Kohn-Sham Density Functional Theory (DFT) has been the most successful tool to probe the electronic structure, mainly the ground-state total energies and densities of many condensed matter systems has led to the development of various databases such as Materials Project (MP), Inorganic Crystal Structure Database (ICSD), and many others. These databases ignited the interest of the material science community towards Machine Learning (ML), leading to the development of a new sub-field in material science called material-informatics, which aims to uncover the interrelation between known features and material properties. ML techniques can handle and identify relationships in complex and arbitrarily high-dimensional spaces data, which are almost impossible for human reasoning. Unlike DFT, the ML approach uses data from past computations or experiments. In many cases, ML models have shown their superiority over DFT in terms of accuracy and efficiency in predicting various physical and chemical properties of materials.
The incorporation of material property data obtained from atomistic simulations is crucial important to make continuous progress in data-driven methods. In this direction, we use DFT with Perdew-Burke-Ernzerhof (PBE), and Heyd–Scuseria–Ernzerhof (HSE) functionals, to introduce a family of mono-layer isostructural semiconducting tellurides M2N2Te8, with M = {Ti, Zr, Hf} and N = {Si, Ge}. These compounds have been identified to possess direct band gaps that are tunable from 1.0 eV to 1.3 eV, which are well suited for photonics and optoelectronics applications. Additionally, in-plane transport behavior is observed, and small electron and hole (0.11-015 Me) masses are identified along the dominant transport direction. High carrier mobility is found in these compounds, which shows great promise for applications in high-speed electronic devices. Detailed analysis of electronic structures reveals the presence of metal center bicapped trigonal prism as the structural building blocks in these compounds; a common feature in most of the group V chalcogenides helps to understand the atomic origins of promising properties of this unique class of 2D telluride materials.
Atomistic simulations based on DFT theory played a vital role in the development of data-driven materials discovery process. However, the resource-based constraints have limited the high-throughput discovery process by using DFT. The main motivation of our work towards the application of machine learning in material science is to assist the discovery process using available material property data in various databases. Incorporation of physical principles in a network-based machine learning (ML) architecture is a fundamental step toward the continued development of artificial intelligence for materials science and condensed matter physics. In this work, as inspired by the Pauling’s rule, we propose that structure motifs (polyhedral formed by cations and surrounding anions) in inorganic crystals can serve as a central input to a machine learning framework for crystalline inorganic materials. We demonstrated that an unsupervised learning algorithm Motif2Vec is able to convert the presence of structural motifs and their connections in a large set of crystalline compounds into unique vectors. The connections among complex materials can be largely determined by the presence of different structural motifs, and their clustering information is identified by our Motif2Vec algorithm. To demonstrate the novel use of structure motif information, we show that a motif-centric learning framework can be effectively created by combining motif information with the recently developed atom-based graph neural networks to form an atom-motif hybrid graph network (AMDNet). Taking advantage of node and edge information on both atomic and motif level, the AMDNet is more accurate than a single graph network in predicting electronic structure related material properties such as band gaps. The work illustrates the route toward the fundamental design of graph neural network learning architecture for complex materials properties by incorporating beyond-atom physical principles.
Due to the limitations in resources, it is not feasible to synthesize hundreds of thousands of materials listed in various databases by experiment or compute their detailed properties by using various electronic structure codes and state-of-the-art computational tools. Hence, the identification of an alternative route to screen such databases is very desirable. If identified, this route would be very helpful in reducing the material search space for any application. Categorizing materials based on their structural building blocks is very important to study the underlying physics and to understand the possible mechanisms for any application. Based on structure motifs, we purpose a novel way to categorize, analyze, and visualize the material space called a material network. The connection between any two nodes in this network is determined by using the calculated similarity value (Tanimoto-coeffecient) between each motif and its surrounding information, encoded in terms of a feature vector of length 64. By mapping a known compound, the network thus constructed can be used to screen compounds for the desired application. All the connections of the mapped compound are identified and extracted as a subgraph for further analysis. In our test screening for the transparent conducting oxides (TCO), the proposed network is successful in identifying compounds that are already listed as TCO in the literature. Thus, this indicates its usefulness in reducing the search space for the new TCO materials and various applications. This motif-based material network can serve as an alternate route for functional material discovery and design. / Physics
| Identifer | oai:union.ndltd.org:TEMPLE/oai:scholarshare.temple.edu:20.500.12613/355 |
| Date | January 2020 |
| Creators | banjade, Huta, 0000-0002-6074-5392 |
| Contributors | Yan, Qimin, Yan, Qimin, Perdew, John P., Ruzsinzsky, Adrienn, Carnevale, Vincenzo |
| Publisher | Temple University. Libraries |
| Source Sets | Temple University |
| Language | English |
| Detected Language | English |
| Type | Thesis/Dissertation, Text |
| Format | 151 pages |
| Rights | IN COPYRIGHT- This Rights Statement can be used for an Item that is in copyright. Using this statement implies that the organization making this Item available has determined that the Item is in copyright and either is the rights-holder, has obtained permission from the rights-holder(s) to make their Work(s) available, or makes the Item available under an exception or limitation to copyright (including Fair Use) that entitles it to make the Item available., http://rightsstatements.org/vocab/InC/1.0/ |
| Relation | http://dx.doi.org/10.34944/dspace/339, Theses and Dissertations |
Page generated in 0.0025 seconds