What kind of convergence should a family have?
In finite-dimensional spaces, a bounded sequence has a convergent subsequence. A sequence of holomorphic functions needs a more careful question: bounded where, and convergent in what sense? Uniform convergence on an entire open disk is often too strong, while pointwise convergence gives too little control. Complex analysis naturally uses locally uniform convergence: uniform convergence on every compact subset of the domain.
Throughout, is a domain, meaning a nonempty connected open set, and is a family of holomorphic functions on .
Definition 2.3.1
(Finite-valued normality and local uniform boundedness)
- is normal (in the finite-valued sense) if every sequence in has a subsequence converging locally uniformly on to a finite-valued holomorphic function.
- is locally uniformly bounded if for each compact there is a finite constant with for every and every .
The constant may depend on , but not on the chosen function. Some conventions also permit a limit of infinity in the spherical metric. The unit Normal families and persistence of zeros uses that broader convention. There, the constants form a normal family; here they are neither locally bounded nor normal. The converse direction below is specific to the finite-valued convention.
Theorem 2.3.2
(Montel's theorem: finite-valued version)
A family of holomorphic functions on is locally uniformly bounded if and only if it is normal in the finite-valued sense.
The proof uses three earlier results: the Cauchy integral formula and its derivative estimate; the Arzelà–Ascoli theorem on a compact metric space; and the Weierstrass theorem that a locally uniform limit of holomorphic functions is holomorphic. It occupies the next two sections: first a lemma on one compact set, then the passage to all compact sets at once.
How holomorphicity turns a bound into equicontinuity
Lemma 2.3.3
(Local bounds give equicontinuity)
If is locally uniformly bounded on , then for every compact the restrictions are uniformly bounded and equicontinuous.
Proof
(A common derivative bound on a compact neighborhood)
Fix a nonempty compact . Because lies a positive distance from the complement of , choose so that the closed -neighborhood of is a compact subset of . Local uniform boundedness supplies one finite bound for all functions in on this neighborhood.
For any in the closed -neighborhood of , the closed disk centered at of radius lies in the larger compact neighborhood. Cauchy's derivative estimate therefore gives
If and , the whole segment between them lies in the closed -neighborhood of . Integrating the derivative along that segment yields
The bound is independent of , so the restrictions to are equicontinuous; they are uniformly bounded by . The segment is used only for nearby points, so need not be convex or connected.
An arbitrary uniformly bounded family of continuous functions need not be equicontinuous: on is bounded by , yet its slopes at grow without bound. The lemma is where holomorphy enters.
One subsequence for every compact set
Choose compact sets whose union is . For example,
omitting initial empty sets; when , the distance to the empty complement is read as infinity. These sets are closed and bounded in and lie inside . A compact is bounded and has positive distance from , so it lies in some .
Proof
(Locally bounded implies normal: nested subsequences and a diagonal)
Let be a sequence in a locally uniformly bounded family . By the lemma and Arzelà–Ascoli, it has a subsequence converging uniformly on . From that subsequence choose a further one converging uniformly on , and continue. Let the diagonal sequence take its th function from the th nested subsequence. For any fixed , all diagonal terms from index onward belong to the th subsequence, so they converge uniformly on .
The limits on the various agree on overlaps, so they define a single finite-valued function on . Every compact subset lies in some , so the diagonal converges to locally uniformly. The Weierstrass theorem makes holomorphic. Hence is normal.
Proof
(Normal implies locally bounded)
Suppose is normal but not uniformly bounded on some compact . For every positive integer choose and with . Normality gives a subsequence converging uniformly on to a finite-valued holomorphic . The continuous is bounded on , say by , and uniform convergence gives on for all large . This contradicts . Hence is locally uniformly bounded.
The two directions have quite different characters, and it helps to see exactly what each hypothesis is doing.
Proof X-ray
Three ingredients and how they lock together
The lemma uses holomorphy once, through Cauchy's estimate on disks of radius around points of the -neighborhood; this is why the bound is taken on the larger -neighborhood rather than on . It produces equicontinuity, the hypothesis Arzelà–Ascoli needs and which boundedness alone cannot give.
