English

Automatic Differentiation for Second Renormalization of Tensor Networks

Strongly Correlated Electrons 2020-07-07 v2 Computational Physics

Abstract

Tensor renormalization group (TRG) constitutes an important methodology for accurate simulations of strongly correlated lattice models. Facilitated by the automatic differentiation technique widely used in deep learning, we propose a uniform framework of differentiable TRG (\partialTRG) that can be applied to improve various TRG methods, in an automatic fashion. Essentially, \partialTRG systematically extends the concept of second renormalization [PRL 103, 160601 (2009)] where the tensor environment is computed recursively in the backward iteration, in the sense that given the forward process of TRG, \partialTRG automatically finds the gradient through backpropagation, with which one can deeply "train" the tensor networks. We benchmark \partialTRG in solving the square-lattice Ising model, and demonstrate its power by simulating one- and two-dimensional quantum systems at finite temperature. The deep optimization as well as GPU acceleration renders \partialTRG manybody simulations with high efficiency and accuracy.

Keywords

Cite

@article{arxiv.1912.02780,
  title  = {Automatic Differentiation for Second Renormalization of Tensor Networks},
  author = {Bin-Bin Chen and Yuan Gao and Yi-Bin Guo and Yuzhi Liu and Hui-Hai Zhao and Hai-Jun Liao and Lei Wang and Tao Xiang and Wei Li and Z. Y. Xie},
  journal= {arXiv preprint arXiv:1912.02780},
  year   = {2020}
}

Comments

13 pages, 9 figures

R2 v1 2026-06-23T12:37:19.314Z