So for each query q in Q, we can consider only
the keys k in K that are close to q.