The Mean-Value Property and the Maximum Principle

The averaging miracle

In the last guide you met harmonic functions — solutions of Laplace's equation Laplacian u = u_xx + u_yy = 0 — as the steady states of diffusion, the shapes a membrane or a temperature settles into once all change has stopped. That is the physics. Now comes the property that gives harmonic functions their entire character, and it is almost unreasonably clean. Take any harmonic u, pick any point, and draw a circle around it (a sphere in 3D). The value of u at the centre equals the average of u over that whole circle. Not approximately — exactly, for every radius that fits inside the region.

u(center)  =  average of u over the circle of radius r

           =  (1 / 2*pi*r) * (integral of u around the circle)

  holds for EVERY r small enough that the disk stays inside the region

The mean-value property: a harmonic function's value at the centre is the literal average of its values on any surrounding circle.

This is the mean-value property, and it is worth sitting with how strong it is. A generic function does whatever it likes; a harmonic one is bound by an infinite list of averaging constraints, one for every point and every radius at once. Intuitively, Laplacian u measures how much u at a point exceeds the average of its immediate neighbours — that is exactly what the second derivatives add up to. Setting that excess to zero says: nowhere does u stick up above, or dip below, its local surroundings. The mean-value property is just this neighbour-balancing, integrated out to circles of any size.

Why the average locks in

Here is the short argument, because it really is short and it shows where Laplacian u = 0 does the work. Define M(r) to be the average of u over the circle of radius r about a fixed centre. As r grows, the circle sweeps outward, so the rate of change of M with r is governed by how u changes in the outward (normal) direction, summed around the circle. The divergence theorem turns that boundary sum into the integral of Laplacian u over the enclosed disk. But u is harmonic, so Laplacian u = 0 everywhere inside — the integral is zero, M'(r) = 0, and the average never moves as the circle grows or shrinks.

Now shrink the radius all the way to zero. As r goes to 0 the circle collapses onto the centre, so its average must approach the centre value u(center). Since M(r) is constant in r, that limit is its value at every r: the average over any circle equals u(center). That is the whole proof. Notice it is genuinely two-way — one can show that any continuous function with the mean-value property is automatically harmonic and, as a bonus, infinitely differentiable. So Laplace's equation and the averaging rule are not merely related; they are the same statement wearing two outfits.

No interior peaks: the maximum principle

The averaging rule has an immediate, vivid consequence. Suppose u attained its maximum value at some interior point p. Draw a small circle around p. The centre value u(p) is the average of the values on that circle — but an average can never exceed the largest number being averaged, and it equals the largest only if every value on the circle equals it too. So u must already be at its maximum all the way around that circle. Slide the circle, repeat, and the maximum spreads through the entire connected region: u is constant. This is the strong maximum principle — a non-constant harmonic function cannot reach its maximum (or, by the same logic applied to -u, its minimum) at any interior point.

The practical face of this is the weak maximum principle: on any closed bounded region, a harmonic function attains its maximum and minimum on the boundary. The interior can hold no surprises — no hidden spike rising above the rim, no secret dip below the floor. Picture a soap film stretched on a bent wire loop: the film is harmonic, and nowhere does it bulge above the highest point of the wire or sag below the lowest. The boundary frame caps everything inside. The maximum principle is exactly this guarantee, and it is the workhorse of the whole theory.

Uniqueness falls out for free

Here is the payoff that makes the maximum principle indispensable. Suppose you want to solve the Dirichlet problem — find a harmonic u inside a region taking prescribed values on the boundary — and you worry there might be two different answers. Could two harmonic functions share the same boundary data yet differ inside? The maximum principle says no, and the proof is three lines. Let u and v both be harmonic with the same boundary values, and form their difference w = u - v.

w = u - v is harmonic too (Laplacian is linear, so Laplacian w = Laplacian u - Laplacian v = 0).
On the boundary, u and v agree, so w = 0 there. The boundary maximum AND minimum of w are both 0.
By the maximum principle, w's interior values lie between its boundary max and min — that is, between 0 and 0. So w = 0 throughout.
Therefore u = v everywhere. The Dirichlet problem has at most one harmonic solution.

That is uniqueness, gift-wrapped. It is half of what it means for a problem to be well-posed: existence (a solution exists), uniqueness (only one), and stability (small data changes give small solution changes). The maximum principle hands you uniqueness directly, and it also hands you stability — if two boundary datas differ by at most some small epsilon, the same w-argument bounds the interior difference of the solutions by epsilon as well. So a tiny error in the boundary measurement can only cause a tiny error inside. The elliptic Dirichlet problem is beautifully stable, in sharp contrast to running the heat equation backward in time, which is famously ill-posed precisely because no such principle protects it.

What the principle buys you everywhere

Once you have the maximum principle, an entire toolbox opens. Comparison: if two harmonic functions are ordered on the boundary, they stay ordered inside — a fence on the rim is a fence everywhere. A priori bounds: before you even know a solution exists, you already know it cannot escape the range of its boundary data, which is exactly the leverage that later proves existence. And a striking global theorem, Liouville's theorem: a harmonic function on the whole plane that stays bounded must be constant. The averaging pressure, applied over arbitrarily large circles, simply will not let a bounded harmonic function vary at all.

It is worth being honest about the boundaries of the boundary principle. It needs a bounded region — on an unbounded domain a harmonic function can grow without ever taking a max anywhere, so you must add a condition at infinity. It is a statement about harmonic functions specifically; the moment you switch to Laplace's equation with a source, Laplacian u = f (Poisson's equation, two guides ahead), the value at a point is the boundary average plus a correction from the source, and pure averaging no longer holds. And the principle controls u itself, not its slope — a harmonic function can have steep gradients at the boundary even while its values stay tame, which is exactly what the Hopf boundary lemma sharpens for the Neumann problem in the next guide.

Step back and feel how much came from one sentence. 'The value at the centre is the average around it' gave us smoothness, the absence of interior extrema, uniqueness and stability of the Dirichlet problem, comparison, a priori bounds, and Liouville's theorem — the entire backbone of elliptic theory, all resting on an idea a child could state. The next guide turns this machinery loose on the actual boundary-value problems, the Dirichlet and Neumann problems, where we stop asking what harmonic functions can't do and start building the ones we want.