The question the mean-value property could not answer
In the previous guide you met the mean-value property: a harmonic function at the centre of a circle equals the plain average of its values around that circle. That is a beautiful fact, but read it carefully and you will notice it only speaks about the centre. It tells you the value at one privileged point from the whole boundary, and it says nothing at all about an off-centre point. Yet the off-centre points are exactly what a physicist wants: if u is the steady temperature of a metal disk and you fix the temperature all around the rim, what is the temperature at some point halfway out toward the edge? The mean-value property shrugs.
This is the Dirichlet problem, the central problem of potential theory in the plane. Stated cleanly: given a continuous function f on the boundary circle of a disk, find a function u that is harmonic inside the disk and that approaches f as you walk out to the boundary. The Dirichlet problem is the mathematical skeleton of a hundred physical situations — steady heat with prescribed edge temperatures, an electrostatic potential with charged plates, the displacement of a soap film clamped to a wire loop. In every case the interior obeys Laplace's equation and the boundary is dictated from outside. The goal of this guide is a formula that solves it.
The Poisson integral: a weighted average that sees every interior point
Here is the leap. The mean-value property averages the boundary with EQUAL weight, and so it can only see the centre, where every boundary point is genuinely equidistant. To reach a point z that is off-centre — closer to one arc of the rim than the other — we must weight the boundary unequally, leaning harder on the nearby arc and lightly on the far one. The Poisson kernel is precisely this set of weights. For the unit disk, the value of the Poisson integral formula at an interior point z = r e^(i phi) is a weighted average of the boundary data f(e^(i theta)), where the weight is the Poisson kernel P_r(phi - theta) = (1 - r^2) / (1 - 2 r cos(phi - theta) + r^2).
u(r e^(i phi)) = (1 / 2 pi) * integral_0^(2 pi) P_r(phi - theta) f(e^(i theta)) d theta Poisson kernel: P_r(phi - theta) = (1 - r^2) / (1 - 2 r cos(phi - theta) + r^2) at the centre r = 0: P_0 = 1 -> u(0) = ordinary average of f (mean-value property!) properties: P_r > 0 and (1/2 pi) integral_0^(2 pi) P_r d theta = 1 (a true averaging weight)
Look at how the kernel behaves and the whole mechanism becomes visible. The numerator 1 - r^2 is the same everywhere on the boundary; it is the denominator that does the steering. When phi - theta is small — that is, when the boundary point e^(i theta) sits in the direction we are aiming at — the cosine is near 1 and the denominator (1 - r)^2 is small, so the weight P_r is LARGE. When the boundary point is on the far side, the cosine is near -1, the denominator (1 + r)^2 is large, and the weight is small. So the kernel automatically spotlights the part of the rim nearest to z. As r climbs toward 1 and z slides out to the boundary, that spotlight sharpens into a spike, and the average locks onto the single boundary value right in front of it.
Why this formula truly solves the problem
A formula is only a solution if it does two jobs: the interior it produces must be harmonic, and the boundary it approaches must be the given f. The Poisson integral passes both tests, and it is worth seeing why, because the reasons are honest and not magical. For the first, the kernel P_r(phi - theta), viewed as a function of the interior point z, is itself harmonic in z — it is the real part of a simple holomorphic expression, (w + z)/(w - z) with w = e^(i theta) on the boundary. An average of harmonic kernels, weighted by fixed numbers f(e^(i theta)), is again harmonic. So u inherits harmonicity straight from the kernel.
For the second job — that u reaches f at the boundary — recall the spotlight. As z moves out to a boundary point e^(i theta_0), the kernel piles almost all of its weight onto a tiny arc around theta_0, and because P_r integrates to exactly 1, the weighted average becomes essentially f(e^(i theta_0)). The technical name for a kernel that does this is an approximate identity: a family of weights that concentrate to a single point in the limit. The one genuine hypothesis we need is that f is continuous, so that the values on that tiny arc are all close to f(e^(i theta_0)); a wildly discontinuous f could disagree with the limit exactly at its jumps.
And uniqueness is a gift from the previous guide. Suppose two harmonic functions both took the boundary values f. Their difference is harmonic and equals zero on the entire boundary. By the maximum principle for harmonic functions, a function harmonic inside and zero on the boundary cannot rise above zero anywhere (its maximum lives on the boundary) and cannot dip below zero anywhere (apply the same to its negative), so it is identically zero. The two solutions were the same all along. This is why we may speak of the solution: the Poisson integral does not just find a solution, it finds the only one.
A tiny worked example: the off-centre temperature
Numbers make this real. Heat the right half of the rim (where cos theta > 0) to 1 degree and the left half to 0 degrees, a step function on the circle. What is the steady temperature at the centre? The mean-value property answers instantly: equal weighting of a function that is 1 on half the circle and 0 on the other half gives exactly 1/2. By symmetry that matches intuition — dead centre feels equal pull from a hot side and a cold side. Now ask the question the mean-value property could not: what is the temperature partway out toward the hot side, say at z = 1/2 on the positive real axis?
Now the Poisson kernel earns its keep. At z = 1/2 (so r = 1/2, phi = 0) the kernel leans toward theta = 0, the heart of the hot arc, and away from the cold left side. Without grinding out the integral, you can already predict the qualitative answer with confidence: the temperature must be ABOVE 1/2, because the weighting now favours the hot half. Carrying out the Poisson integral confirms it — the value comes out to about 0.69. The point closer to the hot rim runs hotter than the centre, exactly as a real heated plate would, and the formula delivered the precise number, not just the trend. This is the payoff the mean-value property could never give.
From the disk to everywhere: conformal transplant
The disk is one shape; the world is full of others — a half-plane, the inside of a square, the region outside an airfoil. Solving the Dirichlet problem afresh on each would be hopeless. The escape is one of the most beautiful unifications in the subject: harmonicity is preserved by conformal maps. If w = g(z) is a holomorphic, one-to-one map carrying a domain D onto the unit disk, and U is harmonic on the disk, then U composed with g is harmonic back on D. A conformal map is a perfect translator between the two boundary-value problems, so a single solved case — the disk — unlocks a whole family of domains.
- Find a conformal map w = g(z) taking your domain D onto the unit disk, and note where it sends the boundary of D onto the boundary circle.
- Transport the boundary data: a value prescribed at a boundary point of D becomes the value at its image on the unit circle, giving a Dirichlet problem ON the disk.
- Solve that disk problem with the Poisson integral, producing a harmonic U on the disk.
- Pull the answer back: u(z) = U(g(z)) is harmonic on D and matches the original boundary data — the solution on your real domain.
Two honest caveats keep this from being a free lunch. First, you must actually FIND the conformal map, and for awkward domains that can be the hard part — for the simplest shapes a Mobius transformation does it, but a polygon needs the heavier Schwarz-Christoffel machinery. Second, the Riemann mapping theorem guarantees that a map onto the disk EXISTS for any simply connected domain that is not the whole plane, but it is non-constructive — it promises the translator without handing you its words, and it pointedly excludes the entire plane (where, by Liouville's theorem, a bounded harmonic function must be constant, so the boundary-value game changes entirely).
What the Poisson integral hands you next
Two consequences fall straight out of the formula and deserve a name. Because the kernel P_r is strictly positive, a non-negative boundary data forces a non-negative u, and squeezing the kernel between its smallest and largest values turns this into a two-sided estimate: at any fixed interior radius, the value of a positive harmonic function is trapped between fixed multiples of its centre value. That is Harnack's inequality, and Harnack's inequality is the quantitative muscle behind much of potential theory — it says positive harmonic functions cannot vary too wildly, and it powers convergence theorems for increasing sequences of harmonic functions.
There is also a converse worth savouring, because it ties this guide back to the very start of the rung. We built the Poisson integral to SOLVE for a harmonic function from boundary data. But run the logic the other way: it can be shown that a continuous function satisfying the mean-value property at every point is automatically harmonic, and is automatically the Poisson integral of its own boundary values. So the mean-value property, the Poisson integral, harmonicity, and being the real part of a holomorphic function are four faces of one object. The thread from guide one — harmonic functions as the real parts of holomorphic ones — and the thread of this guide meet here in a single knot.
The final guide of this rung takes the conformal idea seriously as physics. Once you accept that a harmonic u always has a harmonic partner v, the pair combines into a single holomorphic complex potential f = u + i v, and its derivative encodes a flow — of heat, of charge, of an ideal fluid. Conformal maps then become a design tool: bend a simple flow in the disk into the flow around a real obstacle. Everything you have built here — the Poisson solution, the conformal transplant, the maximum principle — becomes the engine room of that physical picture.