Two regions everyone agrees to standardize on
By now you have a toolbox — Mobius transformations, the cross-ratio, the three-point principle — and a reason to use it: transplanting a problem from an awkward region to an easy one. This last guide aims that toolbox at the single most important transplant in the whole subject, the one named maps were built for. The two regions that get to play 'easy' are the upper half-plane, the set of z with positive imaginary part, and the unit disk, the set of w with |w| < 1. Almost every textbook normalizes onto one of these two, and the bridge between them is what we construct here.
Why two standards rather than one? Because each shines for a different shape of problem. The half-plane has a perfectly straight boundary, the real axis, which makes it ideal when your data live along a line — a heated wall, a flat electrode, the edge of a fluid channel. The disk has a boundary you can travel around forever, parametrized by a single angle, which makes Fourier series and rotational symmetry feel natural. Having a clean conformal map between them means you are never stuck: solve wherever the geometry is friendliest, then carry the answer across.
Building the Cayley transform with three points
Let us actually construct the map, and let us do it the principled way rather than pulling a formula out of a hat. The previous guide proved that a Mobius transformation is pinned down completely by where it sends any three points. So we will choose three landmarks on the boundary of the half-plane — the real axis together with the point at infinity — and decree where they should go on the boundary of the disk, the unit circle. A boundary that maps to a boundary is the key idea: if the edge of the half-plane lands on the edge of the disk, the inside has nowhere to go but the inside.
Here is a clean choice of three points and their targets. Send the real-axis point z = -1 to w = -i (bottom of the circle), z = 0 to w = -1 (left of the circle), and z = 1 to w = i (top of the circle). As z runs along the real axis from left to right, w runs around the unit circle, so the line maps onto the circle. Feeding these three correspondences into the cross-ratio equation and solving — exactly the three-point machinery from the last guide — produces one famous map: the Cayley transform, w = (z - i) / (z + i).
Cayley transform: w = (z - i) / (z + i) boundary check (z real): z = -1 --> w = (-1 - i)/(-1 + i) = -i (bottom of circle) z = 0 --> w = (0 - i)/(0 + i) = -1 (left of circle) z = 1 --> w = (1 - i)/(1 + i) = i (top of circle) z = oo --> w = 1 (right of circle) for real z: |z - i| = |z + i| => |w| = 1 (the line lands ON the circle)
Checking it really lands inside the disk
Sending the boundary to the boundary is necessary but not yet enough — it does not by itself tell you whether the inside goes to the inside of the disk or to the outside. A Mobius map could just as well turn the half-plane inside-out, sending its interior to everything beyond the circle. So we owe ourselves one honest interior check. There is a beautiful way to read the answer straight off the formula: for any z, the quantity |w| = |z - i| / |z + i| compares two distances — the distance from z down to the point i, and the distance from z down to the point -i.
Now picture where i and -i sit: i is up in the upper half-plane, -i is its mirror image down in the lower half-plane. Take any z with positive imaginary part — any genuine interior point of the half-plane. It is strictly closer to i (which lives upstairs with it) than to -i (which lives downstairs). So |z - i| is smaller than |z + i|, which makes the ratio |w| strictly less than 1. Every interior point of the half-plane lands strictly inside the disk. The map does not turn anything inside-out; it carries the upper half-plane faithfully onto the open unit disk.
How many maps does the disk allow itself?
The Cayley transform is one bridge, but it is not the only one — and asking how many there are uncovers a deep rigidity. A conformal map of a region onto itself is called an automorphism. Every automorphism of the unit disk turns out to be a Mobius map of the shape w = e^(i alpha) (z - a) / (1 - a-bar z), where a is any interior point and alpha is any rotation angle. Geometrically you get to do two free things: slide one chosen point a to the center, and then spin the disk by alpha. That is exactly three real degrees of freedom — the same count as the three points a Mobius map lets you nominate.
