Keyboard shortcuts

Press ← or → to navigate between chapters

Press ? to show this help

Press Esc to hide this help

Metadata
bibkey: friedlandgaubert2019spectral
authors: Shmuel Friedland and Stephane Gaubert
year: 2019
title: "Spectral inequalities for nonnegative tensors and their tropical analogues"
doi: null
url: https://arxiv.org/html/1804.00204v3
claim: "Finite supported normalized circulations give the log spectral radius variational formula; positive left/right Perron vectors give an attainer."
strata_touched:
  - D5/S3/Combinatorics/Graph/Pressure/WeightedWalks
  - D5/S3/Combinatorics/Graph/Pressure/CyclePressure
license: citation-only
triage: anchor

Finite supported pressure

Verified locator

Primary source: https://arxiv.org/html/1804.00204v3, version 3, section 6.1, “Entropic characterization of the spectral radius of a nonnegative matrix”. Lemma 15 identifies normalized circulations as convex combinations of supported cycle occupation measures. Lemma 16 relates occupation measures to stationary row-stochastic matrices, including reducible measures. Theorem 17, equation (6.2), gives the entropy maximum for the actual nonnegative matrix spectral radius; equation (6.1) specifies the normalized circulation constraints. Lemma 18, equations (6.3) and (6.4), gives the inequality and positive left/right Perron attainer under its stated irreducibility and common-support hypotheses. This locator was checked against that declared primary HTML version on 2026-10-18.

Application and formal scope

Set mu(i,j)=pi(i)Q(i,j), and use W(i,j)=exp(-c(i,j)/tau) on the original support. Multiplying Theorem 17’s entropy maximum by -tau gives mean cost minus tau times stationary entropy. Zero-mass rows can be completed on original graph successors. The owning WeightedWalks exposition gives the zero-aware KL derivation and attained minimum using existing finite stationary existence and KL nonnegativity; this application has no separate retained API.

The Lean content constructs the actual Perron data from the attributed Apache-2.0 MCMC source and relates unrestricted, repeated supported walks to real weighted matrix powers at every length, including zero. Positive-vector bounds give growth at every start without aperiodicity, and zero supported costs count the same walk type. CyclePressure constructs an actual minimizing simple cycle, bounds every nonempty closed walk, and proves the actual maximum outdegree bracket and right-hand zero-temperature limit using a potential comparison and a Perron estimate along walks. These are formalizations of classical mathematics and carry no new priority claim. The cited section is the literature anchor for the variational formulation; it is not asserted to state the Lean unrestricted-walk encoding or every subsequent comparison step.