converges in tdvp function

ITensor Julia Questions
, ,
1

Hey all,
I am trying to do an non-unitary time evolution to my system and I am running with a warning:
“warning: applyexp not converged in 30 steps” after using the tdvp function. Is there any limitation about the delta t we can put in there ?

using ITensors
            ψ_τ= tdvp(H_syk,ψ_τ,-m)
            normalize!(ψ_τ)

where m can range from 1 to 10. I know that this is a very brute delta t, but I wanted to know if there is any way that I can time developing with a large m’s (even larger than that one) ?

2

You could try the exponentiate solver backend by including the keyword argument solver_backend="exponentiate". That will use the exponentiate function from KrylovKit.jl which is more sophisticated and uses restarts to allow you to reach longer times. Of course, if you use too large of a time step you will have higher projection error/Trotter error.

3

Hi. I am using C++ version of TDVP, is there similar things like solver_backend=“exponentiate” that exist there? I am just confused what that warning implies! Because I have tried my TDVP code with increasing sweeps.niter(), but that warning does not go away.

4

Sorry to hear you’re having an issue. Since this is an older post (from February) can you please post a new topic (you can still link to this one) and tell us more information, such as what time step size you are using?

5

Thank you for your reply. I have created a new topic for my query in the link.