Codecov Report

Attention: 2 lines in your changes are missing coverage. Please review.

Comparison is base (56fcb93) 72.90% compared to head (1ef03c5) 72.87%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

@@            Coverage Diff             @@
##             main     #130      +/-   ##
==========================================
- Coverage   72.90%   72.87%   -0.03%     
==========================================
  Files          71       71              
  Lines        4097     4100       +3     
==========================================
+ Hits         2987     2988       +1     
- Misses       1110     1112       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

This also only works for ProjTTN of an operator with vanishing flux. The cleaner way would be to compute the flux of the environment in order to compute the flux of the tensor resulting from the contraction, allocate the resulting tensor filled with zeros, and in-place add the actual result of the contraction into it (all inside the if clause).

@b-kloss could you explain this a bit more? Under what situation does isempty(Hv) return true, and when that happens, what is the type, block structure, etc. of the storage of Hv? I think that would help to determine a fix to use.

This happens in the very first call to PH(phi) in from KrylovKit.exponentiate when the action of the projected Hamiltonian on the state vanishes. E.g. in a region of a product state that's inert under the local Hamiltonian, let's say a single domain wall initial condition for a tight-binding model in 1D on an MPS, more than two sites away from the domain-wall. If it's helpful I can include a minimal example of 2-site TDVP for fermionic systems with autofermion sign where this happens.

Hv = ITensor ord=2
Dim 1: (dim=1|id=867|"19,20")' <Out>
 1: QN("Nf",10,-1) => 1
Dim 2: (dim=2|id=441|"Fermion,Site,n=20")' <Out>
 1: QN("Nf",0,-1) => 1
 2: QN("Nf",1,-1) => 1
NDTensors.BlockSparse{Float64, Vector{Float64}, 2}
 1×2

storage(Hv) = Float64[]
Hv = ITensor ord=2
nzblocks(Hv) = Block{2}[]

Superseded by #1318, so I'll close this.