Abstract
We have developed a reinforcement learning agent that often finds a minimalsequence of unknotting crossing changes for a knot diagram with up to 200crossings, hence giving an upper bound on the unknotting number. We have usedthis to determine the unknotting number of 57k knots. We took diagrams ofconnected sums of such knots with oppositely signed signatures, where thesummands were overlaid. The agent has found examples where several of thecrossing changes in an unknotting collection of crossings result in hyperbolicknots. Based on this, we have shown that, given knots $K$ and $K'$ that satisfysome mild assumptions, there is a diagram of their connected sum and $u(K) +u(K')$ unknotting crossings such that changing any one of them results in aprime knot. As a by-product, we have obtained a dataset of 2.6 million distincthard unknot diagrams; most of them under 35 crossings. Assuming the additivityof the unknotting number, we have determined the unknotting number of 43 atmost 12-crossing knots for which the unknotting number is unknown.