Kahan summation

In the paper that was sent and reference to me by one of the team members (to my work email account), re: Longshaw and Rogers (2015). "Automotive fuel cell sloshing under temporally and spatially varying high acceleration using GPU-based Smoothed Particle Hydrodynamics (SPH)" doi:10.1016/j.advengsoft.2015.01.008. ", they mentioned implementing the Kahan summation in the paper in order to reduce the error that is introduced for longer running SPH simulations and also errors introduced due to them using single-precision calculations and the second order Verlet integration scheme.

Has anybody here studied this in greater detail?

I am also noticing this as well where if I run, for example, the Case_Dambreak3D longer than the example case specifies, (5 to 10 seconds total simulation time), the number of particles that are excluded (particle out, due to invalid rhop) starts to increase to several thousand particles.

I'm not really certain that I understand why the density will change if this is supposed to be an incompressible fluid/case.

I've only just started using DualSPHysics less than a week ago, so please, do pardon my ignorance.

Thank you.


  • DualSPHysics uses a weakly compressible SPH scheme, therefore nothing is "truly" incompressible, but most of the time, using correct parameters, like speed of sound at least 10 times higher than max velocity in sim., then the density fluctuations have been found to be around 1%.

    Kind regards
Sign In or Register to comment.