Inconsistent normalization when computing expectation value

[20akshay00]

    D = 1 # bond dimension 
    N = 2 # physical space

    operator = TensorMap(rand, Float64, ℂ^N ← ℂ^N)

    # un-normalized random state
    state = FiniteMPS([TensorMap(rand, ComplexF64, ℂ^D ⊗ ℂ^N ← ℂ^D), TensorMap(rand, ComplexF64, ℂ^D ⊗ ℂ^N ← ℂ^D)])

    val1 = expectation_value(state, 1 => operator) + expectation_value(state, 2 => operator)
    val2 = expectation_value(state, @mpoham sum(operator{i} for i in vertices(FiniteChain(2))))

    val1 ≈ val2 # false 

    # normalized random state
    state = FiniteMPS([TensorMap(rand, ComplexF64, ℂ^D ⊗ ℂ^N ← ℂ^D), TensorMap(rand, ComplexF64, ℂ^D ⊗ ℂ^N ← ℂ^D)], normalize = true)

    val1 = expectation_value(state, 1 => operator) + expectation_value(state, 2 => operator)
    val2 = expectation_value(state, @mpoham sum(operator{i} for i in vertices(FiniteChain(2))))

    val1 ≈ val2 # true

expectation_value seems to normalize the state if the operator acts on all sites (or if its an MPOHamiltonian?), but not if the operator is local. Seems like an edge case that isn't encountered usually and I don't know if its intentional, but probably should be documented.

[lkdvos]

Ah, I think it should always normalize, otherwise you should call dot. Thanks for bringing it up!