AMAR: Adaptive Mesh and Algorithm Refinement Modeling

The Adaptive Mesh and Algorithm Refinement (AMAR) multiscale modeling framework allows for active coupling between mesh-based partial differential equation solvers at the continuum level along with particle-based solvers at the molecular level.

The AMAR approach typically discretizes using finite-element or volume methods, along with particle-based solvers at the molecular level, e.g., molecular-dynamics or kinetic Monte-Carlo (KMC) simulations. AMAR technology allows for increased resolution at locations of interesting physical phenomena, e.g., catalyst/electrode surfaces. Using the AMAR approach, in these refined locations, the continuum solver is replaced with a different physical approach (e.g., KMC simulations)—referred to as algorithm refinement—while transferring information back and forth to the higher scale models in real-time.

The AMAR approach addresses several shortcomings of multiscale modeling:

• Active two-way coupling between computational models at different scales
• Accurate bridging of scales without intermediate surrogate models that typically do not span the entire parameter space
• Automated adoption of valid computational models in regions where a higher-level model is invalid.

The framework includes a compressible flow solver, a generalized particle flow solver, MARBLES (GitHub) a Lattice Boltzmann solver, and a KMC-based vapor deposition solver (developed on top of SPPARKS), as well as the third-party libraries needed to build the codes.

We have performed coupled simulations of the compressible adaptive-mesh-based flow solver to a particle-based KMC solver. The target application of the coupled solver framework is the simulation of vapor deposition for halide perovskite crystal growth. The simulation domain encompasses the multi-species gas flow to a deposition substrate, where the deposition and crystal growth will be captured through KMC solver simulations.

Development Team

AMAR’s development team includes Marc Henry de Frahan, Ethan Young, Hari Sitaraman, and Ross Larsen.

Contacts