RIKEN, IBM and Cleveland Clinic run quantum chemistry on 12,635-atom system

Quantum Chemistry at the 12,000‑Atom Scale

Researchers from RIKEN, IBM and the Cleveland Clinic have pushed the limits of hybrid quantum‑classical (HQC) simulation by tackling protein‑ligand complexes containing 12,635 atoms. By fragmenting the molecules and distributing quantum sampling across two 156‑qubit IBM processors (ibmcleveland and ibmkobe), the team ran 21,006 circuits for over 239 hours, collecting 3 × 10⁹ measurement outcomes. Classical supercomputers (Fugaku, Miyabi‑G and ROQUO) handled the fragment wavefunction calculations, achieving 72.5 % parallel efficiency on 128 GPU nodes.

From Fragments to Full Proteins

The workflow decomposes a large biomolecule into smaller fragments that can be solved on a quantum processor while the surrounding environment is treated with a lower‑cost method. Two key algorithmic advances—ExtSQD and TrimSQD—enable accurate fragment solutions and a mixed‑basis approach (3‑21G for ligands and nearby residues, STO‑3G for the rest). This strategy allowed the team to surpass the previous HQC record of 303 atoms, delivering energies within 0.1 mHa of high‑level classical methods for the largest fragments.

Scaling Quantum Resources

Although the IBM devices host 156 physical qubits each, the calculation used up to 94 qubits simultaneously, distributed across the two processors. The work demonstrates that scaling quantum resources, together with exascale classical hardware, can address chemically relevant systems far beyond the reach of earlier quantum‑embedding studies.

Implications for Drug Discovery

Accurate protein‑ligand binding predictions are a cornerstone of structure‑based drug design. By matching coupled‑cluster (CCSD) accuracy while handling systems of this size, the HQC workflow shows promise for exploring binding energetics of realistic biological targets, potentially accelerating early‑stage screening.

Looking Ahead

The study highlights two complementary paths forward: further hardware improvements (more qubits, lower error rates) and algorithmic refinements that reduce quantum resource demands. As quantum processors mature and classical supercomputers continue to scale, hybrid approaches like this could become a standard tool for tackling the most challenging problems in molecular modeling.

--- For a deeper dive, see the pre‑print by Kenneth M. Merz et al. on arXiv.