This study presents an integrated approach combining Density Functional based Tight Binding (DFTB) calculations with machine learning (ML) techniques to predict the density of states (DOS) in pristine and Zn-doped MgO nanoparticles (NPs). A range of over 60 ML models, including linear models, tree-based ensembles, and neural networks, were evaluated for predictive performance. Among these, the weighted k-nearest neighbor (wkNN) algorithm, particularly when using triweight and biweight kernels, consistently outperformed others, achieving a median RMSE of 0.241 for pristine MgO and 0.386 for Zn-doped samples. The models demonstrated robust performance across various doping concentrations (5–25%) and NP sizes (0.8 nm and 0.9 nm), with minimal impact of doping levels on prediction accuracy. This integration of DFTB with ML offers a powerful and efficient framework for accelerating electronic property predictions in materials science, supporting the rapid design of advanced materials for applications in electronics, catalysis, and energy storage. The code and data are publicly available at: https://github.com/KurbanIntelligenceLab/DOS-Nanoparticles-Weighted-kNN.
1 December 2024 in Machine Learning: Science and Technology
Simulating complex and large materials is a challenging task that requires extensive domain knowledge and computational expertise. This study introduces Pure2DopeNet, an innovative multimodal neural network that tackles these challenges by integrating image and text data to accurately predict the physical properties of doped compounds, specifically Carbon (C)-doped TiO2 and Sulfur (S)-doped ZnO nanoparticles. The model achieves quantum mechanical level accuracy, comparable to density functional tight binding (DFTB), across various doping levels, demonstrating its capability to determine the properties from a single simulation of the pure compound. Pure2DopeNet outperforms traditional deep learning architectures such as ResNet, ViT, and CoAtNet, delivering superior accuracy, faster performance, and reduced dependence on domain expertise. This approach highlights the potential of multimodal machine learning to revolutionize materials science by making high-fidelity simulations more accessible and efficient, opening paving the way for material discovery and the exploration of novel properties.