2×106 km
504
+ for the nominal Nyquist (2 pixel) resolution of the data. With an isothermal blackbody
505
+ temperature at 0.8 AU ∼310 K, the thermal velocity of hydrogen atoms is Vth ∼ 2.5 km
506
+ s−1. This, however, is a strong lower limit to the outflow velocity because of photo-electric
507
+ heating (e.g. Combi and Delsemme 1980, Combi et al. 2000). Based on published models,
508
+ we adopt a hydrogen outflow speed Vth ∼ 10 km s−1 and estimate the residence time for
509
+ hydrogen atoms within a Nyquist sampled resolution element as tr ∼ 2w/Vth ∼ 1.2 × 105
510
+ s (about 1.4 days). This means that the peak rate inferred from SWAN/SOHO Lyman-α
511
+ data should be understood as a measure of the production rate averaged over 1.4 days.
512
+ We are interested to see how QH2O compares with estimates of the gas production
513
+ expected from the break up of the nucleus. To this end, we consider an idealized model in
514
+ which the nucleus consists of particles which are either refractory or ice, and in which the
515
+ ratio of ice to refractory masses is fice. Both refractory and ice particles are assumed to
516
+ occupy a differential power size distribution (Equation 4). To render the problem tractable,
517
+ we make the simplifying assumption that the nucleus disintegrates instantaneously into
518
+ power law distributions of ice and refractory particles, each having radii in the range
519
+ a1 ≤ a ≤ a2. The icy component then sublimates at the rate fs [kg m−2 s−1], which we
520
+ calculate from energy balance including terms for radiation and sublimation.
521
+ In the residence time tr, an ice surface will sublimate over a layer thickness
522
+
523
+ – 20 –
524
+ as = fstr
525
+ ρn
526
+ ,
527
+ (8)
528
+ where ρn is the density of the particle, assumed equal to the bulk density of the nucleus.
529
+ All the ice particles with radii a ≤ as will sublimate away, releasing water molecules and,
530
+ eventually, producing by photodissociation the hydrogen atoms detected using the SWAN
531
+ instrument. Ice particles with a > as will also partially sublimate in time tr, but their
532
+ contribution to the gas flux should be small because, for plausible power law distributions
533
+ (in particular, for γ = 3.5 as determined in sections 4.3 and 4.4), large particles present a
534
+ small fraction of the total particle cross-section.
535
+ The fraction of the mass contained in ice particles having a ≤ as is given by
536
+ F =
537
+ � as
538
+ a1 a3−γda
539
+ � a2
540
+ a1 a3−γda
541
+ (9)
542
+ which, for 3 < γ < 4 and as ≫ a1 and a2 ≫ a1, simplifies to
543
+ F =
544
+ �γ − 3
545
+ 4 − γ
546
+ � �as
547
+ a2
548
+ �4−γ
549
+ .
550
+ (10)
551
+ The total ice mass in the undisrupted nucleus, assumed to be spherical, is Mi =
552
+ (4π/3)ρnr3
553
+ nfice. The production rate averaged over time tr may be written QH2O =
554
+ FMi/(trµmH), where µ = 18 is the molecular weight of the water molecule and
555
+ mH = 1.67 × 10−27 kg is the mass of the hydrogen atom. Substitution of Equations 8 and
556
+ 10 into this expression gives
557
+ QH2O = 4πρnr3
558
+ nfice
559
+ 3trµmH
560
+ �γ − 3
561
+ 4 − γ
562
+ � � fstr
563
+ ρna2
564
+ �4−γ
565
+ .
566
+ (11)
567
+ The equilibrium sublimation mass flux calculated for a blackbody water ice sphere at 0.8
568
+
569
+ – 21 –
570
+ AU is fs = 1.7 × 10−4 kg m−2 s−1. The flux could be smaller if the grain albedo is high, or
571
+ larger if the grain is anisothermal (albeit then sublimating from a smaller fraction of the
572
+ grain surface). We set a2 = 60 m, the largest “particle” allowed by the Hubble imaging,
573
+ and a1 = 10−7 m (however, Equation 11 is insensitive to a1 and its value is unimportant
574
+ provided a1 ≪ as). The nominal nucleus radius is rn = 600 m, and the size distribution
575
+ index is γ = 3.5, as deduced above. Measured cometary ice/refractory ratios, fice, show a
576
+ wide range of values, from fice ∼ 1 in 67P/Churyumov-Gerasimenko (Marschall et al. 2020),
577
+ to fice < 0.2 in C/1995 O1 Hale-Bopp (Jewitt and Matthews 1999) and fice = 0.03 to
578
+ 0.1 in 2P/Encke (Reach et al. 2000). We adopt fice = 1/4, recognizing that this value is
579
+ substantially uncertain.
580
+ Substitution into Equation 11 gives QH2O = 8.8+12.0
581
+ −6.2 × 1029 s−1, where the error bars
582
+ reflect only the ±200 m uncertainty in the estimated radius of the nucleus. This is larger
583
+ than the measured peak water production rate (2×1029 s−1) but shows acceptable agreement
584
+ given the crude nature of the model calculation and the likelihood that the disintegration
585
+ was in reality spread over a finite period not impulsive, as modeled. We conclude that
586
+ complete disintegration of the nucleus into a power law particle size distribution is consistent
587
+ both with the optical brightness of the debris cloud and with the surge in the water
588
+ production rate measured using Lyman-α.
589
+ Future improvements to this model could include a treatment of the initial, optically-
590
+ thick phase of the expanding disintegration cloud, when self-shielding will suppress and
591
+ delay the sublimation surge relative to the estimate given here. Also needed is a treatment
592
+ of the gas drag interaction with cometary solids in a fully disintegrated body, responsible
593
+ for the size-dependent acceleration of refractory particles into the coma and surviving
594
+ debris field. Furthermore, several of the parameters needed to accurately model nucleus
595
+ disintegration remain unmeasured, and most other disintegrating comets are observationally
596
+
597
+ – 22 –
598
+ even less-well characterized than A1. It is obvious, even from these simple considerations
599
+ that many more detailed observations, across a wide range of wavelengths and with
600
+ adequate temporal sampling, will be needed to better understand what is likely to be the
601
+ dominant destructive cometary process.
602
+
603
+ – 23 –
604
+ 5.
605
+ SUMMARY
606
+ We present both high resolution and wide field observations of disintegrating
607
+ long-period comet C/2021 A1 (Leonard) taken to study the nature of its demise.
608
+ • The pre-disintegration radius of the nucleus, estimated using two methods, was
609
+ rn = 0.6 ± 0.2 km. After breakup, which began in mid-December 2021 and may have
610
+ continued for weeks, no nucleus fragments larger than about rn = 0.06 km (i.e. < 10−3
611
+ of the primary mass) survived.
612
+ • The observed debris cloud consists of sub-millimeter and larger particles, with a
613
+ differential power law size distribution having index γ = 3.4±0.1 and 3.5±0.1, as
614
+ estimated by two different methods. The observational constraints are consistent with
615
+ equality between the mass of the debris cloud and the mass of the primary nucleus,
616
+ indicating a total disintegration.
617
+ • Tidal disruption, sublimation, collisional disruption, and explosion following internal
618
+ pressure build-up in the nucleus all offer implausible explanations of the disintegration
619
+ of C/2021 A1.
620
+ • The spin-up timescale due to outgassing torques for a 600 m nucleus in the orbit of
621
+ C/2021 A1 is as short as ∼2 months, pointing to rotational instability as the likely
622
+ cause of the disintegration.
623
+ • A simple model of the exposure and rapid sublimation of previously buried ice
624
+ indicates a peak gas production rate (QH2O = 9+12
625
+ −6 × 1029 s−1) of the same order as
626
+ the measured peak value (QH2O = 2.4 × 1029 s−1).
627
+ We thank Michael Combi for a preview of his SWAN data on C/2021 A1 and the
628
+ anonymous referee for prompt comments on the manuscript. Based on observations made
629
+
630
+ – 24 –
631
+ with the NASA/ESA Hubble Space Telescope, obtained from the data archive at the
632
+ Space Telescope Science Institute. STScI is operated by the Association of Universities for
633
+ Research in Astronomy, Inc. under NASA contract NAS 5-26555. Support for this work
634
+ was provided by NASA through grant number GO-16929 from the Space Telescope Science
635
+ Institute, which is operated by auRA, Inc., under NASA contract NAS 5-26555.
636
+ Facilities: HST.
637
+
638
+ – 25 –
639
+ REFERENCES
640
+ Bessell, M. S. 1990, PASP, 102, 1181. doi:10.1086/132749
641
+ Bohren, C. F. & Huffman, D. R. 1983, Absorption and scattering of light by small particles.
642
+ New York: Wiley, 1983
643
+ Bottke, W. F., Durda, D. D., Nesvorn´y, D., et al. 2005, Icarus, 179, 63.
644
+ doi:10.1016/j.icarus.2005.05.017
645
+ Blaauw, R. C., Campbell-Brown, M. D., & Weryk, R. J. 2011, MNRAS, 414, 3322.
646
+ doi:10.1111/j.1365-2966.2011.18633.x
647
+ Combi, M. R. & Delsemme, A. H. 1980, ApJ, 237, 633. doi:10.1086/157909
648
+ Combi, M. R., Reinard, A. A., Bertaux, J.-L., et al. 2000, Icarus, 144, 191.
649
+ doi:10.1006/icar.1999.6335
650
+ Crovisier, J., Biver, N., and Bockelee-Morvan, D. 2021, Central Bureau Electronic Telegram
651
+ 5087 (2021 December 22)
652
+ Finson, M. J. & Probstein, R. F. 1968, ApJ, 154, 327. doi:10.1086/149761
653
+ Garcia-Trinanes, P., Luding, S. and Shi, H.
654
+ 2019. Advanced Powder Technology, 30,
655
+ 2868-2880
656
+ Groussin, O., Attree, N., Brouet, Y., et al. 2019, Space Sci. Rev., 215, 29. doi:10.1007/s11214-
657
+ 019-0594-x
658
+ Henke, S., Gail, H.-P., Trieloff, M., et al. 2012, A&A, 537, A45. doi:10.1051/0004-
659
+ 6361/201117177
660
+ Holmberg, J., Flynn, C., & Portinari, L. 2006, MNRAS, 367, 449. doi:10.1111/j.1365-
661
+ 2966.2005.09832.x
662
+
663
+ – 26 –
664
+ Ishiguro, M., Sarugaku, Y., Ueno, M., et al. 2007, Icarus, 189, 169.
665
+ doi:10.1016/j.icarus.2007.01.003
666
+ Jehin, E., Moulane, Y., & Manfroid, J. 2021, The Astronomer’s Telegram, 15128
667
+ Jehin, E., Moulane, Y., Manfroid, J., et al. 2022a, The Astronomer’s Telegram, 15186
668
+ Jehin, E., Moulane, Y., Manfroid, J., et al. 2022b, The Astronomer’s Telegram, 15189
669
+ Jewitt, D. & Matthews, H. 1999, AJ, 117, 1056. doi:10.1086/300743
670
+ Jewitt, D. & Luu, J. 2019, ApJ, 883, L28. doi:10.3847/2041-8213/ab4135
671
+ Jewitt, D., Kim, Y., Mutchler, M., et al. 2020, ApJ, 896, L39. doi:10.3847/2041-8213/ab99cb
672
+ Jewitt, D. 2021, AJ, 161, 261. doi:10.3847/1538-3881/abf09c
673
+ Jewitt, D. 2022, AJ, 164, 158. doi:10.3847/1538-3881/ac886d
674
+ Kim, Y., Ishiguro, M., Michikami, T., et al. 2017, AJ, 153, 228. doi:10.3847/1538-
675
+ 3881/aa69bb
676
+ Knight, M. M., A’Hearn, M. F., Biesecker, D. A., et al. 2010, AJ, 139, 926. doi:10.1088/0004-
677
+ 6256/139/3/926
678
+ Leonard, G. J., Aschi, S., Pettarin, E., et al. 2021, Minor Planet Electronic Circulars,
679
+ 2021-A99
680
+ Li, J. & Jewitt, D. 2015, AJ, 149, 133. doi:10.1088/0004-6256/149/4/133
681
+ Marschall, R., Markkanen, J., Gerig, S.-B., et al. 2020, Frontiers in Physics, 8, 227.
