Various small fixes and improvements to RandExpand

[lkdvos]

• The changebonds(state, ::RandExpand) algorithm now no longer needs to multiply and decompose a random twosite state, and instead generates the correct vectors directly.
• The dynamic tolerances now properly handle iter = 0
• The InfiniteMPOHamiltonian environments now properly clamp the Krylov dimension if the MPS bond dimension is low.

[lkdvos]

@AFeuerpfeil, tagging you here

[codecov]

Codecov Report

❌ Patch coverage is 58.13953% with 18 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/states/abstractmps.jl	21.73%	18 Missing ⚠️

Files with missing lines	Coverage Δ
src/algorithms/changebonds/randexpand.jl	100.00% <100.00%> (ø)
src/environments/abstract_envs.jl	92.10% <100.00%> (+0.21%)	⬆️
src/utility/dynamictols.jl	75.00% <100.00%> (+2.27%)	⬆️
src/states/abstractmps.jl	51.76% <21.73%> (-13.39%)	⬇️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[AFeuerpfeil]

Hi Lukas,

Thanks, the subspace expansion itself looks good now.

Calculating the environments currently errors:

nested task error: SpaceMismatch("ProductSpace(⊞(Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 6, 3)=>1))) ≠ ProductSpace(⊞(Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 0, 3)=>1)))")

I also identified the reason: When calling right_virtualspace in the initialize_environments, it constructs the right virtualspace of AL and not AR, which are different, as you are using different expansions for the two:

right_virtualspace(ket, i) = Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 0, 3)=>1)

TensorKit.space(ket.AL[i]) = (Rep[U₁ × Cyclic{36} × U₁]((1, 35, 1)=>1, (4, 29, 4)=>1) ⊗ Rep[U₁ × Cyclic{36} × U₁]((-1, 1, -1)=>1, (2, 1, 2)=>1)) ← Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 0, 3)=>1)

TensorKit.space(ket.AR[i]) = (Rep[U₁ × Cyclic{36} × U₁]((1, 35, 1)=>1, (-2, 35, -2)=>1) ⊗ Rep[U₁ × Cyclic{36} × U₁]((-1, 1, -1)=>1, (2, 1, 2)=>1)) ← Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 6, 3)=>1)

[lkdvos]

Thanks for checking and investigating! I pushed some more changes that should hopefully resolve the issue, it would be nice if you could check again!

[AFeuerpfeil]

Sure, I will look into that today.

[AFeuerpfeil]

I have 2 comments:

1. In your rewrite, AR' is only a Vector{TensorMap}, in order to fix that I currently use return _expand!(ψ, AL′, PeriodicVector(AR′)).
2. I still get a very similar error:

ERROR: TaskFailedException

    nested task error: SpaceMismatch("ProductSpace(⊞(Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1))) ≠ ProductSpace(⊞(Rep[U₁ × Cyclic{36} × U₁]((0, 0, 0)=>1, (3, 0, 3)=>1)))")

[AFeuerpfeil]

I observe some peculiarities with the truncscheme and when subsequently running VUMPS get this error

