arXiv:2609.25462
Sharper Bounds for the Complex Grothendieck Constant
Steven Heilman, Chris Jones, Giulio Malavolta
Abstract
We show that $1.4<K_G^{\mathbb{C}}<1.404898554746$, where $K_G^{\mathbb{C}}$ is the complex Grothendieck constant. The upper bound improves on Haagerup's bound of $1.40490913\ldots$ from 1987, and the lower bound improves on Davie's bound of $1.33807$ from 1984. The upper bound combines ideas from the real and noncommutative complex cases of Grothendieck's inequality.
AI Summary
The paper addresses bounds on the complex Grothendieck constant $K_G^{\mathbb C}$, which controls the gap between optimization over unit Hilbert-space vectors and over complex phases. It proves $1.4<K_G^{\mathbb C}<1.404898554746$. The upper bound improves Haagerup’s 1987 bound $1.40490913\ldots$ by approximately $1.06\times10^{-5}$. The lower-bound argument gives $K_G^{\mathbb C}>4000/2857\approx1.40007$, improving both the historical bound $1.33807$ and an independently obtained bound $1.355846\ldots$ discussed in the introduction. The upper bound uses two-stage preprocessing and a mixture of six Gaussian rounding kernels. Five branches apply phase rounding after different tensor-space preprocessors; the sixth uses radially perturbed phase functions and conjugation symmetrization. An inverse-repair lemma converts the nearly linear mixed kernel into an exactly linear valid kernel. Its certified slope exceeds $0.71179516600840569540$. For the lower bound, the paper bounds the Gaussian operator norm of $T_n=P_{1,0}-(37/50)P_{2,1}$ uniformly in dimension by $2857/4000<5/7$. The proof combines Hermite decompositions, dimension-independent cubic-chaos moment bounds, and a reduction to scalar potentials with finite optimization certificates. Both bounds rely on analytic reductions followed by rigorous numerical certification; supporting verification code is provided. The exact constant remains unknown, including the conjectured value $1.404575934663742\ldots$. The paper also reports that directly adapting limiting Krivine schemes, which give the best known real-case bounds, did not succeed here.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 21 pages