Aitomistic provides cutting-edge AI solutions for atomistic simulations, built on decades of expertise in quantum chemistry, machine learning, and computational science. Our tools empower researchers by making simulations accurate, fast, and accessible, supported by flexible deployment options and tailored training programs—all developed with the highest scientific rigor to bridge advanced research and practical applications.​

What is Aitomistic

6月18日.mp4

Have you ever wondered how to train a neural network potential if you have several datasets generated at different levels of theory? Our all-in-one learning strategy elegantly solves this problem by simultaneously training a potential on different datasets. The resulting potentials can make predictions targeting different levels of theory.

medium_1764676622115.jpg

small_1764676622115.jpg

thumbnail_1764676622115.jpg

1764676622115.jpg

, where we used it to train a universal potential OMNI-P1, which can make predictions at both DFT (wB97X/def2-TZVPP) and coupled cluster (CCSD(T)*/CBS) levels. OMNI-P1 is included in the 

medium_1764676598758.jpg

small_1764676598758.jpg

thumbnail_1764676598758.jpg

1764676598758.jpg

The all-in-one learning strategy is general and can be applied for training on more levels of theory. In many respects, all-in-one learning is a better alternative to transfer learning, which requires several stages of training and yields two different models.

Due to the uniqueness of the OMNI-P1 model, you can choose the target level of theory to make predictions at. Default is CCSD(T)*/CBS as a better level. After you choose the level, you can use the model for typical simulations, such as 

Below is an example for geometry optimizations in a Python script:

initmol = ml.molecule.from_xyz_file('init.xyz')

omnip1 = ml.methods(method='omnip1', level='cc')

geomopt = ml.optimize_geometry(model=omnip1, initial_molecule=initmol)

final_mol.write_file_with_xyz_coordinates(filename='final.xyz')

The simulations with OMNI-P1 can be run online on the 

, as usual. This version only works for the CHNO elements.

Please also let us know what you would like us to improve to make your experiences with MLatom and Aitomistic better. Simply reply to this email or 

large_thumbnail_2_04.jpg

medium_thumbnail_2_04.jpg

small_thumbnail_2_04.jpg

thumbnail_thumbnail_2_04.jpg

thumbnail_2_04.jpg

Our all-in-one learning strategy elegantly solves this problem by simultaneously training a potential on different datasets.

Aitomistic Updates: One to Rule Them All——A Universal Interatomic Potential Learning across Quantum Chemical Levels

In this week’s newsletter, we’d like to highlight a powerful feature of MLatom (an 

 supported by Aitomistic) as a tool for analysis of your calculations, e.g., with Gaussian. You can use prepare beautiful Jupyter notebooks and convert them into interactive the html files to browse together with your colleagues the molecules and spectra. You can use these materials instead of or in addition to your Supporting Information files in PDF. For this, 

An example of this functionality is a web version of such a Jupyter Notebook for one of our publications: 

. In our study Alternating Donor–Acceptor Thienoacenes Featuring Up to 23 Linearly Fused Rings published in Org. Lett., earlier this year, we used the UAIQM extension of the AIQM2 method to optimize the 714-atom structure in ca. 88 min on 32 CPUs. Good luck doing that with DFT! The agreement with the crystallographic structure was quite good, given the flexibility of the noncovalently-bound complex with many alkyl groups. In this study, TD-DFT calculations were performed on smaller systems with Gaussian.

medium_8a17c02a-cca7-40fe-acfd-8756a9d91fc9.png

small_8a17c02a-cca7-40fe-acfd-8756a9d91fc9.png

thumbnail_8a17c02a-cca7-40fe-acfd-8756a9d91fc9.png

8a17c02a-cca7-40fe-acfd-8756a9d91fc9.png

MLatom was then used to load the Gaussian output files and convert them to the molecule format as mol = ml.molecule.load(’myfile.log’, format=’gaussian’). Why is this important? It gives you a simple way to convert your old Gaussian calculations into a machine-readable database format with MLatom, which can be later used for training ML models. It is also easy to analyze calculations once loaded in Jupyter Notebook with MLatom, you can, e.g., view molecules by running a command in a cell mol.view().

 as a template for your projects. It shows how to prepare summaries of calculations listing energies and thermochemical properties, number of imaginary frequencies, etc.

medium_61b0f42f-8f2b-4148-b65a-14a4bb6d3a81.png

small_61b0f42f-8f2b-4148-b65a-14a4bb6d3a81.png

thumbnail_61b0f42f-8f2b-4148-b65a-14a4bb6d3a81.png

61b0f42f-8f2b-4148-b65a-14a4bb6d3a81.png

One particularly useful application is to analyze the UV/vis absorption spectra, which you can plot with MLatom’s built-in functions such as ml.spectra.plot_uvvis(molecule=mol, ...) (see tutorial) and compare to the experimental spectra, with functions for spectra alignment and plotting line spectra. Examples of how to print out the Tables for your SI with the list of excitations and their character are provided too.