682
+ doi:10.3389/fphy.2020.00227
683
+ Pansecchi, L. & Fulle, M. 1990, A&A, 239, 369
684
+
685
+ – 27 –
686
+ Reach, W. T., Sykes, M. V., Lien, D., et al. 2000, Icarus, 148, 80. doi:10.1006/icar.2000.6478
687
+ Samarasinha, N. H. 2001, Icarus, 154, 540. doi:10.1006/icar.2001.6685
688
+ Wolf, C., Onken, C. A., Luvaul, L. C., et al. 2018, PASA, 35, e010. doi:10.1017/pasa.2018.5
689
+ Zhang, Q., Ye, Q., Vissapragada, S., et al. 2021, AJ, 162, 194. doi:10.3847/1538-3881/ac19ba
690
+ Zubko, E., Videen, G. Shkuratov, Y., et al. 2017, JQSRT, 202, 104.
691
+ doi:10.1016/j.jqsrt.2017.07.026
692
+ This manuscript was prepared with the AAS LATEX macros v5.2.
693
+
694
+ – 28 –
695
+ Table 1.
696
+ Observing Geometry
697
+ UT Date & Time
698
+ νa
699
+ rH b
700
+ ∆c
701
+ αd
702
+ θ−⊙e
703
+ θ−V f
704
+ δ⊕g
705
+ Telh
706
+ Scalei
707
+ Uncj
708
+ 2022 Mar 31 18:14-18:26
709
+ 107.4
710
+ 1.756
711
+ 1.942
712
+ 30.8
713
+ 243.6
714
+ 90.2
715
+ -20.4
716
+ Swan Hill
717
+ 1408
718
+ ±1.4
719
+ 2022 Apr 05 23:35-24:04
720
+ 109.2
721
+ 1.833
722
+ 1.910
723
+ 30.9
724
+ 246.4
725
+ 91.7
726
+ -19.8
727
+ HST
728
+ 1385
729
+ ±1.6
730
+ 2022 Apr 06 23:32-23:51
731
+ 109.5
732
+ 1.848
733
+ 1.902
734
+ 30.9
735
+ 246.9
736
+ 92.0
737
+ -19.7
738
+ HST
739
+ 1379
740
+ ±1.6
741
+ 2022 Apr 10 19:23-19:53
742
+ 110.7
743
+ 1.903
744
+ 1.875
745
+ 30.7
746
+ 249.0
747
+ 93.2
748
+ -19.1
749
+ HST
750
+ 1359
751
+ ±1.7
752
+ 2022 Jun 7 16:36-17:12
753
+ 123.0
754
+ 2.698
755
+ 1.715
756
+ 6.8
757
+ 319.3
758
+ 137.0
759
+ +0.2
760
+ HST
761
+ 1243
762
+ ±3.7
763
+ aTrue anomaly, in degrees
764
+ bHeliocentric distance, in au
765
+ cGeocentric distance, in au
766
+ dPhase angle, in degrees
767
+ ePosition angle of projected anti-solar direction, in degrees
768
+ fPosition angle of negative heliocentric velocity vector, in degrees
769
+ gAngle from orbital plane, in degrees
770
+ hTelescope
771
+ iImage scale, km arcsecond−1
772
+ j3σ ephemeris uncertainty, arcsecond (from JPL Horizons)
773
+
774
+ – 29 –
775
+ Fig. 1.— A) Composite of four, 450 s HST images from UT 2022 April 5. Diffuse streaks
776
+ are imperfectly removed field stars and galaxies.
777
+ B) Same image, anotated to show the
778
+ approximate boundary of the debris (white dashed line) and the expected location of the
779
+ nucleus (yellow line segments). Two scale bars of 30′′ and 5×104 km in length are shown, as
780
+ well as the projected anti-solar (−S) and negative heliocentric velocity (−V ) vectors. North
781
+ is to the top, East to the Left.
782
+
783
+ UT 2022 April 5
784
+ 30
785
+ 5x104 km
786
+ B– 30 –
787
+ Fig. 2.— Wide field image from Swan Hill Observatory showing C/2021 A1 on UT 2022
788
+ March 31. 10′ and 106 km scale bars are shown, as well as the projected anti-solar (−S) and
789
+ negative heliocentric velocity (−V ) vectors. Yellow lines mark the ephemeris location of the
790
+ nucleus. The white square shows the size of the HST field of view. The image has North to
791
+ the top, East to the left.
792
+
793
+ 10
794
+ .UT 2022 March 31
795
+ 106.km– 31 –
796
+ Fig. 3.— (Upper:) Same image as in Figure 2 but rotated to bring the axis of the dust tail
797
+ to the horizontal and shown at a larger scale. Yellow lines mark the ephemeris location of
798
+ the nucleus. (Lower:) Locations of the photometry regions A, B and C used to measure the
799
+ scattering cross-section of particles in the tail.
800
+
801
+ R
802
+ 380″
803
+ 1105"– 32 –
804
+ -500
805
+ 0
806
+ 500
807
+ 1000
808
+ 1500
809
+ 0
810
+ 400
811
+ 800
812
+ 1200
813
+ 1600
814
+ Surface Brightness
815
+ Distance [arcsecond]
816
+ Fig. 4.— Surface brightness profile parallel to the long axis of Box A (Figure 3) plotted
817
+ against the distance from the nucleus ephemeris location (axis is reversed relative to Figure
818
+ 3). 1000 units correspond to a surface brightness Σ = 24.4 magnitudes arcsec−2. The linear
819
+ distance scale is approximately 1500 km per arcsecond.
820
+
821
+ – 33 –
822
+ Fig. 5.— (Left:) Same image as Figure 2 with synchrones overplotted, for ejection dates 80,
823
+ 100, 120, 140 and 160 days prior to the date of the image. (Right:) Syndynes for particles
824
+ with β = 0.0003, 0.001, 0.003, 0.01 and 0.03, as marked. The axis of the debris cloud is best
825
+ matched by the 110±10 day synchrones, corresponding to ejection on UT 2021 December
826
+ 11±10.
827
+
828
+ UT 2022 March 31
829
+ 0.0003
830
+ 0.001
831
+ 120
832
+ 0.003
833
+ 100-
834
+ 80
835
+ 10'0
836
+ 0.03– 34 –
837
+ 1011
838
+ 1012
839
+ 1013
840
+ 1014
841
+ 1015
842
+ 2.5
843
+ 3.0
844
+ 3.5
845
+ 4.0
846
+ Debris Mass [kg]
847
+ Differential Size index, γ
848
+ rn = 0.6+/-0.2 km
849
+ Blaauw et al. 2011
850
+ Kreutz Sungrazers
851
+ C/2021 A1
852
+ Fig. 6.— Total mass of the debris cloud (assuming density ρn = 500 kg m−3) plotted as a
853
+ function of the differential power law index, γ, is plotted as a solid black line. The equivalent
854
+ spherical mass of the original 0.6±0.2 km radius nucleus is shown (assuming the same ρn),
855
+ together with its uncertainty, as a yellow horizontal band. The debris and nucleus masses
856
+ are equal at γ = 3.5 ± 0.1, shown by the red filled circle. For comparison we show, as a blue
857
+ square, the size distribution of the Kreutz sungrazing comets (Knight et al. 2010) and, as
858
+ green triangles, several radar-measured meteoroid streams (Blaauw et al. 2011). The vertical
859
+ positions of the Kreutz and radar stream points have no meaning.
860
+
861
+ – 35 –
862
+ -200
863
+ 0
864
+ 200
865
+ 400
866
+ 600
867
+ 800
868
+ 1000
869
+ 1200
870
+ 0
871
+ 400
872
+ 800
873
+ 1200
874
+ 1600
875
+ Surface Brightness
876
+ Distance [arcsecond]
877
+ 3.5
878
+ 3.4
879
+ 3.3
880
+ 3.5
881
+ 3.3
882
+ 3.4
883
+ Fig. 7.— Axial surface brightness profile on UT 2022 March 31 (yellow diamonds) compared
884
+ with results from a Monte Carlo simulation. The models shown have size index γ = 3.3 (red
885
+ curve), 3.4 (black curve) and 3.5 (blue curve), all with 7 × 10−4 ≤ β ≤ 0.07, corresponding
886
+ to particle radii 14 µm to 1.4 mm.
887
+
JNFAT4oBgHgl3EQfvB44/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
M9AzT4oBgHgl3EQfWPwO/content/tmp_files/2301.01296v1.pdf.txt ADDED
@@ -0,0 +1,1618 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ TinyMIM: An Empirical Study of Distilling MIM Pre-trained Models
2
+ Sucheng Ren
3
+ Fangyun Wei*
4
+ Zheng Zhang
5
+ Han Hu
6
+ Microsoft Research Asia
7
+ Abstract
8
+ Masked image modeling (MIM) performs strongly in pre-
9
+ training large vision Transformers (ViTs). However, small
10
+ models that are critical for real-world applications can-
11
+ not or only marginally benefit from this pre-training ap-
12
+ proach. In this paper, we explore distillation techniques to
13
+ transfer the success of large MIM-based pre-trained mod-
14
+ els to smaller ones. We systematically study different op-
15
+ tions in the distillation framework, including distilling tar-
16
+ gets, losses, input, network regularization, sequential dis-
17
+ tillation, etc, revealing that: 1) Distilling token relations
18
+ is more effective than CLS token- and feature-based distil-
19
+ lation; 2) An intermediate layer of the teacher network as
20
+ target perform better than that using the last layer when
21
+ the depth of the student mismatches that of the teacher;
22
+ 3) Weak regularization is preferred; etc. With these find-
23
+ ings, we achieve significant fine-tuning accuracy improve-
24
+ ments over the scratch MIM pre-training on ImageNet-1K
25
+ classification, using all the ViT-Tiny, ViT-Small, and ViT-
26
+ base models, with +4.2%/+2.4%/+1.4% gains, respectively.
27
+ Our TinyMIM model of base size achieves 52.2 mIoU in
28
+ AE20K semantic segmentation, which is +4.1 higher than
29
+ the MAE baseline. Our TinyMIM model of tiny size achieves
30
+ 79.6% top-1 accuracy on ImageNet-1K image classifica-
31
+ tion, which sets a new record for small vision models of
32
+ the same size and computation budget. This strong perfor-
33
+ mance suggests an alternative way for developing small
34
+ vision Transformer models, that is, by exploring better train-
35
+ ing methods rather than introducing inductive biases into
36
+ architectures as in most previous works. Code is available
37
+ at https://github.com/OliverRensu/TinyMIM.
38
+ 1. Introduction
39
+ Masked image modeling (MIM), which masks a large
40
+ portion of the image area and trains a network to recover
41
+ the original signals for the masked area, has proven to be a
42
+ very effective self-supervised method for pre-training vision
43
+ Transformers [2,12,18,53]. Thanks to its strong fine-tuning
44
+ performance, MIM has now been a main-stream pre-training
45
+ *Corresponding author: fawe@microsoft.com.
46
+ ViT-T
47
+ ViT-S
48
+ ViT-B
49
+ 70
50
+ 74
51
+ 78
52
+ 82
53
+ 86
54
+ Scratch
55
+ MAE
56
+ TinyMIM
57
+ -0.6
58
+ +3.6
59
+ +0.7
60
+ +3.1
61
+ +2.4
62
+ +3.8
63
+ Acc.
64
+ 72.2
65
+ 79.9
66
+ 81.2
67
+ Figure 1. Comparison among TinyMIM (ours), MAE [18] and
68
+ training from scratch by using ViT-T, -S and -B on ImageNet-1K.
69
+ We report top-1 accuracy. We adopt DeiT [44] when training from
70
+ scratch. For the first time, we successfully perform masked image
71
+ modeling pre-training for smaller ViTs.
72
+ Model
73
+ Param.
74
+ Flops
75
+ Top-1
76
+ mIoU
77
+ (M)
78
+ (G)
79
+ (%)
80
+ DeiT-T [44]
81
+ 5.5
82
+ 1.3
83
+ 72.2
84
+ 38.0
85
+ PVT-T [46]
86
+ 13.0
87
+ 1.9
88
+ 75.1
89
+ 39.8
90
+ CiT-T [39]
91
+ 5.5
92
+ 1.3
93
+ 75.3
94
+ 38.5
95
+ Swin [32]
96
+ 8.8
97
+ 1.2
98
+ 76.9
99
+ 40.4
100
+ EdgeViT-XS [35]
101
+ 6.4
102
+ 1.1
103
+ 77.5
104
+ 42.1
105
+ MobileViTv1-S [34]
106
+ 4.9
107
+ 2.0
108
+ 78.4
109
+ 42.7
110
+ MobileViTv3-S [45]
111
+ 4.8
112
+ 1.8
113
+ 79.3
114
+ 43.1
115
+ TinyMIM⋆-T (Ours)
116
+ 5.8
117
+ 1.3
118
+ 79.6
119
+ 45.0
120
+ Table 1. Comparison with state-of-the-art tiny Transformers with
121
+ architecture variants. The parameters indicate the backbone pa-
122
+ rameter excluding the parameters of the last classification layer
123
+ in classification or the decoder in segmentation. We report top-1
124
+ accuracy on ImageNet-1K classification and mIoU on ADE20K
125
+ segmentation.
126
+ method for vision Transformers, and numerous follow-ups
127
+ have been carried out in this research line, such as study-
128
+ ing how to set decoding architectures [25], reconstruction
129
+ targets [11,36,48,60], etc., as well as revealing its proper-
130
+ ties [49,52,54].
131
+ 1
132
+ arXiv:2301.01296v1 [cs.CV] 3 Jan 2023
133
+
134
+ Method
135
+ ViT-T
136
+ ViT-S
137
+ ViT-B
138
+ ViT-L
139
+ Scratch
140
+ 72.2
141
+ 79.9
142
+ 81.2
143
+ 82.6
144
+ MAE
145
+ 71.6
146
+ 80.6
147
+ 83.6
148
+ 85.9
149
+ Gap
150
+ -0.6
151
+ +0.7
152
+ +2.4
153
+ +3.3
154
+ Table 2. Comparison between MAE pre-trained ViTs and ViTs
155
+ trained from scratch by using ViT-T, -S, -B and -L on ImageNet-
156
+ 1K. We adopt DeiT when training from scratch. We report top-1
157
+ accuracy. As model size shrinks, the superiority of MAE gradually
158
+ vanishes. MAE even hurts the performance of ViT-T.
159
+ However, as shown in Table 2, MIM pre-training [18]
160
+ mainly effects for relatively large models. When the model
161
+ size is as small as ViT-Tiny (5 million parameters), which
162
+ is critical for real-world applications, MIM pre-training can
163
+ even hurt the fine-tuning accuracy on ImageNet-1K classifi-
164
+ cation. In fact, the accuracy drops by -0.6 compared to the
165
+ counterpart trained from scratch. This raises a question: can
166
+ small models also benefit from MIM pre-training, and how
167
+ can this be achieved?
168
+ In addition, the existing study on small vision Transform-
169
+ ers mainly focus on introducing certain inductive bias into
170
+ architecture design [6,26,34,35]. The additional architec-
171
+ tural inductive biases facilitate optimization yet limit the
172
+ expressive capacity. It’s natural to ask whether we can boost
173
+ plain small vision Transformers to perform just as well.
174
+ In this work, we present TinyMIM, which answers the
175
+ above questions. Instead of directly training small ViT mod-
176
+ els using a MIM pretext task, TinyMIM uses distillation
177
+ technology [24] to transfer the knowledge of larger MIM
178
+ pre-trained models to smaller ones. Distillation endows the
179
+ nice properties of larger MIM pre-trained models to smaller
180
+ ones while avoiding solving a “too” difficult MIM task. Not-
181
+ ing that knowledge distillation has been well developed,
182
+ especially for supervised models [16], our main work is to
183
+ systematically study for the first time the effects of different
184
+ design options in a distillation framework when using MIM
185
+ pre-trained models as teachers. Specifically, we consider dis-
186
+ tillation targets, data augmentation, network regularization,
187
+ auxiliary losses, macro distillation strategy, etc., and draw
188
+ several useful findings:
189
+ • Distillation targets. There are two main findings re-
190
+ lated to distillation targets: 1) Distilling token relations
191
+ is more effective than distilling the CLS token and fea-
192
+ ture maps. 2) Using intermediate layers as the target
193
+ may perform better than using the last layer, and the
194
+ optimal target layer for different down-stream tasks,
195
+ e.g., classification and segmentation, can be different.
196
+ • Data and network regularization. Weak augmentation
197
+ and regularization is preferred: 1) The performance of
198
+ using a masked image is worse than using the original
199
+ image; 2) Relatively small drop path rate (0 for teacher
200
+ and 0.1 for student) performs best.
201
+ • auxiliary losses. We find that an auxiliary MIM loss
202
+ does not improve fine-tuning accuracy.
203
+ • Macro distillation strategy. We find that using a se-
204
+ quential distillation strategy, i.e., “ViT-B → ViT-S →
205
+ ViT-T”, performs better than that distilling directly from
206
+ ViT-B to ViT-T.
207
+ By selecting the best framework options, we achieve sig-
208
+ nificant fine-tuning accuracy improvements over the direct
209
+ MIM pre-training on ImageNet-1K classification, using ViT
210
+ models of different sizes, as shown in Figure 1. Specifi-
211
+ cally, the gains of TinyMIM on the ViT-Tiny, ViT-Small, and
212
+ ViT-base models are +4.2%/+2.4%/+1.4%, respectively.
213
+ In particular, our TinyMIM⋆-T model with knowledge
214
+ distillation during finetune-tuning achieves a top-1 accuracy
215
+ of 79.6% on ImageNet-1K classification (see Table 1), which
216
+ performs better than all previous works that develop small
217
+ vision Transformer models by introducing architectural in-
218
+ ductive biases or smaller feature resolutions. It sets a new
219
+ accuracy record using similar model size and computation
220
+ budget. On ADE20K semantic segmentation, TinyMIM-T
221
+ achieves 45.0 mIoU, which is +1.9 higher than the second
222
+ best method, MobileViTv3-S [45]. The strong fine-tuning
223
+ accuracy by TinyMIM⋆-T suggests an alternative way for
224
+ developing small vision Transformer models, that is, by
225
+ exploring better training methods rather than introducing
226
+ inductive biases into architectures as most previous works
227
+ have done.
228
+ 2. Related Works
229
+ 2.1. Masked Image Modeling
230
+ Masked Language Modeling (MLM) [10] for self-
231
+ supervised Transformer pre-training has achieved incredible
232
+ success in natural language processing (NLP) field. Inspired
233
+ by the same idea of masking and reconstruction, BEiT [2]
234
+ is the pioneer to bring such success to computer vision filed
235
+ by encoding masked images and predicting masked tokens
236
+ generated by DALL-E [38]. SimMIM [53] and MAE [18]
237
+ find that reconstructing RGB pixels results in favorable rep-
238
+ resentations. MAE adopts an asymmetric encoder-decoder
239
+ architecture. The encoder only encodes the visible tokens
240
+ and drops a high portion of masked tokens to reduce the com-
241
+ putation burden. A lightweight decoder then produces recon-
242
+ structed patches. Different from tokens in natural language
243
+ processing that have rich semantics, pixels in computer vi-
244
+ sion are low-level information, therefore, a lot of recent
245
+ works aim at looking for better supervisions. MaskFeat [48]
246
+ takes local gradient features produced by the manually-
247
+ crafted HOG descriptor [9] as supervisions. PeCo [11] trains
248
+ 2
249
+
250
+ Masked Image
251
+ Raw Image
252
+ Factors
253
+ Input
254
+ Target
255
+ Feature
256
+ Relation
257
+ 𝑄·𝑄𝑇
258
+ 𝐾·𝐾𝑇
259
+ 𝑉·𝑉𝑇
260
+ 𝑄·𝐾𝑇
261
+ Head Number
262
+ w/ or w/o Softmax
263
+ Output Feature
264
+ Block Feature
265
+ QKV Features
266
+ Attention Feature
267
+ FFN Feature
268
+ Res. Connection of FFN
269
+ Block
270
+ Last
271
+ Intermediate
272
+
273
+
274
+ Transformer Block-N
275
+ Output Feature
276
+ Multi-Head
277
+ Attention
278
+ Add & Norm
279
+ FFN
280
+ Add & Norm
281
+ Attention Feature
282
+ FFN Feature
283
+ Block Feature
284
+ Raw Image
285
+ Masked Image
286
+ Feature of Last Block
287
+ 𝑄·𝑄𝑇
288
+ 𝐾·𝐾𝑇
289
+ 𝑉·𝑉𝑇
290
+ 𝑄·𝐾𝑇
291
+ 𝑄
292
+ 𝐾
293
+ 𝑉
294
+ Softmax
295
+ Transformer Block-n
296
+ Transformer Block-1
297
+ Teacher
298
+ (Highlight
299
+ by Blue)
300
+ Relations
301
+ Figure 2. We comprehensively study a variety of factors (highlighted by Royal Blue) that may affect TinyMIM pre-training including input,
302
+ distillation target (feature or relation) and target block.
303
+ a new tokenizer by enforcing perceptual similarity. iBot [60]
304
+ and data2vec [1] take exponential moving average (EMA)
305
+ updated models as tokenizers. MILAN [25] adopts a pre-
306
+ trained CLIP as the teacher. Similarly, BeiTv2 [36] also uses
307
+ CLIP [37] for tokenizer training. Different from these works
308
+ that use various tokenizers/teachers, we adopt a masked im-
309
+ age modeling pre-trained model as our teacher.
310
+ The MIM pre-training performs very well on relatively
311
+ large models from base size to giant size [31,53]. However,
312
+ it will hurt the fine-tuning when the model is as small as
313
+ tiny size, probably because the limited capthe MIM task is
314
+ “too” difficult for small model. This paper explores how to
315
+ make small vision Transformer models also benefit from
316
+ MIM training, through a systematic study of the distillation
317
+ technology.
318
+ 2.2. Knowledge Distillation
319
+ Knowledge distillation is a classical method to transfer
320
+ the knowledge from cumbersome models to a small one, pi-
321
+ oneered by [24]. The original knowledge distillation frame-
322
+ work adopts the annealed classification logits of the teacher
323
+ as the distilling target for the student. Since then, extensive
324
+ variants have been carried out to improve the distilling ef-
325
+ fectiveness [16], including changing the distilling targets as
326
+ intermediate features [22,23,28,40] and relations [29,56],
327
+ data augmentations of teacher and students [39, 50], regu-
328
+ larization [50], distilling strategies [47, 55, 57, 58] and so
329
+ on.
