method of steepest descent

The method of steepest descent is Laplace's method moved into the complex plane. Many important integrals come with a complex exponential, like the integral of e^(M f(z)) dz along some path, where f is analytic and M is large. The integrand oscillates and grows in complicated ways, so there is no obvious peak to localise around. The brilliant move is to deform the path of integration — which Cauchy's theorem lets us do freely for an analytic integrand — onto a new path where the integral becomes a clean, peaked, Laplace-type problem.

How it works: write f(z) = u(z) + i v(z). The size of the integrand is governed by e^(M u). We look for a saddle point z0 where f'(z0) = 0, then steer the contour through z0 along the direction in which u decreases as fast as possible away from the saddle (the path of steepest descent). Along this special path the oscillating phase v stays constant, so the wild oscillations switch off and the integrand becomes a real, sharply peaked bump that Laplace's method handles directly. The leading result has the same shape as Laplace's: roughly e^(M f(z0)) times sqrt(2 pi / (M |f''(z0)|)), with a phase factor fixed by the orientation of the descent path.

This is the workhorse for the large-argument asymptotics of Bessel functions, Airy functions, the gamma function in the complex plane, and many integral transforms; it is also the mathematical heart of saddle-point evaluation in statistical mechanics and quantum field theory. The honest caveat: choosing the right contour through the right saddle is genuinely an art — there can be several saddle points, and you must include exactly those the deformed path actually crosses (the Stokes phenomenon describes how this set can change abruptly), or your answer will be wrong.