Volume Overview

Exact solutions are indispensable reference points, but they are exceptional. Most useful quantum predictions come from identifying a solvable structure, a limiting regime, a bound, or an asymptotic observable and then controlling what has been neglected.

This chapter is the routing layer for that process. It does not replace the method pages. It explains what each family of methods promises, what controls it, what can invalidate it, and what evidence turns a formal calculation into a trustworthy approximation.

The central discipline is:

$$\text{question} \longrightarrow \text{observable} \longrightarrow \text{dimensionless regime} \longrightarrow \text{method} \longrightarrow \text{independent checks}.$$

What Counts as a Controlled Approximation?

An approximation is controlled when there is a stated reason that the omitted contribution is smaller than the retained one in the regime of interest. That reason need not always be a convergent power series.

Kind of control	Typical statement	What the result provides
Perturbative hierarchy	$epsilon\ll1$ and successive terms scale as powers of $epsilon$	An order-by-order estimate
Spectral separation	Couplings are small compared with relevant gaps	Weak mixing or a reduced Hamiltonian
Variational inequality	The trial space is a subset of the admissible domain	A one-sided energy bound
Asymptotic regime	$S/\hbar\gg1$, $ka\ll1$, or another scaled variable is extreme	A leading limiting form, often with an asymptotic series
Conservation law	Unitarity, symmetry, or flux conservation constrains the answer	Exact checks on an approximate calculation
Numerical comparison	Results stabilize under a controlled refinement	Empirical error evidence for a specified computation

These controls have different logical force. A rigorous bound is stronger than a small next term; a small next term is stronger than visual agreement with one plot. The Small Parameters and Error Estimates page distinguishes bounds, estimates, and diagnostics in detail.

Three labels should not be confused:

• Exact means exact within explicitly stated model assumptions.
• Approximate means some contribution has been omitted and its size must be assessed.
• Asymptotic means the error has a limiting hierarchy for fixed truncation order; the infinite series need not converge.

A result can be both exact and model-dependent. It can also be asymptotically controlled without being a convergent approximation for fixed parameter value.

The Method Families

The volume organizes methods by the structure that makes a calculation possible.

Problem structure	Natural method	Control variable or theorem	Primary output	First warning sign
Solvable $H_0$ plus weak static $V$	Time-independent perturbation theory	Matrix elements divided by spectral gaps	Energies and eigenstates	Small or vanishing denominator
Coupled degenerate subspace	Degenerate perturbation theory	Separation from states outside the chosen subspace	Level splitting and adapted basis	Important states omitted from the subspace
Weak time-dependent drive	Time-dependent perturbation theory	Drive strength, detuning, and duration	Transition amplitude or probability	Secular growth or strong population transfer
Dense continuum of final states	Fermi’s golden rule	Weak coupling and an intermediate-time window	Transition rate	Recurrences, threshold structure, or narrow bandwidth
Unknown low-lying state	Variational or Rayleigh–Ritz method	Rayleigh quotient and min–max structure	Upper bound or Ritz spectrum	Trial family misses symmetry or length scales
Slowly varying local wavelength	WKB or semiclassics	$\hbar/S$ or local adiabaticity	Phase, spectrum, or tunneling exponent	Turning point, caustic, or interference catastrophe
Incoming and outgoing asymptotic flux	Scattering theory	Boundary conditions and unitarity	Amplitude, cross section, or phase shift	Incorrect normalization or long-range asymptotics
Weak scattering potential	Born series	Small repeated-scattering corrections	Approximate scattering amplitude	Resonance or strong phase shift
Central potential	Partial-wave expansion	Angular momentum decomposition	Phase shifts and partial cross sections	Too few partial waves or inelastic channels
Separated sectors or frequencies	Effective Hamiltonian	Coupling-to-gap or inverse-frequency ratio	Reduced dynamics	Leakage, resonance, or uncontrolled truncation
Exponentially small tunneling	WKB or instanton method	Euclidean action divided by $\hbar$	Exponent and, with more work, prefactor	Wrong saddle or omitted fluctuation mode

The Approximation Map gives the conceptual taxonomy. Choosing an Approximation Method turns it into a decision procedure.

Perturbative, Variational, and Asymptotic Thinking compares the logical claims made by the main approximation cultures and explains how scattering and effective descriptions fit among them.

Common Failure Modes is the reusable diagnostic for recognizing when those controls have broken and choosing a better local method.

Notation and Conventions fixes the symbols, normalizations, boundary prescriptions, and translation rules used throughout the volume.

