We know how to find the law of motion (Ito process) of the value function: $$V_t(x)=E\Big{[}\int^{T}_te^{-r (s-t)}f(s,X_s)ds+e^{-r (T-t)}g(T, X_{T})|\mathcal{F}_t\Big{]}$$ such that $$dX_t=\mu(t,X_t)dt+\sigma(t,X_t)dB_t, \quad X_t =x$$ and $\mathcal{F}_t$ is the filtration generated by the brownian motions up to time t. It is, using the martingale representation theorem: $$dV_t=rV_tdt-f(t,X_t)dt+\theta_tdB_t$$ For some square integrable process $\theta_t$.
However, in an application, I'm struggling to find the law of motion of the change in the value function when one perturbs the initial condition $x$ ie the law of motion of $W_t(x)=\dfrac{\partial V_t}{\partial x}$. Any hints or reference on how I could proceed to find it?