On the density of the supremum of nonlinear SPDEs

This paper establishes that the supremum of the solution to a one-dimensional nonlinear stochastic partial differential equation, which encompasses models like the stochastic heat and linearized Cahn-Hilliard equations, admits a density with respect to Lebesgue measure by employing Malliavin calculus and the Bouleau-Hirsch criterion to overcome challenges related to the solution's argmax set and the nondegeneracy of its Malliavin derivative.

The Gist

Imagine you are watching a very complex, chaotic weather system on a small, flat island. This system is governed by a set of rules (the equation in the paper) that mix two things:

1. Predictable forces: Like wind blowing steadily or the ground heating up (these are the smooth parts of the equation).
2. Pure chaos: Like sudden, unpredictable lightning strikes or random gusts of wind that hit the island at any second and any spot (this is the "noise" or $\dot{W}$ in the math).

The scientists in this paper are trying to answer a very specific question about this weather system: "If we look at the highest point the temperature reaches on the island over a whole day, can we describe the probability of that peak being any specific number?"

In math-speak, they are asking if the "supremum" (the absolute maximum) of this random field has a density.

The Big Picture: Why is this hard?

Think of the weather system as a giant, wiggly, 3D surface that changes every millisecond.

• Pointwise: If you pick one specific spot and one specific time, it's relatively easy to say, "There's a 10% chance it's 20°C." We know how to describe the probability for a single point.
• The Peak: But the "highest point" is tricky. It's like trying to find the tallest person in a crowd of a million people who are all jumping up and down randomly. The "winner" (the maximum) changes constantly. The location of the peak is random, and the value of the peak is random.

Usually, when you take the maximum of a random process, it can get "stuck" on certain values or behave in ways that make it impossible to assign a smooth probability curve (a density). The authors want to prove that for this specific type of weather system, the peak does have a smooth, well-behaved probability curve.

The Three Scenarios (Regimes)

The paper looks at three different types of "islands" or physical laws:

1. The Heat Equation (Regime i & ii): Imagine a thin metal plate. Heat spreads out smoothly, but random sparks hit it. The plate has edges that are either frozen (Dirichlet) or insulated (Neumann).
2. The Cahn-Hilliard Equation (Regime iii): Imagine a mixture of oil and water trying to separate. This involves a "fourth-order" rule, which is much more rigid and complex than simple heat spreading. It's like the material has a "stiffness" that resists bending.

The authors prove that for all three of these scenarios, the "highest point" of the system behaves nicely.

The Secret Weapon: Malliavin Calculus

How did they prove this? They used a tool called Malliavin Calculus.

The Analogy:
Imagine you are trying to find the highest point on a foggy mountain range. You can't see the whole mountain at once.

• Standard Calculus is like looking at the slope at your feet.
• Malliavin Calculus is like having a special pair of glasses that lets you see how the entire mountain would shift if you tweaked the "random wind" (the noise) just a tiny bit.

It asks: "If I change the random lightning strike at 2:00 PM by a tiny amount, how much does the location of the highest peak move?"

If the peak moves significantly whenever you tweak the noise, then the peak is "sensitive" and "alive." This sensitivity is what allows us to draw a smooth probability curve (the density). If the peak didn't move at all when you changed the noise, the probability would be stuck on a single number, and you couldn't draw a curve.

The Two Main Hurdles

To prove the peak has a density, the authors had to clear two hurdles:

1. The "Smoothness" Hurdle (Continuity)
They had to prove that the "sensitivity" of the peak changes smoothly.

• Analogy: Imagine the sensitivity is a rubber sheet stretched over the mountain. They had to prove that this rubber sheet doesn't have any sharp tears or holes. If it's smooth, we can do the math. They proved that even though the noise is chaotic, the "sensitivity" of the system is actually quite smooth and predictable.

2. The "Non-Degeneracy" Hurdle (The Argmax Problem)
This is the hardest part. They had to prove that the peak is never in a place where the system is "dead" or "frozen."