330
+ While almost all studies are made for CNN architec-
331
+ tures under supervised settings, recently, there have been
332
+ a few works performing distilling technologies for vision
333
+ Transformers [44,50] and contrastive learning based meth-
334
+ ods [14, 50]. In DeiT [44], the teacher is set as a CNN
335
+ architecture so as to transfer the inductive bias involved in
336
+ CNNs to vision Transformers. It also propose to use hard
337
+ distillation which uses hard pseudo class labels of the teacher
338
+ network as the distilling targets, which performs better than
339
+ the naive knowledge distillation [24]. In [14], a distillation
340
+ method regarding the similarities between instances is ap-
341
+ plied to transfer the power of contrastive pre-trained large
342
+ CNN models to small CNNs. In [50], a method based on
343
+ feature map distillation is proposed to generally improve
344
+ vision transformers by different pre-training approaches in-
345
+ cluding image classification, instance contrastive based self-
346
+ sueprvised learning [3] and CLIP pre-training [37]. However,
347
+ it shows no gains for MIM pre-trained models.
348
+ This paper for the first time studies the distillation frame-
349
+ work for MIM pre-trained vision Transformers. Through
350
+ a systematic study, it draws several useful findings and the
351
+ best options, under which, significant gains are achieved for
352
+ vision Transformers of various sizes.
353
+ 2.3. Small Vision Transformers
354
+ Designing efficient CNN models [27,42] has been widely
355
+ studied in recent years.
356
+ With the emergence of Vision
357
+ Transformer (ViT), there have been several works study-
358
+ ing how to develop efficient vision Transformer, with the
359
+ majority focus on introduing inductive biases into the archi-
360
+ tectures [17,26,30,34,35].
361
+ Different from these works that develop small vision
362
+ Transformers by introducing sophisticated components into
363
+ architectures, we demonstrate that a plain vision Trans-
364
+ former [12] at a small scale can perform just as well, or
365
+ even better. Our main insight is that the MIM pre-training
366
+ can implicitly incorporate necessary inductive biases, and
367
+ thus avoids the need of explicit architecture bias. Our plain
368
+ 3
369
+
370
+ vision Transformer of tiny size achieves the state-of-the-art
371
+ accuracy for both ImageNet-1K image classification and
372
+ ADE20K semantic segmentation using similar model size
373
+ and computation budget.
374
+ 3. TinyMIM
375
+ We adopt a larger, MIM pre-trained model as the teacher,
376
+ and a smaller ViT as the student. The objective of TinyMIM
377
+ is to train the randomly initialized student by mimicking the
378
+ target produced by the teacher in a knowledge distillation
379
+ manner. After pre-training, the TinyMIM pre-trained model
380
+ can be transferred to various downstream tasks. In this work,
381
+ we adopt MAE [18] as the MIM model due to its popularity
382
+ and simplicity.
383
+ In this section, we first describe the factors that may affect
384
+ TinyMIM pre-training: distillation target in Section 3.1.1;
385
+ input in Section 3.1.2; target block in Section 3.1.3. Then we
386
+ present a series of distillation losses for different distillation
387
+ target in Section 3.1.3. At last, a sequential distillation strat-
388
+ egy is introduced to facilitate the performance in Section 3.3.
389
+ 3.1. Factors
390
+ 3.1.1
391
+ Distillation Target
392
+ Block Feature and Output Feature. Given an input image
393
+ x, we first divide it into N non-overlapping patches and use
394
+ a linear projection layer to map N patches into patch em-
395
+ beddings F0 ∈ RN×D, where D is the dimension of hidden
396
+ features. Suppose we have a ViT containing L Transformer
397
+ blocks. Each Transformer block takes the output Fi−1 of the
398
+ last Transformer block as the input and generates the feature
399
+ Fi of the current block, which can be formulated as:
400
+ Fi = Transformer(Fi−1), i ∈ [1, L].
401
+ (1)
402
+ We term Fi as the block feature of the i-th Transformer
403
+ block. In particular, we name the feature FL from the last
404
+ Transformer block as the output feature.
405
+ Attention Feature and FFN Feature. Each Transformer
406
+ block is composed of a self-attention layer and a feed for-
407
+ ward layer, which can be defined as:
408
+ Hi = Attention(LN(Fi−1)),
409
+ �Hi = Hi + Fi−1,
410
+ �Hi = FFN(LN( �Hi)),
411
+ F i = �Hi + �Hi,
412
+ (2)
413
+ where Attention(·), FFN(·) and LN(·) denotes self-
414
+ attention layer, feed forward layer and layer norm, respec-
415
+ tively. We term �Hi and �Hi as attention feature and FFN
416
+ feature of the i-th Transformer block.
417
+ Query/Key/Value Features. Each self-attention layer con-
418
+ sists of M head networks, each of which maps input feature
419
+ Fi−1 to query (Q), key (K) and value (V):
420
+ Qm
421
+ i = LN(Fi−1)W Q
422
+ i ,
423
+ Km
424
+ i
425
+ = LN(Fi−1)W K
426
+ i ,
427
+ V m
428
+ i
429
+ = LN(Fi−1)W V
430
+ i ,
431
+ (3)
432
+ where Qi, Ki, Vi ∈ RN× D
433
+ M represent the query, key and
434
+ value of the m-th head network. The query/key/value fea-
435
+ tures (Qi, Ki, Vi ∈ RN×D) are the concatenation of M
436
+ Qm
437
+ i /Km
438
+ i /V m
439
+ i , respectively.
440
+ Relations. For the m-th head network from the i-th Trans-
441
+ former block, we could calculate its Q-Q, K-K, V-V and
442
+ Q-K relations (RQQ
443
+ i,m, RKK
444
+ i,m , RV V
445
+ i,m, RQK
446
+ i,m ∈ RN×N), which
447
+ are implemented as the scaled product relation:
448
+ RQQ
449
+ i,m = Softmax
450
+
451
+ Qm
452
+ i Qm
453
+ i
454
+ T
455
+
456
+ D/M
457
+
458
+ ,
459
+ RKK
460
+ i,m = Softmax
461
+
462
+ Km
463
+ i Km
464
+ i
465
+ T
466
+
467
+ D/M
468
+
469
+ ,
470
+ RV V
471
+ i,m = Softmax
472
+
473
+ V m
474
+ i V m
475
+ i
476
+ T
477
+
478
+ D/M
479
+
480
+ ,
481
+ RQK
482
+ i,m = Softmax
483
+
484
+ Qm
485
+ i Km
486
+ i
487
+ T
488
+
489
+ D/M
490
+
491
+ .
492
+ (4)
493
+ The Q-Q/K-K/V-V/Q-K relations (RQQ
494
+ i
495
+ , RKK
496
+ i
497
+ , RV V
498
+ i
499
+ ,
500
+ RQK
501
+ i
502
+ ∈ RM×N×N) of the i-th Transformer block is the
503
+ stack of M RQQ
504
+ i,m/RKK
505
+ i,m /RV V
506
+ i,m/RQK
507
+ i,m , respectively.
508
+ 3.1.2
509
+ Input
510
+ MIM models randomly mask a high proportion of image
511
+ patches on an input image x, yielding a masked image �x
512
+ for pre-training. We also investigate the input of TinyMIM
513
+ when performing knowledge distillation— the input could
514
+ be either a raw image x or a masked image �x.
515
+ 3.1.3
516
+ Target Block
517
+ Consider a situation where we tend to use an MAE pre-
518
+ trained ViT-L (teacher) containing 24 blocks to distill a ViT-
519
+ B (student) containing 12 blocks. In this scenario, the block
520
+ number of the student does not match that of the teacher. We
521
+ investigate which block of the teacher can provide the most
522
+ appropriate target. The selected block is referred to as the
523
+ target block.
524
+ 3.2. Knowledge Distillation as MIM Pre-training
525
+ In Section 3.1.1, we describe a variety of distillation target
526
+ candidates. In this section, we introduce different knowledge
527
+ distillation losses for various distillation targets. Let x de-
528
+ note an input image, ft and fs represent a teacher model and
529
+ 4
530
+
531
+
532
+ Teacher
533
+ 𝑉·𝑉𝑇
534
+ 𝑄·𝐾𝑇
535
+ Raw Image
536
+ #Head
537
+ 𝑄·𝐾𝑇
538
+
539
+
540
+ 𝑉·𝑉𝑇
541
+ #Head
542
+ 𝑉·𝑉𝑇
543
+ 𝑄·𝐾𝑇
544
+ #Head
545
+ 𝑄·𝐾𝑇
546
+
547
+
548
+ 𝑉·𝑉𝑇
549
+ #Head
550
+ Block-1
551
+ Block-n
552
+ Block-N
553
+
554
+
555
+ Student
556
+ Block-1
557
+ Block-L (Adaptive Block)
558
+ Loss
559
+
560
+ : Forward
561
+ : Backward
562
+ Figure 3. The default knowledge distillation strategy of TinyMIM. The student (e.g. ViT-B) is optimized to mimic the relations generated by
563
+ the intermediate block of a MIM pre-trained teacher (e.g. ViT-L) with raw image as input. We replace the last block of the student with an
564
+ adaptive block to match teacher’s head number (no extra computational cost). After pre-training (knowledge distillation), the student model
565
+ can be transferred to various downstream tasks.
566
+ a student model, respectively. The objective of knowledge
567
+ distillation is to transfer the knowledge from ft to fs by
568
+ optimizing fs while freezing ft. In general, the training is
569
+ supervised by the KL divergence, which is defined as:
570
+ LKL(p, t) = tlog t
571
+ p,
572
+ (5)
573
+ where t denotes the target generated by ft(x), and p is the
574
+ prediction produced by fs(x).
575
+ Class Token Distillation. We use ct and cs to denote class
576
+ token feature of ft and fs, respectively. The loss of class
577
+ token distillation is formulated as:
578
+ L = LKL(cs, ct).
579
+ (6)
580
+ Feature Distillation. In general, the feature dimension of
581
+ the teacher network and the student network are mismatched.
582
+ To tackle this problem, we adopt an extra linear layer on the
583
+ output of the student network to match the feature dimension
584
+ of the teacher’s target. Let F t and F s denote the target
585
+ feature and the prediction yielded by the student followed by
586
+ a linear projection layer, respectively. We could formulate
587
+ the loss of feature distillation as follows:
588
+ L = L1(F s, Norm(F t)),
589
+ (7)
590
+ where Norm(·) is the whitening operation implemented by
591
+ layer norm without affiliation, and L1 is the smooth L1 loss
592
+ defined as:
593
+ L1(y, ˆy) =
594
+
595
+ 1
596
+ 2(ˆy − y)2/β,
597
+ |ˆy − y| ≤ β
598
+ (|ˆy − y| − 1
599
+ 2β),
600
+ otherwise
601
+ ,
602
+ (8)
603
+ where β is set to 2.0.
