---
title: Non-uniform Berry–Esseen Bounds
url: https://www.emergentmind.com/topics/non-uniform-berry-esseen-bounds
type: topic
---

# Non-uniform Berry–Esseen Bounds

A non-uniform Berry–Esseen (BE) bound is a refinement of classical central limit theorem (CLT) rates, providing pointwise bounds on the Kolmogorov distance between the distribution of a normalized sum or functional and the normal law, with an explicit dependence on the approximation point $z$ rather than only a global supremum. Such bounds take the form $|\mathbb{P}(T \le z)-\Phi(z)| \le$ (error term decaying in $z$), and are crucial for precise quantitative control in the moderate and large deviations regimes, especially for rare-event probability estimation and statistical inference involving extreme quantiles. The literature provides a spectrum of techniques—Fourier smoothing, Stein's method, Malliavin calculus, and concentration inequalities—producing a variety of non-uniform BE bounds across classical, dependent, functional, martingale, and self-normalized settings.

## 1. Fundamental Forms and Historical Milestones

The archetype for non-uniform BE bounds arises from sums of independent real random variables $X_1,\dots,X_n$ with $E[X_i]=0$, $B^2 = \sum_{i=1}^n \operatorname{Var}(X_i)$, and $A = \sum_{i=1}^n E|X_i|^3<\infty$. The original bound of Nagaev (1965/1976) and subsequent sharp versions state
\[
|P(S/B>z) - P(Z>z)| \le C_{\mathrm{nu}} \frac{A}{B^3} \frac{1}{1+z^3}, \qquad z \ge 0
\]
where $S = \sum X_i$ and $Z\sim N(0,1)$. The uniform (classical) BE bound lacks the explicit $z$-dependence and is typically, for iid, $C_u n^{-1/2} E|X_1|^3$. The best-known uniform constant $C_u$ is within 15% of optimality, but $C_{\mathrm{nu}}$ has only recently been reduced to $4.5$ [1302.0516].

Advancements include:
- Smoothing inequalities delivering optimal constants [1301.2828, 1302.0516].
- Non-uniform bounds for functionals (chaos, martingales, Poisson, self-normalized, exchangeable pair regimes).
- Replacement of polynomial decay $1/(1+|z|^3)$ by exponential $e^{-cz^2}$ as achievable under additional tail or moment conditions.

## 2. Methodological Paradigms

### 2.1 Fourier Analytic Smoothing and Truncation

The classical approach by Nagaev involves a blend of truncation, Cramér–Esséen tilting for moderate deviations, and exponential tail inequalities beyond the uniform-case range. The introduction of smoothing inequalities (cf. Prawitz filter, symmetric kernels) enables direct passage from characteristic function bounds to non-uniform error estimates, bypassing delicate explicit density/Edgeworth expansions, and yielding improvements such as
\[
|P(S/B > z)-P(Z > z)| \le \frac{C A}{B^3 (1+z^3)}
\]
with $C$ as small as $10$–$12$ for suitable filtering and numerical optimization [1301.2828, 1302.0516].

### 2.2 Stein's Method and Concentration

Stein's method, when coupled with concentration inequalities, provides a flexible framework for both normal and non-normal approximation in classical and dependent settings. The central technical element is the construction and control of the solution $f_z$ to the Stein equation,
\[
f_z'(x) - x f_z(x) = 1\{x \leq z\} - \Phi(z)
\]
with explicit bounds on $f_z$, $f_z'$, and higher derivatives allowing one to obtain non-uniform Kolmogorov error bounds via tailored moment and concentration controls [1109.0741, 2502.10804, 2506.17061].

### 2.3 Malliavin–Stein Theory

In functionals of Gaussian, Poisson, or Rademacher processes (e.g., multiple Wiener–Itô integrals), the Malliavin–Stein approach gives non-uniform BE bounds depending on Malliavin-derivative-based quantities such as
\[
R(F) := \sqrt{E\left(1 - \langle D F, -D L^{-1} F\rangle_{\mathfrak{H}}\right)^2}
\]
and controls the error via
\[
|P(F\le z) - \Phi(z)| \le R(F)\{P(|F| > |z|/2) + 2e^{-z^2/4}\}
\]
with $F$ in an appropriate Sobolev space $\mathbb{D}^{1,2}$ [2409.01550, 2409.09439].

