Why a branch cut contributes two different values
A contour can encode an integral along a branch cut twice. The upper and lower sides trace the same positive real axis in opposite directions, but a noninteger complex power takes different boundary values on the two sides. The difference between those values is what leaves a useful multiple of the desired real integral.
Theorem 5.4.1
(The integral of over the half-line)
For real ,
The integral is ordinary and convergent: near zero its integrand is at most , and for it is at most . The restrictions and control these two endpoints respectively. The proof occupies the next two sections.
The contour and the residue
Definition 5.4.2
(The slit-plane branch and the integrand)
On choose the branch with , and define
Use a positively oriented keyhole contour bounding a slit annulus and hugging the positive axis: travel outward just above the cut, around a large circle, inward just below the cut, and clockwise around a small circle. The contour encloses the simple pole at . Since , its residue is
Common mistake
(The argument of is fixed by the branch)
On this branch , so . Taking instead gives the residue , and the same computation returns . For this is , a negative value for the integral of a positive function.
The pole lies inside the contour; zero lies in the excluded small disk. For finite radii one may use rays at angles and , circular arcs between them, and first let with fixed. On each side the integrands converge uniformly on , which is compact and avoids the branch point. On the upper ray, the boundary value tends to . On the lower ray, the argument tends to , so the boundary value gains the factor ; that ray is traversed inward. Thus the two radial pieces contribute
Common mistake
(The cut must lie along the slit of the keyhole)
With the principal branch, , the integrand is continuous across the positive axis, so the upper and lower radial integrals cancel and (3) becomes . Its cut, the negative axis, passes through the pole inside the contour, so the residue theorem does not apply as stated. The branch cut has to be the ray along which the keyhole is slit.
The arcs vanish
For , on the large arc and because is real. Its contribution has absolute value at most
For , on the small arc , so its absolute contribution is at most . The arc bounds are uniform in . After taking the ray limit, let and ; the positive integrals increase to . The residue theorem, (2), and (3) therefore yield
Since and , division proves (1).
The proof is now complete, and each hypothesis has been used exactly once.
Proof X-ray
Where the branch, the orientation and each hypothesis enter
The branch with argument in fixes the phase on the lower side; traversing that side inward supplies the minus sign in . Orientation and branch are separate facts, and reversing one of them without adjusting the calculation changes the answer. The hypothesis is used only on the large arc, where , and only on the small arc, where . Finally, gives , so the factor can be divided out. For an integer the two radial pieces would cancel and the contour would say nothing about .
A check at one half
Worked example 5.4.3
(The case checked by substitution)
Taking gives
The substitution gives the same answer independently: the integral becomes .
Worked example 5.4.4
(Integrals of )
For an integer , substitute , so and :
Here lies in . For the formula gives , the value of ; for it gives . Edge case: corresponds to , and indeed diverges.
Worked example 5.4.5
(Rescaling the denominator)
For and , the substitution gives
For and the value is . Independently, gives .
Counterexample mode
The formula does not extend as an ordinary integral to the endpoints
At , the integrand is and diverges near zero. At , it is and diverges at infinity. A formal expression with is not a finite value of either improper integral.
Practice with solutions
Quick check
(For , what is the value of the integral in (1)?)
Evaluate .
Solution · Answer
.
Quick check
(Why does the lower side of the cut contribute a minus sign in (3)?)
Separate orientation from phase.
Solution · Answer
The lower side is traversed from toward , which reverses the integral. The factor is a separate effect of the branch; the phase and the orientation are distinct.
Quick check
(Which endpoint fails if , and which fails if ?)
Compare the integrand with a power of near each endpoint.
Solution · Answer
Infinity fails for , since makes the integrand at least for . Zero fails for , since makes it at least for .
Quick check
(Why is inside the keyhole domain while is not?)
Look at the two radii.
Solution · Answer
The contour encloses the region between the radii off the slit, and lies there. The inner circle excludes a neighborhood of the branch point , where the integrand is not holomorphic.