Arzelà–Ascoli then acts on one compact set at a time. The nested extraction makes each new subsequence refine the previous ones, and the diagonal is the one sequence that is eventually inside every refinement. The condition , together with the distance-to-boundary term in , guarantees that every compact subset of is covered by a single ; without it the diagonal would converge on each but possibly not on some other compact set. The Weierstrass theorem is used only at the end, to keep the limit holomorphic.
The converse uses only that a uniform limit of continuous functions on a compact set is bounded. Finite-valuedness is essential there: under the spherical convention, converges to and the argument has no finite bound to start from.
Local convergence can coexist with boundary growth
Worked example 2.3.4
(Powers on the unit disk)
Let and . For any compact , some satisfies on . Therefore
The whole sequence converges locally uniformly to zero, and the family is locally uniformly bounded by one, hence normal. Nevertheless, for every ,
as can approach one from below. Thus convergence to zero is not uniform on the entire open disk. There is no contradiction: the open disk is not a compact subset of itself.
Worked example 2.3.5
(The exhaustion of the unit disk)
For , , and is automatic. Hence
For this is the single point , a degenerate but nonempty compact set; is the closed disk of radius , whose interior contains . In general , so lies in the interior of . A compact satisfies , and once . So one diagonal subsequence controls every compact subset of the disk.
Concept lensStructural
Montel as a Heine–Borel theorem for holomorphic functions
With the exhaustion , put
Because every compact subset of lies in some , exactly when locally uniformly, and the Weierstrass theorem makes the holomorphic functions a closed set for . In this metric space, finite-valued normality says that has compact closure. Montel's theorem is the analogue of the finite-dimensional fact that bounded sets have compact closure, with local uniform boundedness playing the part of boundedness.
Worked example 2.3.6
(The same exponentials on two half-planes)
Let , so .
On the left half-plane , every is at most , so the family is locally uniformly bounded and therefore normal. Concretely, a compact has for some , and : the whole sequence tends to locally uniformly.
On the right half-plane , the singleton is compact and is unbounded. The family is not locally bounded, so by Montel's theorem it is not normal in the finite-valued sense. Directly, every subsequence has , so no subsequence has a finite limit at .
Counterexample mode
Pointwise bounds are not the same quantifiers
Claim: a family that is bounded at each point is bounded on each compact set. For continuous functions on this fails. Let
The function vanishes outside . For each fixed this interval eventually lies to the left of , and because . So each sequence is eventually zero and the family is pointwise bounded. Yet , so no common bound exists on the compact set .
These functions are continuous, not holomorphic; they show only that the two quantifier patterns differ. Montel's theorem asks for the stronger one, a single bound for the whole family on each compact .
Common mistake
(Choosing a separate subsequence for each compact set)
Arzelà–Ascoli on , then separately on , and so on, gives a different subsequence for each set. Normality requires one subsequence that converges on every compact set. The nested extraction and the diagonal are what produce it; listing the separate subsequences proves nothing about a single one.
Common mistake
(Bounding derivatives instead of values)
Equicontinuity alone is not enough. The constants all have , so they are equicontinuous, but they have no finite limit. Montel's theorem needs bounds on the values; the derivative bound is a consequence of them, not a replacement.
Exercises and complete solutions
Quick check
(Does converge locally uniformly on ? Uniformly on all of ?)
Separate the supremum over a compact subset from the supremum over the whole disk.
Solution · Answer
Locally uniformly, yes: on each compact there is with . Uniformly on , no: for every , so the supremum of the error does not tend to .
Quick check
(Can a subsequence of converge locally uniformly to a finite-valued function on ?)
Look at one point.
Solution · Answer
No. At the values are , so the subsequence does not even converge pointwise to a finite value there. Equivalently, the family is unbounded on the compact set , so Montel's theorem rules out finite-valued normality.
Quick check
(Does converge locally uniformly on ?)
Every compact set lies in some disk .
Solution · Answer
Yes, to . A compact lies in some disk , so . The same estimate shows that the family is uniformly bounded on each compact set, by . It is not uniform on all of , since .
Normality concerns every sequence in the family and requires one extracted subsequence to work on all compact subsets. The derivative estimate gives compactness on each one; the diagonal argument makes those local choices compatible.