## 3. Key Theorems and Typical Non-uniform Bounds

### 3.1 Sums of Independent Variables

| Setting                    | Pointwise Bound                                       | Reference          |
|----------------------------|------------------------------------------------------|--------------------|
| Classical iid sum          | $\frac{A}{B^3 (1+z^3)}$                              | [1302.0516, 1301.2828] |
| Exponential decay (3rd mom)| $C n^{-1/2} E|X_1|^3 e^{-z/2}$                       | [1109.0741]        |
| Higher moments $p$         | $C_p \sum E|X_i|^p / [(1+|z|)^p \sigma^p ]$          | [1109.0741]        |

### 3.2 Exchangeable Pairs

For an exchangeable pair $(W, W')$ with
\[
E[W'|W] = (1-\tau)W + R
\]
and under finite $2p$-moment, unbounded $\Delta:=W-W'$, the general non-uniform BE bound reads
\[
|\Pr[W \le z] - \Phi(z)| \le \frac{C(p)\{1+E|W|^{2p}\}}{(1+|z|)^p} \cdot \mathcal{E}(a)
\]
with $\mathcal{E}(a)$ collecting $L^2$ deviations from ideal conditional variance, regression remainders, and large/small jump controls [2502.10804, 1909.13477].

### 3.3 Functionals of Gaussian and Poisson Processes

For $F \in \mathbb{D}^{1,2}$, $E[F]=0$, $\mathrm{Var}\,F=1$,
\[
|P(F \le x) - \Phi(x)| \le C (1+|x|)^k \cdot \Delta(F)
\]
with $\Delta(F) = (E[1-\langle D F, -D L^{-1} F\rangle_{\mathfrak{H}}]^2)^{1/2}$. For multiple Wiener–Itô integrals $F=I_q(f)$, $\Delta(F)\le C_q\sqrt{E[F^4]-3}$ and explicit exponential tail controls are available for $P(|F|>x)$ [2409.09439, 2409.01550].

### 3.4 Martingales

For a martingale difference array satisfying the conditional Bernstein condition,
\[
|\P(S_n \le x)-\Phi(x)| \le C (1 + x^2) (\varepsilon |\ln\varepsilon| + \delta/(1+|x|)) \exp(-\hat{x}^2/2)
\]
with model-specific expressions for $\hat{x}$ in terms of $\varepsilon, \delta$ controlling tail and quadratic variation deviations [1608.05217, 2104.14063].

## 4. Applications and Model-Specific Instantiations

- **Multiple Wiener–Itô integrals:** Non-uniform bounds decay exponentially in $|z|$, critical for CLTs in fixed Wiener chaos [2409.01550, 2409.09439].
- **Exponential functionals of Brownian motion:** Quantitative non-uniform bounds track the deviation of the normalized functional from normality in settings with heavy tailed or log-normal upper tails [2409.01550].
- **Spin Model Macroscopic Observables:** In mean-field Curie–Weiss and $N$-vector models, non-uniform bounds for magnetization or squared spin length achieve optimal $n^{-1/2}$ uniform rates with an additional decay in $z$ [2502.10804, 2506.17061].
- **Martingale Regression and Self-normalization:** Adapted regression statistics and self-normalized CLTs inherit polynomial or exponential tail decay via non-uniform BE inequalities, sidestepping strong moment or independence assumptions [1608.05217, 2104.14063].
- **Studentized U-Statistics:** Necessary correction terms of exponential type are introduced to avoid failures in naive non-uniform bounds due to vanishing denominators, ensuring valid rates for $t$-statistic and higher-degree kernels [2303.08619].
- **Weakly Dependent Sequences:** Stationary Markov chains, dynamical systems, and products of random matrices allow non-uniform BE and Edgeworth expansion with rate $O(n^{-1/2}(1+|x|)^{-m})$ under general cumulant-derivative controls [2210.07204].

