Every day I seem to stumble onto new and interesting scientific computing projects in computational chemistry. They run the gamut from hobby projects and one-off ideas to high-powered science teams with strong followings.

Personally, I suspect the Covid pandemic has driven people to spend more time on computational projects they believe are important and useful. In some sense, it’s nice to work on projects like that. New ideas are worth exploring, but there’s never enough time to try them all.

This article lists some of those projects.

  • Real Space Multigrid DFT (RMGDFT)

  • DFT-FE

    Two computational chemistry packages for running Kohn-Sham density functional theory. Both have a cmake-based install. RMGDFT has both CPU and GPU versions. DFT-FE uses an energy-level projection method to greatly speed up convergence for metals, and is planning a CUDA-enabled release soon.

  • PySCF

  • Horton

    PySCF and Horton are python front-ends to libint and libxc. They let you get your hands dirty with Gaussian orbital coefficients for wave-functions. While they can calculate Hartree-Fock orbitals, in my opinion, their real utility is in letting you get at the matrix elements for orbital-orbital interactions. For anyone teaching a modern course in computational chemistry, I’d recommend working these packages into your course material.

  • Python Simulation Interface for Molecular Modeling

    This is a set of python scripts for driving polymer simulations with LAMMPS and Cassandra. The research group that developed this is using it regularly to create and simulate hydrocarbon melts with various packing fractions - so I would expect it work very well for this purpose. I had created something like this for a side-project as a postdoc, but never took the time to document, and publish it. Kudos to this group. Disclaimer: I haven’t tested this package.

  • DFTB+

    DFTB+ simulates molecules using a tight-binding approximation to the Schrodinger equation. Essentially, it computes energies using quantum mechanics on the valence orbitals only, using simplified orbital shapes. That still puts it in a different leauge than forcefield-based methods, or even AI methods for that matter. I would consider this program great for a simulations project because it’s very well documented and mostly stand-alone, which simplifies the build and install process. Disclaimer: I haven’t tested this package.

  • ddcMD

    This program is a highly efficient, barebones, GPU-accelerated MD engine. It is essentially what won the 2020 Gordon Bell special prize for Covid research (by simulating a coarse-grained model of the entire coronavirus). Yes, they narrowly beat out our team… It’s written up in a JCP article. Unfortunately, the code itself is still newly published, so not enough users have nagged them asking for more and better documentation. Go ahead and create an issue!

  • Gromacs

    Old programs can learn new tricks, it seems. I added Gromacs to the list here because its recent re-design lets it be used as a python module. Previously, this was accomplished by an external team that wrapped the gromacs library. This is no longer needed! Gromacs has a “gmxapi” python module that is working its way toward providing more APIs that reach into gromacs.

  • chemparam

    Finally, a shameless advertisement. The chemparam project creates traditional CHARMM and gromacs forcefield parameter files for small molecules. It uses any ab-initio force data you have on-hand as input to force-matching, which give bond, angle, torsion, and even Lennard-Jones and electrostatic charge parameters. There are some hooks in there to use data from NWChem, but you need some compilation skills and a special patch to get that to work. Of course, it also lets you input force data from your own sources.

Well, that’s been a quick wrap-up of some of the coolest developments I’ve seen lately. Hopefully it can serve as a jumping-off point for your next hobby project.