Rename tdvp Functions

[emstoudenmire]

Main changes

• Rename the following functions:
  • tdvp -> alternating_update (this does not apply to the user-facing functions in tdvp.jl)
  • tdvp_step -> update_step
  • tdvp_sweep -> update_sweep
  • tdvp_local_update -> local_update
  • _tdvp_compute_nsweeps -> _compute_nsweeps
• Rename the following files:
  • tdvp_generic.jl -> alternating_update.jl
  • tdvp_step.jl -> update_step.jl
• Export the mpo function (needed for tests to pass)

List of commits

• Rename files
• Rename most tdvp functions to alternating_update (and tdvp_sweep to update_sweep, tdvp_step to update_step)
• Export the mpo function
• Define tdvp interface function taking custom solver

[mtfishman]

Thanks! A next step would be to remove the t argument from the alternating_update function since of course that doesn't make much sense for many solvers. I'll start an issue with improvements to make to alternating_update.

[emstoudenmire]

I'll look out for that issue. There's a question of how much we want to do in this PR. Id be happy to change the t argument. Another issue is whether we want to keep all 3 (?) versions of the tdvp user interface function. For the keyword args cleanup I was planning to submit a separate PR.

[mtfishman]

Yes, I wasn't suggesting doing that in this PR, I'm tracking future changes here: #60.

[mtfishman]

Looks like the function:

tdvp(solver, t, A, x; kwargs...)

is missing.

[mtfishman]

I formatted the package in #61, guess it caused some merge conflicts but hopefully they are easy to resolve.

[emstoudenmire]

I decided to just remove those failing tests, which correspond to the tdvp function allowing multiple permutations of its arguments. Thinking about it more, I think we should just have a single interface with one argument ordering.

[mtfishman]

That sounds good. So is the interface now:

tdvp(A, t, x; kwargs...) # Uses the default solver, which will be `KrylovKit.exponentiate`.
tdvp(solver, A, t, x; kwargs...) # Specify the solver

where in both cases it computes exp(At)|x⟩ (i.e. evolving to total time t), and the step size is determined from the keyword arguments?

[emstoudenmire]

Yes, correct that's the interface now following this PR. I did happen to leave the other "permuted" overloads of tdvp without a solver. How about I go ahead and remove those also in this PR? So then it would be only the ones you said and the others would error (not be defined).

[mtfishman]

Yeah, let's remove those as well.

[emstoudenmire]

(Seeing some test errors about is_tree which I think are unrelated to this PR.)

[mtfishman]

Those will be fixed by #65.

[emstoudenmire]

Great, I thought so, thanks for clarifying.

[mtfishman]

Ready to merge? Looks like the tests were fixed by ITensor/NamedGraphs.jl#31.

[emstoudenmire]

Yea this one’s ready to merge, thanks. Then I’ll keep working my way through the items in the issue you opened.

On Mon, Feb 13, 2023 at 12:49 PM Matt Fishman ***@***.***> wrote: Ready to merge? Looks like the tests were fixed by ITensor/NamedGraphs.jl#31 <ITensor/NamedGraphs.jl#31>. — Reply to this email directly, view it on GitHub <#59 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHJ3RKG2VYSXWSYOSUEASDWXJXZFANCNFSM6AAAAAAUOMYXJQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- -=Miles Stoudenmire=- ***@***.*** ***@***.*** http://itensor.org/miles/