A Parallel Dmrg Strategy For A Class Of Lattice Model

The density matrix renormalization group (DMRG) is a high-precision nu-merical method for one-dimesional strongly correlated systems, but it needs long cpu time and a lot of memory for a big lattice system. It’s possible for us to study a bigger lattice system with a parallel DMRG algorithm.This paper first describes the DMRG algorithm for a class of lattice model and analyzes the mount of computation and storage. Then a parallel strategy of DMRG with all nodes loading balance and only a little communications is designed for the diagonalization process and the truncation of operators. A implementation based on hybrid OpenMP-MPI programming model shows that there are speedup and scalability for diagonalization process.