Machine learning based prediction of optical properties in two-dimensional Mo-W-S-Se-Te transition-metal dichalcogenide alloys through physics-informed sampling
This work combined ab initio optical-property calculations with a tabular foundation-model regression to predict the real and imaginary components of the frequency-dependent dielectric function for Mo-W-S-Se-Te TMD alloys, and predicted derived optical quantities, including refractive index, extinction coefficient, and absorption coefficient.
Abstract
Two-dimensional transition-metal dichalcogenide (TMD) alloys provide a compositionally tunable platform for controlling the optical and electronic properties. However, systematic prediction of their dielectric response across multicomponent alloy spaces remains challenging owing to the combinatorial cost of first-principles calculations. In this work, we combined ab initio optical-property calculations with a tabular foundation-model regression to predict the real and imaginary components of the frequency-dependent dielectric function for Mo-W-S-Se-Te TMD alloys. A dataset of 99 alloy structures spanning binary, ternary, quaternary, and quinary compositions was generated using density functional theory (DFT). The resulting polarization-dependent dielectric spectra were used to train a tabular prior-fitted network (TabPFN) and evaluated against the conventional Extra Trees and XGBoost models. To accommodate the in-context capacity limit of TabPFN, we introduced a non-uniform, physics-informed energy subsampling strategy that concentrates sampling in the optically active region above the band gap, where interband absorption is strongest. Trained solely on quaternary alloys, our TabPFN reconstructed the dielectric spectra of held-out quaternary compositions with an R2>0.98 and a mean absolute error below 0.10 for all four dielectric components, outperforming both baselines while requiring no gradient-based training or hyperparameter tuning. Our model further predicted derived optical quantities, including refractive index, extinction coefficient, and absorption coefficient. Additionally, our model generalized in a zero-shot manner to binary, ternary, and quinary alloys absent from the training set, with quinary predictions achieving an R2>0.97.
Machine learning has become an effective tool for accelerating materials discovery by predicting material properties with far lower computational cost than exhaustive first-principles calculations. However, its application to dopant engineering in two-dimensional (2D) oxide semiconductors remains relatively unexplored. In this work, we combine supervised machine learning with density functional theory-derived data from the Materials Project database to predict the electronic, thermodynamic, transport, and optical properties of undoped and doped 2D tin monoxide (SnO). Several supervised regression algorithms are evaluated for predicting the band gap, formation energy, carrier mobility, electrical conductivity, and optical absorption coefficient. Among the models considered, gradient boosting regression provides the highest predictive accuracy and consistently reproduces the complex structure–property relationships of both undoped and doped systems. Feature-importance analysis reveals that electronic band descriptors, particularly band width, followed by band asymmetry and electronic anisotropy, are the primary factors governing band-gap prediction, while structural descriptors provide complementary contributions. The predictions indicate that monolayer SnO has a wider band gap, slightly higher formation energy, and higher carrier mobility than bulk SnO due to quantum confinement. Mg and Zn substitution allow systematic tuning of the material properties over a moderate concentration range. The band gap changes only slightly with doping, while Mg incorporation is thermodynamically more favorable than Zn, whose stability decreases with increasing concentration. Dopant-induced impurity scattering reduces carrier mobility and electrical conductivity, whereas optical absorption is enhanced in both Mg- and Zn-doped systems. These results demonstrate that machine learning can efficiently identify promising dopant configurations while providing reliable predictions of dopant-dependent material properties. The proposed framework offers a practical approach for the computational design of doped 2D oxide semiconductors for future optoelectronic and sensing applications.
This study presents a comparative machine-learning investigation for predicting the optical absorption coefficient of truncated pyramidal GaN/AlxGa1−xN quantum dots. The physical dataset is generated by solving the three-dimensional Schrödinger equation using a coordinate-transformation method combined with the finite-difference method (FDM). The coordinate transformation maps the sloping boundaries of the truncated pyramidal geometry onto a regular computational domain, enabling an accurate representation of the quantum-dot shape and facilitating its numerical treatment using the FDM. The absorption coefficient is then calculated as a function of photon energy for different alloy compositions. Using photon energy and alloy composition as input features, Artificial Neural Network (ANN), Random Forest (RFR), Decision Tree (DT), and k-Nearest Neighbor (KNN) models are developed and evaluated. A second-degree polynomial regression model is also considered as a classical baseline. Under the point-wise random 80/20 split, all models show excellent agreement with the numerical results, with R2 values close to unity. KNN generally provides the lowest prediction errors across most alloy compositions, whereas ANN achieves slightly lower MSE and RMSE values at x=0.5. Furthermore, leave-one-composition-out validation identifies ANN as the most effective model for predicting unseen compositions, achieving a mean R2 of 0.848 and an NRMSE of 7.19%. These findings demonstrate that KNN is particularly effective for local interpolation within the sampled domain, while ANN provides stronger composition-wise generalization. The proposed framework offers an efficient surrogate for computationally demanding numerical simulations of the optical properties of quantum nanostructures.
T. Brahim, A. Bouazra, B. Basha et al.· Mathematics· 0 citations
Hafnium oxide (HfO2) is the cornerstone high-k dielectric in modern silicon technology. Since the constraints of silicon device fabrication rule out replacing the material itself, dopant incorporation is the principal means available to engineer its band gap and dielectric constant within existing process flows. However, dopant selection is still largely empirical due to the coupled interplay among thermodynamic stability, electronic insulation, and dielectric response. Here, we present a high-throughput computational framework integrating special quasi-random structures (SQS), machine-learning potentials (SevenNet), and graph neural networks (ALIGNN) to systematically evaluate doped-HfO2 compositions across three dopants (Al, Si, Y) and two technologically relevant polymorphs (monoclinic and orthorhombic). Our analysis uncovers a fundamental design principle: formation energy, band gap, and dielectric constant are decoupled parameters requiring application-specific prioritization rather than simultaneous optimization. Yttrium achieves the lowest formation energy (-3.763 eV/atom) and favors orthorhombic phase stabilization at process-compatible thermal budgets; silicon preserves near-pristine band gaps (around 5.72 eV) critical for suppressing leakage in gate dielectric applications; and aluminum enables concentration-tunable band gap widening (5.6-5.9 eV) suited for voltage scaling. Validation against experimental literature and density functional theory (DFT) confirms quantitative accuracy (0.02 eV band gap error for Si-doping, mean absolute error less than 0.001 eV/atom formation energy). This framework provides rational, property-decoupled guidance for dopant engineering in HfO2-based dielectrics and related high-k oxide systems.
Zunair Masroor, B. Gu, Wonjoong Kim et al.· 0 citations
An explainable machine-learning framework was developed for dielectric constant prediction using 52,168 crystalline materials extracted from the Joint Automated Repository for Various Integrated Simulations (JARVIS-DFT) database, demonstrating the complementary roles of electronic structure and elemental chemistry.
D. Pundhir, Ashok Kumar· Applied Physics A· 0 citations
A semiempirical extended tight-binding approach (GFN1-xTB) is employed to compute the electronic properties of a dataset of MOFs, and it is shown that GFN1-xTB approximates MOF band gaps well, as compared to semilocal DFT.
A. Jose, A. Walsh· Journal of Chemical Theory a...· 0 citations
A unified multiscale system that entails the implementation of density functional theory (DFT), machine learning (ML), and device-level simulation to hasten the search and development of high-performance perovskite solar cell materials is presented.
Sameer Pandey, N. Shukla, Vishal K. Sharma et al.· Applied Nanoscience· 0 citations