Prof. JIANG Bin’s research team at the University of Science and Technology of China (USTC) has made a groundbreaking discovery in the field of atomic simulations. Their development of the universal field-induced recursively embedded atom neural network (FIREANN) model has revolutionized the study of atomic systems by accurately simulating system-field interactions with exceptional efficiency. This research, published in Nature Communications on October 12, has the potential to significantly advance our understanding of complex chemical, biological, and material systems at the microscopic level.
Atomic simulations play a crucial role in unraveling the mysteries of various systems, enabling us to comprehend their spectra and dynamics. A key aspect of these simulations is accurately representing high-dimensional potential energy surfaces (PESs). In recent years, atomistic machine learning (ML) models have been widely adopted for this purpose. However, traditional ML models only consider isolated systems and fail to capture the complex interactions between external fields and these systems. External fields can have a profound impact, altering chemical structures and controlling phase transitions through field-induced electronic or spin polarization. Therefore, there is an urgent need for a novel ML model that incorporates external fields.
In response to this challenge, Prof. JIANG’s research team proposed an innovative all-in-one approach. They began by treating the external field as virtual atoms and employed embedded atom densities (EADs) as descriptors for the atomic environment. By deriving field-induced EADs (FI-EADs) from a linear combination of field-dependent orbital and coordinate-based orbitals of atoms, they successfully captured the essence of the interaction between external fields and systems. This led to the development of the FIREANN model.
The unique feature of the FIREANN model is its ability to accurately correlate various response properties of a system—such as dipole moment and polarizability—with potential energy changes that depend on external fields. This breakthrough provides a highly accurate and efficient tool for simulating the spectroscopy and dynamics of complex systems under external fields.
To verify the capability of the FIREANN model, the research team conducted dynamic simulations of N-methylacetamide and liquid water under the influence of a strong external electric field. The results demonstrated both high accuracy and efficiency. Importantly, for periodic systems, the FIREANN model overcomes the inherent multiple-value issue of polarization by training solely using atomic forces data.
The significance of this research lies in its ability to address the lack of accurate external field representation in ML models. By filling this crucial gap, the FIREANN model opens up new possibilities for molecular simulations in the fields of chemistry, biology, and materials science.
In conclusion, Prof. JIANG Bin’s research team at USTC has achieved a major breakthrough in atomic simulations with their development of the FIREANN model. By accurately simulating system-field interactions, this model enables a deeper understanding of complex systems at the microscopic level. This pioneering research has the potential to advance scientific exploration in chemistry, biology, and materials science, and pave the way for future discoveries.