1: \begin{abstract}

2: We present an algorithm for solving stochastic

3: heat equations, whose key ingredient is a non-uniform

4: time discretization of the driving Brownian motion $W$.

5: For this algorithm we derive an error bound

6: in terms of its number of evaluations of one-dimensional

7: components of $W$. The rate of convergence depends on

8: the spatial dimension of the heat equation and on the decay of the 

9: eigenfunctions of the covariance of $W$.

10: According to known lower bounds, our algorithm

11: is optimal, up to a constant, and this optimality 

12: cannot be achieved by uniform time discretizations.

13: \end{abstract}

14: