🔢 mathematics

A Kolmogorov fourth-moment bound on Poisson chaos via a martingale core

This paper establishes a Kolmogorov fourth-moment bound for Poisson chaos variables by constructing a finite-count martingale core that extends fixed-chaos estimates to the sole assumption of a finite fourth moment, thereby removing previous technical assumptions and proving that the Kolmogorov distance to normality is bounded by a constant multiple of the fourth-moment excess.

Original authors: Guangqu Zheng

Published 2026-08-03

📖 8 min read🧠 Deep dive

CC BY 4.0

Original authors: Guangqu Zheng

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to figure out if a mysterious collection of random events is actually following a very specific, predictable pattern known as the "bell curve" or "normal distribution." This pattern is the golden rule of statistics; it shows up everywhere, from the heights of people in a city to the errors in a GPS signal. But sometimes, nature throws a curveball, and you need a way to measure just how close your random events are to this perfect bell shape.

In the world of probability, there's a famous rule called the "Fourth Moment Theorem." Think of the "fourth moment" as a special score you calculate for your random data. If this score hits a specific number (3), it usually means your data is behaving exactly like a bell curve. For a long time, mathematicians could prove this rule worked perfectly, but only if they were allowed to make some very strict, almost magical assumptions about the data—like assuming the data was perfectly smooth or bounded in a way that real-world data rarely is. It was like having a rule that only worked if you promised the data wouldn't get too messy. The big question was: Can we prove this rule works even when the data is messy, as long as it doesn't explode into infinity?

This paper, written by Guangqu Zheng, tackles that exact problem. The author introduces a clever new tool called a "martingale core," which acts like a smart filter or a set of stepping stones. Instead of trying to analyze the messy data all at once, this tool breaks the problem down into smaller, manageable chunks that stay within the same "family" of patterns (called Poisson chaos). By using this method, the author proves that the Fourth Moment Theorem holds true even without those strict, magical assumptions. The result is a precise mathematical guarantee: if your random variable has a finite fourth moment (meaning it doesn't get infinitely wild), the distance between it and a perfect bell curve is bounded by a specific formula involving the number 15.6 and the fourth moment. This removes the need for the old, restrictive rules and confirms that the bell curve is a much more robust friend to random data than previously thought.

The Story of the Messy Data and the Magic Filter

Imagine you are trying to predict the weather. You have a giant bucket of data points representing raindrops, wind gusts, and temperature spikes. In the world of math, this bucket is often modeled by something called a Poisson process. Think of a Poisson process like a rainstorm where drops fall randomly. Sometimes you get a huge cloudburst (a big jump), and sometimes it's just a drizzle. Mathematicians love these processes because they describe so many real-world things, from the number of emails you get in an hour to the arrival of buses at a stop.

Now, imagine you have a specific function, let's call it F, that takes all these random raindrops and turns them into a single number. Maybe F is the total amount of water collected in a bucket after an hour. The big question is: Does F look like a nice, smooth bell curve?

For decades, mathematicians knew that if you calculated a specific score for F (called the fourth moment) and it equaled 3, then F was almost certainly a bell curve. This is the "Fourth Moment Theorem." However, there was a catch. To prove this, previous mathematicians had to assume that F was "nice" in a very specific way. They had to assume that F and its "derivatives" (which are like measuring how much F changes if you add one more raindrop) were all perfectly bounded and behaved themselves. It was like saying, "This rule works, but only if you promise the raindrops won't be too big or too weird."

The problem is, in the real world, we can't always make that promise. What if the raindrops are huge? What if the data is messy? The old proofs broke down. They relied on techniques that required the data to be smooth, like a polished marble statue. But real data is more like a pile of jagged rocks.

The Magic Filter: A Martingale Core

Enter Guangqu Zheng and his "martingale core." Imagine you have a pile of jagged rocks (your messy data) and you want to see if they form a perfect circle (the bell curve). You can't just look at the whole pile; it's too chaotic. So, you build a special filter.

This filter is a martingale. In simple terms, a martingale is a way of looking at your data step-by-step. Imagine you are watching a movie, but you only see one frame at a time. As you watch more frames, your understanding of the movie gets better and better. A martingale is a sequence of guesses that gets closer and closer to the truth as you get more information.

But here's the genius part: Zheng didn't just build any filter. He built a filter that respects the "chaos" of the data. In math, there's a concept called Poisson chaos, which is like a family of patterns. If your data belongs to a certain "chaos family," you want your filter to keep it in that same family while it cleans it up.

