---
title: Feynman–Kac Formula & Regime-Switching Diffusions
url: https://www.emergentmind.com/topics/feynman-kac-formula-63ee39d8-d061-4ea9-acc6-d5bbcbabf17c
type: topic
---

# Feynman–Kac Formula & Regime-Switching Diffusions

The Feynman–Kac formula establishes a fundamental correspondence between linear parabolic (and certain partial integro-differential) equations and expectations of functionals of stochastic processes. In the context of regime-switching jump-diffusion processes—where both continuous Itô dynamics, Lévy jump events, and a finite-state Markov chain-driven regime-switching act jointly—the formula provides a probabilistic representation of solutions to a broad class of partial integro-differential equations (PIDEs) with terminal, initial, and boundary conditions. The regime-switching mechanism couples discrete Markovian environments with stochastic dynamics and jump events, admitting both analytical tractability and flexible modeling of complex evolution phenomena across finance, engineering, and applied mathematics [1702.01495].

## 1. Regime-Switching Jump-Diffusions: Model Structure

A regime-switching jump-diffusion consists of a process $(X_t, \alpha_t)$, where $X_t$ is $\mathbb{R}^n$-valued and $\alpha_t$ is a finite-state Markov chain on $M = \{1,\ldots, m\}$. The continuous dynamics of $X_t$ are governed by coefficients $b$, $\sigma$, and $y$, which depend jointly on the current state $X_t$ and regime $\alpha_t$. Jumps occur according to a Poisson random measure $N(dt, dz)$ with Lévy compensator $\nu(dz)\,dt$, respecting integrability conditions on $|z|^2$ with respect to $\nu$.

The SDE reads:
\[
dX(t) = b(X(t),\alpha(t))\,dt +
         \sigma(X(t),\alpha(t))\,dW(t) +
         \int_{\mathbb{R}^n \setminus \{0\}} y(X(t^-),\alpha(t^-),z)\,\widetilde{N}(dt,dz),
\]
with the Markov chain $\alpha_t$ evolving according to state-dependent transition rates $q_{ij}(x)$:
\[
\mathbb{P}\{\alpha(t+\Delta) = j \mid X(t) = x, \alpha(t) = i\} = q_{ij}(x)\Delta + o(\Delta), \quad i \ne j.
\]
This hybrid construction enables the full generator (see Section 3) to capture both local differential, jump, and discrete switching effects [1702.01495].

## 2. Feynman–Kac Representation for Integro-Differential Problems

The Feynman–Kac formula connects solutions of a backward Kolmogorov-type PIDE with expectations over the regime-switching jump-diffusion paths. Consider the backward Cauchy problem:
\[
\partial_t u(t,x,i) + \mathcal{L}u(t,x,i) - c(x,i)\,u(t,x,i) = 0, \qquad u(T, x, i) = f(x, i),
\]
where $c(x,i)$ is a nonnegative, continuous discount or killing rate, $f$ is a terminal datum, and $\mathcal{L}$ is the full infinitesimal generator.

The Feynman–Kac representation asserts:
\[
u(t,x,i) = \mathbb{E}_{x,i} \Bigg[\exp\left(-\int_t^T c(X_s, \alpha_s)\,ds\right)\, f(X_T, \alpha_T)\Bigg],
\]
where the expectation is taken over the solution path $(X_s, \alpha_s)$ starting from $(x, i)$ [1702.01495]. This correspondence is justified by applying Itô’s formula to the process $e^{-\int_t^s c(X_r, \alpha_r) dr} u(s,X_s, \alpha_s)$.

## 3. Generator Structure and Integro-Differential Decomposition

The full generator $\mathcal{L}$ acting on suitable test functions $g(x,i)$ ($C^2$ in $x$ for each $i$) is:
\[
\begin{aligned}
\mathcal{L} g(x,i) &= D_x g(x,i) \cdot b(x,i) +
    \frac{1}{2} \operatorname{Tr}\big[\sigma(x,i)\sigma(x,i)^\top D_x^2 g(x,i)\big] \\
    &\quad + \sum_{j \ne i} q_{ij}(x) [g(x,j) - g(x,i)] \\
    &\quad + \int_{\mathbb{R}^n \setminus \{0\}} \Big\{g(x+y(x,i,z), i) - g(x,i) - D_x g(x,i) \cdot y(x,i,z)\Big\} \nu(dz).
\end{aligned}
\]
It is customary to split $\mathcal{L} = L_{\text{diff+jump}}^{(i)} + L_{\text{switch}}^{(i)}$, separating the continuous (diffusion and jump) and discrete (switching) contributions [1702.01495].

