Asymptotic Analysis

Asymptotic analysis studies approximations in a limiting regime: a small parameter tends to zero, a quantum number becomes large, a distance goes to infinity, or an action scale becomes large compared with $\hbar$.

The central idea is not that an infinite series converges. It is that a finite number of terms can approximate a quantity with a controlled remainder in the limit being studied.

Why Quantum Mechanics Needs It

Quantum mechanics uses asymptotic reasoning whenever exact solutions are unavailable but a limiting regime is physically meaningful. Common examples include:

• perturbation theory in a small dimensionless coupling;
• WKB expansions in powers of $\hbar/S$;
• stationary-phase approximations to oscillatory integrals;
• saddle-point expansions of path integrals;
• high-quantum-number and large-distance limits;
• scattering wavefunctions in asymptotic regions;
• tunneling exponents and exponentially small effects.

The approximation volume explains many physical methods. This page supplies the mathematical grammar: big-O notation, asymptotic series, stationary phase, and saddle points.

Prerequisites

You should be comfortable with:

• sequences and series from Sequences, Series, and Convergence;
• limits, derivatives, and integrals from Real Analysis Essentials;
• contours and analytic continuation from Complex Analysis Essentials for saddle-point methods.

Big-O and Little-O Notation

Write

$$f(\epsilon)=O(g(\epsilon)) \qquad \text{as }\epsilon\to0$$

if there are constants $C>0$ and $\epsilon_0>0$ such that

$$\lvert f(\epsilon)\rvert \le C\lvert g(\epsilon)\rvert$$

for $0\lt\epsilon\lt\epsilon_0$.

Write

$$f(\epsilon)=o(g(\epsilon)) \qquad \text{as }\epsilon\to0$$

if

$$\frac{f(\epsilon)}{g(\epsilon)} \to0.$$

Thus $o(g)$ is smaller than $g$ in the limit, while $O(g)$ is at most comparable to $g$ up to a constant.

For example,

$$\sin\epsilon = \epsilon - \frac{\epsilon^3}{6} + O(\epsilon^5)$$

as $\epsilon\to0$. This says the neglected terms are bounded by a constant times $\epsilon^5$ for sufficiently small $\epsilon$.

Asymptotic Expansions

An expression

$$F(\epsilon) \sim \sum_{n=0}^{\infty} a_n\epsilon^n \qquad \text{as }\epsilon\to0$$

means that, for each fixed $N$,

$$F(\epsilon) - \sum_{n=0}^{N} a_n\epsilon^n = O(\epsilon^{N+1}) \qquad \text{as }\epsilon\to0.$$

The words “for each fixed $N$” matter. The definition does not say that the infinite series converges to $F(\epsilon)$ at a fixed nonzero value of $\epsilon$.

Many useful quantum expansions are asymptotic rather than convergent. Adding terms can improve the approximation at first, then eventually make it worse. The best truncation often occurs near the smallest term.

Convergent Versus Asymptotic

A convergent power series has a radius of convergence and defines a function inside that radius. An asymptotic series is instead a hierarchy of approximations in a limit.

The distinction is visible in error logic:

• for a convergent series, taking more terms eventually approaches the sum;
• for an asymptotic series, taking more terms helps only while the terms continue to decrease in the relevant regime.

This is why a result such as

$$F(\epsilon) = a_0+a_1\epsilon+a_2\epsilon^2 + O(\epsilon^3)$$

is often more honest than writing a formal infinite series. The displayed formula says exactly what has been controlled.

Regular and Singular Limits

A regular perturbation changes the answer smoothly as the parameter tends to its limiting value. A singular perturbation changes the character of the problem.

In quantum mechanics, singular behavior appears when:

• a small parameter multiplies the highest derivative;
• a turning point makes a WKB denominator vanish;
• a continuum threshold changes analytic structure;
• a near degeneracy makes perturbative denominators small;
• an exponentially small term is invisible to every power of the expansion parameter.

For example, WKB expansions fail at classical turning points because the local momentum $p(x)$ vanishes and the usual amplitude $1/\sqrt{p(x)}$ diverges. The repair uses local Airy Functions and connection formulas rather than more terms of the same local expansion.