Zheng's "martingale core" works by counting the exact number of events in small, specific boxes. Instead of looking at the whole storm at once, he looks at how many raindrops fell in a tiny square, then a slightly bigger square, and so on. He creates a sequence of these counts that gets finer and finer.

Here is the magic trick:

It keeps the family: When he filters the data, the result stays in the same "chaos family." It doesn't get mixed up with other types of patterns.
It smooths the rocks: The filter turns the jagged, messy data into "step functions." Imagine taking your pile of jagged rocks and arranging them into neat, flat steps. This makes the data much easier to analyze mathematically.
It converges: As the steps get smaller and smaller, the filtered data gets closer and closer to the original messy data.

The Big Discovery

Using this magic filter, Zheng was able to prove something huge. He showed that you don't need those strict, magical assumptions about the data being "nice" anymore.

He proved that if your random variable F has a finite fourth moment (which basically means the data isn't infinitely wild), then the distance between F and a perfect bell curve is bounded.

The formula he found is:
Distance ≤ 15.6 × √(E[F⁴] − 3)

Let's break that down:

E[F⁴] − 3 is the "score" we mentioned earlier. If it's 0, the data is a perfect bell curve. If it's bigger, the data is further away.
15.6 is a constant number that acts as a safety margin.
The square root (√) means that as the score gets smaller, the distance shrinks very quickly.

This is a massive improvement over previous work. Before, to get a similar result, mathematicians had to assume the data was "locally bounded" (Assumption A and Aloc). Zheng's paper says: "Nope, we don't need that. As long as the fourth moment is finite, the rule holds."

Why This Matters

Think of it like a safety net. Before this paper, if you had a weird, messy dataset, you couldn't be sure if the bell curve rule applied unless you checked a bunch of extra, difficult conditions. Now, you have a simpler, more robust rule. If the fourth moment is finite, you can trust the bell curve approximation.

The paper also gives us a new way to look at the "derivatives" of the data. It proves that if the original data has a finite fourth moment, then all its "change rates" (the Malliavin derivatives) also have finite fourth moments. This is like saying if the whole mountain is stable, then every single rock on the mountain is also stable.

In short, Zheng built a bridge. On one side is the messy, real world of random events. On the other side is the perfect, predictable world of the bell curve. The bridge is made of "martingale cores" and "step functions," and it allows us to walk across without needing to pretend the world is perfect. We just need to know that the data isn't infinitely wild, and the math takes care of the rest.

The paper doesn't just suggest this; it proves it with rigorous mathematics. It removes the old, restrictive assumptions and replaces them with a single, clear condition: a finite fourth moment. This makes the Fourth Moment Theorem a much more powerful tool for scientists and statisticians who deal with the messy, unpredictable reality of the world.

Technical Summary: A Kolmogorov Fourth-Moment Bound on Poisson Chaos via a Martingale Core

Problem Statement
The paper addresses a gap in the quantitative central limit theory for functionals of Poisson random measures. While the "Fourth Moment Theorem" (established for Gaussian fields by Nualart and Peccati [NP05] and later extended to Poisson spaces by Döbler and Peccati [DP18a]) asserts that convergence of the fourth moment to 3 implies convergence in distribution to a standard normal, the known quantitative bounds in the Poisson setting relied on restrictive regularity assumptions. Specifically, the Kolmogorov distance bound derived in [DP18a] required Assumption A (integrability of specific Malliavin derivatives and products) and Assumption Aloc (local validity of Assumption A). These assumptions were acknowledged by their authors as likely artifacts of the proof techniques (Mecke-type identities and spectral representations) rather than intrinsic requirements for the fourth-moment phenomenon. The primary goal of this work is to establish a Kolmogorov distance bound for Poisson multiple integrals under the sole assumption of a finite fourth moment, thereby removing the need for Assumptions A and Aloc.

Methodology: The Finite-Count Martingale Core
The central methodological innovation is the construction of a "finite-count martingale core." The author observes that standard martingale approximations (e.g., based on dyadic partitions or spatial truncations) fail to simultaneously satisfy three necessary properties:

Preservation of the Poisson chaos order (staying within a fixed $C_q$ ).
Regularization of the kernel (ensuring boundedness and finite-measure support to satisfy Assumption A).
Convergence in $L^p$ (specifically $L^4$ ) to the original random variable.

