Hello,
I have a question regarding the ITensorInfiniteMPS package. I understand it is not fully complete yet, but from reading this forum I got the impression that parts of it could already be used.
I tried running the finite_mps_to_infinite_mps.jl example, but it spits an error ERROR: v₁ᴿᴺ not hermitian after the call to orthogonalize. Perhaps there is a simple fix to make it work again?
I am trying to understand the logic behind the infinitemps_approx function in the src/infinitemps_approx.jl file. Is it described in some paper perhaps or are there some diagrams somewhere? From going trough the code, here is my understanding:
we start from creating a random guess for the infinite MPS. Then on lines 62,69 we compute the dominant eigenvectors of the mixed transfer matrix between the guess infinite MPS and the finite region of the finite MPS.
Then, lines 82-87 compute the partial contraction between these boundary matrices and the partial overlap of the finite and infinite MPS. It seems that the resulting tensors should be of the same shape as the boundary matrices (L0, RN). But here is where I am getting confused, because based on this, I would assume that inside the loop beginning on line 92, the C1,C2 (lines 94,95) should be scalars since they correspond to the contraction of these left and right boundary matrices. But this does not make sense given the rest of the code.
Thanks