Compactness for functions
A sequence of bounded numbers has a convergent subsequence, but a sequence of functions requires a topology before convergence has a meaning. The topology used throughout is uniform convergence on compact subsets, together with a spherical version that allows the limit .
Definition 2.1.1
(Normal family)
Let be a family of holomorphic functions on a domain . The family is normal if every sequence in has a subsequence that either
- converges uniformly on compact subsets of to a holomorphic function, or
- converges locally uniformly to in the spherical metric: for every compact , .
The second alternative is needed because a sequence such as should count as convergent on the Riemann sphere, even though it has no finite limit.
Definition 2.1.2
(Locally bounded family)
A family of functions on is locally bounded if for every compact there is with for every and every .
The constant may depend on but not on ; that uniformity over the family is the whole content of the definition.
Quick check
(Each function in is bounded on every compact set. Is locally bounded?)
Compare the order of the quantifiers in the definition.
Solution · Answer
No. Local boundedness asks for one bound valid for every member of the family. On any compact , , which is unbounded in , so no common exists. Bounding each function separately is a weaker statement.
Montel's theorem for locally bounded families
Theorem 2.1.3
(Montel's theorem)
A locally bounded family of holomorphic functions on a domain is normal. More precisely, every sequence in the family has a subsequence converging uniformly on compact subsets of to a finite holomorphic function.
Proof
(Equicontinuity on compacts and a diagonal subsequence)
Fix a compact . Choose so small that the compact set lies in . For the closed disk lies in , so Cauchy's derivative estimate gives
If and , the segment from to lies in , so for every . Hence the family is equicontinuous on , and it is pointwise bounded by . The Arzelà–Ascoli theorem gives, from any sequence, a subsequence converging uniformly on .
Now take a compact exhaustion of with contained in the interior of , so that every compact subset of lies in some . Extract a subsequence converging uniformly on , from it a further subsequence converging uniformly on , and so on. The diagonal sequence, whose -th term is the -th term of the -th subsequence, is from its -th term onwards a subsequence of the -th subsequence. It therefore converges uniformly on every , hence on every compact subset of . A locally uniform limit of holomorphic functions is holomorphic, so the limit is a finite holomorphic function.
The proof has two separate parts: an analytic estimate that turns a bound on into a bound on , and a purely topological extraction.
Proof X-ray
Where holomorphy enters and where it does not
Holomorphy is used exactly once, in Cauchy's derivative estimate. It converts the hypothesis, a bound on the values of , into a Lipschitz bound, which is equicontinuity. For real-differentiable or merely continuous functions this step fails: is bounded by on but has no uniformly convergent subsequence. The margin matters because the estimate at a point of needs the bound on a disk around it, so the bound is taken on the larger compact set .
The rest is topology. Arzelà–Ascoli works on one compact set at a time, and different compact sets could a priori demand different subsequences. The nested extraction and the diagonal sequence remove that dependence; the condition that lies in the interior of is what makes every compact set sit inside one . The final appeal to the locally uniform limit theorem is again complex analysis, and it is why the limit stays holomorphic rather than merely continuous.
Local boundedness is sufficient, not necessary, for normality in the spherical definition. The constant functions tend locally uniformly to infinity and form a normal sequence, yet are not locally bounded.
Zeros under a limit
The second result describes what a locally uniform limit can do to zeros. It rests on Rouché's theorem on a small circle.
Theorem 2.1.4
(Hurwitz's theorem, zero-free form)
Let be connected and let zero-free holomorphic functions on converge locally uniformly to . Then either has no zeros in or .
Proof
(Rouché on a circle around a zero)
Suppose is not identically zero and . Since is connected, the zeros of are isolated, so there is a closed disk whose boundary circle contains no zero of . On the compact set , has a positive minimum . Uniform convergence on gives on for all large . Rouché's theorem then says that and have the same number of zeros inside , counted with multiplicity. That number is at least one for , so has a zero, contradicting zero-freeness of .
