Quantum many-body systems (QMBs) are some of the most challenging physical systems to simulate numerically. Methods involving approximations for tensor network (TN) contractions have proven to be viable alternatives to algorithms such as quantum Monte Carlo or simulated annealing. However, these methods are cumbersome, difficult to implement, and often have significant limitations in their accuracy and efficiency when considering systems in more than one dimension. In this paper, we explore the exact computation of TN contractions on two-dimensional geometries and present a heuristic improvement of TN contraction that reduces the computing time, the amount of memory, and the communication time. We run our algorithm for the Ising model using memory optimized x1.32x large instances on Amazon Web Services (AWS) Elastic Compute Cloud (EC2). Our results show that cloud computing is a viable alternative to supercomputers for this class of scientific applications.
翻译:量子体多系统(QMBs)是模拟数字的最具挑战性的物理系统(QMBs ) 。 采用粒子网络收缩近似法( TN) 已被证明是可替代计算法( 量子蒙特卡洛 ) 或模拟肛交等的可行替代方法。 然而,这些方法繁琐、难以实施,而且当考虑一个以上层面的系统时,其准确性和效率往往受到相当大的限制。 在本文中,我们探索了二维地理分布的TN收缩的精确计算,并展示了TN收缩的超常性改进,从而减少了计算时间、内存量和通信时间。 我们在亚马逊网络服务(AWS) Elatictute Clod (EC2) 上使用记忆优化的X1.32x大实例运行了我们的Ising模型算法。 我们的结果表明,云计算是这一类科学应用的超级计算机的可行替代方法。