Why encode zeros?
A polynomial is determined up to a nonzero constant by its zeros, with multiplicity. An entire function may have infinitely many zeros, and a naive product of factors need not converge. For example, the factors record the positive integers, but at their partial products tend to zero. We need a convergence condition and, when it fails, factors whose first-order errors cancel. Throughout, compact convergence means uniform convergence on every compact subset of the indicated domain.
Isolated zeros and factorization
Definition 1.1.1
(Zero of order )
A holomorphic function has a zero of order at if near , where is holomorphic and .
Its Taylor expansion and the connectedness argument in the identity theorem below show that a holomorphic function not identically zero has only isolated zeros in a connected domain. Consequently, the zeros of a nonzero entire function can accumulate only at infinity. Multiplicity is essential: and have the same zero set but different local behavior.
Theorem 1.1.2
(Identity theorem)
If two holomorphic functions on a connected domain agree on a set having a limit point in the domain, then they agree everywhere.
Proof
(An isolated zero contradicts a limit point)
Their difference has a zero at the limit point. If it is not identically zero nearby, its first nonzero Taylor coefficient makes that zero isolated, a contradiction. Analytic continuation along overlapping disks, or the open-and-closed argument for points with a zero neighborhood, extends the conclusion to the connected domain.
Counterexample mode
The limit point must lie in the domain
Claim: a holomorphic function that vanishes on a convergent sequence of distinct points is identically zero. On , the function vanishes at every with , and these points converge to ; yet . The limit point lies outside , where the local Taylor argument cannot be run. The repair is the stated hypothesis: the limit point must belong to the connected domain.
Proposition 1.1.3
(Zero-free entire functions have logarithms)
An entire function without zeros has an entire logarithm: for some entire function .
Proof
(Integrating the logarithmic derivative)
Since is entire, its integral is path-independent in ; an antiderivative satisfies , and the nonzero constant can be absorbed into .
Thus, if two nonzero entire functions have identical zeros with multiplicities, their quotient extends to a zero-free entire function and they differ by for some entire . Zeros alone do not determine an entire function up to a constant: and already show this.
Infinite products
Theorem 1.1.4
(Absolutely convergent infinite products)
For complex numbers , if , then converges; its limit is nonzero provided no factor is zero. More generally, if are holomorphic and for every compact , the partial products converge uniformly on compact sets to a holomorphic function. On a compact set where the factors have no zeros, the limit is nonzero.
The proof occupies the next subsection.
Why the product criterion works
Proof
(Uniform Cauchy tails and a logarithm for the tail)
Fix a compact set and write . The hypothesis says . For a finite tail, expanding the product and then replacing every term by its absolute value gives
The last inequality follows from for . The right-hand side tends to zero as , uniformly in and . To compare two whole partial products, multiply this tail difference by the earlier product. Its modulus is at most on . Thus the partial products are uniformly Cauchy on and converge uniformly there. Since this works on every compact set, the limit is holomorphic by the locally uniform limit theorem.
Why can the limit not acquire an extra zero? Choose so large that on for . The power series
defines a holomorphic logarithm of near , with on . The sum of these logarithms converges uniformly on , so the limiting tail is the exponential of a finite-valued sum and is nonzero. Consequently the full product is the finite initial product times a nonzero tail. If no factor is identically zero, then at any chosen point the zero order is the sum of the orders of the finitely many factors that vanish there. If a factor is identically zero, the whole product is identically zero. For the order statement, apply the compact convergence criterion on a slightly larger closed disk to obtain a nonzero holomorphic tail near the point, and factor each vanishing initial holomorphic function by its local power of .
The argument has two halves, convergence and nonvanishing, and they use the hypothesis differently.
Proof X-ray
Head, tail, and why a logarithm is needed for zeros
The proof splits the product into a finite head and a tail. For convergence, the tail's deviation from is bounded by , which uses only and ; the head is bounded by , so the partial products are uniformly Cauchy. Nonvanishing is the subtle half, because a limit of nonzero partial products can be zero, as shows. The logarithms , available once , satisfy , so summability is used a second time, inside the logarithm, to write the tail as the exponential of a uniformly convergent series. An exponential never vanishes, and all zeros of the product come from the finite head.
The sufficient condition is not necessary, as an ordered product with cancellation shows.
Counterexample mode
A convergent product with divergent absolute sum
Claim: if converges, then . Put and for . Each consecutive pair has product
The even partial products are and the odd partial products tend to , so this ordered product converges to although diverges. Cancellation can therefore produce convergence beyond the absolute criterion; the criterion remains valuable because it gives a uniform tail bound and control of zeros without relying on such cancellation. For nonnegative the claim is true, since .
Definition 1.1.5
(Elementary factors)
For a sequence of nonzero numbers with , the elementary factors are
For , the logarithm of begins at degree . Hence on there is a constant with .
