Sharp constants for sums of matrix moduli when 1<p<2
Year: 2026
Type: preprint
Abstract: Unrefereed research candidate. In one sentence: for 1 < p < 2 this paper finds the exact Schatten-norm constant for sums of matrix moduli when there are at least as many matrices as the matrix size, gives a new general upper bound ((1+√m)/2)1−1/p, and certifies in exact arithmetic that rank-one matrices do not give the answer to Lee’s two-matrix problem. The question. For numbers, |a+b| ≤ |a|+|b|. A matrix A has a natural absolute value |A| = (A*A)1/2, but for matrices the size of a sum can exceed the size of the sum of the absolute values. Measuring size by the Schatten p-norm (p = 1 is the trace norm, p = 2 the Frobenius norm), one asks for the smallest constant c with ‖A1+…+Am‖p ≤ c ‖|A1|+…+|Am|‖p. What was known. Eun-Young Lee (2010, 2012) asked for the best constant for two matrices. Lin and Zhang (2022) settled p = 2; Tang and Zhang extended the question to m matrices and conjectured a formula; Qiu (2026) proved it for p ≥ 2; below p = 2 the formula fails, by counterexamples of Zeng, Liu and Ratnavelu, of Pang and of Qiu, all built from rank-one matrices. The survey of Bourin and Zhang (2026) lists the range 1 < p < 2 as open. Results (1 < p < 2, n-by-n matrices). (1) When m ≥ n the best constant is an explicit one-variable maximum Dp,n; it depends only on the rank of the sum of the moduli, every equality case is classified, and for m < n the constant is strictly smaller. Consequently the best constant for rank-one summands is Dp,min(m,n) in every dimension. (2) The best constant is log-convex in 1/p, so interpolating between p = 1 and p = 2 gives the dimension-free bound ((1+√m)/2)1−1/p, better than the previous m(p−1)/(2p) for every m ≥ 2 (for two matrices at p = 3/2: 1.0648 instead of 1.1225, while the true value is at least 1.0364). (3) For two 3-by-3 matrices a computer-assisted proof in exact rational and interval arithmetic shows that the best rank-one value is beaten at nine exponents between 11/10 and 17/10. How. The sum is written as a product of two block matrices and Hölder’s inequality is applied; the problem reduces to comparing two Rényi entropies of the spectrum, whose extremal shape is one large eigenvalue and the rest equal. This extends T. Zhang’s Frobenius-norm argument; the matching examples are Pang’s, with a re-tuned parameter. Limits. The exact constant for two matrices of arbitrary size (Lee’s problem) for 1 < p < 2 remains open, as does the case m < n; the interpolation bound is not sharp; the certified excess over the rank-one value is tiny (about 10−5 to 10−7) and covers nine specific exponents, not an interval. Contents. The paper (LaTeX source and PDF) with Figure 1; the certificate script certify_rank_gap.py with its exact rational witnesses; verify_schatten.py (128 numerical checks of the scalar constant, equality families, adversarial searches, the bounds table and asymptotics, with negative controls); the figure script; recorded outputs; the bibliography from registry records; a machine-readable index; a research-gate record; a rating-free review summary; a replay receipt and SHA-256 sums. Assurance. Written proofs, except one computer-assisted proof in exact arithmetic (Proposition 2.5). Internal AI-assisted review only (five reviewer roles, an adversarial review and audit by a second AI model, a clean replay, a re-review and a proof audit): not peer review, independent reproduction or formal verification. Creator: Anonymous. Text and data CC0-1.0; code MIT.
Author: Anonymous
Language: English
Subfield: Applied Mathematics
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233583545
