Rectangular matrix multiplication bound improved via shared-leg entropy
Rectangular matrix multiplication exponents govern the cost of graph and dynamic-programming algorithms, so a proven bound of ω(1,1/2,1)=2 turns into concrete runtime drops, including the O(n^2.4999) shortest-path result. Theoretical computer scientists and algorithm engineers tracking how OpenAI's entropy-based technique spreads through complexity theory will want the proof details. The arXiv note is open access, so the full constructions can be checked directly.