Partial Differential Equations
The Maximum Principle: Why Harmonic Functions Peak on the Boundary
Solve Laplace's equation Δu = 0 on a domain, and the solution can never bulge upward in the interior: its largest and smallest values are always attained on the boundary. This single rigidity fact — that a harmonic function has no interior peaks or valleys unless it is constant — is the engine behind uniqueness of solutions, continuous dependence on data, the entire regularity theory of elliptic equations, and probabilistic formulas like u(x) = 𝔼ₓ[g(B_τ)].
Precisely, the weak maximum principle says: if u ∈ C²(Ω) ∩ C(Ω̄) is subharmonic (Δu ≥ 0) on a bounded open set Ω ⊂ ℝⁿ, then max_{Ω̄} u = max_{∂Ω} u. The strong maximum principle (Hopf) sharpens this: if a connected Ω contains an interior point where u attains its maximum, then u is constant throughout Ω.
- FieldElliptic partial differential equations / potential theory
- Weak formmax over Ω̄ of u equals max over ∂Ω of u, for Δu ≥ 0
- Strong form (Hopf)Interior max ⇒ u constant, on connected Ω
- Key hypothesesΩ bounded (weak), Ω connected (strong); u ∈ C²(Ω) ∩ C(Ω̄)
- Proof techniquesPerturbation with ε|x|², mean value property, Hopf boundary lemma
- Named for / yearsGauss (1840s), E. Hopf (1927 weak, 1952 boundary lemma)
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What the principle claims, precisely
Let Ω ⊂ ℝⁿ be open and bounded, and let u ∈ C²(Ω) ∩ C(Ω̄) satisfy the subharmonicity inequality Δu ≥ 0 in Ω, where Δ = ∑ᵢ ∂²/∂xᵢ² is the Laplacian. The weak maximum principle asserts
max_{x ∈ Ω̄} u(x) = max_{x ∈ ∂Ω} u(x).
In words: the interior can never beat the boundary. Applying the same statement to −u (which is superharmonic, Δ(−u) ≤ 0) gives the companion minimum principle for superharmonic functions. A harmonic function (Δu = 0) is both, so both its max and its min live on ∂Ω.
The strong maximum principle, due to Eberhard Hopf, adds a hypothesis of connectedness and a sharp dichotomy: if Ω is connected and u attains its maximum M at some interior point x₀ ∈ Ω, then u ≡ M everywhere in Ω. No slow-motion interior bump is possible — touch the max inside once and you are pinned to a constant.
The picture: no interior bumps allowed
The intuition is that a harmonic function is its own local average. The mean value property makes this exact: if Δu = 0 then for every ball B_r(x₀) ⊂ Ω,
u(x₀) = (1/|∂B_r|) ∫_{∂B_r(x₀)} u dS = (1/|B_r|) ∫_{B_r(x₀)} u dx.
A value that equals the average of its neighbors cannot strictly exceed all of them. If u(x₀) were a strict interior maximum, every nearby value would be ≤ u(x₀), yet the average would have to equal u(x₀) — forcing every neighbor to equal u(x₀) too. Physically, u models a steady-state temperature or an electrostatic potential with no interior sources: heat with no heater cannot spontaneously form a hot spot hotter than everything around it. Δu ≥ 0 (subharmonic) is the one-sided version — u lies below its spherical averages, so it can bend upward like a taut membrane but never crest in the interior.
Key idea of the proof: the ε|x|² trick and Hopf's lemma
Weak principle. The clean trick handles the equality case Δu = 0 that ordinary calculus can't. At an interior maximum the Hessian is negative semidefinite, so Δu = trace(D²u) ≤ 0 — the wrong direction to get a contradiction from Δu ≥ 0. Fix this by perturbing: set u_ε = u + ε|x|², so Δu_ε = Δu + 2nε > 0 strictly. A function with strictly positive Laplacian cannot have an interior max (its Hessian trace would be ≤ 0 there), so max u_ε lives on ∂Ω. Let ε → 0⁺; boundedness of Ω keeps ε|x|² uniformly small, and the conclusion passes to u.
Strong principle. Suppose the interior set {u = M} is nonempty and proper. Pick a point in {u < M} nearer to the max set than to ∂Ω, grow a ball until it just touches {u = M} at a boundary point p. The Hopf boundary lemma says the outward normal derivative there is strictly positive, ∂u/∂ν(p) > 0 — contradicting that p is an interior maximum where ∇u = 0. Connectedness lets this argument propagate {u = M} across all of Ω.
Worked example: uniqueness for the Dirichlet problem
The flagship payoff is uniqueness. Suppose u₁, u₂ ∈ C²(Ω) ∩ C(Ω̄) both solve the Dirichlet problem Δu = f in Ω, u = g on ∂Ω, with Ω bounded. Let w = u₁ − u₂. Then Δw = 0 in Ω and w = 0 on ∂Ω. By the maximum principle applied to the harmonic w,
max_{Ω̄} w = max_{∂Ω} w = 0, and min_{Ω̄} w = min_{∂Ω} w = 0,
so 0 ≤ w ≤ 0, i.e. w ≡ 0 and u₁ = u₂. A concrete instance: on the unit disk Ω = {x² + y² < 1} with boundary data g(θ) = cos θ, the unique harmonic solution is u = x = r cos θ. Its maximum value 1 and minimum −1 are both achieved only on the boundary circle, exactly at (1,0) and (−1,0); the interior value at the center is u(0) = 0, precisely the average of cos θ over the circle — the mean value property in action.
