Langevin equations with non-Gaussian thermal noise: Valid but superfluous

This paper demonstrates that for a classical Brownian oscillator, the generalized Langevin equation with linear dissipation satisfies the Jarzynski equality at finite times only if the thermal noise is Gaussian (beyond seventh-order perturbation), rendering non-Gaussian variants superfluous for evaluating properties beyond linear or quadratic noise dependence.

The Gist

Imagine you are trying to predict how a tiny, jittery particle (like a speck of dust in water) moves around. Scientists use a famous mathematical recipe called the Langevin Equation to describe this motion.

For over a century, everyone has assumed that the "noise" or random jiggling hitting the particle follows a very specific, bell-curve pattern called Gaussian noise. Think of this like a perfectly smooth, predictable distribution of raindrops: most are average size, a few are tiny, a few are huge, but they follow a strict, symmetrical rule.

However, in the real world, things aren't always perfectly smooth. Sometimes the "rain" might be a bit lumpy or irregular (non-Gaussian). For a long time, scientists have wondered: Can we use the same Langevin recipe if the noise is lumpy instead of smooth?

This paper, written by Alex V. Plyukhin, answers that question with a surprising twist: You can use the recipe, but it's pointless.

Here is the breakdown using simple analogies:

1. The "Perfect" vs. "Approximate" Recipe

The author distinguishes between two ways we use this equation:

• The Exact Case: If the physics of the system is perfectly simple (like a specific model where the water molecules are all identical and behave linearly), the noise is naturally Gaussian. In this case, the recipe works perfectly for everything.
• The Approximate Case: Most of the time, we use the recipe as a shortcut (an approximation) for complex systems. In these complex systems, the noise might actually be "lumpy" (non-Gaussian).

2. The "Short-Term Memory" Test

To test if the recipe works, the author didn't just wait to see if the particle settled down after a long time (which is the usual test). Instead, he looked at what happens during a very short, specific event: a quick "pulse" that changes the stiffness of the particle's environment, like a sudden squeeze.

He used a famous rule in physics called the Jarzynski Equality. Think of this rule as a "truth detector." It says that if you calculate the average "work" done on the particle in a specific way, the result must equal 1. If your math gives you anything other than 1, your recipe is broken.

3. The "Seven-Step" Limit

The author ran the math through a "lumpy noise" recipe and checked the truth detector at every step of the process.

• Steps 1 through 7: The recipe worked perfectly! The "truth detector" read 1, even though the noise was lumpy.
• Step 8 and beyond: The recipe started to fail. The "truth detector" only read 1 again if the noise was perfectly smooth (Gaussian). If the noise was lumpy, the result was wrong.

4. The Big Conclusion: "Superfluous"

This leads to the paper's main point, which is summarized in the title: "Valid but Superfluous."

• Valid: The equation with lumpy noise isn't "wrong" in a way that breaks physics immediately. It works fine for simple things.
• Superfluous (Useless): The only things the equation can calculate correctly with lumpy noise are simple, straight-line (linear) or square (quadratic) relationships.
  • The Analogy: Imagine you have a fancy, high-tech calculator that can handle complex, weird numbers. But, you discover that it only gives you the right answer for simple addition and multiplication. If you try to use it for complex division, it fails.
  • Since the simple things (addition/multiplication) don't actually care if the numbers are weird or smooth, you might as well just use the standard calculator (Gaussian noise). There is no benefit to using the "lumpy" version because it doesn't give you any new or different correct answers for the things it can calculate.

The Takeaway

If you want to study complex, "lumpy" noise effects, you cannot just use the standard Langevin equation. You would need a much more complicated, higher-level equation that the paper suggests doesn't exist in the simple form we usually use.

So, the paper concludes: Don't bother trying to use the standard Langevin equation with non-Gaussian noise. It's like trying to use a bicycle to fly; it might roll fine on the ground (for simple things), but it won't get you where you need to go for complex tasks, and you'd be better off just using a car (the Gaussian model) for the tasks the bike can actually do.

Technical Summary

Technical Summary: Langevin equations with non-Gaussian thermal noise: Valid but superfluous

Problem Statement
The generalized Langevin equation (GLE) with linear dissipation is a standard framework for describing open mesoscopic systems. While it is overwhelmingly common to assume the thermal noise $\xi(t)$ in the GLE is a Gaussian stochastic process, this assumption is often phenomenological or derived from specific models (e.g., the Caldeira-Leggett model) where the bath consists of linearly coupled harmonic oscillators. In more general microscopic derivations, such as those involving mode coupling or ideal gas baths, the resulting noise is manifestly non-Gaussian.

