ITensorTDVP.jl v0.4

[mtfishman]

Closes #14 and closes #35.

Some breaking changes and new features:

• Remove tdvp(t, operator, state; kwargs...) and tdvp(operator, state, t; kwargs...).
• tdvp(operator, t, state; kwargs...) now interprets t as the total time, i.e. outputs the approximation of apply(exp(t * operator), state). Before, it was interpreting t as the time step, which was confusing and also inconsistent with the docstring for tdvp. It also brings the API for ITensorTDVP.tdvp/ITensorMPS.tdvp closer to that of KrylovKit.exponentiate, and my plan would be to rename tdvp to exponentiate in the future. To set the tdvp time step, the recommended API will now be either tdvp(operator, t, state; nsteps, kwargs...) to set a time step of t / nsteps or tdvp(operator, t, state; time_step, kwargs...) which will set nsteps = Int(t / time_step). nsweeps is kept as an alias for nsteps for backwards compatibility and consistency with other solvers like dmrg, linsolve, etc.
• The time step is now removed as an argument to the local solvers and moved to a keyword argument time_step.
• Observer keyword psi changed to state.
• Drop support for ITensor AbstractOberver interface.
• Syntax for passing solver/updater kwargs is changed from tdvp(operator, t, init; nsteps, cutoff, solver_tol=1e-12) to tdvp(operator, t, init; nsteps, cutoff, updater_kwargs=(; tol=1e-12)) (same for ITensorTDVP.dmrg). Syntax tries to match ITensorNetworks solvers as much as possible¹².
• Specifying a custom local updater/solver with tdvp(operator, t, init; nsteps, cutoff, updater=...) (same for ITensorTDVP.dmrg, ITensorTDVP.dmrg_x, and ITensorTDVP.linsolve). Syntax tries to match ITensorNetworks solvers as much as possible¹². Before a solver/updater could be passed as the first argument to certain solvers like tdvp but now support is consistent across all solvers through the updater keyword argument.

To-do:

• In contract(::Algorithm"fit", operator::MPO, state::MPS; init, kwargs...), require that init has the unique indices of operator (i.e. for standard MPO and MPS they should be primed).
• Make a News item on the README.
• Make sure the examples are using the latest recommended syntax.
• Make an announcement on ITensor discourse and Slack.
• Make a new breaking release of ITensorMPS.jl v0.2 using ITensorTDVP.jl v0.4.

@emstoudenmire

Footnotes

1. https://github.com/ITensor/ITensorNetworks.jl/blob/v0.11.6/test/test_treetensornetworks/test_solvers/test_tdvp.jl ↩ ↩²

2. https://github.com/ITensor/ITensorNetworks.jl/blob/v0.11.6/test/test_treetensornetworks/test_solvers/test_dmrg.jl ↩ ↩²

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 91.78886% with 28 lines in your changes are missing coverage. Please review.

Project coverage is 87.56%. Comparing base (21db181) to head (a97489e).
Report is 1 commits behind head on main.

Files	Patch %	Lines
src/reducedconstantterm.jl	80.32%	12 Missing ⚠️
src/tdvp.jl	74.07%	7 Missing ⚠️
src/reducedlinearproblem.jl	85.71%	4 Missing ⚠️
src/sweep_update.jl	97.67%	2 Missing ⚠️
src/timedependentsum.jl	91.66%	2 Missing ⚠️
src/alternating_update.jl	94.73%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #74      +/-   ##
==========================================
+ Coverage   85.99%   87.56%   +1.57%     
==========================================
  Files          19       18       -1     
  Lines         621      571      -50     
==========================================
- Hits          534      500      -34     
+ Misses         87       71      -16     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.