Work

Unique minimizers under prescribed marginals: answers to two questions of Gurvits and Leake

Year: 2026
Type: preprint
Abstract: Gurvits and Leake asked whether, for every n and every trace vector t of sum n sufficiently close to the all-ones vector, the scalar tuple (1/n)(t₁I, …, tₙI) uniquely minimizes the mixed discriminant over positive semidefinite n-tuples with traces tᵢ and sum I (Question 1.7), and whether n^(−n)(Σtᵢxᵢ)ⁿ uniquely minimizes the coefficient of x₁⋯xₙ over n-homogeneous real-stable polynomials with value one and gradient t at the all-ones vector (Question 1.8). We answer both questions affirmatively with explicit neighborhoods. For real-stable polynomials the coefficient excess dominates the centered-root dispersion and the coefficient distance to the power, with explicit constants in every degree and the radius 1/8 in degree three; for Lorentzian polynomials the minimizer persists with linear coefficient stability, which yields a mixed-volume stability bound. For the mixed discriminant we prove the balanced reserve MD(Y) − n^(−n) ≥ bₙ Σ‖Yᵢ − I/n‖²_F on doubly stochastic tuples and an explicit trace-perturbed reserve of quadratic order whose radius exceeds that of the general polynomial route by many orders of magnitude. We identify the exact region Σtᵢ² < n²/(n−1) in which the scalar tuple is a strict local minimizer, and we show that for every n ≥ 3 global minimality fails at trace vectors strictly inside this region. An exact rational 3×3 example at ℓ¹-distance 24/25 bounds the optimal radius for n = 3, where the two questions are equivalent. For n = 2 both questions have affirmative answers at every trace vector.
Language: English
Open Access status: green
OpenAlex ID: W7234502199
Cites: 1
Cited by:
Related to: 10