# $C^0$ estimate for solutions of elliptic PDE with Neumann BC

I am interested in a reference for (or counterexample to) a particular $C^0$ estimate for solutions of the Laplace equation with Neumann boundary conditions. More precisely, let $(M,g)$ be a $C^\infty$ compact Riemannian manifold with non-empty smooth boundary $\partial M$. Let $\Delta_g=-div_g\nabla$ denote the Laplace operator and $\nu$ the outward pointing unit normal on $\partial M$.

Consider (strong) solutions $u$ to the following system \begin{cases} \Delta_gu=f&\text{ in }M\\ \partial_\nu u=h&\text{ on }\partial M \end{cases} where $f\in C^\infty(M)$ and $h\in C^\infty(\partial M)$ satisfy $\int_M fdvol_g=\int_{\partial M}hd\sigma_g$ (the smoothness condition is for simplicity but the later condition is necessary for the problem to be well-posed). Here $dvol_g$ denotes the Riemannian volume form and $d\sigma_g$ denotes its pullback under the inclusion $\partial M\to M$. To specify $u$ uniquely, let us also suppose that it has zero average value i.e. $\int_Mudvol_g=0$. My question is this: does there exist $C>0$, independent of $f$ and $h$, such that $\sup_M|u|\leq C(\sup_M|f|+\sup_{\partial M}|h|)$?

Most classical sources (Evans, Gilbarg+Trudinger, and Aubin) either only consider Dirichlet conditions (in which case the analogous estimate is essentially just the maximum principle) or only provide Schauder estimates for the Neumann problem -- this is not quite what I want.

Does anyone know of a resource for this kind of estimate or of a counterexample?

• It might be easier to study a first order system for a vector field: Set $w=\nabla u$. Then the system becomes $-\text{div}(w)=f$ in $M$ and $\nu\cdot w=h$ on $\partial M$. Since $u$ must have a zero somewhere (assuming connectedness), $\sup_M|u|\leq\text{diam}(M)\sup_M|w|$, so it suffices to estimate $w$ uniformly. Sep 23, 2014 at 19:42
• If f and h are bounded, they are in every $L^p$. You can then get what you want by using $L^p$ estimates and Sobolev embedding. Sep 24, 2014 at 0:07
• they are both smooth and the domain is compact, so they are certainly bounded. Sep 27, 2014 at 18:40

Yes, I think the estimate you propose is true. As a simple case take $$f = 0$$ and $$M = B_1 \subset \mathbb{R}^n$$. By adding a constant assume that $$\inf_{B_1} u = 0$$ and $$\sup_{B_1}u = K,$$ and note that these extrema are on the boundary. Note that $$u(0)$$ is closer to $$0$$ or $$K$$, say $$u(0) > K/2$$ without loss of generality. Then by the Harnack inequality we have $$u > cK$$ in $$B_{1/2}$$, so the function $$c(n)K(1-|x|^{2-n})$$ is a lower barrier in the annulus $$B_1-B_{1/2}$$. In particular, at the point where $$u$$ takes its minimum we have $$c(n)K which gives the estimate.

In a more general domain we can argue similarly; take an interior ball touching the point on the boundary where $$u = 0$$ and argue that $$u$$ is no larger than $$C\sup|h|$$ at its center, and similarly take a ball touching the point where $$u = K$$ and argue that $$u$$ is larger than $$K - C\sup|h|$$ at its center. If $$K$$ is much larger than $$\sup|h|$$ we contradict the Harnack inequality (with a constant depending on the geometry of $$M$$).

Finally, if $$f$$ is nonzero similar arguments should work since we still have the Harnack inequality (now depending on $$\sup|f|$$) and for a barrier we can take $$|x|^{-\alpha}$$ for $$\alpha$$ large depending on $$\sup|f|$$.

Suppose for simplicity that $$M$$ is a compact n-dimensional submanifold of $$\mathbb{R}^n$$ with boundary. Extend $$f$$ to all $$\mathbb{R}^n$$ so that it's $$0$$ outside $$M$$ and define the function $$u_1(x) := \displaystyle \int_{\mathbb{R}^n}\phi(y-x)f(y)dy$$ where $$\phi$$ is the fundamental solution to the Laplace equation. Then $$-\Delta u_1 = f$$ in $$M$$, and we have the following estimates:

$$\rVert u_1 \rVert_{C^0(M)} + \rVert \nabla u_1 \rVert_{C^0(M)} \leq C_1 \lVert f \rVert_{C^0(M)}$$

Now we find an integral representation of a solution to the homogeneous system using single layer potentials. Define $$S (g) := \int_{\partial M} \phi(x-y) g(y) dy$$ for $$g \in C^0(\partial M)$$. It is known that $$S$$ is a compact bounded linear operator on $$C^{0}(\partial M)$$ into $$C^0(M)$$ (check "The Laplace Equation" by Medkova, Proposition 6.7.1). It is also known that there exists an isomorphism $$T$$ on $$C^0(\partial M)$$ such that if $$v = S(\psi_1)$$ and $$\psi_2 = T(\psi_1)$$ then $$\Delta v= 0$$ on $$M$$ and $$\frac{\partial v}{\partial \nu} = \psi_2$$ on $$\partial M$$. And so define $$u_2:= S\circ T^{-1} (h-\frac{\partial u_1}{\partial \nu})$$, which will then solve

$$\Delta u_2 = 0 \quad \text{in} \ M, \qquad \frac{\partial u_2}{\partial \nu} = h-\frac{\partial u_1}{\partial \nu} \quad \text{in} \ \partial M$$ And so we have

$$\lVert u_2 \rVert_{C^0(M)} \leq \lVert S \rVert \lVert T^{-1} \rVert \left(\lVert h\rVert_{C^0(\partial M)}+\lVert \frac{\partial u_1}{\partial \nu} \rVert_{C^0(\partial M)} \right) \leq C_2(\lVert h\rVert_{C^0(\partial M)} + \lVert f\rVert_{C^0(M)})$$

Finally, let $$C_3 = -\int_M (u_1+u_2)$$; then $$u:= u_1 +u_2 + C_3$$ solves $$-\Delta u = f \quad \text{in} \ M, \qquad \frac{\partial u}{\partial \nu} = h \quad \text{in} \ \partial M, \qquad \int_M u = 0$$ and we get

$$\lVert u\rVert_{C^0( M)} \leq \lVert u_1\rVert_{C^0(M)} + \lVert u_2\rVert_{C^0( M)} + |C_3| \leq C \left( \lVert h\rVert_{C^0(\partial M)} + \lVert f\rVert_{C^0( M)} \right) + |C_3|$$ and the desired estimate follows.