Can Artificial Viscosity model real viscosity?


In some papers they argue that based on a paper Monaghan has written it is possible for a 2D Wendland kernel to say that:

Which means that one can turn alpha into a fixed parameter based on a physical value instead of a free-parameter!

I notice though that this is not included in DualSPHysics wiki, which makes me think that this kind of manevour is not actually okay. Would any of the developers or more knowledgeable people about the theory please explain why/why not, this is okay to do?

It would be a great help for me atleast, but also others at large.

Monaghan's work on it can be found here:

Chapter 8.1, specifically equation 8.8

Kind regards


  • One thing you may already note in the DualSPHysics implementation is that the artificial viscosity only acts when pair of particles come closer. See This means that artificial viscosity will slow down colliding particles but will not hinder the formation of particle gaps, which is fixed by other means. The idea of becoming closer or farther away is an artefact of the numerical discretisation in "particles".

    Viscosity in a continuum medium acts if there is a gradient of velocity, and works towards evening out differences in both directions: the slower become faster, the faster become slower.

    My two cents on artificial and physical viscosity.

  • I think that is a good way to think about it, and why artificial viscosity as it is now will never be able to model real viscosity accurately, only in shocktube simulations.

    Thanks for your comment.

  • Hi,

    perhaps you can find some more insight into it in the paper from Meringolo et al. (2019) where the authors explicitly link the artificial viscosity parameter to the cell Reynolds number, aiming at solving in SPH all the main vorticity scales or, at least, tending to it.

    Here the link to the paper:

Sign In or Register to comment.