Stationary Phase

The model oscillatory integral is

$$I(\epsilon) = \int_a^b A(x) \exp \left[ \frac{i}{\epsilon}S(x) \right]dx, \qquad \epsilon\to0^+.$$

If $S'(x)$ is nonzero on an interval and there are no boundary contributions, rapid oscillations tend to cancel. The leading contributions usually come from stationary points

$$S'(x_\star)=0.$$

For a nondegenerate stationary point with $S''(x_\star)\ne0$, the leading local contribution is

$$I(\epsilon) \sim A(x_\star) \exp \left[ \frac{i}{\epsilon}S(x_\star) \right] \exp \left[ i\frac{\pi}{4} \operatorname{sgn}S''(x_\star) \right] \sqrt{ \frac{2\pi\epsilon} {\lvert S''(x_\star)\rvert} }.$$

This formula is the finite-dimensional ancestor of the semiclassical statement: phases far from stationary action cancel, while neighborhoods of classical paths add coherently.

If $S''(x_\star)=0$, the ordinary formula fails. A degenerate stationary point requires a different scaling, often involving Airy Functions or other special functions.

Saddle-Point and Steepest-Descent Methods

Stationary phase is the oscillatory real-variable version of a broader complex method. Consider

$$I(\lambda) = \int_C A(z)e^{\lambda\Phi(z)}\,dz, \qquad \lambda\to+\infty.$$

Saddles satisfy

$$\Phi'(z_\star)=0.$$

If the contour can be deformed to a path of steepest descent through a nondegenerate saddle, the leading contribution has Gaussian form:

$$I(\lambda) \sim A(z_\star)e^{\lambda\Phi(z_\star)} \sqrt{ \frac{2\pi} {-\lambda\Phi''(z_\star)} },$$

with the square-root branch determined by the contour orientation and the descent direction.

Several cautions are essential:

• not every saddle lies on a contour reachable from the original integral;
• multiple saddles can compete;
• a saddle can turn on or off across Stokes lines;
• branch cuts and poles crossed during contour deformation contribute separately.

These are not technical decorations. They determine tunneling exponents, instanton sectors, resonance approximations, and the large-order behavior of perturbation theory.

Worked Example: Gaussian Scaling

The simplest saddle-point estimate is the Gaussian integral

$$I(\lambda) = \int_{-\infty}^{\infty} e^{-\lambda x^2/2}\,dx, \qquad \lambda>0.$$

The exponent has a saddle at $x=0$. Since

$$-\frac{\lambda x^2}{2}$$

is already quadratic, the saddle approximation is exact:

$$I(\lambda) = \sqrt{\frac{2\pi}{\lambda}}.$$

This scaling explains why quadratic fluctuations around a saddle often produce determinants and factors of the large parameter to a half power. In many-dimensional Gaussian integrals, the second derivative becomes a Hessian matrix.

Worked Example: Stationary Phase Scaling

With a regulator understood, the oscillatory Gaussian

$$I(\epsilon) = \int_{-\infty}^{\infty} \exp \left( \frac{i x^2}{2\epsilon} \right)dx$$

has a stationary point at $x=0$. The stationary-phase formula gives

$$I(\epsilon) \sim e^{i\pi/4} \sqrt{2\pi\epsilon}.$$

The exact regulated calculation gives the same leading phase and scaling. The factor $e^{i\pi/4}$ is the one-dimensional version of the phase produced by an oscillatory Gaussian determinant.

WKB and Semiclassical Connections

WKB begins with an ansatz such as

$$\psi(x) = A(x) \exp \left[ \frac{i}{\hbar}S(x) \right],$$

then organizes the Schrödinger equation in powers of $\hbar$ relative to the classical action scale. The leading term gives the Hamilton–Jacobi equation, and the next term gives a transport equation for $A(x)$.

The asymptotic small parameter is not $\hbar$ as a dimensionful number. It is a ratio such as

$$\epsilon \sim \frac{\hbar}{S_{\mathrm{cl}}}, \qquad \epsilon\ll1.$$

This is why WKB Approximation works best for slowly varying potentials and high actions, and why Semiclassical Propagator is organized by stationary phase of $e^{iS/\hbar}$.

Exponentially Small Terms

Power-series asymptotics can miss terms that are smaller than every power of the expansion parameter. For $A>0$,

$$e^{-A/\epsilon}$$

satisfies

$$e^{-A/\epsilon} = o(\epsilon^N)$$

for every fixed $N$ as $\epsilon\to0^+$.

