[Paper Review] The nonlinear stochastic heat equation with rough initial data:a summary of some new results

Le Chen, Robert C. Dalang|arXiv (Cornell University)|Oct 5, 2012

Stochastic processes and financial applications16 references3 citations

TL;DR

This paper establishes existence, uniqueness, and moment estimates for the nonlinear stochastic heat equation on ℝ with rough initial data, including non-tempered measures and Dirac delta functions, via a novel Picard iteration approach that avoids Gronwall's lemma. It derives sharp upper and lower bounds for all p-th moments (p ≥ 2), determines growth indices, and proves Hölder continuity of the solution for t > 0 despite irregular initial conditions, with explicit formulas for the parabolic Anderson model's second moment and phase transitions in Lyapunov exponents.

ABSTRACT

This is a preliminary announcement of results in the PhD. thesis of the first author concerning the nonlinear stochastic heat equation in the spatial domain $\R$, driven by space-time white noise. A central special case is the parabolic Anderson model. The initial condition is taken to be a measure on $\R$, such as the Dirac delta function, but this measure may also have non-compact support and even be non-tempered (for instance with exponentially growing tails). Existence and uniqueness is proved without appealing to Gronwall's lemma, by keeping tight control over moments in the Picard iteration scheme. Upper and lower bounds on all $p$-th moments $(p\ge 2)$ are obtained. These bounds become equalities for the parabolic Anderson model when $p=2$. The growth indices introduced by Conus and Khoshnevisan (2010) are determined and, despite the irregular initial conditions, Hölder continuity of the solution for $t>0$ is established.

Motivation & Objective

• To extend the existence and regularity theory of the nonlinear stochastic heat equation to initial data that are not necessarily bounded or tempered, including signed measures with exponential growth.
• To establish moment estimates for all p ≥ 2 without relying on Gronwall’s lemma, using tight control in Picard iteration.
• To determine the exact growth indices (Lyapunov exponents) for the solution, particularly for the parabolic Anderson model.
• To prove Hölder continuity of the solution in space and time for t > 0, even when the initial data is highly irregular.
• To provide explicit formulas for the second moment and two-point correlation functions in key cases, such as Lebesgue measure and Dirac delta initial data.

Proposed method

• A modified Picard iteration scheme is employed to construct the solution, with moment bounds controlled directly rather than via Gronwall’s inequality.
• The initial data is assumed to be a signed Borel measure μ satisfying |μ| * Gν(t,·)(x) < ∞ for all t > 0 and x ∈ ℝ, allowing non-compact and non-tempered supports.
• For the parabolic Anderson model (ρ(u) = λu), explicit formulas are derived for the second moment and two-point correlation functions using the heat kernel and hyperbolic functions.
• Growth indices (Lyapunov exponents) are computed via asymptotic analysis of moment growth, with phase transitions identified based on parameters like β and λ.
• Sample path regularity is established using Hölder continuity estimates on the stochastic integral and heat kernel components, distinguishing between the initial condition term J₀ and the iterated integral term I.
• The analysis leverages the Jordan decomposition of μ and properties of the heat kernel Gν(t,x) = (2πνt)⁻¹ᐟ² exp(−|x|²/(2νt)) to control spatial and temporal behavior.

Experimental results

Research questions

• RQ1 Can the nonlinear stochastic heat equation be solved with initial data that are not bounded or tempered, such as measures with exponentially growing tails or Dirac delta functions?
• RQ2 What are the exact upper and lower bounds for the p-th moments of the solution when p ≥ 2, and how do they behave for the parabolic Anderson model?
• RQ3 What are the precise values of the upper and lower Lyapunov exponents (growth indices) for the solution, and how do they depend on the initial data and the nonlinearity parameter λ?
• RQ4 Does the solution exhibit Hölder continuity in space and time for t > 0 even when the initial data is rough (e.g., a measure or discontinuous function)?
• RQ5 What phase transitions occur in the growth behavior of the second moment as a function of the initial data's decay rate β, and are the derived bounds sharp?

Key findings

• For the parabolic Anderson model with μ = δ₀, the second moment satisfies E[|u(t,x)|²] = (λ²/(2ν)) e^(λ⁴t/(4ν)) Φ(λ²√(t/(2ν))) Gν/₂(t,x), with explicit asymptotic behavior.
• The upper and lower Lyapunov exponents for the second moment are equal and given by λ²/2 when β ≥ λ²/(2ν), and by βν/2 + λ⁴/(8βν) when 0 < β ≤ λ²/(2ν), proving a sharp phase transition.
• For the parabolic Anderson model with initial data μ(dx) = e^(-β|x|)dx, the exact second moment Lyapunov exponent is λ²/2 when β ≥ λ²/(2ν), and βν/2 + λ⁴/(8βν) otherwise, confirming sharpness of bounds.
• The solution is almost surely Hölder continuous in time with index 1/4− and in space with index 1/2− for t > 0, even when the initial data is a signed Borel measure satisfying the integrability condition.
• For initial data μ = Lebesgue measure or μ = δ₀, explicit formulae are derived for the two-point correlation functions, matching known results but now under broader initial data assumptions.
• The paper proves that the solution is not continuous at t = 0 when μ = δ₀, as ||I(t,x)||₂² → ∞ at x = 0 and → 0 elsewhere as t → 0⁺, indicating a singularity in the initial stochastic integral.