Accurate force fields are necessary for predictive molecular simulations. However, developing force fields that accurately reproduce experimental properties is challenging. Here, we present a machine learning directed, multiobjective optimization workflow for force field parametrization that evaluates millions of prospective force field parameter sets while requiring only a small fraction of them to be tested with molecular simulations. We demonstrate the generality of the approach and identify multiple low-error parameter sets for two distinct test cases: simulations of hydrofluorocarbon (HFC) vapor–liquid equilibrium (VLE) and an ammonium perchlorate (AP) crystal phase. We discuss the challenges and implications of our force field optimization workflow.

A fully atomistic model designed to reproduce the molecular weight and average functionality typical of commercial hydroxyl-terminated polybutadiene (HTPB) formulations is presented. The structural and dynamical properties of HTPB were investigated utilizing molecular dynamics (MD) simulations. Spatial conformations of the HTPB chains were characterized by using the radius of gyration and end-to-end distance; HTPB chain dynamics were described by the end-to-end vector autocorrelation function. Terminal hydroxyl (OH) groups were found to associate in dynamic aggregates of various sizes. The distribution of the OH aggregate sizes and their relative stabilities were calculated. The significance of OH association in relation to curing reactions is discussed.

The diffusion of a CO 2 /CH 4 mixture in carbon nanotube (CNT) bundles was studied using molecular simulations. The effect of diameter and temperature on the diffusion of the mixture was investigated. Our results show that the single-file diffusion occurs when CO 2 and CH 4 are confined in CNTs of diameter less than 1.0 nm. In CNTs of diameter larger than 1.0 nm, both molecules diffuse in the Fickian style. The transition from single-file to Fickian diffusion was demonstrated for both CO 2 and CH 4 molecules. A dual diffusion mechanism was observed in the studied (20, 0) CNT bundle, single-file diffusion of CO 2 in the interstitial sites of (20, 0) CNT bundle and Fickian diffusion of CO 2 and CH 4 in the pores. For CO 2 , the interaction energies (CO 2 -CO 2 and CO 2 -CNT) are larger than that of CH 4 in all cases. But only a very small difference in the diffusion coefficient was observed between CO 2 and CH 4 . Temperature has a negligible effect on the difference between diffusion coefficients of CO 2 and CH 4 in the studied CNT bundles. The adsorption, diffusion and permeation selectivities are discussed and compared, and the adsorption is demonstrated to be the rate limiting step for the separation of CO 2 /CH 4 in CNT bundle membranes.

The applicability of an atomistic Class II force field to capture the properties of the orthorhombic crystal phase of ammonium perchlorate was investigated. Structural and dynamical behaviors including density, lattice parameters, bulk modulus, infrared spectrum, and rotational dynamics were calculated from the trajectories of molecular dynamics (MD) simulations. Properties calculated from MD were compared to available experimental data over a range of temperatures, including those significantly higher than the parameterization temperature of 10 K.

A computational method for preserving the equilibrated structure of a fully atomistic model of a prepolymer melt while introducing cross-links is introduced. An atomistic model of hydroxyl-terminated polybutadiene (HTPB) is cured using isophorone diisocyanate (IPDI) to form an HTPB-IPDI elastomer. After cross-linking, the elastomer is quenched to several different temperatures and subsequently stressed via both compressive and tensile deformations to assess mechanical properties as a function of temperature. This study establishes the framework for investigating the degradation and embrittlement of HTPB-IPDI binders by molecular simulation.