Detailed molecular simulations to investigate multicomponent diffusion models

Authors

  • Harshit A. Patel,

    1. The Isermann Dept. of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180
    Search for more papers by this author
  • Shekhar Garde,

    1. The Isermann Dept. of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180
    Search for more papers by this author
  • E. Bruce Nauman

    Corresponding author
    1. The Isermann Dept. of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180
    • The Isermann Dept. of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180
    Search for more papers by this author

Abstract

The theoretical treatment of multicomponent diffusion is complicated by the generally unknown dependence of diffusivities on the local concentration of species. As pointed out by Nauman and Savoca in 2001, the standard treatment uses an n-1 by n-1 matrix of diffusion coefficients for an n-component system and can give anomalous and non-physical results when there is no dominant component and when the various components have significantly different diffusivities. Although theoretically resolved by postulating diffusivities with suitable concentration dependence, there has been no practical resolution of this problem short of unrealistic, exhaustive experimentation. Nauman and Savoca proposed two models for multicomponent diffusion that produce only physically possible results, but they were unable to suggest which model was better. This article reports on molecular dynamic experiments that were designed to differentiate between the models. Specifically studied were ternary, liquid mixtures of ethane, octane, and hexadecane. It was found that the proportional flux model agrees with the molecular simulations. No cross-diffusion was observed in agreement with this model and in contrast to the alternative, pair-wise flux model. The proportional flux model is easy to implement and requires a minimum of data, although detailed, compositional dependent diffusion coefficients can be incorporated into the model when such data are available. © 2005 American Institute of Chemical Engineers AIChE J, 2006

Ancillary