Investigation of coarse-grained models across a glass transition

ABSTRACT

Due to their computational efficiency, coarse-grained (CG) models have become increasingly popular for simulating soft condensed matter. At least in principle, bottom-up CG models can reproduce the properties of all-atom (AA) models that are observable at the CG resolution. Unfortunately, the resulting effective potentials vary with thermodynamic state point, which can significantly limit the range of densities and temperatures for which the CG model is valid. In this study, we revisit these considerations for a 3-site CG model of ortho-terphenyl (OTP), which is a representative glass former. We employ force-matching and self-consistent pressure matching to parameterize the CG models. The resulting models accurately reproduce the OTP pair structure and pressure–volume equation of state at each state point for which they were parameterized. Above the glass transition, the effective potentials vary monotonically with temperature and density, as expected for molecular liquids. However, below the glass transition, these simple trends do not hold. Nevertheless, the effective potentials generally appear more sensitive to density than temperature. Moreover, despite this state-point dependence, the potentials appear reasonably transferable in the sense that they reasonably describe OTP across a fairly wide density and temperature range that spans the glass transition. Interestingly, the glass phase potentials appear most accurate and transferable. Conversely, the potentials parameterized near the glass transition appear least accurate and transferable.

KEYWORDS:

• Multiscale modeling
• transferability
• coarse-graining
• ortho-terphenyl
• glass
• phase transition

Acknowledgments

The authors gratefully acknowledge the financial support of the National Science Foundation (Grant No. CHE-1565631). RJS would like to thank Kathryn Lebold for many helpful discussions in analyzing the coarse-grain models, as well as Michael DeLyser for assistance with the computational methods. Parts of this research were conducted with XSEDE resources awarded by TG-CHE150090 and TG-CHE170062. This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number ACI-1548562.^[81] Additional computations were conducted with Advanced CyberInfrastructure computational resources provided by the Institute for Cyber Science at the Pennsylvania State University. (http://ics.psu.edu)

Additional information

Funding

This work was supported by the National Science Foundation [CHE-1565631]; National Science Foundation [ACI-1548562].