arXiv:2610.24542
Comment on "The polarization hierarchy for polynomial optimization over convex bodies, with applications to nonnegative matrix rank"
Julius A. Zeiss
Abstract
We revisit the feasibility and convergence claims for the polarization hierarchy of Plávala, Ligthart and Gross [Linear Algebra Appl.\ 723 (2025) 15--32]. Counterexamples show that the stated assumption is insufficient for the feasibility characterization and for convergence of the hierarchy with the polarization constraint. We establish corrected results under a stronger algebraic condition and supply a missing compactness argument.
AI Summary
This comment examines whether the polarization hierarchy for polynomial optimization over compact convex state spaces has the feasibility and convergence properties previously claimed. The issue is whether a polarization constraint $\Pi(M)=0$ forces the positive-semidefinite residual second moment $M=\int f(a,b)\otimes f(a,b)\,d\mu$ to vanish. The stated two-vector assumption does not suffice. A projection onto traceless symmetric $3\times3$ matrices satisfies that assumption but annihilates $I_3$. Simplex constructions then give an infeasible problem whose hierarchy remains feasible at every level, and a feasible optimization problem with $p_n^\Pi=-1$ for every $n\ge2$, although $p^*=0$. The corrected condition is $\ker\Pi\cap\mathrm{PSD}(V)=\{0\}$. It is necessary for a universal feasibility characterization and equivalent to excluding cancellation among $D=\dim V$ positive rank-one terms; $D$ cannot generally be reduced. A semidefinite extreme-point rank bound shows that the original two-vector condition suffices when $\operatorname{rank}(\Pi|_{\mathrm{Sym}^2(V)})\le4$. Under the corrected condition, de Finetti representation establishes feasibility, while diagonal compactness extraction supplies a missing step proving monotone convergence of optimal values. The three polarization maps proposed in the original paper already satisfy the correction; the stronger hierarchy and reported finite-level infeasibility certificates remain valid. No general convergence rate or detection-level bound is obtained, and finite-level exactness or convergence of minimizers is not asserted. An example with $\Pi=\mathrm{id}$ has $p_n=-1/n\to0$, without finite exactness.
Generative AI tools supported manuscript preparation, mathematical analysis, and code development.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 25 pages, Comment on https://doi.org/10.1016/j.laa.2025.05.019 (arXiv:2406.09506)