Study of Acid Suppressed Thickener Technology Using Density Functional Theory and Machine Learning Techniques

Acid Suppressed Thickener Technology;pKa;Solvation Free Energy;Density Functional Theory;Machine Learning;Materials Science and Engineering;Engineering;Materials Science and Engineering

Wu, WenkunTuteja, Anish ;

University of Michigan

关键词: Acid Suppressed Thickener Technology;    pKa;    Solvation Free Energy;    Density Functional Theory;    Machine Learning;    Materials Science and Engineering;    Engineering;    Materials Science and Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/145967/wuwenkun_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Hydrophobically modified ethylene oxide urethane (HEUR) rheology modifiers, which are water-based polyurethane formulations manufactured by Dow Coating Materials, a division of the Dow Chemical Company, are often added to interior and exterior water-based Latex paint formulations to control their viscosity.The thickening efficiency of the HEUR rheol-ogy modifier is controlled by the pH of the solvent, as this affects the protonation-deprotonation equilibrium of the amine hydrophobe group at the end of the rheology modi-fier polymer chain.The principal quantity characterizing this equilibrium is the acid disso-ciation constant (pKa) of the hydrophobe group, which identifies the transition between high and low viscosity of the suspension.To gain a better understanding of the functioning of the hydrophobe molecular groups, and to develop novel hydrophobes that meet specific per-formance characteristics, it is important to accurately predict the pKa based on first princi-ples calculations, and use it as a first evaluation criterion for a rapid screening of candidate hydrophobe molecules. A main source of error in the pKa calculation is the value of solvation free energy of the molecule in its charged state.We therefore develop new methods to increase the accuracy of the solvation free energy calculation for charged species without excessive increase the computational expense.This includes a hybrid cluster-continuum model approach, where explicit solvent molecules are added to the traditionally employed continuum solvation model, and a molecular dynamics (MD sampling procedure that eliminates the costly ener-gy minimization step.Using test molecules for pKa calculations, we systematically exam-ine the convergence behavior in terms of number of explicit water molecules that need to be included in the cluster-continuum model, the influence of the dielectric constant attributed to the continuum, and the placement of a counter ion for charge neutrality for the accurate calculation of the solvation free energy.We establish that the MD sampling method yields results comparable the energy minimization procedure during density functional theory (DFT) calculations, but at 100 times the speed.When calculating the solvation free energy and the pKa calculation of a known hydrophobe, ethoxylated bis(2-ethylhexy)amine, we find that including explicit water molecules and a fragment of the latex polymer in its local en-vironment both significantly improve the results. Finally we develop an informatics-based approach that employs a transferable machine learning (ML) model, trained and validated on a limited amount of experimental data, to predict the solvation free energies of new ionic species at a reasonable computational cost.We compare three different ML methods – linear ridge regression, support vector regression and random forest regression, and find that the model trained by the random forest regres-sion method yields the predictions with the lowest mean absolute error.A feature selection analysis shows that the atomic fraction feature, which reflects the chemical constitution of the hydrophobe, plays the most important role in the solvation free energy prediction.Add-ing the Wiener index, a measure of the molecular topology, and the solvent accessible sur-face area of the molecules further improve the performance of the model.Accordingly, our ML model predicts the solvation energies of ionic species, including our test hydrophobe molecule, with similar accuracy as atomistic modeling using first-principles calculations.