Photostability is a key property of agrochemicals since insufficient photostability will lead to damaged efficacy in application. Industrial assessment of the photostability of agrochemicals often involves large scale experimental tests. Therefore, robust theoretical methods for modelling photostability are necessary to accelerate the discovery of agrochemicals. In this thesis, computational models for the assessment of the photostability of agrochemicals against both direct and indirect photodegradation were developed.
The stabilities of Strobilurin molecules against direct photoisomerization and photodegradation were examined using their molecular structures and excited state properties. The excitation state energies of the Strobilurins were fed to a support vector machine (SVM) to ‘predict’ their stability. An unsupervised clustering model and a Naïve Bayesian classifier against photoisomerization were built to study the structural–stability relationships. The prediction accuracy of the SVM was the highest, while the unsupervised clustering model and Bayesian classifier were computationally less expensive.
The indirect photodegradation of Cycloxydim in the presence of Chlorothalonil was studied. The reaction process involved intermolecular hydrogen transfer and the reactant dimer was formed via excitation energy transfer (EET) from triplet Chlorothalonil to ground state Cycloxydim. Electron transfer followed by proton transfer was suggested as the reaction mechanism. Oxidation rates of Phenylurea herbicides by dissolved organic matter (DOM) were modelled using electron affinity-based and bond strength-based descriptors. The oxidation reaction rates had a negative relationship with oxidation potential and vertical ionization potential of the herbicides. Solvent effects of water and acetonitrile on the photochemical fate of Cycloxydim were examined with different models: continuum solvent alone, and continuum solvation in combination with solvent molecules and QM/MM/MD. Explicit solvent representations were helpful for examining the solvent–solute interactions.
In summary, computationally inexpensive descriptors can be used for screening candidate molecules with similar photochemical fate, while calculation of potential energy profiles and intermolecular interactions would be required for mechanism examination.