A zero set tells us where an entire function vanishes, but it does not tell us how rapidly the function grows. Multiplication by an exponential changes growth without adding zeros. To make this distinction quantitative, we compare three measurements: the maximum modulus on a circle, the decay of Taylor coefficients, and the density of zeros. The first two determine the same growth order. The third is bounded by that order, but it may be much smaller.
Measuring growth on circles
Definition 1.4.1
(Order of a nonconstant entire function)
Let be nonconstant and entire, and write . Its order is
Since for a nonconstant entire function, the double logarithm is defined for all sufficiently large . We use natural logarithms. A nonzero constant is assigned order zero by convention; the identically zero function is excluded when discussing zeros.
Indeed, if remained bounded, the maximum modulus principle would make bounded on the plane, and Liouville's theorem would make it constant. Since is nondecreasing, unboundedness gives the stated limit.
Common mistake
(Order zero does not mean bounded or polynomial)
Order zero only says that grows more slowly than every power . Every nonconstant polynomial has order zero although it is unbounded, and the first worked example below exhibits a nonpolynomial entire function of order zero. Boundedness, by Liouville's theorem, means constancy.
Theorem 1.4.2
(Order from Taylor coefficients)
If is nonconstant and entire, then
For large the denominator is positive. For a polynomial the right-hand side is defined to be zero. When infinitely many coefficients are nonzero, assigning zero to indices with gives the same limsup. The equality also allows the value .
Proof
(From growth to coefficient decay)
First suppose , and choose . By the definition of limsup, for all sufficiently large . Cauchy's estimate therefore gives, for such ,
Choose ; it is eventually large enough. For every nonzero ,
Thus the coefficient limsup is at most . Letting proves one inequality. Independently, the entire power series has , so the denominators are indeed positive eventually. If , this direction is automatic.
Proof
(From coefficient decay back to growth)
Let denote the coefficient limsup and suppose . Pick ; eventually every nonzero coefficient satisfies
Choose so the coefficient bound holds for . For sufficiently large , set . Split the Taylor series at , isolating finitely many initial coefficients. For , the factor is at most , so the tail is at most . For the remaining terms, treating as a real variable shows
Indeed its maximum occurs at . Hence the at most middle terms contribute at most . The finitely many initial terms grow at most polynomially. Consequently for large , with a constant depending on and . This implies ; let . When the reverse inequality is automatic. For a polynomial, the convention gives coefficient value zero and gives order zero directly.
Proof X-ray
Why the two directions use different tools
The two directions are not mirror images. From growth to coefficients, Cauchy's estimate bounds one coefficient by the whole circle, and the only choice is the radius: minimizes , which is where the term comes from.
Going back, a bound on each coefficient must control an infinite sum, so the key move is the cutoff . Beyond it the hypothesis gives the ratio and a geometric tail. Below it every term is at most by a one-variable maximization, and there are only such terms, which costs . The argument closes because both costs are in the exponent, so ; the finitely many coefficients where the bound may fail contribute only a polynomial.
Examples of order
Worked example 1.4.3
(Exponentials and a zero-order transcendental function)
For , , the maximum on is : the upper bound follows from , and equality holds at . Thus . Likewise has , so its order is infinite.
Consider instead
The root test makes it entire; infinitely many coefficients are nonzero, so it is not a polynomial. Its coefficient ratio is , hence . Zero order allows unbounded, transcendental functions.
Worked example 1.4.4
(Fractional order from factorial coefficients)
For a positive integer , the series is entire. The elementary estimate gives
For this is . For it exhibits order without requiring a closed form.
Counting zeros with multiplicity
Definition 1.4.5
(Counting function and exponent of convergence)
Let list the nonzero zeros of a nonzero entire function, repeated according to multiplicity, and let count those with . Define the exponent of convergence
with . A finite or empty list has exponent zero. The zero at the origin is excluded because its reciprocal is undefined; it is recorded separately by its multiplicity.
Convergence at one exponent implies convergence at every larger exponent, since only finitely many zeros lie inside the unit disk.
Theorem 1.4.6
(Counting exponent and growth order)
For a nonzero entire function,
The first equality concerns the nonzero zero list alone. The inequality may be strict, and neither statement specifies whether the reciprocal-power series converges at .
Proof
(Why the two zero counts agree)
If , then , so the counting limsup is at most . Taking the infimum gives one inequality, also when .
Conversely, if the counting limsup is , choose . For all large , . In the shell , the sum of is at most . The shell bounds are summable. Finitely many zeros inside the first shell add a finite amount. Hence the reciprocal series converges for every , giving . If , the opposite inequality already forces .
Proof
(Jensen's formula bounds zero density)
Factor , where . On , ; for nonconstant , this gives , and a nonzero constant has no zeros. Jensen's zero-counting estimate from the preceding chapter, applied to , gives
For finite and any , the right side is . Therefore the counting limsup is at most ; letting proves . The inequality is automatic when the order is infinite. This proof includes zeros with multiplicity and does not require a zero-free circle: use the continuous-radius version of Jensen's formula, or take zero-free radii decreasing to .
Counterexample mode
Zeros do not determine growth
and have the same empty zero list, so both have , but their orders are and . Even for an infinite zero list, and have the same zeros with the same multiplicities. The first has order one, while the second has order two: gives their upper bounds; gives the first lower bound. Along positive real the second has modulus . Thus the inequality on zero density cannot be reversed.
Worked example 1.4.7
(Order one half and zero density for )
The series is entire and equals whenever , for either choice of . Since ,
The zeros are , : exactly at , and give the same . Each is simple, because there. Thus counts the with , about of them, and . At the critical exponent, diverges. This is the edge case in which the zero-density inequality is an equality.
Common mistake
(The exponent does not decide convergence at the exponent itself)
At the critical exponent, the answer depends on the list. Positive integers have exponent one and diverges. In contrast, the list for also has exponent one, but its reciprocal sum converges. Both lists have no finite accumulation point and can be realized as zero sets by the Weierstrass product theorem. The exponent records the threshold, not the endpoint behavior.
Exercises
Exercise 1.4.8
(Find the order of . Justify both an upper and a lower bound.)
Foundation
Solution(3)
- 1
Upper bound
On , for large , so the order is at most three.
- 2
Lower bound on the positive axis
At the value is , so .
- 3
Conclude
The order is exactly three.
Exercise 1.4.9
(Show that is entire and find its order.)
Foundation
Solution(2)
- 1
Entire
The ratio of consecutive coefficients is , so the series converges on all of .
- 2
Apply the coefficient formula
, so the order is .
Because: .
Exercise 1.4.10
(The nonzero zeros of an entire function are exactly , , all simple. Find their convergence exponent. Must the function have order zero?)
Core
Solution(3)
- 1
Exponent of convergence
converges for every , so .
- 2
A product of order zero
Key step
For , the factors with contribute at most each to , and the rest at most . So and has order zero.
Because: At most indices have , and .
- 3
The order need not be zero
has the same zeros. Along , gives , and the product bound gives the upper bound, so its order is exactly two.
Exercise 1.4.11
(Suppose an entire function has finite order . Show that converges, counting multiplicity. Can equality alone decide convergence?)
Core
Solution(3)
- 1
Below order two
Since , some convergent exponent lies below two, and monotonicity in the exponent gives convergence at two.
- 2
Two functions of order two
Key step
has no zeros, so the sum is empty. The function has order two and simple nonzero zeros , for , so its sum at exponent two is , which diverges.
Because: , and at , .
- 3
Conclude
Order exactly two does not decide convergence at exponent two.