A Berry--Esseen Bound of Order (1+ n)/sqrt n for Peng's Central Limit Theorem
summary
The gist
We derive a quantitative central limit theorem under sublinear expectations from a spatial endpoint regularity estimate for the one-dimensional G-heat equation The final replacement step is estimated…
In short
This work derives a quantitative central limit theorem under sublinear expectations for the one-dimensional G-heat equation. It establishes a Berry–Esseen Bound of order (1+ log n)/√n for bounded Lipschitz tests when the third absolute moment is finite and lower variance is positive, providing a precise rate of convergence.
Key concepts
- Quantitative Central Limit Theorem under Sublinear Expectations
- This theorem investigates how the G-normal distribution converges to a normal distribution when using sublinear expectations instead of standard linear ones. It provides specific error bounds for this convergence, which is crucial for understanding the limiting behavior of stochastic processes in these non-standard settings.
- Endpoint Regularity of the G-heat Equation
- This concept involves proving regularity properties (like Lipschitz continuity) for solutions to the one-dimensional G-heat equation at specific points, particularly near time zero. The paper shows that the second spatial derivative of the solution, [uxx(t, ·)], is Lipschitz in space away from $t=0$, which is a key input for bounding convergence errors.
- Nonlinear Stein Estimates
- These are advanced mathematical tools used to bound the difference between a G-normal expectation and its true expectation. The paper uses these estimates, specifically Lemma 3.2, to derive an integral term that is essential for establishing the main convergence proof of the theorem.
Terminology used across episodes
This episode discusses
- A Berry--Esseen Bound of Order (1+ n)/sqrt n for Peng's Central Limit Theorem · Paper Radio
- A New Central Limit Theorem under Sublinear Expectations
The paper
A Berry--Esseen Bound of Order (1+ n)/sqrt n for Peng's Central Limit Theorem · Read on arXiv
State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences · School of Mathematical Sciences, University of Chinese Academy of Sciences
We derive a quantitative central limit theorem under sublinear expectations from a spatial endpoint regularity estimate for the one-dimensional G-heat equation. For bounded Lipschitz initial data, the second spatial derivative and the time derivative of its solution are Lipschitz in space and 1/2-Hölder in time away from the initial time. In particular, [u xx(t, times)] Lip at most C G (ϕ)/t. Combining this estimate with the nonlinear Stein identity established by Song and a replacement argument yields a bound of order (1+ n)/sqrt n for bounded Lipschitz tests when the third absolute moment is finite and the lower variance is positive. The final replacement step is estimated directly, avoiding the nonintegrable singularity at time zero.
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: We're looking at the paper: "A Berry--Esseen Bound of Order (1+ n)/sqrt n for Peng's Central Limit Theorem".
Emmy: Here's the gist of the paper. We derive a quantitative central limit theorem under sublinear expectations from a spatial endpoint regularity estimate for the one-dimensional G-heat equation The final replacement step is estimated directly, avoiding the nonintegrable singularity at time zero Main Convergence Result The theorem establishes a Berry–Esseen Bound of order (1+ log n)/√n for bounded Lipschitz tests when the third absolute moment is finite and the lower variance is positive This bound holds under the condition that Eb
X1: = Eb
-X1: = 0, 0 < α:= -Eb
-X2/: , m3:= Eb
X13: < ∞, and AG = 8βπα2 In particular, for uniformly bounded tests with Lipschitz constant at most one, the error is O((1 + log n)/√n) Endpoint Regularity of the G-heat Equation The authors prove endpoint regularity for the one-dimensional G-heat equation, showing that for bounded Lipschitz initial data, the second spatial derivative and the time derivative of its solution are Lipschitz in space and 1/2-H¨older in time away from the initial time Specifically,
uxx(t, ·): Lip ≤ CG Lip(ϕ)/t This regularity is derived using a one-dimensional linear smoothing estimate and subsequent comparison arguments involving smooth approximations of the G-heat equation Nonlinear Stein Estimates The proof relies on nonlinear Stein estimates to bound the difference between the G-normal expectation and the true expectation Lemma 3.2 provides an interpolating identity, leading to a bound of NG
F: − N
F: ≤ 2m3 Z 1/0
F''s Lip: ds This is applied to the Stein interpolation, yielding the key integral term in the main convergence proof Proof of Convergence The main convergence theorem proceeds by defining sequences of expectations and using independence and semigroup properties to establish a bound on ai − ai−1 Summing these bounds, combined with the endpoint regularity estimate
uxx(t, ·): Lip ≤ AGL/t, leads to the final bound of order (1+ log n)/√n The separation of the last replacement step introduces a cutoff at t = 1/n, which accounts for the log n term Scope and Limitations The endpoint estimate improves the regularity input from an unspecified exponent γ ∈ (0, 1) to spatial Lipschitz continuity However, substituting this into an integral extending to time zero is not justified because the bound AGL/t is nonintegrable there The endpoint assertion confirms the Lipschitz continuity of uxx, not the existence of a continuous third derivative This illustrates that C3 regularity does not hold across joining points References
Crandall M G, Ishii H, Lions P L. [Ada: User’s guide to viscosity solutions of second order partial differential equations.
