QN ITensor and HDF5 Format Issue

ITensor Julia Questions
1

Dear admins and moderators,

I designed and executed the following program, but an error occurred:

function MakeVidalMPS(psi, maxdim)
    Gm=[]; Ld=[]

    U, Sv, V = svd(psi[1] * psi[2] , inds(psi[1]); maxdim=maxdim, cutoff=eps(Float64))
    Sv /= norm(Sv)
    push!(Gm, U)
    push!(Ld, Sv)
    push!(Gm, noprime(V))

    Gm[2], Sv, Gm[1] = svd(Gm[2] * Gm[1], inds(Gm[2]); maxdim=maxdim, cutoff=eps(Float64))
    Sv /= norm(Sv)
    push!(Ld, Sv)

    return Gm, Ld
end

sites = siteinds("S=1/2", 2; conserve_sz=true)
states = [isodd(n) ? "Up" : "Dn" for n in 1:2]
psi = MPS(sites,states)

Gamma, Lambda = MakeVidalMPS(psi, 100)

f_lambda = h5open("Lambd.h5", "w")
    write(f_lambda, "Lambda", Lambda[2])
close(f_lambda)

The error message was as follows:

ERROR: MethodError: no method matching strides(::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2})
The function `strides` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  strides(::StaticArraysCore.MArray)
   @ StaticArrays ~/.julia/packages/StaticArrays/MSJcA/src/abstractarray.jl:19
  strides(::LinearAlgebra.Adjoint{<:Real, <:AbstractMatrix})
   @ LinearAlgebra ~/.julia/juliaup/julia-1.11.0+0.x64.linux.gnu/share/julia/stdlib/v1.11/LinearAlgebra/src/adjtrans.jl:349
  strides(::LinearAlgebra.Adjoint{<:Real, <:AbstractVector})
   @ LinearAlgebra ~/.julia/juliaup/julia-1.11.0+0.x64.linux.gnu/share/julia/stdlib/v1.11/LinearAlgebra/src/adjtrans.jl:346
  ...

Stacktrace:
 [1] stride(A::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2}, k::Int64)
   @ Base ./abstractarray.jl:595
 [2] _check_invalid(dataset::HDF5.Dataset, buf::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2})
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:501
 [3] write_dataset(dataset::HDF5.Dataset, memtype::HDF5.Datatype, buf::NDTensors.DiagBlockSparse{…}, xfer::HDF5.DatasetTransferProperties)
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:521
 [4] write_dataset(dataset::HDF5.Dataset, memtype::HDF5.Datatype, buf::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2})
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:521
 [5] write_dataset(parent::HDF5.Group, name::String, data::NDTensors.DiagBlockSparse{Float64, Vector{…}, 2}; pv::@Kwargs{})
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:320
 [6] write_dataset(parent::HDF5.Group, name::String, data::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2})
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:315
 [7] write(parent::HDF5.Group, name::String, data::NDTensors.DiagBlockSparse{Float64, Vector{Float64}, 2}; pv::@Kwargs{})
   @ HDF5 ~/.julia/packages/HDF5/Z859u/src/datasets.jl:342
 [8] write(parent::HDF5.File, name::String, T::ITensor)
   @ ITensorsHDF5Ext ~/.julia/packages/ITensors/fUsvl/ext/ITensorsHDF5Ext/itensor.jl:11
 [9] top-level scope
   @ REPL[23]:1
Some type information was truncated. Use `show(err)` to see complete types.

I believe the issue is that the ITensor, which I want to save in the HDF5 file, has a QN. In fact, ITensors without a QN can be saved to the HDF5 file without any issues. Even though MPS has a QN, it can be saved without any problems. Why is it that ITensor cannot be saved, but MPS can?

2

It looks like we haven’t defined writing diagonal block sparse ITensors to HDF5, which is the format of the Lambda tensors. As a workaround, you could try:

write(f_lambda, "Lambda", denseblocks(Lambda[2]))
3

Thanks to your answer, it works as expected! I appreciate your response to my question very much!

4

Glad that works for now. Of course, ideally we would directly define saving and loading for DiagBlockSparse, maybe I’ll look into that this week if I have time.

5

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.