Perturbative Reasoning

For a static problem, write

$$H(\lambda)=H_0+\lambda V,$$

and seek an eigenpair as a formal expansion,

$$\begin{aligned} E_n(\lambda) &= E_n^{(0)} +\lambda E_n^{(1)} +\lambda^2E_n^{(2)}+\cdots, \\ \lvert n(\lambda)\rangle &= \lvert n^{(0)}\rangle +\lambda\lvert n^{(1)}\rangle +\lambda^2\lvert n^{(2)}\rangle+\cdots. \end{aligned}$$

The bookkeeping symbol $\lambda$ is not itself proof of smallness. For nondegenerate state mixing, the physically relevant quantities include

$$\epsilon_{mn} = \frac{\lvert\lambda V_{mn}\rvert} {\lvert E_n^{(0)}-E_m^{(0)}\rvert}, \qquad m\ne n.$$

If any materially coupled state has $\epsilon_{mn}\not\ll1$, the ordinary nondegenerate expansion is not controlled for that state. Exact degeneracy makes the denominator vanish; near degeneracy can be just as consequential. The remedy is to identify the relevant subspace and diagonalize the perturbation or an effective Hamiltonian there.

Start with Nondegenerate Perturbation Theory and Degenerate Perturbation Theory. The Anharmonic Oscillator is the standard calibration model.

Time-Dependent Transitions

For a weak drive $V(t)$, the first-order transition amplitude from an unperturbed state $\lvert i\rangle$ to $\lvert f\rangle$ is

$$c_f^{(1)}(t) = -\frac{i}{\hbar} \int_{t_0}^{t} dt'\, e^{i\omega_{fi}t'}V_{fi}(t'), \qquad \omega_{fi} = \frac{E_f-E_i}{\hbar}.$$

This formula already displays the three controls: the matrix element sets the coupling strength, the oscillatory phase sets resonance and cancellation, and the integration interval sets how long perturbation theory has had to accumulate amplitude.

For a sufficiently smooth continuum of final states and an appropriate time window, the transition probability can become approximately linear in time. The resulting rate is

$$\Gamma_{i\to f} = \frac{2\pi}{\hbar} \lvert V_{fi}\rvert^2 \rho(E_f),$$

with the energy-conservation condition understood. This is not a universal long-time identity. It assumes weak depletion of the initial state, a continuum dense enough to suppress visible recurrences, and times long enough to resolve energies but short enough that the perturbative description remains valid.

The chapter map and time-scale hierarchy are in Time-Dependent Perturbation Theory and Transitions. Continue from there to First-Order Transition Probability and Fermi’s Golden Rule.

Variational Reasoning

For a self-adjoint Hamiltonian bounded below, a normalized admissible trial state satisfies

$$E_0 \le \mathcal R[\psi] \equiv \frac{\langle\psi\rvert H\lvert\psi\rangle} {\langle\psi\vert\psi\rangle}.$$

The guarantee concerns the energy. It does not imply that every local observable, correlation function, or tail of the optimized trial state is accurate. A useful variational calculation therefore reports:

1. why the trial family lies in the operator or quadratic-form domain;
2. which exact symmetries the family respects;
3. which length scales and asymptotic behaviors it can represent;
4. how the estimate changes when the trial family is enlarged;
5. whether independent observables agree with known limits or benchmarks.

The Variational Principle establishes the bound. Rayleigh–Ritz Method turns it into a finite-dimensional generalized eigenvalue problem.

Asymptotic and Semiclassical Reasoning

Semiclassical methods exploit a large phase rather than a weak potential. In one dimension, define

$$p(x)=\sqrt{2m\bigl(E-V(x)\bigr)}.$$

Away from turning points, the leading WKB form in a classically allowed region is

$$\psi_{\pm}(x) \approx \frac{C_{\pm}}{\sqrt{p(x)}} \exp\left[ \pm\frac{i}{\hbar} \int^x p(x')\,dx' \right].$$

A local diagnostic is

$$\eta(x) = \frac{\hbar\lvert p'(x)\rvert} {\lvert p(x)\rvert^2} \ll1.$$

The condition fails where $p(x)=0$. A turning point is therefore not a place to substitute blindly into a divergent WKB amplitude; it requires a local uniform approximation and a connection formula. In higher dimensions the corresponding obstructions include caustics and changes in saddle structure.

An asymptotic series can be useful even when it diverges. For fixed truncation order $N$ and small $\epsilon$, the defining statement is of the form

$$F(\epsilon) - \sum_{n=0}^{N}a_n\epsilon^n = o(\epsilon^N), \qquad \epsilon\to0,$$

with the precise remainder convention stated. Adding terms beyond the smallest term can make a divergent asymptotic approximation worse. Independent comparison at the edge of the intended regime is therefore part of responsible use.

