Improved Upper Bounds for Slicing the Hypercube
Researchers used reasoning LLMs to discover a construction that improves a combinatorial geometry bound from 1971.
Deep Dive
A team of researchers including Duncan Soiffer and others used the AI tool CPro1 to prove an improved upper bound for slicing all edges of an n-dimensional hypercube. They proved S(n) ≤ ⌈4n/5⌉, beating the previous 1971 bound of ⌈5n/6⌉. The key breakthrough was CPro1, which uses reasoning LLMs and automated hyperparameter tuning to search for mathematical constructions, finding 8 hyperplanes that slice the 10-dimensional cube.
Why It Matters
Demonstrates AI's growing role in pure mathematics research, automating the search for proofs and constructions previously done manually.