This paper presents MatterGen, a model that generates stable, diverse inorganic materials across the periodic table and can further be fine-tuned to steer the generation towards a broad range of property constraints. To enable this, the authors introduce a new diffusion-based generative process that produces crystalline structures by gradually refining atom types, coordinates, and the periodic lattice.

This paper introduces LLaDA, a diffusion model trained from scratch under the pre-training and supervised finetuning (SFT) paradigm. LLaDA models distributions through a forward data masking process and a reverse process, parameterized by a vanilla Transformer to predict masked tokens. https://arxiv.org/abs/2502.09992

Domains registered within days of this post. Not a dataset - just a wordlist

Here we showcase a few key latent features and their relationships to each other, points colored by their Tc. We animate from 0 K to ~130 K.

Superconductor candidates sampled at a target Tc of 130k

Generated image from "A hairy frog" using DALL-E 3 from OpenAI.

Generated model from an image using the StabilityAI API.

Generated image from "A marble sculpture of a human male with white background" using the StabilityAI API.

400 .cif files of candidate structures property condition generated by MatterGen where tc = 298.15K

Interactive plot of predicted vs. true Tc on the evaluation set.

Visualizing the counts of materials in the training and evaluation dataset by their Tc. First bin is non-superconductors, the rest are ranges of 20 K increments.

Using the 256 dimensional latent space output from the Orb model, we visualize the 3DSC(MP) dataset using UMAP with direction from Tc labels. Hover a point to see Tc, formula, and Material Project identifier.

Using the 256 dimensional latent space output from the Orb model, we visualize the 3DSC(MP) dataset using t-SNE and UMAP. The UMAP projection has been given the target for learning a manifold that keeps similar Tc materials close together.

Authors introduce Orb, a family of universal interatomic potentials for atomistic modeling of materials. Orb models are 3-6 times faster than existing universal potentials, stable under simulation for a range of out of distribution materials and, upon release, represented a 31% reduction in error over other methods on the Matbench Discovery benchmark. https://arxiv.org/abs/2410.22570

Authors present MatterSim, a deep learning model actively learned from large-scale first-principles computations, for efficient atomistic simulations at first-principles level and accurate prediction of broad material properties across the periodic table, spanning temperatures from 0 to 5000 K and pressures up to 1000 GPa. https://arxiv.org/abs/2405.04967

Not exactly the most rigorous test, but this side-by-side comparison shows the difference between running MD locally (M2 Macbook Air) and on a proper server (g4dn.2xl with T4 GPU). Each log is actually 100 simulation steps too.

Molecular dynamics simulation temperature ramping NaCl 3x3x3 supercell from 0 K to 300 K

Molecular dynamics simulation temperature ramping H2O 3x3x3 supercell from 0 K to 300 K

Simulating ice into water