bibkey: bourinlee2021inradius
authors: J.-C. Bourin and E.-Y. Lee
year: 2021
title: “Eigenvalue inequalities for positive block matrices with the inradius of the numerical range”
doi: 10.1142/S0129167X22500094
url: https://arxiv.org/abs/2111.15180v1
claim: “Let . If the inequality \\left\\| \\begin{bmatrix} A &X \\\\ X^* &B\\end{bmatrix}\\right\\|_\\infty \\le \\|A+B\\|_\\infty holds for all positive block-matrix with as off-diagonal block, then is essentially Hermitian.”
strata_touched:
- D5/S3/Quantum/BlockNorm/SpikeEdgeEstimate
- D5/S3/Quantum/BlockNorm/EssentiallyHermitian license: citation-only triage: anchor
Positive block matrices and essentially Hermitian off-diagonal blocks
Verified locator
DOI: 10.1142/S0129167X22500094
Source: https://arxiv.org/abs/2111.15180v1
Conjecture 3.3 is on printed page 6 of the arXiv v1 PDF. Crossref identifies the same title and authors with the IJM DOI above. Theorem 3.2 on the same page quotes Hayashi’s Theorem 2.5 (arXiv:1808.00181).
Conjecture 3.3, verbatim, with its displayed formula:
Let . If the inequality
holds for all positive block-matrix with as off-diagonal block, then is essentially Hermitian.
The preceding sentence is: “If W(T) is line segment, then T is a so-called
essentially Hermitian matrix.” The binding convention in
https://github.com/the-omega-institute/trureturing/issues/13752 encodes this
as the literal affine Hermitian expression , with
and . The claim quantifies over every
, all complex matrices , and every positive
semidefinite completion. The norm is the Euclidean operator norm, identified
in Lean with Matrix.toEuclideanCLM by definitional equality.
Hayashi’s quoted theorem says, verbatim:
Suppose that is invertible with distinct singular values. If the inequality
holds for all positive block-matrix with as off-diagonal block, then is normal.
Proposition 3.4 on printed page 6 states the Frobenius-norm characterization: the same inequality for every positive completion holds if and only if is normal. This is a proved result in the source; the operator-norm claim has the stronger conclusion of an affine Hermitian matrix. The argument for other unitarily invariant norms is not supplied here.
Formal mechanism
Positive scalar shifts force the top and bottom spectral edges of to sum to zero for every Hermitian . Rank-one probes , with , have a two-sided edge expansion with error for and . The coefficient of forces normality; the coefficient of , tested with the difference of two projected rank-one matrices, forces their equality. Unitary diagonalization then makes the spectrum collinear and reconstructs .
Code provenance
The private normal-operator and normal-matrix diagonalization proofs are
adapted from the TauCeti contributors, commit
7c8d7117e41432d613582fc600dcb0a3be7dba30, under Apache-2.0.
Copyright (c) 2026 The Tau Ceti contributors. The license text is available
at https://www.apache.org/licenses/LICENSE-2.0. No TauCeti Lake dependency
is used. Pinned Mathlib supplies the simultaneous eigenspace decomposition
of the self-adjoint real and imaginary parts and the subordinate
orthonormal basis construction.