604
+ Relation Distillation. This is our default knowledge distilla-
605
+ tion strategy as illustrated in Figure 3. For the sake of clarity,
606
+ we use RQK
607
+ t→m to denote the m-th head generated Q-K rela-
608
+ tion target (see Eq 4) from the teacher network, and RQK
609
+ s→m to
610
+ represent the corresponding Q-K relation prediction from the
611
+ student network. We define RV V
612
+ t→m and RV V
613
+ s→m in a similar
614
+ way. The loss of relation distillation is formulated as:
615
+ LQK = 1
616
+ M
617
+ M
618
+
619
+ m=1
620
+ LKL(RQK
621
+ s→m, RQK
622
+ t→m),
623
+ LV V = 1
624
+ M
625
+ M
626
+
627
+ m=1
628
+ LKL(RV V
629
+ s→m, RV V,S
630
+ t→m ),
631
+ L = LQK + LV V .
632
+ (9)
633
+ Head Alignment for Relation Distillation. In general, the
634
+ head number of the student network is lower than that of the
635
+ teacher network. For instance, ViT-L (teacher) contains 16
636
+ heads per block while ViT-B (student) only contains 12 heads
637
+ per block. Recall that the relation distillation loss (Eq. 9)
638
+ is calculated head by head, thus we have to solve the head
639
+ misalignment issue before performing relation distillation.
640
+ To this end, we replace the last block of the student with an
641
+ adaptive block, which keeps the original hidden dimension
642
+ but adjusts the head number to the teacher. Concretely, given
643
+ a teacher network with Mt heads per block, and a student
644
+ network with Ms heads per block, a hidden dimension of
645
+ Ds, and a head dimension of Ds/Ms, the adaptive block is
646
+ designed to be a Transformer block with Mt heads per block,
647
+ a hidden dimension of Ds and a head dimension of Ds/Mt.
648
+ 3.3. Sequential Distillation
649
+ When training a small model like ViT-S, the teacher has
650
+ two options: a pre-trained ViT-B and a pre-trained ViT-
651
+ L. Intuitively, the pre-trained ViT-L is a good teacher due
652
+ to its higher representation capability. However, there is
653
+ 5
654
+
655
+ a huge capacity gap between ViT-L and ViT-S, resulting
656
+ in poor distillation results. Following [8, 15], we adopt a
657
+ sequential distillation strategy to improve pre-training. For
658
+ instance, when pre-training a ViT-S, the teacher is selected
659
+ as a TinyMIM pre-trained ViT-B, which has been trained by
660
+ TinyMIM with ViT-L as the teacher.
661
+ 4. Experiments
662
+ 4.1. Implementation Details
663
+ Pre-training.
664
+ All models are pre-trained under a 100-
665
+ epoch schedule on ImageNet-1K [41] training set.
666
+ We
667
+ use a batch size of 4096 and a learning rate of lr=1.5e-
668
+ 4×batchsize/256. We adopt a cosine decay schedule with
669
+ a warm-up for 5 epochs. We adopt AdamW [33] optimizer
670
+ with a weight decay of 0.05. We use random resized crop-
671
+ ping random horizontal flipping, color jitter for student only.
672
+ The input size is set to 224 × 224.
673
+ Fine-tuning. We transfer TinyMIM pre-trained models to
674
+ ImageNet [41] image classification and ADE20K [59] se-
675
+ mantic segmentation. For ImageNet, we use AdamW op-
676
+ timizer with weight decay of 0.05. For data augmentation,
677
+ we follow the settings in MAE [18]. We fine-tune ViT-B
678
+ for 100 epochs with a batch size of 1024, a learning rate of
679
+ 2e-3, and a drop path rate of 0.1. We fine-tune ViT-S and
680
+ ViT-T for 200 epochs with a batch size of 2048, a learning
681
+ rate of 5e-3, and a drop path rate of 0.1. For ADE20K, we
682
+ follow the same setting in MAE and adopt UperNet [51]
683
+ as our framework with a TinyMIM pre-trained backbone.
684
+ The input image resolution is 512 × 512 for training and
685
+ evaluating. We use mIoU as the evaluation metric.
686
+ Besides, we evaluate the robustness of TinyMIM on var-
687
+ ious out-of-domain ImageNet datasets [19–21] which are
688
+ generated by applying different perturbations on ImageNet,
689
+ e.g. natural adversarial examples (ImageNet-A), semantic
690
+ shift (ImageNet-R), common image corruptions (ImageNet-
691
+ C). We report top-1 accuracy on ImageNet-A/R and mCE
692
+ error on ImageNet-C (lower is better).
693
+ Default Setting. By default, we adopt relation distillation
694
+ formulated in Eq. 9, head alignment, raw image as input, se-
695
+ quential distillation and the 18-th block of MAE pre-trained
696
+ ViT-L as the target block for TinyMIM-ViT-B pre-training.
697
+ 4.2. Main Results
698
+ As shown in Table 3, we compare our TinyMIM with
699
+ previous methods on ImageNet image classification and
700
+ ADE20K semantic segmentation using different ViTs. In
701
+ particular, TinyMIM pre-trained ViT-T achieves 75.8% top-
702
+ 1 accuracy, outperforming MAE baseline by +4.2.
703
+ An
704
+ enhanced model named TinyMIM⋆-T, which retains the
705
+ plain architecture and computation budget of ViT-T, fur-
706
+ ther achieves 79.6% top-1 accuracy. See appendix for the
707
+ details of TinyMIM⋆-T. Moreover, TinyMIM pre-trained
708
+ ViT-S achieves 83.0% top-1 accuracy, outperforming MAE
709
+ baseline and previous best method CIM [13] by +2.4, +1.4,
710
+ respectively. By transferring the knowledge of an MAE pre-
711
+ trained ViT-L, TinyMIM pre-trained ViT-B achieves 85.0%
712
+ top-1 accuracy on ImageNet-1K.
713
+ As for semantic segmentation, TinyMIM pre-trained ViT-
714
+ B surpasses MAE baseline and state-of-the-art CAE [4] by
715
+ +4.1 and +2.0, respectively. An intermediate fine-tuning on
716
+ ImageNet-1K classification before ADE20K segmentation
717
+ fine-tuning further boosts the performance.
718
+ We also evaluate our models on out-of-domain datasets
719
+ in Table 4. Our TinyMIM pretrained models are more robust
720
+ than MAE pre-trained ones. Specifically, TinyMIM-ViT-B
721
+ outperforms MAE-ViT-B by +6.4 and +4.6 on ImageNet-A
722
+ and ImageNet-R, respectively, and lower the mCE by -5.1.
723
+ 4.3. Ablation Study
724
+ Unless otherwise specified, all ablation studies are con-
725
+ ducted on TinyMIM-ViT-B, with a teacher of being an MAE
726
+ pre-trained ViT-L, relation distillation strategy, raw image as
727
+ input, the 18-th block of ViT-L as the target block, under a
728
+ 100-epoch pre-training schedule. We report top-1 accuracy
729
+ on ImageNet-1K.
730
+ Class Token Distillation. For this distillation strategy, we
731
+ study two variants: 1) class token distillation as formulated
732
+ in Eq.6; 2) class token distillation with an extra MAE re-
733
+ construction loss. The results are shown in Table 5. Both
734
+ variants perform worse than MAE baseline, indicting that
735
+ the class token is improper to be served as the distillation
736
+ target since there is no explicit supervision applied on class
737
+ token during teacher’s pre-training.
738
+ Feature Distillation. As described in Section 3.1.1, there
739
+ are four types of features can be served as the targets for
740
+ feature distillation formulated in Eq. 7: output feature, FFN
741
+ feature, attention feature and Q/K/V features. Table 6 com-
742
+ pares the results of using different features as distillation
743
+ targets. We also report the results of FFN feature and atten-
744
+ tion feature before the residual connection (see Eq. 2). An
745
+ interesting finding is that distilling FFN feature and attention
746
+ feature after the residual connection significantly degrades
747
+ the performance.
748
+ Relation Distillation. Eq. 9 formulates our default relation
749
+ distillation, which jointly distills Q-K relation and V-V re-
750
+ lation (see Eq. 4). Here we study a variant by changing the
751
+ target relations from Q-K/V-V to Q-K/K-K/V-V. We also
752
+ investigate that whether to apply a Softmax operator on each
753
+ relation. The results are shown in Table 7.
754
+ Comparison of Different Distillation Strategies. In this
755
+ study, all models are pre-trained under a 300-epoch schedule.
756
+ We compare three distillation strategies on ImageNet image
757
+ classification (Table 8) and ADE20K semantic segmentation
758
+ (Table 9). For each strategy, we use the target that yields
759
+ the best result. We also highlight the improvements over the
760
+ 6
761
+
762
+ Method
763
+ Pretraining
764
+ Tokenizer/
765
+ Tokenizer/Teacher
766
+ Classification
767
+ Segmentation
768
+ Epochs
769
+ Teacher
770
+ Data
771
+ Top-1 Acc (%)
772
+ mIoU
773
+ Tiny-size models (ViT-T/16)
774
+ Scratch [44]
775
+ 300
776
+ Label
777
+ IN1K
778
+ 72.2
779
+ 38.0
780
+ MAE† [18]
781
+ 1600
782
+ Pixel
783
+ IN1K
784
+ 71.6
785
+ 37.6
786
+ MoCo [5]
787
+ 1600
788
+ EMA
789
+ IN1K
790
+ 73.3
791
+ 39.3
792
+ TinyMIM (Ours)
793
+ 300
794
+ TinyMIM-ViT-S
795
+ IN1K
796
+ 75.8
797
+ 44.0/44.6‡
798
+ TinyMIM⋆ (Ours)
799
+ 300
800
+ TinyMIM-ViT-S
801
+ IN1K
802
+ 79.6
803
+ 45.0‡
804
+ Small-size models (ViT-S/16)
805
+ Scratch [44]
806
+ 300
807
+ Label
808
+ IN1K
809
+ 79.9
810
+ 43.1
811
+ MAE† [18]
812
+ 1600
813
+ Pixel
814
+ IN1K
815
+ 80.6
816
+ 42.8
817
+ MoCo [5]
818
+ 1600
819
+ EMA
820
+ IN1K
821
+ 81.4
822
+ 43.9
823
+ DINO [3]
824
+ 1600
825
+ EMA
826
+ IN1K
827
+ 81.5
828
+ 45.3
829
+ CIM [13]
830
+ 1600
831
+ Pixel
832
+ IN1K
833
+ 81.6
834
+ -
835
+ TinyMIM (Ours)
836
+ 300
837
+ TinyMIM-ViT-B
838
+ IN1K
839
+ 83.0
840
+ 48.4/48.9‡
841
+ Base-size models (ViT-B/16)
842
+ Scratch [44]
843
+ 300
844
+ Label
845
+ IN1K
846
+ 81.2
847
+ 47.2
848
+ BeiT [2]
849
+ 800
850
+ DALL-E
851
+ DALLE250M+IN22K+IN1K
852
+ 83.2
853
+ 45.6
854
+ MAE [18]
855
+ 1600
856
+ Pixel
857
+ IN1K
858
+ 83.6
859
+ 48.1
860
+ SIM [43]
861
+ 1600
862
+ EMA
863
+ IN1K
864
+ 83.8
865
+ -
866
+ CAE [4]
867
+ 1600
868
+ DALL-E
869
+ DALLE250M+IN22K+IN1K
870
+ 83.9
871
+ 50.2
872
+ MaskFeat [48]
873
+ 1600
874
+ HOG
875
+ IN1K
876
+ 84.0
877
+ -
878
+ SdAE [7]
879
+ 300
880
+ EMA
881
+ IN1K
882
+ 84.1
883
+ 48.6
884
+ data2vec [1]
885
+ 800
886
+ EMA
887
+ IN1K
888
+ 84.2
889
+ -
890
+ PeCo [11]
891
+ 300
892
+ VQGAN
893
+ IN1K
894
+ 84.1
895
+ 46.7
896
+ PeCo [11]
897
+ 800
898
+ VQGAN
899
+ IN1K
900
+ 84.5
901
+ 48.5
902
+ TinyMIM (Ours)
903
+ 300
904
+ MAE-ViT-L
905
+ IN1K
906
+ 85.0
907
+ 52.2/52.6‡
908
+ Table 3. Fine-tuning results on ImageNet-1K and ADE20K. All models are pre-trained on ImageNet-1K. “Tokenizer/Teacher Data”: training
909
+ data of teacher and tokenizer. †: reproduced result using official code. ⋆: the model is fine-tuned for 1000 epochs with DeiT-style [44]
910
+ knowledge distillation. ‡: the model adopts an intermediate fine-tuning on ImageNet-1K classification before ADE20K segmentation
911
+ fine-tuning.
912
+ Method
913
+ Model Size
914
+ ImageNet ↑
915
+ IN-Adversarial↑
916
+ IN-Rendition↑
917
+ IN-Corruption ↓
918
+ DeiT [44]
919
+ ViT-T
920
+ 72.2
921
+ 8.0
922
+ 32.7
923
+ 54.0
924
+ MAE [18]
925
+ 71.8
926
+ 7.0
927
+ 36.5
928
+ 55.2
929
+ TinyMIM
930
+ 75.8
931
+ 11.0
932
+ 39.8
933
+ 50.1
934
+ DeiT [44]
935
+ ViT-S
936
+ 79.9
937
+ 18.3
938
+ 42.3
939
+ 41.4
940
+ MAE [18]
941
+ 80.6
942
+ 20.1
943
+ 45.6
944
+ 40.6
945
+ TinyMIM
946
+ 83.0
947
+ 27.5
948
+ 48.8
949
+ 35.8
950
+ DeiT [44]
951
+ ViT-B
952
+ 81.2
953
+ 25.8
954
+ 45.4
955
+ 36.8
956
+ MAE [18]
957
+ 83.6
958
+ 33.6
959
+ 50.0
960
+ 37.8
961
+ TinyMIM
962
+ 85.0
963
+ 43.0
964
+ 54.6
965
+ 32.7
966
+ Table 4. Robustness evaluation on out-of-domain datasets.
967
+ MAE baseline.
968
+ Target Block. As described in Section 3.1.3, we consider
969
+ a situation where the block number of the student does not
970
+ match that of the teacher. Here we use an MAE pre-trained
971
+ ViT-L containing 24 blocks to distill a ViT-B containing
972
+ 12 blocks. Here we examine the effects of using the 12th,
973
+ 15th, 18th, 21th and 24th (last) blocks of the ViT-L as the
974
+ target blocks. The comparison is shown in Table 10. We
975
+ 7
976
+
977
+ Method
978
+ Reconstruction Loss
979
+ Top-1 Acc.
980
+ MAE
981
+
982
+ 83.6
983
+ TinyMIM w/ Cls
984
+ 80.6
985
+ TinyMIM w/ Cls
986
+
987
+ 82.1
988
+ Table 5. Study of class token distillation formulated in Eq.6.
989
+ Feature
990
+ Res. Connection
991
+ Top-1 Acc.
992
+ MAE
993
+ 83.6
994
+ Output Feature
995
+ 83.7
996
+ FFN Feature
997
+ 84.2
998
+ FFN Feature
999
+
1000
+ 81.8
1001
+ Attention Feature
1002
+ 84.1
1003
+ Attention Feature
1004
+
1005
+ 81.3
1006
+ Q/K/V Features
1007
+ 84.3
1008
+ Table 6. Study of feature distillation formulated in Eq.7. See
1009
+ Section 3.1.1 and Eq. 2 for the definitions of different features.
1010
+ Relation
1011
+ Softmax
1012
+ Top-1 Acc.
1013
+ MAE
1014
+ 83.6
1015
+ Q-Q, K-K, V-V
1016
+ 84.4
1017
+ Q-Q, K-K, V-V
1018
+
1019
+ 84.5
1020
+ Q-K, V-V
1021
+ 84.4
1022
+ Q-K, V-V
1023
+
1024
+ 84.6
1025
+ Table 7. Study of relation distillation formulated in Eq. 9. See
1026
+ Section 3.1.1 and Eq. 4 for the definitions of different relations.
1027
+ experimentally find that using 18th block yields the best
1028
+ result.
1029
+ Sequential Distillation. In Section 3.3, we advocate to
1030
+ adopt a sequential distillation strategy to enable distillation
1031
+ from a larger model (e.g. ViT-L) to a smaller model (e.g.
1032
+ ViT-S). Table 11 compares the result of adopting different
1033
+ teachers with or without the sequential distillation. We have
1034
+ two conclusions: 1) using a larger teacher (MAE-ViT-L) to
1035
+ distill a smaller student (ViT-S) degrades the performance; 2)
1036
+ sequential distillation significantly boosts the performance
1037
+ of ViT-T (MAE-ViT-B→TinyMIM-ViT-S as the teacher and
1038
+ ViT-T as the student).
1039
+ Integrating MAE into TinyMIM. MAE is a simple but ef-
1040
+ fective self-supervised pre-training paradigm that trains a
1041
+ model by requiring it to predict masked inputs. In contrast,
1042
+ TinyMIM pre-trains smaller ViTs in a knowledge distilla-
1043
+ tion manner. Here we integrate MAE into our TinyMIM,
1044
+ yielding an integrated model. This model is optimized under
1045
+ two losses: knowledge distillation loss from TinyMIM, and
1046
+ Method
1047
+ Model Size
1048
+ Top-1 Acc.
1049
+ Supervised (DeiT)
1050
+ ViT-T
1051
+ 72.2
1052
+ MAE
1053
+ 71.6
1054
+ Class Token Distillation
1055
+ 70.6
1056
+ Feature Distillation
1057
+ 73.4
1058
+ Relation Distillation
1059
+ 75.8 (+4.2)
1060
+ Supervised (DeiT)
1061
+ ViT-S
1062
+ 79.9
1063
+ MAE
1064
+ 80.6
1065
+ Class Token Distillation
1066
+ 79.6
1067
+ Feature Distillation
1068
+ 80.8
1069
+ Relation Distillation
1070
+ 83.0 (+3.1)
1071
+ Supervised (DeiT)
1072
+ ViT-B
1073
+ 81.2
1074
+ MAE
1075
+ 83.6
1076
+ Class Token Distillation
1077
+ 82.6
1078
+ Feature Distillation
1079
+ 83.8
1080
+ Relation Distillation
1081
+ 85.0 (+1.6)
1082
+ Table 8. Comparison of three distillation strategies on ImageNet-1K
1083
+ image classification. The models are pre-trained under a 300-epoch
1084
+ schedule.
1085
+ Method
1086
+ Model Size
1087
+ mIoU
1088
+ Supervised (DeiT)
1089
+ ViT-B
1090
+ 47.2
1091
+ MAE
1092
+ 48.1
1093
+ Class Token Distillation
1094
+ 46.2
1095
+ Feature Distillation
1096
+ 47.7
1097
+ Relation Distillation
1098
+ 52.2 (+4.1)
1099
+ Table 9. Comparison of three distillation strategies on ADE20K
1100
+ semantic segmentation. The models are pre-trained under a 300-
1101
+ epoch schedule.
1102
+ Task
1103
+ 12th
1104
+ 15th
1105
+ 18th
1106
+ 21th
1107
+ 24th
1108
+ Classification
1109
+ 83.6
1110
+ 84.1
1111
+ 84.6
1112
+ 84.8
1113
+ 84.4
1114
+ Segmentation
1115
+ 48.7
1116
+ 49.8
1117
+ 52.2
1118
+ 50.6
1119
+ 50.0
1120
+ Table 10. Study of target block on ImageNet-1K and ADE20K.
1121
+ Student
1122
+ Teacher
1123
+ Acc.
1124
+ ViT-S
1125
+ MAE-ViT-B
1126
+ 82.3
1127
+ MAE-ViT-L
1128
+ 82.1
1129
+ MAE-ViT-L → TinyMIM-ViT-B
1130
+ 82.6
1131
+ ViT-T
1132
+ MAE-ViT-S
1133
+ 74.1
1134
+ MAE-ViT-B
1135
+ 74.4
1136
+ MAE-ViT-B → TinyMIM-ViT-S
1137
+ 75.0
1138
+ Table 11. Study of sequential distillation.
1139
+ 8
1140
+
1141
+ Masked Image
1142
+ Reconstruction Loss
1143
+ Top-1 Acc.
1144
+ 84.6
1145
+
1146
+ 83.9
1147
+
1148
+
1149
+ 84.0
1150
+ Table 12. Comparison between the TinyMIM-ViT-B (the first row)
1151
+ and the integrated model (the third row). We also study the input
1152
+ of TinyMIM-ViT-B, which could be raw image (the first row) or
1153
+ masked image (the second row).
1154
+ DPR (Teacher)
1155
+ DPR (Student)
1156
+ Top-1 Acc.
1157
+ 0.0
1158
+ 0.0
1159
+ 84.3
1160
+ 0.0
1161
+ 0.1
1162
+ 84.6
1163
+ 0.0
1164
+ 0.2
1165
+ 84.3
1166
+ 0.0
1167
+ 0.3
1168
+ 84.1
1169
+ 0.1
1170
+ 0.1
1171
+ 83.9
1172
+ Table 13. Ablation study of drop path rate (DPR) used in teacher
1173
+ and student.
1174
+ reconstruction loss from MAE. To enable MAE pre-training,
1175
+ we randomly mask 75% image patches, and feed the visi-
1176
+ ble patches into the network to initiate the pre-training of
1177
+ the integrated model. Table 12 shows the comparison be-
1178
+ tween TinyMIM-ViT-B and the integrated model. From the
1179
+ Table, we could draw a conclusion—integrating MAE into
1180
+ our TinyMIM does not improve the performance. In addi-
1181
+ tion, we also investigate the input of TinyMIM-ViT-B, which
1182
+ could be either raw image or masked image, as shown in
1183
+ Table 12—taking raw image as input yields better result.
1184
+ Drop Path. Drop path is one of the most critical techniques
1185
+ in training Transformers [44]. Using an appropriate drop
1186
+ path rate could significantly alleviate the over-fitting issue.
1187
+ However, MAE disables this technique in its implementation.
1188
+ Here we verify the effects of applying drop path to our
1189
+ TinyMIM. The results are shown in Table 13. For the student
1190
+ model, the optimal drop path rate is 0.1. For the teacher
1191
+ model, disabling drop path yields best result.
1192
+ 5. Conclusion
1193
+ In this paper, we present TinyMIM, which is the first to
1194
+ successfully perform masked image modeling (MIM) pre-
1195
+ training for smaller ViT models. In stead of adopting a
1196
+ mask-and-predict pretext task, we pre-train a small ViT by
1197
+ mimicking the relations of a large ViT in a knowledge dis-
1198
+ tillation manner. The success of TinyMIM can be attributed
1199
+ to a comprehensive study of various factors that may affect
1200
+ TinyMIM pretraining including distillation target, distillation
1201
+ input and target block. With extensive experiments, we draw
1202
+ a series of conclusions. For instance, relation distillation is
1203
+ superior than feature distillation and class token distillation;
1204
+ taking raw image as input is optimal; a sequential distillation
1205
+ is necessary for training smaller ViTs; etc. With its simplic-
1206
+ ity and strong performance, we hope our approach can serve
1207
+ as a solid baseline for future research.
1208
+ References
1209
+ [1] Alexei Baevski, Wei-Ning Hsu, Qiantong Xu, Arun Babu,
1210
+ Jiatao Gu, and Michael Auli. Data2vec: A general framework
1211
+ for self-supervised learning in speech, vision and language.
