McKean Feynman–Kač Paradigm

• The McKean Feynman–Kač paradigm is a framework that unifies nonlinear stochastic dynamics with probabilistic representations and non-conservative PDEs through mean-field interactions and exponential weighting.
• It employs a coupled system of SDEs and linked PDEs, where Feynman–Kač weights encapsulate reaction and killing effects to reflect changes in the system's total mass.
• The approach supports numerical methods via interacting particle systems and finds applications in stochastic control, quantum geometry, and noncommutative frameworks.

The McKean Feynman–Kač paradigm designates a general framework that unifies nonlinear stochastic dynamics of mean-field (McKean–Vlasov) type with Feynman–Kac probabilistic representations. It describes systems where the evolution of a process and an associated (possibly non-conservative) PDE are linked through expectations involving exponential functionals. This structure arises both in the analysis of nonlinear PDEs, interacting particle systems, and in the mathematical formulation of a broad class of stochastic models with feedback, reaction, or killing phenomena.

1. Mathematical Formulation of the McKean Feynman–Kač Paradigm

The central object is a coupled system: a mean-field stochastic differential equation (SDE) together with a linking relation involving a Feynman–Kac-type exponential weight. The canonical formulation is as follows (Lieber et al., 2018, Izydorczyk et al., 2019):

Let $(W_t)_{t\ge0}$ be a $d$-dimensional Brownian motion, $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$ bounded and measurable, initial law $\mathbf{u}_0$, and coefficients

$$b_0: [0,T]\times\mathbb{R}^d \to \mathbb{R}^d, \quad b: [0,T]\times\mathbb{R}^d \times \mathbb{R} \to \mathbb{R}^d, \quad \Lambda: [0,T]\times\mathbb{R}^d\times\mathbb{R} \to \mathbb{R}$$

(bounded, Lipschitz in the last variable). The McKean–Feynman–Kač equation is

$$\begin{cases} Y_t = Y_0 + \int_0^t \Phi(s, Y_s) dW_s + \int_0^t [b_0(s, Y_s) + b(s, Y_s, u(s, Y_s))] ds \ Y_0 \sim \mathbf{u}_0 \ \int_{\mathbb{R}^d} \varphi(x) u(t,x)\,dx = \mathbb{E}\bigg[\varphi(Y_t) \exp\left\{\int_0^t \Lambda(s, Y_s, u(s, Y_s)) ds \right\}\bigg],\; \forall \varphi \in C_b(\mathbb{R}^d) \end{cases}$$

Formally, $u(t,x)$ is the weighted density of $Y_t$ and satisfies a non-conservative semilinear PDE: $$\partial_t u = L_t^* u - \nabla_x \cdot [b(t,x, u(t,x)) u] + \Lambda(t,x,u(t,x)) u$$ with $$L_t^*\varphi = \frac12 \sum_{i,j} \partial_{x_i x_j}^2[a_{ij}(t,x)\varphi] - \sum_j \partial_{x_j}[b_{0,j}(t,x)\varphi]$$, $d$0.

When $d$1, this reduces to the classical McKean–Vlasov scenario; for non-zero $d$2, the system is non-conservative (mass is not preserved) and supports reaction terms (Lieber et al., 2018, Izydorczyk et al., 2019, Morale et al., 6 Feb 2026).

2. Representation Theorems and Analytic Structure

Under general conditions (boundedness, weak regularity, uniform non-degeneracy), existence, uniqueness, and pathwise uniqueness for the SDE component are established by relating solutions of the MFKE to solutions of the associated non-conservative PDE (Lieber et al., 2018, Izydorczyk et al., 2019). Theorems guarantee:

• If $d$3 solves the MFKE and $d$4, then $d$5 is a weak solution of the PDE.
• Conversely, if $d$6 solves the PDE (in weak/mild sense) and $d$7, $d$8 are bounded, then there exists a unique-in-law process $d$9 solving the SDE, and the linking Feynman–Kač relation is satisfied.
• Additional regularity (Lipschitz $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$0) yields strong existence and pathwise uniqueness for the SDE.

Key tools include: Itô’s formula for functionals $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$1, martingale problem approach (Stroock–Varadhan), existence of fundamental solution for parabolic PDEs, and analytic $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$2 fixed-point arguments for the PDE (Lieber et al., 2018, Izydorczyk et al., 2019).

3. Probabilistic Representations and Dual Approaches

Two principal probabilistic mechanisms capture the non-conservative PDE's dynamics on the stochastic level:

• Feynman–Kač Weighting: Each trajectory is exponentially weighted by a functional $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$3, and $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$4 is reconstructed as an expectation over the (possibly interacting) SDE (Izydorczyk et al., 2019, Morale et al., 6 Feb 2026).
• Survival/Killing Interpretation: The reaction term is interpreted as a random killing rate; the law $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$5 is the sub-probability corresponding to surviving particles up to time $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$6. This leads to alternate particle systems where survival times are governed by the reaction functional (Morale et al., 6 Feb 2026).