Why the hypotheses matter — and clean counterexamples
Boundedness is essential. On the unbounded half-space Ω = {x ∈ ℝⁿ : x₁ > 0}, the function u = x₁ is harmonic and vanishes on the boundary {x₁ = 0}, yet u → +∞ inside: max_{∂Ω} u = 0 while sup_Ω u = ∞. The ε|x|² perturbation blows up when Ω is unbounded, and the conclusion genuinely fails.
Connectedness is essential for the strong form. On Ω = B₁ ∪ B₂ (two disjoint balls), the locally constant function u = 0 on B₁, u = 1 on B₂ is harmonic and hits its max at every interior point of B₂ without being globally constant.
The zeroth-order term matters. For Lu = Δu + c(x)u the principle can fail when c > 0: u = sin x₁ solves u'' + u = 0 on (0, π) with u = 0 at both endpoints, yet u > 0 inside — an interior positive max despite zero boundary data. This is why the standard elliptic maximum principle requires c ≤ 0, and connects directly to the spectrum of −Δ: eigenfunctions are exactly where positivity breaks.
Why it matters: what the principle unlocks
The maximum principle is the backbone of elliptic and parabolic PDE theory. Uniqueness and stability: ‖u₁ − u₂‖_{L^∞(Ω)} ≤ ‖g₁ − g₂‖_{L^∞(∂Ω)}, so solutions depend continuously on boundary data — well-posedness in Hadamard's sense. Comparison principles: sub/supersolutions sandwich the true solution, powering Perron's method for existence via upper envelopes of subharmonic functions. Harnack's inequality and interior gradient estimates, and thence the De Giorgi–Nash–Moser regularity theory, all descend from maximum-principle machinery. Probability: the principle is the analytic face of the martingale property of B_t under a harmonic u, giving the representation u(x) = 𝔼ₓ[g(B_τ)] where τ is the exit time of Brownian motion from Ω. It also drives geometric analysis (Bochner formulas, Cheng–Yau, the Omori–Yau maximum principle on manifolds), viscosity-solution theory for fully nonlinear equations, and unique continuation. Few one-line inequalities carry so much of a subject.
| Feature | Weak maximum principle | Strong maximum principle (Hopf) |
|---|---|---|
| Domain hypothesis | Ω bounded, open | Ω open, connected |
| Operator hypothesis | Δu ≥ 0 (subharmonic); or Lu ≥ 0, L uniformly elliptic, no zeroth-order term | Same, plus needs interior sphere / mean value structure |
| Conclusion | max_{Ω̄} u = max_{∂Ω} u | Interior max attained ⇒ u ≡ const |
| Main tool | Perturb u_ε = u + ε|x|², Δu_ε > 0 | Mean value property + Hopf boundary lemma |
| Typical use | Comparison, a priori sup bounds | Uniqueness, unique continuation, Harnack |
| Fails if... | Ω unbounded (u = x₁ on a half-space) | Ω disconnected, or c(x) > 0 term added |
Frequently asked questions
Why can't ordinary calculus (second derivative test) prove the weak maximum principle directly?
At an interior maximum the Hessian D²u is negative semidefinite, so its trace Δu is ≤ 0. For strictly superharmonic functions (Δu > 0) that's an immediate contradiction. But harmonic functions only satisfy Δu = 0, which is consistent with a degenerate interior max, so the naive test is inconclusive. Hopf's fix is to perturb to u_ε = u + ε|x|² with Δu_ε = 2nε > 0, apply the strict case, then let ε → 0.
What is the difference between the weak and strong maximum principles?
The weak principle says the maximum over Ω̄ is attained somewhere on the boundary ∂Ω — but it might also be attained inside. The strong (Hopf) principle says that if the max is attained at any interior point of a connected domain, then u is constant. So the strong form rules out nonconstant functions from ever touching their max inside, which the weak form permits.
Why is boundedness of Ω needed in the weak maximum principle?
Boundedness makes the perturbation ε|x|² uniformly small so the conclusion survives the limit ε → 0. Without it the principle is false: on the half-space {x₁ > 0}, the harmonic function u = x₁ is 0 on the boundary but unbounded inside. On unbounded domains you must add a growth condition at infinity (a Phragmén–Lindelöf hypothesis) to recover a maximum principle.
Does the maximum principle hold for general second-order elliptic operators, not just Δ?
Yes, for L u = ∑ aⁱʲ(x) ∂ᵢⱼu + ∑ bⁱ(x) ∂ᵢu with (aⁱʲ) uniformly elliptic and bounded coefficients, provided there is no zeroth-order term or the zeroth-order coefficient satisfies c(x) ≤ 0. The Hopf construction and boundary lemma go through. If c(x) > 0 the principle can fail, as u = sin x on (0,π) shows for u'' + u = 0.
How does the maximum principle relate to Brownian motion?
They are two faces of the same fact. If u is harmonic in Ω and B_t is Brownian motion, then u(B_t) is a local martingale, so its expectation is preserved. Stopping at the exit time τ from Ω gives u(x) = 𝔼ₓ[u(B_τ)] = 𝔼ₓ[g(B_τ)] — a weighted average of boundary values. Since an average never exceeds the max of what's averaged, u(x) ≤ max_{∂Ω} g, which is precisely the maximum principle.
Is there a maximum principle for parabolic equations like the heat equation?
Yes. For u_t = Δu on Ω × (0,T], the maximum over the closed cylinder is attained on the parabolic boundary — the initial slice t = 0 together with the lateral boundary ∂Ω × [0,T] — but never at an interior or final-time point unless u is constant up to that time. This parabolic maximum principle gives uniqueness and comparison for heat flow, and underlies Nash–Moser estimates and Ricci-flow analysis.