Zvonkin Transformation: A Key Stochastic Solution

• Zvonkin Transformation is an analytical technique for regularizing stochastic differential equations (SDEs) with irregular drift coefficients.
• This method leverages a non-linear transformation using a parabolic partial differential equation (PDE) to reduce SDEs to globally Lipschitz conditions.
• It enables the derivation of strong well-posedness and key inequalities like Krylov and Harnack for challenging stochastic systems.

The Zvonkin transformation is a fundamental analytical technique for addressing stochastic differential equations (SDEs) whose drift component is irregular, singular, or merely integrable in time and space. By constructing a non-linear, often time-dependent transformation mapped via the solution of a well-posed parabolic partial differential equation (PDE), this method regularizes the drift, typically reducing the SDE to one with (global) Lipschitz coefficients. This approach underpins strong well-posedness results for classes of SDEs that are otherwise out of reach for classical stochastic analysis, and it is instrumental in the derivation of sharp Krylov and Harnack-type inequalities.

1. Analytical Framework and the Defining Parabolic PDE

Consider an SDE driven by Brownian motion $(W_t)_{t\in[0,T]}$ in $\mathbb{R}^d$ of the form

$dX_t = b(t,X_t)dt + \sigma(t,X_t)dW_t,$

with drift $b$ having decomposition $b = b_1 + b_0 + b_2$, where $b_1$ is globally Lipschitz, $b_0$ is singular but satisfies an $L^p$–$L^q$ Krylov-type integrability, and $b_2$ is bounded. The Zvonkin transformation begins by solving a vector-valued, backward parabolic PDE for a function $u: [0,T]\times\mathbb{R}^d \to \mathbb{R}^d$: $$\partial_t u^i + \mathrm{tr}(a\nabla^2 u^i) + (b_1 + b_0 + b_2)\cdot \nabla u^i - \lambda u^i = -b_0^i + \lambda u^i$$ with terminal condition $u^i(T,x) = 0$. The matrix $a = \frac{1}{2}\sigma\sigma^*$ is assumed uniformly elliptic and continuous. The key is to move the problematic component $b_0$ to the right-hand side, enabling its removal from the transport structure of the PDE (Yuan et al., 2019).

2. $L^p$-$L^q$ Estimates and Sobolev Regularity

The a priori estimates hinge on weighted Sobolev spaces for

$$b_0 \in L^q([0,T];L^p(\mathbb{R}^d)), \quad \frac{d}{p}+\frac{2}{q}<1,$$

guaranteeing solution regularity: $$u \in W^{1,q}([0,T]; L^p_w)\cap L^q([0,T]; W^{2,p}_w), \quad w(x) = (1+|x|^2)^{-2}.$$ The norm bounds,

$$\|u\|_{L^q W^{2,p}_w} + \|\partial_t u\|_{L^q L^p_w} \leq C\|b_0\|_{L^qL^p},$$

as well as higher-order estimates and the crucial property $\|\nabla u\|_{L^\infty}<1$ for large $\lambda$ facilitate the structural invertibility of the Zvonkin map and control of the nonlinearity (Yuan et al., 2019).

3. The Zvonkin Map and Drift Regularization

Define the time-dependent diffeomorphism $\Phi_t(x) = x + u(t,x)$ and corresponding process $Y_t = \Phi_t(X_t)$. Application of Itô's formula yields

$$dY_t = [b_1 + b_2 + \lambda u - b_0](t,X_t)dt + [I + \nabla u](t,X_t)\sigma(t,X_t)dW_t + \partial_t u(t,X_t)dt,$$

in which the carefully constructed PDE cancels the singular drift $b_0$. Changing variables to $Y$-space produces an SDE with all drift and diffusion coefficients globally Lipschitz or bounded, enabling classical strong well-posedness and stochastic flow results (Yuan et al., 2019).

4. Krylov Estimates for Integrable Singular Drifts

The Zvonkin framework provides a pathway to explicit Krylov-type estimates. Supposing $X_t$ solves the SDE under the above integrability assumptions, for every nonnegative $f\in L^q([0,T];L^p)$ and every stopping time $S\le T$,

$$\mathbb{E}\left[\int_S^T f(s,X_s)ds\,\middle|\,\mathcal{F}_S\right] \leq C(1+\|f\|_{L^q([S,T];L^p)}).$$

The key argument employs mollification and weighted Sobolev relaxation to handle $b_0\cdot\nabla u$, followed by an iterative argument in $(p,q)$-space to saturate the Krylov range $\frac{d}{p}+\frac{2}{q}<1$ (Yuan et al., 2019).

5. Harnack Inequalities via the Zvonkin Method

Post-transformation, the new SDE supports standard coupling or Girsanov arguments, yielding both log-Harnack and power-Harnack inequalities for the transition semigroup $P_T$. For example, if $\sigma$ is additionally Hölder continuous,

$(P_T f(y))^p \leq P_T(f^p)(x) \exp\{C_p |x-y|^2\},$

for all $f\ge0$, $p>1$. The derivation follows from coupling by change of measure, moment estimates on the Girsanov density, and the application of Hölder's inequality, then pulling back these inequalities through the Zvonkin map for the original process (Yuan et al., 2019).

6. Extensions and Variations

Several advanced extensions of the Zvonkin transformation have been developed:

• For SDEs with merely Hölder or Besov regular drift and possibly unbounded coefficients, the Zvonkin map solves a time-inhomogeneous parabolic PDE in weighted or function space contexts, enabling strong well-posedness and stochastic flows of diffeomorphisms even under weak regularity (Moritoki et al., 8 Jan 2026, Ørke, 3 Jan 2025).
• In the McKean–Vlasov or distribution-dependent context, the Zvonkin approach adapts to master-type PDEs on $$[0,T]\times\mathbb{R}^d\times \mathcal{P}_2(\mathbb{R}^d)$$, inducing regularization not only in space but in the measure variable via smoothing estimates for the associated parametrix expansion (Raynal, 2015, Huang et al., 2019).
• When SDEs are driven by non-Gaussian (e.g., $\alpha$-stable) noise, Zvonkin transformations involve nonlocal, integro-differential PDEs for the correction term, under suitable assumptions on the drift's regularity (Huang et al., 2019).
• In domains with reflection (possibly with non-smooth, time-dependent boundaries), the Zvonkin method is coupled with Neumann boundary-value parabolic PDEs, carefully transferring the drift removal and reflection structure through functional analytic and stochastic flow arguments (Yang et al., 2020).
• For SDEs with random or path-dependent coefficients, the transformation relies on solving backward stochastic Kolmogorov equations, with the differentiability in the noise direction handled via Malliavin calculus (Zhao, 2020).

7. Summary and Impact

The Zvonkin transformation provides a constructive bridge between analytic PDE estimates and probabilistic SDE theory, converting equations with singular or low-regularity drift into settings suitable for standard stochastic analysis. It allows for strong well-posedness, sharp probabilistic estimates (Krylov, Harnack), and enables the analysis of numerical approximation schemes (e.g., Euler–Maruyama), broadening the classes of SDEs treatable with rigorous mathematical tools (Yuan et al., 2019, Moritoki et al., 8 Jan 2026, Ørke, 3 Jan 2025, Raynal, 2015, Zhao, 2020, Huang et al., 2019, Yang et al., 2020).