A research team from MIPT, Skoltech, and Dukhov Research Institute of Automatics, headed by ArtemOganov, employed a machine learning technique for modeling the behavior of uranium and aluminum in the crystalline and liquid phases at various pressures and temperatures.
This is a slider. Credit: MIPT Press Office
Such simulations of chemical systems are capable of predicting their properties under a variety of conditions prior to performing experiments, enabling further work with only the materials considered to be the most promising. The research findings have been published in the journal Scientific Reports.
Prompt advances in science over the last 100 years have led to the discovery of an astonishing number of inorganic and organic compounds, chemical reactions, and protein and lipid structures. However, with all these new structures and molecules, a great amount of time is needed for studying their makeup, physical and biochemical properties, and also for testing the models of their behavior under different conditions and their possible interactions with various other compounds. Such research can presently be accelerated by the use of computer modeling.
The modeling technique that is currently dominant refers to the force field approach. It uses a set of parameters describing a given biochemical system. These comprise of bond lengths and angles, and charges, among others. However, this technique is not capable of accurately reproducing the quantum mechanical forces at play in molecules. Accurate quantum mechanical calculations are considered to be time-consuming. Besides, they only allow predictions of the behavior of samples that are at best several hundred atoms large.
Chemists are greatly interested in machine learning approaches to molecular modeling. They enable models that are trained on comparatively small data sets obtained through quantum mechanical calculations. Such models are then capable of replacing quantum mechanical calculations, as they are just as accurate and need about 1,000 times less computing power.
Progress made by machine learning tools modeling interactions between atoms
The Researchers employed machine learning for modeling the interactions between atoms in crystalline and liquid aluminum and uranium. Aluminum is considered to be a well-studied metal whose chemical and physical properties are known to Scientists. By contrast, uranium was chosen as there is conflicting published data on its chemical and physical properties, which the Researchers sought to define more accurately.
The paper presents an in depth study of such material properties as the phonon density of states, entropy and the melting temperature of aluminum.
The magnitudes of interatomic forces in crystals can be used to predict how atoms of the same element will behave under different temperatures and in a different phase. By the same token, you can use the data on the properties of a liquid to find out how the atoms will behave in a crystal. This means that by finding out more about the crystal structure of uranium, we can eventually reconstruct the entire phase diagram for this metal. Phase diagrams are charts indicating the properties of elements as a function of pressure and temperature. They are used to determine the limits to the applicability of a given element.
Ivan Kruglov, the Computational Materials Design Laboratory, MIPT
The Researchers compared experimental results in order to ensure that the data yielded by computer simulations is valid. The method employed by the team was in good agreement with earlier experiments. The information obtained with the approach based on machine learning had a lower error rate when compared to the modeling techniques making use of force fields.
In this study, the Authors bring about improvements on their 2016 results in terms of the speed and accuracy of atomic system modeling using machine learning.
The Russian Science Foundation supported the study reported in this article.