A functional Breuer--Major theorem with Poisson noise and a sharp regularity condition

arXiv:2510.26216 · math.PR · Submitted 2025-10-30 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: I'm Ada.

Emmy: And I'm Emmy, here with Kolya.

Ada: We're looking at the paper: "A functional Breuer--Major theorem with Poisson noise and a sharp regularity condition".

Emmy: Here's the gist of the paper. A functional Breuer-Major theorem with Poisson noise extends the classical result for Gaussians to stationary sequences arising from a Poisson point process by establishing a functional central limit theorem.

Ada: Extension of the Central Limit Theorem The paper extends the functional Breuer-Major theorem for Gaussians to the Poisson case, where the stationary sequence arises from a Poisson point process.

Emmy: The main result establishes a functional central limit theorem for the process defined by when Xn arises from a Poisson point process.

Ada: This is achieved through two primary steps: first, proving the tightness of the sequence of processes and second, proving the convergence of finite-dimensional distributions.

Title and authors: Ada: So, we're looking at this paper titled "A functional Breuer--Major theorem with Poisson noise and a sharp regularity condition". It extends the classical results that deal with Gaussians to a new setting where the underlying sequence isn't Gaussian but comes from a Poisson point process. It also adds this part about needing a sharp regularity condition for the function phi we are studying.

Emmy: That sounds like they're taking something that works nicely for random variables, which is the Breuer-Major theorem, and trying to make it work when the noise structure is different, specifically coming from a Poisson point process. The authors are Kong and Wang from Peking University.

Kolya: So what does this mean practically? Are we talking about a more general class of sequences where the correlations aren't as simple as in a standard Gaussian setting?

Ada: Exactly. The core idea is extending the Breuer-Major theorem, which originally dealt with stationary sequences arising from Gaussian processes, over to stationary sequences that arise from a Poisson point process. The new part they introduce is this regularity condition on the function phi.

Emmy: That sounds like they are trying to figure out what kind of smoothness or control we need on our function phi for this theorem to hold in the Poisson context. It's about making sure that even with this different noise, we can still get convergence results.

Kolya: So, if we think about it simply, they’re showing that if your noise comes from a Poisson process and your function has enough regularity according to their new condition, you can still get the same kind of convergence as in the standard Gaussian case.

The paper's summary: Ada: To summarize what this paper does, they establish a functional central limit theorem for these processes defined by that Poisson point process. They do this by first proving that the sequence of processes is tight, which they do using an Lp spectral gap inequality specific to the Poisson point process.

Emmy: So tightness means that even though we have this more complex noise structure from the Poisson process, we can still control the moments of our processes in a way that guarantees they don't just wander off into infinity as n gets larger. That’s a key step in proving convergence.

Kolya: The spectral gap inequality is mentioned, and I want to know what that actually tells us about the tail behavior of these moments? Does it give us any specific bounds on how fast those moments grow?

Ada: It provides a bound on the p-th moment of F in terms of Malliavin derivatives. That's because in the Poisson case, you don't have the Meyer inequality that we often rely on for controlling those moments, so they had to use this different approach to control them.

Emmy: And then for proving convergence of finite-dimensional distributions, they use a Fourier expansion technique on phi(X u) to decouple it from the random variable X u. This turns the problem into something resembling a trigonometric case that fits with their method of moments.

Kolya: So, by using this Fourier expansion to match it to the method of moments, are they overcoming some issue we usually have when applying those methods in Poisson settings?

Ada: Yeah, they are tackling the failure of the classical method of moments in this specific case because of bounds on the derivatives. They show that this decoupling helps reduce it down to a trigonometric case where their method works properly.

The paper's improvements: Emmy: Now let's talk about what they suggest as improvements or additions to the original idea, because they introduce some new tools and conditions. They introduce a class of functions with specific regularity properties, defined in Definition one point one <ref:2510.26216#pg1>.

Ada: This definition sets up the necessary conditions for phi, which we can call C gamma M. It involves bounding how fast phi grows with respect to the argument x, and also controlling the smoothness of phi at certain points using a parameter gamma.

Kolya: So this regularity condition is what allows them to get that tightness in the Skorohod space D(

zero one: ) <ref:2510.26216#pg2,in the Skorohod space D( 0, 1>? What's the connection between this function class and controlling those moments?

Emmy: It connects because by assuming these strong decay conditions on the coefficients of the chaos expansion of phi, they can use hypercontractivity to bound the p-th moment of each chaos component. This is what guarantees that the series converges.

