ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let BB and VV be weight matrices satisfying Assumption~, with V≠IdV\neq\mathrm{Id}, and let (ρtβ)t≥0(\rho_t^\beta)_{t\geq 0} be the token-distribution evolution on Sd−1{\mathbb S}^{d-1}. Let FF be the eigenspace of VV associated with its largest eigenvalue, and let FabsF^{\mathrm{abs}} be the eigenspace associated with an eigenvalue of largest magnitude. For a probability measure ρ∞∈P(Sd−1)\rho_\infty\in\mathcal{P}({\mathbb S}^{d-1}), write supp⁡(ρ∞)\operatorname{supp}(\rho_\infty) for its support. Dominant-eigenspace concentration conjecture. There exist weight matrices BB and V≠IdV\neq\mathrm{Id} satisfying Assumption~, and a probability measure ρ∞∈P(Sd−1)\rho_\infty\in\mathcal{P}({\mathbb S}^{d-1}) supported on F∩Sd−1F\cap{\mathbb S}^{d-1} or Fabs∩Sd−1F^{\mathrm{abs}}\cap{\mathbb S}^{d-1} such that

lim⁡t→∞W2(ρtβ,ρ∞)=0.\lim_{t\to\infty}W_2(\rho_t^\beta,\rho_\infty)=0.

Numerical evidence and asymptotic stability results support convergence toward a dominant eigenspace at long times, beyond the O(log⁡β)\mathcal{O}(\log\beta) time scale. The conjecture is trivial when V=IdV=\mathrm{Id}, while the nontrivial case concerns general interactions with V≠IdV\neq\mathrm{Id}; the supplied text does not establish the claim.

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.