Before designing any device, such as a laser, physicists need to determine whether their chosen material is suitable for the specific task. This involves calculating the electronic band structure, which describes the energy states of electrons and determines the material's optical, electrical, mechanical, and other physical properties. Electronic band structure can be determined with first-principles (ab initio) calculations; they do not require experimental data and are based on the fundamental laws of physics.
However, not every physicist is capable of performing first-principles calculations. This is taught separately, including preparing files for calculations, configuring and working with specialized software. Even if a research group has a theoretical physicist with first-principles calculation skills, determining the electronic band structure of a material with a large number of electrons can take an entire day.
Researchers at the School of Physics and Engineering, supported by a developer from ITMO’s Institute of Applied Computer Science, have devised AI-Materials Project, an AI assistant that simplifies first-principles calculations and makes them accessible to a wide range of specialists. It is a chatbot based on a RAG system and a ready-made LLM model, enhanced with the specialized open database of crystalline materials, The Materials Project. This database contains information on the structure of each material, describing the chemical elements and the arrangement of their atoms relative to each other. Based on the crystal structure, the system computes the electronic band structure as well as the optical, electrical, and other physical properties of a material.
AI-Materials Project. Video courtesy of Ivan Iorsh
Thanks to AI-Materials Project, users no longer need to download specialized software or write API requests to obtain information from the database. The AI assistant itself “translates” scientists’ requests from natural language into database queries. For example, you can ask it to find semiconductors with a specific bandgap width, sort perovskites with a specific formula by stability, or list all crystallographic modifications of the mineral fluorite. Separately, for the material of interest, you can calculate the electronic band structure and find out its physical properties. The chatbot returns results in a short period of time – from a few minutes to half an hour, depending on the complexity of the calculation.
“With our AI-Materials Project, scientists who do not major in theoretical physics calculations, professional software, or API query skills can perform several tasks in a single window in a short period of time. Firstly, they can request information from the specialized database. The results will be more accurate, as The Materials Project is trusted by scientists around the world, whereas general AI assistants may hallucinate. Secondly, based on verified information, physicists can run first-principles calculations of material properties or send other complex, multi-component queries to the chatbot. As a result, our solution significantly speeds up the research pipeline, both in preliminary calculations and in preparing illustrations for publication,” says Ivan Iorsh, the project lead, chief researcher at ITMO’s School of Physics and Engineering.
Ivan Iorsh. Photo by ITMO University
With AI-Materials Project, physicists will be able to determine material characteristics faster and more accurately (for example, conductivity, magnetic, optical, and mechanical properties, bandgap width, and light-emission capability). From all the options, they can choose suitable materials for research in laser technologies, optics, thermoelectrics, and microelectronics.
In the future, the developers plan to add several more features to make AI-Materials Project a full-fledged tool in a single window. First, they intend to implement a search feature for scientific literature in the field of materials physics in the ArXiv database. It is physically impossible for a single scientist to keep track of all publications from the past 20 years, so this new feature will help conduct in-depth analysis of scientific literature. Based on this, scientists will be able to identify “white spots” in their field of research and then verify them using newly obtained but not yet interpreted experimental data. The second new feature will be the use of variational algorithms to model quantum materials in quantum chemistry.
The developers presented AI-Materials Project at the 5th Congress of Young Scientists in Sirius. This was a key event of the Decade of Science and Technology, which gathered more than 7,000 participants from about 60 countries – representatives of the academic and university community, state corporations and private businesses, and public associations.
"Our goal is to spark interest in this development among researchers working in the field of materials science. The more scientists start using the system and share their feedback, the faster we will be able to improve it to meet user requirements. At the Congress of Young Scientists, we hope to establish partnerships with companies and research groups advancing AI in scientific discovery," says Ivan Iorsh.
AI-Materials Project is not the only digital assistant created at ITMO. ChemCoScientist, developed by the team of Anna Kalyuzhnaya, a senior researcher at ITMO’s Research Center “Strong AI in Industry,” helps chemists automate scientific research. It can generate new chemical compounds and refine existing ones, develop synthesis algorithms, predict chemical properties, and extract knowledge from chemistry articles.
Translated by Anna Butko
