4
$\begingroup$

Let $A\in C^{\infty}(\mathbb{R}^2)$ be Lebesgue integrable, and $c_1,c_2\in C^{\infty}(\mathbb{R})$ also be Lebesgue integrable. Consider the hyperbolic PDE $$ \begin{cases} \partial_{x,y}u & = A\cdot u\\ u(0,\cdot) & = c_1\\ u(\cdot,0) &= c_2 \end{cases} $$

Is there a known closed-form expression for the solution to this type of PDE?

If not, do we at-least, know the regularity of a solution to this PDE?

Disclaimer: I don't do PDEs so this question may be very simple, but it came up during research in probability theory.

Update: Even a closed-form solution in the special case where $A(x,y)=x^p\,y^q$ for $p,q\ge 0$ and $c_1,c_2$ are constants would be perfect.

$\endgroup$

1 Answer 1

6
+50
$\begingroup$

This is known as the Goursat problem, because the boundary condition are given over two intersecting characteristic lines. Notice the necessary condition (which turns out to be sufficient): $$c_1(0)=c_2(0).$$

There is no closed form in general, but I can explain here the proof of existence and uniqueness, which gives also the regularity of the solution. Amazingly, the proof mimics that for an ODE (usually, proffs for PDEs are much different than for ODEs). You must rewrite the Goursat problem as a fixed point question, in terms of the unknown $(u,v:=\partial_y u)$ for instance. we have $$u(x,y)=c_2(x)+\int_0^yv(x,\eta)d\eta$$ and $$v(x,y)=c_1'(y)+\int_0^x(Au)(\xi,y)d\xi.$$ Therefore $(u,v)$ is the fixed point of the map $T:(u,v)\mapsto (U,V)$ defined by \begin{eqnarray*} U(x,y) & = & c_2(x)+\int_0^yv(x,\eta)d\eta, \\ V(x,y) & = & c_1'(y)+\int_0^x(Au)(\xi,y)d\xi. \end{eqnarray*} Exactly as in the case of ODEs, $T$ is contracting over $C^k\times C^{k-1}([-a,a]\times[-a, a])$ for some finite $a$, provided that $c_{1,2}\in C^k$ and $A\in C^k$. Hence it has a unique fixed point. And $a$ is known in terms of $A$ only. Given a bounded rectangle $R$, you can cover $R$ with overlaping squares of side $2a$, and you get a unique solution on $R$.

Remark. On the qualitative level, the Goursat problem has an amazing property: suppose for instance that $c_1\equiv0$ over ${\mathbb R}$, and that $c_2\equiv0$ over $(-\infty,0]$. Then $$(x<0)\Longrightarrow(u(x,y)=0).$$ In my paper La compacité par compensation pour les systèmes hyperboliques non linéaires de deux équations à une dimension d'espace (JMPA 1986), I used these half-plane-supported solutions, which I called of east-type. There are of course solutions of north-, west- and south-types.

Equivalently, the value of $u$ in the quadrant $x,y>0$ depends only upon the restrictions of $c_1$ on $(0,y)$ and of $c_2$ on $(0,x)$. In the limit as $y\to0$, we obtain a closed formula for the normal derivative $$\partial_yu(x,0)=\int_0^xA(\xi,0)c_2(\xi)d\xi.$$

$\endgroup$
3
  • $\begingroup$ Very interesting. Though I wonder, if $c_1$ and $c_2$ are constants (as they are in my case) and $A$ is rather simple, e.g. a monomial of the form $A(x,y)=x^p\,y^q$ for some $p,q\ge 0$ then can one obtain a closed-form? Since, perhaps, I posed the question in much too general a formulation for that to be possible (more general than I need). $\endgroup$
    – ABIM
    Commented Jan 28 at 1:42
  • $\begingroup$ @ABIM See my edits, whether this is helpful. $\endgroup$ Commented Jan 29 at 8:16
  • $\begingroup$ This is super helpful. Thanks a lot! I think this may be enough to get what I'm looking for :) $\endgroup$
    – ABIM
    Commented Jan 30 at 2:51

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .