Machine Learning Assisted Parameter Tuning on Wavelet Transform Amorphous Radial Distribution Function

TL;DR

This work tackles the challenge of accurately characterizing amorphous structures by enhancing the physics-based WT-RDF framework with machine-learning–guided parameter optimization, yielding WT-RDF+. By treating key parameters as learnable, applying parameter bounding, and employing a selective loss focused on RDF peaks, WT-RDF+ achieves superior predictions of peak positions and amplitudes across Ge–Se binary and Ag–Ge–Se ternary amorphous systems. The approach demonstrates robustness under limited data and outperforms Radial Basis Function and Long Short-Term Memory baselines, particularly at low data ratios, while maintaining parameter efficiency with only five tunable variables. Cross-dataset evaluation confirms improved generalization to unseen compositions, highlighting WT-RDF+ as a reliable tool for structural characterization and the design of phase-change thin films for electronic devices. The work emphasizes the value of integrating physics-based models with targeted ML optimization to achieve accurate, data-efficient predictions in complex amorphous materials.

Abstract

Understanding atomic structures is crucial, yet amorphous materials remain challenging due to their irregular and non-periodic nature. The wavelet-transform radial distribution function (WT-RDF) offers a physics-based framework for analyzing amorphous structures, reliably predicting the first and second RDF peaks and overall curve trends in both binary Ge 0.25 Se 0.75 and ternary Ag x(Ge 0.25 Se 0.75)100-x (x=5,10,15,20,25) systems. Despite these strengths, WT-RDF shows limitations in amplitude accuracy, which affects quantitative analyses such as coordination numbers. This study addresses the issue by optimizing WT-RDF parameters using a machine learning approach, producing the enhanced WT-RDF+ framework. WT-RDF+ improves the precision of peak predictions and outperforms benchmark ML models, including RBF and LSTM, even when trained on only 25 percent of the binary dataset. These results demonstrate that WT-RDF+ is a robust and reliable model for structural characterization of amorphous materials, particularly Ge-Se systems, and support the efficient design and development of phase-change thin films for next-generation electronic devices and components.

Figures (9)

• Figure 1: Illustration of the atomic structure of a material with (a) crystalline structure and (b) amorphous structure
• Figure 2: Illustration of the mean field method that is adapted in the WT-RDF formulation
• Figure 3: Comparison between (a) experimental RDF, and (b) WT-RDF for Ge$_{0.25}$Se$_{0.75}$ (binary) and Ag$_x$(Ge$_{0.25}$Se$_{0.75}$)$_{100-x}$ ($x$ = 5, 10, 15, 20, 25) (ternary A to E). The parameters are $a$ = 0.590; $b$ = 4.600; $K_f$ = 0.106; $\Lambda$ = $-$4.700; $\tilde{C}$ = 0.212.
• Figure 4: The effects of WT-RDF training components on binary data. (a) Parameter optimization without parameter bounding and selective loss; (b) Parameter optimization with parameter bounding; (c) Parameter optimization with selective loss; (d) The result of our proposed WT-RDF${+}$.
• Figure 5: Comparison between WT-RDF (baseline) and our proposed WT-RDF${+}$ under different training data ratio on binary data. (a) 100$\%$; (b) 75$\%$; (c) 50$\%$; (d) 25$\%$.