University of Osaka and Fixstars Advance IQPE Quantum Chemistry Simulation on 1,024 GPUs

Key Takeaways

• Simulated 42-spin-orbital H₂O system on 1,024 GPUs
• Extended quantum chemistry simulation beyond previous 40‑qubit limit
• Used NVIDIA H100 GPUs on AIST’s ABCI‑Q supercomputer
• Implemented new parallel method to overcome inter‑GPU bottlenecks
• Supports future fault‑tolerant quantum computers for drug discovery

Summary

A joint team from Osaka University’s Center for Quantum Information and Quantum Biology and Fixstars Corporation has run one of the world’s largest classical simulations of iterative quantum phase estimation (IQPE) circuits for quantum chemistry, leveraging up to 1,024 NVIDIA H100 GPUs. The effort broke the prior 40‑qubit ceiling, simulating a 42‑spin‑orbital H₂O system and a 41‑qubit Fe₂S₂ circuit. The work relied on the chemqulacs‑gpu simulator and a new parallel‑computing method that eliminated inter‑GPU bottlenecks. The breakthrough expands the molecular problems that can be tested before fault‑tolerant quantum computers become operational.

Pulse Analysis

Quantum chemistry remains a grand challenge for both academia and industry, as accurate electronic‑structure calculations are essential for designing new drugs and advanced materials. Classical methods quickly hit a wall when electron correlation grows, prompting researchers to look toward fault‑tolerant quantum computers. Iterative quantum phase estimation (IQPE) is a cornerstone algorithm that can extract molecular energies with far fewer qubits than traditional QPE, but its validation requires massive classical resources to emulate the quantum circuits before hardware matures. The Osaka‑Fixstars collaboration therefore targeted the most demanding benchmark: a 42‑spin‑orbital water molecule and a 41‑qubit iron‑sulfur cluster, both beyond the 40‑qubit frontier that has limited prior studies.

The technical feat hinged on the chemqulacs‑gpu simulator, optimized for NVIDIA H100 accelerators, and a bespoke parallel‑computing framework that coordinated 1,024 GPUs on AIST’s ABCI‑Q system. By redesigning inter‑GPU communication pathways, the team eliminated the latency that typically stalls large‑scale state‑vector simulations. The result was a stable 48‑hour run that delivered accurate IQPE outcomes for circuits previously deemed infeasible. This achievement not only demonstrates the raw horsepower of modern GPU clusters but also validates the software stack needed to bridge the gap between theoretical quantum algorithms and practical, high‑throughput experimentation.

For industry, the implications are immediate. With a reliable classical sandbox, algorithm developers can now stress‑test quantum chemistry codes on realistic molecular sizes, refining error‑mitigation strategies and resource estimates before quantum hardware is ready. This accelerates the pipeline from academic discovery to commercial applications in pharmaceuticals, where early‑stage molecular screening can shave years off development cycles, and in materials science, where novel catalysts for climate‑friendly processes are urgently needed. Moreover, the university‑industry partnership showcases a model for leveraging national supercomputing assets to push the boundaries of quantum simulation, signaling a maturing ecosystem where HPC and quantum research co‑evolve.