2
$\begingroup$

Consider a 1-dimensional stochastic heat equation on $[0, 1]$, with boundary conditions of Neumann's type: \begin{equation}\left\{ \begin{aligned} &\partial_t u(t, x) = \frac{1}{2}\partial_x^2 u(t, x) - U(u(t)) + \dot W(dt, dx), \\ &\partial_x u(t, 0) = \partial_x u(t, 1) = 0, \\ &u(0, x) = v(x). \end{aligned}\right. \end{equation} Here $U$ is a map from $L^2[0, 1]$ to $L^2[0, 1]$. We assume that $U$ is bounded and Lipschitz continuous, thus the solution exists and is continuous (let the initial condition be continuous in $x$). My question involves two aspects.

(i) If $U = \frac{1}{2}DV$ for some Frechet differentiable real function $V$ on $L^2[0, 1]$ and $D$ is the Frechet derivative operator, then it is easy to expect that the infinite measure on $C[0, 1]$: $$\nu(\mathrm du) = \exp(-V(u))\mu(\mathrm du)$$ is invariant for $u(t)$, where $\mu$ is the measure of a process $\{w_x\}_{x \in [0, 1]}$ such that $w_x - w_0$ is a standard BM on $[0, 1]$ and $w_0$ subjects to the Lebesgue measure on $\mathbb R$. However I can not find reference for this result (since it deals with an infinite invariant measure). Could anyone give me a reference (or a proof) for this?

(ii) If $U = \frac{1}{2}DV + B$, where $V$ is as same as the above, and $B$ satisfies $E_\nu \langle Df, B \rangle_{L^2[0, 1]} = 0$ for all $f \in C_b^1(H)$, would it change the invariant measure?.

Remark: I know a reference [T. Funaki, Nagoya Math. J., 1983] which treats (i) for local nonlinear term, i.e., $V(u)$ is replaced with $\{V(u(x))\}_{x \in [0, 1]}$. It is proved through a broken-line approximation. If no other results maybe I have to go in this way.

$\endgroup$

1 Answer 1

0
$\begingroup$

I check it with the standard Garlerkin method and confirmed that it is right, in both cases (i) and (ii).

Discribe the proof briefly (under (ii)):

Take a CONS of $H$ as $h_1 = 1$ and $h_k(x) = \cos[(k-1)\pi x]$.

First notice that if $V(u) = V^\dagger(\langle u, h_1 \rangle, \ldots, \langle u, h_N \rangle)$ and $\langle B(u), h_k \rangle = B_k^\dagger(\langle u, h_1 \rangle, \ldots, \langle u, h_N \rangle)$ for some $V^\dagger$, $B_k^\dagger \in C_b^2(\mathbb R^N)$ for $k \leq N$, and $B_k^\dagger = 0$ for $k > N$, then it becomes an $N$-dimensional diffusion and a infinite-dimensional OU process, and the result is obvious.

Then take $V_N^\dagger$ and $B_{N, k}^\dagger$ be the marginal expectation, i.e., for $x \in \mathbb R^N$ and $k \leq N$, $$V_N^\dagger(x) = \int_{\mathbb R^\infty} V\left(\sum_{1}^N x_kh_k + \sum_{N+1}^\infty y_kh_k\right)\prod_{N+1}^\infty \Phi_k(y_k)dy_k, $$ $$B_{N, k}^\dagger(x) = e^{2V_N^\dagger(x)} \int_{\mathbb R^\infty} U_k\left(\sum_{1}^N x_kh_k + \sum_{N+1}^\infty y_kh_k\right)\prod_{N+1}^\infty \Phi_k(y_k)dy_k, $$ where $U_k = e^{-2V}\langle B, h_k \rangle$ and $\Phi_k(x) = \sqrt{\lambda_k\pi^{-1}}e^{-\lambda_kx^2}$. This finite-dimensional approximation protect the condition (ii), so we have the invariant measure, due to the previous step. Finally take the limit.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.