A central tension exists in the literature: while Gaussian noise guarantees that all moments of the system relax to correct equilibrium values and satisfies fluctuation theorems, it is unclear whether non-Gaussian noise is fundamentally inconsistent with the GLE. Previous studies have shown that GLEs with non-Gaussian noise fail to recover correct equilibrium values for higher-order moments (e.g., $\langle v^4 \rangle$) at asymptotically long times, yet it has not been rigorously proven that Gaussianity is necessary for the consistency of the GLE in its general form, particularly for non-ergodic systems. Furthermore, recent arguments (e.g., by Speck and Seifert) suggest that fluctuation theorems might hold for non-Markovian processes regardless of noise statistics. This paper investigates whether the GLE with non-Gaussian noise is intrinsically inconsistent or merely limited in its applicability.

Methodology
The author assesses the validity of the GLE with non-Gaussian noise by testing its consistency with the Jarzynski equality (JE) at finite times, rather than relying on asymptotic equilibrium conditions or assuming ergodicity.

1. System Definition: A classical Brownian oscillator with mass $m$ and time-dependent stiffness $k(t)$ is considered. The system interacts with a thermal bath at temperature $T$.
2. Protocol: The stiffness is perturbed by a rectangular pulse of duration $\tau$, switching from $k_0$ to $k$ and back. This constitutes a cyclic process where the free energy difference $\Delta F = 0$.
3. Theoretical Framework: The system is governed by the GLE with linear dissipation and an arbitrary noise statistics satisfying the fluctuation-dissipation relation (FDR) for the second moment. The work $W$ done on the system during the process is calculated.
4. Perturbative Analysis: The average exponential work $\langle e^{-W/T} \rangle$ is evaluated perturbatively in powers of the process duration $\tau$. The solution to the GLE is expressed using Laplace transforms and relaxation functions. The average is computed in two steps:
   • First, averaging over the initial equilibrium distribution of the system's coordinate and velocity.
   • Second, averaging over the noise realizations.
5. Consistency Criterion: The Jarzynski equality requires $\langle e^{-W/T} \rangle = 1$ for any $\tau$. The paper expands this average in a Maclaurin series in $\tau$ and analyzes the coefficients $c_n$. For the JE to hold, all coefficients for $n \geq 1$ must vanish.

Key Results
The analysis reveals a distinct threshold in the order of the expansion where the noise statistics become critical:

• Orders $\tau^1$ through $\tau^7$: The coefficients $c_1$ through $c_7$ depend only on pair (two-time) correlations of the noise and its derivatives. Since these correlations are fixed by the fluctuation-dissipation relation (FDR) regardless of the noise distribution, the coefficients vanish identically for any stationary noise statistics (Gaussian or non-Gaussian). Thus, the GLE is consistent with the JE up to the seventh order in $\tau$ unconditionally.
• Order $\tau^8$ and higher: Starting at the eighth order, the coefficients involve higher-order moments of the noise (e.g., the fourth moment $\langle \xi^4 \rangle$) and higher-order correlations that are not determined by the FDR.
  • The coefficient $\langle c_8 \rangle$ vanishes if and only if $\langle \xi^4 \rangle = 3\langle \xi^2 \rangle^2$, a property specific to Gaussian distributions.
  • Higher-order coefficients (e.g., $\langle c_9 \rangle$, $\langle c_{10} \rangle$) involve mixed moments (e.g., $\langle \xi^3 \dot{\xi} \rangle$) that vanish only if the noise process is fully Gaussian.
  • The paper argues that if the noise were non-Gaussian, the non-vanishing higher-order cumulants would prevent the coefficients from vanishing, thereby violating the Jarzynski equality.

Main Conclusion and Significance
The paper concludes that the GLE with non-Gaussian noise is not inconsistent per se, but it is superfluous for evaluating properties that depend on noise statistics beyond the second order.

• Validity Range: The GLE is a valid approximation for calculating properties that are linear or quadratic in the noise (and its derivatives), as these depend only on the second-order correlations fixed by the FDR. In this regime, the specific statistics of the noise (Gaussian vs. non-Gaussian) are irrelevant.
• Limitation: It is perturbatively inconsistent to use a GLE derived to a specific order (typically second order in a small parameter like the mass ratio) to evaluate properties involving higher-order noise moments. If one requires the calculation of such properties (like the full exponential work average for finite $\tau$), the GLE with non-Gaussian noise fails to reproduce the correct physical results (specifically, the fluctuation theorems) unless the noise is exactly Gaussian.
• Implication: The Gaussianity of Langevin noise is not a fundamental requirement for the existence of the GLE, but rather a necessary condition for the equation to be self-consistent when applied to non-linear functionals of the noise. If the noise is non-Gaussian, the GLE (1) is an incomplete description; a more appropriate description would require a generalized equation of higher perturbation order involving nonlinear dissipation terms.

The author emphasizes that this conclusion holds for both ergodic and non-ergodic systems, as the derivation relies on finite-time consistency with the Jarzynski equality rather than asymptotic thermalization. The work clarifies that while non-Gaussian noise can be derived microscopically, using the standard linear GLE to study its effects on higher-order statistics is a methodological error, rendering the non-Gaussian extension of the standard GLE redundant.