The Signal
PNNL Breaks a 164-Qubit Chemistry Model into 12-Qubit Tasks
Pacific Northwest National Laboratory describes how its Quantum Flow algorithm models propane as a 164-qubit chemistry problem using repeated 12-qubit-equivalent calculations. The simulated workflow recovered about 95% of a coupled-cluster
Quantum.world Team · Written with AI · fact-checked ·
One 164-qubit problem, thousands of smaller calculations
Pacific Northwest National Laboratory’s Quantum Flow algorithm breaks a large chemistry calculation into repeated tasks that each need the equivalent of 12 qubits. In a propane example, the full model corresponds to 164 qubits, while the QFlow workflow uses 56,860 small active-space calculations in each cycle. PNNL published the result in a research highlight on October 30; the paper appeared in The Journal of Physical Chemistry Letters on October 8.
The reported result is a computer simulation, not a demonstration on a 12-qubit quantum processor. The research team ran its variational quantum eigensolver calculations using NWQSim, a quantum circuit simulator, as part of a high-performance computing workflow. The 12-qubit figure describes the size of each simulated active-space problem, not a physical device used to calculate propane.
How QFlow divides the chemistry
Quantum Flow, or QFlow, splits a large correlated-electron problem into smaller active spaces, then uses classical computing to construct effective Hamiltonians that connect those spaces. The active-space calculations optimize pieces of a shared wavefunction; their results feed into an evolving pool of parameters, which informs later calculations. Because each active space is small, the method’s quantum-resource requirement depends on the largest subproblem rather than the full molecule.
For the propane calculation, the researchers used the cc-pVDZ basis set and active spaces containing six electrons in six orbitals. The paper describes the full target as 13 occupied and 69 virtual orbitals, corresponding to a 164-qubit encoding. Its authors write: “In the largest QFlow simulations, we optimize 1.17 million wave function parameters using the equivalent of 12 qubits.” Nicholas P. Bauman, Ajay Panyala, Chenxu Liu, Muqing Zheng, Meng Wang and Karol Kowalski
The 95% result is measured against CCSD
By the third cycle, QFlow recovered an average of about 95% of the correlation energy calculated with the coupled-cluster singles and doubles method, or CCSD. The first cycle’s lowest-energy active space recovered 94%; subsequent cycles added about one percentage point. That comparison is to a CCSD reference, not to an exact solution of the full 164-qubit problem.
The propane calculation ran across 232 CPU cores organized into 116 process groups. The scale of the workflow matters alongside its 12-qubit subproblems: QFlow processes tens of thousands of active spaces, updating and coordinating their results on classical computing infrastructure. In this implementation, quantum-circuit simulation and high-performance computing do the work; the authors do not report a run on quantum hardware.
A useful result, with a clear boundary
The paper also examines water across several basis sets, including a 228-qubit parent problem represented by repeated 12-qubit tasks. Across those water calculations, QFlow recovered roughly 97% of the CCSD correlation energy. The results extend the method beyond the propane case, but the team’s demonstrations remain simulations using a singles-and-doubles model and a variational solver.
That boundary limits what the figures establish. The calculations show that a partitioned workflow can approximate a large chemistry model while keeping each quantum-solver subproblem small; they do not establish quantum advantage or show that a physical quantum computer has simulated propane at this scale. For the headline propane case, the concrete benchmark is 56,860 active spaces per cycle, 1.17 million optimized parameters and about 95% of the CCSD reference correlation energy by cycle three.