Résumé
Thermodiffusion (or thermophoresis) is the phenomenon by which the spatial distributions of constituents of liquid or gas phases become inhomogeneous in response to a temperature gradient. It has been evidenced in a variety of systems and has many practical applications as well as implications in the context of the origins of life. A complete molecular picture of thermophoresis is still missing, and phenomenological approaches are often employed to account for the experimental observations. In particular, the amplitude of the resulting concentration-gradients (quantified by the Soret coefficient) depends on many factors that are not straightforwardly rationalized. All-atom molecular dynamics simulations appear as an exquisite tool to shed light on the molecular origins for this phenomenon in molecular systems, but the practical implementation of thermophoretic settings in silico poses significant challenges. Here, we propose a robust approach to tackle thermophoresis in dilute realistic solutions at the molecular level. We rely on a recent enhanced heat-exchange algorithm to generate temperature-gradients. We carefully assess the convergence of thermophoretic simulations in dilute aqueous solutions. We show that simulations typically need to be propagated on long timescales (hundreds of nanoseconds). We find that the magnitude of the temperature gradient and the box sizes have little effect on the measured Soret coefficients. Practical guidelines are derived from such observations. Provided with this reliable setup, we discuss the results of thermophoretic simulations on several examples of molecular, neutral solutes, which we find in very good agreement with experimental measurements regarding the concentration-, mass-, and temperature-dependence of the Soret coefficient.