One problem asks how to time 42 trains to cut peak power use. In under two hours, agents beat MIPLIB’s best timetable by 2.6%, and later proved no better timetable exists.
We’ve submitted the solutions to MIPLIB for anyone to check.
Claude agents beat MIPLIB's best train timetable by 2.6% and prove it optimal
AIAnthropic says its agents found a train timetable for 42 trains that beats MIPLIB's best known solution by 2.6% in under two hours. The agents later proved that no better timetable exists.
AIWhy it matters.
The post reports a concrete optimization result and a proof of optimality, with solutions released for independent checking on MIPLIB.
Source details
Claude · x.comPublished · added here
Original title: One problem asks how to time 42 trains to cut peak power use. In under two hours, agents beat MIPLIB’s best timetable by 2.6%, and later ...