• The Problem: At the very edges of the island (the boundaries) or at the very start of the day (time zero), the rules might force the system to be static. If the highest point happened to be exactly at the edge at the start, the "sensitivity" might be zero, and the math would break.
• The Solution: They proved that the highest point almost certainly never happens at the edge or at the start time. It happens somewhere in the middle, where the chaos is active. Because the peak is always in the "active zone," the sensitivity is always non-zero, and the probability density exists.

The "Initial Condition" Twist

There was one special case. If the starting temperature map (the initial condition) was perfectly flat or had a weird shape, the peak might have started at the edge.
To fix this, the authors added a small assumption: The starting map must be "rough" enough (specifically, Hölder continuous).

• Analogy: If you start with a perfectly flat pancake, the first bump might happen right at the edge. But if you start with a slightly bumpy pancake, the highest point will almost certainly pop up somewhere in the middle where the random sparks hit. This "roughness" guarantees the peak stays in the safe zone.

The Conclusion

In simple terms, this paper says:

"Even though these random, chaotic systems are incredibly complex, the highest point they reach is not a 'freak accident' that defies probability. It follows a smooth, predictable pattern. We can calculate the odds of the peak being 50°C, 51°C, or 52°C, just like we can for a single point."

This is a big deal because it allows scientists and engineers to better predict extreme events (like the highest flood level or the hottest temperature) in complex systems, knowing that these extremes follow a reliable mathematical law.

Technical Summary

Here is a detailed technical summary of the paper "On the Density of the Supremum of Nonlinear SPDEs" by Karali, Stavrianidi, Tzirakis, and Zoubouloglou.

1. Problem Statement

The paper investigates the existence of a probability density function (with respect to the Lebesgue measure) for the supremum of solutions to a class of one-dimensional nonlinear stochastic partial differential equations (SPDEs).

The specific equation studied is:
$$\frac{\partial u}{\partial t} = -\kappa \frac{\partial^4 u}{\partial x^4} + \rho \frac{\partial^2 u}{\partial x^2} + b(u) + \sigma(u)\dot{W}$$
posed on a bounded spatial domain $[0, 1]$ and time interval $[0, T]$, where $\dot{W}$ is space-time white noise. The authors analyze three distinct regimes based on the parameters $\kappa$ and $\rho$:

1. Regime (i): $\kappa = 0, \rho = 1$ (Stochastic Heat Equation with Dirichlet boundary conditions).
2. Regime (ii): $\kappa = 0, \rho = 1$ (Stochastic Heat Equation with Neumann boundary conditions).
3. Regime (iii): $\kappa > 0$ (Linearized Stochastic Cahn–Hilliard equation with Neumann boundary conditions).

The Core Challenge: While the existence of densities for pointwise evaluations $u(t, x)$ of SPDE solutions is well-established using Malliavin calculus, proving the existence of a density for the supremum $\sup_{(t,x) \in K} u(t, x)$ is significantly more difficult. This is because the supremum is a non-smooth functional of the random field, and standard criteria for density existence (like the Bouleau-Hirsch criterion) require verifying non-degeneracy of the Malliavin derivative specifically on the set where the maximum is attained (the argmax set).

2. Methodology

The authors employ Malliavin Calculus, specifically adapting the Bouleau-Hirsch criterion (developed by Nualart and Vives) for suprema of random fields. To prove the existence of a density, they must verify three conditions for the random variable $F = \sup_{(t,x) \in K} u(t, x)$:

1. Integrability: $F \in \mathbb{D}^{1,2}$ (Malliavin differentiable with square-integrable derivative).
2. Regularity of the Derivative: The Malliavin derivative process $D_{s,y}u(t,x)$ must possess a continuous version as an $L^2$-valued process.
3. Non-degeneracy: The Malliavin derivative must be non-degenerate almost surely on the set of maximizers ($S_K = \arg\max_{(t,x) \in K} u(t,x)$).

