About

What is Asparagus?

Asparagus is our attempt to make the creation of potential energy surfaces with machine learning as easy as possible.

Why is it called Asparagus?

Asparagus are very easy to grow but they grow separate to each other. In this code, we made an effort to put together all the different parts of the process of creating a potential energy surface with machine learning that have been separately developed in our research group for the last few years.

Why are we lying? It is called Asparagus because Kai does not like Asparagus and Luis likes to make bad jokes.

What can Asparagus do?

  • Sampling: Asparagus can sample the potential energy surface using a variety of methods. See the sampling page for more information.

  • Training: Asparagus can train a neural network to reproduce the sampled data. See the training page for more information.

  • Production: Asparagus can use the trained neural network to perform molecular dynamics simulations. See the production page for more information.

  • Evaluation: Asparagus can evaluate the performance of the neural network. See the evaluation page for more information.

  • Tools: Asparagus can perform a variety of useful tasks. See the tools page for more information.

What is next for Asparagus?

We are always looking for ways to improve Asparagus. If you have any suggestions, please let us know by opening an issue on GitHub or drop a line to the main developers. Some of the things we are currently working on are:

  • Adding new architectures (MACE or Nequip) Maybe ANI or general Behler-Parinello.

  • Uncertainty Quantification methods (Ensemble, Deep Evidential Regression, Others)

  • Interface with other MD codes: LAMMPS

  • Solvation models for ML/MM simulations

  • Active learning cycles.

  • Automatize transfer learning