Exponentially improved stationary phase approximations for double integrals (Q1902346)

The usual procedure of obtaining an asymptotic expansion of, say, an integral can be improved by including terms that are exponentially small compared with the terms in the expansion. Even for moderate values of the asymptotic parameter and using a few terms in the expansion, the numerical computation can be more successful when including exponentially small terms. The paper describes the approach for an oscillatory double integral over a domain in the real plane and uses surfaces of steepest descent through complex saddle points. Details are given for an integral with no real critical points, that is, the integral \(\iint_D \exp[i\lambda (z^3+ z) (w^2+ 1)]dz dw\), where \(\lambda\) is large and \(D\) is a bounded domain in the real \((z, w)\) plane. If \(D\) contains the origin then the asymptotic expansion of the integral includes a contribution from the complex saddle \((i, 0)\).