Managed to croak my way through my talk with cold sweats (ill, not hungover -- honest), DM/email if you want the slides, otherwise here's some useful links
English
Matthew Evans / @[email protected]
2.5K posts
Matthew Evans / @[email protected]
@funroll_loops
Comp. matsci postdoc at @UCLouvain_be & @Matgenix interested in materials discovery and data management for open science, also https://t.co/sDCho3BPom
King's Lynn, UK / Brussels, BE Присоединился Mayıs 2011
1K Подписки352 Подписчики
In sunny Boston for #f24mrs! Just putting the "finishing" touches on my talk from the public library ready for Tuesday at 13:30. Come listen to me talk about OPTIMADE, datalab and their relevance to broadening access to AI in materials research!
English
@jsnover Posted there first, with extra bonus characters! bsky.app/profile/funrol…
English
@funroll_loops Come join us at the other place.
bsky.app/profile/materi…
English
Matthew Evans / @[email protected] ретвитнул
We have been on Bluesky for a while and we would love for you to join us there as well! 🦋
Follow us ▶️go.bsky.app/QUQJbsd
English
same username over there, though hoping to use this as an excuse to stop doomscrolling
English
@anuroopsriram @bkmi13 @RickyTQChen @bwood_m Interesting work! Will you be releasing the new crystal structures? Once out there, there's a whole raft of us who want to screen them for our favourite properties to prioritise some for expt. validation
English
I’m excited to share our latest work on generative models for materials called FlowLLM.
FlowLLM combines Large Language Models and Riemannian Flow Matching in a simple, yet surprisingly effective way for generating materials.
arxiv.org/abs/2410.23405
@bkmi13 @RickyTQChen @bwood_m
English
@RowanSci @ccdc_cambridge If you support the MP API, it should be straight forward to support OPTIMADE (optimade.org) which will give you access to OQMD, AFLOW, NOMAD, COD, MCloud, Alexandria etc, including unified search, would be happy to help!
English
Periodic calculations can be submitted on Rowan just like molecular calculations.
You can input periodic structures from files, use the Materials Project API, or draw them yourself in our molecule editor.
(Support for the @ccdc_cambridge API is planned.)
English
Rowan now supports periodic systems!
We're launching today with two periodic engines: xTB and Open Materials 24.
Periodic calculations are great for modeling crystal structures, aqueous systems, solvent mixtures, battery electrolytes, ionic liquids, and surface catalysis.
English
Matthew Evans / @[email protected] ретвитнул
Presented #semanticClimate's work in 5-starification of climate documents for its sustainable future. My takeaway from the panel: #TheTechWeWant is simple, plural and fosters community. Software should be rooted around the planet with goals other than economic growth. @OKFN
English
@MJCliffe Not sure it's entirely what you want but I've worked with Xerus in the past, which wraps GSAS-II with a Bayesian optimizer and can screen OPTIMADE and other dbs (github.com/pedrobcst/Xerus), or the new iteration of fullprofapp is open source (gitlab.org/d7081/fullprof…)
Ath, Belgique 🇧🇪 English
East, England 🇬🇧 English
So happy to have graduated yesterday from @unibirmingham alongside Andrea Iliceto.
Thanks again to @condensedmatter for your supervision and to the whole Morris Group!
English
@MJCliffe @TheRealBenjiB also worth a visit for a pint if you're in Brussels
English
@MJCliffe @TheRealBenjiB Bit of an overwhelming schedule but I tune into whatever is on kioskradio.com/schedule fairly often
English
@digital_rsc If you are interested in getting involved we still hold open monthly meetings online (details at optimade.org), and fingers crossed there will be another CECAM workshop next year
x.com/cecamEvents/st…
CECAM@cecamEvents
The workshop "Open Databases Integration for Materials Design" returns to CECAM-HQ (& online) this week! The @optimade_ consortium was first established to create a comprehensive #API capable of accessing all #materials #databases, to address the issue of scattered databases.
English
@digital_rsc As always, thanks to all the other authors and contributors who have volunteered their time over the past years. Looking forward to seeing how people build on our efforts! [3/2]
English
Our paper on the latest OPTIMADE developments is now in print in @digital_rsc (doi.org/10.1039/D4DD00…), accompanying the latest version of the specification (changelog: github.com/Materials-Cons…) [1/2]
CECAM@cecamEvents
The large-scale use and exchange of data is a prerequisite for AI in materials research. A major international collaboration, including #CECAM and @nccr_marvel, now presents an extended version of the #OPTIMADE standard. Read more ➡️ cecam.org/news-details/i…
English
@bkmi13 Lovely work! Do you plan to release any of the DFT stable structures your model has already generated? We have some nice tools for doing this within the OPTIMADE consortium (optimade.org) which can (in principle) get them in front of experimentalists for verification
English
In De Novo Generation (DNG), we aim to predict stable configurations, along with atom types. FlowMM produces Stable, Unique, and Novel crystals at a competitive rate compared to leading, open models. It does so at between 1/3rd and 1/35th the cost (in terms of integration steps)!
English
Announcing our new model for materials!
FlowMM...
- Generates stable & novel materials efficiently
- Predicts crystal structure accurately
- Generalizes Riemannian Flow Matching to point clouds w/ periodic boundaries
arxiv.org/abs/2406.04713
@RickyTQChen @anuroopsriram @bwood_m
GIF
English
@Andrew_S_Rosen @Robert_Palgrave @MolecularXtal I would add that some of the most egregious examples in the first release of structures were pathological even under DFT (numerical issues from overlapping atoms etc), I think these have since been removed from the Google drive data release
Brandon, England 🇬🇧 English
@Robert_Palgrave @MolecularXtal ... then the model did exactly what it was supposed to. But it inherits all the approximations of the underlying exchange-correlation functional and the fact that the definition of "stable" here is not the same as realizable. 3/3
English
This is a very interesting article, and I'd like to expand on "But what are the risks of using data that isn’t real?"
Computational chemistry is inherently about model systems and approximations. The materials being studied, by definition, do not precisely match experiment. 🧵
PKamat@KamatlabND
👇 Why are computational chemists making up their data?: chemistryworld.com/features/why-a…
English
@dr_k_choudhary Fair enough, thanks for responding, the scale just seems a bit off from what we usually report. Fingers crossed for Nb2B then!
English
@funroll_loops There are materials which have been synthesized at similar energy above hull values. A materials scientist would know, ehull is not the ultimate criteria for stability. Anyway, it's a proof of concept work, more detailed work to follow.
English
Delighted to share a preprint on #AtomGPT. We demonstrate the computational discovery of a novel #superconductor candidate material using a combination of #LLM, #DFT, #ML surrogate models, and a force-field (#MLFF).
arxiv.org/abs/2405.03680
#jarvisnist #nist #mistral #gpt2 #ai
English