Marcel M

Our second agent, El Agente Estructural (“Structural” in Spanish) is a multimodal, natural-language–driven agent for molecular geometry generation and manipulation.

🔗 www.arxiv.org/abs/2602.048...
[1/7]

Chemical Space Exploration with Artificial “Mindless” Molecules We introduce MindlessGen, a Python-based generator for creating chemically diverse, “mindless” molecules through random atomic placement and subsequent geometry optimization. Using this framework, we ...

🚀 JCIM: "Chemical Space Exploration with Artificial Mindless Molecules"

We present MindlessGen, an open-source tool for generating chemically diverse "mindless" molecules, and the MB2061 benchmark set with high-level reference data to test methods on unconventional systems.

doi.org/10.1021/acs....

We are delighted to announce that our perspective article, “Steering towards safe self-driving laboratories (SDLs),” has been accepted for publication in Nature Reviews Chemistry.

Link: www.nature.com/articles/s41...

Headed to Accelerate 2025? You cannot miss the presentations from The Matter Lab. We've got your back--here's your ultimate cheat sheet so you don't miss a thing.

#Accelerate2025 #AIinScience
[1/2]

You can now use g-xTB @grimmelab.bsky.social with ORCA via the ExtOpt feature! Check out our new tutorial and learn how to use it in GOAT, NEB-TS and more.

www.faccts.de/docs/orca/6....

#ORCAqc #FACCTs #gxTB #CompChem #QuantumChem

We are working in this direction. However, analytical expressions for the nuclear gradient (or at least their implementation) get much more complicated in ab-initio methods, when using an atom-in-molecule-adaptive basis set.

Excited states-support is a feature that will also be available with g-xTB in the future (in the final implementation). Stay tuned! :)

📢 Update on our g-xTB release published on ChemRxiv:
We’ve uploaded a Linux executable of the current development version of g-xTB on GitHub, along with a simple usage guide:
🔗 github.com/grimme-lab/g...

⚠️ Please note:
This is a preliminary release — currently Linux-only, using numerical gradients.

GitHub - grimme-lab/g-xtb: Development versions of the g-xTB method. Final implementation will not happen here but in tblite (https://github.com/tblite/tblite). Development versions of the g-xTB method. Final implementation will not happen here but in tblite (https://github.com/tblite/tblite). - grimme-lab/g-xtb

You can try it directly here:

github.com/grimme-lab/g...

Happy to receive any feedback, particularly cases where it does not work as expected.

g-xTB excels in areas where SQM and even DFT often struggle:
✅ Transition-metal thermochemistry
✅ Spin-state energies
✅ Orbital energy gaps
✅ Reaction barriers
And all that at a fraction of DFT cost.

g-xTB is built to replace GFN2-xTB in all applications.
It cuts MAEs by half, improves SCF convergence, and even beats B3LYP-D4 for reaction barriers — all with just 30–50% more computational cost than GFN2-xTB.

Chemical Space Exploration with Artificial ”Mindless” Molecules We introduce MindlessGen, a Python-based generator for creating chemically diverse, “mindless” molecules through random atomic placement and subsequent geometry optimization. Using this framework, we ...

g-xTB is trained and validated on an extremely diverse molecular set — including actinides and "mindless molecules" (see also: chemrxiv.org/engage/chemr...)
Fully parameterized for Z = 1–103, it’s designed to perform reliably across the entire periodic table.

Some key highlights of g-xTB — our first general-purpose xTB method delivering DFT accuracy at SQM speed.
It tackles not only geometries, frequencies, and NCIs ("GFN"), but also strong thermochemistry and electronic properties with unprecedented accuracy for a semiempirical method.
🔗 #compchem

Two of them are at #WATOC2025 this week and ready to share all the details about the method you’ve been waiting for:
📍 @thfroitzheim.bsky.social — Thursday, Session B1, 9:20 AM
📍 S. Grimme — Thursday, Session A2, 10:20 AM

Don’t miss it!

Big thanks to my amazing co-workers: @thfroitzheim.bsky.social, Stefan Grimme, and Andreas Hansen! 🎉

g-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 ga...

After years of development and preparatory works which you might have seen on this profile, a major milestone is achieved:
g-xTB marks not just an evolution, but a revolution in the capabilities of semiempirical quantum chemistry. Convince yourself! A thread.
🔗 chemrxiv.org/engage/chemr...
#compchem

I see it more as a form of art 😂

I immediately loved the optical appearance of the molecules in this figure when I created it. 😂 But yeah, "unhinged" is very accurate! That's exactly what we wanted. 🤓

Chemical Space Exploration with Artificial ”Mindless” Molecules We 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.

#RobSelects preprint of the week #ChemRxiv: Benchmarking density functional approximations with a systematic set of randomly generated molecules. #compchem https://doi.org/10.26434/chemrxiv-2025-rdsd0

The Bond Capacity Electronegativity Equilibration Charge Model (EEQBC) for the Elements Z=1–103 The accurate and efficient assignment of atomic partial charges is crucial for many applications in theoretical and computational chemistry, including polarizable force fields, dispersion corrections, a...

Check out our new EEQBC model!

It delivers accurate and robust atomic charges for all elements up to Z=103. By incorporating bond capacitors, we eliminate most artificial CT while preserving the simplicity and efficiency of classical charge equilibration:

doi.org/10.26434/che...

#compchem

Say Hello to the Bannwarth group at Bluesky and give them a follow for great science! 👋 @bannwarthlab.bsky.social 🚀🔬

Thank you for your question! While an energy expression in the context of density-corrected DFT can still be conceptually very inspiring, we are currently working on a “real” xTB successor, called g-xTB.
This plot about the accuracy of the barrier heights compared to DFT gives a good impression. 💡

Thank you for your question! While an energy expression in the context of density-corrected DFT can still be conceptually very inspiring, we are currently working on a “real” xTB successor, called g-xTB.
This plot about the accuracy of the barrier heights compared to DFT gives a good impression. 💡

Our vDZP basis set utilized in the ⍵B97X-3c composite DFT method is now also available via www.basissetexchange.org (API-based: github.com/MolSSI-BSE/b...). 🎉
Many thanks to @Susi Lehtola & coworkers for jointly providing it there!

This is a question I can only answer with a certain bias, as we are actively developing xTB and related tight-binding methods (which have their roots in DFTB). From this point of view, I would answer “No, xTB has become the standard, at least for molecular systems with less than about 2000 atoms.” 🤓

2. See this answer: bsky.app/profile/mrcl...
Thus, I consider models that have a built-in quantum chemical foundation in them as semiempirical (in the sense of theoretical chemistry/quantum chemistry methods).

1. Good suggestion. I wasn't sure if I should consider it semiempirical, since in a sense it could also be a precursor to KS-DFT.

Personally, I would consider the idea of machine-learning potentials or force fields as empirical (not semiempirical), since they derive their behavior mainly from the emulation of reference data (e.g. DFT) and carry only a limited amount of physics (e.g. no quantized energy levels).

Happy to announce the release of @avogadro.cc 1.100 (not quite 2.0 yet) with a pile of new features, bug fixes, etc.

Can't fit the whole release notes here, but more on the forum:
discuss.avogadro.cc/t/avogadro-1...

Thanks for the remark, I will add his name to the NDO part.