Quantum-Based Molecular Dynamics Simulations Using Tensor Cores

Tensor cores, along with tensor processing units, represent a new form of hardware acceleration specifically designed for deep neural network calculations in artificial intelligence applications. cores provide extraordinary computational speed and energy efficiency but the caveat that they were contractions (matrix–matrix multiplications) using only low-precision floating-point operations. Despite this perceived limitation, we demonstrate how can be applied high to challenging numerically sensitive problem quantum-based Born–Oppenheimer molecular dynamics, which requires highly accurate electronic structure optimizations conservative force evaluations. The interatomic forces are calculated on-the-fly from an is obtained generalized network, where naturally takes advantage exceptional power allows performance excess 100 Tflops on single Nvidia A100 GPU. Stable dynamics trajectories generated framework extended Lagrangian combines long-term stability, even when approximate charge relaxations evaluations limited accuracy by noisy conditions caused core A canonical ensemble simulation scheme also presented, additional numerical noise absorbed into Langevin-like dynamics.