| Feynman–Kac formula | |
|---|---|
| Name | Feynman–Kac formula |
| Introduced | 1940s |
| Field | Mathematical physics, Probability theory |
| Related | Feynman path integral, Schrödinger equation, Stochastic differential equation |
Feynman–Kac formula
The Feynman–Kac formula is a mathematical identity that connects certain linear partial differential equations with expected values of functionals of stochastic processes. It provides a rigorous bridge between the probabilistic description of diffusion processes and the analytic description used in quantum mechanics and statistical mechanics, enabling probabilistic methods to solve evolution equations arising in quantum and thermal contexts.
The Feynman–Kac formula relates parabolic PDEs, typically of Schrödinger or heat type, to expectations over stochastic trajectories such as Brownian motion or solutions of Itô diffusions. In quantum physics it clarifies connections between the operator formalism (e.g., the Hamiltonian and the Schrödinger equation) and the path-integral perspective introduced by Richard Feynman. The formula underpins rigorous constructions of Euclidean field theory via probabilistic methods and serves as a mathematical underpinning for Monte Carlo simulation techniques used in lattice quantum field theory and many-body quantum problems. It is closely associated with work by Mark Kac and provides an analytic counterpart to the Feynman path integral heuristic.
In its simplest form, the Feynman–Kac formula states that the solution u(x,t) to a backward parabolic PDE (L + V)u = ∂_t u, with terminal condition u(x,0)=f(x), can be represented as an expectation u(x,t) = E_x[exp( - ∫_0^t V(X_s) ds ) f(X_t)], where X_t is a diffusion with generator L (often the Laplacian Δ/2 generating Brownian motion). Variants include time-dependent potentials V(t,x), magnetic vector potentials via complex-valued weights (related to the Wiener integral and Feynman–Kleinert methods), and versions for operators on manifolds or graphs. There are also operator-theoretic formulations linking semigroups e^{-tH} for self-adjoint Hamiltonians H to expectations over path measures, and extensions to complex-time (analytic continuation) that approach the real-time Feynman path integral.
The formula interprets solutions of certain evolution equations as averages over random paths weighted by exponential functionals, providing a probabilistic representation of the propagator in imaginary time (Euclidean signature). This connects directly to the Feynman path integral under Wick rotation t → iτ: the Euclidean path integral becomes a mathematically tractable Wiener-type expectation, while the original oscillatory Feynman integral is formal. The representation uses named constructs such as Itô's lemma, the Girsanov theorem for changes of drift, and the Cameron–Martin theorem for shifts in Gaussian measures, each of which plays a role in adapting the path measure to specific potentials or gauge fields.
In quantum mechanics, the Feynman–Kac formula is used to compute ground-state energies, spectral bounds, and heat-kernel traces of Schrödinger operators, which are central to perturbation theory and variational estimates. In statistical physics it yields representations of partition functions and correlation functions for classical and quantum systems in the Euclidean picture. Practical uses include path-integral Monte Carlo methods for bosonic many-body systems, analysis of Anderson localization via random potentials, and rigorous constructions in constructive quantum field theory and stochastic quantization frameworks. Key institutions and works linked to development and application include research at Princeton University, Harvard University, and foundational texts by authors such as Barry Simon and Elliott Lieb.
Proofs typically start from considering the Markov semigroup generated by L and applying Itô calculus to f(X_t) exp(-∫V). Using Itô's formula one shows the martingale property of appropriately weighted functionals; taking expectations yields the PDE solution. Alternative derivations proceed from the Trotter–Kato product formula and operator splitting for e^{-t(H+V)}, or via discrete random-walk approximations that converge to Brownian motion (Donsker's theorem) and Riemann-sum approximations of the action. For magnetic and spin systems, path-integral measure adjustments employ the Stratonovich integral and spin-coherent-state representations.
Extensions generalize the basic formula to non‐linear settings (via Feynman–Kac type representations for McKean–Vlasov equations), jump processes (using Lévy processes and nonlocal generators), and infinite-dimensional situations such as stochastic partial differential equations (SPDEs) and Gaussian measures on function spaces relevant to quantum field theory. On manifolds, the formula uses the Laplace–Beltrami operator and stochastic development; on discrete networks it relates to random walks and electrical network theory. Connections to spectral theory, Kato's inequality, and the theory of semigroups of operators are central to analyzing stability and domain questions.
Concrete examples include the free-particle heat kernel (recovering Gaussian transition densities), the harmonic oscillator where Mehler kernels arise, and the hydrogen-like Coulomb potential where path integral numerics estimate bound states. Computational methods based on the Feynman–Kac representation include importance-sampled Monte Carlo, diffusion Monte Carlo, and stochastic mesh methods; these are implemented in computational physics packages and used in simulations at facilities such as Argonne National Laboratory and CERN for model problems. Analytical approximations use short-time asymptotics, semiclassical (WKB) expansions, and perturbative series linked to Feynman diagrammatics.
Category:Mathematical physics Category:Stochastic processes Category:Quantum mechanics