Researchers built a hybrid quantum-classical workflow that can drive ab initio molecular dynamics (AIMD) for small molecules and liquid solutions using current quantum hardware. The key idea is to run a chemistry-inspired quantum circuit (called a LUCJ ansatz) on a superconducting quantum processor, then post-process the measurement outcomes with a method named Sample-based Quantum Diagonalization (SQD). For gas-phase and solution benchmarks, SQD recovered energies and analytical nuclear forces that agree with exact full configuration interaction (FCI) results in the same small basis to within about 1 kcal per mole, and produced stable molecular-dynamics trajectories.

How the team did it: they connected a standard molecular-dynamics engine (Amber/sander) to a quantum-chemistry stack. Quick provided one- and two-electron integrals, PySCF built the quantum Hamiltonian in a minimal STO-3G basis, and IBM superconducting hardware executed the LUCJ circuit to produce measurement bitstrings. SQD took those bitstrings, reconstructed a small set of important determinants (configurations of electrons), diagonalized the Hamiltonian in that subspace, and returned energies and analytical nuclear gradients for the dynamics.

The tests were concrete and deliberately small so the results could be checked exactly. The quantum-driven runs used active spaces that include all electrons in the STO-3G minimal basis: (10 electrons, 8 orbitals) for ammonia (NH3), (10e, 9o) for methane (CH4), and (10e, 7o) for water (H2O). In this setup the classical reference is FCI in the same basis, so it is an exact benchmark for comparison. The SQD workflow sampled between 200 and 800 measurement bitstrings per batch; for CH4 this sampled about 1.3–5.0% of the full 15,876-determinant space, for NH3 about 6.4–25.5% of 3,136 determinants, and for H2O about 45–100% of 441 determinants.