Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp01bk128f048
DC FieldValueLanguage
dc.contributor.authorYue, Shuwen
dc.contributor.otherChemical and Biological Engineering Department
dc.date.accessioned2021-10-04T13:49:27Z-
dc.date.available2021-10-04T13:49:27Z-
dc.date.created2021-01-01
dc.date.issued2021
dc.identifier.urihttp://arks.princeton.edu/ark:/88435/dsp01bk128f048-
dc.description.abstractMolecular simulation predictions of thermodynamic and transport properties of fluids such as water, electrolyte solutions, and CO2 are of considerable interest to energy, environmental, and industrial applications. The reliability and accuracy of these predictions are contingent on the molecular models used in simulation. Here, we investigate the predictive capabilities of several classes of molecular models, from simple empirical force fields to high dimensional machine learning (ML) models, in order to provide insight on the necessary physics for representing complex fluids. We first evaluate empirically derived polarizable, non-polarizable, and scaled charge models in representing the dynamic properties of aqueous electrolyte solutions. While polarizability improves structural and dynamic predictions, there re- main insufficient physics for achieving quantitative accuracy. The advent of ML frameworks applied to molecular models has made way for far more descriptive representations of water and electrolyte solutions, combining ab initio levels of accuracy with classical level computational costs. However, the lack of explicit long-range interactions in ML models remains a fundamental caveat. We investigate the consequences of this localized representation for various thermodynamic regimes of water and electrolyte solutions. We then construct ML models based on the SCAN DFT functional for several species of alkali halide electrolyte solutions which give thermo- dynamic properties with excellent agreement with experiments and dynamic properties which significantly improve upon that of conventional empirical force fields. Finally, we constructed many-body polarizable models of CO2 and assessed the influence of functional form flexibility and training set quality on bulk thermodynamic properties. The results in this thesis illustrate the limitations and scope by which several classes of molecular models, from empirical force fields to ML models, can be utilized reliably. Additionally, new ML models of electrolyte solutions and CO2 constructed in this work provide promising avenues toward studying complex fluid behavior from first principles perspectives.
dc.format.mimetypeapplication/pdf
dc.language.isoen
dc.publisherPrinceton, NJ : Princeton University
dc.relation.isformatofThe Mudd Manuscript Library retains one bound copy of each dissertation. Search for these copies in the library's main catalog: <a href=http://catalog.princeton.edu>catalog.princeton.edu</a>
dc.subjectElectrolyte solutions
dc.subjectMolecular simulations
dc.subjectWater
dc.subject.classificationMolecular physics
dc.titleThermodynamic and transport properties of molecular fluids: From empirical force fields to machine-learning models