Recent searches

No recent searches

Search options

Only available when logged in.
sigmoid.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A social space for people researching, working with, or just interested in AI!

Administered by:

Server stats:

598
active users

sigmoid.social: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.4.2

#compchem

3 posts3 participants0 posts today
Robert Pollice

#RobSelects preprint 1 of the week #ChemRxiv: Reaching density functional approxmation accuracy at the cost of extended tight-binding quantum chemistry. #compchem doi.org/10.26434/chemrxiv-2025

ChemRxivg-xTB: A General-Purpose Extended Tight-Binding Electronic Structure Method For the Elements H to Lr (Z=1–103)We present g-xTB, a next-generation semi-empirical electronic structure method derived from tight-binding (TB) approximations to Kohn–Sham density functional theory (KS-DFT). Designed to bridge the gap between semi-empirical quantum mechanical (SQM) approaches and DFT in terms of accuracy, robustness, and general applicability, g-xTB targets the performance of the ωB97M-V range-separated hybrid density functional with large basis sets while maintaining TB speed. Key innovations include an atom-in-molecule adaptive atomic orbital basis, a refined Hamiltonian incorporating range-separated approximate Fock exchange, up to fourth-order charge-fluctuation terms with a novel first-order electronic contribution, and atomic correction potentials (ACPs), as well as a charge-dependent semi-classical repulsion function. Parameterized on extended and extremely diverse molecular training sets – including “mindless molecules” – g-xTB achieves excellent accuracy across a broad chemical space, including the actinide elements. Benchmarking against around 32k relative energies across thermochemistry, conformational energetics, non-covalent interactions, and reaction barriers shows that g-xTB consistently outperforms GFN2-xTB, often reducing mean absolute errors by half. Notably, it achieves a WTMAD-2 of 9.3 kcal mol−1 on the full GMTKN55 benchmark, comparable to low-cost DFT methods. It also shows substantial improvements for transition-metal complexes, relative spin state energies, and orbital energy gaps – areas where many SQM and even DFT methods often struggle. In summary, g-xTB offers DFT-like accuracy with minimal computational overhead compared to its predecessor, GFN2-xTB, making it a robust, minimally empirical, transferable, and efficient alternative to machine learning interatomic potentials for a wide range of molecular simulations. It is proposed as a general replacement for the GFNn-xTB family and, in many practical cases, a viable substitute for low- and mid-level DFT methods.
Robert Pollice

#RobSelects preprint of the week #ChemRxiv: Benchmarking density functional approximations with a systematic set of randomly generated molecules. #compchem doi.org/10.26434/chemrxiv-2025

ChemRxivChemical Space Exploration with Artificial ”Mindless” MoleculesWe introduce MindlessGen, a Python-based generator for creating chemically diverse, “mindless” molecules through random atomic placement and subsequent geometry optimization. Using this framework, we constructed the MB2061 benchmark set, containing 2061 molecules with high-level PNO-LCCSD(T)-F12 reference data for dissociation reactions. This set provides a challenging benchmark for testing, validation, and training of density functional approximations (DFAs), semiempirical methods, force fields, and machine learning potentials using molecular structures beyond the conventional chemical space. For DFAs, we initially hypothesized that highly parameterized functionals might perform poorly on this set. However, no consistent relationship between fitting strategy and accuracy was observed. A clear Jacob’s ladder trend emerges, with ωB97X-2 achieving the lowest mean absolute error (MAE) of 8.4 kcal·mol−1 and r²SCAN-3c offering a robust cost-efficient alternative (19.6 kcal·mol−1). Furthermore, we discuss the performance of selected semiempirical methods and contemporary machine learned interatomic potentials.
Evan Spotte-Smith (they/them)

I have some bittersweet news: I won't be starting my faculty career in the Chemical Engineering Department at Carnegie Mellon University. Come Sept., I will be joining the School of Chemistry at @ucddublin.bsky.social as an Ad Astra Fellow and Asst. Prof. of "digital chemistry". #ChemSky #CompChem

Bluesky SocialUniversity College Dublin (@ucddublin.bsky.social)Official source of news from University College Dublin (UCD) - Ireland's Global University. Est. 1854. One of Europe's leading research-intensive universities
ma𝕏pool

UMA: A Family of Universal Models for Atoms
ai.meta.com/research/publicati

family of Universal Models for Atoms (UMA), designed to push the frontier of speed, accuracy, and
generalization. UMA models are trained on half a billion unique 3D atomic structures (the largest
training runs to date) by compiling data across multiple chemical domains, e.g. molecules, materials,
and catalysts.

#ml#deepLearning#compChem
TrendingLive feeds
About