When @jrib_ first invited me to help training MACE for matbench discovery, I would never imagine the universal MACE I trained will become "a foundational model for atomistic materials chemistry." Explore our amazing team effort at

@XirtamEsrevni @Robert_Palgrave Agree with all points. Phase competition at finite temperatures and kinetics effects play a crucial role, so do electrons for many functional materials.

RT @XirtamEsrevni: Nice tool, glad they open-sourced it, but: 1. I believe @Robert_Palgrave has opined, critically, on this statement in…

@xie_tian Congratulations, MSR and the wonderful team!

RT @xie_tian: Excited to finally announce the publication of MatterGen on Nature. MatterGen represents a new paradigm of materials design w…

@jla_gardner Very awesome work. I ran into it last week and was thinking this was done in past few months

@jrib_ @XirtamEsrevni @TimothyDuignan Same! NEB examples example are so much needed

@Andrew_S_Rosen @PrefectIO but fortunately we have retries so usually not too many tries each job can complete within the job running window

@Andrew_S_Rosen @PrefectIO In many. I am being conservative and only scale up to 50-100. Perhaps could achieve higher throughput. Would love to try on NERSC but I guess it is not allowed

@Andrew_S_Rosen @PrefectIO Yes! dask-jobqueue SLURMCluster you shared! And all tasks in one parent flow. Could easily hit the cap though due to prefect API rate limit. I am on free plan :$

RT @BenBlaiszik: For those of you looking for a big helping of LLMs for Thanksgiving, here you go! 🤖🦃 (link in next post) In this work, we…

@AriWagen Such a great addition to #MLIPArena Wonder what would be your perspective on how to fairly compare models trained on DFTs of molecular and periodic systems. In MLIPArena, we aim to provide both task-specific and target-agnostic evals :)

This is a major step to lead the entire field/community toward the right direction that real experiments can testify. Congrats to the team for the achievement!

@Andrew_S_Rosen @jrib_ @lonepair good model stands the test of time ;) I wish I could put down other work and have insane compute and resource to train another MACE on bigger dataset 😅

RT @AhlawatParamvir: Introducing SRME: a new metric for further assessment of foundational machine learning interatomic potentials! 👉A sm…

@XirtamEsrevni @RhysGoodall @MarkNeumannnn Thanks for pointing that out @RhysGoodall Also we never guarantee Arena will be "living" although we are working toward a longstanding one. This work is pretty much WIP unlike matbench discovery. It's good to receive early feedback but imo it's our decision to prioritize todos

@XirtamEsrevni @MarkNeumannnn They are barastat. Models' stability will also reflect on fluctuation. Less stable models or non-conservative models tend to cause instability. We are working on equation of states and will add those to arena too.

@MarkNeumannnn @XirtamEsrevni It is clear that energy-based metrics are gamable. We haven't updated MACE-MP-0 and the version one works well for less gamable metrics like thermal conducvity. Would you ask us to retract the paper submission for you? credit to wonderful TC test source: