Existence of strong solution to SDEs with non-Lipschitzian drift

Consider the SDE: $$dX_t=b(X_t)dt+dW_t\quad X_0=x$$ If $b$ is bounded Borel function, using Zvonkin's Transform, one can prove there exists a unique strong solution.

I want to know if we assume $b$ satisfies some better condition but non-Lipschitz, like $b\in C^\alpha$, can we get the existence of strong solution to this equation without using Zvonkin's transform.

I tried to consider the Eular approximation: $$X^n_t=X^n_{t_n}+b(X_{t_n})(t-t_n)+(W_t-W_{t_n})\quad X^n_0=x$$ where $t_n=[t2^n]/2^n$. But I can prove the relatively compact property of $\{X^n_{.}\}$ in the metric space $$d_F(X^n,X^m)=E\frac{\sup_{t\leq 1}|X_t^n-X_t^m|}{1+\sup_{t\leq 1}|X_t^n-X_t^m|}$$ when $b\in C^\alpha$.

• I removed my answer, since I misunderstood your question. – Nawaf Bou-Rabee Mar 24 '15 at 20:35

This question has recently been addressed in detail in the article "Averaging along irregular curves and regularisation of ODEs" by Catellier and Gubinelli. They give a purely analytical condition on the "irregularity" of $W$ and the regularity of $b$ which guarantees that the fixed point equation $$x(t) = x_0 + W(t) + \int_0^t b(x(s))\,ds$$ admits a unique solution in the space of continuous functions. In particular, for any fixed $b \in C^\alpha$ with $\alpha>0$, their condition is satisfied for almost every sample path of a Brownian motion. If $W$ is even more irregular (for example a typical trajectory of fractional Brownian motion with small Hurst parameter), then they can even build solutions for distributional drifts $b$. Since the constructions are purely analytic, the solutions are automatically strong.
• Many thanks for your help. But I feel confused that why the solutions are strong? Because we don't know any measurable property of $T: W_. \rightarrow x_.$ – Guohuan Zhao Mar 25 '15 at 3:20
• Thanks, I know their approach. But I want to use some technique to solve the following problem: $X_t=x+\int_0^tb(X_s)ds+L_t$ where $L_t$ is the cylindrical $\alpha-$stable process ($L_t=(L^1_t,...L_t^d)$ where $L_t^i$ is independent 1-d $\alpha-$stable process with $\alpha\leq 1$). In "On the construction and Malliavin differentiability of solutions of Lévy noise driven SDE’s with singular coefficients" the author discussed the related problem when $L_t$ is $\alpha-$stable process when $\alpha> 1$. I don't how the construct a strong solution directly without using Zvonkin transform in my case. – Guohuan Zhao Mar 25 '15 at 3:22