2
$\begingroup$

Let $D \subset \mathbf{R}^n$ be the unit ball, and $u \in C^2(D)$ be a solution of the linear elliptic PDE \begin{equation} a^{ij} D_{ij} u + b^i D_i u + cu = 0 \quad \text{in $D$}, \end{equation} where the coefficients are regular, say of class $C^d$ for some integer $d \geq 1$.

Assume that $u(0) = 0$, $Du(0) = 0$. Strong unique continuation means that $u$ has finite order of vanishing at the origin unless it vanishes identically: there exists $N > 0$ so that $r^{-n} \int_{D_r} u^2 \notin O(r^N)$ as $r \to 0$ if $u \not \equiv 0$.

Question. Under the given hypotheses, do the partial derivatives $D_k u$ of $u$ also satisfy the strong unique continuation property?

$\endgroup$
8
  • $\begingroup$ Can you clarify your question? Does what hold under what hypotheses? It is not clear from your question which hypotheses you want to share between the known result and that which you want to ask about. $\endgroup$ Commented Jun 4, 2021 at 16:22
  • $\begingroup$ In particular, if your question is: if $u$ solves a second order linear elliptic PDE as above, and you know that $Du$ vanishes to infinite order at the origin, is $u$ constant? Then the answer is "not in general". $\endgroup$ Commented Jun 4, 2021 at 16:26
  • $\begingroup$ @WillieWong I've reworded the question slightly - is it clearer now? I think your second comment is precisely what I meant to ask. Could you share the example you have in mind? $\endgroup$
    – Leo Moos
    Commented Jun 4, 2021 at 18:20
  • $\begingroup$ If you don't assume $u(0) = 0$, then you have easy counterexamples: take any positive smooth function that is constant near the origin. Define $c = \frac{1}{u}(a^{ij}\partial^2_{ij} u + b^i \partial_i u)$. This example can be ruled out with with additional hypotheses on $c$. $\endgroup$ Commented Jun 4, 2021 at 20:09
  • $\begingroup$ @WillieWong That's good to keep in mind, thanks - I've seen similar examples in a couple of papers. Regarding the second part, if e.g. $c \equiv 0$ would there be an affirmative answer to the question? $\endgroup$
    – Leo Moos
    Commented Jun 5, 2021 at 6:54

1 Answer 1

2
$\begingroup$

In general there's no way what you want (as clarified in this comment) can be true.

Let $n = 3$, and fix $k = 3$. Consider the function $u = (x_1)^2 - (x_2)^2$ which is harmonic.

Let $\Phi: D \to D$ be a diffeomorphism that is the identity at a neighborhood of the origin. Then the Laplacian pulls back to a Laplace-Beltrami operator $L$ on $D$, and the pullback $\tilde{u} = \Phi^*u$ is a solution to $L\tilde{u} = 0$. Note that $L$ has no 0-th order terms. Note that $\tilde{u}$ satisfies $\tilde{u}(0) = \nabla \tilde{u}(0) = 0$, and that $\partial_3 \tilde{u}$ vanishes to infinite order at origin (since this holds of $u$ and $\Phi$ is identity near the origin).

However, for general $\Phi$, the function $\partial_3\tilde{u}$ does not vanish identically in $D$.

$\endgroup$
2
  • $\begingroup$ Makes sense - thanks for the answer and apologies for the ambiguous formulation! $\endgroup$
    – Leo Moos
    Commented Jun 5, 2021 at 18:52
  • $\begingroup$ Incidentally, when $n \leq 2$ and $c \equiv 0$ what you want holds. $\endgroup$ Commented Jun 11, 2021 at 17:44

You must log in to answer this question.

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