Chollet's permanent inequality for diagonally dominant factors and every graph Laplacian
Chollet's conjecture (1982) asserts per(H ∘ T) ≤ per(H) per(T) for positive semidefinite matrices. We prove the stronger Bapat–Sunder bound per(H ∘ T) ≤ per(T) ∏ H_ii whenever H is Hermitian with positive diagonal and scaled diagonally dominant, for every positive semidefinite T. In particular Chollet's inequality holds for every pair of weighted graph Laplacians, with arbitrary magnetic phases, and for all their principal submatrices. This answers the question of Pant and Singh whether per(L_G ∘ L_G) ≤ per(L_G) ∏_v deg_G(v) for every graph G; previously it was known for graphs whose blocks are bipartite or cycles. The proof writes the permanent as an expectation over the cycles of a random map and bounds each cycle by one of its edges, which gives a sharper matching bound with an exact equality case. Spectral radius one is the sharp threshold for this bound. A second, entrywise criterion covers correlation matrices whose squared off-diagonal row sums are at most 2/5. As an application we determine the exact optimum of generalized boson bunching over all internal-state Gram matrices in the same class, together with its minimal internal dimension.
Authors
- Scott Schubr (ORCID: https://orcid.org/0009-0009-7357-4439)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24430650
- Primary Topic
- Matrix Theory and Algorithms
- Type
- preprint