Algorithms for molecular dynamics and geometry optimization on quantum hardware

Performing ab initio molecular dynamics and geometry optimization on quantum computers promises chemically exact simulations with transformative implications for catalysis and biophysics. However, progress is currently constrained by hardware noise, the high cost of gradient calculations, and limited qubit availability. Here, we demonstrate how machine learning and surrogate modeling can bridge the gap between theoretical potential and practical utility. By leveraging transfer learning to construct robust potential energy surfaces, our approach mitigates noise and significantly reduces the required number of quantum evaluations. This work establishes a framework for performing molecular dynamics on both near-term and fault-tolerant hardware.