Thermodynamic Accuracy and Transferability of Coarse Grained Models and Applications to Petrochemical Systems

Restricted (Penn State Only)
Dunn, Nicholas Jeremiah
Graduate Program:
Doctor of Philosophy
Document Type:
Date of Defense:
September 12, 2018
Committee Members:
  • William George Noid, Dissertation Advisor
  • William George Noid, Committee Chair
  • Vincent Henry Crespi, Committee Member
  • Michael Anthony Hickner, Committee Member
  • Kristen Ann Fichthorn, Outside Member
  • coarse grained modeling
  • asphaltenes
  • bocs
  • molecular dynamics
Asphaltenes are a problematic fraction of crude oil known for their propensity for aggregation during oil extraction and processing. This aggregation is an expensive problem for the petroleum industry, as it is difficult to predict, prevent, or reverse. Asphaltenes have been difficult to study via experimental methods due to the complexity of the asphaltene fraction as well as its propensity for aggregating at very low concentrations. As a result, computational techniques such as molecular dynamics (MD) simulations are an appealing approach to studying asphaltenes. However, simulating relatively large-scale, slowly evolving processes such as mesoscale aggregation at an all-atom (AA) level of resolution is infeasible using even the latest high-performance computing hardware. Coarse-grained (CG) modeling has emerged as a method of reducing the computational complexity of AA MD simulations by coarsening out degrees of freedom and grouping atoms together into CG sites. This reduction in resolution simplifies simulations using the resulting models, extending the length- and time-scales accessible by MD simulation. In order to be useful for studying real systems, CG models must incorporate the correct physics to accurately describe the system they represent. There are two main methods for incorporating these physics into CG models via parame- terization: top-down and bottom-up modeling. Top-down CG models use simple functional forms for their interaction potentials and are parameterized to reproduce experimentally observable properties of the target system. The resulting models accurately reproduce the targeted properties and are transferable to other state points, but may not accurately represent the fine structural details of the system. Bottom-up CG models are parameterized using information from simulations of an underlying AA model and may use more complex functional forms. The correct potential for a bottom-up CG model is the potential of mean force (PMF). The PMF contains all of the information necessary for the CG model to reproduce all properties of the AA model at the CG level of representation. However, the PMF is too complex to determine or use in simulation, so bottom-up models almost always use a potential that is an approximation to the configuration-dependent portion of the PMF. As a result of neglecting the state-point dependence of the PMF, these approximations generally do not provide accurate descriptions of the thermodynamic properties of the underlying AA model and are not transferable away from the state point of their parameterization. This work implements methods for improving the thermodynamic accuracy and transferability of bottom-up CG models and studies petrochemical systems as examples for demonstrating these methods. Chapters 2, 3, and 4 of this work examine the volume-dependence of the PMF using bottom-up CG models of the the petrochemical solvents heptane and toluene as example systems. We implement the volume-dependent pressure correction devised by Das and Andersen for use with bottom-up CG models, and demonstrate that this method obtains qualitative but not quantitative agreement with the PV equation of state of the underlying AA model. We extend this pressure-matching method with a self-consistent iterative procedure that generates CG models that quantitatively reproduce the PV equation of state of the underlying AA model. We further demonstrate this method for use in parameterizing a transferable CG model that is accurate across a range of system compositions. Chapter 5 presents the open-source release of the BOCS software package used to parameterize the bottom-up models in Chapters 2-4. Finally, Chapter 6 presents a top-down toy model for asphaltenes to study nanoaggregate formation over a range of solvent conditions and molecular structures.