Laplace's method (source code)

= Laplace's method
{c}
{wiki}

= Laplace method
{c}
{synonym}

<Laplace's method> estimates an <integral> whose real exponential weight concentrates near a phase maximum. A useful precise case is a smooth real phase $\phi$ on a compact <interval>, with a unique interior maximum $t_0$, $\phi''(t_0)<0$, and a continuous amplitude $a$ satisfying $a(t_0)\ne0$. As $x\to+\infty$,
$$
\int a(t)e^{x\phi(t)}dt
\sim a(t_0)e^{x\phi(t_0)}\sqrt{\frac{2\pi}{x|\phi''(t_0)|}}.
$$
The quadratic <Taylor expansion> of the phase and the rescaling $s=\sqrt{x}(t-t_0)$ turn the local contribution into a <Gaussian integral>. Outside any fixed neighborhood the compact-domain strict maximum supplies a positive phase gap, making the remainder exponentially smaller. On an unbounded interval one must separately check the <localization condition for Laplace's method>.<NIST Digital Library of Mathematical Functions>[§2.3(iii), Laplace integrals and endpoint localization]{ref}

For example,
$$
\int_{-1}^1e^{-xt^2}dt\sim\sqrt{\pi/x}.
$$
The peak width is $x^{-1/2}$. A nonstationary endpoint maximum has width $x^{-1}$ and a different prefactor, given by <Laplace endpoint estimate>. If the amplitude vanishes at the maximum, the displayed equivalence cannot be used with a zero right side: higher amplitude terms or cancellation, as in <endpoint asymptotics after odd Gaussian cancellation>, decide the leading contribution.