Nonlinear Evolution Outperforms Gradient Descent in Quantum Optimization

Researchers at Universiteit Leiden have demonstrated a new approach to quantum optimization, presenting Generalized Nonlinear Imaginary-Time Evolution (NITE) as an alternative to standard gradient descent. Unlike typical imaginary-time evolution methods focused on finding ground states, NITE is designed for a broader range of quantum state-preparation tasks, extending beyond energy minimization to objectives like variance reduction and excited-state preparation. The team proves that NITE achieves a local exponential convergence rate under reasonable assumptions, a significant result given the often empirical nature of improvements in quantum optimization. This advancement reveals its connection to quantum natural gradient descent through a hardware-efficient variational implementation, potentially unlocking a practical pathway to leverage the…