## 5. Technical Proof Elements and Optimality

### 5.1 Stein–Malliavin Integration by Parts

In Gaussian and Poisson functionals, the key representation
\[
P(F \leq z) - \Phi(z) = E[f_z'(F)(1 - \langle D F, -D L^{-1} F\rangle)] - E[f_z(F)]E[F]
\]
enables a split of the error over small and large $|F|$ via tailored bounds for $f_z$ and $f_z'$. The Cauchy–Schwarz step and concentration inequalities enable exponential or polynomial tail decay, with constants coming from precise operator norms or Malliavin derivatives [2409.01550, 2409.09439].

### 5.2 Moment Truncation and Smoothing

Smoothing inequalities using compactly supported filters applied to the characteristic function produce directly the necessary $1/(1+z^3)$ factor and uniform constants, as in
\[
|P(S>Bz)-\Phi(z)| \le \frac{C A}{B^3 (1+z^3)},
\]
with further refinements tied to the chosen filter, truncation scales, and analysis of remainder terms in the Edgeworth expansion [1302.0516, 1301.2828].

## 6. Optimality, Limitations, and Extensions

- The $(1+|z|)^k$ prefactor is optimal whenever only a finite $2k$th moment is assumed for $F$, and cannot in general be replaced by exponential decay unless stronger sub-Gaussian cores or concentration hold [2409.09439].
- In functionals of Poisson and Rademacher variables, second-order Poincaré inequalities enable more granular control involving Malliavin operators up to order 2, permitting explicit decomposition of the error into leading and higher-order components [2409.09439].
- For Studentized statistics, naive non-uniform bounds may fail due to rare events where the denominator is vanishingly small. Here, adding a (sharp, necessary) exponentially small correction term restores validity [2303.08619].
- In exchangeable-pair-based normal approximation, unbounded difference settings are now included with explicit control on large jumps, polynomial tail factors, and remainders [2502.10804, 1909.13477].

## 7. Summary Tables: Representative Non-uniform Berry–Esseen Bounds

| Class              | Error Form                 | Decay in $z$         | Key Condition                          | Reference          |
|--------------------|---------------------------|----------------------|----------------------------------------|--------------------|
| iid sum            | $C n^{-1/2}(1+|z|)^{-3}$  | Polynomial           | $E|X|^3<\infty$                        | [1302.0516]        |
| Sums (exp decay)   | $C e^{-z/2}$              | Exponential          | $E|X|^3<\infty$                        | [1109.0741]        |
| Martingale         | $C (1+z^2) e^{-\hat{x}^2/2}$ | Exponential      | Conditional Bernstein condition         | [1608.05217]       |
| Functionals (Gaussian/Poisson) | $C (1+|z|)^k \Delta(F)$       | Polynomial  | $E|F|^{2k}<\infty$, $F\in \mathbb{D}^{1,2}$   | [2409.09439]       |
| Exchangeable pair  | $C (1+E|W|^{2p})/(1+|z|)^p$ | Polynomial         | $E|W|^{2p}$ finite                     | [2502.10804]       |

## References

- "More on the nonuniform Berry--Esseen bound" [1302.0516]
- "On the nonuniform Berry--Esseen bound" [1301.2828]
- "Non-uniform Berry-Esseen bounds via Malliavin-Stein method" [2409.01550]
- "Non-uniform Berry--Esseen bounds for Gaussian, Poisson and Rademacher processes" [2409.09439]
- "Improved nonuniform Berry--Esseen-type bounds" [1109.0741]
- "Non-uniform Berry--Esseen bounds for exchangeable pairs with applications..." [2502.10804]
- "Non-uniform Berry-Esseen Bound by Unbounded Exchangeable Pair Approach" [1909.13477]
- "Nonuniform Berry-Esseen bounds for Studentized U-statistics" [2303.08619]
- "Nonuniform Berry-Esseen bounds for martingales..." [1608.05217]
- "Non-uniform Berry-Esseen theorem and Edgeworth expansions..." [2210.07204]
- "Nonuniform Berry-Esseen bound for self-normalized martingales" [2104.14063]
- "Non-uniform bounds for non-normal approximation via Stein's method..." [2506.17061]

Source: https://www.emergentmind.com/topics/non-uniform-berry-esseen-bounds