LaSDI (Latent Space Dynamics Identification) enables a fast and accurate solution process on various partial differential equations (i.e., Burgers’ equations, radial advection problem, nonlinear heat conduction problem), achieving more speed-ups and less relative error with respect to the corresponding full order models. The repo includes four examples: 1D Burgers, 2D Burgers, a radial advection example as from MFEM, and a time-dependent diffusion example.

The accompanying repo gLaSDI (“greedy” LaSDI, or Parametric Physics-informed Greedy Latent Space Dynamics Identification) provides a framework for accurate, efficient, and robust data-driven reduced-order modeling of high-dimensional nonlinear dynamical systems. gLaSDI’s autoencoder discovers intrinsic nonlinear latent representations of high-dimensional data, while dynamics identification models capture local latent-space dynamics.