Discovering unusual structures from exception using big data and machine learning techniques

Abstract

a b s t r a c tRecently, machine learning (ML) has become a widely used technique in materials science study. Most work focuses on predicting the rule and overall trend by building a machine learning model. However, new insights are often learnt from exceptions against the overall trend. In this work, we demonstrate that how unusual structures are discovered from exceptions when machine learning is used to get the relationship between atomic and electronic structures based on big data from high-throughput calculation database. For example, after training an ML model for the relationship between atomic and electronic structures of crystals, we find AgO 2 F, an unusual structure with both Ag 3+ and O 2 2À , from structures whose band gap deviates much from the prediction made by our model. A further investigation on this structure might shed light into the research on anionic redox in transition metal oxides of Li-ion batteries.

Discovery of space aromaticity in transition–metal monoxide crystal Nb3O3 enabled by octahedral Nb6 structural units

“…Although it is crucial for understanding the properties of the solids, the structural units of them are usually inaccessible. In our previous work, we have applied machine learning to get the relationship between atomic and electronic structures and uncover unusual structural units from exceptions [38]. Our previous work on cathode material LiNi 0.5 Mn 0.5 O 2 has revealed that the d-p-d Möbius aromaticity within the fragmented Mn 6 -ring units contribute to stability and the interlayer ordering of the material [39].…”

Recent progress on discovery and properties prediction of energy materials: Simple machine learning meets complex quantum chemistry

“…Those inaccurate prediction results should not ascribe to algorithms but the lack of training data. Notably, Pan et al [3] developed a material structure database and studied the bandgap of material structure by ML method. Thirdly, developing the technical databases for characterizing energy materials is important.…”

“…Energy materials are materials used in energy harvesting, storage and conversion filed such as photovoltaics, batteries, fuel cells and photo(electro)catalysis, etc. Energy materials including intermetallic alloys, metal/covalent-or-ganic frameworks, carbon materials, perovskite and spinel materials play an important role in global economic development [1][2][3][4][5]. For instance, as an efficient ammonia synthesis materials, iron (Fe)-based catalysts invented in the early 20th century have promoted food production, which feeds billions of people [6][7][8].…”

Deep learning predicts boiling heat transfer

“…However, the boiling experiments in these studies are conducted on one-dimensional (1D) wires, which cannot represent the complex and volatile bubble motions associated with realistic two-dimensional (2D) or three-dimensional (3D) surfaces. Unfortunately, the results from many past models were hard to physically comprehend as they relied on abstract input features such as groups of pixels or principal components 51 . In addition to this, there have been no such an effort to practice machine learning based computer vision link bubble dynamics and boiling processes.…”