ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let G\ninRm×nG\nin\mathbb{R}^{m\times n} have independent standard-normal entries. For constraint density α=lim⁡n→∞m/n\alpha=\lim_{n\to\infty}m/n and κ∈R+\kappa\in\mathbb{R}_+, consider the statistical SBP S(G,κ,α)\mathbf{\mathcal S}(G,\kappa,\alpha). Define its algorithmic threshold by

αa(κ)=max⁡{α | lim⁡n→∞P(S(G,κ,α) is solvable in polynomial time)=1}.\alpha_a(\kappa)=\max\left\{\alpha\ \middle|\ \lim_{n\to\infty}\mathbb{P}\bigl(\mathbf{\mathcal S}(G,\kappa,\alpha)\text{ is solvable in polynomial time}\bigr)=1\right\}.

Let ψˉrd(r)=ψˉrd\bar{\psi}_{rd}^{(r)}=\bar{\psi}_{rd}, with ψˉrd\bar{\psi}_{rd} and p^,q^,c^,γ^sq\hat{\mathbf p},\hat{\mathbf q},\hat{\mathbf c},\hat{\gamma}_{sq} as in the cited definitions and theorem. SBP algorithmic threshold conjecture. There is a non-increasing convergent sequence αc(r)(κ)\alpha_c^{(r)}(\kappa) defined by

αc(r)(κ)={α | ψˉrd(r)(p^,q^,c^,γ^sq)=0},\alpha_c^{(r)}(\kappa)=\left\{\alpha\ \middle|\ \bar{\psi}_{rd}^{(r)}(\hat{\mathbf p},\hat{\mathbf q},\hat{\mathbf c},\hat{\gamma}_{sq})=0\right\},

such that

αc(r)(κ)≥αa(κ),lim⁡r→∞αc(r)(κ)=α^a(κ)=αa(κ).\alpha_c^{(r)}(\kappa)\geq\alpha_a(\kappa),\qquad \lim_{r\to\infty}\alpha_c^{(r)}(\kappa)=\hat{\alpha}_a(\kappa)=\alpha_a(\kappa).

Since αc(κ)=αc(2)(κ)\alpha_c(\kappa)=\alpha_c^{(2)}(\kappa), the statistical computational gap is

SCG=αc(κ)−αa(κ)=αc(2)(κ)−lim⁡r→∞αc(r)(κ).SCG=\alpha_c(\kappa)-\alpha_a(\kappa)=\alpha_c^{(2)}(\kappa)-\lim_{r\to\infty}\alpha_c^{(r)}(\kappa).

The claim proposes that the fl-RDT hierarchy converges to the SBP algorithmic threshold, with the finite-rr quantities providing upper estimates; the source gives no resolution, so the conjecture remains open.

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.