1: \begin{abstract}

2:   One way of getting insight into non-Gaussian measures, posed on

3:   infinite dimensional Hilbert spaces, is to first obtain best fit

4:   Gaussian approximations, which are more amenable to numerical

5:   approximation.  These Gaussians can then be used to accelerate

6:   sampling algorithms.  This begs the questions of how one should

7:   measure optimality and how the optimizers can be obtained.  Here, we

8:   consider the problem of minimizing the distance with respect to

9:   relative entropy.  We examine this minimization problem by seeking

10:   roots of the first variation of relative entropy, taken with respect

11:   to the mean of the Gaussian, leaving the covariance fixed. Adapting

12:   a convergence analysis of Robbins-Monro to the infinite dimensional

13:   setting, we can justify the application of this algorithm and

14:   highlight necessary assumptions to ensure its convergence.

15: \end{abstract}

16: