Why should growth constrain zeros?
The product constructions show how to build a function from a prescribed zero sequence. The reverse question is quantitative: if an entire function grows slowly, how densely can its zeros occur? Jensen's formula answers by converting a boundary average of into a weighted count of interior zeros. Zeros are counted with multiplicity throughout.
The formula
Theorem 1.3.1
(Jensen's formula)
Let be holomorphic on a neighborhood of the closed disk , with and no zeros on . Let be its zeros in , repeated by multiplicity. Then
There are only finitely many zeros in the closed disk: otherwise compactness gives an accumulation point in a slightly larger connected disk and the identity theorem forces , contrary to . The boundary hypothesis keeps the logarithm continuous on the circle. Jensen's formula has extensions when boundary zeros occur, but they require a limiting argument and are not being used here.
Proof
(Divide out the zeros with disk automorphisms)
For a nonzero in the disk set
Its denominator vanishes only at , which lies outside the closed disk, so is holomorphic on a neighborhood of the closed disk. For we have , so . Divide by the finite product of these factors. Each zero is canceled to its full multiplicity, giving a zero-free holomorphic function on a neighborhood of the closed disk. It has a holomorphic logarithm on a slightly larger disk: has an antiderivative there, and the multiplicative constant can be absorbed. Hence is harmonic and its value at is its average over the circle. On the boundary, . At zero, , so
Combining the last two equalities proves the formula. This proof uses an actual zero-free quotient and the mean-value property, not a formal logarithm across the zeros of .
Proof X-ray
Why cancel the zeros before averaging?
The obstacle is that is singular at a zero, so the mean-value property cannot be applied to it directly. Each factor removes one zero while having modulus one on the boundary. The quotient is therefore zero-free on a neighborhood of the disk and has a holomorphic logarithm. Its boundary average is unchanged; its value at the origin records exactly for each removed zero. The hypothesis makes the left side finite, and the zero-free boundary makes the right side an integral of a continuous function. The mean-value property closes the argument.
Common mistake
(Applying the formula when )
If the term is and the formula as stated is meaningless. The repair is to factor out the zero first: if has a zero of order at , write with , apply the formula to , and add the boundary mean of , which is .
A counting estimate
Definition 1.3.2
(Zero-counting function and maximum modulus)
For a nonzero entire function , let be the number of zeros of in , counted with multiplicity, and let .
Jensen's formula turns a bound on into a bound on .
Corollary 1.3.3
(Zero counting from growth)
If is entire with and , then
In particular .
Proof
(Drop the nonnegative terms)
Suppose first that has no zero on . Every zero in contributes at least to the sum in Jensen's formula, and the other contributions are nonnegative. The boundary average of is at most . For an arbitrary , select zero-free radii decreasing to from above, which is possible because zeros are isolated, and pass to the limit using continuity of ; the same inequality follows. Taking gives the second bound.
Corollary 1.3.4
(A convergence exponent for the zeros)
If for all sufficiently large , where , then , and converges for every , with zeros repeated by multiplicity.
Proof
(Dyadic shells)
The first claim follows from the previous bound with ; if has a zero of order at , apply that bound to , whose maximum modulus on is for , and add . Only finitely many zeros satisfy . Group the others into dyadic shells . The contribution of shell is at most , a summable geometric bound.
This gives a sufficient exponent, not necessarily the smallest one, and it says nothing about the endpoint .
Common mistake
(Convergence can fail at the endpoint exponent)
It is tempting to conclude from that also converges. Take , with . Its zeros are the nonzero integers, each simple. Since , we have for , so and works. Yet . The corollary is sharp in the sense that cannot be replaced by ; here every gives convergence.
Worked examples
Worked example 1.3.5
(A single linear factor)
For with , we have and one zero inside, so the formula gives the mean of as . For , there are no zeros inside and the mean is . The two answers reflect which side of the circle contains the zero; together the mean is .
Worked example 1.3.6
(A zero-free function)
For , , whose average over is zero, equal to . The sum over zeros is empty. Rapid growth along some directions, here , need not create any zeros: the counting estimate is only an upper bound.
Worked example 1.3.7
(A growth bound of order two)
Suppose for all sufficiently large and . Then for large , so
and the reciprocal-power sum over nonzero zeros converges for every . This does not prove divergence at : a function with only finitely many zeros, such as a polynomial times , also satisfies such a growth bound.
Practice and solutions
Quick check
(Apply Jensen's formula to , where . What is the mean of on ?)
Both zeros lie inside the circle.
Solution · Answer
Here , so the mean equals .
Quick check
(If an entire with obeys for all large , give an explicit upper bound for .)
Use the counting estimate with .
Solution · Answer
Since and , the estimate gives for all sufficiently large .
Quick check
(Why cannot the theorem be applied directly to at the origin, and how can one remove that obstruction?)
Find the order of the zero at .
Solution · Answer
Here , so is not finite. Factor out the known zero: with for and . Apply the formula to at a radius avoiding its zeros and account separately for the double zero at : the mean of on the radius- circle is .