Additive Functionals of Stochastic Processes

• Additive functionals of stochastic processes are constructed by integrating observables along sample paths, addressing both continuous and jump dynamics.
• They underpin key limit theorems such as the SLLN, CLT, and LIL, providing insight into ergodic behavior and scaling limits.
• Advanced analysis incorporates Feynman–Kac formalisms, Dirichlet forms, and rough path techniques, linking spectral theory with stochastic calculus.

Additive functionals of stochastic processes are random variables or processes constructed as (typically pathwise) sums or integrals of observables along the trajectories of a base stochastic process, with constructions encompassing both continuous and jump processes, interacting particle systems, Markov/Feller processes, stochastic differential equations (including path-dependent, McKean–Vlasov, G-Lévy, Volterra, and structured processes), and broad areas such as rough path theory, high-frequency statistics, ergodic theory, and large deviations.

1. Definitions and Canonical Constructions

Additive functionals are generally formulated by integrating observables along sample paths. For a process $X=(X_t)_{t\geq0}$ on a filtration $(\mathcal F_t)$, the classical forms are:

• Continuous time: $A_t = \int_0^t f(X_s)\,ds$, or in a more general form, $$A_t = \int_0^t a(X_s)\,ds + \int_0^t b(X_s) \circ dX_s$$ (Stratonovich or Itô), or as mixtures including jump terms and measure dependence (Dieball et al., 2022).
• Discrete time: $S_n = \sum_{k=1}^n f(X_k)$.
• General Markov/Girsanov/Feynman–Kac/Dirichlet forms: Additive functionals comprise continuous martingale parts, Lebesgue/zero-quadratic variation parts, and jump sums, often within Fukushima-type decompositions (Kuwae, 2010, Chen et al., 2014, Walsh, 2011).

For path-dependent processes, observables $f$ may depend on the memory segment $X_t(\cdot)$ rather than just $X_t$ (Bao et al., 2019). Additive functionals of interacting particle systems or processes on infinite-dimensional spaces may involve sums over spatial and/or temporal indices (Gonçalves et al., 2011).

2. Limit Theorems: Law of Large Numbers, CLT, LIL

Fundamental probabilistic structures underpin the large-time and large-scale behavior of additive functionals:

• Strong Law of Large Numbers (SLLN): Under exponential mixing and regularity, $A_t(f)/t \to \pi(f)$ a.s.; the invariant measure $\pi$ and path-Lipschitz structure play key roles (Bao et al., 2019).
• Central Limit Theorem (CLT): When $\pi(f)=0$, rescaled functionals converge: $t^{-1/2}A_t(f) \to N(0, D_f^2)$, where $D_f^2$ involves Poisson-corrector integrals (Bao et al., 2019).
• Law of Iterated Logarithm (LIL): Refined fluctuations governed by the Strassen ball in the functional convergence topology (Bao et al., 2019).

These principles extend to particle systems, where diffusive rescaling leads to fractional Brownian motion or non-Gaussian fluctuation fields (Gonçalves et al., 2011), and to ergodic Markov models under rough path convergence (Deuschel et al., 2019).

3. Functional Calculus, Spectral Theory, Feynman–Kac Formalism

The analysis of time-integrated observables (density, current, work, heat) within stochastic thermodynamics utilizes Feynman–Kac tilting, which computes moments/generating functions via PDEs for tilted generators. The formalism comprises:

• Itô and functional calculus derivations: Backward Kolmogorov equations and Dyson series expansions yield explicit cumulant representations (Dieball et al., 2022).
• Principal eigenvalue asymptotics: Long-time cumulant generating functions and large deviation rate functions are characterized by leading spectral data of the tilted generator (Dieball et al., 2022).
• Generalizations: These results unify methodologies across stochastic calculus, path integrals, and functional perturbation series.

Quasi-ergodic theorems for additive functionals under Feynman–Kac weighted observation yield functional LLN, quasi-stationary distributions, and convex large deviation rate functions parameterized by spectral data (Kim et al., 2024).

