Bottom-up coarse-grained models with predictive accuracy and transferability for both structural and thermodynamic properties of heptane-toluene mixtures

Abstract

This work investigates the promise of a "bottom-up" extended ensemble framework for developing coarse-grained (CG) models that provide predictive accuracy and transferability for describing both structural and thermodynamic properties. We employ a force-matching variational principle to determine system-independent, i.e., transferable, interaction potentials that optimally model the interactions in five distinct heptane-toluene mixtures. Similarly, we employ a self-consistent pressure-matching approach to determine a system-specific pressure correction for each mixture. The resulting CG potentials accurately reproduce the site-site rdfs, the volume fluctuations, and the pressure equations of state that are determined by all-atom (AA) models for the five mixtures. Furthermore, we demonstrate that these CG potentials provide similar accuracy for additional heptane-toluene mixtures that were not included their parameterization. Surprisingly, the extended ensemble approach improves not only the transferability but also the accuracy of the calculated potentials. Additionally, we observe that the required pressure corrections strongly correlate with the intermolecular cohesion of the system-specific CG potentials. Moreover, this cohesion correlates with the relative "structure" within the corresponding mapped AA ensemble. Finally, the appendix demonstrates that the self-consistent pressure-matching approach corresponds to minimizing an appropriate relative entropy.

Simultaneous Description of Equilibrium, Interfacial, and Transport Properties of Fluids Using a Mie Chain Coarse-Grained Force Field

“…Among them, CG model based on a rigorous bottom-up scheme (i.e. CG models developed on the basis of a FG model) are be very promising for general purpose [11][12][30][31][32][33][34]. For thermophysical fluid properties predictions, CG models based on a top-down parameterization strategy (i.e.…”

Assessing the transferability of common top-down and bottom-up coarse-grained molecular models for molecular mixtures

“…The extended ensemble approach of Mullinax and Noid, 97 in which the coarse-grained potential is parametrised for an ensemble of systems simultaneously, allows for concentration transferable models, although at the cost of some representability compared to models parametrised for a single system. This approach was extended by Dunn and Noid 98 to include volume potentials, yielding a set of related models which, together, can give both temperature and concentration transferability. The linear pressure correction which has been applied in this work is very effective at correcting the coarse-grained pressure for a given state-point, without requiring the inclusion of any other parameters in the coarse-grained force field.…”

“…63 This was later extended to a volume matching method, which can act as a pressure correction. 64,65 However, this requires including an additional terms (volume) in the pair potential. To avoid these additional terms, we applied a pressure correction to the forcematched potentials using the iterative ramp correction described in Equation 4 (noting that this does not strictly address the statepoint dependence).…”

“…Within a force matching methodology, it is usually accepted that the pressure will not be predicted correctly and that these simulations should be run at constant volume, although pressure correction based on a volume-dependent term has been proposed. 63,64 However, Figure 2 shows that a simple linear pressure correction can successfully be applied to force matched potentials within the hybrid scheme employed here.…”

Direct route to reproducing pair distribution functions with coarse-grained models via transformed atomistic cross correlations

“…39 These issues can be sensitively dependent on chosen cut-off distances and enforced thermodynamic constraints, 39 and may be exacerbated as the complexity of the model increases due to increasing the number of distinct interactions or as the complexity of the optimization landscape increases through (i) enforcement of model restrictions (e.g., fixing a subset of parameters based on a given reference model) or (ii) incorporation of reference data from multiple state points. Moreover, structure-based potentials can be employed as a starting point for constructing CG models that also accurately represent thermodynamic [40][41][42] and dynamic [43][44][45] properties of the underlying reference model through a pressure-matching variational principle and the Mori-Zwanzig formalism, respectively. This motivates the development of direct optimization methods that can be efficiently combined with these methodologies into a unified optimization procedure.…”