[ Info: VUMPS init:     obj = +2.216860652296e+03       err = 0.0000e+00
┌ Warning: Constructing an MPS from tensors that are not full rank
└ @ MPSKit ~/.julia/dev/MPSKit/src/states/infinitemps.jl:191
ERROR: SpaceMismatch("Incompatible spaces for permuting")
Stacktrace:
  [1] TensorKit.TreeTransformer(transform::Function, p::Tuple{…}, Vdst::TensorMapSpace{…}, Vsrc::TensorMapSpace{…})
    @ TensorKit ~/.julia/packages/TensorKit/JuisY/src/tensors/treetransformers.jl:161
  [2] _treetransposer
    @ ~/.julia/packages/TensorKit/JuisY/src/tensors/treetransformers.jl:199 [inlined]
  [3] #treetransposer##5
    @ ~/.julia/packages/TensorKit/JuisY/src/auxiliary/caches.jl:113 [inlined]
  [4] get!(default::TensorKit.var"#treetransposer##5#treetransposer##6"{…}, lru::LRUCache.LRU{…}, key::Tuple{…})
    @ LRUCache ~/.julia/packages/LRUCache/ZH7qB/src/LRUCache.jl:169
  [5] treetransposer(Vdst::TensorMapSpace{…}, Vsrc::TensorMapSpace{…}, p::Tuple{…}, ::TensorKit.GlobalLRUCache)
    @ TensorKit ./none:-1
  [6] treetransposer
    @ ./none:-1 [inlined]
  [7] treetransposer
    @ ~/.julia/packages/TensorKit/JuisY/src/tensors/treetransformers.jl:193 [inlined]
  [8] add_transpose!
    @ ~/.julia/packages/TensorKit/JuisY/src/tensors/indexmanipulations.jl:450 [inlined]
  [9] transpose!
    @ ~/.julia/packages/TensorKit/JuisY/src/tensors/indexmanipulations.jl:181 [inlined]
 [10] repartition!(tdst::TensorMap{…}, tsrc::TensorMap{…})
    @ TensorKit ~/.julia/packages/TensorKit/JuisY/src/tensors/indexmanipulations.jl:238
 [11] gauge_orth_step!(it::MPSKit.IterativeSolver{…}, state::@NamedTuple{…})
    @ MPSKit ~/.julia/dev/MPSKit/src/states/ortho.jl:312
 [12] iterate(it::MPSKit.IterativeSolver{…}, state::@NamedTuple{…})
    @ MPSKit ~/.julia/dev/MPSKit/src/states/ortho.jl:287
 [13] iterate(it::MPSKit.IterativeSolver{MPSKit.RightCanonical, @NamedTuple{…}})
    @ MPSKit ~/.julia/dev/MPSKit/src/states/ortho.jl:286
 [14] (::MPSKit.var"#uniform_rightorth!##0#uniform_rightorth!##1"{…})()
    @ MPSKit ~/.julia/dev/MPSKit/src/states/ortho.jl:271
 [15] #30
    @ ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:103 [inlined]
 [16] with_logstate(f::LoggingExtras.var"#30#31"{…}, logstate::Base.CoreLogging.LogState)
    @ Base.CoreLogging ./logging/logging.jl:540
 [17] with_logger
    @ ./logging/logging.jl:651 [inlined]
 [18] withlevel(f::MPSKit.var"#uniform_rightorth!##0#uniform_rightorth!##1"{…}, level::Base.CoreLogging.LogLevel; verbosity::Int64)
    @ LoggingExtras ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:99
 [19] withlevel (repeats 2 times)
    @ ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:98 [inlined]
 [20] uniform_rightorth!(::Tuple{…}, A::PeriodicVector{…}, C₀::TensorMap{…}, alg::MPSKit.RightCanonical)
    @ MPSKit ~/.julia/dev/MPSKit/src/states/ortho.jl:262
 [21] gaugefix!
    @ ~/.julia/dev/MPSKit/src/states/ortho.jl:143 [inlined]
 [22] #gaugefix!#105
    @ ~/.julia/dev/MPSKit/src/states/ortho.jl:122 [inlined]
 [23] gaugefix!
    @ ~/.julia/dev/MPSKit/src/states/ortho.jl:111 [inlined]
 [24] InfiniteMPS(AL::PeriodicVector{…}, C₀::TensorMap{…}; kwargs::@Kwargs{…})
    @ MPSKit ~/.julia/dev/MPSKit/src/states/infinitemps.jl:202
 [25] gauge_step!(it::MPSKit.IterativeSolver{…}, state::MPSKit.VUMPSState{…}, ACs::PeriodicVector{…})
    @ MPSKit ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:156
 [26] iterate
    @ ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:89 [inlined]
 [27] iterate
    @ ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:88 [inlined]
 [28] (::MPSKit.var"#459#460"{…})()
    @ MPSKit ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:70
 [29] #30
    @ ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:103 [inlined]
 [30] with_logstate(f::LoggingExtras.var"#30#31"{MPSKit.var"#459#460"{…}}, logstate::Base.CoreLogging.LogState)
    @ Base.CoreLogging ./logging/logging.jl:540
 [31] with_logger
    @ ./logging/logging.jl:651 [inlined]
 [32] withlevel(f::MPSKit.var"#459#460"{…}, level::Base.CoreLogging.LogLevel; verbosity::Int64)
    @ LoggingExtras ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:99
 [33] withlevel (repeats 2 times)
    @ ~/.julia/packages/LoggingExtras/VLO3o/src/verbosity.jl:98 [inlined]
 [34] dominant_eigsolve(operator::InfiniteMPOHamiltonian{…}, mps::InfiniteMPS{…}, alg::VUMPS{…}, envs::MPSKit.InfiniteEnvironments{…}; which::Symbol)
    @ MPSKit ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:67
 [35] dominant_eigsolve
    @ ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:54 [inlined]
 [36] find_groundstate
    @ ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:51 [inlined]
 [37] find_groundstate(mps::InfiniteMPS{…}, operator::InfiniteMPOHamiltonian{…}, alg::VUMPS{…})
    @ MPSKit ~/.julia/dev/MPSKit/src/algorithms/groundstate/vumps.jl:51
 [38] top-level scope
    @ ~/.julia/dev/ITensorTensorKitInterface/scripts/benchmarks/ergodicity.jl:93
Some type information was truncated. Use `show(err)` to see complete types.

[AFeuerpfeil]

Thanks. The error is gone, but at least in my examples the GMRES for the environment does not converge after expanding leading to weird errors and Lapack Exceptions, when the local update is ill-defined.
Let's discuss that tomorrow, if you have some time in the afternoon.

[lkdvos]

@AFeuerpfeil as we discussed I abandoned this approach, after realizing that the proper way of doing this is to sequentially expand and sweep through the unit cell. I will open a new PR for that.

[AFeuerpfeil]

Hi Lukas,
OK. Does that mean that our idea to add some noise to the expanded isometries did not work?

[lkdvos]

I'm currently trying that out, but I felt like at that point it no longer fits the changes I'm doing to changebonds so I didn't want to mix this in here. (Basically I want to still merge the small improvements I have here but start fresh on the other ideas)

[AFeuerpfeil]

I agree.
Even if our approach works, for most applications like nearest-neighbour interactions, this 2-site expansion is still useful and does not change the state in contrast to our idea. Therefore, it is reasonable to keep it and for our idea to overload a new function like approx_changebonds or so...

[VictorVanthilt]

G2G for me!