Theorem 4.4 — finite sampling-kernel error along the Euler trajectory
ProvedCritWinSpin.numeric_kernel_trajectory_tvcritical-windowsamplerspin-systemstheorem-4-4
Fix the finite numerical Euler data and a number of steps . Let be its actual rounded trajectory law and let be the output law obtained by averaging the finite acceptance and spin draws over the trajectory's random-bit blocks.
Under the stated fuel, decoding, and kernel assumptions, its distance from the ideal kernel mixture satisfies
The explicitly defined error budget contains the integral discretization error and the finite-bit errors of the Bernoulli acceptance draw and the coordinate spin draws. This is the finite-kernel approximation component of the sampler in Theorem 4.4.
Preamble
import Definitions.Def_CritWinSpin_SamplerNumericalData set_option autoImplicit false set_option linter.all false noncomputable section open Classical IsingLTL.BeliefProp MeasureTheory CritWinSpin open CritWinSpin.LocalTM2 CritWinSpin.LocalTM2.Compiled CritWinSpin.Implementation
Formal statement
theorem CritWinSpin.numeric_kernel_trajectory_tv {N : ℕ} (S : SpinLaw) (hS : IsGHS S) (C : ℕ) (hC : 0<C)
(D : EulerData N) (T : ℕ) (hs : tau S < (D.sCap:ℝ))
(hfSpin : 2 / (spinPrecision C D.k:ℝ)^3+2 ≤ D.fuel.length)
(hTake : D.k+1 ≤ D.fuel.length)
(hfIntegral : 2 / (1/((D.k:ℝ)+1))^3+2 ≤ (D.fuel.length:ℝ))
(hnil : decConfig S N [] = none)
(r : Option (Config S (Fin N)) → ℝ)
(hmean : ∀ y, r y = FiniteDraw.mean (fun w : Fin T → Fin (N*gaussianNoiseBits D.k) → Bool =>
kernelWordLaw S C D (finitePath (D.next S) (fun _ => 0) w) y))
(Ck : ℝ) (hCk : 0 ≤ Ck)
(hLip : ∀ z w : Fin N → ℝ,
tvDist (kernelK S (D.sCap:ℝ) (castM D.L) (castV D.h) z)
(kernelK S (D.sCap:ℝ) (castM D.L) (castV D.h) w) ≤
Ck * Real.sqrt (N:ℝ) * Real.sqrt (normSq (z-w))) :
tvDist r
(fun x => ∫ z, SpinKernel.pointKernel S (D.sCap:ℝ) (castM D.L) (castV D.h) z x
∂D.trajectoryLaw S T (fun _ => 0)) ≤
KernelPrecisionBudget.error S C N D.k (D.sCap:ℝ) := by sorry
Source
Liu–Zhang–Ren, The Critical Window for Sampling and Learning-to-Sample in Spin Systems, STOC 2027 submission (manuscript sha256 f77584da…), Section 4, pp.15–17, equations (23)–(24). Finite acceptance/spin-kernel implementation estimate for Theorem 4.4.