Inheritance diagram of tenpy.algorithms.tdvp

Engine(psi, model, options, **kwargs)

Deprecated old name of TDVPEngine.

H0_mixed(Lp, Rp)

Class defining the zero site Hamiltonian for Lanczos.

H1_mixed(Lp, Rp, W)

Class defining the one site Hamiltonian for Lanczos.

H2_mixed(Lp, Rp, W0, W1)

Class defining the two sites Hamiltonian for Lanczos.

TDVPEngine(psi, model, options[, environment])

Time dependent variational principle algorithm for MPS.

Module description

Time Dependant Variational Principle (TDVP) with MPS (finite version only).

The TDVP MPS algorithm was first proposed by [haegeman2011]. However the stability of the algorithm was later improved in [haegeman2016], that we are following in this implementation. The general idea of the algorithm is to project the quantum time evolution in the manyfold of MPS with a given bond dimension. Compared to e.g. TEBD, the algorithm has several advantages: e.g. it conserves the unitarity of the time evolution and the energy (for the single-site version), and it is suitable for time evolution of Hamiltonian with arbitrary long range in the form of MPOs. We have implemented the one-site formulation which does not allow for growth of the bond dimension, and the two-site algorithm which does allow the bond dimension to grow - but requires truncation as in the TEBD case.


This is still a beta version, use with care. The interface might still change.


long-term: Much of the code is similar as in DMRG. To avoid too much duplicated code, we should have a general way to sweep through an MPS and updated one or two sites, used in both cases.