Improved hybrid parallel strategy for density matrix renormalization group method
Chen Fu-Zhou1, Cheng Chen1, 2, Luo Hong-Gang1, 2, †
       

The relative performance improvement of Algorithm 1 with different numbers of DMRG kept states, comparing to the previous work. Optimization on the CPU (□) increases about 20% performance for our largest D. The total improvement (°) is above 40% for D = 4096, and decreases as D increases.