Researcher uses AI to predict probabilities for chemical processes

Dr. Michael F. Peintinger, from the Mulliken Center for Theoretical Chemistry at the University of Bonn, has been exploring the potential of artificial intelligence in quantum chemistry. His research focuses on how AI can be utilized to predict the probabilities of various chemical processes.

A crucial aspect of chemical properties is the electronic structure, which pertains to the distribution and behavior of electrons within a substance. These electrons dictate whether a material conducts electricity, the color of a dye, the stability of a crystal, and the speed of chemical reactions.

Peintinger's work aims to understand the adsorption of molecules onto solid surfaces, such as water on a thin layer of boron nitride. This process involves placing the molecule at different locations within a model, orienting it in various ways, and comparing the energies. The arrangement with the lowest energy is deemed the most favorable.

Quantum chemistry, a branch of chemistry, applies the principles of quantum mechanics to chemical problems. It involves calculating the probability of an electron being found at a specific location, as electrons do not follow fixed paths. However, solving these equations exactly is only feasible for a single electron, such as in a hydrogen atom. In more complex systems, like diamond, many electrons interact, necessitating approximations that depend on the specific question being addressed.

Vibe-qc, a program developed almost entirely by generative AI, is a significant breakthrough. It represents the first program of its kind, created through a system where AI agents, or digital programmers, write and test the code under scientific guidance. This process was remarkably swift, with the first comprehensive version ready after approximately 10 weeks.

However, Peintinger emphasizes that while the program functions well, it must undergo thorough reviews and comparisons with independent software to ensure its accuracy. The program's accuracy depends on the assumptions made in the model, and its predictions must be verified against experimental data.

The development of vibe-qc has made it possible for a single researcher to run large-scale quantum chemistry programs, previously requiring extensive programming effort. The program is open-source, with its source code freely available for anyone to use, modify, and distribute under the Mozilla Public License.