Last week we introduced a few new routes to the Materials Science API from . This work is part of a broader effort to create a suite of tools that eventually can be commanded by an AI agent for automated material design and discovery.
Create a supercell
Create a supercell from a material
This route is primarily for aesthetics and observing the symmetry of a CIF file. Depending on the size of your unit cell, I've seen things get pretty laggy so be aware.
Calculate the energy above the hull of a material
Calculate the energy above the convex hull (eV/atom) of a material
This route creates a phase diagram and calculates the energy above the convex hull to estimate the stability of a material. Usually this requires DFT and expensive computations, but this route approximates everything with an MLIP. Results in seconds, not hours or days. I'll post a more thorough writeup of this route as it's quite involved on the backend.
Relax a structure
Relax a material structure
Using Orb v3, this route takes a CIF file and attempts to relax it into its lowest energy state. For many chemical systems, you'll notice there is a rather large difference in what is considered relaxed compared to the DFT relaxed materials found on MP or another database.
Over the next couple weeks I'll be working to expand this suite of tools. Still, the focus will be "good enough" machine learning estimates of essential material characterizations. Users should be able to take a novel material and find out if it's worth pursuing
Discover other posts like this one
After wrestling with Mattergen finetuning for longer than I would've liked to, I pivoted back to simple property conditioned generation on Zn-Mg-H systems per 's recommendation. Each generated system
For simplicity I feel like we can frame this as purely focusing on the materials discovery, knowing that the broader goal could still be the Bell Labs 2.0 Logo draft, tried to go Skunkworks style cart
MatterGen is a diffusion model built for materials discovery published by Microsoft, trained on materials datasets Alexandria, ICSD (licensed data so it isn't publicly released), and Materials Project