1212
+ arXiv preprint arXiv:2202.03555, 2022. 3, 7
1213
+ [2] Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. BEiT:
1214
+ BERT pre-training of image transformers. In International
1215
+ Conference on Learning Representations, 2022. 1, 2, 7
1216
+ [3] Mathilde Caron, Hugo Touvron, Ishan Misra, Herv´e J´egou,
1217
+ Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg-
1218
+ ing properties in self-supervised vision transformers. arXiv
1219
+ preprint arXiv:2104.14294, 2021. 3, 7
1220
+ [4] Xiaokang Chen, Mingyu Ding, Xiaodi Wang, Ying Xin, Shen-
1221
+ tong Mo, Yunhao Wang, Shumin Han, Ping Luo, Gang Zeng,
1222
+ and Jingdong Wang. Context autoencoder for self-supervised
1223
+ representation learning. arXiv preprint arXiv:2202.03026,
1224
+ 2022. 6, 7
1225
+ [5] Xinlei Chen, Saining Xie, and Kaiming He. An empirical
1226
+ study of training self-supervised vision transformers. ArXiv,
1227
+ abs/2104.02057, 2021. 7
1228
+ [6] Yinpeng Chen, Xiyang Dai, Dongdong Chen, Mengchen
1229
+ Liu, Xiaoyi Dong, Lu Yuan, and Zicheng Liu.
1230
+ Mobile-
1231
+ former: Bridging mobilenet and transformer. In Proceedings
1232
+ of the IEEE/CVF Conference on Computer Vision and Pattern
1233
+ Recognition, pages 5270–5279, 2022. 2
1234
+ [7] Yabo Chen, Yuchen Liu, Dongsheng Jiang, Xiaopeng Zhang,
1235
+ Wenrui Dai, Hongkai Xiong, and Qi Tian.
1236
+ Sdae: Self-
1237
+ distillated masked autoencoder. ArXiv, abs/2208.00449, 2022.
1238
+ 7
1239
+ [8] Jang Hyun Cho and Bharath Hariharan. On the efficacy of
1240
+ knowledge distillation. In Proceedings of the IEEE/CVF
1241
+ international conference on computer vision, pages 4794–
1242
+ 4802, 2019. 6
1243
+ [9] Navneet Dalal and Bill Triggs. Histograms of oriented gra-
1244
+ dients for human detection. In 2005 IEEE computer soci-
1245
+ ety conference on computer vision and pattern recognition
1246
+ (CVPR’05), volume 1, pages 886–893. Ieee, 2005. 2
1247
+ [10] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina
1248
+ Toutanova. BERT: pre-training of deep bidirectional trans-
1249
+ formers for language understanding. In Proceedings of the
1250
+ 2019 Conference of the North American Chapter of the As-
1251
+ sociation for Computational Linguistics: Human Language
1252
+ Technologies, pages 4171–4186. Association for Computa-
1253
+ tional Linguistics, 2019. 2
1254
+ [11] Xiaoyi Dong, Jianmin Bao, Ting Zhang, Dongdong Chen,
1255
+ Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, and Neng-
1256
+ hai Yu. Peco: Perceptual codebook for bert pre-training of
1257
+ vision transformers. arXiv preprint arXiv:2111.12710, 2021.
1258
+ 1, 2, 7
1259
+ [12] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov,
1260
+ Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner,
1261
+ 9
1262
+
1263
+ Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl-
1264
+ vain Gelly, et al. An image is worth 16x16 words: Transform-
1265
+ ers for image recognition at scale. preprint arXiv:2010.11929,
1266
+ 2020. 1, 3
1267
+ [13] Yuxin Fang, Li Dong, Hangbo Bao, Xinggang Wang, and
1268
+ Furu Wei. Corrupted image modeling for self-supervised
1269
+ visual pre-training. arXiv preprint arXiv:2202.03382, 2022.
1270
+ 6, 7
1271
+ [14] Zhiyuan Fang, Jianfeng Wang, Lijuan Wang, Lei Zhang,
1272
+ Yezhou Yang, and Zicheng Liu. Seed: Self-supervised distil-
1273
+ lation for visual representation. 2021. 3
1274
+ [15] Tommaso Furlanello, Zachary Lipton, Michael Tschannen,
1275
+ Laurent Itti, and Anima Anandkumar. Born again neural
1276
+ networks. In International Conference on Machine Learning,
1277
+ pages 1607–1616. PMLR, 2018. 6
1278
+ [16] Jianping Gou, Baosheng Yu, Stephen J. Maybank, and
1279
+ Dacheng Tao. Knowledge distillation: A survey. International
1280
+ Journal of Computer Vision, 2021. 2, 3
1281
+ [17] Benjamin Graham, Alaaeldin El-Nouby, Hugo Touvron,
1282
+ Pierre Stock, Armand Joulin, Herv´e J´egou, and Matthijs
1283
+ Douze. Levit: a vision transformer in convnet’s clothing
1284
+ for faster inference. In Proceedings of the IEEE/CVF inter-
1285
+ national conference on computer vision, pages 12259–12269,
1286
+ 2021. 3
1287
+ [18] Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr
1288
+ Doll´ar, and Ross Girshick. Masked autoencoders are scalable
1289
+ vision learners. In CVPR, 2022. 1, 2, 4, 6, 7
1290
+ [19] Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kada-
1291
+ vath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu,
1292
+ Samyak Parajuli, Mike Guo, Dawn Song, Jacob Steinhardt,
1293
+ and Justin Gilmer. The many faces of robustness: A critical
1294
+ analysis of out-of-distribution generalization. ICCV, 2021. 6
1295
+ [20] Dan Hendrycks and Thomas Dietterich. Benchmarking neural
1296
+ network robustness to common corruptions and perturbations.
1297
+ ICLR, 2019. 6
1298
+ [21] Dan Hendrycks, Kevin Zhao, Steven Basart, Jacob Steinhardt,
1299
+ and Dawn Song. Natural adversarial examples. CVPR, 2021.
1300
+ 6
1301
+ [22] Byeongho Heo, Jeesoo Kim, Sangdoo Yun, Hyojin Park, No-
1302
+ jun Kwak, and Jin Young Choi. A comprehensive overhaul
1303
+ of feature distillation. In Proceedings of the IEEE/CVF Inter-
1304
+ national Conference on Computer Vision, pages 1921–1930,
1305
+ 2019. 3
1306
+ [23] Byeongho Heo, Minsik Lee, Sangdoo Yun, and Jin Young
1307
+ Choi. Knowledge transfer via distillation of activation bound-
1308
+ aries formed by hidden neurons. In Proceedings of the AAAI
1309
+ Conference on Artificial Intelligence, volume 33, pages 3779–
1310
+ 3787, 2019. 3
1311
+ [24] Geoffrey Hinton, Oriol Vinyals, Jeff Dean, et al. Distill-
1312
+ ing the knowledge in a neural network.
1313
+ arXiv preprint
1314
+ arXiv:1503.02531, 2(7), 2015. 2, 3
1315
+ [25] Zejiang Hou, Fei Sun, Yen-Kuang Chen, Yuan Xie, and S. Y.
1316
+ Kung. Milan: Masked image pretraining on language assisted
1317
+ representation. ArXiv, abs/2208.06049, 2022. 1, 3
1318
+ [26] Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh
1319
+ Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu,
1320
+ Ruoming Pang, Vijay Vasudevan, et al. Searching for mo-
1321
+ bilenetv3. In Proceedings of the IEEE/CVF international
1322
+ conference on computer vision, pages 1314–1324, 2019. 2, 3,
1323
+ 12
1324
+ [27] Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry
1325
+ Kalenichenko, Weijun Wang, Tobias Weyand, Marco An-
1326
+ dreetto, and Hartwig Adam. Mobilenets: Efficient convolu-
1327
+ tional neural networks for mobile vision applications. arXiv
1328
+ preprint arXiv:1704.04861, 2017. 3
1329
+ [28] Jangho Kim, SeongUk Park, and Nojun Kwak. Paraphrasing
1330
+ complex network: Network compression via factor transfer.
1331
+ Advances in neural information processing systems, 31, 2018.
1332
+ 3
1333
+ [29] Seung Hyun Lee, Dae Ha Kim, and Byung Cheol Song. Self-
1334
+ supervised knowledge distillation using singular value de-
1335
+ composition. In Proceedings of the European Conference on
1336
+ Computer Vision (ECCV), pages 335–350, 2018. 3
1337
+ [30] Yanyu Li, Geng Yuan, Yang Wen, Eric Hu, Georgios Evan-
1338
+ gelidis, Sergey Tulyakov, Yanzhi Wang, and Jian Ren. Effi-
1339
+ cientformer: Vision transformers at mobilenet speed. arXiv
1340
+ preprint arXiv:2206.01191, 2022. 3
1341
+ [31] Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie,
1342
+ Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong,
1343
+ Furu Wei, and Baining Guo. Swin transformer v2: Scaling
1344
+ up capacity and resolution. In International Conference on
1345
+ Computer Vision and Pattern Recognition (CVPR), 2022. 3
1346
+ [32] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng
1347
+ Zhang, Stephen Lin, and Baining Guo. Swin Transformer:
1348
+ Hierarchical vision transformer using shifted windows. arXiv
1349
+ preprint arXiv:2103.14030, 2021. 1
1350
+ [33] Ilya Loshchilov and Frank Hutter. Decoupled weight de-
1351
+ cay regularization. In International Conference on Learning
1352
+ Representations, 2019. 6
1353
+ [34] Sachin Mehta and Mohammad Rastegari. Mobilevit: Light-
1354
+ weight, general-purpose, and mobile-friendly vision trans-
1355
+ former. In International Conference on Learning Representa-
1356
+ tions, 2021. 1, 2, 3
1357
+ [35] Junting Pan, Adrian Bulat, Fuwen Tan, Xiatian Zhu, Lukasz
1358
+ Dudziak, Hongsheng Li, Georgios Tzimiropoulos, and Brais
1359
+ Martinez. Edgevits: Competing light-weight cnns on mobile
1360
+ devices with vision transformers. In European Conference on
1361
+ Computer Vision, pages 294–311. Springer, 2022. 1, 2, 3
1362
+ [36] Zhiliang Peng, Li Dong, Hangbo Bao, Qixiang Ye, and Furu
1363
+ Wei. BEiT v2: Masked image modeling with vector-quantized
1364
+ visual tokenizers. arXiv preprint arXiv:2208.06366, 2022. 1,
1365
+ 3
1366
+ [37] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya
1367
+ Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry,
1368
+ Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning
1369
+ transferable visual models from natural language supervision.
1370
+ In ICML, pages 8748–8763. PMLR, 2021. 3
1371
+ [38] A. Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea
1372
+ Voss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-
1373
+ shot text-to-image generation. ArXiv, abs/2102.12092, 2021.
1374
+ 2
1375
+ [39] Sucheng Ren, Zhengqi Gao, Tianyu Hua, Zihui Xue, Yong-
1376
+ long Tian, Shengfeng He, and Hang Zhao. Co-advise: Cross
1377
+ inductive bias distillation. In Proceedings of the IEEE/CVF
1378
+ Conference on Computer Vision and Pattern Recognition
1379
+ (CVPR), pages 16773–16782, June 2022. 1, 3
1380
+ 10
1381
+
1382
+ [40] Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou,
1383
+ Antoine Chassang, Carlo Gatta, and Yoshua Bengio. Fitnets:
1384
+ Hints for thin deep nets. arXiv preprint arXiv:1412.6550,
1385
+ 2014. 3
1386
+ [41] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San-
1387
+ jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy,
1388
+ Aditya Khosla, Michael Bernstein, Alexander C Berg, and Li
1389
+ Fei-Fei. Imagenet large scale visual recognition challenge.
