High-Throughput Molecular Dynamics Simulations and Validation of Thermophysical Properties of Polymers for Various Applications

Abstract

Recent advances in graphics processing unit (GPU) hardware and improved efficiencies of atomistic simulation programs allow for the screening of a large number of polymers to predict properties that require running and analyzing long molecular dynamics (MD) trajectories. This paper outlines a MD simulation workflow based on GPU MD simulation and the refined optimized potentials for liquid simulation (OPLS) OPLS3e force field to calculate glass transition temperatures (T gs) of 315 polymers for which Bicerano reported experimental values [BiceranoJ. Bicerano, J. Prediction of Polymer PropertiesMarcel Dekker Inc.New York1996]. Applying the workflow across this large set of polymers allowed for a comprehensive evaluation of the protocol performance and helped in understanding its merits and limitations. We observe a consistent trend between predicted T g values and empirical observation across several subsets of polymers. Thus, the protocol established in this work is promising for exploring targeted chemical spaces and aids in the evaluation of polymers for various applications, including composites, coatings, electrical casings, etc. During the stepwise cooling simulation for the calculation of T g, a subset of polymers clearly showed an ordered structure developing as the temperature decreased. Such polymers have a point of discontinuity on the specific volume vs temperature plot, which we associate with the melting temperature (T m). We demonstrate the distinction between crystallized and amorphous polymers by examining polyethylene. Linear polyethylene shows a discontinuity in the specific volume vs temperature plot, but we do not observe the discontinuity for branched polyethylene simulations.

Benchmarking Machine Learning Models for Polymer Informatics: An Example of Glass Transition Temperature

“…77−79 It is acknowledged that MD conditions are not exactly consistent with experiments such as the MD's high cooling rate on the nanosecond time scale. 78,80−82 Nevertheless, a consistent trend between MD simulated T g and experimental observation has been demonstrated by Afzal et al 83 over 315 common polymers. To verify the reliability of our MD simulations, we also selected 100 polymers in our Data set_1 test set for MD simulations through K-means clustering (K = 100).…”

“…Thus, the PCFF force field is particularly suitable for the molecular simulation of polymer’s T g value . A cooling process simulation generates the specific volume vs temperature curve, from which rubbery phase and glassy phase are identified, and their intersection gives the T g value. − It is acknowledged that MD conditions are not exactly consistent with experiments such as the MD’s high cooling rate on the nanosecond time scale. ,− Nevertheless, a consistent trend between MD simulated T g and experimental observation has been demonstrated by Afzal et al over 315 common polymers. To verify the reliability of our MD simulations, we also selected 100 polymers in our Data set_1 test set for MD simulations through K -means clustering ( K = 100).…”

Design of Organic Electronic Materials With a Goal-Directed Generative Model Powered by Deep Neural Networks and High-Throughput Molecular Simulations

“…Owing to the computational cost of predicting glass transition temperatures ( T g ) using physics-based simulations ( Afzal et al, 2021 ), we opted to use a QSPR model instead. The QSPR model was trained to predict T g based on 250 known OLED materials from the literature ( Naito and Miura, 1993 ; Fujikawa et al, 2000 ; Shirota, 2000 ; Yin et al, 2003 ; Kimura et al, 2005 ; Xu and Chen, 2005 ; Gao et al, 2007 ; Shirota and Kageyama, 2007 ).…”

United-Atom Molecular Dynamics Study of the Mechanical and Thermomechanical Properties of an Industrial Epoxy

“…Under such conditions, sufficient chain relaxation cannot occur, and the rubbery state acts as a glassy state. Interestingly, Afzal et al [ 33 ] found that the discrepancy between measured and calculated T g is consistent within a simulation protocol, and hence could be accounted for with proper calibration. Alternatively, the difference between experimental and computational rates could potentially be bridged using time-temperature superposition [ 24 , 31 ].…”