Open benjaminvillalonga opened 4 years ago
The only major blocker I see for this is that initial ordering of tensor indices (which uses the contraction ordering) is currently part of the flatten_grid_of_tensors
operation. If we can separate those steps, it should be possible to reuse the grid for multiple contraction orderings.
See #267.
In some cases the tensor network might take long compared to contracting it. In those cases, it makes sense to compute several amplitudes for a single build.