Janosh
Computational materials scientist. I work at Periodic Labs on high-throughput density functional theory (DFT) and machine-learning force fields (MLFFs) for atomistic simulations.
Recent Work
Materials ProjectWidely used database, website, API and OSS ecosystem built for computing properties of inorganic materials.
atomate2
atomate2 is a library of computational materials science workflows used by the Materials Project and beyond.
Matbench DiscoveryBenchmark for machine learning energy models simulating a real-world materials discovery campaign.
MatPESA foundational DFT potential energy dataset for materials covering 89 elements and emphasizing data diversity and quality (at PBE and r2SCAN level).
MLIP PES softening
Overcoming systematic softening in universal machine learning interatomic potentials by fine-tuning
TorchSimTorch-native, batchable, atomistic simulations.
DielectricsPushing the Pareto front of band gap and permittivity with ML-guided dielectrics discovery incl. experimental synthesis.
PhD ThesisTowards Machine Learning Foundation Models for Materials Chemistry — Matbench Discovery, ML-guided dielectric discovery and the MACE-MP foundation model.
CHGNetPretrained universal neural network potential for charge-informed atomistic modeling published on the Sep 2023 cover of NMI.
Open Source
pymatgen
One of the largest and most popular open source materials analysis codes that defines classes for structures, molecules, slabs, etc. and interfaces seamlessly with various other materials codes. It also powers the Materials Project.
- Matbench Discovery
Benchmark for machine learning energy models simulating a real-world materials discovery campaign.
MatterVizA library of Svelte components for building interactive web apps with performant chemistry visualizations like periodic tables, Bohr atoms, nuclei, heatmaps, scatter plots.
pymatvizA toolkit for visualizations in materials informatics to complement pymatgen.
MultiSelectKeyboard-friendly, accessible and customizable multi-select web component.
atomate2 is a library of computational materials science workflows used by the Materials Project and beyond.
DiagramsTypst and LaTeX diagrams of concepts in physics/chemistry/ML.
- CHGNet
Pretrained universal neural network potential for charge-informed atomistic modeling published on the Sep 2023 cover of NMI.
jobflowjobflow is a library for writing computational workflows. It provides the plumbing underlying atomate2 and was adopted by several other workflow libraries.
Tensorboard ReducerReduce multiple PyTorch TensorBoard runs to new events/CSV/JSON. Good for model ensembles.
MatCalc
A Python library for calculating materials properties from ML force field potential energy surfaces.
Normalizing FlowsCurated list of resources for learning and using normalizing flows, a powerful tool in ML for modeling probability distributions.
Overcoming systematic softening in universal machine learning interatomic potentials by fine-tuning
- TorchSim
Torch-native, batchable, atomistic simulations.
- Dielectrics
Pushing the Pareto front of band gap and permittivity with ML-guided dielectrics discovery incl. experimental synthesis.
MACE Foundation Models
Fast and accurate machine learning interatomic potentials with higher order equivariant message passing.
- MatPES
A foundational DFT potential energy dataset for materials covering 89 elements and emphasizing data diversity and quality (at PBE and r2SCAN level).
- Materials Project
Widely used database, website, API and OSS ecosystem built for computing properties of inorganic materials.
Notes on Physics
This is a compilation of notes and solutions to problem sheets for some of the physics lectures I took, most of them in Heidelberg. Hopefully, they can be useful to others. If you find errors, please open an issue.
