Machine-Learning Classification to Predict the Dimensionality of Hybrid Organic-Inorganic Halide Perovskites

Aug 5, 2026

Low dimensional hybrid perovskites deliver strong performance in photovoltaic applications through exceptional optical and electronic properties. Structure understanding remains essential to evaluate stability while searching for new candidates. This thesis develops a machine-learning framework to predict perovskite dimensionality by learning polyhedral network connectivity across directions. Trained on 588 reported structures spanning 0D–2D, the model reaches high class precisions and predicts dimensionality for 40,556 new HOIPs, followed by similarity-driven selection of promising organic cations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages50

Reading Time2 h 6 min

AccessFree to download

Chapter 1 Introduction

1.1 Motivations for materials science

1.2 Motivations for machine learning

Chapter 2 Methods

2.1 Dataset generation

2.2 Feature generation

University of South Carolina Scholar Commons Theses and Dissertations Summer 2023 Machine-Learning Classification to Predict the Dimensionality of Hybrid Organic-Inorganic Halide Perovskites Yatiwelle Koralalage Samuditha Sandaru Yatiwella Follow this and additional works at: [https://scholarcommons.sc.edu/etd](https://scholarcommons.sc.edu/etd) Part of the Chemistry Commons Recommended Citation Yatiwella, Y. (2023) . Machine-Learning Classification to Predict the Dimensionality of Hybrid OrganicInorganic Halide Perovskites. (Master's thesis) . Retrieved from [https://scholarcommons.sc.edu/etd/7492](https://scholarcommons.sc.edu/etd/7492) [This Open Access Thesis is brought to you by Scholar Commons. It has been accepted for inclusion in Theses and](This Open Access Thesis is brought to you by Scholar Commons. It has been accepted for inclusion in Theses and)[ ](This Open Access Thesis is brought to you by Scholar Commons. It has been accepted for inclusion in Theses and)[Dissertations by an authorized administrator of Scholar Commons. For more information](Dissertations by an authorized administrator of Scholar Commons. For more information), please contact [digres@mailbox.sc.edu](digres@mailbox.sc.edu). Machine-Learning classification to predict the dimensionality of hybrid organic-inorganic halide perovskites by Yatiwelle Koralalage Samuditha Sandaru Yatiwella Bachelor of Science University of Colombo 2017 Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Science in Chemistry College of Arts and Sciences University of South Carolina 2023 Accepted by: Christopher Sutton, Director of Thesis Andrew Greytak, Reader Ann Vail, Dean of the Graduate School Acknowledgments Words cannot express my gratitude to Professor Christopher Sutton, who gave me valuable instructions on my journey; for his support. I also could not have undertaken this journey without the support of Professor Andrew Greytak and Professor Mark Berg, who generously provided knowledge and expertise. Additionally, this endeavor would not have been possible without the generous support from my group colleagues Sourin, Dipannoy, Jack, Dr. Adihkari, and Dr. Karimitari. I also thank our outside collaborator Professor Steve for his feedback on our project. I am also grateful to my teachers, the University of South Carolina; Prof. Sophya Garashchuk, Prof. Michael L. Myrick, Prof. Dmitry V. Peryshkov, and the teachers of my undergraduate university; the University of Colombo, SriLanka and all the other teachers who impacted and inspired me. Most importantly I would like to thank my parents and parents-in-law and my siblings for their incredible support and encouragement and for all they have done to make my life happier. My heartfelt gratitude goes to my dear wife, who supported me in numerous ways to accomplish this target. Lastly, I would be remiss in not mentioning my university colleagues from the Department of Chemistry and all my friends for their invaluable help and friendship. Abstract Low dimensional hybrid perovskites have demonstrated remarkable performance in photovoltaic applications, primarily due to their exceptional optical and electronic properties. As the search for potential candidates for novel materials continues, understanding the structure of these materials is crucial for investigating their stability. In this study, we implement a framework to find novel material based on a machinelearning model. The machine learning model was trained to predict the dimensionality of the polyhedral network based on the connectivity of the polyhedral network indifferent directions. The polyhedral connectivity of low-dimensional structures can be classified into three dimensions: 0D, 1D, and 2D. This ML model was trained on 588 unique experimentally reported structures and these structures include halide perovskites as well as related, non-perovskite halometallate structures. These str

FAQs

What dimensionality classes does the machine-learning model aim to predict for hybrid perovskites?

It classifies polyhedral connectivity into three dimensionalities: 0D, 1D, and 2D based on connectivity across directions.

How was the model trained and what input representations were used?

The model was trained on 588 experimentally reported structures and used a graph representation from the GROVER self-supervised message passing transformer for organic cations, together with seven features for inorganic/halide elements.

How does the model help discover missing or new HOIP compounds and organic cations?

After learning from known compounds, it predicts dimensionality for 40,556 new HOIPs and performs a similarity search using GROVER to identify potentially interesting organic cations not yet examined for HOIPs.

Aug 5, 2026

Upload by Olivia Brown