Renormalisation of Singular SPDEs with Correlated Coefficients

This paper establishes the local well-posedness of the generalized parabolic Anderson model and the $\phi^{K+1}_2$-equation on the two-dimensional torus with random, noise-correlated coefficients by proving that naive renormalisation fails due to variance blow-up and instead demonstrating convergence through the use of random renormalisation functions supported by novel stochastic estimates.

The Gist

Here is an explanation of the paper "Renormalisation of Singular SPDEs with Correlated Coefficients" using simple language and creative analogies.

The Big Picture: Predicting the Unpredictable

Imagine you are trying to predict the weather, but the atmosphere isn't just a blank canvas; it's made of a strange, shifting material that changes how wind and rain move through it. Furthermore, the material itself is reacting to the wind.

This paper tackles a specific type of mathematical problem called a Singular Stochastic Partial Differential Equation (SPDE).

• Singular: The equations involve "white noise," which is like static on a radio that is infinitely loud and chaotic at every single point. It's too messy to solve with standard math.
• Stochastic: It involves randomness.
• Correlated Coefficients: This is the paper's main twist. Usually, mathematicians assume the "material" (the coefficients) and the "noise" (the weather) are independent. But in this paper, the authors study what happens when the material and the noise are best friends—they are linked. If the noise spikes, the material changes instantly.

The Problem: The "Variance Explosion"

In the world of these equations, to get a sensible answer, mathematicians have to perform a trick called Renormalisation.

The Analogy: The Leaky Bucket
Imagine trying to measure the water level in a bucket that has a hole in the bottom (the noise).

1. The Old Way (Deterministic): If the hole is fixed and predictable, you can just subtract a constant amount of water to get the real level. It's like saying, "Okay, I lose 1 cup an hour, so I'll just add 1 cup back in my calculation."
2. The New Problem (Correlated): In this paper, the hole isn't fixed. The size of the hole changes depending on how much water is currently splashing around.
3. The Disaster: The authors discovered that if you try to use the "Old Way" (subtracting a fixed number) when the hole is changing with the splash, your calculation goes crazy. The error doesn't just get big; it explodes to infinity. They call this "Variance Blow-up." It's like trying to balance a scale where the weights keep changing size based on how hard you push them.

The Solution: The "Smart, Localized" Fix

Since a fixed number doesn't work, the authors had to invent a new tool. Instead of subtracting a single constant (like "minus 5"), they subtract a function.

The Analogy: The Custom Tailor

• Old Method: You buy a "one-size-fits-all" suit. It fits okay in some places but is terrible in others.
• New Method: The authors act like a master tailor. They create a custom suit for every single point in the city (or the mathematical space).
  • If the noise is wild in one neighborhood, the "renormalisation function" adjusts the math specifically for that neighborhood.
  • If the noise is calm in another, the function adjusts differently there.

They call these Random Renormalisation Functions. They are "smart" because they look at the local environment (the specific coefficient field) and adjust the math instantly to cancel out the chaos.

How They Proved It: The "Graph Detective"

To prove that their new method actually works and doesn't explode, they had to do some heavy lifting with probability theory.

The Analogy: The Traffic Map
Imagine the math problem as a massive, tangled web of traffic.

• The "noise" is cars driving erratically.
• The "coefficients" are the road conditions (potholes, speed limits).
• The authors had to prove that even with the cars and roads interacting, the traffic flow (the solution) remains smooth and predictable.

They used a technique called the Hairer–Quastel Criterion.

• The Metaphor: Think of this as a set of traffic rules for a graph (a map of connections). They drew the math problem as a diagram with dots (vertices) and lines (edges).
• They checked if the "traffic" on these lines was heavy enough to cause a jam (infinity).
• By combining Heat Kernel Asymptotics (looking at how heat spreads to understand how the noise spreads) and Gaussian Integration by Parts (a fancy way of rearranging the math to cancel out the messy parts), they proved that their "Custom Tailor" method keeps the traffic flowing smoothly.

Why Does This Matter?

You might ask, "Who cares about these weird equations?"

1. Real Materials: In the real world, materials aren't perfect. A ferromagnet (like a magnet) or a porous rock has imperfections. If you try to model how heat or electricity moves through it, those imperfections often react to the flow itself. This paper gives us the math to model those real-world, messy interactions.
2. Statistical Mechanics: It helps physicists understand how systems behave near "critical points" (like water turning to ice), where tiny fluctuations can cause massive changes.
3. The Future: This is a "Step 0." It's the foundation. Now that they know how to handle the "simple" cases where noise and material are friends, they can build up to solve even more complex, 3D versions of these problems.

Summary in One Sentence

The authors solved a messy math problem where the rules of the game change based on the random noise itself, proving that you can't use a "one-size-fits-all" fix, but you can use a smart, custom-made mathematical adjustment to keep the solution from exploding.

Technical Summary

Here is a detailed technical summary of the paper "Renormalisation of Singular SPDEs with Correlated Coefficients" by Nicolas Clozeau and Harprit Singh, published in SIGMA (2026).

1. Problem Statement

The paper addresses the local well-posedness of two classes of singular Stochastic Partial Differential Equations (SPDEs) on the two-dimensional torus ($\mathbb{T}^2$) where the diffusion coefficient is random and correlated with the driving noise.

The two equations studied are:

1. The Generalised Parabolic Anderson Model (g-PAM):
   $$\partial_t u - \sum_{i,j} a_{ij}(x) \partial_i \partial_j u = \sum_{i,j} f_{ij}(u) \partial_i u \partial_j u + g(u)\xi$$
   where $\xi$ is spatial white noise.
2. The $\phi^{K+1}_2$-equation:
   $$\partial_t u - \sum_{i,j} a_{ij}(t, x) \partial_i \partial_j u = -u^K + \xi$$
   where $\xi$ is space-time white noise.

