infsiteinds with QNs and arbitrary initial state

[nbaldelli]

If I try to initialize an infiniteMPS with an arbitrary initial state, for example

N = 2 # Unit cell size
initstate(n) = isodd(n) ? "↑" : "0" #quarter filling
s = infsiteinds("Electron", N; initstate, conserve_qns = true)
ψ = InfMPS(s, initstate)

I get this error

ERROR: Dividing ITensors.QNVal(Nf, 1, -1) by 2, the resulting QN value is not an integer
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:35
  [2] /(qnval::ITensors.QNVal, n::Int64)
    @ ITensorInfiniteMPS ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:10
  [3] #103
    @ ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:17 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/StaticArrays/PUoe1/src/mapreduce.jl:75 [inlined]
  [5] _map
    @ ~/.julia/packages/StaticArrays/PUoe1/src/mapreduce.jl:40 [inlined]
  [6] map
    @ ~/.julia/packages/StaticArrays/PUoe1/src/mapreduce.jl:31 [inlined]
  [7] /
    @ ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:17 [inlined]
  [8] shift_flux_to_zero(s::Vector{Index{Vector{Pair{QN, Int64}}}}, flux::QN)
    @ ITensorInfiniteMPS ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:56
  [9] shift_flux_to_zero(s::Vector{Index{Vector{Pair{QN, Int64}}}}, initstate::Function)
    @ ITensorInfiniteMPS ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:38
 [10] infsiteinds(site_tag::String, n::Int64; translator::Function, initstate::Function, kwargs::Base.Pairs{Symbol, Bool, Tuple{Symbol}, NamedTuple{(:conserve_qns,), Tuple{Bool}}})
    @ ITensorInfiniteMPS ~/.julia/packages/ITensorInfiniteMPS/Ogcvb/src/infinitecanonicalmps.jl:64

If I understand correctly, the idea is that to not accumulate a QN flux passing from on cell to the following, the total flux of the unit cell is set to zero by calling shift_flux_to_zero that reassigns the quantum numbers by removing the average defined as total flux / unit cell size. This works smoothly if the flux of all quantum numbers is a multiple of the unit cell size.

In my previous example the total charge flux = 1 is not a multiple of 2 and the shift_flux_to_zero returns that error.

Maybe to solve the issue one can just catch these cases, multiply every single quantum number by the size of the unit cell, apply the shift to construct the MPS and restore the "physical' QNs by dividing by the unit cell size when computing observables. Would that work?

[mtfishman]

You may just have to use a larger unit cell for this case (e.g. N=4).

[nbaldelli]

In that case I would have an average flux of "Nf" per tensor of 2 / 4 and indeed if I try to set N=4 I get the same error with

ERROR: Dividing ITensors.QNVal(Nf, 2, -1) by 4, the resulting QN value is not an integer

[nbaldelli]

Keeping this open until #69 is successfully merged

[mingruyang]

#69 seems not work for me. After the fix, I still get "Flux not invariant under one unit cell translation, not implemented" error when calling InfMPS.

[mtfishman]

@mingruyang could you share a minimal example with a stacktrace as a comment in #69?

[mtfishman]

I'll close for now since it should be fixed by #69, which is now merged into main. @nbaldelli, please try it out and report any issues you might come across in a new issue.

@mingruyang, I guess the issue you report is either fixed by #69 or actually related to subspace expansion, which is being tracked and discussed in #75? If you see another issue beyond the one being tracked in #75 please report it as a new issue with a minimal reproducible code example and stacktrace.