Asymptotic error for the Milstein scheme for SDEs driven by continuous semimartingales (Q2496506)

The author considers a general \(q\)-dimensional stochastic differential equation driven by a vector of continuous semi-martingales \(Y \in R^d\) on \([0,1]\) of the form \(X_t=x_0 + \int_0^t f(X_s) dY_s\). He defines an approximation process \(X_t^n \in R^q\), given componentwise for \(1\leq i\leq q\) as \[ X_t^{ni} = X_{n(t)}^{ni} + f^i(X_{n(t)}^{n})(Y_t -Y_{n(t)}) + \text{tr}([f(X_{n(t)}^{n})]^T D f^i(X_{n(t)}^{n})\int_{n(t)}^t Y_s-Y_{n(s)}dY_s^T), \] where \(n(t)= k/n\), if \(k/n< t < (k+1)/n\), \(\text{tr}(A)\) is the sum of the diagonal elements of some matrix \(A\) and \(Df^i=(f_k^{ij})\) with \(f_k^{ij}:= \frac{\partial f^{ij}}{\partial x_k}\). Defining the error process of the above scheme as \(U_t^n=X_t^n-X_t\) the author proves via tightness arguments under certain conditions on \(f\) and \(Y\) that \(nU^n\) converges weakly to a non-zero process \(U\) which satisfies a certain linear stochastic differential equation.