Ada: So, it’s not just a general statement about Poisson noise; they are saying that for this convergence to work, you need your function phi to belong to this specific class C gamma M, where gamma and M are parameters controlling the growth and smoothness.

Kolya: I'm curious about what the limitations of this method are here. Does requiring a function to be in this class impose a restriction on the types of functions we can analyze?

Ada: Yes, it does. The requirement is quite specific regarding how fast phi grows and its local oscillation control at points, which is what they call the sup notation in Definition one point one <ref:2510.26216#pg1>. It’s a technical constraint that makes the theorem applicable to these specific classes of functions.

Conclusion: Ada: So, to wrap up this paper, they establish the functional Breuer--Major theorem with Poisson noise and a sharp regularity condition by proving tightness using an Lp spectral gap inequality for the Poisson point process. They also show convergence of finite-dimensional distributions through Fourier expansion and moment matching.

Emmy: The main implication is that we can extend these central limit theorems to stationary sequences arising from Poisson point processes, provided our function phi satisfies their specific regularity condition, which they call C gamma M. This gives us a way to analyze the limiting behavior of these processes in this new noise setting.

Kolya: From a probabilistic standpoint, the convergence of finite-dimensional distributions is shown by relating moments to the variance term varX n j=one b j t j and showing that the expectation of powers scales with 2p times that variance term. This lets us control how much things fluctuate in law.

Ada: That scaling is important because it shows how the convergence behaves across different moments, and they even show a specific convergence for T >= dZn, which relates to the process converging to something like mu B in law.

Emmy: It's a solid piece of work because they manage to use tools from spectral analysis and Fourier techniques to handle the complexity introduced by the Poisson noise structure. But they admit that requiring that function phi to be in the class C gamma M is a necessary constraint for their methods to work.

Kolya: I just want one final thought on what this means for applications. If we have a real-world process generated by a Poisson process, and it fits into this function class, then we can trust that our approximations based on the central limit theorem will converge to the right object.

Ada: That’s right, it gives us a framework for understanding these kinds of limits in more complex noise environments. We'll keep an eye on this work as we look at how these tools might apply elsewhere.

Fanhao Kong, Haiyi Wang

Peking University

math.PR

Submitted: 2025-10-30

Updated: 2026-11-10

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 87/100

The gist: A functional Breuer-Major theorem with Poisson noise extends the classical result for Gaussians to stationary sequences arising from a Poisson point process by establishing a functional central limit…

Key concepts

Functional Central Limit Theorem (FCLT)
This theorem shows that a sequence of stochastic processes, derived from a Poisson point process, converges in distribution to a standard Brownian motion. This is crucial because it describes the long-term, continuous behavior of the discrete random points.
Tightness via Spectral Gap Inequality
Tightness ensures that the sequence of processes does not wander too wildly over time. For Poisson processes, this is achieved using an $L^p$ spectral gap inequality. This inequality provides bounds on moments related to Malliavin derivatives, which are necessary because standard inequalities fail in the Poisson setting.
Fourier Expansion for Decoupling
The convergence of finite-dimensional distributions is analyzed by expanding the function $\phi(X_u)$ into a Fourier series. This technique separates the function $\phi$ from the process $X_u$, transforming a complex problem into a simpler trigonometric case that matches established methods for analyzing moments.

Terminology

Summary

A functional Breuer-Major theorem with Poisson noise extends the classical result for Gaussians to stationary sequences arising from a Poisson point process by establishing a functional central limit theorem. p. 0

Extension of the Central Limit Theorem

The paper extends the functional Breuer-Major theorem for Gaussians to the Poisson case, where the stationary sequence arises from a Poisson point process. The main result establishes a functional central limit theorem for the process defined by when Xn arises from a Poisson point process. p. 1 This is achieved through two primary steps: first, proving the tightness of the sequence of processes and second, proving the convergence of finite-dimensional distributions. p. 1

Tightness via Spectral Gap Inequality

The tightness of the process (T≥dYn)n≥1 is established by using an Lp spectral gap inequality for the Poisson point process to control moments. Specifically, Proposition 3. p. 33 provides a bound on the p-th moment of F in terms of Malliavin derivatives. p. 3 This technique is employed because the lack of Meyer inequality in the Poisson case necessitates this alternative approach to control moments. p. 1

Convergence through Fourier Expansion

The convergence of finite-dimensional distributions is achieved by Fourier expanding phi(Xu) to decouple it from Xu, reducing the problem to a trigonometric case matching the method of moments. This approach overcomes the failure of the classical method of moments in the Poisson case due to derivative bounds. p. 1

Convergence of Finite-Dimensional Distributions

The convergence of finite-dimensional distributions is proven by showing that for fixed N, t1,..., tN and b1,.., bN, the process Yen converges to µB in law. p. 2 This relies on the convergence of expectations involving multiple Wiener-Ito integrals and a decomposition of Yen into terms that converge to Brownian motion components. p. 0

Convergence of Truncated Terms

The convergence of the truncated terms Bn,l,m is shown by decomposing the summation over partitions into a main part and a remainder part. The remainder part is controlled using Lemma 3. p. 16 to show that its limit as n→∞ is zero. p. 1 This confirms that the convergence of finite-dimensional distributions also holds for T≥dZn. p. 3

Convergence of Truncated Processes

The convergence of the truncated processes in terms of their L2 norm is shown by showing that the difference between Yeε,m,n and Yen converges to zero as m→∞. This is achieved by bounding the difference using Lemma 4. p. 112 and applying dominated convergence theorem. p. 0

Convergence of Finite-Dimensional Distributions (Final)

The final convergence for the finite-dimensional distributions is established by showing that limε→0 limm→∞ µ2ε,m = µ2. This is done by controlling the term X∞k=d1k! p. 1ρk(θ) ≲ (1 + θ1)dα(1 + θ2)dα < ∞. p. 2 This leads to the convergence of the expectation to µ2, which is given by Dϕb⊗2,X∞k=d1k! p. 0ρkE. The convergence of finite-dimensional distributions also holds for T≥dZn. p. 3 p. 3 p. 3 p. 3 p. 3 p. 3

Convergence of Finite-Dimensional Distributions (T≥dZn)

The convergence for T≥dZn is established by showing that E[nt−[nt]√n T≥dϕ(X[nt])2] → 0 as n→∞. This follows from the fact that the difference between Yn and Zn converges to zero in a suitable sense. p. 3 The convergence of finite-dimensional distributions also holds for T≥dZn.

Convergence of Finite-Dimensional Distributions (Yen)

The convergence of Yen to µPNj=1 bjBtj in law is shown by relating the moments to the variance term varXNj=1 bjBtj and showing that limn→∞ E(Yeε,m n)2p = (2p − 1)!!varXNj=1 bjBtj pµ2p. This implies Yeε,m n converges in law to µε,m PNj=1 bjBtj. p. 2 The convergence of finite-dimensional distributions also holds for T≥dZn.

Convergence of Finite-Dimensional Distributions (Yen - Yen, epsilon→0)

The convergence of the difference between Yeε,n and Yen in L2(omega) is shown to be zero as ε→0. This is established by bounding the difference using Lemma 4. p. 1 p. 111 and showing that the resulting series converges. p. 5 The convergence of finite-dimensional distributions also holds for T≥dZn.

Convergence of Finite-Dimensional Distributions (Yen - Yen, m→∞)

The convergence of the difference between Bn,l,m and Bel,m is shown to be zero as n→∞. This is established by showing that the remainder term A2p(n) converges to zero. p. 1 The convergence of finite-dimensional distributions also holds for T≥dZn.

Convergence of Finite-Dimensional Distributions (Yen - Yen, n→∞)

The convergence of the difference between Yeε,n and Yen in L2(omega) is shown to be zero as n→∞. This is established by bounding the difference using Lemma 4.10 and showing that the resulting series converges. p. 5 The convergence of finite-dimensional distributions also holds for T≥dZn.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improve convergence of finite-dimensional distributions by showing that for Yen, Yen:= 1/sqrt(n) X N j=1 b j [nt X j]−1 u=0 T≥dϕ (Xu) (4.1) converges to mu B PN j=1 bjBtj in law.

  2. Enhance the ability to analyze the convergence of moments for truncated processes by using the result that lim n→∞ E(Yeε,m n) 2p = (2p − 1)!! varX N j=1 bjBtj p µ 2p ε,m as a tool for controlling finite-dimensional distributions.

  3. Enable the analysis of process behavior for Poisson point processes by utilizing the L p spectral gap inequality of Poisson point process to control moments, which is used to prove tightness in D([0, 1]).

  4. Strengthen the understanding of local bounds for functions by applying Lemma 2.4 and Lemma 2.5 to bound phiˆ⊗N M0+2,RK⃗‖ in terms of the process norm Y N j=1 (1 + Ki)−γ0, which controls regularity.

  5. Develop methods to handle the complexity of the chaos expansion for Poisson processes by using the Fourier expansion phi(Xu) = ⟨phi, e b iθXu⟩θ to separate phi and Xu when applying the method of moments for finite-dimensional distributions.

Sources

Related papers