Both viewpoints yield equivalent macroscopic PDEs. The Feynman–Kač version is particularly suitable for analytic approaches and fixed-point arguments, while the killing-time approach facilitates probabilistic intuition and efficient simulation algorithms.

Path-dependence in MFKE is realized through dependencies of $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$7 and $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$8 on historical functionals of the empirical density, leading to non-Markovian dynamics in the law and the need for pathwise analysis (Morale et al., 6 Feb 2026).

4. Extensions: Geometry, Wasserstein Spaces, and Conditioned Laws

The framework has been extended to measure-valued and conditional McKean–Vlasov SDEs, especially in the presence of common noise or image-dependence, with the associated PDEs defined on the $$\Phi : [0,T]\times\mathbb{R}^d \to \mathbb{R}^{d\times d}$$9-Wasserstein space $\mathbf{u}_0$0 (Wang, 2019). In this setting:

• SDEs depend on the conditional law given common noise, and
• The associated PDE is a nonlinear Schrödinger-type PDE on $\mathbf{u}_0$1 involving Lions’ derivatives and intrinsic second-order operators.

The Feynman–Kač representation for this infinite-dimensional PDE is analogous: the value function is given as the expectation of terminal data and a running source, exponentiated against the path-integral of a potential, evaluated along the conditional McKean process (Wang, 2019).

This setting enables strong well-posedness results under monotonicity conditions weaker than standard Lipschitz conditions, promoting strong solutions and ergodicity for the measure-valued diffusions.

5. Computational Algorithms and Particle Approximations

A core feature of the McKean–Feynman–Kač paradigm is the rigorous connection to interacting particle systems and probabilistic numerical schemes:

• Weighted particle approximations employ the exact dynamics and Feynman–Kač exponential weights to form kernel-based approximations for $\mathbf{u}_0$2 (Izydorczyk et al., 2019).
• In survival-time (killed) particle systems, empirical measures of surviving particles converge to sub-probability solutions of the non-conservative PDE (Morale et al., 6 Feb 2026).
• Particle filters with Feynman–Kač weights are further optimized using optimal transport couplings (Ensemble Transform Particle Filters), achieving variance reduction and improved consistency in high dimensions by deterministic optimal transport rather than resampling (Cheng et al., 2013).
• Forward-backward SDE (FBSDE) approaches for stochastic control deploy McKean–Markov branched sampling, reweighting, and entropy-regularized regressions for efficient solution of associated HJB or Kolmogorov equations (Hawkins et al., 2022).
• Propagation of chaos results guarantee that as the number of particles grows, the empirical distribution of the (possibly weighted or killed) particles converges to the solution of the mean-field equations.

6. Impact on Nonlinear PDEs, Stochastic Control, and Geometric Representations

The MFKE admits diverse applications:

• Numerical PDE Solution: MFKE-based particle schemes underpin robust Monte Carlo methods for nonlinear and non-conservative PDEs, including those with path-dependent or reaction terms (Izydorczyk et al., 2019, Morale et al., 6 Feb 2026).
• Optimal Control: Nonlinear Feynman–Kač representations provide probabilistic solutions to value functions of stochastic optimal control problems for McKean–Vlasov dynamics (HJB equations on Wasserstein space), extendable to open-loop controls via randomized DPP and FBSDE (Bayraktar et al., 2016, Hawkins et al., 2022).
• Quantum and Information Geometry: Functorial frameworks reinterpret Feynman–Kač correspondences as monoidal functors associating path-space stochastic processes with flows on statistical or information-geometric manifolds; these generalize Chapman–Kolmogorov equations and variational principles (Sakthivadivel, 2022).
• Geometry and Noncommutative Contexts: MFKE is realized on vector bundles, with connections to noncommutative geometry and probabilistic representations of objects such as the Chern character and JLO cocycle via operator trace Feynman–Kač formulas (Boldt et al., 2020).

7. Summary of Distinctive Phenomena and Theoretical Features

The McKean–Feynman–Kač paradigm generalizes the mean-field interaction framework by introducing non-conservative effects and path-dependent feedback at the microscopic scale:

• The exponential (Feynman–Kač) weight introduces birth/death or reaction phenomena, leading to semilinear, non-conservative PDEs.
• The law of the SDE is replaced by a weighted measure, with the weight depending on trajectory functionals, thus breaking the classical Markov property in the law.
• Non-conservativity ($\mathbf{u}_0$3) results in possible loss/gain of total mass, and phenomena such as extinction, blowup, or equilibrium selection.
• The theory provides control over weak, strong, and pathwise uniqueness regimes under regularity/monotonicity assumptions and is robust to path-dependent or degenerate diffusion settings (Lieber et al., 2018, Izydorczyk et al., 2019, Wang, 2019).
• All these features are unified in a rigorous analytic-probabilistic duality that facilitates theoretical analysis and efficient computation for broad classes of mean-field and non-linear stochastic systems.

The McKean–Feynman–Kač paradigm is thus foundational in modern analysis of nonlinear stochastic dynamics, probabilistic numerics, and mathematical modeling where mean-field structure and reaction/killing effects are essential.