DMRG with Subspace Expansion on Symmetry-Protected Tensor Networks (C. Hubig)
- Datum: 23.03.2017
- Uhrzeit: 14:00 - 15:00
- Vortragende(r): Claudius Hubig
- Department für Physik, Ludwig-Maximilians-Universität München
- Raum: Seminar room Theory Division - B 2.46
- Gastgeber: MPQ, Theory Division
In this talk, I will discuss two separate methods which can be used to improve the computational efficiency of DMRG and related methods on matrix-product states and beyond. The first component is the implementation of both abelian and non-abelian symmetries in an entirely general way suitable also for higher-rank tensors as encountered in e.g.tree tensor network states. The second ingredient, the subspace expansion, allows for a fully single- site DMRG algorithm with favourable linear scaling in the local dimension of the tensor network. Even for common problems, this results in a considerable speed-up over the traditional two-site DMRG method or the density matrix perturbation approach for ground-state search at reduced algorithmic complexity. Additionally, the subspace expansion can potentially be used in a large set of other algorithms, such as the TDVP or the variational application of a matrix-product operator onto a matrix-product state.