1: \begin{abstract}

2: 

3: The understanding of adaptive algorithms for SDEs is an open

4: area where many issues related to both convergence and stability (long time

5: behaviour) of algorithms are unresolved. This paper considers a very simple

6: adaptive algorithm, based on controlling only the drift component

7: of a time-step. Both convergence and stability are studied.

8: 

9: The primary issue in the convergence analysis is that the

10: adaptive method does not necessarily drive the time-steps to zero

11: with the user-input tolerance. This possibility must be quantified and

12: shown to have low probability.

13:   

14: The primary issue in the stability analysis is ergodicity.

15:   It is assumed that the noise is

16:   non-degenerate, so that the diffusion process is elliptic, and the

17:   drift is assumed to satisfy a coercivity condition. The SDE is

18:   then geometrically ergodic (averages converge to statistical equilibrium

19:   exponentially quickly). If the drift is not linearly

20:   bounded then explicit fixed time-step approximations, such as the

21:   Euler-Maruyama scheme, may fail to be ergodic.  In this work, it is

22:   shown that the simple adaptive time-stepping strategy cures this

23:   problem.  In addition to proving ergodicity,

24:   an exponential moment bound is also proved,

25: generalizing a result known to hold for the SDE itself.

26: 

27: \vspace{0.2in}

28: 

29: \noindent{\sc Key Words:} 

30: Stochastic Differential Equations,  Adaptive Time-Discretization,

31: Convergence, Stability, Ergodicity, Exponential Moment Bounds.

32: \end{abstract}

33: