Skip to content

[已并入 #783|复用接口] topological Fourier spectral sample #784

Description

@LightChainr

动机

当前 #773 已经把 square-site topological rank observable

X(omega)=r_black(omega)-1 in {-1,0,1}

的 several finite derivatives统一到 pivotal geometry:

M'(p)=sum_i E[Delta_i X],
M_ij=E[Delta_i Delta_j X].

#737 的 spatial-source Hessian 也正是 M_ij 的 Fourier transform。另一方面,critical percolation crossing 的 Fourier spectral sample / noise sensitivity 已有成熟理论。这里要判断:topological rank observable 是否有一个自然的 p-biased spectral sample,使 thermal derivative、pivotal-pair atlas、source Hessian 和 dynamical noise sensitivity成为同一套 Fourier levels 的不同读出。

Exact finite algebra(先验证)

对 product Bernoulli(p),用标准 p-biased orthonormal basis

phi_i=(omega_i-p)/sqrt(pq), q=1-p.

定义

X-E X = sum_{S!=empty} Xhat(S) prod_{i in S} phi_i.

需要逐项证明并在 L=3,4 exact control 验证:

Xhat({i}) = sqrt(pq) E[Delta_i X],
Xhat({i,j}) = pq E[Delta_i Delta_j X], i!=j.

因此

M'(p) = (pq)^(-1/2) sum_i Xhat({i}),
M_ij  = (pq)^(-1) Xhat({i,j}).

PR737 的 spatial Hessian spectrum 应与 level-2 signed Fourier coefficients逐点 Fourier 对上。

Spectral-sample measure

定义非空随机集合 S_X

P(S_X=S) = Xhat(S)^2 / Var(X).

则 noise operator / resampling covariance satisfies

Cov(X(omega),X(omega_epsilon))/Var(X)
 = E[(1-epsilon)^|S_X|].

以及

E |S_X|
 = [pq / Var(X)] sum_i E[(Delta_i X)^2].

注意 Delta_i X 可为 0,1,2,所以 direct rank-jump-two 以平方权重 4 进入 Dirichlet energy。这与普通 Boolean crossing 不同,必须保留。

主要问题

A. critical spectral size

若 ordinary four-arm pivotal scaling控制非零 Delta_i X,候选

E |S_X| ~ L^(3/4).

检验 square / triangular 的 L=3,4 exact values;若 #775 返回 L=5–8 rank-sector polynomials仍不足以算完整 Fourier energy,则说明还缺哪些 fixed-site jump histograms。

B. matching parity across Fourier levels

对 self-matching triangular site at p=1/2,configuration complement gives X(1-omega)=-X(omega). Since each centered basis factor changes sign,应该 exact 有

Xhat(S)=0 for every even |S|.

这是比 g_2=0 强得多的 all-level parity theorem。请证明并在 tiny exact control 验证。

对 square/matching pair,derive cross-model relation between level-k coefficients under p <-> 1-p;不要错误宣称单模型 even levels消失。

C. level-2 signed vs absolute activity

#769 当前计算

J_ij=E Delta_i Delta_j X

是 level-2 Fourier coefficient本身;而 spectral sample使用 J_ij^2(乘 (pq)^2)作为 level-2 probability mass。比较:

sum J_ij       signed cancellation,
sum |J_ij|     absolute mixed influence,
sum J_ij^2     Fourier/spectral energy.

三者可能有不同 arm fusion semantics。不要混称同一个 exponent。

D. dynamical noise sensitivity

若 normalized spectral size grows like L^(3/4),自然 decorrelation scale是

epsilon_L ~ L^(-3/4).

这与 thermal window exponent同数但概念不同;应检查是否由同一 pivotal measure控制。若能把 critical topological-rank process under dynamical percolation接到已有 spectral-sample theorem,给 theorem/gap boundary。

E. source-Hessian spectral bridge

对 translation-invariant torus,把 level-2 coefficients按 displacement d 组织:

Xhat_2(d)=pq J(d).

则 PR737 的 Fourier mode eigenvalues是 Xhat_2(d) 的空间 Fourier transform(除 pq 归一化)。给 exact formula与 L=4 controls。

文献边界

优先 primary text:

  • Benjamini–Kalai–Schramm noise sensitivity;
  • Schramm–Steif;
  • Garban–Pete–Schramm spectral sample / pivotal measure;
  • near-critical/dynamical percolation where relevant。

区分普通 crossing indicator 已证明的结果与本 rank-valued observable需要的新适配。

交付

notes/topological-fourier-spectral-sample-YYYYMMDD.md
scripts/... L3/L4 exact Fourier control
results/.../topological-fourier-control.json

最终至少给:

  1. exact finite Fourier identities;
  2. self-matching all-even-level vanishing theorem/control;
  3. source-Hessian = level-2 Fourier bridge;
  4. topological spectral size / noise scale的 theorem, conditional theorem, or explicit gap。

不做

  • 不把 squared Fourier mass 与 signed M''混为一谈;
  • 不从 L<=8 fit noise exponent作为测量;
  • 不把 crossing spectral-sample theorem自动宣称适用于 X;
  • 不改 original-U contract / STATUS。

相关:draft #773, #737, #768/#769, #775/#776

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    priority:P2Deferred research or on-demand support; no default new compute allocation.

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions