T. Verstraelen

The inŕuence of ŕuctuating charges or charge ŕow on the dynamic linear response properties of isolated molecules from the TS42 database is evaluated, with particular emphasis on dipole polarizability and C6 dispersion coefficients. Two new descriptors are deőned to quantify the charge-ŕow contribution to response properties, making use of the recoupled dipole polarizability to separate isotropic and anisotropic components. Molecular polarizabilities are calculated using the “frequency-dependent atom-condensed Kohn-Sham density functional theory approximated to second orderž, i.e. the ACKS2ω model. With ACKS2ω, the charge-ŕow contribution can be constructed in two conceptually distinct ways, which appear to yield compatible results. The charge-ŕow contribution is signiőcantly affected by molecular geometry and the presence of polarizable bonds, in line with previous studies. We show that the charge-ŕow contribution qualitatively reproduces the polarizability anisotropy. The contribution to the anisotropic C6 coefficients is less pronounced, but cannot be neglected. The effect of ŕuctuating charges is only negligible for small molecules with at most one non-hydrogen atom. They become important and sometimes dominant for larger molecules or when highly polarizable bonds are present, such as conjugated, double or triple bonds. Charge ŕow contributions cannot be explained in terms of individual atomic properties, because they are affected by non-local features such as chemical bonding and geometry. Therefore, polarizable force őelds and dispersion models can beneőt from the explicit modeling of charge ŕow.

Explicit-electron force fields introduce electrons or electron pairs as semiclassical particles in force fields or empirical potentials, which are suitable for molecular dynamics simulations. Even though semiclassical electrons are a drastic simplification compared to a quantum-mechanical electronic wave function, they still retain a relatively detailed electronic model compared to conventional polarizable and reactive force fields. The ability of explicit-electron models to describe chemical reactions and electronic response properties has already been demonstrated, yet the description of short-range interactions for a broad range of chemical systems remains challenging. In this work, we present the electron machine learning potential (eMLP), a new explicit electron force field in which the short-range interactions are modeled with machine learning. The electron pair particles will be located at well-defined positions, derived from localized molecular orbitals or Wannier centers, naturally imposing the correct dielectric and piezoelectric behavior of the system. The eMLP is benchmarked on two newly constructed data sets: eQM7, an extension of the QM7 data set for small molecules, and a data set for the crystalline beta-glycine. It is shown that the eMLP can predict dipole moments, polarizabilities, and IR-spectra of unseen molecules with high precision. Furthermore, a variety of response properties, for example, stiffness or piezoelectric constants, can be accurately reproduced.

Explicit-electron force fields introduce electrons or electron pairs as semiclassical particles in force fields or empirical potentials, which are suitable for molecular dynamics simulations. Even though semiclassical electrons are a drastic simplification compared to a quantum-mechanical electronic wave function, they still retain a relatively detailed electronic model compared to conventional polarizable and reactive force fields. The ability of explicit-electron models to describe chemical reactions and electronic response properties has already been demonstrated, yet the description of short-range interactions for a broad range of chemical systems remains challenging. In this work, we present the electron machine learning potential (eMLP), a new explicit electron force field in which the short-range interactions are modeled with machine learning. The electron pair particles will be located at well-defined positions, derived from localized molecular orbitals or Wannier centers, naturally imposing the correct dielectric and piezoelectric behavior of the system. The eMLP is benchmarked on two newly constructed data sets: eQM7, an extension of the QM7 data set for small molecules, and a data set for the crystalline beta-glycine. It is shown that the eMLP can predict dipole moments, polarizabilities, and IR-spectra of unseen molecules with high precision. Furthermore, a variety of response properties, for example, stiffness or piezoelectric constants, can be accurately reproduced.

This work presents a novel parametrization for the ReaxFF formalism as a means to investigate reaction processes of chlorinated organic compounds. Force field parameters cover the chemical elements C, H, O, Cl and were obtained using a novel optimization approach involving relaxed potential energy surface scans as training targets. The resulting ReaxFF parametrization shows good transferability, as demonstrated on two independent ab initio validation sets. While this first part of our two-paper series focuses on force field parametrization, we apply our parameters to the simulation of chlorinated dibenzofuran formation and decomposition processes in Part II.

Viscosity is a crucial property of liquid lubricants, and it is theoretically a well-defined quantity in molecular dynamics (MD) simulations. However, no standardized protocol has been defined for calculating this property from equilibrium MD simulations. While best practices do exist, the actual calculation depends on several ad hoc decisions during the post-processing of the raw MD data. A common protocol for calculating the viscosity with equilibrium MD simulations is called the time decomposition method (TDM). Although the TDM attempts to standardize the viscosity calculation using the Green–Kubo method, it still relies on certain empirical rules and subjective user observations, e.g., the plateau region of the Green–Kubo integral or the integration cut-off time. It is known that the TDM works reasonably well for low-viscosity fluids, e.g., at high temperatures. However, modified heuristics have been proposed at high pressures, indicating that no single set of rules works well for all circumstances. This study examines the effect of heuristics and ad hoc decisions on the predicted viscosity of a short, branched lubricant molecule, 2,2,4-trimethylhexane. Equilibrium molecular dynamics simulations were performed at various operating conditions (high pressures and temperatures), followed by post-processing with three levels of uncertainty quantification. A new approach, “Enhanced Bootstrapping”, is introduced to assess the effects of individual ad hoc parameters on the viscosity. The results show a strong linear correlation (with a Pearson correlation coefficient of up to 36%) between the calculated viscosity and an ad hoc TDM parameter, which determines the integration cut-off time, under realistic lubrication conditions, particularly at high pressures. This study reveals that ad hoc decisions can lead to potentially misleading conclusions when the post-processing is performed ambiguously.