1: \begin{abstract}

2: In a recent paper, de Forcrand has pointed out that

3: matrix inversions using Gaussian noise need not be iterated to

4: full convergence, but instead may be solved

5: approximately and treated as a heatbath. Gaussian noise

6: however is not optimal for minimizing variance. It shown here how his algorithm

7: may be generalized to a mixture of Gaussian and Z(N) noise,

8: resulting in a smaller effective variance for some operators.

9:  

10: \end{abstract}

11: