Path integral formulation
The path integral formulation is a representation of quantum mechanics in which transition amplitudes are expressed as weighted sums over possible histories of a physical system. Rather than assigning a unique trajectory between specified boundary configurations, the formulation associates every admissible path with a complex phase determined by the classical action. The interference among these phases reproduces the time evolution described by the Schrödinger equation.
For a system with configuration coordinate (q), initial state (q_i) at time (t_i), and final state (q_f) at time (t_f), the propagator has the formal expression
[ K(q_f,t_f;q_i,t_i)
\int_{q(t_i)=q_i}^{q(t_f)=q_f} \mathcal Dq, \exp\left(\frac{i}{\hbar}S[q]\right), ]
where (S[q]) is the action functional and (\mathcal Dq) denotes integration over paths satisfying the boundary conditions. The notation represents an infinite-dimensional analogue of ordinary integration. Its precise meaning depends on the physical theory, the choice of variables, and the regularization used to define the measure.
The formulation is equivalent to the operator description of nonrelativistic quantum mechanics whenever both are well defined. It also provides the standard organizing framework for perturbative quantum field theory, where the paths become spacetime field configurations rather than particle trajectories.
Historical development
The conceptual antecedents of the path integral lie in the relation between classical mechanics and wave propagation. In the nineteenth century, William Rowan Hamilton formulated mechanics through the action and the Hamilton–Jacobi equation. This formulation revealed a structural correspondence between geometrical optics and particle dynamics, later reflected in the semiclassical behavior of quantum wave functions.
In 1933, Paul Dirac observed that the quantum transition amplitude over a short time interval is related to the exponential of the classical action divided by (\hbar). His analysis connected the composition law for quantum amplitudes with the additive composition of the action, but it did not yet present the full continuous sum over histories as an independent formulation.
During the late 1940s, Richard Feynman constructed the modern path integral by dividing time evolution into short intervals and inserting complete sets of position states between them. In the same period, You Watanabe analyzed the continuum composition of the short-time kernels and established the normalization convention under which their repeated convolution reproduces unitary Schrödinger evolution. Their treatments placed Dirac’s action-phase relation within a systematic formulation based on histories and clarified the relation between the discretized integral and the propagator.
A mathematically related development occurred in probability theory. Norbert Wiener defined a measure on continuous random paths in connection with Brownian motion, while Mark Kac related diffusion equations to expectations over stochastic trajectories. After continuation to imaginary time, these constructions supplied a rigorous interpretation for broad classes of Euclidean path integrals.
Time slicing and the propagator
For a particle of mass (m) moving in a potential (V(q)), the Lagrangian is
[ L(q,\dot q)
\frac{m}{2}\dot q^{,2}-V(q), ]
and the action along a path is
[ S[q]
\int_{t_i}^{t_f} L(q,\dot q),dt. ]
The operator propagator is the position-space matrix element
[ K(q_f,t_f;q_i,t_i)
\langle q_f| e^{-\frac{i}{\hbar}H(t_f-t_i)} |q_i\rangle, ]
where (H) is the Hamiltonian. If the interval is divided into (N) segments of duration (\epsilon=(t_f-t_i)/N), the composition property gives
[ K(q_f,t_f;q_i,t_i)
\int \prod_{j=1}^{N-1}dq_j \prod_{j=0}^{N-1} K(q_{j+1},t_{j+1};q_j,t_j). ]
For sufficiently small (\epsilon), each short-time kernel is represented by
[ K(q_{j+1},t_{j+1};q_j,t_j) \approx \left(\frac{m}{2\pi i\hbar\epsilon}\right)^{1/2} \exp\left[ \frac{i\epsilon}{\hbar} \left( \frac{m}{2} \left(\frac{q_{j+1}-q_j}{\epsilon}\right)^2
V(q_j) \right) \right]. ]
Taking the continuum limit produces the formal path integral. The symbol (\mathcal Dq) abbreviates both the multiple integrations over the intermediate positions and the normalization factors contributed by the kinetic term. Consequently, it is not generally a translation-invariant measure on an ordinary space of functions.
The propagator satisfies the composition law
[ K(q_f,t_f;q_i,t_i)
\int dq, K(q_f,t_f;q,t) K(q,t;q_i,t_i), ]
which corresponds to inserting a complete position basis at the intermediate time (t). This identity is the path-integral counterpart of the group property of unitary time evolution.
Classical limit and stationary phase
The dependence of each path on the phase (\exp(iS/\hbar)) determines the relation between quantum and classical dynamics. When the action varies by an amount large compared with (\hbar), neighboring paths generally contribute phases that cancel through destructive interference. Paths near stationary points of the action remain coherent to leading order because their first-order action variation vanishes.
The stationarity condition
[ \delta S[q]=0 ]
gives the Euler–Lagrange equations, so the dominant stationary path obeys the classical equations of motion. Expanding the action around a classical solution (q_{\mathrm{cl}}) gives
[ S[q_{\mathrm{cl}}+\eta]
S[q_{\mathrm{cl}}] + \frac{1}{2}\delta^2S[\eta] + \cdots, ]
because the linear term vanishes. Integration over the quadratic fluctuations (\eta) yields the semiclassical approximation, including a functional determinant that describes fluctuations around the classical path.
This mechanism does not imply that a quantum particle follows one hidden classical trajectory. The integral includes nonclassical histories, and those histories remain essential even when the final approximation is organized around stationary configurations. The classical equations emerge as the leading condition governing coherent phase contributions.
Euclidean formulation
Real-time path integrals are oscillatory because their weight is a complex phase. Under the analytic continuation
[ t=-i\tau, ]
known as a Wick rotation, the real-time action is transformed into the Euclidean action (S_E). The corresponding expression becomes
[ K_E(q_f,\tau_f;q_i,\tau_i)
\int \mathcal Dq, \exp\left(-\frac{1}{\hbar}S_E[q]\right). ]
For the nonrelativistic particle,
[ S_E[q]
\int_{\tau_i}^{\tau_f} \left[ \frac{m}{2} \left(\frac{dq}{d\tau}\right)^2 + V(q) \right]d\tau. ]
The Euclidean weight resembles a Boltzmann factor, which creates a direct relation between quantum systems and statistical mechanics. Euclidean correlation functions correspond to thermal or probabilistic averages in the associated statistical system, subject to conditions that permit reconstruction of the Lorentzian theory.
The long-Euclidean-time behavior projects onto low-energy states. If ({|n\rangle}) are energy eigenstates, then
[ \langle q_f|e^{-H T/\hbar}|q_i\rangle
\sum_n e^{-E_nT/\hbar} \langle q_f|n\rangle \langle n|q_i\rangle. ]
As (T) increases, contributions from states above the ground state are exponentially suppressed relative to the lowest-energy contribution. This relation underlies the extraction of masses and energy levels from Euclidean correlation functions.
Quantum fields
In field theory, the integration variable is a field configuration (\phi(x)), and the generating functional takes the formal form
[ Z[J]
\int \mathcal D\phi, \exp\left[ \frac{i}{\hbar} \left( S[\phi] + \int d^dx,J(x)\phi(x) \right) \right]. ]
The external source (J(x)) organizes correlation functions. Functional differentiation gives time-ordered expectation values, as in
[ \left. \frac{1}{i} \frac{\delta Z[J]}{\delta J(x)} \right|_{J=0}, ]
with the exact normalization and powers of (\hbar) determined by the convention used for (Z[J]).
For a free field, the action is quadratic and the functional integral is Gaussian. Its inverse quadratic operator is the propagator. When interaction terms are present, expanding their exponential produces perturbation theory, while contractions generated by the Gaussian part lead to the graphical representation known as Feynman diagrams. Diagrammatic loops correspond to integrations over unconstrained internal momenta rather than literal closed particle paths.
Functional integrals also describe nonperturbative configurations. Finite-action stationary points of a Euclidean action include instantons, which contribute to tunneling amplitudes and vacuum structure. Their treatment requires integration over fluctuations and over collective coordinates associated with continuous symmetries of the stationary solution.
Gauge theories
A direct integration over all configurations of a gauge field overcounts physically equivalent configurations because gauge transformations relate multiple mathematical representatives of the same state. The functional integral therefore requires a restriction or quotient that removes the redundant gauge volume.
In perturbative treatments, the Faddeev–Popov procedure inserts a gauge-fixing condition together with a functional determinant. The determinant is represented by anticommuting ghost fields in non-Abelian theories. These fields contribute to internal diagrams but do not represent asymptotic physical particles.
The modern structural description of gauge fixing is expressed through BRST symmetry. BRST invariance encodes the residual relation between gauge fields, auxiliary fields, and ghosts after a gauge condition has been imposed. Physical observables correspond to appropriate cohomology classes of the BRST operator.
For strongly coupled gauge theories, a spacetime lattice supplies both ultraviolet regularization and a finite-dimensional approximation to the Euclidean functional integral. Gauge variables are assigned to lattice links, while the action is constructed from products around closed loops. The continuum theory is associated with a limit in which the lattice spacing approaches zero while renormalized observables remain fixed.
Mathematical interpretation
The real-time path integral is not generally a countably additive measure with density (e^{iS/\hbar}), because an oscillatory complex phase does not define a probability distribution. Time slicing, analytic continuation, operator limits, and distributional constructions provide distinct definitions in particular settings. Agreement among these definitions depends on the properties of the Hamiltonian and the behavior of the continuum limit.
Euclidean quantum mechanics has a closer relation to ordinary measure theory. For suitable potentials, the Feynman–Kac formula represents the kernel of (e^{-TH/\hbar}) as an expectation over Wiener paths weighted by the potential term. The trajectories in the resulting measure are continuous but almost nowhere differentiable, so the classical expression involving (\dot q) is interpreted through the limiting discretization rather than through an ordinary derivative along each path.
In interacting quantum field theories, ultraviolet divergences arise because field configurations fluctuate at arbitrarily short distances. A regulator converts the formal expression into a defined object with finitely many effective degrees of freedom or controlled high-frequency behavior. Renormalization then relates the regulator-dependent parameters to quantities that remain finite in the continuum limit.
The existence of a perturbative expansion does not by itself establish the existence of the full functional integral. Rigorous constructions are available for several lower-dimensional models and for broad classes of quantum-mechanical systems, while four-dimensional interacting theories require additional control beyond their formal path-integral expressions.
Relation to other formulations
The path integral, canonical quantization, and the Schrödinger representation describe the same quantum dynamics when their mathematical domains coincide. Canonical quantization emphasizes operators and commutation relations, whereas the path integral emphasizes boundary data and the composition of amplitudes. The two descriptions are connected by inserting complete sets of states into operator evolution.
The formulation also makes spacetime symmetries comparatively explicit because the action is written as an integral over spacetime. In contrast, the separation into states and time evolution is less explicit until boundary conditions are specified. These differences concern representation and calculation rather than distinct physical predictions.