Accelerating Quantum Simulations via Machine-Learned Basis Optimization

Quantum simulations hold the potential to transform molecular modeling but are constrained by limited computational resources. We present a general data-driven framework that learns compact molecular representations, enabling faster and more accurate hybrid quantum–classical simulations. By leveraging physical symmetries and electronic correlations, the model generates reduced subspaces that preserve essential quantum accuracy while greatly lowering computational demands. When incorporated into quantum algorithms, the learned representations accelerate convergence and enhance numerical stability. This approach establishes a scalable path toward automated basis optimization and efficient Hamiltonian reduction, broadening the applicability of quantum simulation across chemistry and materials science.