Towards overcoming the entanglement barrier when simulating long-time evolution

Quantum many-body systems out of equilibrium pose some of the most intriguing questions in physics. Unfortunately, numerically keeping track of time evolution of states under Hamiltonian dynamics constitutes a severe challenge for all known methods. Prominently, tensor network methods are marred by an entanglement blowup, which allows to simulate systems following global quenches only to constant time. In this work, we present a scheme that allows to significantly extend the simulation time for interacting fermionic or equivalent spin systems. In the past when keeping track of evolution in one-dimensional real space, the subspace parametrised by real-space matrix product states satisfying area laws - often dubbed the "physical corner" of Hilbert space - was chosen as variational set. In contrast, if the manifold containing both tensor network states and fermionic mode transformations is chosen, significantly longer times can be achieved. We argue and our results suggest that in many cases it is genuine correlations between modes that is the actual limiting factor: The system at hand is for intermediate times contained in the "physical corner", but a different one than what is commonly assumed.