Emmy: Bulletin of the American Mathematical Society, 1992, 27(1): 1–67]
Krylov N V. [Ada: On Shige Peng’s central limit theorem.
Title and authors: Ada: So we're looking at this paper called "A Berry--Esseen Bound of Order (one plus n)/sqrt n for Peng's Central Limit Theorem" <ref:2610.15419#pg1,A Berry--Esseen Bound of Order (1>. This work tackles how to get a quantitative central limit theorem under sublinear expectations for sums of independent variables.
Emmy: It’s about taking the general idea from Peng’s central limit theorem and making it concrete with some actual bounds. The title suggests they're improving the rate of convergence.
Kolya: I'm interested in what this actually means for us, Ada; is this just a theoretical tweak, or does it translate into something practical for how we model these random sums?
Ada: Well, the core idea here is to derive this convergence rate using an endpoint regularity estimate from the one-dimensional G-heat equation. They show that they can get a bound of order (one plus log n) over the square root of n for bounded Lipschitz tests, provided we have a finite third absolute moment and positive lower variance <ref:2610.15419#pg1,a bound of order (1>.
Emmy: That’s interesting because it connects probability to partial differential equations through that G-heat equation setup. It suggests this isn't just a statement about random variables in isolation; it has a geometric underpinning.
Kolya: The caveat I see immediately is that they require the third absolute moment of the variable to be finite, which limits the types of random processes we can apply this to. Also, what about those constants in the bound? Do they hold up when we apply this to more complex test classes than just Lipschitz functions?
Ada: The paper specifies that for uniformly bounded tests with a Lipschitz constant at most one, the error is bounded by something as small as (one plus log n) over the square root of n <ref:2610.15419#pg1>. They also mention that this holds under specific conditions on alpha, beta, and m3 values.
Emmy: That specific bound of (one plus log n) over the square root of n is what they claim for those simple tests; it’s a concrete quantitative statement about how fast the approximation gets better as we add more variables <ref:2610.15419#pg1>.
Kolya: So, if we're talking about practical applications, what does this improvement actually mean? Does it allow us to use this central limit theorem in scenarios where the third moment is infinite, or are those conditions just necessary for their proof right now?
Ada: The paper explicitly states that they assume the third absolute moment is finite for this particular result. They also discuss how they can use endpoint regularity to get a spatial Lipschitz continuity estimate for the second derivative of the G-heat equation solutions, specifically showing that
uxx(t, ·): Lip is bounded by something proportional to C times Lip of the initial data divided by t.
Title and authors: Emmy: That spatial regularity estimate is key; it’s what lets them build the link between their PDE analysis and their probabilistic convergence proof. It shows that the solution behaves nicely in space away from time zero, which is crucial for controlling the error term.
Kolya: But you mentioned something about using smooth approximations of the heat equation solutions to get those estimates, right? What does that actually look like in terms of bounding things?
Ada: They use a one-dimensional linear smoothing estimate and then comparison arguments involving smooth approximations of the G-heat equation to derive that spatial Lipschitz bound. They also show how they can utilize nonlinear Stein estimates, which is a technique for bounding the difference between the G-normal expectation and the true expectation.
Emmy: The nonlinear Stein identity is where they get that integral term in their main convergence proof; it’s a powerful tool for bridging that gap between what we expect and what actually happens under sublinear expectations. It shows how structure in the PDE relates directly to the probabilistic error.
Kolya: So, when you look at the overall convergence proof, how do these regularity estimates combine with the Stein identity to actually yield that final bound of (one plus log n) over sqrt n <ref:2610.15419#pg1>? Is it a direct sum of things?
Ada: The main convergence theorem sets up sequences of expectations and uses independence and semigroup properties to establish a bound on the difference between successive expectations, which they then sum up. Combining those bounds with the endpoint regularity estimate, specifically
uxx(t, ·): Lip is bounded by C times Lip of the initial data divided by t, leads to that final order (one plus log n) over the square root of n <ref:2610.15419#pg1>.
