The peps-torch performs gradient optimization of two-dimensional iPEPS tensor networks. It is primarily composed of two parts: various iPEPS defined through one of the iPEPS classes and their environments computed by the corner-transfer matrix algorithm.
In order to optimize an iPEPS define a loss function which computes the variational energy with respect to target Hamiltonian. Generally, to evaluate the energy, first compute the environment of iPEPS and then the appropriate reduced density matrices. Afterwards, the individual terms of Hamiltonian can be computed with these reduced density matrices. There are several examples of energy computation for different spin models, e.g. spin=2 AKLT model. Finally, invoke the optimization.
- Abelian-symmetric iPEPS
- Linear Algebra