One extra word: injective
By the time you reach this rung you have a holomorphic function feeling like an old friend: differentiable once in the complex sense, hence infinitely differentiable, hence a convergent power series, hence rigid in all the ways the earlier rungs taught. Geometric function theory asks one sharper question of such a function. Not merely is it holomorphic, but is it one-to-one? A holomorphic map f on a domain is called univalent — the old word is schlicht, German for 'plain' or 'simple' — when f(z_1) = f(z_2) forces z_1 = z_2. It never glues two distinct input points to the same output. A univalent function is exactly a holomorphic function that is also injective on its domain.
That one extra word changes everything, because injectivity is a global demand and holomorphy is built from local data. The power series only knows the function near a point; injectivity is a promise about the whole domain at once. Squaring, f(z) = z^2, is as holomorphic as can be, yet it is not univalent on the whole plane: it sends z and -z to the same value, folding the plane in two. But restrict it to the right half-plane, where z and -z can never both live, and on that smaller domain z^2 becomes injective. Univalence is not a property of a formula; it is a property of a formula together with the domain you put it on.
Why the disk, and why we normalize
The natural home for this study is the unit disk D = { |z| < 1 }, and there is a deep reason. From the conformal-mapping rung, the Riemann mapping theorem says that any simply connected domain other than the whole plane is the conformal image of the disk under some univalent map. So the univalent functions on the disk are not one example among many — they are, up to a change of coordinates, the catalogue of all simply connected planar regions and how the disk gets stretched onto them. Understanding univalent maps of D is understanding every conformal picture there is. (Recall the theorem is non-constructive and pointedly excludes the whole plane, which has no such map.)
But there is too much freedom. If f is univalent on D, so is a f(z) + b for any constants a (nonzero) and b — translating and scaling the image is still injective, and gives essentially the same geometry shifted and resized. To get a clean theory we pin down this freedom by normalizing. We agree to study only univalent functions f on D with f(0) = 0 (the center goes to the origin) and f'(0) = 1 (the map starts out at unit speed, neither magnifying nor shrinking infinitesimally at the center). Every univalent f can be brought to this form by subtracting f(0) and dividing by f'(0), so we lose no real generality — we have just chosen a canonical representative of each geometric shape.
Class S = { f holomorphic and univalent on D = {|z|<1} }
with the two normalizations
f(0) = 0
f'(0) = 1
so the Taylor series at 0 reads
f(z) = z + a_2 z^2 + a_3 z^3 + a_4 z^4 + ...
\_/ \____________ the interesting part ___________/
forced by
normalizationThis normalized family is the famous class S — S for schlicht. Because f(0) = 0 the constant term is 0, and because f'(0) = 1 the linear coefficient is exactly 1, so every member has a power series f(z) = z + a_2 z^2 + a_3 z^3 + ... . The geometry of f is now entirely encoded in the higher coefficients a_2, a_3, a_4, ... . The grand program of this rung — and it really is the spine of the whole subject — is to discover how large those coefficients are allowed to be when the function is forced to stay injective. Injectivity is a strong constraint; it should leave a fingerprint on the numbers a_n.
The hero of the class: the Koebe function
Every good theory has an extremal example that pushes every bound to its limit, and for class S that example is the Koebe function. Start from the simple geometric series 1/(1 - z) = 1 + z + z^2 + ... and build k(z) = z / (1 - z)^2. A quick way to see its series: differentiating 1/(1 - z) gives 1/(1 - z)^2 = sum n z^(n-1), and multiplying by z lands you on k(z) = sum n z^n = z + 2 z^2 + 3 z^3 + 4 z^4 + ... . Notice the coefficients are a_n = n: as large and tidy as you could ask for. The Koebe function is in S, and its coefficients grow exactly linearly.
