Tensor software and algorithms for quantum simulation by Edgar Solomonik, University of Illinois скачать в хорошем качестве

Tensor software and algorithms for quantum simulation by Edgar Solomonik, University of Illinois

Tensor computations are central numerical primitives for computational modeling of quantum systems. These include standard linear algebra operations on multidimensional data (tensors) as well as specialized methods for decomposition of tensors and for optimization of tensor networks. I will discuss challenges and opportunities in this area and present our work on algorithms and software in this domain. Specifically, I will introduce new methods for (1) handling symmetries in tensors, (2) numerical optimization for tensor decomposition, and (3) contraction of 2D tensor networks. On the software side, I will also describe three efforts: (1) Cyclops, a library for distributed-memory tensor computations, (2) AutoHOOT, tensor-algebra-centric library for automatic differentiation, and (3) Koala, a library for simulation of 2D quantum systems. Special focus will be given to the Koala library, which provides efficient parallel algorithms for approximate simulation of 2D quantum circuits (https://arxiv.org/abs/2006.15234).