Saddle-point method
Asymptotic evaluation of contour integrals and coefficient formulas using stationary points of the phase.
The saddle-point method is a complex-analytic method for estimating integrals with a large parameter,
as . Here and are analytic near the relevant part of the contour . One deforms , without crossing singularities, through stationary points of along directions on which decreases.
Nondegenerate saddle
A point is a nondegenerate saddle of if
Near such a point,
so integration along a steepest-descent direction reduces locally to a Gaussian integral. The resulting leading factor has order
Its complex phase and square-root branch depend on the orientation of the deformed contour, so there is no contour-independent formula obtained merely by inserting .
Coefficient asymptotics
For an analytic generating function
Cauchy's formula gives
and a positive saddle radius is commonly chosen to satisfy
Define
and let . Under additional admissibility hypotheses that ensure this saddle dominates the rest of the circle,
Analyticity and the saddle equation alone do not imply this formula; one also needs uniform local approximation and decay away from the saddle.
Relation to Laplace's method
The method is the complex analogue of Laplace's method. Both rely on:
- identifying the dominant point (maximum/saddle),
- making a local quadratic approximation,
- evaluating a Gaussian integral.
Practical notes
- Choosing the correct contour (often a steepest descent path) is part of the method; it ensures the contribution away from the saddle is negligible.
- Multiple saddle points can contribute, in which case the leading term may be a sum of their contributions.
- Degenerate saddles, where , require higher-order expansions and have different powers of .