Emmy: And they do a specific thing there—they introduce a cutoff at time t equals one over n in the last replacement step. That specific choice introduces that log n term into the error bound, which is what makes it (one plus log n) and not just (one over sqrt n) <ref:2610.15419#pg1>.
Kolya: That sounds like a very delicate construction. What's the limitation they explicitly state about their method? Is there a point where this approach completely breaks down, or is it just restricted by the moment conditions they set at the beginning?
Ada: They do state that substituting that endpoint estimate into an integral extending to time zero isn't fully justified because that specific bound, proportional to C times Lip of the initial data divided by t, becomes nonintegrable there.
Title and authors: Emmy: That’s a crucial piece of information; it tells us that while they can get good control away from time zero, they can't easily extend this specific type of regularity all the way back to the very beginning of time. It confirms that C3 regularity doesn't hold across those joining points.
Kolya: So, to summarize what we have: for bounded Lipschitz tests with a finite third moment and positive lower variance, this paper provides a quantitative central limit theorem under sublinear expectations with an error bound of (one plus log n) over the square root of n <ref:2610.15419#pg1,a quantitative central limit theorem under sublinear expectations>.
Ada: Exactly. And they show how endpoint regularity on the one-dimensional G-heat equation allows them to get that specific rate by combining it with nonlinear Stein identities and a clever replacement argument that handles the singularity at time zero by stopping at t equals one over n.
Emmy: It’s a sophisticated way to use PDE analysis—the G-heat equation—to get probabilistic estimates, which is pretty impressive structural work. It shows how the geometry of the underlying space dictates the speed of convergence in these kinds of limit theorems.
Kolya: For someone who just listens to this show, what does this mean for their actual work? Does it suggest a new way to look at these types of sums, or is it just a very precise tool for existing models?
Ada: It suggests that when you have certain structural conditions on the underlying space and the moment assumptions are met, you can achieve this specific convergence rate. It gives researchers a concrete target for what they should be aiming for when trying to prove central limit theorems in these settings.
Emmy: And it points towards using nonlinear Stein methods more effectively as a way to bridge those gap between the abstract expectation and the actual distribution of the normalized sums. The structure there is quite elegant.
Kolya: I think it's a very precise tool, definitely. It’s not solving every problem, but for problems where you can check those moment conditions, this gives you a much tighter error control than maybe what was available before.
Ada: Right. So that’s the essence of "A Berry--Esseen Bound of Order (one plus n)/sqrt n for Peng's Central Limit Theorem" <ref:2610.15419#pg1,A Berry--Esseen Bound of Order (1>. It’s about getting a concrete rate by tying together spatial regularity of the G-heat equation with nonlinear Stein identities and careful time truncation.
Emmy: A solid piece of work that really grounds the probabilistic theory in some hard geometric analysis. We should keep an eye on how they handle those lower variance conditions in future extensions.
Kolya: Definitely, we’ll be watching how these techniques translate when we move to higher dimensions or more complex underlying geometries. That’s where the real test will be for this paper and related ideas.
The paper's summary: Ada: So, this paper is putting together everything we've talked about—the heat equation regularity and those Stein estimates—to finally give us a concrete error rate for Peng's central limit theorem under sublinear expectations.
Emmy: Exactly, Ada. They’re essentially showing that the structure of the G-heat equation isn't just some abstract mathematical object; it provides a tangible control mechanism for how fast these random sums converge to their normal distribution.
Kolya: From a probabilistic standpoint, what does this (one plus n over the square root n) bound actually tell us about the tail behavior of the error term? Is that rate achievable in more complex scenarios?
Ada: It tells us that when you have a finite third absolute moment and positive lower variance, you can guarantee this specific rate for bounded Lipschitz tests. It's a quantitative statement about how much error you can expect to see when approximating the true expectation with the G-normal expectation.
Emmy: The way they get that rate is through this endpoint regularity estimate on the one dimensional G-heat equation. They show that even though they can't guarantee infinite differentiability at time zero, they can control the Lipschitz continuity of the second spatial derivative away from that point. That spatial Lipschitz bound is key because it controls the growth of the error term in their main convergence proof.
Kolya: So, it’s not just a general statement about convergence; it's tied directly to this specific PDE regularity. What’s missing is that they explicitly state that this endpoint estimate doesn't hold all the way down to time zero because the bound becomes nonintegrable there. That means their control is localized near the start of time.
Ada: Right, that’s the limitation they flag, Kolya. They can't just apply that spatial Lipschitz control across every single point in space and time. But they do introduce a cutoff at one over n for the last part of their proof, which is what lets them get that extra log n term into the final bound.
Emmy: It’s a very neat construction to handle that singularity by truncating the time interval just enough to tame the growth of the spatial derivative, and it's how they manage to get that (one plus log n) over square root n rate instead of something simpler.