The Core Challenge:
In standard singular SPDE theory (e.g., Hairer's Regularity Structures), coefficients are typically deterministic or independent of the noise. Here, the coefficient field $a(x)$ is constructed as $A(h)$, where $h = \sigma * \xi + \mu$. This creates a correlation between the environment (diffusivity) and the noise.

• The Variance Blow-up Phenomenon: The authors demonstrate that if one attempts to use standard deterministic renormalisation constants (as is common in constant-coefficient settings), the variance of the solution (or the model) blows up as the regularisation parameter $\delta \to 0$. This occurs because the correlation induces a non-trivial dependence that deterministic counterterms cannot cancel.

2. Methodology

The authors employ the framework of Regularity Structures (Hairer, 2014), adapting it to handle variable, random coefficients.

A. Structural Assumptions

• The coefficient field is defined as $a = A(h)$, where $A$ is a smooth, uniformly elliptic matrix-valued function, and $h$ is a smoothed version of the noise ($h = \sigma * \xi + \mu$).
• The noise $\xi$ is mollified via $\xi_\delta = \xi * \rho_\delta$.

B. Random Renormalisation Functions

Instead of subtracting deterministic constants, the authors introduce random renormalisation functions $c_\delta(x)$ and $c^{ij}_\delta(x)$.

• These functions are derived from the asymptotic expansion of the heat kernel (Green's function) associated with the operator $\partial_t - \sum a_{ij}(x)\partial_i \partial_j$.
• Specifically, they utilize the leading-order term of the heat kernel (Levy's construction) to define kernels $G_x(y)$ and $G_x^i(y)$.
• The counterterms are defined as convolutions of these kernels with the mollifier:
  $$c_\delta(x) = \int G_x(y) (\rho_\delta * \rho_\delta)(y) dy$$
  This choice ensures that the counterterms adapt locally to the realization of the coefficient field $a(x)$.

C. Stochastic Estimates

The primary technical hurdle is proving that the "model" (the collection of stochastic objects required to solve the fixed-point problem) converges.

• Non-Wiener Chaos: Because the coefficients depend on the noise, the stochastic objects do not lie in a finite Wiener chaos. Standard moment equivalence arguments (used in deterministic coefficient settings) fail.
• Tools Used:
  1. Isserlis' Theorem: Used to expand high-order moments of Gaussian variables into sums of products of covariances.
  2. Gaussian Integration by Parts: Used to handle the derivatives of the random coefficients with respect to the noise.
  3. Hairer–Quastel Bounds: The authors apply and slightly modify the Hairer–Quastel criterion (originally for constant coefficients) to bound the resulting singular integrals. They represent these integrals as directed graphs and verify the "super-renormalisability" conditions (Assumption B.1/B.3 in the paper) to ensure convergence.

3. Key Contributions

1. Proof of Variance Blow-up for Deterministic Counterterms:
   The paper rigorously proves (Proposition 1.3) that for the g-PAM with correlated coefficients, using deterministic renormalisation constants leads to either mean or variance blow-up. This necessitates the use of random, state-dependent counterterms.

2. Construction of Random Renormalisation Functions:
   The authors explicitly construct the random counterterms $c_\delta(x)$ based on the local geometry of the coefficient field $a(x)$. They show that these functions converge to limits that depend on the realization of the noise, effectively "renormalising" the correlation.

3. Extension of Regularity Structures to Correlated Environments:
   The paper extends the solution theory of singular SPDEs to the case where the environment is not independent of the driving noise. This is a significant step towards understanding stochastic homogenisation in singular regimes.

4. Technical Innovation in Stochastic Estimates:
   The authors develop a robust method for bounding moments of models in correlated settings. By combining heat kernel asymptotics with the Hairer–Quastel criterion, they establish the necessary stochastic bounds without relying on the finite Wiener chaos structure.

4. Main Results

• Theorem 1.7 (g-PAM): Establishes local well-posedness for the g-PAM equation with random, correlated coefficients. It proves the existence of a random time $T>0$ and a unique solution $u$ such that the regularised solutions $u_\delta$ converge to $u$ in probability, provided the random renormalisation functions $c_\delta$ are used.
• Theorem 1.9 ($\phi^{K+1}_2$): Establishes local well-posedness for the $\phi^{K+1}_2$ equation under the same correlated setting. The solution is constructed using a fixed-point argument in weighted Hölder spaces, utilizing the generalized Hermite polynomials $H_K(u_\delta, c_\delta)$ for renormalisation.
• Convergence: In both cases, the limit solution is independent of the choice of the mollifier $\rho$, confirming the robustness of the renormalisation scheme.

5. Significance and Outlook

• Physical Relevance: The model describes physical systems where the medium's properties (diffusivity) are intrinsically linked to the external forcing (e.g., a ferromagnet with random bonds correlated to external fields, or population dynamics in a random environment).
• Stochastic Homogenisation: The authors position this work as a "step-0" towards the study of stochastic homogenisation of singular SPDEs. By providing a reference solution theory for the correlated case, they enable precise statements about how these singular equations behave under scaling limits.
• Future Directions: The paper suggests that while the current methods work for these "simplest" singular SPDEs, extending them to more singular equations (like $\phi^4_3$) would be extremely tedious. It calls for the development of more systematic tools (e.g., algebraic approaches) to handle correlated coefficients in higher dimensions or more complex geometries.

In summary, this paper resolves a critical gap in the theory of singular SPDEs by demonstrating that correlation between coefficients and noise fundamentally alters the renormalisation procedure, requiring random, spatially varying counterterms to achieve well-posedness.