Features reconstruction errors between the latent spaces of several universal MLIPs
No day goes by without a new universal #ML potential. But how different they really are? Sanggyu and Sofiia tried to give a quantitative answer by comparing the reconstruction errors between their latent-space features. If you are curious, check out the #preprint arxiv.org/html/2512.05...
Congrats to 🧑🚀 Sergey Pozdnyakov who received a distinction (best 8% of theses at @materials-epfl.bsky.social) for his PhD thesis "Advancing understanding and practical performance of machine learning interatomic potentials". Поїхали 🚀! infoscience.epfl.ch/entities/pub...
📢 PET-MAD is here! 📢 It has been for a while for those who read the #arXiv, but now you get it preciously 💸 typeset by @natcomms.nature.com Take home: unconstrained architecture + good train set choices give you fast, accurate and stable universal MLIP that just works™️ www.nature.com/articles/s41...
👏 Congratulations to Prof. @micheleceriotti.bsky.social (@labcosmo.bsky.social) and Prof. Roland Logé (LMTM)
on their promotion to Full Professor of Materials Science in the School of Engineering!
More info: actu.epfl.ch/news/appoint...
A cartoon explaining how mild finite-temperature conditions induce disorder and dynamical reconstruction on the surfaces of lithium thiophosphates
📢 Now out on @physrevx.bsky.social energy, journals.aps.org/prxenergy/ab... from 🧑🚀 @dtisi.bsky.social and Hanna Türk, our #PET -powered study of the dynamic reconstruction of LPS surfaces, and how it affects their structure, stability and reactivity.
Just before my last week in @labcosmo.bsky.social. Our metatensor and metatomic paper is out! A collection of the hard work we’ve done at @labcosmo.bsky.social to make atomistic machine learning easier to use for experts and not alike.
Scheme of the GNN architecture of the FlashMD method.
📢 Running molecular dynamics with time steps up to 64fs for any atomistic system, from Al(110) to Ala2? Thanks to 🧑🚀 Filippo Bigi and Sanggyu Chong, with some help from Agustinus Kristiadis, this is not as crazy as it sounds. Let us briefly introduce FlashMD⚡ arxiv.org/html/2505.19...
Plot of the potential and total energy trends along a MD trajectory, showing drift for non-conservative forces, and how that is fixed using multiple time stepping.
The PET-MAD universal forcefield mingled with the dark side, and got twice as fast 🚀. Read on, or head to the 🧑🍳📖 atomistic-cookbook.org/examples/pet..., if you are curious of what this is all about. #atomistic-cookbook #compchem #machinelearning #mlip🧵
A dynamical surface morphology implies changing electronic structure properties of the surface. We found large fluctuations of the electrostatic potential during dynamics, which will be crucial for the surface reactivity.
We studied the evolution of the SOAP descriptors of the tetrahedra on the surface, for an atomistic analysis of their dynamics. We time-averaged the SOAP vector to rule out the effects of fast-moving Li atoms
The surface reconstruction increases the atom density and reduces the Li diffusion
The main takehome is: static calculation are not enough!!
After MD relaxation, the effect of temperature reduces the differences in surface energy and coordination number of the different surface cuts, in agreement with a general amorphisation of the surfaces.
Last week, we continued @labcosmo.bsky.social's study of solid-state electrolytes, focusing on the properties of the LPS surfaces arxiv.org/abs/2504.11553. Many thanks to my office mate Hanna Tuerk, who did a lot of the work
Find (many) more benchmarks and tests in the preprint arxiv.org/html/2503.14..., try PET-MAD for yourself, and let us know if you manage to break it - we're already working towards PET-MAD-2 😅 #compchem @nccr-marvel.bsky.social @materials-epfl.bsky.social @erc.europa.eu @cscsch.bsky.social
@labcosmo.bsky.social continuing its strong performance with Davide Tisi's talk on transport mechanisms in solid-state electrolytes!
Polar plot showing the errors of several machine-learning potential of different test sets. Smaller is better here!
Plots showing the evaluation time per atom for several machine-learning potentials as a function of the number of atoms in a simulation. Smaller is better
📢 PET-MAD has just landed! 📢 What if I told you that you can match & improve the accuracy of other "universal" #machinelearning potentials training on fewer than 100k atomic structures? And be *faster* with an unconstrained architecture that is conservative with tiny symmetry breaking? Sounds like 🧑🚀