Tensor-Parallel Emulation of Quantum Circuits with Block-Cyclic Distributed Matrix Product States

Researchers Jakub Adamski and Oliver Thomson Brown have published a groundbreaking paper introducing a tensor-parallel emulation method for quantum circuits using block-cyclic distributed matrix product states (MPS). Their approach addresses a critical gap in distributed-memory tensor network methods by expanding MPS site tensors beyond local memory constraints through a novel distribution scheme where individual dense tensors are scattered across computational indices. This represents a significant advancement over existing parallel techniques, which have been limited to direct contraction and sampling problems rather than full quantum state evolution.

The team's technical innovation centers on replacing slower singular value decomposition (SVD) with pivoted QR factorization, enabling more efficient tensor operations. They demonstrated their method's capabilities by approximately emulating Google's classically difficult random circuit sampling (RCS) benchmark, achieving unprecedented bond dimensions of 16,384 on 32 nodes of the ARCHER2 supercomputer. This performance surpasses state-of-the-art methods by three orders of magnitude in accuracy, marking a substantial leap forward in classical quantum circuit simulation.

The research has practical implications beyond benchmarks, advancing experiments involving quantum phase estimation circuits and offering a scalable, naturally load-balanced distribution formula. The approach is compatible with other types of parallelism and has the potential to enhance numerous algorithms based on dense tensor networks. By pushing the quantum-classical computational phase boundary, this work unlocks new opportunities for simulating quantum systems that were previously computationally prohibitive, potentially accelerating quantum algorithm development and verification.