Let $\Omega \subset \mathbb R^n$ be a subdomain with Lipschitz boundary, i.e. locally any part of the boundary looks like the graph of a Lipschitz continuous function, after some affine coordinate transformation.

Suppose we are given a "partition" $\Gamma_D$ and $\Gamma_N$ of the boundary, s.t. these sets are submanifolds of $\mathbb R^n$ with Lipschitz boundary by themselves, and their intersection has measure zero.

Let us be given $g \in L^2(\Gamma_D)$ and $h \in L^2(\Gamma_N)$ and some function $f \in L^2(\Omega)$. We want to solve Poisson's equation with mixed boundary conditions

$\operatorname{div}\operatorname{grad} u = f$ over $\Omega$

$u_{|\Gamma_D} = g$ over $\Gamma_D$

$\operatorname{grad} u_{|\Gamma_N} \cdot n = h$ over $\Gamma_N$

It is standard to prove well-posedness of these problems if either $\Gamma_D$ or $\Gamma_N$ is the empty set. I have not found a rigorous proof of well-posedness for general mixed boundary conditions in the standard books like, say, Gilbarg-Trudinger. On the other hand, certain papers suggest the boundary parts are required to meet at an angle that is not 180° in the case of Lipschitz boundaries, so the boundary is necessarily non-smooth.

These influences appear confusing to me. I do not know how to learn more about this. Could please give a reference where to learn more about the Poisson problem with mixed boundary conditions?

EDIT: In order to motivate why this is interesting and why it confuses me, I would like to point to the paper Ott, Brown: The mixed problem for the Laplacian in Lipschitz domains and R.M. Brown. The mixed problem for Laplace’s equation in a class of Lipschitz domains. On the other hand, in numerical analysis lectures that I attend, this question is usually swept under the rug and one deals freely with mixed boundary conditions. So either I don't know the well-posedness results for simplicial domains, or the numerical examples all belong to the well-posed case.

  • $\begingroup$ You'll need some additional global conditions regarding your domain and its boundary. For example, is the boundary connected? What about $\Gamma_N \cap \Gamma_D$? As to your angle condition - I would then start by looking at special solutions in corner domains, e.g. $u_\alpha(x,y) = \Re (x+iy)^{\alpha}$ in domains like $\Omega_\gamma = \{(x,y) | x \ge 0, \, 0 \le y \le \gamma x\}$ to understand in what sense one can expect maximal regularity to hold. For example, for which $\alpha, \gamma$ is $u_\alpha \in H^{3/2}_{loc}(\Omega_\gamma)$? $\endgroup$ Nov 26 '11 at 18:01

Conditions for the unique solvability in H1(Ω) are given in Theorem 4 of the paper on discrete maximum principles by Karatson and Korotov in Numer. Math. 99 (2005), 669-698. You need to specialize this result (stated for nonlinear second-order elliptic equations with mixed boundary conditions) to your particular situation.


If we consider the simplified problem $$-\Delta u = f\quad\mbox{in}\quad\Omega,$$ $$u|{\Gamma_D}=0,$$ $$\partial_nu|_{\Gamma_N}=h,$$ Then the weak formulation is to find $u\in H^1_E(\Omega)=\{v\in H^1(\Omega):v|_{\Gamma_D}=0\}$ such that $$(\nabla u,\nabla w)=(f,w)+(h,w)_{\Gamma_N}$$ for all $w\in H^1_E(\Omega)$. By the integrability assumptions on $f,w$, and the trace theorem on $H^1$, the right hand side in a continuous linear functional on $H^1_E(\Omega)$, and the left hand side is a coercive bounded bilinear form. So the Lax Milgram theorem gives us existence and uniqueness in $H^1_E(\Omega)$.

I wanted to approach the more general question that you posed by subtracting some extension $\tilde g$ of the Dirichlet boundary data from $u$, and solve your problem for $v=u-\tilde g$, which would reduce to the problem I mentioned, except for I was unsure how this extension would interact with the Neumann condition, perhaps there are such extensions $\tilde g\in H^1_E(\Omega)$ such that $\partial_n \tilde g|_{\Gamma_N}\in L^2$, in which case the above applies.


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.