🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Existing approaches for modeling molecular dynamics fall short. Blake Duschatko, Xiang Fu, Cameron J. Owen, Yu Xie, Albert Musaelian, #JClinic PI Tommi Jaakkola, and Boris Kozinsky propose a novel ML coarse grained force field approach in a new paper: arxiv.org/abs/2405.19386

Excited to announce a new version of our arXiv paper "Unified Differentiable Learning of Electric Response", now including ferroelectrics in addition to dielectrics! arxiv.org/abs/2403.17207 Anders Johansson Chuin Wei Tan Cameron J. Owen (cr. video) Boris Kozinsky Materials Intelligence Research @ Harvard Harvard SEAS

Announcing "torch.compile: the missing manual": docs.google.com/document/d/1y5… The performance/memory sections are not done yet, but everything else is as much hard won debugging knowledge that I could write down from Meta's deployments of PT2. Feedback welcome!

After a journey started during my PhD, I am happy to share our preprint about BMTI: a new nonparametric density estimator for high dimensional spaces developed together with my great collaborators Aldo Glielmo , Alex Rodriguez and Alessandro Laio! arxiv.org/abs/2407.08094

Our preprint on the BMTI nonparametric density estimator for high-dimensional spaces is out! #ML #density #gradient Aldo Glielmo

New preprint is out: arxiv.org/abs/2407.13643! Here, we present a multimodal technique for understanding surface roughening in nanoparticle catalysts using CO-DRIFTS, X-ray absorption (XAS), and machine learned force fields. The expt. methods provide disparate insights into the 🧵

My latest work is on ChemRxiv. We used Allegro, an ML interatomic potential based on local equivariant representations, to perform large-scale MD simulations for a series of deep eutectic electrolytes./1 doi.org/10.26434/chemr…

New work by T. Bischoff, B. Jäckl, and me: Benchmarking machine-learning potentials for molecular dynamics requires running accelerated simulations. We provide a fully automated benchmark for this. Preprint: doi.org/10.48550/arXiv… Repository: gitlab.com/qmml/h-benchma…

Intrinsic & extrinsic defect chemistry of trigonal Selenium, incl metastable states & non-radiative recombination Combined theory & expt analysis, we find an intrinsic tolerance to 𝘱𝘰𝘪𝘯𝘵 defects, with GBs/interfaces the limiting factor for PV 📈 chemrxiv.org/engage/chemrxi…

🚀 Call for Abstracts! 🚀 Join us at the APS March Meeting symposium "Machine Learning for Atomistic Simulations" to shape this rapidly-evolving field! 📝 Submission Deadline: October 25, 2024 🌐 Submit here: march.aps.org #APS #MarchMeeting #ML Michele Pavanello

Upcoming blog post teaser: "Ways to use torch.compile"

Machine learning can be powerful for understanding defects, but currently sufficient only in select cases. MLFFs (& geometric/electrostatic tools in doped) allow screening for challenging 'non-local' defect reconstructions (split vacancies) in all ICSD/MP solids, w/caveats 🔗👇

Excited to give an invited talk on our ML model for modeling materials' response to electric fields at the Ferro2025 conference, and to see the latest developments on the fundamental physics of ferroelectrics! arxiv.org/abs/2403.17207

The latest version of 𝙙𝙤𝙥𝙚𝙙 (and 𝑺𝒉𝒂𝒌𝒆𝑵𝑩𝒓𝒆𝒂𝒌), our defect modelling python packages, have been released! Incl: - Major efficiency updates - Advanced defect/carrier thermodynamics w/custom constraints - Auto shallow defect handling - CC diagram generation ...🧵👇

For existing MLIPs, lower test errors do not always translate to better performance in downstream tasks. We bridge this gap by proposing eSEN -- SOTA performance on compliant Matbench-Discovery (F1 0.831, κSRME 0.321) and phonon prediction. arxiv.org/abs/2502.12147 1/6

Beyond happy to announce today Allegro-pol, a machine-learning framework that predicts how materials respond to electric fields with quantum-level accuracy, capturing vibrational, dielectric, and ferroelectric behavior at the million-atom scale! 🚀 nature.com/articles/s4146…

Our Allegro-Pol model extended the Allegro architecture to predict how materials respond to external electric fields while enforcing physical rules. It could describe vibrational, dielectric, and ferroelectric behavior for systems up to millions of atoms! nature.com/articles/s4146…

Excited to see our AI model for electric fields featured by Harvard Harvard SEAS ! 🚀 Link to paper 👉 nature.com/articles/s4146… seas.harvard.edu/news/2025/06/m…

Last month, we released a major update to the NequIP framework that fully leverages PyTorch 2.0 compilation for MLIPs. It’s significantly faster, easier to use, and more versatile than before. Preprint: arxiv.org/abs/2504.16068 Code: github.com/mir-group/nequ… nequip.readthedocs.io

Machine learning can be powerful for understanding defects, but currently sufficient only in select cases. MLIPs (& geometric/electrostatic tools in doped) allow screening for challenging 'non-local' defect reconstructions (split vacancies) in all ICSD/MP solids, w/caveats 🔗