• Login
    View Item 
    •   UBIR Home
    • Theses and Dissertations (ETDs)
    • 2018-09-01 UB Theses and Dissertations (public)
    • View Item
    •   UBIR Home
    • Theses and Dissertations (ETDs)
    • 2018-09-01 UB Theses and Dissertations (public)
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    PREDICTING MELTING POINTS OF DEEP EUTECTIC SOLVENTS

    Thumbnail
    View/Open
    sonpal_buffalo_0656m_16123.pdf (3.419Mb)
    Date
    2018
    Author
    Sonpal, Aditya
    Metadata
    Show full item record
    Abstract
    Up until the end of the 20th century, ionic liquids and later, room temperature ionic liquids had been used as solvents for the synthesis of a number of organic and inorganic materials. Owing to their wide liquid range, immiscibility, non volatility, easy synthesis, etc., they were largely favored by scientists everywhere. Until then they were also believed to be green solvents. Their green status was later contested, which paved the way for the discovery of deep eutectic solvents (DES). These solvents were composed of two or more cheap and safe materials which associated with each other through hydrogen bond interactions to create a eutectic mixture. In other words, DES systems had a lower melting point than that of their individual components. In addition to all the advantages of ionic liquids, DES are also green solvents and can be easily and cheaply synthesized. We therefore identified the potential of DES systems and the importance of melting point in characterizing such systems. We use a data driven approach to predict melting points of DES. We first shortlist a set of 36 DES systems, each having Choline Chloride as the hydrogen bond donor. We optimize the geometries of these compounds using DFT frameworks and use that structural information to generate descriptors from the Dragon 7 software to generate our feature space. We use pearson correlation coefficients and genetic algorithm for feature selection and use the resulting feature space to train generalized linear regression algorithms from the 'scikit-learn' python library. In this thesis, we first show that the PBE0 functional with def2-TZVP basis set generates the most accurate geometrical configurations for feature representation. We further prove the supremacy of genetic algorithms over pearson correlation coefficients for feature selection. In conclusion, we show that the linear ridge regression model yields the most accurate predictions for our dataset with PBE0/def2-TZVP. We then highlight some of the challenges associated with predicting melting points and provide suggestions that could possibly help to mitigate them.
    URI
    http://hdl.handle.net/10477/78667
    Collections
    • 2018-09-01 UB Theses and Dissertations (public)

    To add content to the repository or for technical support: Contact Us
     

     

    Browse

    All of UBIRCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsTypesThis CollectionBy Issue DateAuthorsTitlesSubjectsTypes

    My Account

    LoginRegister

    To add content to the repository or for technical support: Contact Us