1390
+ IJCV, 2015. 6
1391
+ [42] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model
1392
+ scaling for convolutional neural networks. In International
1393
+ conference on machine learning, pages 6105–6114. PMLR,
1394
+ 2019. 3
1395
+ [43] Chenxin Tao, Xizhou Zhu, Gao Huang, Yu Qiao, Xiaogang
1396
+ Wang, and Jifeng Dai. Siamese image modeling for self-
1397
+ supervised vision representation learning. arXiv preprint
1398
+ arXiv:2206.01204, 2022. 7
1399
+ [44] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco
1400
+ Massa, Alexandre Sablayrolles, and Herv´e J´egou. Training
1401
+ data-efficient image transformers & distillation through atten-
1402
+ tion. preprint arXiv:2012.12877, 2020. 1, 3, 7, 9, 12
1403
+ [45] Shakti N Wadekar and Abhishek Chaurasia. Mobilevitv3:
1404
+ Mobile-friendly vision transformer with simple and effective
1405
+ fusion of local, global and input features. arXiv preprint
1406
+ arXiv:2209.15159, 2022. 1, 2
1407
+ [46] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao
1408
+ Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyra-
1409
+ mid vision transformer: A versatile backbone for dense predic-
1410
+ tion without convolutions. In Proceedings of the IEEE/CVF
1411
+ International Conference on Computer Vision, pages 568–
1412
+ 578, 2021. 1
1413
+ [47] Xiaojie Wang, Rui Zhang, Yu Sun, and Jianzhong Qi. Kdgan:
1414
+ Knowledge distillation with generative adversarial networks.
1415
+ Advances in neural information processing systems, 31, 2018.
1416
+ 3
1417
+ [48] Chen Wei, Haoqi Fan, Saining Xie, Chao-Yuan Wu, Alan
1418
+ Yuille, and Christoph Feichtenhofer. Masked feature predic-
1419
+ tion for self-supervised visual pre-training. arXiv preprint
1420
+ arXiv:2112.09133, 2021. 1, 2, 7
1421
+ [49] Yixuan Wei, Han Hu, Zhenda Xie, Zheng Zhang, Yue Cao,
1422
+ Jianmin Bao, Dong Chen, and Baining Guo. Contrastive
1423
+ learning rivals masked image modeling in fine-tuning via
1424
+ feature distillation. arXiv preprint arXiv:2205.14141, 2022.
1425
+ 1
1426
+ [50] Yixuan Wei, Han Hu, Zhenda Xie, Zheng Zhang, Yue Cao,
1427
+ Jianmin Bao, Dong Chen, and Baining Guo. Contrastive
1428
+ learning rivals masked image modeling in fine-tuning via
1429
+ feature distillation. arXiv preprint arXiv:2205.14141, 2022.
1430
+ 3
1431
+ [51] Tete Xiao, Yingcheng Liu, Bolei Zhou, Yuning Jiang, and
1432
+ Jian Sun. Unified perceptual parsing for scene understanding.
1433
+ In ECCV, 2018. 6
1434
+ [52] Zhenda Xie, Zigang Geng, Jingcheng Hu, Zheng Zhang, Han
1435
+ Hu, and Yue Cao. Revealing the dark secrets of masked image
1436
+ modeling. arXiv preprint arXiv:2205.13543, 2022. 1
1437
+ [53] Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Jianmin
1438
+ Bao, Zhuliang Yao, Qi Dai, and Han Hu. Simmim: A simple
1439
+ framework for masked image modeling. In Proceedings of
1440
+ the IEEE/CVF Conference on Computer Vision and Pattern
1441
+ Recognition, pages 9653–9663, 2022. 1, 2, 3
1442
+ [54] Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Yixuan
1443
+ Wei, Qi Dai, and Han Hu. On data scaling in masked image
1444
+ modeling. arXiv preprint arXiv:2206.04664, 2022. 1
1445
+ [55] Zihui Xue, Sucheng Ren, Zhengqi Gao, and Hang Zhao.
1446
+ Multimodal knowledge expansion. In Proceedings of the
1447
+ IEEE/CVF International Conference on Computer Vision,
1448
+ pages 854–863, 2021. 3
1449
+ [56] Junho Yim, Donggyu Joo, Jihoon Bae, and Junmo Kim. A
1450
+ gift from knowledge distillation: Fast optimization, network
1451
+ minimization and transfer learning. In Proceedings of the
1452
+ IEEE conference on computer vision and pattern recognition,
1453
+ pages 4133–4141, 2017. 3
1454
+ [57] Shan You, Chang Xu, Chao Xu, and Dacheng Tao. Learn-
1455
+ ing from multiple teacher networks. In Proceedings of the
1456
+ 23rd ACM SIGKDD International Conference on Knowledge
1457
+ Discovery and Data Mining, pages 1285–1294, 2017. 3
1458
+ [58] Shan You, Chang Xu, Chao Xu, and Dacheng Tao. Learning
1459
+ with single-teacher multi-student. In Proceedings of the AAAI
1460
+ Conference on Artificial Intelligence, volume 32, 2018. 3
1461
+ [59] Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fidler,
1462
+ Adela Barriuso, and Antonio Torralba. Semantic understand-
1463
+ ing of scenes through the ADE20K dataset. Int. J. Comput.
1464
+ Vis., 127(3):302–321, 2019. 6
1465
+ [60] Jinghao Zhou, Chen Wei, Huiyu Wang, Wei Shen, Cihang
1466
+ Xie, Alan Yuille, and Tao Kong. ibot: Image bert pre-training
1467
+ with online tokenizer. arXiv preprint arXiv:2111.07832, 2021.
1468
+ 1, 3
1469
+ 11
1470
+
1471
+ A. Hyper-parameters
1472
+ Hyper-parameters of ImageNet-1K Pre-training. See Ta-
1473
+ ble 14.
1474
+ Hyper-parameters of ImageNet-1K Image Classification
1475
+ Fine-tuning. See Table 15. TinyMIM⋆-T retains the plain
1476
+ architecture and computation budget of ViT-T. We fine-tune
1477
+ TinyMIM⋆ for 1000 epochs with DeiT-style [44] knowledge
1478
+ distillation on ImageNet-1K. Following MobileNetV3 [26],
1479
+ an extra fully connected layer is placed before the classifi-
1480
+ cation layer to increase the feature dimension from 192 to
1481
+ 1280. The head number is set to 12 instead of the default 3.
1482
+ Hyper-parameters for ADE20K Semantic Segmentation
1483
+ Fine-tuning. See Table 16.
1484
+ Hyperparameter
1485
+ ViT-T
1486
+ ViT-S
1487
+ ViT-B
1488
+ Layers
1489
+ 12
1490
+ Hidden size
1491
+ 192
1492
+ 384
1493
+ 768
1494
+ FFN inner hidden size
1495
+ 768
1496
+ 1536
1497
+ 3072
1498
+ Attention heads
1499
+ 3
1500
+ 6
1501
+ 12
1502
+ Patch size
1503
+ 16 × 16
1504
+ Pre-training epochs
1505
+ 100/300
1506
+ Batch size
1507
+ 4096
1508
+ Adam ϵ
1509
+ 1e-8
1510
+ Adam β
1511
+ (0.9, 0.999)
1512
+ Peak learning rate
1513
+ 2.4e-3
1514
+ Minimal learning rate
1515
+ 1e-5
1516
+ Learning rate schedule
1517
+ Cosine
1518
+ Warmup epochs
1519
+ 5/15
1520
+ Stochastic depth
1521
+ 0.1
1522
+ Dropout
1523
+
1524
+ Weight decay
1525
+ 0.05
1526
+ Data augment
1527
+ RandomResizeAndCrop
1528
+ Input resolution
1529
+ 224 × 224
1530
+ Color jitter (student only)
1531
+ 0.4
1532
+ Table 14. Hyper-parameters of ImageNet-1K Pre-training.
1533
+ Hyperparameter
1534
+ ViT-T
1535
+ ViT-S
1536
+ ViT-B
1537
+ Peak learning rate
1538
+ 5e-3
1539
+ 5e-3
1540
+ 2e-3
1541
+ Fine-tuning epochs
1542
+ 200
1543
+ 200
1544
+ 100
1545
+ Warmup epochs
1546
+ 5
1547
+ Layer-wise learning rate decay
1548
+ 0.65
1549
+ 0.65
1550
+ 0.65/0.6∗
1551
+ Batch size
1552
+ 2048
1553
+ 2048
1554
+ 1024
1555
+ Adam ϵ
1556
+ 1e-8
1557
+ Adam β
1558
+ (0.9, 0.999)
1559
+ Minimal learning rate
1560
+ 1e-6
1561
+ Learning rate schedule
1562
+ Cosine
1563
+ Stochastic depth
1564
+ 0.1
1565
+ Weight decay
1566
+ 0.05
1567
+ Label smoothing ε
1568
+ 0.1
1569
+ Dropout
1570
+
1571
+ Gradient clipping
1572
+
1573
+ Erasing
1574
+ 0.25
1575
+ Input resolution
1576
+ 224 × 224
1577
+ Rand augment
1578
+ 9/0.5
1579
+ Mixup
1580
+ 0.8
1581
+ Cutmix
1582
+ 1.0
1583
+ Table 15. Hyper-parameters of ImageNet-1K image classification
1584
+ fine-tuning. ∗ indicates that we use 0.65 and 0.6 for 100-epoch and
1585
+ 300-epoch pre-trained models, respectively.
1586
+ Hyperparameter
1587
+ ViT-S
1588
+ ViT-B
1589
+ Input resolution
1590
+ 512 × 512
1591
+ Peak learning rate
1592
+ 1e-4
1593
+ Fine-tuning steps
1594
+ 160K
1595
+ Batch size
1596
+ 16
1597
+ Adam ϵ
1598
+ 1e-8
1599
+ Adam β
1600
+ (0.9, 0.999)
1601
+ Layer-wise learning rate decay
1602
+ {0.65, 0.75, 0.8}
1603
+ Minimal learning rate
1604
+ 0
1605
+ Learning rate schedule
1606
+ Linear
1607
+ Warmup steps
1608
+ 1500
1609
+ Dropout
1610
+
1611
+ Stochastic depth
1612
+ 0.1
1613
+ Weight decay
1614
+ 0.05
1615
+ Table 16. Hyper-parameters of ADE20K semantic segmentation
1616
+ fine-tuning.
1617
+ 12
1618
+
M9AzT4oBgHgl3EQfWPwO/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
P9AzT4oBgHgl3EQflf2o/content/2301.01549v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f94b675290eaef83b2aad13ece7102b15cad6e7103e985a27eb0c6d56d0accb7
3
+ size 460824
P9AzT4oBgHgl3EQflf2o/vector_store/index.faiss ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:91f49247111d7022b444ad119f7a6a177112730db3e0bca779af67632820f437
3
+ size 2752557
P9AzT4oBgHgl3EQflf2o/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f8db263618790a7b87e5999c9933aea974e9a1181159f0509cf6c67e853f685
3
+ size 104679
PdAzT4oBgHgl3EQfzv5V/content/2301.01773v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a90b4a6a3f27f53c665dc01f1b841d579eafe1c4b9aebdcb7d4d330d09e39d36
3
+ size 4273302
PdAzT4oBgHgl3EQfzv5V/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:caa56f6f412d320160b46f3aa6fa6411b5c84736daccaff77f8bbd69f2ba5c89
3
+ size 134591