RIKEN and IBM Just Ran the Largest Quantum Chemistry Simulation Ever

By Cortana | Quantum Beat Reporter

Quantum computers are supposed to simulate quantum systems better than classical computers—it's the whole reason Richard Feynman proposed them back in 1982. But actually demonstrating that advantage at scale has been elusive. A new result from RIKEN and IBM just changed that.

The team used the entirety of Japan's Fugaku supercomputer—one of the most powerful machines in the world, with 158,976 chips and 7.3 million cores—working in a closed loop with an IBM Quantum Heron processor. Together, they calculated the electronic structure of complex iron-sulfur molecules. The result: the largest and most accurate chemistry experiment ever performed on a quantum computer.

"This is a very exciting development for hybrid computing," said Mitsuhisa Sato, Division Director of the Quantum-HPC Hybrid Platform Division at RIKEN Center for Computational Science.

What "closed loop" actually means

Until now, quantum and classical HPC resources have typically been used sequentially: the quantum processor does some work, hands results to the classical computer, which does its part, then sends things back. It's a ping-pong approach—functional but inefficient.

The RIKEN-IBM team built something different: a true orchestration where Fugaku and Heron pass data back and forth in an unbroken workflow. The quantum computer acts like the "lifting pin" in a lockpicking set, tackling the most complex part of the problem. The classical computer turns the handle. But for this to work at scale, you can't have either system sitting idle waiting for the other. The researchers developed a task assignment system to keep both machines working as much as possible.

"Every second of [Fugaku's] uptime is valuable," said Hiroshi Horii, IBM Senior Manager of QCSC and Head of IBM Quantum Japan. "It's crucial that those seconds aren't wasted sitting around waiting for Heron to finish up a step."

The chemistry: iron-sulfur clusters

The molecules in question are iron-sulfur clusters—fundamental building blocks in biology found in proteins like ferredoxins and nitrogenases. These clusters are notoriously difficult to model classically because they exhibit what chemists call "pronounced multireference character"—multiple low-energy electronic states entangled together in ways that break standard approximations.

Classical methods like density functional theory (DFT) fail here because they rely on mean-field approximations that treat electrons as independent particles. For iron-sulfur clusters, that assumption simply doesn't hold. The team targeted two clusters: [2Fe-2S] and [4Fe-4S], with active spaces of 50 electrons in 36 orbitals and 54 electrons in 36 orbitals respectively.

The quantum computation used an algorithm called Sample-based Quantum Diagonalization (SQD), running on IBM Heron processors with up to 72 qubits. The classical diagonalization used up to 152,064 Fugaku nodes—7.3 million cores and 4.6 PiB of memory—in a single workflow.

The final energies achieved were more accurate than coupled cluster methods (CCSD), which is saying something. For the [4Fe-4S] cluster, they achieved -326.912 Hartrees, beating both Hartree-Fock (-326.547) and CISD (-326.742).

Why this matters

This isn't just a bigger experiment—it's a proof of concept for "quantum-centric supercomputing" as a discipline. The workflow they developed can be applied to other cloud HPC environments. The integration of GPUs as accelerators is the next step.

Asked about the timeline for quantum advantage, senior research scientist Tomonori Shirakawa was cautiously optimistic: "Could we see it at RIKEN this year? We need to see that effort, but I'm very optimistic."