Democritus University of Thrace (DUTH)
Μελέτη φυσικοχημικών ιδιοτήτων χημικών πολεμικών ουσιών με τη συνδυαστική χρήση μοριακών προσομοιώσεων και μηχανικής μάθησης
Abstract
dc:descriptionThis dissertation examines the most prevalent and hazardous chemical warfare agents that have posed a significant concern to humanity over the past decades. Specifically, it investigates the three major categories of nerve agents: G-series agents, V-series agents, and A-series agents. These substances belong to the class of highly toxic compounds, and the molecular simulations employed in this study enable their safe and risk-free investigation. The primary objective of this study is to analyze and compute a comprehensive set of thermodynamic properties, transport properties, intermolecular structural properties at the microscopic level, and the spectroscopic behavior of these liquid molecular systems in the infrared region of the radiation spectrum. The infrared spectra were computed using a partial Fourier transform of the time autocorrelation function of the total dipole moment for the studied systems. For each substance examined in this dissertation, two different empirical molecular potential models were applied. These models differ in their methodology for computing the localized charges at the interaction sites. Two approaches were employed: one based on quantum mechanical calculations and another utilizing machine learning techniques. A machine learning technique was also used to determine the initially most stable molecular structure of the compounds. Subsequently, the molecules were modeled using fully flexible molecular potentials, allowing for motion along intramolecular bonds, bond angles, and dihedral angles. For G-series agents, three substances were studied: Sarin, Soman, and Tabun. For V-series agents, three substances were examined: VX, RVX (Russian VX), and CVX (Chinese VX).For A-series agents, six substances were analyzed, as the three principal representatives of this category have two distinct proposed molecular structures available in the international literature. Specifically, for each of the compounds A230, A232, and A234, two different structural models were investigated. Results were compared with the limited experimental data available, and conclusions were drawn regarding the reliability of the molecular models applied. Additionally, a comparative analysis was conducted between the two interaction potential models used for each substance—the quantum mechanical model and the machine learning-based model—to assess their respective performances. A machine learning-based linear regression algorithm was employed to determine the self-diffusion coefficient of each substance in the molecular liquid and to compute the coefficients of the semi-empirical Antoine equation.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Koufou, Antonios
- Κουφού, Αντώνιος
Subjects
dc:subject × 11Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58962