Regression in find_groundstate with DMRG2 and FiniteMPOHamiltonian (v0.13.11)

[DaanVrancken]

Description

After updating to MPSKit v0.13.11, running find_groundstate with DMRG2 on a finite system throws a MethodError. It appears that a Missing value is being passed into a TensorKit function during the construction of the 2-site Hamiltonian derivative. I used TensorKit v0.16.5 and MPSKitModels v0.4.6.

This code worked as expected in v0.13.10.

Reproducible Example

using MPSKit, TensorKit, MPSKitModels

# Model parameters
const t = 1.0
const U = 3.0
const mu = U / 2

# Construct Hamiltonian from Infinite MPO
H = hubbard_model(InfiniteChain(2); U, t, mu)
H_f = periodic_boundary_conditions(H, 12)

# Initialize State
Vspaces = fill(Vect[fℤ₂](0 => 10, 1 => 10), 11)
psi = FiniteMPS(physicalspace(H_f), Vspaces)

# Algorithm
alg = DMRG2(; 
    maxiter=1000, 
    trscheme=trunctol(; atol=10e-2), 
    tol=10e-5, 
    verbosity=2
)

# Throws MethodError: no method matching spacetype(::Type{Missing})
find_groundstate(psi, H_f, alg)

Stacktrace

ERROR: MethodError: no method matching spacetype(::Type{Missing})
This error has been manually thrown, explicitly, so the method may exist but be intentionally marked as unimplemented.

Stacktrace:
 [1] spacetype(::Type{Missing})
   @ TensorKit ...\src\spaces\vectorspaces.jl:398
 [2] sectortype(::Type{Missing})
   @ TensorKit ...\src\spaces\vectorspaces.jl:348
 [3] sectortype(x::Missing)
   @ TensorKit ...\src\spaces\vectorspaces.jl:347
 [4] prepare_operator!!(H::MPSKit.JordanMPO_AC2_Hamiltonian{...}, ...)
   @ MPSKit ...\src\algorithms\derivatives\hamiltonian_derivatives.jl:304
 [5] prepare_operator!!
   @ ...\src\algorithms\derivatives\derivatives.jl:232 [inlined]
 [6] AC2_hamiltonian(site::Int64, below::FiniteMPS, operator::FiniteMPOHamiltonian, ...)
   @ MPSKit ...\src\algorithms\derivatives\hamiltonian_derivatives.jl:195
...

[DaanVrancken]

Thanks, @lkdvos, for looking into this. This indeed seems to fix the minimal working example above.

Unfortunately, when I include symmetries, I seem to get a related error. Here is a new minimal example, now with MPSKit v0.13.12.

using MPSKit, TensorKit, MPSKitModels

const t = 1.0
const U = 3.0
const mu = U / 2

H_u1_su2 = hubbard_model(ComplexF64, U1Irrep, SU2Irrep, InfiniteChain(2); U, t, mu = U / 2);
charges = fill(FermionParity(1) ⊠ U1Irrep(1) ⊠ SU2Irrep(0), 2);
H_u1_su2 = MPSKit.add_physical_charge(H_u1_su2, charges);
H = periodic_boundary_conditions(H_u1_su2, 12)

pspaces = physicalspace.(Ref(H))
vspaces = accumulate(fuse, pspaces)[1:end-1]
psi = FiniteMPS(pspaces, vspaces)

alg = DMRG2(; 
           maxiter=1000, 
           trscheme=trunctol(; atol=10e-2), 
           tol=10e-5, 
           verbosity=2
        )

find_groundstate(psi, H, alg)

Stacktrace:

Stacktrace:
  [1] spacecheck_transform(f::typeof(transpose), Vdst::TensorMapSpace{GradedSpace{ProductSector{…}, TensorKit.SortedVectorDict{…}}, 3, 3}, Vsrc::TensorMapSpace{GradedSpace{ProductSector{…}, TensorKit.SortedVectorDict{…}}, 4, 2}, p::Tuple{Tuple{Int64, Int64, Int64}, Tuple{Int64, Int64, Int64}})
    @ TensorKit ...\.julia\packages\TensorKit\nhfGA\src\tensors\indexmanipulations.jl:423
  [2] spacecheck_transform
    @ ...\.julia\packages\TensorKit\nhfGA\src\tensors\indexmanipulations.jl:419 [inlined]
  [3] add_transpose!
    @ ...\.julia\packages\TensorKit\nhfGA\src\tensors\indexmanipulations.jl:503 [inlined]
  [4] planaradd!(C::TensorMap{ComplexF64, GradedSpace{…}, 3, 3, Vector{…}}, A::TensorMap{ComplexF64, GradedSpace{…}, 4, 2, Vector{…}}, p::Tuple{Tuple{…}, Tuple{…}}, α::VectorInterface.One, β::VectorInterface.One, backend::TensorOperations.DefaultBackend, allocator::TensorOperations.DefaultAllocator)
    @ TensorKit ...\.julia\packages\TensorKit\nhfGA\src\planar\planaroperations.jl:35
  [5] planaradd!
    @ ...\.julia\packages\TensorKit\nhfGA\src\planar\planaroperations.jl:9 [inlined]
  [6] planaradd!
    @ ...\.julia\packages\TensorKit\nhfGA\src\planar\planaroperations.jl:5 [inlined]
  [7] prepare_operator!!(H::MPSKit.JordanMPO_AC2_Hamiltonian{TensorMap{…}, TensorMap{…}, TensorMap{…}, MPSKit.MPODerivativeOperator{…}}, backend::TensorOperations.DefaultBackend, allocator::TensorOperations.BufferAllocator{Memory{…}})
    @ MPSKit ...\.julia\packages\MPSKit\7FTCR\src\algorithms\derivatives\hamiltonian_derivatives.jl:304
  [8] prepare_operator!!
    @ ...\.julia\packages\MPSKit\7FTCR\src\algorithms\derivatives\derivatives.jl:232 [inlined]
  [9] AC2_hamiltonian(site::Int64, below::FiniteMPS{…}, operator::FiniteMPOHamiltonian{…}, above::FiniteMPS{…}, envs::MPSKit.FiniteEnvironments{…}; prepare::Bool)
    @ MPSKit ...\.julia\packages\MPSKit\7FTCR\src\algorithms\derivatives\hamiltonian_derivatives.jl:195
...

[lkdvos]

I'll try and have a look ASAP, thanks for reporting (again)!

[lkdvos]

Apologies for taking something before getting to this, I think I spotted my mistake and hopefully this is now also fixed. Would you mind trying this out before I hastily merge this this time?

[DaanVrancken]

Yes, this works, thanks!