large_dc0b7f92-4e2e-47a7-b56c-d339ceaec121.png

medium_dc0b7f92-4e2e-47a7-b56c-d339ceaec121.png

small_dc0b7f92-4e2e-47a7-b56c-d339ceaec121.png

thumbnail_dc0b7f92-4e2e-47a7-b56c-d339ceaec121.png

dc0b7f92-4e2e-47a7-b56c-d339ceaec121.png

Please also let us know what you would like us to improve to make your experiences with MLatom and Aitomistic better. J

In this week’s newsletter, we’d like to highlight a powerful feature of MLatom (an open-source software supported by Aitomistic) as a tool for analysis of your calculations, e.g., with Gaussian. 

Aitomistic Updates: Tools for analyzing your simulations

In this weekly briefing, you will learn about the aitomistic updates, including implementations in the newly released open-source 

, and a preview of upcoming course updates and features for premium users.

Starting from MLatom 3.19.1, which we released this week, you can perform optimization of the conical intersections. Mikolaj implemented the penalty-constraint approach of Levine et al. [J. Phys. Chem. B 112, 405–413 (2008)] and 

 for you, which you can run on Aitomistic Hub as usual. 

large_1763558980111.jpg

medium_1763558980111.jpg

small_1763558980111.jpg

thumbnail_1763558980111.jpg

1763558980111.jpg

From this tutorial, you will learn that in this implementation, the geometry is optimized to minimize the average energy of two electronic states while penalizing the energy gap between them. It is an approximated method, which is particularly useful if you do not have nonadiabatic couplings (NACs). We have already used this method in 

The tutorial shows two examples: optimization of conical intersection of fulvene with a pre-trained MS-ANI model and cyclohexadiene with a semi-empirical ODM2 method (the method I have developed when I was a post-doc in Max Planck Institute with late Walter Thiel). As usual with MLatom, you can visualize the optimized geometries and check their geometric parameters seamlessly in the Jupyter Notebook. Optimization of conical intersections is now as simple as:

ml.ci_opt(molecule=fulvene, model=mlmodel)

We keep improving open-source MLatom’s molecular dynamics engine, and Yi-Fan has included his 

 in the MLatom 3.19.1 release. You can do dynamics as usual by defining thermostat as, e.g.:

Langevin = ml.md.Langevin_thermostat(temperature=300,molecule=init_mol,tau=0.001)

New publication on active learning with MLatom

Last Tuesday, a paper by Matheus de Oliveira Bispo and co-workers was published describing the MELTS program, interfacing Newton-X and MLatom to perform active learning to enable ML-accelerated surface-hopping dynamics. This program is of particular interest to the big Newton-X community.

medium_1763559306232.jpg

small_1763559306232.jpg

thumbnail_1763559306232.jpg

1763559306232.jpg

Matheus de Oliveira Bispo, Rafael Souza Mattos, Max Pinheiro Jr, Bidhan Chandra Garain, Pavlo O. Dral, Mario Barbatti. 

Interested in the ins and outs of active learning?

Active learning for nonadiabatic dynamics on 

Active learning for ground-state simulations on 

Our Premium users enjoy early access to the updates. Among the new updates:

Fine-tuning of OMNI-P2x for your excited-state problem, enabling precise UV/vis spectra and more

Fine-tuning of AIQM models for your tasks

 on computational chemistry and AI, with webinars

More tokens for autonomous simulations with our Agents

Interested in becoming a Premium user? Please contact us!

Please also let us know what you would like us to improve to make your experiences with Aitomistic better. Simply reply to this email or 

In this weekly briefing, you will learn about the aitomistic updates, including implementations in the newly released open-source MLatom 3.19.1, a new publication, and a preview of upcoming course updates and features for premium users.

Aitomistic Updates: Conical Intersection Optimization, Langevin Thermostat MD, and New Active Learning Research

Here is your weekly dose of exciting updates around aitomistic simulations from our team, including IRC, a video lecture and new publication, and preview of upcoming features.

IRC (intrinsic reaction coordinate) calculations

Many of you use Aitomistic for performing transition state calculations with 

 and other methods. The next logical step is performing IRC calculations on the obtained structure – something that was unavailable for online computations until recently. We have listened to your requests, and we are happy to share that this essential feature is now available for online computations and in 

, implemented by Xinxin – the author of AIQM2!

