\( \newcommand\D{\mathrm{d}} \newcommand\E{\mathrm{e}} \newcommand\I{\mathrm{i}} \newcommand\bigOh{\mathcal{O}} \newcommand{\cat}[1]{\mathbf{#1}} \newcommand\curl{\vec{\nabla}\times} \newcommand{\CC}{\mathbb{C}} \newcommand{\NN}{\mathbb{N}} \newcommand{\QQ}{\mathbb{Q}} \newcommand{\RR}{\mathbb{R}} \newcommand{\ZZ}{\mathbb{Z}} \)
UP | HOME

Laplace's Method

Table of Contents

For "sharply peaked" integrands, we can approximate the integral since the dominant contribution comes from the neighborhood of a single point (the peak). Laplace-type integrals are of the form

\begin{equation} I(x) = \int^{b}_{a} f(t)\E^{x\phi(t)}\,\D t \end{equation}

as \(x\to\infty\) and are prime candidates for using Laplace's method.

1. Examples

Consider, as \(x\to\infty\),

\begin{equation} I(x) = \int^{10}_{-10}\E^{-xt^{2}}\,\D t. \end{equation}

This is a Laplace-type integral, with \(f(t)=1\) and \(\phi(t)=-t^{2}\). But we can't apply integration-by-parts since \(\phi'(0)=0\).

A crude estimate: draw the curve then determine its width when

\begin{equation} \E^{-xt^{2}}\sim 1 \end{equation}

which happens when \(t=1/\sqrt{x}\). If we pretend the integrand is a rectangle, we get \(I(x)\sim\mathcal{O}(1/\sqrt{x})\). And it's correct!

Let's examine the tail. The tail region where \(|t|\gg 1/\sqrt{x}\) contribute exponentially small terms. We don't need to examine the tails: they're transcendentally small terms, they do not affect the asymptotic expansion. They're "subdominant".

Now, we take

\begin{equation} \begin{split} I(x) &= \int^{10}_{-10}\E^{-xt^{2}}\,\D t=\left(\int^{\infty}_{-\infty}\E^{-xt^{2}}\,\D t\right) + \mbox{(TST)}\\ &= \sqrt{\frac{\pi}{x}} + \mbox{(TST)} \end{split} \end{equation}

where "TST" stand for "transcendentally small terms".

Question: How do we know the remainder terms really are "transcendentally small terms"?

Solution. We observe the remainder term is

\begin{equation} R(x) = -2\int^{\infty}_{10}\E^{-xt^{2}}\,\D t. \end{equation}

We need to come up with a bound to reflect the remainder term really is a transcendentally small term. For \(t\gt 10\) we see \(t^{2}\gt 10t\), hence

\begin{equation} \begin{split} |R(x)| &\leq 2\int^{\infty}_{10}\E^{-x10t}\,\D t = \left.\frac{\E^{-10xt}}{-10x}\right|^{t=\infty}_{t=10}\\ &\leq\frac{\E^{-100x}}{100x}. \end{split} \end{equation}

Hence we have our hard bound

\begin{equation} R(x)\lt \frac{\E^{-100x}}{100x} \end{equation}

which is clearly as transcendentally small term.

Now consider

\begin{equation} I(x) = \int^{\infty}_{-\infty}\E^{-x\cosh(t)}\,\D t. \end{equation}

To leading order,

\begin{equation} \cosh(t)\approx 1 + \frac{t^{2}}{2} + \mathcal{O}(t^{4}) \end{equation}

as \(t\to 0\). When \(t\) is large, we're out in the tails, and it does not matter that this approximation is no good! Then we have

\begin{equation} \begin{split} I(x) &\sim\int^{\infty}_{-\infty}\exp\left[-x\left(1 + \frac{t^{2}}{2}\right)\right]\D t\\ &\sim\E^{-x}\int^{\infty}_{-\infty}\E^{-xt^{2}/2}\,\D t=\sqrt{\frac{2\pi}{x}}\E^{-x}. \end{split} \end{equation}

Again, the remainder is a transcendentally small term.

2. Higher-Order Terms

Last Updated 2021-06-01 Tue 10:00.