Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I've managed to reduce a question in probability to the following simple looking PDE: $$ u_t = -t u_x + \frac{1}{2} u_{xx}, {\rm ~for~} x>0, \, t \in \mathbb{R} \;, $$ with a limiting initial condition: $$ u(x,t) \to 1, {\rm ~as~} t \to -\infty \;, $$ a boundary condition at $x=0$: $$ u_x(0,t) = \left\{ \begin{array}{lc} -2 a t u(0,t) \;, & t \le 0 \\ 0 \;, & t \ge 0 \end{array} \right. \;, $$ where $a > 0$ is a constant, and a suitable boundary condition as $x \to \infty$: either $u \to 1$ or $u$ is bounded.

All I want is an exact expression for $p = \lim_{t \to \infty} u(0,t)$, which represents the particular probability that I am after, but I don't know how to do this or if it's possible. I'm hoping there's some sort of method out there that I don't know about to derive $p$ without having to solve the whole boundary value problem. Any ideas?

A few comments:

1) I've solved it numerically (with finite differences) and $u(x,t)$ looks nice and smooth and only takes values between 0 and 1.

2) I don't expect the piecewise nature of the boundary condition at $x=0$ to be a major issue. Presumably we can solve up to $t=0$, pause, and then solve for $t>0$. For $t \le 0$, the boundary condition is a time-dependent Robin boundary condition, which I've never seen dealt with before.

3) To me the major problem seems to be the explicit time-dependency in the PDE and the boundary condition at $x=0$. I can get rid of the time-dependency in the PDE by defining $$ v(x,t) = {\rm e}^{\frac{1}{6} t^3 - t x} u(x,t) \;, $$ which gives $$ v_t = -x v + \frac{1}{2} v_{xx} \;, $$ $$ v_x(0,t) = \left\{ \begin{array}{lc} -(1+2a)tv(0,t) \;, & t \le 0 \\ -tv(0,t) \;, & t \ge 0 \end{array} \right. \;. $$

4) To the boundary value problem in $v$, I can't get separation of variables, or Laplace transforms in time, or some sort of half Fourier transform in space, to work.

5) Incidentally, $\lim_{t \to \infty} u(x,t)$, should be independent of $x$, that is, pointwise, $u(x,t)$ approaches the constant value $p$.

share|cite|improve this question
Perhaps you can take the laplace transform and then use the Final value theorem to find $f(\infty)$ – Navin Oct 29 '12 at 1:54
That's true, but I don't see how to usefully take the Laplace transform to begin with. – djws Oct 29 '12 at 8:42


Let $\begin{cases}v=x-\dfrac{t^2}{2}\\w=t\end{cases}$ ,

Then $\dfrac{\partial u}{\partial x}=\dfrac{\partial u}{\partial v}\dfrac{\partial v}{\partial x}+\dfrac{\partial u}{\partial w}\dfrac{\partial w}{\partial x}=\dfrac{\partial u}{\partial v}$

$\dfrac{\partial^2u}{\partial x^2}=\dfrac{\partial}{\partial x}\left(\dfrac{\partial u}{\partial v}\right)=\dfrac{\partial}{\partial v}\left(\dfrac{\partial u}{\partial v}\right)\dfrac{\partial v}{\partial x}+\dfrac{\partial}{\partial w}\left(\dfrac{\partial u}{\partial v}\right)\dfrac{\partial w}{\partial x}=\dfrac{\partial^2u}{\partial v^2}$

$\dfrac{\partial u}{\partial t}=\dfrac{\partial u}{\partial v}\dfrac{\partial v}{\partial t}+\dfrac{\partial u}{\partial w}\dfrac{\partial w}{\partial t}=-t\dfrac{\partial u}{\partial v}+\dfrac{\partial u}{\partial w}$

$\therefore-t\dfrac{\partial u}{\partial v}+\dfrac{\partial u}{\partial w}=-t\dfrac{\partial u}{\partial v}+\dfrac{1}{2}\dfrac{\partial^2u}{\partial v^2}$

$\dfrac{\partial u}{\partial w}=\dfrac{1}{2}\dfrac{\partial^2u}{\partial v^2}$

Let $u(v,w)=V(v)W(w)$ ,

Then $V(v)W'(w)=\dfrac{1}{2}V''(v)W(w)$




$\therefore u(v,w)=\int_{-\infty}^\infty C_1(s)e^{\frac{ws^2}{2}+vs}~ds+\int_{-\infty}^\infty C_2(s)e^{\frac{ws^2}{2}-vs}~ds$

$u(x,t)=\int_{-\infty}^\infty C_1(s)e^{\frac{ts^2}{2}+\left(x-\frac{t^2}{2}\right)s}~ds+\int_{-\infty}^\infty C_2(s)e^{\frac{ts^2}{2}-\left(x-\frac{t^2}{2}\right)s}~ds$

$u(x,t)=\int_{-\infty}^\infty C_1(s)e^{\frac{t}{2}\left(s^2+\left(\frac{2x}{t}-t\right)s\right)}~ds+\int_{-\infty}^\infty C_2(s)e^{\frac{t}{2}\left(s^2-\left(\frac{2x}{t}-t\right)s\right)}~ds$

$u(x,t)=\int_{-\infty}^\infty C_1(s)e^{\frac{t}{2}\left(s^2+\left(\frac{2x}{t}-t\right)s+\left(\frac{x}{t}-\frac{t}{2}\right)^2-\left(\frac{x}{t}-\frac{t}{2}\right)^2\right)}~ds+\int_{-\infty}^\infty C_2(s)e^{\frac{t}{2}\left(s^2-\left(\frac{2x}{t}-t\right)s+\left(\frac{x}{t}-\frac{t}{2}\right)^2-\left(\frac{x}{t}-\frac{t}{2}\right)^2\right)}~ds$

$u(x,t)=e^{-\frac{t}{2}\left(\frac{x}{t}-\frac{t}{2}\right)^2}\int_{-\infty}^\infty C_1(s)e^{\frac{t}{2}\left(s+\frac{x}{t}-\frac{t}{2}\right)^2}~ds+e^{-\frac{t}{2}\left(\frac{x}{t}-\frac{t}{2}\right)^2}\int_{-\infty}^\infty C_2(s)e^{\frac{t}{2}\left(s-\frac{x}{t}+\frac{t}{2}\right)^2}~ds$

share|cite|improve this answer
Yes, you've removed the drift term from the PDE, but now the location of the boundary, $x=0$, is $v=-t^2/2$, so unfortunately I don't see how this helps. – djws Dec 17 '12 at 20:39

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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