🔢 mathematics

Large Deviations for Stochastic Evolution Equations in the Critical Variational Setting

This paper establishes the large deviation principle for quasilinear stochastic evolution equations in a critical variational setting using a weak convergence approach, requiring no monotonicity or compact embedding assumptions and thereby resolving a 15-year open problem regarding equations like the 2D Navier-Stokes and Boussinesq systems with gradient noise on unbounded domains.

Original authors: Esmée Theewis, Mark Veraar

Published 2026-02-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Esmée Theewis, Mark Veraar

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 trying to predict the weather. You have a super-complex computer model that simulates wind, rain, and temperature. Usually, this model runs perfectly. But what if you add a tiny bit of "static" or random noise to the system? Maybe a tiny, unpredictable gust of wind that your sensors can't quite catch.

The Big Question: If you make that random noise smaller and smaller (almost zero), does the weather pattern settle down into a single, predictable path? And if it does deviate from that path, how likely is that to happen?

This paper is about answering that question for a very specific, very difficult type of mathematical weather model called a Stochastic Evolution Equation.

Here is the breakdown of what the authors, Esmée Theewis and Mark Veraar, achieved, using some everyday analogies.

1. The Problem: The "Rigid" Rules

For decades, mathematicians had a set of strict rules (called "monotonicity") to prove that these weather models behave nicely. Think of these rules like a traffic light system.

The Old Way: To prove the traffic flows smoothly, you had to assume every car (the math variables) strictly obeyed the red and green lights. If a car tried to run a red light (a non-monotonic behavior), the old math tools broke down.
The Limitation: Many real-world phenomena, like fluid dynamics (water swirling in a river or air moving around a plane), don't always follow these strict "traffic light" rules. They are chaotic. Also, previous math tools struggled if the river was infinitely long (unbounded domains) or if the noise was "gradient noise" (noise that depends on how fast the water is changing, not just where it is).

2. The New Tool: The "Flexible Framework"

The authors used a new, more flexible framework developed recently (the "critical variational setting").

The Analogy: Instead of a rigid traffic light system, imagine a flexible dance floor.
In this new framework, the "dancers" (the math equations) don't need to follow strict traffic rules. They can move more freely, as long as they don't explode off the dance floor.
This framework allows for "gradient noise" (noise that reacts to the speed of the dance) and works even if the dance floor is the size of the entire universe (unbounded domains).

3. The Main Achievement: The "Large Deviation Principle" (LDP)

The paper proves the Large Deviation Principle. Let's break that down:

The Scenario: You have a main path (the "deterministic" path) that the system wants to follow.
The Noise: You add a tiny bit of randomness (noise).
The Result: The system usually stays very close to the main path. But sometimes, it takes a wild, unlikely detour.
The LDP: This is a mathematical formula that tells you exactly how unlikely those wild detours are. It's like a "risk score" for the weather model. It tells you: "The chance of the wind blowing in this crazy direction is $10^{-100}$ ."

Why is this hard?
Usually, to prove this, mathematicians need to use a trick called "compactness." Imagine trying to prove that a crowd of people will stay in a small room. If the room is small and closed (compact), it's easy. But if the room is an infinite hallway (unbounded domain), people can wander off forever, and the math gets messy.

The Authors' Trick: They didn't rely on the room being small. Instead, they used a "strong approximation" technique. Think of it like using a high-resolution camera to track the dancers. Even in an infinite hallway, if you zoom in close enough and track the movement step-by-step, you can prove they aren't wandering off randomly.

4. The "Skeleton" Equation

To calculate the "risk score" (the rate function), the authors had to solve a special "Skeleton Equation."

The Analogy: Imagine you want to know the most efficient way to drive from New York to LA. The "Skeleton Equation" is the perfect, frictionless, no-traffic version of that drive.
The authors proved that this "perfect drive" always exists and is unique, even with their flexible rules. This was a major hurdle because, in the past, proving this "perfect drive" existed required those strict "traffic light" rules they were trying to avoid.

5. The Real-World Impact: Fluid Dynamics

The paper isn't just abstract math; it solves a 15-year-old mystery in fluid dynamics.

The Navier-Stokes Equations: These are the equations that describe how water and air move. They are famous for being incredibly difficult (one of the "Millennium Prize Problems").
The Breakthrough: The authors proved that their new "risk score" formula works for 2D Navier-Stokes equations with gradient noise on unbounded domains.
Why it matters: Previous attempts to solve this had a "gap" in the logic (like a missing step in a recipe). The authors filled that gap. Now, we can mathematically guarantee how likely it is for a fluid to behave strangely, even if the fluid is in an infinite ocean and the noise depends on the fluid's speed.

Summary

Think of this paper as upgrading the GPS for chaotic systems.

Before: The GPS only worked if the roads were straight, the cars were obedient, and the map was small.
Now: The GPS works on winding, chaotic roads, with rebellious cars, and even if the map is infinite.
The Result: We can now accurately predict the odds of a "wild detour" in complex systems like weather, ocean currents, and plasma, even when the math gets messy and the noise is tricky.

The authors didn't just fix a small bug; they rewrote the engine to handle a whole new class of difficult problems that were previously unsolvable.

1. Problem Statement

The paper addresses the Large Deviation Principle (LDP) for solutions to small-noise stochastic evolution equations (SEEs) of the form:
$dY^\varepsilon(t) = -A(t, Y^\varepsilon(t)) dt + \sqrt{\varepsilon} B(t, Y^\varepsilon(t)) dW(t)$
where $\varepsilon \in (0, 1]$ is a small parameter, $W$ is a cylindrical Brownian motion, and the equation is set in a critical variational framework.

