TheoremComplete

Maximum Principle for Elliptic PDEs

Theorem6.2Strong Maximum Principle for Harmonic Functions

Let uu be harmonic (2u=0\nabla^2 u = 0) in a connected open domain ΩRn\Omega \subset \mathbb{R}^n, continuous on Ωˉ\bar{\Omega}. Then: (1) uu attains its maximum and minimum on Ω\partial\Omega (weak maximum principle); (2) if uu attains its maximum at an interior point, then uu is constant (strong maximum principle). More generally, for Lu=(au)+cu0\mathcal{L}u = -\nabla\cdot(a\nabla u) + cu \geq 0 with c0c \geq 0, uu satisfies maxΩˉumaxΩu+\max_{\bar\Omega} u \leq \max_{\partial\Omega} u^+.


Proof

Proof

Weak maximum principle. Suppose maxΩˉu=M\max_{\bar\Omega}u = M is attained at an interior point x0Ωx_0 \in \Omega. At a maximum: 2u(x0)0\nabla^2 u(x_0) \leq 0 (sum of second derivatives is non-positive). But 2u=0\nabla^2 u = 0, so 2u(x0)=0\nabla^2 u(x_0) = 0, which is consistent. A direct argument is needed.

Proof via mean value property. For harmonic uu: u(x0)=1BrBr(x0)udxu(x_0) = \frac{1}{|B_r|}\int_{B_r(x_0)} u\,dx (mean value property over balls). If u(x0)=M=maxuu(x_0) = M = \max u, then since uMu \leq M everywhere: M=1BrBrudxMM = \frac{1}{|B_r|}\int_{B_r} u\,dx \leq M. Equality implies u=Mu = M a.e. on BrB_r. By continuity, u=Mu = M on Br(x0)B_r(x_0).

Strong maximum principle. Let S={xΩ:u(x)=M}S = \{x \in \Omega : u(x) = M\}. We showed SS is open (every point of SS has a ball neighborhood in SS). SS is also closed in Ω\Omega (as the preimage of {M}\{M\} under the continuous function uu). Since Ω\Omega is connected: S=S = \emptyset or S=ΩS = \Omega. If uu attains its maximum at any interior point, SS \neq \emptyset, so S=ΩS = \Omega and uu is constant.

Mean value property (proof sketch). For harmonic uu and ball Br(x0)ΩB_r(x_0) \subset \Omega: define v(ρ)=1BρBρudSv(\rho) = \frac{1}{|\partial B_\rho|}\oint_{\partial B_\rho} u\,dS (surface average). Differentiate: v(ρ)=1BρBρurdS=1BρBρ2udx=0v'(\rho) = \frac{1}{|\partial B_\rho|}\oint_{\partial B_\rho}\frac{\partial u}{\partial r}\,dS = \frac{1}{|\partial B_\rho|}\int_{B_\rho}\nabla^2 u\,dx = 0. So v(ρ)=v(0)=u(x0)v(\rho) = v(0) = u(x_0), proving the mean value property. \square


ExamplePhysical Interpretation

For steady-state heat conduction (2T=0\nabla^2 T = 0, no sources): the temperature cannot have a local maximum or minimum inside the domain. Hot spots and cold spots must occur on the boundary. This is physically intuitive: heat flows from hot to cold, so an interior maximum would cause outward heat flow, contradicting the steady state.

RemarkUniqueness from Maximum Principle

The maximum principle implies uniqueness for Dirichlet problems: if 2u1=2u2=f\nabla^2 u_1 = \nabla^2 u_2 = f with u1=u2u_1 = u_2 on Ω\partial\Omega, then w=u1u2w = u_1 - u_2 is harmonic with w=0w = 0 on Ω\partial\Omega. By the maximum principle: maxw0\max w \leq 0 and minw0\min w \geq 0, so w=0w = 0.