Fix 2d heisenberg example and add error for bad unit cell inputs

[ryanlevy]

As reported in https://itensor.discourse.group/t/vumps-example-2d-heisenberg-model-interaction-term/2052?u=ryanlevy

The 2D heisenberg example misses some operators, which is fixed in this version. I've also included a warning to help users catch this in the future, happy to take suggestions on alternative "fixes"

[ryanlevy]

I'm unable to replicate the errors from the CI at the moment (on 1.10.4), all tests pass locally for me

[mtfishman]

It looks like it may be specific to Julia 1.11, it was just released and the CI started testing against that automatically.

[mtfishman]

Here is a minimal code that reproduces the test failure in Julia 1.11:

using ITensors: ITensor, Index
i = Index(2)
a = reshape([ITensor(), ITensor(i)], 1, 2)
b = reshape([ITensor(2.0), ITensor(2.0, i)], 2, 1)
a * b

which leads to an error:

ERROR: StackOverflowError:
Stacktrace:
      [1] promote_result(::Type, ::Type, ::Type{NDTensors.EmptyNumber}, ::Type{Bool})
        @ Base ./promotion.jl:338
      [2] promote_type
        @ ./promotion.jl:318 [inlined]--- the above 2 lines are repeated 79982 more times ---

so it is related to performing matrix multiplication of matrices of ITensors where some of the ITensors are not initialized.

[mtfishman]

It looks like there were a few missing promote_rule definitions for EmptyNumber, defining the following:

using NDTensors: EmptyNumber
Base.promote_rule(::Type{T}, ::Type{EmptyNumber}) where {T<:Number} = T
Base.promote_rule(::Type{Bool}, ::Type{EmptyNumber}) = Bool

seems to fix the issue. I'll make a PR to NDTensors.jl.

[mtfishman]

Here are more minimal examples that error with stack overflows without those new definitions:

promote_type(Bool, EmptyNumber)
promote_type(EmptyNumber, Bool)

[ryanlevy]

Thanks @mtfishman ! I can check things in InfiniteMPS once that's sorted (there looks like some deprecated Vararg calls I can open a PR for later)

[mtfishman]

ITensor/ITensors.jl#1541 is now merged, that should fix the test failures in Julia 1.11.

[mtfishman]

Looks good, thanks Ryan! I'll merge once tests pass.