AI and the World's Hard Problems · Quantum physics & quantum computing · arXiv · 2026-10-19
Geometric inflation of deviations challenges neural quantum states in dynamics of quantum Ising models
Neural quantum states are a method for simulating non-equilibrium dynamics of strongly correlated quantum systems, yet they prove surprisingly hard to get accurate in some simple regimes. Using quench dynamics of a 1D quantum Ising model as a benchmark, the authors found much tighter bounds on the required parameter count than before, ruling out representational limits as the main obstacle. They instead attribute the difficulty to a geometric inflation of small deviations under the TDVP, where rotation of the quantum geometric tensor's null space makes previously irrelevant parameter errors significant, and suggest the neural network's nonlinearity as the source.