The paper introduces a new sparse estimator of the precision matrix for high dimensional multivariate longitudinal data. Based on block Cholesky decomposition, we impose a banded block structure on the Cholesky factor and sparsity on the innovation variance matrices using two penalties. An efficient alternative convex optimization algorithm is developed using ADMM algorithms. The resulting estimators are shown to converge in an optimal rate of Frobenius norm and the exact bandwidth recovery is established for the precision matrix. Simulations show that the proposed estimator outperforms its competitors.