Integrating a family of holomorphic functions
An integral can define a function of a complex parameter. The important question is whether holomorphy survives integration. A pointwise formula for the integrand is not enough: one needs control that is uniform on compact parameter sets.
Theorem 1.13.1
(Holomorphy of parameter integrals)
Let be open and a -finite measure space. Suppose is measurable for the product of the Borel -algebra on and , is holomorphic on for almost every , and for each compact there is with for all and almost every . Then
is holomorphic on , and
Holomorphy comes from Morera's theorem; formula (1) needs a separate domination of difference quotients.
Proof
(Morera for holomorphy, Cauchy estimates for the derivative)
For each closed triangle , the boundary is compact, so there and Fubini's theorem justifies interchanging the contour and parameter integrals. The integral of around is zero for almost every by Goursat's theorem, so . Dominated convergence on compact sets shows that is continuous, and Morera's theorem makes holomorphic.
To justify (1), fix and a closed disk . Cauchy's derivative estimate on circles of radius around points of gives there, outside a common null set. For , integrating along the segment from to bounds the difference quotient by that same integrable function. The derivative is measurable as an almost-everywhere limit of measurable difference quotients. Dominated convergence therefore permits differentiation inside the integral.
Proof X-ray
Where domination is spent, and why the larger disk matters
The hypothesis of compact domination is used three times, each time on a different compact set: on the boundary of a triangle, to make the double integral absolutely convergent so that Fubini applies; on a neighborhood of a point, to make continuous by dominated convergence, which Morera requires; and on the larger disk , to control derivatives. -finiteness is used only for Fubini.
The key move is the passage from a bound on to a bound on . Pointwise differentiability of each gives no rate that is uniform in ; Cauchy's estimate converts a bound on a circle of radius into a derivative bound with the explicit factor . A bound on the integrand on a slightly larger disk therefore yields an integrable dominating function for all difference quotients at once, which is what dominated convergence needs.
Gamma's initial domain
The gamma function is the first and most important example.
Definition 1.13.2
(The gamma function on the right half-plane)
For , define
using the ordinary real logarithm for .
On a compact set in this half-plane choose . Since , for the modulus of the integrand is at most ; for , it is at most . Both bounds are integrable. The theorem therefore makes holomorphic there and gives
Proposition 1.13.3
(Functional equation of )
For ,
Consequently for every integer .
Proof
(Integration by parts with vanishing endpoints)
For ,
The endpoint terms have moduli as and as , the latter because . Letting and gives . Also , and induction gives .
Common mistake
(An off-by-one in the factorial formula)
The identity is , not . For instance , while . The shift comes from the exponent in (2): corresponds to .
Continuing and locating the poles
The functional equation, read backwards, defines further to the left.
Theorem 1.13.4
(Meromorphic continuation of )
extends to a meromorphic function on whose only poles are at . Each pole is simple, and
Proof
(Define strip by strip and glue by the identity theorem)
For an integer , use (3) to define
away from zeros of the denominator. For two choices the formulas agree on by repeated use of (3). Their common half-plane with the finitely many denominator zeros removed is connected, so the identity theorem makes them agree throughout that overlap. Thus they define a meromorphic continuation to the plane, with possible poles only at .
They are genuine simple poles. At , choose in (4). The numerator tends to , while all denominator factors except tend to nonzero numbers. Hence
where the empty product for is one.
This argument establishes the poles and residues; it does not prove that the continuation has no zeros.
Common mistake
(The integral formula is not the continuation)
For the value is defined by (4), not by (2). Substituting into (2) gives , which diverges at , whereas the continued function has the finite value computed below.
Worked examples
Worked example 1.13.5
(Half-integer values of )
Substituting , , in (2) and using ,
Moving right with (3): and . Moving left with (4), gives , and gives
The signs alternate between consecutive poles, as the residues predict.
Worked example 1.13.6
(Residues at and )
At take : , and , so the residue is . At take :
A larger gives the same answer, since (4) defines the same function for every admissible .
Worked example 1.13.7
(Checking domination for )
Let on . On a compact choose with on ; then for , and . The theorem gives holomorphy, and (1) gives
Direct computation agrees: and . The edge of the domain shows the role of the hypothesis: at the would-be bound is not integrable on , and there, so the integral is not absolutely convergent at any point of that line, although is holomorphic at every point of it except .
A missing domination cannot be ignored
Counterexample mode
Entire integrands can have a restricted integral domain
For each , the function is entire. Nevertheless,
has a meromorphic continuation with a pole at zero; the integral itself diverges there. Near there is no integrable function dominating uniformly in : at its modulus is one for all . The parameter integral cannot be declared entire merely because every integrand is entire.
Practice with solutions
Quick check
(Use (3) and to compute .)
Apply the functional equation three times.
Solution · Answer
.
Quick check
(What is the residue of at ?)
Use (4) with .
Solution · Answer
, so the residue is .
Quick check
(Why does the bound near in (2) require ?)
Compute the modulus of .
Solution · Answer
, and is finite precisely for . On a compact parameter set one therefore needs a positive lower bound for .
Quick check
(Does (4) by itself prove that the gamma function has no zeros?)
Ask what (4) says about the numerator.
Solution · Answer
No. Formula (4) describes a continuation and its poles, but a zero of would give a zero of ; excluding zeros requires a further argument.