4
$\begingroup$

I am trying to understand the exact implications between local solvability and a general version of the Cauchy-Kovalevskaya (CK) theorem, in the context of PDEs.

Let $\Delta(x,u^{(n)})=0$ be a system of PDEs of order $n$. Here $x$ is the vector of independent variables, and $u^{(n)}$ is the vector of dependent variables and all their derivatives up to order $n$. According to Olver ("Applications of Lie groups to Differential Equations", 2/e, Ch. 2, Def. 2.70), $\Delta$ is locally solvable if the variety it induces on $\mathbb{R}^{|x|+|u^{(n)}|}$

$V(\Delta):=\{(x_0,u_0^{(n)}):\Delta(x_0,u_0^{(n)})=0\}$

coincides with its solution variety

$S(\Delta):={\small \{(x_0,u_0^{(n)})\in V(\Delta) : \text{ $\exists$ an analytic solution $U$ of $\Delta$ in a neighborhood of $x_0$ s.t. } U^{(n)}(x_0)=u_0^{(n)} \}}$

Olver shows that if $\Delta$ is in Kovalevskaya form then it is locally solvable (Corollary 2.74, p. 163; this is indeed an easy consequence of CK theorem). He then states that the same result still holds when $\Delta$ is in general Kovalevskaya form:

enter image description here

I am struggling to convince myself of the validity of this statement. For instance, consider the 2nd order system $\Delta$ in the independent variables $t,x$ and the dependent variables $u,v$: \begin{align*} u_t & = v\\ v_{tt} &= u_x\,. \end{align*} From what I gather, this system is not locally solvable. Indeed, there are differential consequencences that are not captured algebraically by the two equations above, such as $u_{tx}=v_x$. So there are points $(x_0,u_0^{(2)})\in V(\Delta)$ s.t. (with obvious notation) $u_{0,tx}\neq v_{0,x}$, hence not in $S(\Delta)$.

Yet, $\Delta$ is in general Kovalevskaya form, is it?

Edit: for reference, I paste Olver's original definition. Note that $pr^{(n)}f$ denotes the $n$-th prolongation of $f$.

enter image description here

$\endgroup$
  • $\begingroup$ Couldn't you find all solutions by solving the first order system \begin{align*} u_t &= v\\ v_t &= w\\ w_t &= u_x \end{align*} with initial data $(u,v,w)$ along $t = 0$? $\endgroup$ – Deane Yang Jul 28 at 14:37
  • $\begingroup$ Your system is "equivalent" to the old one in terms of solutions. However the point here is not finding the solutions, but solving the apparent contradiction I have pointed out. Cheers $\endgroup$ – Michele Jul 28 at 15:35
  • $\begingroup$ Why is this system not locally solvable? Doesn’t $u_{0,tx} = v_{0,x}$ follow from the assumption that the equations hold at $(x_0,t_0)$ up to second order. $\endgroup$ – Deane Yang Jul 28 at 22:33
  • $\begingroup$ This is in the definitiion of $S(\Delta)$, but not in that of $V(\Delta)$. That is, there are 2nd order differential consequences (e.g. $u_{tx}=v_x$) that are not algebraic consequences. $\endgroup$ – Michele Jul 29 at 7:21
  • 1
    $\begingroup$ Here, it appears that local solvability means that any second order jet that solves the system can be extended to a local smooth solution (which is not the standard definition of local solvability). To be a second order solution, the three equations and the first partial derivatives with respect to both $x$ and $t$ of the first two equations must hold. In particular, if a 2-jet is a solution at $(x_0,t_0)$, $u_{tx} = v_x$ then holds at $(x_0,t_0)$. So in fact this equation is a consequence of the jet being in $S(\Delta)$. $\endgroup$ – Deane Yang Jul 29 at 16:28
1
$\begingroup$

The system you wrote down First, let's assume everything is smooth. \begin{align*} u_t &= v\\ v_{tt} &= u_x\\ \end{align*} is equivalent to the first order system \begin{align*} u_t &= v\\ v_t &= w\\ w_t &= u_x \end{align*} in the sense that $(u,v)$ is a solution to the first system if and only if $(u,v,w)$ is a solution to the second system, where we set $w = v_t$.

By Cauchy-Kovalevski, given any analytic functions $u_0(x), v_0(x), w_0(x)$, there exists a unique analytic solution $(u,v,w)$ to the second system such that $u(x,0) = u_0(x), v(x,0) = v_0(x), w(x,0) = w_0(x)$. This is equivalent to saying that given any analytic functions $u_0, v_0, w_0$, there exists a unique analytic solution $(u,v)$ to the first system such that $u(x,0) = u_0(x)$, $v(x,0) = v_0(x)$, and $v_t(x,0) = w(x)$, which is what Olver asserts. The fact that $u_{tx} = v_x$ is a consequence of the equations and need not be specified in the initial data.

As for Olver's definition of local solvability, an element of $V(\Delta)$ for the original system consists of a $2$-jet $(x_0,t_0,u(x_0,t_0), u_x(x_0,t_0), u_t(x_0,t_0), v(x_0,t_0), v_x(x_0,t_0), v_t(x_0,t_0))$ that satisfies the system up to second order. On other words, at $(x_0,t_0)$, \begin{align*} u_t &= v\\ u_{tx} &= v_x\\ u_{tt} &= v_t\\ v_{tt} &= u_x \end{align*} In particular, if a $2$-jet lies in $V(\Delta)$, then $u_{tx} = v_x$ does hold for that $2$-jet at $(x_0,t_0)$. Given such a $2$-jet, you can extend $u$, $v$, and $v_t$ arbitrarily to initial data along $t = 0$ and solve the system as described above. In particular, the initial data is assumed to satisfy $u_{tx} = v_x$ at $(x_0,t_0)$.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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