2023-12-10, 15:30–16:00 (Asia/Taipei), NYCU
In this talk, we will introduce tensor network and our tensor network library Cytnx. We first introduce the graphical tensor notation and provide simple examples showing how to use Cytnx to implement a tensor contraction. We then explain in detail the basic elements in Cytnx, and how to perform tensor contractions and tensor decompositions. We then show benchmark the performance of Cytnx
with another tensor library ITensor. In the end we will give a summary and discuss future
goals of the library.
Cytnx (pronounced as sci-tens) is a tensor network library designed for classical/quan-
tum physics simulations. It supports C++ and Python with almost identical interface
and syntax, such that users can effortlessly switch between the two languages. Aiming
at a quick learning process for new users of tensor network algorithms, the interfaces
resemble the popular scientific libraries such as numpy, Scipy, and PyTorch. Symmetries
present in physical systems can be easily defined and implemented in tensors. In ad-
dition, we provide a useful tool called Network that allows users to store large tensor
networks and perform the contractions in an optimal order that can be automatically
computed. Tensor calculations can be done on both CPUs and GPUs. We benchmark
against a well-known tensor library ITensor and see a similar performance for both li-
braries.
As questions at slido
Yes, previous knowledge expected
Language –Mandarin talk w. English slides
I'm currently a graduate student from Department of Physics, National Taiwan University, studying strongly correalated system using tensor network methods.