Some problems about MPI Parallelism

Dear ITensor team,
Thanks for your reply and thanks for the wonderful library about TensorNetwork. Recently, i perform some DMRG calculations with MPI parallelism. I have a question about MPI parallelism

I am using the latest version of ITensor c++ and i find some documents about how to MPI parallelism

  1. ITensor
  2. ITensor
    i use the second method since it’s simple to realized. BUT i find i can’t realized MPI parallelism when the model don’t use QNs (MPI parallelism work when model use QNs ). So my question is : is the MPI parallelism only work for the model which use QNs? if no, can you provide the simple way to perform MPI parallelism for QNs not conserve calculation ? Many thanks!
    Best regards,

Hi, the MPI parallelism tools (MailBox and related classes) is just for sending data like ITensors and other objects between nodes. Therefore it does not automatically provide an MPI-parallel implementation of DMRG or other algorithms, and those have to be written on top. Does that answer your question? If not, could you be more precise about what you’d like to do?

When you say you can’t realize MPI parallelism when the model doesn’t use QNs, did you mean that you tried sending a dense ITensor between nodes and it didn’t work? What did you try and could you provide a small code example?


Hi Miles,
Thanks for your reply. I have solved the problem by reinstall intensor in my lab. Maybe the problem is due to install issues, anyway, thank you very much !
Best regards,

1 Like

Thanks for letting us know! Glad you figured it out.