Stratonovich SPDE: Theory and Applications

• Stratonovich SPDEs are stochastic evolution equations defined with the Stratonovich integral, preserving the classical chain rule essential for geometric and physical consistency.
• They involve an intricate Itô–Stratonovich conversion that introduces drift correction terms and are analyzed using variational methods, Feynman–Kac representations, and chaos expansions.
• Their applications span fluid dynamics, fluctuating hydrodynamics, and stochastic control, ensuring model invariance and regularity in complex infinite-dimensional settings.

A Stratonovich Stochastic Partial Differential Equation (SPDE) is a stochastic evolution equation in which stochastic integrals are understood in the Stratonovich sense, i.e., preserving the classical chain rule and invariance properties familiar from deterministic calculus. Stratonovich SPDEs are encountered in fields ranging from mathematical physics and fluid dynamics to hydrodynamic limits in statistical mechanics. Their analytical and probabilistic properties, as well as their relation to Itô-based SPDE theory, have motivated a substantial body of mathematical research, targeting foundational questions regarding solution concepts, regularity, stochastic calculus in infinite dimensions, geometric invariance, and physically meaningful modeling.

1. Foundations and Definitions

In its general form, a Stratonovich SPDE is written as

$$d\Psi_t = A(t, \Psi_t)\,dt + G(t, \Psi_t) \circ dW_t,$$

where $A$ is a (possibly nonlinear, unbounded) deterministic operator, $G$ is a (possibly time-dependent and nonlinear) noise operator, $W_t$ is a cylindrical or Q-Wiener process on a Hilbert space, and $\circ$ denotes a Stratonovich integral. In contrast to the Itô formulation, the Stratonovich integral is defined either as the limit of midpoint (symmetric) Riemann sums or through appropriately mollified approximations, enabling direct application of classical calculus rules and geometric (e.g., variational, invariance) principles (Goodair, 5 Aug 2025, Honkonen, 2011).

The Stratonovich approach is particularly suited to physical models in which the stochasticity arises as the scaling limit of deterministic systems (e.g., via homogenization, rough paths, or physical modeling frameworks such as fluctuating hydrodynamics) or in models requiring coordinate-invariant chain rule properties (as in geometric mechanics and Hamiltonian systems) (Holm, 2019, Ciotir et al., 2015).

2. Itô–Stratonovich Conversion in SPDEs

The transition from Stratonovich to Itô calculus in the finite-dimensional case involves a correction drift term resulting from the quadratic variation of the noise. In infinite dimensions and with nonlinear or unbounded noise coefficients, this conversion is substantially more technical. Specifically, if $G$ is decomposed along an orthonormal basis $\{e_i\}$ as $G(t, \Psi_t)(e_i) =: G_i(t, \Psi_t)$, then the Itô–Stratonovich conversion formula reads: $$d\Psi_t = A(t, \Psi_t)\,dt + \frac{1}{2} \sum_{i=1}^\infty D_u G_i(t, \Psi_t)[G_i(t, \Psi_t)]\,dt + G(t, \Psi_t)\,dW_t,$$ where $D_u G_i$ denotes the Fréchet (Hilbert-space) derivative of $G_i$ with respect to $\Psi_t$ (Goodair, 5 Aug 2025). The corrector term ensures that the Stratonovich form, when translated to the Itô calculus, fully accounts for the interaction between the noncommutativity of the operator $G$ and the stochastic variation of $W_t$. This framework is essential for a rigorous analysis of SPDEs with transport-type noise, as in fluid dynamics and geophysical models.

In applications, the noise operator $G$ is often unbounded (acting as a differential operator), nonlinear, and time-dependent, motivating work in variational settings (e.g., Gelfand triples) and the use of martingale techniques, stopping-time truncations, and careful control of the quadratic variation via localization and approximation (Goodair, 5 Aug 2025, Ciotir et al., 2015).

3. Solution Theory: Existence, Uniqueness, and Feynman–Kac Representations

The solution concept for Stratonovich SPDEs generally adopts the variational (weak or mild) formulation, often justified via Feynman–Kac representations and Picard iterations. In equations of the form

$$\frac{\partial u}{\partial t} = \mathcal{L} u + u \circ \dot{W},$$

with $\mathcal{L}$ the generator of a (possibly nonlocal) Markov process and $\dot{W}$ Gaussian noise (possibly colored in time and/or space), the mild solution in the Stratonovich sense is constructed as the limit of mollified (regularized) stochastic convolutions. For example, in (Song, 2015, Song et al., 2023) the solution admits a Feynman–Kac-type formula,

$$u(t,x) = \mathbb{E}^{X} \left[u_0(X_t^x) \exp \left\{ \int_0^t \int_{\mathbb{R}^d} K_{t-s}(X_t^x-y) W(ds, dy) \right\}\right],$$

where $X_t^x$ is a Lévy or Feller process with generator $\mathcal{L}$, $K_{t-s}$ is an approximation kernel, and the integral is interpreted via the Stratonovich construction. Moment formulas and intermittency properties are derived from multi-point Feynman–Kac representations involving independent copies of $X$ and multiple integrations with respect to the noise (Song, 2015, Song et al., 2023, Balan, 2021).

Uniqueness and stability often require exponential integrability of double time (and space) integrals encoding the covariance structure of the noise and the process $X$: e.g., $$\mathbb{E}\left[\exp\left(\lambda \int_0^t \int_0^t |r-s|^{-\beta_0} \gamma(X_r - X_s)\,dr ds\right)\right] < \infty$$ for suitable spectral conditions (Song, 2015, Song et al., 2023).

4. Analytical and Regularity Properties

Stratonovich solutions typically inherit continuity, regularity, and invariance properties from the underlying SPDE and noise structure.

• Hölder Continuity: Sufficient spectral integrability of the noise and regularity of the generator enable proofs of Hölder continuity (in space and/or time) for the solution (Song, 2015, Song et al., 2023). For example, if the kernel and spectral measure satisfy appropriate moment bounds, the solution is Hölder continuous with positive exponent in each argument.
• Law Regularity: Malliavin calculus is applied to prove that, under non-degeneracy and invertibility of the Malliavin derivative, the law of the solution at fixed points is absolutely continuous and may possess a smooth density (assuming suitable moment and support conditions for the noise and underlying process) (Song et al., 2023).
• Nonnegativity and Support: Feynman–Kac representations guarantee that positivity or nonnegativity of the initial data generally propagates to the solution, ensuring physicality in models where the unknown (e.g., density or temperature) must remain positive (Song et al., 2023, Song, 2015).
• Well-posedness for Irregular Nonlinearities: For singularities (e.g., square-root nonlinearities such as in Dean–Kawasaki models or total variation flows), kinetic solution frameworks and entropy inequalities allow the treatment of coefficients or noise operators that are highly singular or discontinuous in the natural function space (Fehrman, 25 Apr 2025, Ciotir et al., 2015).

5. Structural and Geometric Aspects

• Chain Rule and Functional Calculus: A key attribute of the Stratonovich integral is its preservation of the chain rule in the infinite-dimensional (field-theoretic or functional-analytic) setting, making it possible to apply classical variational methods and construct dynamic actions with correct symmetry properties (Honkonen, 2011).
• Translation and Flow Invariance: Solutions of certain Stratonovich SPDEs can be viewed as flows or translations on function or distribution spaces, as in the translation-invariant models for Lévy-driven equations in spaces of tempered distributions, where the chain rule plays a central technical role. Such constructions can be directly linked to the transformation properties of the underlying stochastic flow (Bhar et al., 2018, Li et al., 2010).
• Spatial and Temporal Corrections in Discretization: Spatial versions of the Itô–Stratonovich correction arise in finite-difference and finite-element approximations of nonlinear SPDEs, revealing that the shape of the limit equation (and resulting correction terms) depends on whether the discretization preserves the chain rule. For example, forward (one-sided) spatial differences in discretizing Burgers-type SPDEs introduce a correction analogous to the temporal Itô–Stratonovich term, leading to drift modifications involving the Laplacian of the nonlinearity (Hairer et al., 2010).
• Functional Integral and Dynamic Action: The different interpretations (Itô versus Stratonovich) manifest in stochastic field theory as ambiguities in the functional Jacobian determinant and dynamic action, which impact not only correlation and response functions but also perturbative expansions, with explicit differences in loop (tadpole) terms capturing the correction (Honkonen, 2011).

6. Applications and Physical Models

Stratonovich SPDEs are widely used for physically consistent stochastic modeling in:

• Fluid Dynamics and Geophysical Flows: Models with transport noise, where the conservation of geometric structures under stochastic flows (e.g., through the Kelvin circulation theorem or variational/Hamiltonian principles) demands Stratonovich integrals (Holm, 2019, Goodair, 5 Aug 2025).
• Fluctuating Hydrodynamics: SPDEs such as the Dean–Kawasaki equation derive from particle-based models and exhibit non-stationary and singular noise, requiring advanced well-posedness theory involving kinetic solutions and new forms of entropy regularization (Fehrman, 25 Apr 2025).
• Stochastic Control and Estimation: Vector bilinear Stratonovich SDEs capture physically accurate uncertainty in dynamic circuits, robotics, and nonlinear control systems, with estimation theory adapted to the Stratonovich framework for rigorous moment and state estimation (Rathore et al., 2019).
• Thermodynamics: In stochastic thermodynamics, different stochastic calculus choices (including anti-Itô or kinetic interpretations) correspond to physically different interpretations of fluctuation-driven drift; the Stratonovich and related prescriptions emerge naturally from physical invariance and transformation properties (Tsekov, 2015, Arenas et al., 2014).

7. Generalizations, Expansions, and Calculation Methods

• Iterated Integral Expansions: Moments and local approximations to Stratonovich SPDE solutions exploit expansions in iterated integrals (stochastic Taylor expansions), with expectation values computable through explicit combinatorial rules, simplifying analysis and numerical schemes (Ladroue, 2010).
• Integral Definitions Beyond Semimartingales: The Stratonovich integral can be generalized to integration with respect to general stochastic measures through symmetric Riemann sums, permitting the definition and analysis of SPDEs outside the framework of classical semimartingales, and ensuring the validity of the chain rule and existence/uniqueness of solutions in this extended context (Radchenko, 2016).
• Chaos Expansions: Stratonovich solutions admit chaos (multiple Wiener–Itô integral) expansions which differ structurally from those of Skorohod/Itô solutions; the comparison elucidates the precise algebraic corrections and reveals the role of higher-order variations, especially with more restrictive spectral conditions (Balan, 2021).

---

Stratonovich SPDEs, through their geometric fidelity and analytic structure, provide a framework that unifies modeling demands from stochastic physics, reliable numerical discretization, function space analysis, and the rigorous probabilistic interpretation of infinite-dimensional stochastic processes. Their analysis and application require an overview of variational methods, martingale problem formulations, Malliavin calculus, kinetic regularization, and careful handling of singularities and invariance principles in both drift and noise. The resulting theory is essential across the paper of transport, diffusion, and hydrodynamic fluctuation in random environments.