Machine learning encompasses tools and algorithms that are now becoming popular in almost all scientific and technological fields. This is true for molecular dynamics as well, where machine learning offers promises of extracting valuable information from the enormous amounts of data generated by simulation of complex systems. We provide here a review of our current understanding of goals, benefits, and limitations of machine learning techniques for computational studies on atomistic systems, focusing on the construction of empirical force fields from ab initio databases and the determination of reaction coordinates for free energy computation and enhanced sampling.

Sets of atoms collectively behaving as rigid bodies are often used in molecular dynamics to model entire molecules or parts thereof. This is a coarse-graining strategy that eliminates degrees of freedom and supposedly admits larger time steps without abandoning the atomistic character of a model. In this paper, we rely on a particular factorization of the rotation matrix to simplify the mechanical formulation of systems containing rigid bodies. We then propose a new derivation for the exact solution of torque-free rotations, which are employed as part of a symplectic numerical integration scheme for rigid-body dynamics. We also review methods for calculating pressure in systems of rigid bodies with pairwise-additive potentials and periodic boundary conditions. Finally, simulations of liquid phases, with special focus on water, are employed to analyze the numerical aspects of the proposed methodology. Our results show that energy drift is avoided for time step sizes up to 5 fs, but only if a proper smoothing is applied to the interatomic potentials. Despite this, the effects of discretization errors are relevant, even for smaller time steps. These errors induce, for instance, a systematic failure of the expected equipartition of kinetic energy between translational and rotational degrees of freedom.

-The Boundary Driven Non-Equilibrium Molecular Dynamics (BD-NEMD) method is employed to evaluate Soret coefficients of binary mixtures. Using a n-decane/n-pentane mixture at 298 K, we study several parameters and conditions of the simulation procedure such as system size, time step size, frequency of perturbation, and the undesired warming up of the system during the simulation. The Soret coefficients obtained here deviated around 20% when comparing with experimental data and with simulated results from the literature. We showed that fluctuations in composition gradients and the consequent deviations of the Soret coefficient may be due to characteristic fluctuations of the composition gradient. Best results were obtained with the smallest time steps and without using a thermostat, which shows that there is room for improvement and/or development of new BD-NEMD algorithms.