As machine learning becomes increasingly important in science and engineering, it holds the promise to provide a universal approach applicable to various systems to investigate their crystalline phase transitions. Here, we build and train accurate artificial neural networks that can distinguish tiny energy difference, which is crucial to predict the crystalline phase transitions. Employing the trained artificial neural networks in Monte Carlo simulations as the surrogate energy function, we apply this approach to monochalcogenides, including bulk and two-dimensional monolayer SnTe and GeTe, investigating their crystalline phase transitions. The machine-learning approach, when viewed as providing a universal mathematical structure, can be transferred to the investigation of other materials when the training data set generated with ab initio methods are available. Moreover, the machine-learning approach resolves the difficulties associated with constructing the effective Hamiltonian for monochalcogenides, showing great potential with its accuracy and efficiency.
The huge axial ratio of supertetragonal perovskites can induce large polarization due to the coupling between the unit cell and ion displacements, which often results in higher Curie temperature or better ferroelectric properties. With the progress of vertical strain engineering, experimentally available supertetragonal raises much interest in the relation between its intrinsic polarization and structural characteristics, such as the axial ratio and ion displacements that can be obtained with a high-resolution (scanning) transmission electron microscope nowadays. However, the interpretation and use of such obtained structural information are not without peril. Here, employing first-principles calculations, we report on the relationship between the polarization and the structural information for supertetragonal . Our investigation shows that, unlike normal , the strong sublinear correlation raises the difficulty to quantitatively estimate the spontaneous polarization for the axial ratio larger than 1.15. In addition, the relative displacement between cations (i.e., Pb and Ti) proves to be an unsuitable structural parameter to determine the spontaneous polarization, emphasizing the importance of obtaining the positions of anions, albeit presumably more difficult, along with cations.
Relaxor ferroelectrics, which form a peculiar class of functional materials, are often composed of complex perovskites , as represented by where the compositional ordering of Mg and Nb is believed to be essential to its relaxor properties. In this work, analysis using a first-principles-based model shows that, while the electrostatic interactions are important, the nearest neighbor assumption, which was used for metallic alloys, can be adopted to understand the compositional ordering in . Numerical simulations with the Kawasaki Monte Carlo method can model the experimentally observed compositional ordering by maximizing the number of the unlike pairs (or the Bethe's parameter), which is the overriding factor that determines the ordering. Subtle points of configuration energy degeneracy are also discussed, which explains the partial disorder inherently present in such systems.
25 July 2022 in Journal of Physics: Condensed Matter
The prediction of magnetic phase transitions often requires model Hamiltonians to describe the necessary magnetic interactions. The advance of machine learning provides an opportunity to build a unified approach that can treat various magnetic systems without proposing new model Hamiltonians. Here, we develop such an approach by proposing a novel set of descriptors that describes the magnetic interactions and training the artificial neural network (ANN) that plays the role of a universal magnetic Hamiltonian. We then employ this approach and Monte Carlo simulation to investigate the magnetic phase transition of two-dimensional monolayer chromium trihalides using the trained ANNs as energy calculator. We show that the machine-learning-based approach shows advantages over traditional methods in the investigation of ferromagnetic and antiferromagnetic phase transitions, demonstrating its potential for other magnetic systems.