Key Challenges Addressed:

Critical Nonlinearities: The coefficients $A$ (drift) and $B$ (diffusion) allow for "critical" growth rates, meaning the nonlinearities grow at the limit of what standard Sobolev embeddings allow.
Absence of Monotonicity: Unlike classical variational settings, the authors do not assume the weak monotonicity condition (often required for well-posedness).
Gradient Noise: The diffusion coefficient $B$ is allowed to have flexible growth, including terms involving gradients of the solution (transport noise), which typically breaks standard energy estimates.
Unbounded Domains: The framework does not require a compact embedding of the space $V$ into $H$ (Gelfand triple), allowing the theory to apply to equations on unbounded spatial domains (e.g., $\mathbb{R}^d$ ).

2. Methodology

The authors employ the Weak Convergence Approach (introduced by Budhiraja and Dupuis) to prove the LDP. This approach avoids the difficult discretization and contraction principle arguments used in earlier works. Instead, it relies on verifying two main conditions regarding the "skeleton equation" (the deterministic control problem associated with the LDP).

Core Technical Steps:

The Critical Variational Setting: The authors utilize the framework developed in [6] (Agresti & Veraar). This setting decomposes coefficients into a quasilinear part ( $A_0, B_0$ ) and a semilinear part ( $F, G$ ). It relies on complex interpolation spaces $V_\beta = [V^*, V]_\beta$ to handle critical growth conditions where the Lipschitz constants depend polynomially on interpolation norms.
Well-posedness of the Skeleton Equation:
- The skeleton equation is a deterministic ODE in infinite dimensions: $u' = -A(t, u) + B(t, u)\psi$ .
- Proving global well-posedness for this equation is non-trivial due to the lack of monotonicity and the presence of gradient noise.
- Strategy: The authors first prove well-posedness for a linearized version using the method of continuity and maximal regularity theory. They then use a fixed-point argument to establish local existence and a blow-up criterion combined with energy estimates to extend this to global existence.
Weak Convergence Verification:
- Condition (ii) (Continuity): They prove that the map from the control $\psi$ to the skeleton solution $u_\psi$ is continuous with respect to the weak topology of $L^2$ . This requires delicate estimates using critical interpolation inequalities and a combination of deterministic and stochastic Gronwall inequalities.
- Condition (iii) (Stochastic Continuity): They prove that the perturbed stochastic solution converges in probability to the skeleton solution as $\varepsilon \to 0$ . This involves a cut-off argument to handle the lack of global boundedness in the $H$ -norm and utilizes a stochastic Gronwall lemma to control the growth of the diffusion term.

3. Key Contributions

Removal of Monotonicity: The paper establishes LDP results without assuming the weak monotonicity condition, which is a significant generalization over previous variational approaches.
Handling Gradient Noise: The framework explicitly accommodates diffusion coefficients with gradient noise (e.g., $(b \cdot \nabla)u$ ), which is crucial for fluid dynamics models.
Unbounded Domains: By avoiding the requirement for compact embeddings ( $V \hookrightarrow H$ ), the results apply to equations on unbounded domains (like $\mathbb{R}^2$ ), a setting where previous compactness-based arguments fail.
Resolution of Open Problems: The authors explicitly resolve a gap in the literature regarding the 2D Navier-Stokes equations with gradient noise on unbounded domains, a problem that had remained open for over 15 years.

4. Main Results

Theorem 2.6 (Main Theorem):
Under the critical variational assumptions (Assumption 2.2) and a coercivity condition (2.7), the family of solutions $(Y^\varepsilon)$ satisfies the LDP on the maximal regularity space $MR(0, T) = C([0, T]; H) \cap L^2(0, T; V)$ .
The rate function $I: MR(0, T) \to [0, \infty]$ is given by:
$I(z) = \frac{1}{2} \inf \left\{ \int_0^T \|\psi(s)\|_U^2 ds : \psi \in L^2(0, T; U), \, z = u_\psi \right\}$
where $u_\psi$ is the unique strong solution to the skeleton equation driven by $\psi$ .

Applications (Section 5):
The theory is applied to a broad class of fluid dynamics and reaction-diffusion equations, including:

2D Navier-Stokes equations with gradient noise on arbitrary open sets (bounded or unbounded).
3D Tamed Navier-Stokes equations with gradient noise.
Cahn-Hilliard, Allen-Cahn, and Swift-Hohenberg equations.
Boussinesq equations and various reaction-diffusion systems.

5. Significance

This work represents a major advancement in the theory of stochastic partial differential equations (SPDEs).

Unification: It unifies the treatment of quasilinear and semilinear equations under a single "critical" variational umbrella, removing the need for case-by-case monotonicity assumptions.
Fluid Dynamics: It provides the first rigorous LDP proof for the 2D Navier-Stokes equations with gradient noise on unbounded domains. Previous attempts (e.g., [45]) were incomplete due to gaps in proving the compactness of sublevel sets for the rate function in this specific setting.
Methodological Innovation: The use of maximal regularity theory and specific approximation techniques to replace compactness arguments opens the door for analyzing SPDEs in settings where traditional compact embedding methods are unavailable.

In summary, Theewis and Veraar have successfully extended the Large Deviation Principle to a much broader and more physically relevant class of stochastic evolution equations, solving long-standing open problems in the mathematical theory of fluid dynamics.