On two questions of Shapiro on Jacobi scaling
Year: 2026
Type: preprint
Abstract: For a symmetric positive definite matrix B of order n, let R*(n) be the supremum of κ(J(B))/min_D κ(DBD), where J(B) is the Jacobi scaling of B and D runs over the positive diagonal matrices. Van der Sluis proved that R*(n) ≤ n, and Shapiro that R*(n) ≥ n/2. We compute R*(n) for every n: R*(n) = n/2 for 2 ≤ n ≤ 5, and for n ≥ 6 it exceeds n/2 and is algebraic of degree at most 3; for instance, R*(6) = 3.0224406… is the real root of 2y³ − 21y² + 384y − 1024, and R*(12) = 7.0153277… is the positive root of y³ + 224y² + 432y − 14400. For n ≥ 3 the supremum is not attained, and the bound κ(J(B)) ≤ 1 + R*(n)(κ_opt(B) − 1) is strict for every non-diagonal B. We also show that R*(n) = n − 2√n + 2 + O(n^(−1/2)), and that R*(n)/n is strictly increasing from n = 5 on; this answers Shapiro's two questions. For block-Jacobi scaling we lower the factor in Demmel's bound to max{1, ϑ(Ḡ) − 1}, where ϑ is the Lovász theta function and Ḡ is the complement of the block graph, and show that this factor is exact on every block graph when the blocks have size at least 2; on K_b (b ≥ 3) the supremum is b − 1 if and only if at most one block has size 1. Version 2 (November 19, 2026). Changes from version 1: R*(n) is now computed exactly for every n (version 1 gave bounds for n ≥ 6, e.g. 3.02244 ≤ R*(6) ≤ 3.02753). The constant in the upper bound is now best possible: R*(n) ≤ n − 2√n + R*(9) − 3 for every n ≥ 3, with equality only at n = 9, where R*(9) − 3 = 2.0131022… (version 1: R*(n) < n − 2√n + 121/60). Odd cycles: R*(C_m) = R*_C(C_m) = (1 + sec(π/m))/2 = ϑ(C̄_m)/2 for odd m ≥ 3, for real and complex matrices. Mixed block sizes: the block supremum depends only on min(n_v, 2), and on K_b (b ≥ 3) it equals b − 1 if and only if at most one block has size 1. Every proof is by hand; the scripts in the supplement are checks only.
Author: Paiva, Gabriel Costa de Mello
Language: English
Topic: Matrix Theory and Algorithms
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7235042428