Key Technical Strategies:

• Kolmogorov-Type Continuity: To handle the supremum inside the expectation (Condition 2), the authors prove anisotropic Kolmogorov continuity estimates for the Malliavin derivative $D_{s,y}u(t,x)$. This involves deriving sharp $L^p$ bounds for the Malliavin derivative using the mild solution formulation and Picard iterations.
• Handling the Argmax Set (Condition 3):
  • Unlike Gaussian fields where the maximizer is often unique, the nonlinear SPDE solutions here do not guarantee a unique maximizer.
  • The authors avoid proving uniqueness. Instead, they show that the Malliavin derivative is non-degenerate on every compact subset of the interior $(0, T) \times (0, 1)$.
  • They then prove that the argmax set almost surely does not intersect the boundary $\{t=0\}$ or the spatial boundaries (in the Dirichlet case) under specific assumptions on the initial data.
• Small Ball Estimates: A crucial part of the proof involves establishing "small ball" estimates for the process $\gamma(t, x) = \int_0^T \int_0^1 |D_{s,y}u(t,x)|^2 dy ds$. They show that the probability of $\gamma(t, x)$ being close to zero decays rapidly, ensuring non-degeneracy.
• Initial Data Assumption: For the global supremum over $[0, T] \times [0, 1]$, they impose a local Hölder continuity condition on the initial datum $u_0$ near its maximizer (Assumption 1.2). This ensures the solution immediately "escapes" the initial condition, preventing the supremum from being attained at $t=0$ where the Malliavin derivative vanishes.

3. Key Contributions

• First Result for Nonlinear SPDEs: This is the first paper to establish the existence of a density for the supremum of nonlinear SPDEs. Previous works (e.g., Dalang and Pu) were limited to linear Gaussian cases.
• Generalization to Fourth-Order Equations: The results extend beyond the stochastic heat equation to include the linearized stochastic Cahn-Hilliard equation (Regime iii), which involves a fourth-order spatial derivative.
• Refined Analysis of the Argmax Set: The paper provides a novel approach to the non-degeneracy condition without requiring the almost sure uniqueness of the maximizer, a significant hurdle in non-Gaussian settings.
• Regularity of the Malliavin Derivative: As a byproduct, the authors establish Hölder continuity properties for the Malliavin derivative of the solution as an $L^2$-valued process, matching the regularity of the underlying field.

4. Main Results

The paper proves the following theorems:

• Theorem 1.3 (Regimes i & ii): For the stochastic heat equation (Dirichlet or Neumann), assuming Lipschitz coefficients and uniform ellipticity of $\sigma$:

  • The supremum over any compact subset $K \subset (0, T] \times (0, 1)$ (Dirichlet) or $K \subset (0, T] \times [0, 1]$ (Neumann) admits a density.
  • If the initial condition $u_0$ satisfies the local Hölder condition near its maximizer, the supremum over the entire domain $[0, T] \times [0, 1]$ admits a density.

• Theorem 1.4 (Regime iii): For the linearized stochastic Cahn-Hilliard equation ($\kappa > 0$) with Neumann boundary conditions:

  • The supremum over any compact subset $K \subset (0, T] \times [0, 1]$ admits a density.
  • The proof follows a similar structure to Theorem 1.3 but utilizes different kernel estimates (Green's function for the bi-Laplacian operator) and different Hölder exponents (time regularity is $3/8$vs$1/4$ for the heat equation).

5. Significance

• Theoretical Advancement: The work bridges a gap between the theory of pointwise densities and supremum densities for SPDEs, moving from linear/Gaussian models to nonlinear, non-Gaussian settings.
• Methodological Robustness: The techniques developed (specifically the handling of the argmax set without uniqueness and the derivation of Malliavin derivative continuity) provide a template for analyzing suprema of other complex stochastic systems.
• Applications: Understanding the density of the supremum is critical for applications involving exit probabilities (e.g., calculating the probability that a physical system exceeds a critical threshold) and risk management in stochastic modeling. The ability to handle nonlinear drift and diffusion terms makes these results applicable to a wider range of physical phenomena modeled by SPDEs.

In summary, the paper successfully overcomes the technical difficulties associated with the non-Gaussian nature of nonlinear SPDEs and the lack of smoothness of the supremum functional, providing a rigorous proof for the existence of a density for the maximum of the solution.