IRC calculations are as simple as a three-line input file or a couple of lines in a Python script, with rich features for plotting and analysis. A 

We are continuously improving Hub & Lab to make the user experience smoother. Here, your feedback is crucial - if you do not tell us that something is not working for you, we will not be able to make it work. Recently, we have substantially simplified operations with files and folders so that you can now easier manage them, e.g., copy and paste, duplicate, rename, etc, with as little hassle as possible.

thumbnail_Weixin Image_2025-12-03_170220_579.png

Weixin Image_2025-12-03_170220_579.png

Release of Aitomic add-ons to MLatom, version 0.92

We have also upgraded add-ons to MLatom, which now contain 

. The new method retains the performance of AIQM2 but is extended to more elements, now covering CHNOFSCl and targeting coupled-cluster accuracy. Upgrade includes many further improvements such as support of higher versions of PyTorch. 

large_1763040362474.jpg

medium_1763040362474.jpg

small_1763040362474.jpg

thumbnail_1763040362474.jpg

1763040362474.jpg

We have also updated its installation instructions at 

On Monday, a paper describing a novel methodology for machine learning of nonadiabatic couplings with high accuracy based on physics-informed descriptors (energy gradient difference) appeared in JPCL. Methodology is based on implementations in MLatom, including FSSH (fewest switches surface hopping), which can be performed entirely with ML, with QM, and in hybrid ML/QM combinations as implemented by Jakub (paper’s first author).

small_1763040449203.jpg

thumbnail_1763040449203.jpg

1763040449203.jpg

Jakub Martinka, Lina Zhang, Yi-Fan Hou, Mikołaj Martyka, Jiří Pittner*, Mario Barbatti*, 

, 16, 11732–11744. DOI: 10.1021/acs.jpclett.5c02810.

If you wish to learn more about MD (molecular dynamics) with ML, which you can do on Aitomistic Hub & Lab, we have given an 

. The lecture covered many important topics, such as pitfalls and state-of-the-art of machine learning interatomic potentials including the universal models and their fine-tuning. In addition, the lecture has covered emerging, radically different ways of performing and analyzing molecular dynamics by 

? You can subscribe to watch more videos.

In the coming days and weeks, we will roll out several big upgrades such as improved AI agents, optimization of conical intersections, upgraded OMNI-P2x models for excited-state simulations with tutorials about UV/vis spectra simulations and fine-tuning for NAMD, etc. Stay tuned!

Aitomistic Updates: IRC Calculations, AIQM3, and New ML Research

Truly amazing: Autonomous computational chemistry calculations done by AI are no longer a distant future; it is a vibrant present. Soon, most such computations will be done by AI.

large_Weixin Image_20250821122627.png

medium_Weixin Image_20250821122627.png

small_Weixin Image_20250821122627.png

thumbnail_Weixin Image_20250821122627.png

Weixin Image_20250821122627.png

In our demo below, the user prompts our Aitomia on the Aitomistic Hub to calculate the reaction energy and thermodynamics of Diels-Alder reaction for ethene and 1,3-butadiene. Aitomia accomplishes these calculations from scratch, even without the provided initial coordinates of the reactants and products.

large_Weixin Image_20250821122619.png

medium_Weixin Image_20250821122619.png

small_Weixin Image_20250821122619.png

thumbnail_Weixin Image_20250821122619.png

Weixin Image_20250821122619.png

The time to solution was just 5 min on a single CPU, delivering accurate geometries and energies calculated with our state-of-the-art AIQM2 model.

The same exercise took our expert team member to do in 20+ minutes from scratch by setting up manually initial xyz files for all involved species, performing, and analyzing the calculations. The results were the same, but the human cost was much higher.

This Deep Thinking mode, enabling increasingly complex autonomous computational chemistry simulations, is already enabled for selected users on the

. Autonomous simulations with Aitomia are available on Aitomistic Hub since May 11, 2025, for free for all registered users. See 

This Deep Thinking mode, enabling increasingly complex autonomous computational chemistry simulations, is already enabled for selected users on the Aitomistic Hub.

Elevate your aitomistic simulations
Atomistic simulationsPowered by AI

What is Aitomistic

Products

News

Aitomistic Updates: One to Rule Them All——A Universal Interatomic Potential Learning across Quantum Chemical Levels

Aitomistic Updates: Tools for analyzing your simulations

Aitomistic Updates: Conical Intersection Optimization, Langevin Thermostat MD, and New Active Learning Research

Aitomistic Updates: IRC Calculations, AIQM3, and New ML Research

Aitomia: Reaction simulation in 5 minutes without human help

Elevate your aitomistic simulationsAtomistic simulationsPowered by AI

What is Aitomistic

Products

News

Aitomistic Updates: One to Rule Them All——A Universal Interatomic Potential Learning across Quantum Chemical Levels

Aitomistic Updates: Tools for analyzing your simulations

Aitomistic Updates: Conical Intersection Optimization, Langevin Thermostat MD, and New Active Learning Research

Aitomistic Updates: IRC Calculations, AIQM3, and New ML Research

Aitomia: Reaction simulation in 5 minutes without human help

Elevate your aitomistic simulations
Atomistic simulationsPowered by AI