Why does this matter for the half-plane-to-disk map? Because it tells you the map is essentially unique once you nail down a little data. The Cayley transform is one valid bridge; compose it with any disk automorphism and you get another valid bridge. So the freedom in choosing 'the' map from half-plane to disk is precisely the three degrees of freedom an automorphism supplies. Fix where one interior point goes (say, which point becomes the center) and one direction there, and the map is then pinned down uniquely — no remaining wiggle.
And the rigidity goes deeper than counting. The Schwarz-Pick lemma says that no holomorphic map of the disk to itself can increase a certain natural 'hyperbolic' distance between points, and the maps that preserve it exactly are precisely the automorphisms above — nothing else. So the disk is not a soft region you can deform freely; its holomorphic self-maps are tightly constrained, and the only ones that are genuine symmetries are the Mobius automorphisms. That hidden geometry is why a single explicit formula like Cayley's is enough: there simply is not much room for the map to be anything else.
Cashing it in: transplanting a problem onto the disk
Time to spend everything we built. Suppose you must find a steady temperature in the upper half-plane whose values along the real-axis boundary are prescribed — a classic boundary-value problem. Steady temperature is a harmonic function, a solution of Laplace's equation, and that is exactly the kind of object a conformal map preserves: composing a harmonic function with a holomorphic map gives back a harmonic function. So the plan is to carry the whole problem across the Cayley bridge to the disk, solve it there where Fourier methods are easy, and carry the answer home.
- Map the region: send the upper half-plane to the unit disk with the Cayley transform w = (z - i) / (z + i).
- Carry the boundary data along the same map, so the temperatures prescribed on the real axis become prescribed values on the unit circle.
- Solve on the disk: find the harmonic function inside the disk matching those circle values — on the disk this is a standard Poisson-integral or Fourier-series computation.
- Map the answer back: compose with the inverse Cayley transform to land the solution back on the half-plane, where it solves the original problem.
Notice what the conformality bought us at the last step. Because the map preserved angles, the geometry of the solution comes back undistorted: heat flow lines still cross level curves of temperature at right angles, just as physics demands, because right angles are exactly what a conformal map keeps. We did not merely shuffle a formula around; we moved a physical problem to a friendlier room and brought the physics back intact. This is the entire reason the rung opened by proving that holomorphic maps preserve angles — every guide since has been assembling the machine that makes this transplant trustworthy.
The honest big picture, with its caveats
Step back and the whole rung snaps into a single sentence: a holomorphic map with non-zero derivative preserves angles, the Mobius maps are the cleanest such maps and shuffle lines-and-circles freely, three points pin one down, and the Cayley transform spends all of that to identify the half-plane with the disk. Behind it stands the Riemann mapping theorem, which promises that any simply connected region other than the whole plane can be mapped conformally onto the disk — so the disk is the universal 'easy' region, and learning to map onto it is learning to simplify almost everything.
But be honest about the fine print, the same fine print this whole subject keeps insisting on. The Riemann mapping theorem is non-constructive: it guarantees a map exists yet hands you no formula, which is exactly why the explicit Cayley and Joukowski maps remain precious — when you can write the map down, you have something the abstract theorem never gives. It also excludes the entire plane: there is no conformal map from the whole plane onto the disk, a fact forced by Liouville's theorem back in the holomorphy rung, since a bounded entire function must be constant. And the region must be simply connected — no holes — or the clean identification fails.
One last honest note, so you do not over-promise to yourself. Conformal mapping is local-angle-faithful, not distance-faithful: the map stretches different places by different amounts, so a tidy grid in the half-plane comes back warped on the disk even though every crossing stays a right angle. It also lives in two dimensions — the trick that two harmonic-and-conjugate functions package into one holomorphic function is special to the plane and has no higher-dimensional twin. Within those honest boundaries, though, you now hold a complete, usable craft: take a hard planar region, conformally carry it to the disk or half-plane, solve, and carry the answer back with its angles intact.