Dataset and High-Throughput Screening Results for Machine-Learning-Guided Discovery of Relaxor Ferroelectric Ceramics

This repository provides the dataset and screening results used in the study of machine-learning-guided discovery of relaxor ferroelectric ceramics.

The repository contains:

• ONNL relaxor dataset.xlsx    Dataset used for model training and validation. The dataset contains composition, processing conditions, and dielectric properties collected from the literature.

• 250804 element_master.xlsx    Element descriptor table used to compute high-order physicochemical features for A-site and B-site elements.

• HTS_relaxor_pipeline_logProduct_A3_B2_step001.csv    High-throughput screening (HTS) results generated by the ensemble neural network model.    The screening explores the compositional design space of A3–B2 mixed perovskites with step size 0.01 and applies charge neutrality and tolerance-factor constraints.

• Supplementary_Table_8_NBT_ST_family_benchmark.xlsx    Benchmark dataset used for Supplementary Table 8. This file compiles 74 reported SNBT/NBT–ST-family compositions and compares their room-temperature permittivity and upper-temperature-side dielectric stability. The dataset includes the dielectric-permittivity changes at 85, 105, and 125 °C, corresponding to the X5R, X6R, and X7R upper-temperature-side criteria, respectively. It was used to identify compositions satisfying |Δε/ε25| ≤ 15% across the simultaneous X5R/X6R/X7R upper-side stability window and to benchmark the present SNBTS compositions against related literature systems.