Source-linked AI summary
Alibaba Cloud Quantum Development Platform: Large-Scale Classical Simulation of Quantum Circuits
Fang Zhang, Cupjin Huang, Michael Newman, Junjie Cai, Huanjun Yu, Zhengxiong Tian, Bo Yuan, Haihong Xu, Junyin Wu, Xun Gao, Jianxin Chen, Mario Szegedy, Yaoyun Shi
TL;DR
The paper studies classical simulation of revised random quantum circuits, whose simulation difficulty has motivated prior benchmarking. It uses a refactored tensor-network simulator deployed across Alibaba Cloud resources and reports distributional and scaling results for Bristlecone circuits.
Problem
Revised random quantum circuits are substantially harder to simulate than earlier circuits and require benchmarking on large computing systems.
Method
The authors deploy a refactored tensor-network simulator across Alibaba Cloud ECS instances, using preprocessing and distributed subtasks coordinated through OSS.
Results
The calculated Bristlecone-70 amplitude distribution closely matches the Porter-Thomas form, while larger batches reduce execution time per amplitude until runtime stabilizes.
Takeaways & Limitations
The experiments support distributed tensor-network simulation of revised random circuits and show that workload size and instance granularity affect efficiency.
Takeaways & Limitations
Bristlecone-70 and Bristlecone-72 are treated as equivalent for simulation because two qubits in the latter can be contracted with their only neighbor.
Abstract
from arXiv · showhide
We report, in a sequence of notes, our work on the Alibaba Cloud Quantum Development Platform(AC-QDP). AC-QDP provides a set of tools for aiding the development of both quantum computing algorithms and quantum processors, and is powered by a large-scale classical simulator deployed on Alibaba Cloud. In this note, we report the computational experiments demonstrating the classical simulation capability of AC-QDP. We use as a benchmark the random quantum circuits designed for Google's Bristlecone QPU {\cite{GRCS}}. We simulate Bristlecone-70 circuits with depth $1 + 32 + 1$ in $0.43$ second per amplitude, using $1449$ Alibaba Cloud Elastic Computing Service (ECS) instances, each with $88$ Intel Xeon(Skylake) Platinum 8163 vCPU cores @ 2.5 GHz and $160$ gigabytes of memory. By comparison, the previously best reported results for the same tasks are $104$ and $135$ seconds, using NASA's HPC Pleiades and Electra systems, respectively ({arXiv:1811.09599}). Furthermore, we report simulations of Bristlecone-70 with depth $1+36+1$ and depth $1+40+1$ in $5.6$ and $580.7$ seconds per amplitude, respectively. To the best of our knowledge, these are the first successful simulations of instances at these depths.
1 Benchmarking and the Experimental Setup
The study benchmarks Alibaba Cloud simulation of Bristlecone random circuits, clarifies equivalences and runtime accounting, and describes the distributed tensor-contraction setup and simulator refactoring.
- Benchmarking and the Experimental Setup: The circuits begin and end with all-Hadamard layers, with t alternating CZ configurations and intervening randomly selected single-qubit gates.
- Benchmarking and the Experimental Setup: The benchmark uses revised random quantum circuits for Bristlecone-70, with prior results reported on NASA Pleiades, Electra, and Sunway systems.Bristlecone-70 is simulation-equivalent to Bristlecone-72 because two qubits can be contracted with their only neighbor.
- Benchmarking and the Experimental Setup: The distributed setup uses 1449 ECS instances with 88 virtual CPU cores and 160 GB memory per instance.An agent divides the tensor contraction into subtasks, distributes them through OSS, and aggregates completed results into an amplitude.
- Benchmarking and the Experimental Setup: Reported runtime excludes preprocessing because preprocessing is performed once independently of the number of amplitudes calculated.
- Benchmarking and the Experimental Setup: The simulator was refactored by swapping tensor-network nodes and edges, improving abstraction while retaining functionally similar algorithms and performance.
2 Benchmarking Results
The benchmarking results evaluate workload balance and amplitude-distribution behavior for Bristlecone-70 simulations on Alibaba Cloud.
- Benchmarking Results: The execution-time estimates use equal-complexity subtasks and predict cluster runtime from the slowest among four representative node configurations.The estimates scale assignments by Na, Ns, and available vCPU cores.
- Benchmarking Results: More calculated amplitudes reduce execution time per amplitude, while sufficiently large workloads stabilize per-amplitude runtime.The reported explanation is improved computational-load balance for larger Na×Nc workloads.
- Benchmarking Results: Using more smaller ECS instances slightly reduces execution time when the total number of vCPU cores is held constant.
- Benchmarking Results: For 200,000 amplitudes from Bristlecone-70 circuits at depth 1 + 28 + 1, the distribution of Np closely matches the Porter-Thomas form Ne^-Np.