Researchers at Aitomistic has published a 

 in npj Computational Materials presents an efficient ML protocol for accelerating trajectory surface hopping dynamics, while tackling many key issues making machine learning of excited states difficult. The protocol introduces a new machine learning interatomic potential based on multi-state learning and accelerated sampling of the critical regions near conical intersections. All the methods, together with the end-to-end active learning are available in 

Quantum chemical simulations of molecular excited states are important for understanding the photochemical and photophysical processes, but they are very computationally intensive, especially when used for nonadiabatic dynamics. Machine learning is an attractive tool to tackle this problem by replacing expensive quantum chemical calculations, however, the progress in ML for excited states is much slower than for ground states due to the complexity of the former (see, e.g., 

large_toc-graph-with_text.webp

medium_toc-graph-with_text.webp

small_toc-graph-with_text.webp

thumbnail_toc-graph-with_text.webp

toc-graph-with_text.webp

To address the key challenges in machine learning of excited states, this work has reported a comprehensive and robust protocol for ML based on an eight-year-long (!) in-depth research. This protocol is based on a multi-state learning model, MS-ANI, capable of learning excited states of various molecules with unprecedented accuracy. The core idea is to take the state number as an input feature, concatenated with the ANI atomic environment vectors, and passed to the neural network for the corresponding element type. Impressively, the accuracy of learning excited states is better or similar to the accuracy of learning ground-state energies, where information on excited-state energies improves the quality of ground-state predictions. 

small_image-1.webp

thumbnail_image-1.webp

image-1.webp

One of the key ingredients to train a robust potential for nonadiabatic dynamics is accelerated sampling of the regions near conical intersections. We have achieved this in this work by introducing gapMD, which ensures enhanced sampling of the low-gap regions by propagating trajectories with the gradient energy difference between adjacent surfaces when needed.

small_image-2.webp

thumbnail_image-2.webp

image-2.webp

Together, these advances enabled end-to-end active learning for generating machine learning models to perform nonadiabatic simulations.

small_image-3.webp

thumbnail_image-3.webp

image-3.webp

For simple cases such as fulvene, this active learning is very fast and converges just after three days on a single GPU card.

large_image-4.webp

medium_image-4.webp

small_image-4.webp

thumbnail_image-4.webp

image-4.webp

Active learning also works for more complex cases and, e.g., can be used to calculate photoreaction quantum yields. We have also discovered long-time-scale oscillations in cis-azobenzene photoisomerization.

small_image-5.webp

thumbnail_image-5.webp

image-5.webp

In this work, we have also shown for the first time that a single model can be used to perform the nonadiabatic dynamics of different molecules. This has laid the foundation for our follow-up study on the first-ever universal excited-state potential OMNI-P2x.

Video lecture about the theory and life-recorded tutorial:

Mikołaj Martyka, Lina Zhang, Fuchun Ge, Yi-Fan Hou, Joanna Jankowska*, Mario Barbatti*, 

 npj Comput. Mater. 2025, 11, 132. DOI: 10.1038/s41524-025-01636-z.

large_npj.png

medium_npj.png

small_npj.png

thumbnail_npj.png

npj.png

The protocol introduces a new machine learning interatomic potential based on multi-state learning and accelerated sampling of the critical regions near conical intersections. 

npj Comput. Mater.: Efficient Machine Learning Protocol For Accelerating Trajectory Surface Hopping Dynamics

 with on-demand resources such as AI models, QM methods, ML and QM software for aitomistic simulations, GUIs, learning materials, and much more. On the Hub, you can run aitomistic simulations without installing any software and collaborate with your colleagues. Aitomistic Hub is a perfect starting point, with free trial resources subject to their availability.

We understand that the trial resources might not satisfy your aitomistic simulation needs. Hence, we offer flexible solutions based on your requirements and resources. You may choose appropriate subscriptions to more resources, i.e., CPU and GPU time, storage, and unlocking more features.

If you like Aitomistic Hub but want to use your hardware recourses with the local, customized, and private version of Aitomistic Hub, we have got you covered too! Check out our Aitomistic Lab powering the Hub.

Last, but not least: do you have a cool software that you want to showcase to others and make available for calculations online? Contact us, and it might be integrated into the Aitomistic Hub.

small_Screenshot 2025-05-06 at 12.18.20.png

thumbnail_Screenshot 2025-05-06 at 12.18.20.png

Screenshot 2025-05-06 at 12.18.20.png

Beta version is live! Try it now and be among the first to explore the latest features—early adopters who provide feedback will get a chance at free perks and discounts for future versions!

Aitomistic Hub Launched!

Are universal machine learning potentials for excited states possible?

Such a potential would be a major breakthrough — enabling key applications like the design of advanced photomaterials.

We’ve already seen successful universal potentials for ground states — ANI-1ccx, MACE-OFF, our own 

 — you name it. Some of these go back to 2017.

But 8 years later, not a single universal potential exists for excited states. So… are there fundamental obstacles preventing them?

 with quality approaching TD-DFT — but in milliseconds, orders of magnitude faster.

 — the simulation finishes in a flash. You can try it yourself online, either on Aitomistic Hub or with a Jupyter notebook on the 

affordable excited-state molecular dynamics

, something existing ML models struggle with.

Want us to go deeper? A full tutorial or even a live Q&A session? Let us know in the comments!

As with any new approach, questions remain —

large_thumbnail_2_04.png

medium_thumbnail_2_04.png

small_thumbnail_2_04.png

thumbnail_thumbnail_2_04.png

thumbnail_2_04.png

Are universal machine learning potentials for excited states possible? We’re excited to share: they are possible. We’ve built the first-ever universal excited-state potential — OMNI-P2x.

News

npj Comput. Mater.: Efficient Machine Learning Protocol For Accelerating Trajectory Surface Hopping Dynamics

Aitomistic Hub Launched!

Meet OMNI-P2x — the First Universal ML Potential for Excited States!