[Paper Review] Function-valued stochastic convolutions arising in integrodifferential equations

Anna Karczewska|ArXiv.org|Dec 26, 2004

Stochastic processes and financial applications19 references3 citations

TL;DR

This paper establishes sufficient conditions for the existence of function-valued stochastic convolutions in the context of stochastic integrodifferential equations that interpolate between the heat and wave equations. By analyzing the covariance structure of spatially homogeneous Wiener noise and leveraging Fujita's fundamental solutions, the author proves the $ L^2_v $-valuedness of stochastic convolutions via Hilbert-Schmidt operator estimates, providing a foundational framework for nonlinear stochastic Volterra equations.

ABSTRACT

We study stochastic convolutions providing by fundamental solutions of a class of integrodifferential equations which interpolate the heat and the wave equations. We give sufficient condition for the existence of function--valued convolutions in terms of the covariance kernel of a noise given by spatially homogeneous Wiener process.

Motivation & Objective

• To establish sufficient conditions for the existence of function-valued stochastic convolutions arising from stochastic integrodifferential equations that interpolate the heat and wave equations.
• To extend the framework of stochastic convolution theory from wave and heat equations to Volterra-type equations, which lack a known fundamental solution representation.
• To analyze the regularity and well-posedness of stochastic integrals of the form $ \int_0^t P(t-s)*b(u(s))\,dW(s) $ in $ L^2_v $-spaces.
• To provide a rigorous foundation for studying nonlinear stochastic Volterra equations, a topic previously unaddressed in the function-valued solution context.

Proposed method

• The paper uses fundamental solutions $ P_\alpha(t,x) $ derived from Fujita’s work on fractional-in-time integrodifferential equations with $ 1 \leq \alpha \leq 2 $.
• It defines the stochastic convolution as $ \mathcal{K}_R(t,u)\eta(x) = \int_\mathbb{R} u(x-y)\eta(x-y) P_\alpha(t,y)\,dy $, restricted to $ [-R,R] $, and analyzes its Hilbert-Schmidt norm.
• The existence of the $ L^2_v $-valued process is established by proving uniform boundedness of the Hilbert-Schmidt norm $ \|\mathcal{K}_R(t,u)\|_{L_{HS}} $ as $ R \to \infty $.
• The proof relies on asymptotic decay estimates of $ P_\alpha(t,x) $ as $ |x| \to \infty $, showing $ P_\alpha(t,x) \to 0^+ $ faster than any polynomial, ensuring integrability.
• The analysis uses the spectral measure $ \nu $ of the spatially homogeneous Wiener process and assumes a hypothesis (H) on the covariance kernel to control the noise structure.
• The limit $ R \to \infty $ is justified by showing the tail integral $ \int_{\widetilde{R}}^R P_\alpha(t,y)\,dy $ is uniformly bounded by a finite $ \mathcal{M}_{\widetilde{R}} $.

Experimental results

Research questions

• RQ1 Under what conditions is the stochastic convolution $ \int_0^t P(t-s)*b(u(s))\,dW(s) $ well-defined as an $ L^2_v $-valued process?
• RQ2 How can the Hilbert-Schmidt norm of the stochastic convolution operator be uniformly bounded as the spatial domain expands to infinity?
• RQ3 What role does the asymptotic decay of the fundamental solution $ P_\alpha(t,x) $ play in ensuring the existence of function-valued solutions?
• RQ4 Can the framework used for wave and heat equations be extended to Volterra-type equations, which lack a standard fundamental solution representation?
• RQ5 What is the impact of the spatial covariance structure of the noise on the regularity of the stochastic convolution?

Key findings

• The stochastic convolution $ \mathcal{K}_R(t,u) $ is well-defined as a Hilbert-Schmidt operator from $ \mathcal{H}_W $ to $ L^2_v $ for all finite $ R $, under hypothesis (H).
• The Hilbert-Schmidt norm $ \|\mathcal{K}_R(t,u)\|_{L_{HS}} $ is uniformly bounded in $ R $, with the bound depending on $ \|u\|_{L^2_v} $ and a constant $ C $, ensuring convergence as $ R \to \infty $.
• The asymptotic behavior $ P_\alpha(t,x) \sim \frac{B_\alpha |x|^{(\alpha-1)/(2-\alpha)}}{\exp[A_\alpha |x|^{2/(2-\alpha)}]} $ as $ |x| \to \infty $ ensures integrability of the tail, which is essential for the limit.
• The limit $ R \to \infty $ is justified by showing $ \int_{\widetilde{R}}^R P_\alpha(t,y)\,dy \to \mathcal{M}_{\widetilde{R}} < \infty $, allowing extension to the full real line.
• The result holds under the assumption that $ b $ is Lipschitz continuous and the noise is spatially homogeneous with a covariance kernel satisfying hypothesis (H).
• The paper establishes the first framework for function-valued solutions to nonlinear stochastic Volterra equations, filling a key gap in the literature.