Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-adaptedPipeline-generatedjudge pass (gpt-6-luna)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Geometric small-set renewal convergence gives an integrable exponential hitting moment

Statement

Assume the Axiom of Choice. Let Q be a probability kernel on any measurable space (X,G), with invariant probability π. Let C∈G satisfy π(C)>0, and suppose Q(x,⋅)≥ϵν(⋅)(x∈C),0<ϵ<1, where ν is a probability measure. Suppose some A<∞ and 0<θ<1 satisfy ∣νQn(C)−π(C)∣≤Aθn(n≥0). For a Q-chain started at x, put σC=inf⁡{n≥0:Xn∈C}. Then some s>1 makes H(x)=Ex[sσC]measurable,π(H)<∞, where s∞=∞. In particular H is finite π-almost everywhere; finiteness at every state is not asserted.

For the same hypotheses, the split construction also has the following return and occupation properties. Put w1(x)=ϵ1C(x) and w0(x)=1−w1(x). Define Kx,1=ν, Kx,0=(Q(x,⋅)−ϵν)/(1−ϵ) on C, and Kx,0=Q(x,⋅) off C. On X×{0,1} use the kernel Q~((x,i),dy×{j})=Kx,i(dy)wj(y) and invariant law π~(dx,i)=wi(x)π(dx). Its atom α=C×{1} has mass p=ϵπ(C)>0. If Z0 has law κ=π~(⋅∣α) and τ=inf⁡{n≥1:Zn∈α}, then some b>1 satisfies Eκbτ<∞, and for every nonnegative measurable g on the split space, π~(g)=p Eκ∑k=0τ−1g(Zk). In particular τ<∞ almost surely and Eκτ=1/p. These additional assertions use exactly the construction and hypotheses above; they do not assert an exponential entrance-time moment from every state.

Facts & Assumptions

Given: AC, the invariant law, small-set minorization, and scalar geometric-convergence bound in the Statement.

[A1]

The Axiom of Choice: AC is assumed, as required by the path-space and Markov suppliers below.

[F1]

Small-set minorization and geometric drift: the displayed minorization permits the probability residual (Q(x,⋅)−ϵν)/(1−ϵ) on C.

[F2]

Canonical Markov chain on path space: probability kernels on arbitrary measurable spaces have canonical path laws under AC.

[F3]

Measurability of integration against a kernel: integration of a nonnegative measurable function against a probability kernel is measurable.

[F4]

Discrete strong Markov property: future bounded path functionals obey the Markov identity at stopping times.

[F5]

Markov property for bounded future path functionals: bounded future-path expectations are measurable functions of the starting state and satisfy the deterministic-time Markov identity.

[F6]

Invariant initial law makes a Markov chain stationary: an invariant initial distribution makes every finite-dimensional law stationary.

[F7]

Monotone convergence for the integral: nonnegative increasing limits pass through integrals.

[F8]

Tonelli's theorem for nonnegative measurable functions on a sigma-finite product: nonnegative double sums and probability integrals may be interchanged.

[F9]

The sum of a complex power series is analytic throughout its open disc of convergence and Every complex analytic function is holomorphic: an absolutely convergent complex power series is holomorphic in its open disk of convergence.

[F10]

A complex power-series sum has complex derivatives of every order, obtained by repeated termwise differentiation: power series can be differentiated repeatedly term by term inside their convergence disk.

[F11]

All higher complex derivatives exist and satisfy Cauchy's integral formula on an interior circle: a holomorphic function has derivatives of every order locally; these derivatives are continuous because each has a further complex derivative.

[F12]

Cauchy's inequalities bound the Taylor coefficients by the circle supremum: if ∣g∣≤B on a radius-δ circle about a contained with its disk in the holomorphic domain, then ∣g(k)(a)∣/k!≤Bδ−k.

[F13]

Invariant and stationary distribution for a Markov kernel: invariance is equality of the initial law with its one-step image.

[F14]

Measure kernel and probability kernel: a probability kernel is measurable in its starting point and countably additive in its target event.

Proof

Given: The hypotheses in the Statement, with AC [A1].

1.1A1F1F2F3F14given

Set w1(x)=ϵ1C(x) and w0(x)=1−w1(x). On X~=X×{0,1} define Kx,1=ν; define Kx,0=(Q(x,⋅)−ϵν)/(1−ϵ) for x∈C, and Kx,0=Q(x,⋅) otherwise. The split kernel is Q~((x,i),B×{j})=∫Bwj(y)Kx,i(dy). Nonnegativity follows from [F1], measurability from [F3], and finite summation over the two flag sections defines a probability on every product-measurable event. Thus [F2] realizes the split path. If the initial flag at x has probabilities wi(x), summing it gives ∑iwi(x)Kx,i=Q(x,⋅). The same summation at each subsequent time shows that the x-coordinate path has the original Q-chain law. The set α=C×{1} is an atom: all its states have the identical outgoing transition obtained by using ν.

2.1F6F13step 1.1given

Put π~(dx,i)=wi(x)π(dx). Summing its incoming transitions first over i gives πQ=π and then assigns the new flag according to wj, proving π~Q~=π~ by [F13]. Let p=π~(α)=ϵπ(C)>0, and let κ=π~(⋅∣α), an ordinary normalized restriction. Write Pα=Pκ; its future after time zero is the same from every atom state. Starting in α, the next coordinate law is ν and subsequent coordinate transitions are Q, so u0=1,un=Pα(Zn∈α)=ϵνQn−1(C)(n≥1). The hypothesis implies ∣un−p∣≤cθn for all n≥0 after increasing a finite c.

3.1F4F7F8F9step 2.1

Let τ=inf⁡{n≥1:Zn∈α} and fn=Pα(τ=n) for n≥1, with f0=0. Decompose a visit at time n≥1 by the first return time and use [F4] and the common atom transition to obtain un=∑j=1nfjun−j. The possible event τ=∞ contributes zero. For ∣z∣<1, absolute convergence and [F8] give U(z)=∑n≥0unzn=11−F(z),F(z)=∑n≥1fnzn. Here ∣F(z)∣≤∣z∣∑nfn≤∣z∣<1, so the denominator is legitimate. Since un−p decays geometrically, [F9] gives U(z)=p/(1−z)+J(z) with J holomorphic on ∣z∣<θ−1. For real t↑1, U(t)→∞; hence F(t)→1 and [F7] gives ∑nfn=1.

3.2F6step 2.1algebra

Let T=inf⁡{n≥0:Zn∈α} and start the split chain from π~. By [F6], Ij=1α(Zj) have mean p and Cov⁡(I0,In)=p(un−p). Expanding the square of N−1∑j=0N−1Ij−p and summing the geometric covariance bound from step 2.1 gives expectation at most c1/N for some finite c1. On {T≥N} this square equals p2, so Pπ~(T≥N)≤c1/(Np2)→0. In particular the stationary split chain hits the atom almost surely; no unproved Harris or skeleton-recurrence theorem is needed.

4.1F7F10F11F12step 3.1

Near z=1 the expression F^(z)=1−1−zp+(1−z)J(z) is holomorphic and agrees with F on its overlap with ∣z∣<1. Indeed the denominator has value p>0 at 1 and remains nonzero on a sufficiently small disk; the quotient is complex differentiable there by the difference-quotient formula for the derivative of a reciprocal. By [F10], for 0<t<1 and k≥0, F(k)(t)=∑n≥kfn(n)ktn−k, where (n)k=n!/(n−k)!. Letting t↑1 and using [F7] and derivative continuity from [F11] gives ∑n≥kfn(n)k=F^(k)(1). Choose a radius δ>0 whose closed disk about 1 lies strictly within this holomorphic neighborhood and a finite bound B for ∣F^∣ on its circle. [F12] then bounds these nonnegative derivative values by k!Bδ−k.

4.2F5F6F7F8step 2.1step 3.2

For bounded nonnegative g, decompose g(ZN) by the last atom visit at times 0,…,N. A last visit at time N−k contributes pEα[g(Zk);τ>k], by stationarity and [F5]. The remaining paths have no atom visit through time N, and their contribution RN lies between zero and ∥g∥∞Pπ~(T>N). Thus π~(g)=p∑k=0NEα[g(Zk);τ>k]+RN. Let N→∞ using step 3.2, then extend from bounded g to all nonnegative measurable g using [F7] and [F8]. This proves the stationary occupation identity π~(g)=pEα∑k=0τ−1g(Zk). At k=0 the atom state has law κ; subsequent transitions are independent of which atom state was chosen, so this argument does not require g to be constant on α.

5.1F8step 3.1step 4.1algebra

For 0<h<δ, the finite binomial expansion for each integer n, followed by [F8], yields Eα[(1+h)τ]=∑k≥0hkk!∑n≥kfn(n)k≤B1−h/δ<∞. Thus some b>1 has Eαbτ<∞. The return is almost surely finite by step 3.1. This argument establishes an exponential moment from positivity and analytic continuation; it does not assume in advance that the original series F converges beyond the unit disk.

6.1F5F7F8step 1.1step 5.1step 4.2∎

Fix 1<s<b and let g(z)=EzsT, allowing infinity. The bounded path functions sT∧l increase to sT, so [F5] and [F7] give measurability and the Markov identity for this nonnegative function by truncation. On α, g=1. For k≥1, on {τ>k}, the residual time to α is τ−k; the Markov identity and step 4.2 give π~(g)=p(1+Eα∑k=1τ−1sτ−k)≤pEα[τsτ]<∞. The last moment is finite because sup⁡j≥1j(s/b)j<∞ and step 5.1 bounds Eαbτ. For the coordinate chain initialized with the prescribed flag, its first C visit σC is no later than its first atom visit T. Consequently H(x)=ExsσC is measurable by the same bounded-path approximation and π(H)≤π~(g)<∞. This proves the conclusion, including the null-set qualification.

Depends on

Used by

Dependency tree · two levels

65 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources