I will use complex variables to show you how to do these integrals. If you don’t know these methods, you can simply use the results. If you want to learn the methods, they are not difficult, and they are very rewarding. An inexpensive addition to your library would be ” Complex Analysis with Applications” (Dover) by Robert Silverman.
You are all very familiar with
and
with .
Use the fact that the exponential function is holomorphic, consequently for any closed curve in a simply connected open subset of (Cauchy’s theorem)
to evaluate (with real)
on the path shown, in which . Break the path into three parts for which , , and
with , and and explicitly compute in the limit as .
Obtain an explicit expression for ( is real)
We notice that the integrand has no singularities within , so
(1)
(the second integral on the second line has exponent
and so vanishes in the limit. Now let
You are all familiar with
which you obtain by the variable change and completing the square. This does not change the path of integration. But what about
The same transformation changes the path of integration to a line in the complex plane parallel to the real axis.
The correct way to evaluate the integral is by contours, consider that
integrated over the illustrated contour, in the limit .
and the choice turns this into
The contributions from the two vertical segments will be zero
Let’s show that
for is just a Gamma function. Evaluate on this contour. Since need not be integral, we have a cut, which I draw from the origin to infinity and stretch out along the negative real axis (you can place it anywhere).
(2)
and we let and . Hopefully you see that the integrals over the arcs vanish,
and we can analytically continue this integral to negative non-integral using the Gamma function functional relation.
Asymptotic formulas
In many applications we will need to evaluate integrals such as
in which but is not infinite. What we will most often do is simply say that , but we may need to know how significant the error is when we make this approximation.
Integration by parts repeatedly gives you
and the error is exponentially small
We will make heavy use of approximations to , which we will obtain from
A standard method for evaluating
is to find the point where is maximal, and rewrite as
and perform the integration; if we have
replacing with if appropriate.
For our case we obtain Stirling’s approximation
and Stirling’s approximation begets the following very useful (for paramagnetic and polymeric models) expression
This technique is known as the saddle-point method.