To resolve this, the paper constructs a specific increasing filtration $(G_n)_{n \ge 1}$ generated by exact Poisson counts on a sequence of finite partitions of the underlying space $Z$ .

Construction: For a finite family of Poisson multiple integrals, the author defines a sequence of finite-dimensional subspaces $H_n \subset L^2(\mu)$ and associated orthogonal projections $P_n$ . The filtration $G_n$ is generated by the random variables $\eta(A_{n,k})$ , representing the exact number of points in disjoint cells $A_{n,k}$ of a refining partition.
Key Property: The conditional expectations $F_{j,n} = \mathbb{E}[F_j | G_n]$ possess a crucial structural advantage: they remain in the original Poisson chaos $C_{q_j}$ (unlike occupancy-based approximations which mix chaoses) and their kernels become bounded step functions with finite-measure support (satisfying the regularity conditions of [DP18a]).
Convergence: Theorem 1.7 proves that for any finite family of integrals in $L^p$ ( $p \ge 2$ ), these approximations converge in $L^p(\Omega)$ to the original variables while preserving the chaos structure.

Key Contributions and Results

Removal of Regularity Assumptions (Theorem 1.9):
The primary result is a Kolmogorov distance bound for a random variable $F$ in a Poisson chaos with unit variance and finite fourth moment. The paper proves:
$d_{\text{Kol}}(F, N) \le 15.6 \sqrt{\mathbb{E}[F^4] - 3}$
This bound matches the order and multiplicative constant of the result in [DP18a] but holds under the minimal assumption $F \in L^4(\Omega)$ . The proof leverages the martingale core to approximate $F$ by a sequence $F_n$ satisfying the regularity assumptions, applies the known bound to $F_n$ , and uses the lower semicontinuity of the Kolmogorov distance to pass to the limit.
Quantitative $L^4$ Estimates for Malliavin Derivatives (Theorem 1.8):
The paper establishes that if a Poisson multiple integral $F$ has a finite fourth moment, then all its iterated Malliavin derivatives $D^r F$ are also in $L^4(\Omega \times Z^r)$ . Specifically, it provides explicit bounds:
$\int_{Z^r} \mathbb{E}[|D^r_{z_1, \dots, z_r} F|^4] \mu(dz_1)\dots\mu(dz_r) \le a_{q,r} \kappa_4(F)$
This implies that the fourth integrability of the random variable forces the fourth integrability of its kernel, a non-trivial regularity result derived via the martingale core.
Graph Closure and Spectral Identities (Proposition 1.10):
The martingale core allows for the simultaneous approximation of products and carré-du-champ operators. The paper proves that the exact identities relating the fourth moment, the kernel, and the carré-du-champ operator (previously known only under regularity assumptions) hold for all $F \in L^4(\Omega) \cap C_q$ .

Significance and Claims
The paper claims to resolve a technical obstruction in the Poisson Malliavin-Stein method. By introducing the finite-count martingale core, the author demonstrates that the regularity conditions (Assumptions A and Aloc) used in [DP18a] are not intrinsic to the fourth-moment phenomenon but are merely technical conveniences that can be bypassed via approximation.

The result extends the "Fourth Moment Theorem" to its most natural setting: the sole assumption of a finite fourth moment.
The methodology provides a "constructive spectral closure," showing that the space of regular functionals (satisfying Assumption A) is dense in the graph norm involving $L^4$ norms of derivatives up to order $q$ .
The work achieves the same optimal rate ( $\sqrt{\kappa_4(F)}$ ) for the Kolmogorov distance as previously achieved for the Wasserstein distance in [DVZ18], but without requiring the more complex infinitesimal exchangeable-pairs construction for the Kolmogorov case. Notably, while [DVZ18] had already established the optimal Wasserstein bound under the sole assumption of a finite fourth moment, this paper specifically targets the more difficult Kolmogorov bound, which had previously required stronger regularity assumptions.

The paper does not propose new applications or future experimental directions but positions itself as a foundational technical improvement that removes unnecessary hypotheses from existing quantitative limit theorems.

The Story of the Messy Data and the Magic Filter

The Magic Filter: A Martingale Core

The Big Discovery

Why This Matters

Technical Summary: A Kolmogorov Fourth-Moment Bound on Poisson Chaos via a Martingale Core

More like this