Why we need compactness for functions
The previous guide settled a foundational fact: if a sequence of holomorphic functions f_n converges locally uniformly on a region, the limit f is again holomorphic, and even the derivatives f_n' converge to f'. That is reassuring — the limit does not fall out of the holomorphic world. But notice what it does NOT give you: it tells you what happens IF a sequence converges, not WHETHER you can find a convergent sequence in the first place. The whole strategy for the Riemann mapping theorem hinges on the second question.
Here is the analogy to hold onto. On the real line, a bounded sequence of numbers need not converge — think 0, 1, 0, 1, ... — but the Bolzano-Weierstrass theorem rescues you: it always has a convergent SUBSEQUENCE. That 'bounded implies a convergent subsequence' is the beating heart of compactness, and it is exactly the tool that lets analysts produce limits on demand. We want the same superpower one level up: not for a bounded sequence of numbers, but for a bounded family of FUNCTIONS. Out of infinitely many candidate maps, we want to extract one subsequence that actually converges to something usable.
A normal family is the name for a family of functions that has precisely this property: every sequence drawn from it has a subsequence converging locally uniformly on the region. So 'normal' is to families of functions what 'bounded' was to sequences of numbers — it is the condition that guarantees you can always squeeze out a convergent subsequence. The rest of this guide answers the obvious follow-up: which families are normal, and how do we ever check it?
The missing ingredient: equicontinuity
Boundedness alone is not enough to make a family of continuous functions normal — and the reason is worth feeling in your bones. Consider, on the real interval [0, 1], the functions f_n(x) = sin(n x). Every one of them is bounded by 1, so the family is uniformly bounded. Yet no subsequence converges (even pointwise the values jump around forever as n grows). The family is bounded but wildly un-normal. What went wrong is that the functions wiggle faster and faster: near any point, increasing n makes the graph steeper without limit, so the functions never settle into a common rhythm.
The cure is a uniformity-of-wiggling condition called equicontinuity. Recall that one function is continuous at a point if a small enough change in input forces a small change in output. A whole family is equicontinuous if you can pick ONE 'small enough' that works for every function in the family at once: given a tolerance epsilon, there is a single delta — the same delta for all members — so that points within delta land within epsilon under every f in the family. The functions sin(n x) fail this badly, because as n grows the delta you would need keeps shrinking; no single delta serves them all.
With equicontinuity in hand, the classical compactness theorem snaps into place: the Arzela-Ascoli theorem says that a family of functions on a compact set is normal — every sequence has a uniformly convergent subsequence — exactly when it is both pointwise bounded AND equicontinuous. Read it as a balance sheet: boundedness controls how big the functions get, equicontinuity controls how fast they can change, and together those two controls are precisely what you need to corner a convergent subsequence. The sin(n x) family had the first and lacked the second, and that single missing ingredient was the whole problem.
Montel's theorem: bounded is enough
Now comes the moment where complex analysis pulls ahead of the real-variable story, and it is genuinely surprising. For HOLOMORPHIC functions you get equicontinuity for free — you do not have to check it separately. Montel's theorem states it cleanly: a family of holomorphic functions on a region is normal as soon as it is locally bounded, meaning that on every compact subset there is a single constant M bounding every function in the family. No equicontinuity hypothesis, no derivative bound — just 'uniformly bounded on compact pieces' and normality follows.
Why on earth does mere boundedness force equicontinuity here, when it failed so badly for sin(n x)? Because holomorphic functions cannot wiggle independently of their size — the Cauchy estimates chain the derivative to the values. The Cauchy integral formula expresses f'(z_0) as an integral of f over a small circle around z_0, and reading off the bound gives |f'(z_0)| <= M / r, where M bounds |f| on that circle of radius r. The point is that the SAME M that bounds the family also bounds all their derivatives. A uniform bound on the derivatives is, by the mean value theorem, exactly a uniform Lipschitz-type control on how fast the functions change — and that is equicontinuity.
Cauchy estimate on a circle of radius r about z_0:
1 max |f| on the circle M
|f'(z_0)| = | --- integral of f / (z - z_0)^2 | <= ------------------------ = ---
2 pi i r r
ONE family bound M ===> ONE derivative bound M / r ===> equicontinuity
(same for every f in the family)So the proof of Montel's theorem is a two-step relay. Step one: locally bounded plus the Cauchy estimates give a uniform bound on the derivatives, hence equicontinuity on every compact subset. Step two: bounded plus equicontinuous is exactly the Arzela-Ascoli hypothesis, which delivers a locally uniformly convergent subsequence. The real-variable example sin(n x) could never get off the ground because there is no Cauchy estimate for merely-continuous functions — only holomorphy ties the wiggling to the size, and that tie is what makes 'bounded implies normal' a complex-analysis miracle with no real-variable analogue.
Two sharper forms, honestly flagged
It is worth knowing that 'locally bounded' is sufficient but not the only route to normality. A far deeper result, Montel's fundamental normality test, says that a family of holomorphic functions on a region is normal if there are merely two fixed values, say a and b, that NONE of the functions ever take. That is astonishing: you assume nothing about size, only that everyone in the family omits the same two points, and normality still follows. It is the deep cousin of Picard's theorems and it powers some of the strongest rigidity results in the subject. We flag it here but will not prove it — it is a much harder theorem than the bounded version.
There is also a one-sided strengthening you should not confuse with Montel. Hurwitz's theorem does not produce subsequences; it controls the ZEROS of a locally uniform limit. It says that if holomorphic f_n converge locally uniformly to a non-constant f, then near any zero of f the functions f_n eventually have zeros too, with the right count. The decisive consequence we will lean on later: if every f_n is injective (one-to-one), the limit f is either injective or constant — injectivity cannot quietly evaporate in the limit except by collapsing to a constant. Keep Hurwitz in your pocket; the proof of the Riemann mapping theorem needs it to guarantee the extremal map does not degenerate.
How this powers the Riemann mapping theorem
Now you can see the whole machine assembling. The plan, fleshed out in the next two guides, is to find the conformal map from a simply connected region onto the unit disk by solving an extremal problem: among all injective holomorphic maps of the region into the disk that fix a chosen point z_0, look for the one whose derivative |f'(z_0)| is as LARGE as possible. The candidate maps all land inside the disk, so they are bounded by 1, so the family is locally bounded — and Montel's theorem instantly tells you it is normal.
Normality is precisely the licence to take a maximizing sequence — maps whose |f'(z_0)| climbs toward the supremum — and extract a subsequence converging locally uniformly to a limit map f. The previous guide guarantees f is holomorphic and that f'(z_0) equals the supremum, so the limit really achieves the maximum; it is the extremal map. Then Hurwitz steps in to rule out the degenerate possibility, forcing f to be injective rather than constant, and the Schwarz lemma forces f to be onto. Each piece you have built does one job, and Montel's theorem is the one that conjures the limit out of thin air.