Connectedness was used only to say that the zeros of a function which is not identically zero are isolated. On a disconnected open set that conclusion holds componentwise, and the theorem fails as stated.
Counterexample mode
Hurwitz on a disconnected open set
Claim: if zero-free holomorphic converge locally uniformly on an open set , the limit is zero-free or identically zero. Take , two disjoint disks, and define on and on . Each is holomorphic and never zero on . The convergence is uniform on all of , and the limit equals on and on . So has zeros, yet . The correct statement for an arbitrary open set is componentwise: on each connected component, the limit is either zero-free or identically zero.
The conclusion also allows the zero function: is zero-free but tends to zero. The theorem does not say the zero set of an arbitrary limit is the pointwise limit of zero sets without further hypotheses.
The same Rouché estimate proves a counting version.
Corollary 2.1.5
(Hurwitz's theorem, counting form)
Let locally uniformly on , and let be a closed disk whose boundary circle contains no zero of . Then for all sufficiently large , and have the same number of zeros in , counted with multiplicity.
Proof
(The same estimate without a contradiction)
As above, , and on for large . Rouché's theorem gives equal zero counts inside , with multiplicity.
In particular, a zero of order of the limit can split into nearby zeros of , counted with multiplicity, but no zeros can appear or vanish inside .
Concept lensAnalytic
Zero counts as contour integrals
By the argument principle, the number of zeros of inside is
Locally uniform convergence also gives uniformly on , and on for large , so uniformly on . Hence , the zero count of . Each is an integer, so a convergent sequence of integers is eventually constant and equal to . This is the counting form again, now read as continuity of an integer-valued integral.
Worked examples
Worked example 2.1.6
(A normal family on the disk)
The functions are bounded by on the unit disk, so Montel's theorem says they form a normal family. Here no extraction is needed: on each disk , , so the full sequence converges locally uniformly to zero. It does not converge uniformly on the whole open unit disk, because for every . Normality only promises convergence on compact subsets.
Worked example 2.1.7
(A double zero splitting)
Let . On every compact set uniformly, since . The double zero of the limit at is represented by the two simple zeros of . On the circle with fixed, once , so both functions have two zeros in . The count with multiplicity is stable, even though the number of distinct zeros changes from one to two.
Worked example 2.1.8
(A family which is not normal)
On , consider . Every subsequence takes the value at and tends to infinity at every . A locally uniform limit in the spherical metric would be a continuous map into the Riemann sphere, but the pointwise limit is at the origin and elsewhere, which is discontinuous at . So no subsequence converges and the family is not normal. Consistently with Montel's theorem, it is not locally bounded near : on , .
Common mistake
(Convergence on compacts is not uniform convergence on the domain)
Normality gives uniform convergence on each compact subset, not on itself. The sequence on the unit disk converges to on every disk , but . Statements that need a uniform bound on all of must be proved separately.
Common mistake
(Counting distinct zeros instead of zeros with multiplicity)
Hurwitz's counting form preserves the number of zeros counted with multiplicity. The number of distinct zeros can change: has two distinct zeros, its limit has one. Arguing that a limit with one zero forces approximants with one zero is wrong.
Practice and solutions
Quick check
(Does converge locally uniformly on ? What does Hurwitz's theorem say about the limit?)
Estimate for small .
Solution · Answer
Yes. For , , using from the power series. The limit is the constant . Each is zero-free and is connected, so Hurwitz allows only a zero-free limit or ; the limit is zero-free, as it must be.
Quick check
(Let . Count the zeros of in for all sufficiently large .)
Compare with on the circle .
Solution · Answer
On , when . Rouché's theorem gives the same number of zeros in as , namely three, counted with multiplicity. Directly, the zeros are the three cube roots of , of modulus exactly when .
Quick check
(Zero-free holomorphic on the unit disk converge locally uniformly to with . What can you conclude?)
The unit disk is connected.
Solution · Answer
By Hurwitz's theorem is either zero-free or identically zero. It has a zero at , so on the whole disk. An example is .