3
$\begingroup$

A stochastic control problem has led me to the following PDE:

State space: $t \in [0,T]$ and $x \in [-1,1]$.

$$4 \frac{\partial f}{\partial t} \frac{\partial^2 f}{\partial x^2} = 1 \quad \forall (t,x) \in [0,T) \times (-1,1)$$ and the boundary conditions are: $$ \begin{split} f(T,x) &= 0\quad \forall x\\ &\text{and} \\ f(t,\pm 1) &= T-t \end{split}$$ Is there an argument that can let me say that a solution exists to this PDE? I am not after a closed form expression obviously.

From the properties of the problem I can argue that $f(t,x) = f(t,-x)$ and $\arg \max_x f(t,x) = 0$. Don't know if it is of any use though.

Thanks.

$\endgroup$
  • 2
    $\begingroup$ By inspection, $f(t,x) = (T-t)(9-x^2)/8$ is a solution. $\endgroup$ – user126920 Jan 29 at 20:29
  • 1
    $\begingroup$ Not sure I see it. The right hand side of the PDE is 1, not $f$. If we evaluate the left hand side with your expression we have $-(T-t)(9-x^2)=-8f(t,x)$. Am I missing something? $\endgroup$ – avk255 Jan 30 at 6:45
  • 2
    $\begingroup$ @StanleySnelson: the initial condition $f(T,x)=0$ for all $x$ implies that, if $f$ is $C^2$ near $t=T$, then $f_{xx}=0$. But the equation gives $4f_tff_{xx}=1$, a contradiction. So there is no $C^2$ solution, with the given boundary conditions. $\endgroup$ – Ben McKay Jan 30 at 8:17
  • 1
    $\begingroup$ Not sure I agree entirely. I follow that $f_{xx} \to 0$ as $t \to T$. However, we could have $f_t \to \infty$ at the same time to have $f$ being $C^{1,2}$ on $(0,T) \times (-1,1)$, isn't it? $\endgroup$ – avk255 Jan 30 at 8:49
  • 1
    $\begingroup$ At least formally, taking the Legendre transform $u$ of $f$ in the $x$ variable, we have that $w := u(x, T-t)$ solves the heat equation. Taking initial data $w(x,0) = |x|$ and transforming back gives a nontrivial solution with $0$ data on the sides. It seems feasible that playing with the initial data for $w$ could give a solution with the desired data (which corresponds to $w_{xx} = 1$ where $w_x = \pm 1$). $\endgroup$ – Connor Mooney Jan 30 at 17:24
1
$\begingroup$

Putting $T=1$, I obtain with Maple:

sol := pdsolve(4*(diff(f(t, x), t))*(diff(f(t, x), x, x)) = 1,
 {f(1, x) = 0, f(t, -1) = 1-t, f(t, 1) = 1-t}, numeric):sol:-plot3d(t = 0 .. 1, x = -1 .. 1);

solution becomes undefined, problem may be ill posed or method may be ill suited to solution

$\endgroup$
  • $\begingroup$ Mathematica reports "NDSolve::ndnum: Encountered non-numerical value for a derivative at t == 1." $\endgroup$ – user64494 Jan 30 at 10:16
  • $\begingroup$ My sense is that, in light of what Ben pointed out above, we must have a blowup for $f_t$ near $T$. That maybe the reason why numerical solutions are showing non existence? $\endgroup$ – avk255 Jan 30 at 18:01

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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