math.PR · 2605.05193 Copy arXiv ID · May 6, 2026 Save Grokability in five inequalities Authors: Paata Ivanisvili , Xinyuan Xie
Organizations: University of California, Irvine
Abstract In this note, we report five mathematical discoveries made in collaboration with Grok, all of which have been subsequently verified by the authors. These include an improved lower bound on the maximal Gaussian perimeter of convex sets in R n \mathbb{R}^n R n , sharper L 2 L_2 L 2 -L 1 L_1 L 1 moment comparison inequalities on the Hamming cube { − 1 , 1 } n \{-1,1\}^n { − 1 , 1 } n , a strengthened autoconvolution inequality, improved asymptotic bounds on the size of the largest g g g -Sidon sets in { 1 , … , n } \{1,\dots,n\} { 1 , … , n } , and an optimal balanced Szarek's inequality.
Explore similar work Jun 30, 2026 · Sungyoon Kim, Mert Pilanci Upper Bounds Inequalities
Jul 25, 2026 · Zhanfu Yang Upper Bounds Conjecture
May 6, 2026 · Ziang Chen, Jaume de Dios Pont, Paata Ivanisvili +2 Inequalities Spectral Norm
Jun 30, 2026 · cs.AI J/K move · Enter open · S save
Sungyoon Kim, Mert Pilanci
Recent work shows that LLM agents can improve sharp-constant inequalities by searching for extremal constructions, which yield upper bounds. We address the complementary side: a lower bound holds for every admissible function and follows from a convex relaxation of the nonconvex problem, with tighter relaxations giving stronger bounds. We instantiate the autoresearch paradigm to discover such relaxations: a coding agent proposes valid tightening constraints, a theory agent verifies each one and searches for counterexamples, and every reported bound is certified by an explicit dual-feasible point checked in rigorous interval arithmetic. On two optimization constants studied by \citet{tao2025alphaevolve} - the first autocorrelation inequality (
C 6.2 C_{6.2} C 6.2 ) and the Erdős minimum-overlap constant (
C 6.5 C_{6.5} C 6.5 ) - we improve the certified lower bounds from
1.28 1.28 1.28 to
1.2937 1.2937 1.2937 and from
0.379005 0.379005 0.379005 to
0.37912 0.37912 0.37912 , respectively.