Visualisation of dam break with 1, 10, 100 million particles

I have uploaded to a YouTube playlist several short videos (20 sec) of the dam break simulation run with 1, 10, 100 million fluid particles. The test case is the 3D dam break from the main suite of examples in DualSPHysics 4.4.003 (10-04-2019).

The videos show the particle positions, the velocity field and the density field. Thanks to the post-processing scripts in DualSPHysics toolbox, I have used the two ParaView rendering modes particles and surface.

I have taken the default XML file and changed the particle spacing in order to achieve the desired particle count, and then let it go. The simulations have been run in double precision.            

particle spacing and number of fluid particles:

dp="0.000928"    Nfp=100,372,573

 dp="0.00200"    Nfp= 10,020,000

 dp="0.00250"     Nfp= 5,126,400

 dp="0.00425"    Nfp= 1,026,480

 dp="0.00450"    Nfp= 882,524 

 dp="0.00500"     Nfp= 644,300


Clearly, simulations should converge to the experimental evidence. So it should not be taken for granted that the particle spacing is the sole parameter to play around with in order to obtain an accurate simulation. I might release more information on validation and on compute performance at a later time.

Nonetheless, this visualisation exercise already shows the dramatic differences that a modeller comes across when the particle count increases, which is technically possible thanks to ever more powerful GPUs. Hope this helps us appreciate the capabilities of SPH and DualSPHysics as well as the care needed in modelling fluid flows appropriately.

For a sneak preview:


  • This is very cool. I would like to hear about the GPU used for simulation and how long it took. I experienced that 12 million particles would take 48 hours with an advanced cloud GPU, so I assume this would be 1½ week approximately in time with 100 million?

    As you also mention making the resolution finer, is not necessarily increasing experimental convergence, but hopefully the result becomes stable numerically, i.e. result at 100 million particles is the same as with 10 million and so forth.

    Kind regards

  • In the case of dam-breaks where the interaction between fluid and dry bottom is key in the dam evolution, the parameter that plays a more important role is the viscosity, at least using DBC.

    So in order to obtain the same dam celerity or the the same agreement with experimental or reference data, the resolution (number of particles) or "dp" should change with the value of artificial viscosity.

    You can also observe this just by playing with the 2-D dam break.

  • This is very interesting ! Thanks for sharing !

  • The video at shows the 'dam break' animation with 100M particles in true time: the animation is just as long as the simulated time, that is 1.5 s. I thought this one renders pretty well the common statement that SPH is particularly suited for simulating rapid and violent flows!

Sign In or Register to comment.