Begin with WKB Approximation, then Turning Points and Connection Formulas and Bohr–Sommerfeld Quantization.

Scattering as an Observable Framework

Scattering theory is more than an approximation scheme. It defines observables by asymptotic preparation and detection. For a short-range potential in three dimensions, one common convention is

$$\psi^{(+)}_{\mathbf k}(\mathbf r) \underset{r\to\infty}{\sim} e^{i\mathbf k\cdot\mathbf r} + f(\Omega) \frac{e^{ikr}}{r}.$$

With incident plane-wave normalization and elastic single-channel kinematics,

$$\frac{d\sigma}{d\Omega} = \lvert f(\Omega)\rvert^2.$$

Every symbol here carries a convention: state normalization, incident flux, outgoing boundary condition, and amplitude normalization. Long-range Coulomb interactions require modified asymptotics, while inelastic or multichannel problems require flux and channel factors.

Approximation enters after the observable has been defined. A weak potential suggests the Born series; a central potential suggests partial waves; a low-energy short-range problem suggests scattering length and effective-range parameters. Unitarity and the optical theorem then supply nontrivial checks.

The route is Scattering Amplitude $\to$ Differential and Total Cross Sections $\to$ Lippmann–Schwinger Equation $\to$ First Born Approximation. For central potentials, continue with Partial-Wave Expansion, Phase Shifts, and the Optical Theorem.

Effective Reasoning and Scale Separation

Let $P$ project onto the sector whose dynamics or spectrum is wanted, and let $Q=I-P$. If the two sectors are separated by a characteristic gap $\Delta$ and coupled by a block $PHQ$, a first diagnostic is

$$\epsilon_{\mathrm{mix}} \sim \frac{\lVert PHQ\rVert}{\Delta}.$$

When this ratio is small, a unitary block-diagonalization or projection method can encode virtual excursions into $Q\mathcal H$ as corrections to an effective Hamiltonian on $P\mathcal H$. The effective description should reproduce specified low-energy eigenvalues, matrix elements, or time evolution to a stated order.

The approximation can fail at resonance, when the eliminated sector is populated appreciably, or when the observable itself probes the eliminated degrees of freedom. An effective Hamiltonian is therefore tied to a scale window and a matching prescription; it is not a universally equivalent smaller matrix.

The chapter map is Effective Hamiltonians and Scale Separation. Continue there to Projection Methods, Schrieffer–Wolff Transformation, Rotating-Wave Approximation, and Magnus Expansion.

One Model, Several Legitimate Questions

Consider the quartic oscillator

$$H = \frac{p^2}{2m} + \frac{1}{2}m\omega^2x^2 + g x^4, \qquad g\gt0.$$

The natural oscillator length is

$$a=\sqrt{\frac{\hbar}{m\omega}},$$

so the dimensionless perturbative parameter is

$$\epsilon = \frac{g a^4}{\hbar\omega} = \frac{g\hbar}{m^2\omega^3}.$$

For $\epsilon\ll1$, first-order perturbation theory gives

$$E_n^{(1)} = \frac{3g}{4} \left(\frac{\hbar}{m\omega}\right)^2 \bigl(2n^2+2n+1\bigr).$$

That is a local statement around the harmonic limit. It is not a uniform approximation for arbitrarily large $n$, because the quartic energy grows faster with excitation than the unperturbed level spacing.

For the ground state, a normalized Gaussian trial function

$$\psi_b(x) = \frac{1}{(\pi b^2)^{1/4}} \exp\left(-\frac{x^2}{2b^2}\right)$$

gives the variational energy

$$E(b) = \frac{\hbar^2}{4mb^2} + \frac{m\omega^2b^2}{4} + \frac{3gb^4}{4}.$$

Minimizing over $b\gt0$ gives an upper bound for every positive $g$, not merely for weak coupling. The trial family may still miss non-Gaussian details of the exact state.

At high excitation, semiclassical quantization instead uses the classical turning points $x_-$ and $x_+$:

$$\int_{x_-}^{x_+} dx\, \sqrt{2m\bigl(E-V(x)\bigr)} = \left(n+\frac{1}{2}\right)\pi\hbar.$$

The three answers address different regimes and carry different guarantees. Asking which one is “best” without specifying the state, coupling, and observable is not a well-posed comparison.

A Reusable Workflow

For any approximation problem, record the following before doing substantial algebra.

1. Define the model and observable

State the Hilbert space, Hamiltonian, boundary conditions, initial state if relevant, and the quantity to be predicted. An energy correction, transition rate, and cross section are different targets even when they involve the same interaction.

2. Expose the scales

Choose characteristic units and nondimensionalize. List couplings, gaps, lengths, momenta, action scales, drive frequencies, and observation times. A dimensional coefficient is never a complete control parameter.

3. Use exact structure first

Apply symmetry, selection rules, conservation laws, positivity, analyticity, and known limiting cases before expanding. Exact zeros and block structure should not be rediscovered through pages of approximate algebra.

4. Match the method to the regime

Identify degeneracies, continua, turning points, thresholds, resonances, and separated sectors. These features often decide the method more strongly than the nominal size of a coupling.

5. State the truncation

Name the retained order and the omitted terms. For a composite expansion, state the order in every parameter; for example, leading order in coupling but next-to-leading order in $1/\Omega$.

6. Validate independently

Use at least two checks when practical:

• dimensions and normalization;
• exact symmetry or conservation law;
• solvable limit;
• unitarity or flux conservation;
• next-order estimate;
• basis, grid, or time-step refinement;
• comparison with direct numerical solution;
• stability under changing a trial family or matching point.

7. Report the domain of trust

Conclude with a sentence of the form: “This result is expected to be accurate through order $\epsilon^p$ for states below the specified threshold, away from the stated resonance.” A bare formula is not a complete approximation result.

Failure Diagnostics

Symptom	Likely cause	First response
State correction is comparable to the unperturbed state	Small denominator or strong mixing	Enlarge and diagonalize the relevant subspace
Transition probability grows beyond unity	Secular perturbative term used too long	Resum, solve the reduced dynamics, or shorten the time window
Variational energy is stable but observables are not	Trial family fits energy but misses local structure	Enlarge the ansatz and test sensitive observables
WKB amplitude diverges	Turning point or caustic	Use connection formulas or a uniform approximation
Born cross section violates unitarity badly	Multiple scattering is not negligible	Use partial waves, integral equations, or numerical scattering
Low-energy cross section changes sharply with a weak parameter change	Near-threshold bound state or resonance	Use scattering length, effective range, or pole analysis
Effective model leaks population	Eliminated sector is resonant or insufficiently separated	Retain more states or revise the scale separation
Higher asymptotic terms increase	Expansion truncated past optimal order	Stop near the least term and estimate the remainder

A failed approximation is often informative. It may reveal a hidden degeneracy, resonance, threshold, new scale, or nonperturbative contribution. The correct response is to diagnose the structure, not to force the original formula to produce a number.

Reading Paths

First advanced-undergraduate pass

1. Approximation Map
2. Choosing an Approximation Method
3. Small Parameters and Error Estimates
4. Nondegenerate Perturbation Theory
5. Variational Principle
6. WKB Approximation
7. Scattering Amplitude

Graduate methods path

Add degenerate perturbation theory, transition probabilities, Fermi’s golden rule, connection formulas, Lippmann–Schwinger theory, Born approximation, partial waves, phase shifts, the optical theorem, and effective Hamiltonians.

QFT preparation path

Prioritize interaction-picture transition amplitudes, continuum normalization, the $S$- and $T$-matrix conventions, unitarity, the optical theorem, stationary-phase reasoning, effective descriptions, and nonperturbative saddles. These pages establish the quantum-mechanical logic; relativistic normalization and field-theoretic amplitudes belong to the QFT bridge.

AMO and quantum-matter path

Prioritize degenerate and time-dependent perturbation theory, selection rules, golden-rule rates, rotating-wave and Magnus approximations, Schrieffer–Wolff transformations, low-energy scattering, and resonance diagnostics. Detailed atomic, molecular, optical, and condensed-matter applications remain in their canonical volumes.

Scope and Canonical Homes

This volume owns the calculational methods and their validity conditions. It does not duplicate:

• exact canonical model solutions, owned by Wave Mechanics and Model Systems;
• general time evolution, pictures, propagators, and Green functions, owned by Quantum Dynamics;
• angular momentum algebra, symmetry, and geometric phases, owned by Symmetry, Angular Momentum, and Spin;
• molecular Born–Oppenheimer applications, many-body Hartree–Fock theory, open-system master equations, or relativistic scattering;
• functional analysis and rigorous operator-theoretic scattering, which belong to Mathematical and Rigorous Quantum Mechanics.

Method pages may introduce enough of an external application to make the approximation concrete. Full derivations should remain at their canonical homes and be cross-linked.

Common Mistakes

• Calling a dimensional coefficient small without comparing it with a physical scale.
• Treating a formal series as evidence of convergence.
• Ignoring degeneracy, resonance, threshold behavior, or a turning point.
• Comparing methods without holding the observable and regime fixed.
• Quoting a variational energy as though it guaranteed a uniformly accurate wavefunction.
• Turning a transition probability into a rate without establishing an intermediate-time continuum regime.
• Using a short-range scattering asymptotic form for an unscreened Coulomb potential.
• Reporting agreement at one parameter value without a convergence or limiting-case study.
• Omitting state, flux, or continuum-normalization conventions.
• Hiding a failed validity test because the final number looks plausible.

Exercises

Classify the control

For each statement, identify the kind of control and one limitation: (a) $E_0\le\mathcal R[\psi]$; (b) $\lvert V_{mn}\rvert/\lvert E_n-E_m\rvert\ll1$; (c) $\hbar\lvert p'\rvert/\lvert p\rvert^2\ll1$; (d) a computed scattering amplitude satisfies the optical theorem through the retained order.

Solution

(a) is a variational upper bound for an admissible trial state and a Hamiltonian bounded below. It bounds the energy, not every observable. (b) is a perturbative mixing diagnostic; it fails near degeneracy and must be checked for all materially coupled states. (c) is a local WKB diagnostic; it necessarily fails at ordinary turning points. (d) is a unitarity consistency check. Passing it does not by itself show that all omitted contributions are small, but failing it badly is strong evidence that the approximation or convention is wrong.

Choose the method

A weak periodic drive couples a discrete initial state to a dense continuum. The drive has acted long enough to resolve the continuum but the initial population is still nearly one. Which method is natural, and which assumptions must be monitored?

Solution

Time-dependent perturbation theory followed by Fermi’s golden rule is natural. One must monitor weak coupling, negligible depletion, a continuum dense and smooth on the finite-time energy-resolution scale, and an observation time shorter than recurrence or saturation times. A sharp threshold, resonance, or structured density of states can invalidate the simple constant-rate description.

Compare two quartic-oscillator estimates

Explain why first-order perturbation theory and a Gaussian variational calculation for the quartic oscillator make logically different claims even when their numerical energies agree at weak coupling.

Solution

First-order perturbation theory gives the coefficient of the energy expansion about $g=0$ and expects omitted corrections of higher order in the dimensionless coupling. The Gaussian variational calculation gives an upper bound after optimization for every $g\gt0$ for which the trial state is admissible. It does not reproduce every perturbative coefficient and does not guarantee an accurate wavefunction. Agreement at weak coupling is a useful cross-check, but the control mechanisms remain different.

Diagnose a failed Born approximation

A first Born calculation for short-range elastic scattering predicts a large low-energy cross section and violates the partial-wave unitarity scale. What physical structure should be investigated before adding more Born terms?

Solution

Investigate whether the potential supports a bound or virtual state near threshold, producing a large scattering length or a resonance. In that regime repeated scattering is essential, and a low-energy effective-range or nonperturbative partial-wave treatment is more informative than blindly adding a few Born terms.

Cross-Links

• Volume Landing Page
• Approximation Map
• Choosing an Approximation Method
• Small Parameters and Error Estimates
• Perturbative, Variational, and Asymptotic Thinking
• Common Failure Modes
• Notation and Conventions
• Time-Dependent Perturbation Theory and Transitions
• Perturbation Theory Benchmarks
• WKB Versus Exact Spectrum Notebook
• Phase Shift Extraction Notebook
• Formula Sheet
• Method Comparison Table
• Approximation Decision Tree

References

• B. Zwiebach, Quantum Physics III, MIT OpenCourseWare 8.06, 2018. Chapters 1, 3, 4, and 7 cover perturbation theory, semiclassics, transitions, and scattering.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020.
• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994.
• A. Messiah, Quantum Mechanics, Dover, 1999.
• C. Cohen-Tannoudji, B. Diu, and F. Laloë, Quantum Mechanics, Vols. 1–2, Wiley, 1977.
• J. R. Taylor, Scattering Theory: The Quantum Theory of Nonrelativistic Collisions, Dover, 2006.
• M. V. Berry and K. E. Mount, “Semiclassical approximations in wave mechanics”, Reports on Progress in Physics 35, 315–397, 1972.
• F. W. J. Olver and R. Wong, DLMF Chapter 2: Asymptotic Approximations, NIST Digital Library of Mathematical Functions.
• T. Kato, Perturbation Theory for Linear Operators, 2nd ed., Springer, 1976.