Non-Gaussian Limits of Long Memory Random Fields

This direction is split into other directions: Wiener chaos expansion. Its last version stays readable and citable here.

Version 1 of 117 papers18 conceptsfrozen 2026-10-04

The Central Limit Theorem (CLT) is the bedrock of statistics, suggesting that the sum of many independent random variables eventually settles into a Gaussian "bell curve." However, this predictability breaks down in systems with long-range dependence (LRD)—where every data point retains a "memory" of distant points in time or space.

In these "long-memory" systems, non-linear transformations don't always converge to Gaussian distributions. Instead, they often converge to exotic, non-Gaussian limits known as Hermite-type or Rosenblatt distributions. Understanding why this happens requires a look at how we decompose randomness into its constituent parts.

The Chaos Decomposition

To analyze a complex non-linear function of a random process, researchers use a technique called Wiener chaos expansion (or Walsh chaos expansion in discrete settings). As Ivan Nourdin and Giovanni Peccati demonstrate in Stein's Method and Stochastic Analysis of Rademacher Functionals, any sufficiently smooth random variable can be represented as an infinite sum of orthogonal components.

Think of this like a Fourier series for randomness. Just as a complex sound wave can be broken down into simple sine waves of different frequencies, a non-linear stochastic process can be broken down into a series of Hermite polynomials. Each term in the series represents a different "order" of interaction with the underlying noise.

The Dominance of the Hermite Rank

The most critical insight in this field is the reduction principle. When we look at a process over a very large scale, we don't need the whole infinite series. Instead, the entire asymptotic behavior of the system is usually determined by just one term: the first non-zero term in the expansion.

This index—the Hermite rank—acts as a structural pivot point:

  1. If the rank is low, the correlations might still be weak enough for the CLT to hold.
  2. If the rank is high or the correlations are sufficiently "long," the system enters the realm of Non-Central Limit Theorems (NCLT).

Marianne Clausel, Murad S. Taqqu, and Ciprian A. Tudor have shown in Large scale behavior of wavelet coefficients of non-linear subordinated processes with long memory that this rank dictates whether a wavelet-transformed process will converge to a standard Gaussian or a more complex "generalized Hermite process."

From Euclidean Space to Hypersurfaces

While much of the early theory focused on simple time series, recent work has expanded these ideas to complex geometries. Andriy Olenko has pioneered the study of random fields on hypersurfaces, such as the surface of a sphere.

In Non-central limit theorems for functionals of random fields on hypersurfaces, Olenko proves that even on these curved manifolds, the Hermite rank remains the deciding factor. This is particularly relevant for fields like cosmology or global climate modeling, where data is naturally collected on a sphere and often exhibits long-range spatial correlations. By identifying the rank, researchers can derive explicit rates of convergence, telling us exactly how much data is needed before the statistical estimators stabilize, as explored by Vo Anh and Olenko in Non-central limit theorems and convergence rates.

Why the Limit Stays Random

A fascinating consequence of long-range dependence appears in Stochastic Partial Differential Equations (SPDEs). In Weak Convergence Approach for Parabolic Equations with Large, Highly Oscillatory, Random Potential, Guillaume Bal describes a sharp transition:

  • Short-range noise: The oscillations average out, and the system becomes deterministic (homogenization).
  • Long-range noise: The "memory" is so strong that the noise never fully averages out. The system remains inherently random, converging to a limit that is itself a Stochastic PDE.

This persistent randomness is captured through the same Wiener chaos expansion used in the Chaos expansion of 2D parabolic Anderson model by Jingyu Huang. By treating the solution as a sum of iterated integrals, researchers can bridge the gap between abstract probability and physical models like polymers moving through random environments.

Summary of Intuition

The transition from Central to Non-Central Limit Theorems is a transition from independence to influence.

  • The Tool: Wiener chaos/Hermite expansion breaks non-linear randomness into a hierarchy of interactions.
  • The Key: The Hermite rank identifies which level of interaction dominates at large scales.
  • The Result: In the presence of long memory, the "first" interaction is so powerful that it prevents the system from ever reaching a Gaussian state, forcing it instead into a Hermite-type distribution.

Papers in this version

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_non_gaussian_limits_of_long_memory_rando_a7a507eb,
  title = {Non-Gaussian Limits of Long Memory Random Fields},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-gaussian-limits-of-long-memory-random-fields/rdr_a7a507eb794c4a5a9ffdc46b5c079e4f},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_non_gaussian_limits_of_long_memory_rando_a7a507eb_v1,
  title = {Non-Gaussian Limits of Long Memory Random Fields},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-gaussian-limits-of-long-memory-random-fields/rdr_a7a507eb794c4a5a9ffdc46b5c079e4f/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}