NCMN/EQUATE Seminar w/ Arti Kashyap
DFT and DFT based Material Databases in the Era of Machine Learning
4:00 pm –
5:00 pm
Jorgensen Hall Room: 110
Abstract:
Machine learning has successfully established itself in a variety of industries, ranging from self-driving automobiles to fraud detection. The materials science community has also started to benefit from it. This has been possible because of the available large number of datasets for materials and their properties, created by high-throughput DFT calculations where an automated input can perform thousands of computations in a single shot. Not only DFT has proven to be a very robust tool for understanding the electronic structure, but the amount of data it has provided for machine learning-based material design is highly valuable.
In the first part of this talk, I will present the use of DFT with an example of e- Fe2O3. It is one of the polymorphs of the most abundant material, that is, iron-oxide but least explored due to its metastable state in the bulk form. The magnetic properties such as the spin arrangement, anisotropy as well as the ferroelectric properties like polarization and its switching for the bulk e- Fe2O3 will be presented. The stability of thin films and the use of stable thin-film surface for the application of e- Fe2O3 in the photoelectrochemical cell will be presented. The stable surface has been hydrolyzed to see the effect of adsorption. The surfaces of the thin-film give rise to the surface states which acts as trapping sites for the electron-hole pairs leading to the loss of charges and a decrease in the efficiency of the devices. To overcome this problem, we have explored the possibility of modelling the heterostructure of two polymorphs, namely ?-Fe2O3 and e- Fe2O3. We find that the alignment of the band edges at the interface of both the materials are such that it gives rise to a rare type- III broken band-gap heterostructure. This heterostructure helps in the charge separation and can result in good PEC device performance. In the second part of the talk, I will focus on the DFT-based databases and their use in predicting material properties using machine learning techniques. Some of the commonly available databases are AFLOWLIB, Materials Project, OQMD, NOMAD, and many more that collect vast quantities of DFT computations in a convenient manner. Use of one such database to understand the magnetic properties of Fe-based compounds will be discussed. The model interpretability and the important role of designing descriptors for better modeling will be presented.
Machine learning has successfully established itself in a variety of industries, ranging from self-driving automobiles to fraud detection. The materials science community has also started to benefit from it. This has been possible because of the available large number of datasets for materials and their properties, created by high-throughput DFT calculations where an automated input can perform thousands of computations in a single shot. Not only DFT has proven to be a very robust tool for understanding the electronic structure, but the amount of data it has provided for machine learning-based material design is highly valuable.
In the first part of this talk, I will present the use of DFT with an example of e- Fe2O3. It is one of the polymorphs of the most abundant material, that is, iron-oxide but least explored due to its metastable state in the bulk form. The magnetic properties such as the spin arrangement, anisotropy as well as the ferroelectric properties like polarization and its switching for the bulk e- Fe2O3 will be presented. The stability of thin films and the use of stable thin-film surface for the application of e- Fe2O3 in the photoelectrochemical cell will be presented. The stable surface has been hydrolyzed to see the effect of adsorption. The surfaces of the thin-film give rise to the surface states which acts as trapping sites for the electron-hole pairs leading to the loss of charges and a decrease in the efficiency of the devices. To overcome this problem, we have explored the possibility of modelling the heterostructure of two polymorphs, namely ?-Fe2O3 and e- Fe2O3. We find that the alignment of the band edges at the interface of both the materials are such that it gives rise to a rare type- III broken band-gap heterostructure. This heterostructure helps in the charge separation and can result in good PEC device performance. In the second part of the talk, I will focus on the DFT-based databases and their use in predicting material properties using machine learning techniques. Some of the commonly available databases are AFLOWLIB, Materials Project, OQMD, NOMAD, and many more that collect vast quantities of DFT computations in a convenient manner. Use of one such database to understand the magnetic properties of Fe-based compounds will be discussed. The model interpretability and the important role of designing descriptors for better modeling will be presented.
Download this event to my calendar
This event originated in Nebraska Center for Materials and Nanoscience.