Exact and numerical solution of stochastic Burgers equations with variable coefficients (Q827473)

The following equations \begin{align*} du&=(A(t)\partial_{zz}u+B(t)u\partial_zu+C(t)\partial_zu+D(t)u)\,dt+E(t)\partial_zu\,dW_t \tag{1}\\ du&=(A(t)\partial_{zz}u+B(t)u\partial_zu+C(t)\partial_zu+D(t)u)\,dt+E(t)\,dW_t \tag{2} \end{align*} with \(u(0,z)=\phi(z)\) for \(z\in\mathbb R\) are considered where \(A\), \(B\), \(C\), \(D\) and \(E\) are real-valued bounded continuous functions of the time variable \(t\), the function \(B\) is strictly positive and \(W\) is a one-dimensional Brownian motion. The solutions can be constructed by an ansatz \(u(t,z)=U(t,X^z(t))\) where \(U\) is a solution of a deterministic Burgers equation and \(X^z\) is a solution of a first or second order linear stochastic ordinary differential equation (LSDE) with the initial condition equal to \(z\). Next, the authors suggest a numerical algorithm for simulations of the above stochastic Burgers equations that includes a step with a stochastic mesh (that happens to be the numerical solutions of the stochastic ordinary differential equations LSDE) on which the deterministic equations are solved numerically via Euler or Runge-Kutta method (central difference in space, finite difference in time). When compared with the stochastic forward Euler method, the proposed stochastic mesh method has a smaller mean absolute error when tested on equations for which explicit solutions of the equations (1) and (2) are known.