Such terms are called exponentially small relative to the power expansion. They are central in tunneling, instantons, level splitting in double wells, and Stokes phenomena. Saying that a perturbative expansion has no such term does not prove the physical effect is absent; it may be nonperturbative in the chosen parameter. Asymptotic Series and Nonperturbative Corrections develops the factorial-growth, optimal-truncation, and tunneling consequences.

Error Discipline

Asymptotic reasoning is useful only when the limiting regime is stated. A mature approximation should say:

• which dimensionless parameter is small or large;
• which order has been retained;
• what size the first neglected term is expected to have;
• where the expansion fails;
• whether the error statement is a rigorous bound or a diagnostic indicator.

The physics-facing version of this discipline is Small Parameters and Error Estimates.

Common Mistakes

• Treating every formal expansion as convergent.
• Writing $O(\epsilon^n)$ without saying which limit is being taken.
• Calling a dimensionful quantity “small” without forming a dimensionless ratio.
• Adding more terms to an asymptotic series after the terms have started growing.
• Applying stationary phase when there is no accessible stationary point or when endpoint contributions dominate.
• Using a nondegenerate saddle formula at a degenerate saddle or caustic.
• Forgetting that exponentially small effects can be invisible to all orders in a power expansion.
• Confusing a physics error indicator with a rigorous error bound.

Cross-Links

• Sequences, Series, and Convergence
• Complex Analysis Essentials
• Small Parameters and Error Estimates
• Semiclassical Limit Overview
• Semiclassical Limit
• Airy Functions
• WKB Approximation
• Semiclassical Propagator
• Asymptotic Series and Nonperturbative Corrections
• Why Path Integrals?
• Large-N and Saddle-Point Methods Preview

References

• C. M. Bender and S. A. Orszag, Advanced Mathematical Methods for Scientists and Engineers, Springer, 1999.
• F. W. J. Olver, Asymptotics and Special Functions, Academic Press, 1974.
• N. Bleistein and R. A. Handelsman, Asymptotic Expansions of Integrals, Dover, 1986.
• M. V. Berry and K. E. Mount, “Semiclassical approximations in wave mechanics,” Reports on Progress in Physics 35, 315-397, 1972.
• M. Brack and R. K. Bhaduri, Semiclassical Physics, Westview Press, 2003.

Exercises

1. Show that $\sin\epsilon=\epsilon+O(\epsilon^3)$ as $\epsilon\to0$.

Solution

The Taylor expansion is

$$\sin\epsilon = \epsilon - \frac{\epsilon^3}{6} + O(\epsilon^5).$$

Therefore

$$\sin\epsilon-\epsilon = O(\epsilon^3).$$

2. Suppose

$$F(\epsilon) \sim a_0+a_1\epsilon+a_2\epsilon^2+\cdots.$$

What does it mean to keep terms through $a_2\epsilon^2$?

Solution

It means using

$$F(\epsilon) = a_0+a_1\epsilon+a_2\epsilon^2 + O(\epsilon^3)$$

as $\epsilon\to0$, provided the expansion is valid in that regime. It does not require the infinite series to converge.

3. Let

$$I(\lambda) = \int_{-\infty}^{\infty} e^{-\lambda x^2/2}\,dx.$$

Find the large-$\lambda$ scaling.

Solution

Rescale $u=\sqrt{\lambda}\,x$. Then $dx=du/\sqrt{\lambda}$ and

$$I(\lambda) = \frac{1}{\sqrt{\lambda}} \int_{-\infty}^{\infty} e^{-u^2/2}\,du = \sqrt{\frac{2\pi}{\lambda}}.$$

Thus $I(\lambda)=O(\lambda^{-1/2})$ as $\lambda\to\infty$.

4. Why can an exponentially small tunneling factor be absent from every term of a power series in $\epsilon$?

Solution

For $A>0$,

$$e^{-A/\epsilon} = o(\epsilon^N)$$

for every fixed $N$ as $\epsilon\to0^+$. It is smaller than all powers of $\epsilon$ in that limit. A power series in $\epsilon$ therefore cannot reveal such a term term-by-term, even though the term may be physically important.