## 4. Existence and Well-Posedness Conditions

Well-posedness of the SDE, moment bounds, and validity of the Feynman–Kac representation depend on global Lipschitz continuity in $x$ (uniform in $i$) for $b$ and $\sigma$, quadratic and linear growth bounds on $y(x,i,z)$, and boundedness of switching rates $q_{ij}(x)$. For boundary value problems, the domain $D \subset \mathbb{R}^n$ must be $C^2$, $c \geq 0$ continuous on $D \cup \partial D$, and boundary data continuous and bounded.

Under these standing assumptions, the regime-switching SDE admits a unique strong solution without explosion, and the associated PIDE has a unique (classical or viscosity) solution matching the Feynman–Kac formula [1702.01495].

## 5. Boundary, Initial, and Nonhomogeneous Value Problems

The Feynman–Kac approach extends to several PIDE scenarios:

- **Backward Cauchy problem (pure terminal data):**
  \[
  \partial_t u + \mathcal{L} u - c u = 0, \quad u(0, x, i) = f(x, i)
  \]
  admits the representation
  \[
  u(t, x, i) = \mathbb{E}_{x,i}\Big[e^{-\int_0^t c(X_s, \alpha_s) ds} f(X_t, \alpha_t)\Big].
  \]
- **Nonhomogeneous terminal-value problem:**
  \[
  \partial_t u + \mathcal{L} u - c u = g(t,x,i), \quad u(T) = f
  \]
  with solution
  \[
  u(t,x,i) =
    \mathbb{E}_{x,i}\left[
    e^{-\int_t^T c} f(X_T) +
    \int_t^T e^{-\int_t^s c} g(s,X_s) ds
    \right].
  \]
- **Dirichlet (boundary) problem:**
  \[
  \mathcal{L}u - c u = 0 \text{ in } D, \quad u = \varphi \text{ on } \partial D,
  \]
  leads to
  \[
  u(x, i) =
    \mathbb{E}_{x,i}\left[
      e^{- \int_0^\tau c(X_s, \alpha_s) ds} \varphi(X_{\tau}, \alpha_{\tau})
    \right],\quad \tau = \inf\{t: X_t \notin D \}.
  \]
For each case, the Feynman–Kac formula provides not only a representation but, under analyticity assumptions, a means of verifying that the expectation satisfies the corresponding PIDE [1702.01495].

## 6. Representative Examples and Application Domains

**(a) Jump-diffusion Black–Scholes with switching:**  
In a risk-neutral regime-switching jump-diffusion Black–Scholes model, the log-stock $S_t$ satisfies:
\[
dS = S \Big[(r(\alpha_t) - \lambda)\, dt + \sigma(\alpha_t) dW + \int y(\alpha_t,z)\,\widetilde{N}(dt,dz)\Big]
\]
The no-arbitrage price $V(t, s, i)$ of a European contingent claim, with payoff $h(S_T, \alpha_T)$, is
\[
V(t,s,i) = \mathbb{E}_{s,i}\left[e^{-\int_t^T r(\alpha_u)\,du} h(S_T, \alpha_T)\right],
\]
which solves a generalized Black–Scholes integro-PDE under the operator $\mathcal{L}$ [1702.01495].

**(b) Two-time-scale switching/arcsine law:**  
In the fast-switching limit ($\alpha^\epsilon$ with generator $Q/\epsilon$, $\epsilon \downarrow 0$) and vanishing drift, the occupation-time functional
\[
\theta^\epsilon(T) = \int_0^T f(X^\epsilon(s)) ds
\]
appropriately normalized converges in law to the classical arcsine law, connecting the weak convergence of the process to universal limit theorems [1702.01495].

## 7. Analytical and Computational Consequences

For regime-switching jump-diffusions, the Feynman–Kac formula serves as both (i) an existence/uniqueness tool (via stochastic representation and classical solution) and (ii) a pathway to practical Monte Carlo computation for local, global, and boundary-value problems. The underlying technical tools include extended Itô formulas for jump-diffusion-switching processes and careful moment and exit-time analysis under the stated Lipschitz and growth constraints [1702.01495]. This framework is indispensable in stochastic modeling and financial engineering for processes with discontinuities, regime shifts, and random switching mechanisms.

Source: https://www.emergentmind.com/topics/feynman-kac-formula-63ee39d8-d061-4ea9-acc6-d5bbcbabf17c