1: \begin{abstract}

2: 		We obtain the optimal Bayesian minimax rate for the unconstrained large covariance matrix of multivariate normal sample with mean zero,  when both the sample size, $n$, and the dimension, $p$, of the covariance matrix tend to infinity. 

3: 		Traditionally the posterior convergence rate is used to compare the frequentist asymptotic performance of priors, but defining the optimality with it is elusive. We propose a new decision theoretic framework for prior selection and define {\it Bayesian minimax rate}.

4: 		Under the proposed framework, we obtain the optimal Bayesian minimax rate for the spectral norm for all rates of $p$. We also considered Frobenius norm, Bregman divergence and squared log-determinant loss and obtain the optimal Bayesian minimax rate under certain rate conditions on $p$. A simulation study is conducted to support the theoretical results. 

5: 	\end{abstract}

6: