1
$\begingroup$

I'm wondering if there might be an explicit solution for the following linear PDE in two space dimensions $(x_1,x_2)$ on the whole space $\mathbb{R}^2$: $$ \partial_t f = {div} \left [\left( \begin{array}{rr} 1/4 & 0 \\ 0 & 1 \\ \end{array}\right) \nabla f + \left( \begin{array}{rr} 1/4 & -4 \\ 4 & 1 \\ \end{array}\right)xf \right ], $$ with $x = (x_1,x_2) \in \mathbb{R}^2,$ $t >0$ and $f=f(t,x_1,x_2).$ For this equation it is known that there is sufficient regular and positive, unique solution if we impose a positive initial state $f_0 \in L^1(\mathbb{R}^2).$ The problem is similar to the diffusion equation, for which, as is well known, there is a fundamental solution on the whole space. The equation can also be written as: $$ \partial_t f = \frac{1}{4} \partial_{x_1 x_1}^2f + \partial_{x_2 x_2}^2f + \left(\frac{1}{4}x_1 - 4x_2 \right) \partial_{x_1} f + \left(4x_1 + x_2 \right) \partial_{x_2} f + \frac{5}{4}f.$$ The (normalized) steady-state $f_{\infty}$, solution of the corresponding elliptic equation $$ 0 = {div} \left [\left( \begin{array}{rr} 1/4 & 0 \\ 0 & 1 \\ \end{array}\right) \nabla f + \left( \begin{array}{rr} 1/4 & -4 \\ 4 & 1 \\ \end{array}\right)xf \right ]$$ is known, it would be $$ f_{\infty}(x_1,x_2) = \frac{1}{2 \pi} {exp}\left(-\frac{1}{2}(x_1^2+x_2^2)\right).$$ Unfortunately I am not that very well versed in PDEs. So are there any results on explicit solutions of such equations on the whole space in 2D? I strongly assume the solution should be very similar to one of the heat equation.

I would be very grateful for any help!

$\endgroup$

1 Answer 1

2
$\begingroup$

The solution to the linear PDE is just the PDF corresponding to $\mathcal{N}(\mu_t, \Sigma_t)$ where $$ \mu_t = \exp(A t) x \;, \quad \text{and} \quad \Sigma_t = 2 \int_0^t \exp(A (t-s) \begin{bmatrix} 1/4 & 0 \\ 0 & 1 \end{bmatrix} \exp(A^T (t-s)) ds $$ where $A= -\begin{bmatrix} 1/4 & -4 \\ 4 & 1 \end{bmatrix}$.

This solution can be easily obtained by using the correspondence between SDE/PDE. In particular, the corresponding SDE is $$ d X_t = A X_t + \sqrt{2} \begin{bmatrix} 1/2 & 0 \\ 0 & 1 \end{bmatrix} d B_t $$ where $B_t$ is a standard $2$-dimensional Brownian motion. This SDE is a non-symmetric Ornstein-Uhlenbeck process due to the presence of a skew term in the drift, but since the SDE is linear, it can be explicitly solved to obtain the transition distribution $\mathcal{N}(\mu_t, \Sigma_t)$. It is also straightforward to verify that $\mathcal{N}(\mu_t, \Sigma_t) \to \mathcal{N}(0,I_{2\times2})$ as $t \to \infty$.

Standard references for this are:

Metafune, G.; Pallara, D.; Priola, E., Spectrum of Ornstein-Uhlenbeck operators in $L ^{p}$ spaces with respect to invariant measures, J. Funct. Anal. 196, No. 1, 40-60 (2002). ZBL1027.47036.

Pavliotis, Grigorios A., Stochastic processes and applications. Diffusion processes, the Fokker-Planck and Langevin equations, Texts in Applied Mathematics 60. New York, NY: Springer (ISBN 978-1-4939-1322-0/hbk; 978-1-4939-1323-7/ebook). xiii, 339 p. (2014). ZBL1318.60003.

$\endgroup$
1
  • 1
    $\begingroup$ Thank you very much for the quick and helpful answer. The reference in particular will certainly help me further. $\endgroup$
    – kumquat
    Commented May 28, 2022 at 4:33

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.