1: \begin{abstract}

2: This paper proposes an adaptive timestep construction 

3: for an Euler-Maruyama approximation of SDEs with a drift 

4: which is not globally Lipschitz.  It is proved that 

5: if the timestep is bounded appropriately, then over

6: a finite time interval the numerical approximation is 

7: stable, and the expected number of timesteps is finite. 

8: Furthermore, the order of strong convergence is the 

9: same as usual, i.e.~order $\halfs$ for SDEs with a 

10: non-uniform globally Lipschitz volatility, and order 

11: $1$ for Langevin SDEs with unit volatility and a drift 

12: with sufficient smoothness.  The analysis is supported 

13: by numerical experiments for a variety of SDEs.

14: \end{abstract}

15: