I'm considering the diffraction problem described in section 3.16 of "Linear and quasilinear elliptic equations" of Ladyzhenskaya and Uraltseva (1968). Let $\Omega$ be an open bounded subset in $\mathbb{R}^n$, and suppose the surface $\Gamma$ partition $\Omega$ into sub-regions $\Omega_1$ and $\Omega_2$ where $\Omega = \Omega_1 \cup \Omega_2$. For simplicity, I assume $\Gamma$ to be the surface $\{x_n=0\}$. Now suppose $u(x)$ satisfies the following condition in $\Omega_1$ and $\Omega_2$ \begin{equation} Lu \equiv \partial_{x_i} (a^{ij}(x) \partial_{x_j}u) + b^i(x) \partial_{x_i}u + c(x) u = f(x) \quad (16.1) \end{equation} and the conditions \begin{equation} u|_{\partial\Omega} = 0, \quad [u]|_\Gamma = 0, \quad \left.\left[ \frac{\partial u}{\partial N} \right]\right|_\Gamma = 0 \quad (16.3) \end{equation} on $\partial\Omega$ and $\Gamma$.

Here the symbol $[v]|_\Gamma$ denotes the jump in the function $v$ as it crosses $\Gamma$. Also, $\displaystyle \left.\frac{\partial u}{\partial N} \right|_\Gamma = a^{ij} (\partial_{x_j}u) \cos(n, x_i)$, where $n$ is the normal to $\Gamma$.

$u \in H^1_0(\Omega)$ is a weak solution of the problem (16.1) and (16.3) if $u$ satisfies \begin{equation} \int_\Omega \left[ a^{ij}(\partial_{x_j}u) (\partial_{x_i}v) -b^i (\partial_{x_i}u) v - cuv \right] \,dx = - \int_\Omega fv \,dx \quad (16.5) \end{equation} for all $v \in H^1_0(\Omega)$. Assume boundedness of the coefficients $a^{ij}, \partial_{x_k}a^{ij}, b^i, c$, the uniformly elliptic condition, and $f \in L^{q/2}(\Omega)$ for $q > n$.

Suppose we already showed $u \in H^2(\Omega_i)$. In (16.5), take $v(x) = \partial_{x_s} \eta$ for $s \le n-1$, where $\eta$ is a sufficiently smooth function with compact support in $K_\rho$, where $K_\rho$ is a sphere inside $\Omega$ with center on $\Gamma$. Let $\Gamma_1 := K_\rho \cap \Gamma$.

Substituting into (16.5) and integrating by parts we get \begin{equation} \int_\Omega \left[ a^{ij} (\partial_{x_s}\partial_{x_j}u) (\partial_{x_i}\eta) + \partial_{x_s} a^{ij} (\partial_{x_j}u) (\partial_{x_i}\eta) + b^i (\partial_{x_i}u) (\partial_{x_s}\eta) + cu(\partial_{x_s}\eta) \right] \,dx = \int_\Omega f (\partial_{x_s} \eta) \,dx. \quad (16.13) \end{equation} The author stated "On the basis of the conditions (16.13), we have $\left.[\partial_{x_s}u] \right|_{\Gamma_1} = 0$ for $s \le n-1$.

I don't understand why it is the case. I'm new to this type and really want to understand it.

Thank you.

EDIT: Following John Jiang's comment, I did the computation out. Substituting $v = \partial_{x_s} \eta$ into (16.5) and integrating by parts the first term, we get (justifying derivatives of $u$ makes sense) \begin{equation} \int_{K_\rho} \left[ -\partial_{x_s}(a^{ij}\partial_{x_j}u)(\partial_{x_i}\eta) - b^i(\partial_{x_i}u)(\partial_{x_s}\eta) - cu(\partial_{x_s}\eta) \right] \,dx + \int_{\Gamma_1} \big[a^{ij}(\partial_{x_j}u)(\partial_{x_i}\eta)\big]\Big|_\Gamma \,dS = -\int_{K_\rho} f(\partial_{x_s}\eta) dx. \end{equation} The integral over $\Gamma_1$ vanishes by the hypothesis that $\displaystyle \left.\left[ \frac{\partial u}{\partial N} \right]\right|_\Gamma = 0$. However, I haven't seen why the tangential derivatives of $u$ across $\Gamma_1$, $[\partial_{x_s}u]|_{\Gamma_1}$ must be 0.


closed as off-topic by Michael Renardy, Stefan Kohl, Wolfgang, Myshkin, Willie Wong Mar 21 '16 at 15:58

This question appears to be off-topic. The users who voted to close gave this specific reason:

  • "MathOverflow is for mathematicians to ask each other questions about their research. See Math.StackExchange to ask general questions in mathematics." – Stefan Kohl, Wolfgang, Myshkin, Willie Wong
If this question can be reworded to fit the rules in the help center, please edit the question.


You can exchange the partial derivatives on $\nu = \partial_{x_s} \eta$. Taking one partial of $u$ is allowed in the whole $\Omega$ since $u \in H^1_0(\Omega) \subset W^1_0(\Omega)$. So $\partial_{x_s} u$ makes sense there. Mixed second partials of $u$ are justified by equation 16.10, which you didn't mention; it's established by assuming $u$ to be a solution of the original equation 16.1 satisfying condition 16.3. I believe 16.10 alone tells you there is no gap of $u$ across $\Gamma$. This is of course in the almost sure sense since $u$ is an equivalence class of functions. Any gap of positive measure in $\Gamma$ would show up as a boundary term upon integration by parts. For the simplest situation refer to this. I encourage you to write down the details carefully.

  • $\begingroup$ I added the details into my question, but I haven't seen why the tangential derivatives of $u$ across $\Gamma$ must be 0. Can you please elaborate more? Also, there is no gap of $u$ across $\Gamma$ by the hypothesis, isn't it? $\endgroup$ – dh16 Mar 20 '16 at 16:48
  • $\begingroup$ If you exchange partial derivative with respect to s and j, if there is a gap in $\partial_s u$ across $\Gamma$, how do you make sense of its derivative with respect to $n$? $\endgroup$ – John Jiang Mar 20 '16 at 22:02
  • $\begingroup$ By gap of u I meant gap of $\partial_s u$. $\endgroup$ – John Jiang Mar 20 '16 at 22:03

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