Weierstrass products can prescribe zeros, but a general product may need factors of increasing genus and says little about the growth of the remaining exponential. Finite order supplies stronger control: one fixed genus is enough, and the exponent left over is a polynomial of bounded degree. This extends the genus-zero argument for order below one; it does not require a lower bound for the product near its zeros.
Canonical factors and the factorization theorem
For a nonzero entire , let . Its order is the infimum of nonnegative such that, for every , as . Assume and put .
Definition 1.12.1
(Canonical factors)
For an integer , the canonical factor of genus is
Each is entire with a single simple zero, at .
Theorem 1.12.2
(Finite-order Hadamard factorization)
Let be nonzero and entire of finite order . Let be its zero multiplicity at and list its nonzero zeros with multiplicity. With , there is a polynomial of degree at most , with , such that
The product converges locally uniformly; the empty product is . Zeros alone therefore determine only up to the indicated exponential polynomial and nonzero constant.
We use Jensen's formula, the zero-free entire logarithm theorem and the standard locally uniform product criterion. Their hypotheses will be checked in the proof.
Proof
(Count zeros, bound the product, control the exponent)
Set , filling the removable value . Removing this factor preserves the order bound. Jensen's formula on a zero-free circle of radius between and shows that the number of nonzero zeros in , counted with multiplicity, is for every : each zero inside contributes at least . Choose
Such numbers exist even if is an integer. Splitting the zeros into dyadic annuli shows , since the contribution from is . In particular , so the canonical product converges locally uniformly, has exactly the prescribed zeros, and satisfies .
We need an upper growth estimate, not a pointwise lower bound. There is a constant such that
For , the logarithmic series gives , so the left side is . For , the definition bounds by , which is because ; the compact intermediate annulus is harmless. Hence
The quotient extends across the zeros to a zero-free entire function. Choose its entire logarithm with , so . Put and . On circles avoiding zeros of , Jensen's formula and give mean . Thus the mean of is at most the mean of , which is by (3). Because , the growth bound for gives mean . The mean of is , so the mean of is twice the mean of and is also .
Write . Taking a Fourier coefficient of on this circle yields for . Thus . Let through radii without zeros. Every with vanishes. Therefore is a polynomial of degree at most , and . This proves (1).
Proof X-ray
Why division near zeros is safe
Finite order enters twice. Through Jensen it bounds the zero count, , which fixes the genus and makes the product converge; through it bounds the positive part of . The room is the hinge: makes the large- part of (2) work, and makes the small- part work, so a single exponent serves both regimes.
The key move is to avoid any lower bound for near its zeros. Jensen's formula with says that the mean of is nonnegative, so the mean of is paid for by the mean of , which (3) controls. Because the mean of is , the one-sided information becomes a bound on the mean of . The proof closes with Fourier coefficients: an bound on the boundary values of the harmonic function kills every Taylor coefficient of of degree above , and the quotient is entire in the first place only because the zeros of match those of with multiplicity.
Examples and a boundary warning
Worked example 1.12.3
(The degree bound is attained, and zeros do not fix )
The function has no zeros and order exactly : , attained at . Here , , the product is empty and , so the bound is attained. By contrast and have identical empty zero sets and value at zero, but differ: their factorizations have and . Zero data and one normalization do not generally determine a finite-order entire function.
Worked example 1.12.4
(The product for )
The zeros of are the integers, all simple. Since and , the order is , so , and . The theorem gives
for some . This product converges absolutely, so its factors may be grouped in pairs, . Both and are even, so away from the integers. Hence the entire function equals identically, and differentiating at gives :
Worked example 1.12.5
(Genus zero: no exponential factor for )
Let , which equals whenever . Since , , while ; so the order is exactly and . The zeros are , , all simple, and , . Now and force :
This edge case is the only one in which the factorization leaves no free exponential beyond the zeros and .
Common mistake
(Pairing factors is not absolute convergence of a genus-zero product)
For the genus-zero product is not absolutely convergent, since diverges; the theorem uses genus , whose factors carry the convergence factor . Pairing gives a convergent product for that particular grouping, but a change of grouping is not a proof that the unpaired genus-zero product converges.
Counterexample mode
Without finite order the exponent need not be a polynomial
One might expect every zero-free entire function to be with a polynomial. The function is zero-free and entire. If with a polynomial, then would be constant, so its derivative would vanish and would be a polynomial, which is false because no derivative of vanishes. The theorem does not apply because , so and the order is infinite. The finite-order hypothesis is exactly what forces the exponent to be a polynomial.
Practice with solutions
Quick check
(A zero-free entire function has order less than and value at zero. What forms can it have?)
Solution · Answer
Since , , and the product is empty, it is for some , including . Each such function qualifies.
Quick check
(Can have order less than ? Explain using its maximum modulus.)
Solution · Answer
No: is not for .
Quick check
(Why does (3) alone not give a positive lower bound for valid on every sufficiently large circle?)
Solution · Answer
An upper bound on gives no lower bound; a circle can pass through a zero of . On one fixed zero-free circle, continuity does give a positive lower bound that may depend on the circle. The proof instead chooses zero-free circles and estimates the mean negative logarithm.