Theorem 1.1.6
(Canonical products of fixed genus)
Let be nonzero with , and let be an integer. If , then
converges locally uniformly and has exactly the listed zeros, counting repetitions.
Proof
(The product criterion on each disk)
Fix . Only finitely many have . For the others, on , so , a summable bound. The product criterion therefore gives locally uniform convergence, and on the disk the zeros of the product are those of the finitely many factors ; the factor vanishes only at , simply.
This is a useful special case of Weierstrass factorization. The general theorem permits the genus to vary with and thus constructs a product for any zero sequence escaping to infinity; that general construction is not proved here.
Differentiating an infinite product
A finite product turns multiplication into addition after division by the product. For an infinite product the same idea is useful, but it requires two separate checks: the product must converge, and division must be performed away from its zeros. It is not enough to differentiate each factor formally.
Theorem 1.1.7
(Logarithmic derivative of a locally uniform limit)
Let be holomorphic on a domain and suppose locally uniformly, where is not identically zero. On every compact subset of ,
Proof
(Keeping the limit away from zero)
The locally uniform limit theorem gives uniformly on . Since has no zero on , continuity and compactness give . For sufficiently large , on , so the quotients converge uniformly. In particular the eventual nonvanishing of is a consequence, not an extra assumption. This argument says nothing about convergence of quotients at zeros of .
Worked example 1.1.8
(Why a correction term survives differentiation)
Define
The criterion above gives local uniform convergence and simple zeros exactly at the positive integers. For its finite partial product, the product rule and the derivative of each exponential give
Fix a compact set , and choose with on . For , , so the summand has modulus at most . The finitely many earlier summands are holomorphic on a neighborhood of , and the tail converges uniformly by the majorant. Applying (1) therefore proves
with locally uniform convergence on that punctured domain. Each term of the uncorrected sum behaves like ; in fact its partial sums diverge, since the difference from converges by the same estimate. The terms come from the exponential factors and cannot be dropped. Formula (2) determines the logarithmic derivative, but multiplying by a nonzero constant leaves that derivative unchanged.
Near a zero , separate its term in (2). The remaining series is holomorphic in a sufficiently small disk about , by uniform convergence on compact subsets. Thus there, where is holomorphic. The coefficient records the simple zero; the formula is valid off the zero, not at it.
Quick check
(Let . Find and show that the logarithmic-derivative series converges locally uniformly away from its zeros. Why can one not cancel the exponential corrections in (2)?)
Bound on a disk once is large.
Solution · Answer
The same argument gives off : on a compact disk the tail is bounded by once exceeds twice its radius; the finitely many earlier terms are holomorphic away from their poles. At zero, and hence . In (2), deleting the corrections leaves divergent harmonic-sized terms, even though the corrected sum converges.
Worked examples
Worked example 1.1.9
(A product without correction)
With , , so defines an entire function with simple zeros at . On , the sum of the tail factor errors is bounded by ; the absolute difference of the tail product from is at most .
Worked example 1.1.10
(A first-order correction)
With , the uncorrected errors have divergent absolute sum for nonzero . But , so converges locally uniformly. Its zeros are exactly the positive integers and all are simple. The exponential is not decorative: it cancels the linear term in the logarithm.
Worked example 1.1.11
(Growth controls a polynomial)
If is entire and for a nonnegative integer , then is a polynomial of degree at most . Cauchy's estimate gives for every . Let when ; those Taylor coefficients vanish. The edge case is Liouville's theorem: a bounded entire function is constant.
Common mistake
Common mistake
(Forgetting a vanishing factor)
A locally uniform product may vanish at a point because a factor vanishes there. The assertion that an infinite product has a nonzero limit requires every factor to be nonzero at that point as well as the summability condition.
Common mistake
(Factors tending to are not enough)
Convergence of the individual factors to does not give a nonzero limit: tends to zero, although . Here , so the summability condition fails.
Practice and solutions
Quick check
(Show that is entire and identify its zeros.)
Find a summable majorant on .
Solution · Answer
The summable majorant on each disk gives local uniform convergence. The zeros are precisely , each simple, by separating the relevant factor from the nonvanishing tail.
Quick check
(Suppose and are nonzero entire functions, have the same zeros with multiplicity, and is bounded after removable singularities are filled in. What can be concluded?)
The quotient is entire; apply Liouville's theorem.
Solution · Answer
The quotient is entire and zero-free; Liouville's theorem makes it a nonzero constant. Hence .
Quick check
(Why does the identity theorem not force to be identically zero although it vanishes at every integer?)
Where would the zero set need a limit point?
Solution · Answer
The integers have no finite accumulation point in ; accumulation at infinity does not meet the hypothesis.