Large Dataset on the Relationship Between Structure and Stability in Glass Wins Open Science Award for Yu

The generation and sharing of a large dataset created as part of his study has won Zheng Yu the 2021 Wisconsin MRSEC Excellence in Open Science Prize. A graduate student in Dr. Bu Wang’s lab at the Grainger Institute for Engineering, Yu generated the data as part of his work investigating the relationship between structure and stability in specialized glasses using computer simulations and machine learning.

“Generating structures is a crucial but laborious step for molecular dynamics simulations of glass. By sharing our data, we hope to facilitate future large-scale studies of structure-property-stability relationships in glass,” said Wang.

The MRSEC Open Science prize recognizes a researcher or team that has demonstrated an exceptional effort or success in the development and dissemination of data for benefit of the scientific community. With the transformative developments in data access and analytics, this is an increasingly important part of modern materials science. This prize seeks to encourage researchers to be innovative in sharing their data.

Thermodynamic stability, describing where a glass is located on the potential energy landscape, is very important to determining its material properties. However, even for some of the most common glass systems, like silica, the structure-stability relationship is not well understood.

The goal of Yu’s work was to investigate the relationship between structure and thermodynamic stability in vitreous silica (quartz glass, the basis of regular glass) using molecular dynamics simulations and machine learning. To obtain a large dataset for better characterizing the relationship, the researchers applied special molecular dynamics simulations to sample the potential energy landscape of silica. With a wide range of stabilities, 24,000 glass structures were generated. The energetic information as well as 70 structural features, such as density and bond angles of these structures, were calculated and summarized in a dataset for machine learning studies. The structures as well as the dataset are available for download from the Materials Data Facility (MDF)

In this study, the researchers found that the thermodynamic stability can be precisely predicted by both linear and nonlinear machine learning models from the structural features. Based on the machine learning models, a set of five structural features that are the most predictive of the silica glass stability was identified.

The most predictive structural features of stability in silica as determined by the data generated in Yu’s study.

Yu hopes the shared data will benefit other researchers. “We have shared the dataset that can be used to reproduce the machine learning results from this work. Furthermore, we have shared a large database of well-characterized silica glass structures with a wide range of stabilities,” he said. The shared data are currently being used by other researchers in the Wisconsin MRSEC.

Yu and the research team have published a paper based on this work titled, “Structural signatures for thermodynamic stability in vitreous silica: Insight from machine learning and molecular dynamics simulations,” in Physical Review Materials this year (

The Wisconsin MRSEC congratulates Yu on his outstanding research and his commitment to maximizing the value and impact of this work through the practices of open science.