How to use a previously DMRG calculated state as starting point for a new H?

Question

Is it possible to read a state (IQMPS) calculated with one Hamiltonian, and then load it as an initial guess for a DMRG with a slightly different Hamiltonian?

If I try to simply load the IQMPS and apply the new Hamiltonian in the DMRG function, I get an error that they don't match.

Answer 1

Good question. The Hamiltonian MPO and your MPS (or IQMPO and IQMPS) must have the same "site" or "physical" indices in order to be compatible. This is one of the reasons there is a separate site set object, such as "SpinHalf" or "Hubbard" to keep track of the site indices.

So to answer your question, you will need to write your site set object (usually called "sites" in the sample codes) to disk alongside your IQMPS.

Here is some sample code showing how to do that:
http://itensor.org/docs.cgi?page=formulas/readwrite_mps

Then you need to read your site set back in before making your new Hamiltonian, and use this previous site set when you initialize (construct) the AutoMPO object. This will ensure that the Hamiltonian has the same sites as your MPS.

If that doesn't solve your problem please let me know and we can troubleshoot further.

Best regards,
Miles

Comments

Thank you - now I understand. The following works (e.g. for a Hubbard system):

    auto sites Hubbard(N);
    readFromFile(siteFile, sites);
    IQMPS psi(sites);
    readFromFile(stateFile, psi);

Then I can set up the Hamiltonian and do DMRG. (Does code format not work in comments?)

---

Great. As a bit of a refinement, what I often do in my codes is to use the function fileExists(siteFile) which returns true if there is a file named siteFile in the current directory. Then if the file is present, I read in the site set; otherwise I create a new one. Up to you, but I find this is a nice way to work.