Kolya: For someone who only listens, what does this mean for their modeling work? It means if you’re working with these kinds of sums and your variables have those moment conditions met, you can expect a much tighter control on the error than you might have thought before.
Ada: It suggests a clear target for future research. Researchers should aim to find ways to extend that endpoint regularity estimate, maybe by looking at different types of nonlinear Stein identities or different underlying geometries, to see if they can push that control further into the singularity at time zero.
The paper's improvements: Kolya: So we've looked at the main convergence result—that (one plus n over the square root n) bound for bounded Lipschitz tests when you have a finite third absolute moment and positive lower variance. Now, let's talk about what this paper suggests can be improved beyond that baseline.
Ada: The authors are pointing out that their method is tied to specific assumptions about the test functions, and they're suggesting a way to get even better rates for those tests. They are looking at how the structure of the G-heat equation allows them to refine the error term calculation.
Emmy: What they’re suggesting is using those endpoint regularity estimates we talked about earlier—the ones that show spatial Lipschitz continuity for the second derivative—and applying them in a different way to see if we can get a cleaner bound on the integral term from the nonlinear Stein estimates.
Kolya: So, if I understand this right, they aren't just looking at one fixed test class anymore; they are trying to find a more robust framework that handles a wider family of tests while still getting that improved convergence rate.
Ada: Exactly. They want to see if we can use the spatial regularity of the solution not just for bounding it, but actively for sharpening the convergence estimate itself across a broader range of conditions. It’s about making the proof more flexible and less dependent on those very strict moment conditions they started with.
Emmy: The implication here is that we might be able to use these geometric tools—the heat equation analysis—to derive stronger probabilistic results for other classes of random sums, not just the ones covered by this specific theorem. It suggests a deeper connection between the geometry of the underlying space and the statistical behavior of these limits.
Kolya: But what’s their actual plan for pushing it further? If we can get a better rate, say something closer to one over n, what kind of new mathematical structure would that imply about these sublinear expectations?
Ada: The paper hints that by getting this improved bound, you might be able to prove convergence under even weaker moment conditions in some cases. It’s about finding the minimal requirements for these types of central limit theorems and showing where the constraints truly lie.
Emmy: They are essentially trying to see if we can move from a theorem that works under strong assumptions to one that works with slightly less demanding assumptions, by leveraging more intricate structural properties of the G-heat equation's solution space.
Conclusion: Ada: So, to wrap this up, we've seen how they use the endpoint regularity of the one dimensional G heat equation to get a quantitative central limit theorem under sublinear expectations with an error of one plus n over the square root n.
Emmy: That means they’ve successfully linked PDE analysis directly into probability theory using those nonlinear Stein estimates, showing how geometric structure dictates statistical convergence speed.
Kolya: For me, the main thing is that this gives us a concrete error control for applications where we can check those third absolute moment conditions and positive lower variance. It's a useful tool for bounding the tail behavior of these sums.
Ada: Exactly, Kolya. It’s about providing a specific target for researchers trying to prove these types of limit theorems in settings involving sublinear expectations.
Emmy: I think what this paper really changes is how we view the bridge between analysis and probability; it shows that the tools from geometric PDEs can be used to precisely measure probabilistic errors.
Kolya: It’s a very precise tool, definitely. It’s not solving every problem, but for problems where you can check those moment conditions, this gives you a much tighter error control than maybe what was available before.
Ada: So we've seen how the authors used the G heat equation regularity to build that specific bound in the paper "A Berry--Esseen Bound of Order (one plus n over the square root n) for Peng's Central Limit Theorem".
Emmy: And it sets a high bar for how we might use those structural estimates to improve convergence rates in other complex systems.
Kolya: I'm curious if this method can be pushed further into higher dimensions, or if those geometric constraints will always limit the strength of the resulting bounds.
More episodes
- 2508.20749-Rates of convergence for extremal spacings in Kakutani's random interval-splitting process
- 2509.05860-Concentration Inequalities for Branching Random Walks
- 2509.19309-Small Prime Primitive Roots in Arithmetic Progressions
- 2509.21830-Noncollapsing for Curvature Flows with Inhomogeneous Speeds
- 2510.08346-Separability and Submetrizability in Locally Convex Spaces
- 2510.27290-Borel Combinatorics of Schreier Graphs of Z-actions
- 2511.13500-Sampling Density for Gabor Phase Retrieval
- 2511.19161-Linear dynamics of random products of weighted shifts
- 2511.21548-Metastability of diffusion processes in narrow tubes
- 2512.01536-Abelian approximations for bicrossed products