1: \begin{abstract}

2: The Poisson model is frequently employed to describe count data, but in a Bayesian context it leads to

3: an analytically intractable posterior probability distribution. In this work, we analyze a variational Gaussian

4: approximation to the posterior distribution arising from the Poisson model with a Gaussian prior. This is

5: achieved by seeking an optimal Gaussian distribution minimizing the Kullback-Leibler divergence from

6:  the posterior distribution to the approximation, or

7: equivalently maximizing the lower bound for the model evidence. We derive an explicit expression for

8: the lower bound, and show the existence and uniqueness of the optimal Gaussian approximation. The lower

9: bound functional can be viewed as a variant of classical Tikhonov regularization that penalizes also the

10: covariance. Then we develop an efficient alternating direction maximization algorithm for solving

11: the optimization problem, and analyze its convergence. We discuss strategies for reducing the computational

12: complexity via low rank structure of the forward operator and the sparsity of the covariance. Further, as an

13: application of the lower bound, we discuss hierarchical Bayesian modeling for selecting the

14: hyperparameter in the prior distribution, and propose a monotonically convergent algorithm for determining

15: the hyperparameter. We present extensive numerical experiments to illustrate the Gaussian approximation and the algorithms.\\

16: {\bf Keywords}: Poisson data, variational Gaussian approximation,  Kullback-Leibler divergence, alternating direction maximization,

17: hierarchical modeling

18: \end{abstract}