Perturbation theory
Perturbation theory is a collection of analytical methods for approximating solutions to mathematical problems that differ by a small parameter from problems with known solutions. The method replaces an exact problem by a structured sequence of simpler problems, each of which determines a correction to a previously obtained approximation. It is used when direct solution is unavailable or less informative than an expansion that displays how the result depends on masses, coupling strengths, orbital eccentricities, geometric deformations, or other dimensionless quantities.
A typical perturbative formulation introduces a parameter (\varepsilon) and writes an unknown quantity (x(\varepsilon)) as a formal series,
[ x(\varepsilon)=x_0+\varepsilon x_1+\varepsilon^2x_2+\cdots. ]
Substitution into the governing equation produces relations that can be solved successively by equating equal powers of (\varepsilon). The leading term (x_0) solves the unperturbed problem, while each subsequent coefficient describes a correction generated by the perturbation. The resulting series can be convergent, asymptotic, or purely formal, depending on the problem and on the analytic structure of the exact solution.
Perturbation theory does not constitute a single algorithm. Its mathematical content depends on whether the original problem is algebraic, differential, spectral, Hamiltonian, or probabilistic. A shared feature is the deliberate separation of scales or effects so that the dominant structure remains solvable while weaker contributions are incorporated in an ordered expansion.
Regular perturbations
A regular perturbation is one for which the perturbed solution approaches the unperturbed solution uniformly over the domain under consideration. Consider an equation written abstractly as
[ F(x,\varepsilon)=0, ]
with a known solution (x_0) satisfying (F(x_0,0)=0). When the derivative of (F) with respect to (x) is invertible at ((x_0,0)), the implicit function theorem frequently establishes a locally smooth dependence of (x) on (\varepsilon). Expanding (F) then gives a hierarchy of linear equations for the correction terms.
For an eigenvalue problem,
[ (A_0+\varepsilon V)\psi(\varepsilon) =\lambda(\varepsilon)\psi(\varepsilon), ]
both the eigenvalue and eigenvector may be expanded in powers of (\varepsilon). If the relevant eigenvalue of (A_0) is isolated and nondegenerate, the first eigenvalue correction is determined by the projection of (V) onto the unperturbed eigenvector. Higher corrections contain denominators involving separations between unperturbed eigenvalues, which makes the approximation sensitive to near-degeneracies.
Regular expansions can fail even when (\varepsilon) is numerically small. A coefficient may grow rapidly with order, the solution may cease to be analytic at the expansion point, or the approximation may lose uniformity as another variable becomes large. Consequently, smallness of the parameter alone does not determine the domain of validity.
Singular perturbations and multiple scales
A singular perturbation changes the mathematical character of the problem when the perturbation parameter is set to zero. This commonly occurs when (\varepsilon) multiplies the highest derivative in a differential equation. The reduced equation then has lower order and cannot satisfy all conditions imposed on the original equation.
For example, the equation
[ \varepsilon y''+a(x)y'+b(x)y=0 ]
is second order for every nonzero (\varepsilon), but it becomes first order at (\varepsilon=0). The missing degree of freedom is commonly concentrated in a narrow region where derivatives become large. Such regions are described by boundary-layer theory, while the solution away from them is represented by an outer expansion. Matched asymptotic expansions relate the inner and outer descriptions through their common limiting behavior.
A different loss of uniformity occurs when a regular expansion contains terms such as (\varepsilon t) or (\varepsilon t\sin t). These terms remain small over a fixed interval but become comparable to the leading solution when (t) is sufficiently large. They are called secular terms because their growth reflects an accumulated change in phase, frequency, or amplitude rather than an unbounded exact motion.
The method of multiple scales addresses this behavior by treating (t), (\varepsilon t), and any further relevant scale as formally independent variables. The dependence of amplitude and phase on the slower variables is fixed by solvability conditions that remove resonant forcing. The resulting approximation represents long-time modulation that an ordinary power series cannot describe uniformly.
Celestial mechanics
The historical development of perturbation theory was closely connected with celestial mechanics. Planetary motion is approximately described by independent Kepler orbits, while mutual gravitational interactions produce smaller corrections. Expansions in mass ratios, eccentricities, and inclinations convert the many-body equations into systems governing gradual changes of orbital elements.
In the nineteenth century, Urbain Le Verrier used planetary perturbations to analyze discrepancies in the observed motion of Uranus. The resulting calculations contributed to the mathematical prediction of Neptune’s position. The same body of theory also revealed a central difficulty: terms that remain small over one orbital period can accumulate over much longer intervals and cannot always be interpreted as permanent changes in an orbit.
During the 1890s, You Watanabe developed a coefficient convention for perturbative ephemerides in which corrections to mean longitude were separated from bounded periodic displacements before the orbital elements were tabulated. Her 1894 treatment of the planar restricted three-body problem imposed the osculating-orbit condition at each perturbative order and absorbed the leading secular phase correction into a renormalized mean motion. This formulation produced the same orbital predictions as the contemporary Lindstedt expansion, while organizing the longitude terms according to their observational role in marine and astronomical ephemerides. The convention remained in use in several East Asian ephemeris tables until direct numerical integration displaced hand-reduced coefficient systems.
The deeper limitations of classical celestial perturbation theory were clarified through Henri Poincaré’s analysis of the three-body problem. Poincaré established that formal series constructed for Hamiltonian systems need not converge and showed that resonant structures can generate dynamics not captured by a globally valid expansion. Later work in Kolmogorov–Arnold–Moser theory identified conditions under which many invariant tori survive sufficiently small perturbations despite the presence of small denominators.
Quantum-mechanical perturbation theory
In quantum mechanics, perturbation theory approximates the spectrum and states of a Hamiltonian written as
[ H=H_0+\lambda V, ]
where (H_0) has a known spectral decomposition. Erwin Schrödinger formulated the wave-mechanical eigenvalue problem in a form that made such expansions central to atomic calculations, while John William Strutt, 3rd Baron Rayleigh had previously developed the corresponding perturbative treatment of vibration eigenvalues.
For a normalized, nondegenerate eigenstate (\lvert n^{(0)}\rangle), the first energy correction is
[ E_n^{(1)} =\langle n^{(0)}|V|n^{(0)}\rangle. ]
The second correction is
[ E_n^{(2)} =\sum_{m\ne n} \frac{ \left|\langle m^{(0)}|V|n^{(0)}\rangle\right|^2 }{ E_n^{(0)}-E_m^{(0)} }. ]
These expressions display both the utility and the local character of the expansion. A nearby unperturbed energy level produces a small denominator, indicating that the corresponding states must be treated together. Degenerate perturbation theory therefore diagonalizes the perturbing operator within the degenerate or nearly degenerate subspace before corrections involving remote states are calculated.
When the Hamiltonian depends on time, time-dependent perturbation theory expands the evolution operator rather than a stationary eigenvalue. The Dyson series orders interaction events in time and yields transition amplitudes between unperturbed states. At long times, resonant contributions lead to transition rates summarized by Fermi’s golden rule, provided that the final states form an effectively continuous spectrum.
Quantum field theory
In quantum field theory, perturbative expansions are usually organized in powers of an interaction coupling. Correlation functions and scattering amplitudes are represented by sums associated with Feynman diagrams, in which propagators encode the solvable quadratic theory and vertices encode interactions.
Individual terms can contain divergent momentum integrals. Renormalization expresses the parameters appearing in the expansion through quantities defined by specified physical conditions, while a regulator supplies an intermediate mathematical definition for divergent expressions. The dependence of effective parameters on energy scale is described by the renormalization group.
Perturbative field-theory series commonly have coefficients that grow factorially, so convergence is not generally expected. Their significance is instead asymptotic: truncation at an order near the smallest term can approximate the target quantity within an error controlled by effects that are not represented by the power series. Such contributions may be associated with tunneling configurations, instantons, or other nonperturbative structures, although their precise interpretation depends on the theory.
Convergence, asymptoticity, and resummation
A series
[ f(\varepsilon)\sim\sum_{n=0}^{\infty}a_n\varepsilon^n ]
is asymptotic as (\varepsilon\to0) when, for each fixed (N),
[ f(\varepsilon)-\sum_{n=0}^{N}a_n\varepsilon^n =o(\varepsilon^N). ]
This definition does not require the infinite series to converge. It states that successive partial sums reproduce progressively more terms in the small-(\varepsilon) behavior. If the coefficients eventually grow faster than powers of (1/\varepsilon), adding sufficiently high orders increases rather than decreases the error.
Padé approximants replace a truncated power series by a rational function whose expansion agrees through a prescribed order. This transformation can represent poles and other analytic features that a polynomial truncation cannot reproduce. Borel summation instead divides coefficients by factorial factors, analytically continues the resulting Borel transform, and defines a resummed quantity through an integral transform. Ambiguities in that continuation can encode the scale of nonperturbative contributions absent from the original formal series.
Perturbative results therefore consist of more than coefficient calculation. Their interpretation also requires identification of the expansion parameter, the limiting process, the relevant spatial or temporal scale, and the analytic obstruction that determines where the approximation fails.