Database Generation and Machine Learning

Machine-Learning-Assisted Accurate Band Gap Predictions of Functionalized MXene, Chem. Mater. 30, 4031 (2018)
Accelerated Data-Driven Accurate Positioning of the Band Edges of MXenes, J. Phys. Chem. Lett., 10, 780, (2019)
Accelerated Discovery of the Valley-Polarized Quantum Anomalous Hall Effect in MXenes, Chem. Mater., 33, 6311–6317 (2021)

Machine learning is proving to be a valuable tool for predicting the properties of materials and discovering new ones, but the accuracy of the resulting models depends on access to high-quality data. To this end, a computational materials database called “aNANt” has been established which currently contains data on over 24,000 MXenes, including optimized structures and various electronic properties [1,2]. aNANt is first such computational materials database from India.

Two highly accurate machine learning models have been developed using this database. The first model uses easily available properties of MXenes, such as boiling and melting points, atomic radii, phases, and bond lengths, to predict accurate GW bandgaps [2]. The second model enables the simultaneous prediction of the absolute positions of both valence band maxima and conduction band minima, which are crucial for applications in electronic devices such as lasers, photocatalysis, photovoltaics, field effect transistors, and optoelectronics [3]. This model has a root-mean-square error of only 0.12 eV with GW level of accuracy. Additionally, a machine learning-based high-throughput screening approach has been developed to search for valley-polarized quantum anomalous Hall (VP-QAH) insulators in MXenes [4]. Using only eight elemental features, a gradient boost multi-out regressor model has been developed to predict both nodal positions of magnetic nodal line semimetals. This approach can be generalized for the prediction of VP-QAH insulators in other materials, making it computationally efficient and potentially useful for dissipationless valleytronic applications. aNANt is currently hosting topological materials, two-dimensional octahedral materials, thermoelectric materials databases as easily usable application along with MXenes database.

Our group is also extensively working on the developing data-driven machine learning approach for structural, thermoelectric, and catalytic materials.

Reference

  1. 1. aNANt: A functional materials database (https://anant.mrc.iisc.ac.in)
  2. 2. A. C. Rajan, A. Mishra, S. Satsangi, R. Vaish, H. Mizuseki, K. R. Lee and A. K. Singh. Machine Learning-Assisted Accurate Band Gap Predictions of Functionalized MXene. Chem. Mater. 30, 4031 (2018)
  3. 3. A. Mishra, S. Satsangi, A. C. Rajan, H. Mizuseki, K. R. Lee and A. K. Singh, Accelerated Data-driven Accurate Positioning of the Band-edges of MXenes, J. Phys. Chem. Lett., 10, 780, (2019)
  4. 4. R. K. Barik, and A. K. Singh, Accelerated discovery of valley polarized quantum anomalous hall effect in MXenes, Chem. Mater., 33, 6311–6317 (2021)