Development of an efficient charge equilibration method for nonlocal neural network potentials with applications in redox chemistry

by Emir Kocer
Doctoral thesis
Date of Examination:2024-12-13
Date of issue:2025-01-06
Advisor:Prof. Dr. Jörg Behler
Referee:Prof. Dr. Jörg Behler
Referee:Prof. Dr. Ricardo A. Mata
Persistent Address: http://dx.doi.org/10.53846/goediss-10972

 

 

Files in this item

Name:Emir_Kocer_PhD_Thesis_revised.pdf
Size:12.1Mb
Format:PDF
Description:PhD Thesis

The following license files are associated with this item:

Abstract

English

This thesis focuses on the development and application of new methodologies for machine learning potentials in atomistic simulations. The primary contribution of this work is the design and implementation of an optimized fourth-generation high-dimensional neural network potential interface within an open-source molecular dynamics software. The method introduces an iterative global charge equilibration scheme, significantly improving computational efficiency and scalability compared to traditional matrix-based solutions. This allows for accurate modeling of long-range electrostatic interactions and charge transfer in large-scale systems. The developed methodology was applied to simulate aqueous iron chloride solutions, demonstrating the first successful modeling of redox reactions using a machine learning potential. The study highlights the capability of the fourth-generation high-dimensional neural network potential to accurately capture electron transfer and oxidation state changes in aqueous environments, offering new opportunities for simulations of redox chemistry in solution. Alongside these advancements, the thesis also reviews key developments in neural network potentials, explores the performance of different structural descriptors, and compares their sensitivity to atomic displacements. Collectively, this work aims to provide a step forward in both the theoretical and practical aspects of machine learning potentials, enabling their use in a wider range of chemical and physical systems.
Keywords: machine learning; physical chemistry; neural networks; nonlocal systems; charge equilibration
 
Statistik