Hi, I have a question about using TDVP:
I have a two-level system (let’s say site 1, 2) connected to some external system. Now there’s some on-site interaction \sim U n_in_1 on site 1, and nothing on site 2. There’s no hopping between 1 and 2 (or between the two-level and the external system), so the on-site densities should remain strictly fixed.
However in practice I’ve found that as the time evolution goes on, there would be some small oscillation in the densities ( 10^{-4} < \Delta \langle n_1\rangle < 10^{-3} ) with some given bond dimension. When I quadrupled the bond dimension, this small oscillation goes away. So this is most probably some numerical errors from the projector.
My question is, is there any other way to hopefully contain this error without going to a larger bond dimension (which is expensive)? What are some probably knobs I can tune in the tdvp() algorithm to try to see where the problem is?