4. Additive Functionals for Generalized Markov and SDEs

Advanced models encompass path-dependent SDEs, McKean–Vlasov equations, Volterra equations, SDEs with G-Lévy noise, and processes with nonregular/parameter-dependent drift:

• Path-dependent SDEs: The Markov property is lifted to segment processes; additive functionals are constructed for observables on the memory path, with mixing in Wasserstein quasi-metrics and limit theorems established via martingale difference sequences (Bao et al., 2019).
• McKean–Vlasov SDEs with jumps: Path independence of additive functionals is characterized via nonlinear PIDEs, involving Lions-derivatives and measure-dependent equations (Qiao et al., 2019).
• G-Lévy SDEs: Path independence becomes equivalent to the solvability of fully nonlinear PIDEs under sublinear expectation, generalizing change-of-measure and risk measure theory in uncertain jump-diffusion models (Qiao et al., 2020).
• Volterra SPDEs: For processes with singular/fractional kernels, path independence is characterized by the solution of parabolic SPDEs with fractional Laplacians; this extends beyond Markovian/semimartingale frameworks (Qiao et al., 2022).
• Nonregular parametric dependence: Weak convergence of additive functionals is exhibited for SDEs with highly nonuniform drifts, including cases with scaling limits to non-Gaussian or skew-diffusive processes (Kulinich et al., 2016).

5. Stochastic Calculus for Dirichlet Forms and Additive Functionals

The stochastic calculus of Markov processes driven by Dirichlet forms provides generalized decompositions:

• Fukushima decomposition: Any suitable function of the process can be written as the sum of a martingale AF, zero energy continuous AF (CAF), and a sum over large jumps, characterized via local domains and capacities (Chen et al., 2014, Kuwae, 2010, Walsh, 2011).
• Zero quadratic variation integration: Unique stochastic integrals exist for AFs of zero QV, allowing an Itô formula for nonsmooth, time-dependent functionals and encompassing pathwise extensions (Klimsiak, 2010, Walsh, 2011).
• Extended Itô and Stratonovich formulas: Chain rules incorporate the energy measures, jump and killing contributions, and zero QV terms, with localization and nest arguments ensuring q.e. validity (Kuwae, 2010).

6. Scaling Limits, Particle Systems, Rough Paths, and Statistical Functional Laws

Special cases and general methodologies:

• Interacting particle systems: Additive functionals under diffusive scaling yield fractional Brownian motion or quadratic field limits. The local Boltzmann–Gibbs principle connects microscopic observables to polynomial expansions in the density (Gonçalves et al., 2011).
• Hayashi–Yoshida estimators: High-frequency statistics of asynchronously observed semimartingales lead to nontrivial LLNs for increment functionals, contingent upon the overlap structure and vanishing properties of the test functions (Martin et al., 2018).
• Scaling limits for synchrony models: Sequences of Markov processes with rapid hitting yield additive functionals converging to Lévy subordinators; explicit Laplace exponents are given for Wright–Fisher, CIR, and drifted Brownian models, with direct application to neural synchrony modeling (Taillefumier et al., 2024).
• Rough path convergence: Under Kipnis–Varadhan conditions, additive functionals of stationary Markov processes converge in rough-path topology to Brownian (Stratonovich) lifts, with explicit area anomaly terms tied to asymmetries of the generator (Deuschel et al., 2019).

7. Excursions, Path Functionals, and Persistence

Excursion-theoretic formulations with zero drift (Lamperti-type) and skew/self-similar processes:

• Excursion sums: Sharp almost-sure asymptotics for sums $\sum X_s^\alpha$ are established for processes with asymptotically zero drift, yielding trichotomies spanning null recurrence, heavy-tailed regimes, and positive recurrence (Hryniv et al., 2012).
• Persistence and exit distributions: Closed-form exponents for time to first passage and exit density distributions are derived for homogeneous additive functionals of skew Bessel processes (and Brownian motion), via analytic continuation and scaling arguments (Profeta, 2019).

---

Collectively, research on additive functionals of stochastic processes integrates functional analytic, probabilistic, spectral, and statistical viewpoints. The rigorous structure of additive functionals—encompassing limiting laws, functional calculus, stochastic integration, generalized decompositions, and scaling regimes—is foundational to ergodic theory, statistical inference, non-equilibrium physics, mean field dynamics, and functional analysis of random systems.