What does it do geometrically? The Koebe function takes the unit disk and maps it univalently onto the entire plane with a single slit removed — the ray from -1/4 straight out to minus infinity along the negative real axis. So it stretches the disk over almost the whole plane, sparing only a thin cut, and the nearest the image ever comes to the origin in the missing direction is the point -1/4. Hold on to that number -1/4: it is no accident, and the next guide's one-quarter theorem will show that no function in S can keep the origin's neighbourhood any clearer than the Koebe function does. It is the boundary case for everything.
The first real bound: |a_2| <= 2
Let us actually catch injectivity leaving a fingerprint, on the very first free coefficient a_2. The clean result, due to Bieberbach, is the bound |a_2| <= 2: for every f in the class S, the second Taylor coefficient satisfies |a_2| <= 2, with equality only for the Koebe function (and its rotations). The Koebe series began z + 2 z^2 + ... , so its a_2 is exactly 2 — it hits the bound dead on, just as the tip above promised it would.
- Start with f in S and form the square-root transform g(z) = sqrt( f(z^2) ). Because f(0) = 0 with a simple zero, f(z^2) has a double zero at 0, so the square root is single-valued and holomorphic near 0 — choosing the branch with g'(0) = 1.
- Check that g is odd and again univalent: g(z) = z + (a_2/2) z^3 + ... , a member of S whose series has only odd powers. Injectivity of f transfers to g through this construction.
- Apply the area theorem (next guide) to 1/g(1/w), an injective map of the exterior of the disk. The area theorem forces the sum of squared coefficients of such a map to be at most 1 — a statement that an enclosed area cannot be negative.
- Read off the very first coefficient inequality from that sum; it says exactly |a_2| <= 2. Equality squeezes all other freedom to zero and pins f down to the Koebe function.
Do not let the brevity fool you — that chain is the genuine logical skeleton, and the engine room is the area theorem, which is exactly why it is the subject of guide 2. The moral worth carrying forward is the shape of the argument: a global geometric fact (an area is nonnegative) gets translated, through a clever transform, into a hard arithmetic bound on a coefficient. Injectivity really did leave its fingerprint, and the fingerprint reads 'at most 2'.
The road ahead: the Bieberbach conjecture and friends
Once you know |a_2| <= 2 and that Koebe with a_n = n hits it, the obvious gamble is irresistible: maybe every coefficient obeys |a_n| <= n for every f in S, with the Koebe function the unique extremal. That is the Bieberbach conjecture, posed in 1916. It is one of the great stories in analysis: easy to state, brutally hard to prove, and it stayed open for sixty-nine years, with a_3, a_4, a_5, a_6 each falling only after serious individual battles. It was finally settled in 1985 by Louis de Branges, and the de Branges theorem — proving |a_n| <= n for all n at one stroke — is the climax of guide 4. The tool that cracked it, Loewner's idea of evolving a univalent map through a family of growing slit domains, is genuinely beautiful, and we will meet it there.
Between here and that summit the rung visits the everyday consequences of staying in S. Guide 2 proves the area theorem and the one-quarter theorem (the image always covers the disk |w| < 1/4). Guide 3 turns the |a_2| <= 2 bound into the Schwarz-lemma-flavoured growth and distortion theorems, which pin |f(z)| and |f'(z)| between explicit ratios depending only on |z| — once again with Koebe sitting on both edges. And guide 5 carves out the most well-behaved sub-families, the starlike and convex univalent functions, where injectivity is guaranteed by a visible geometric shape of the image, together with the Schwarzian derivative, a gadget that detects univalence by measuring how far a map departs from a Mobius transformation.
Keep two honest reminders in your pocket as you climb. First, everything here lives on the disk and assumes the two normalizations f(0) = 0, f'(0) = 1; strip those away and the clean numbers 2, 1/4, n dissolve, because they were measuring deviation from a fixed starting point. Second, univalence is global and subtle — never infer it from a nonvanishing derivative alone, as the e^z example warned. With those caveats kept honest, class S is one of the most rewarding playgrounds in all of analysis, where a single word, injective, blossoms into a sharp and beautiful arithmetic.