DFSPH + Volume mapping: avoiding collisions and particle "squeezing".

[twanverweij]

Dear mr. Bender and team,

First of all a bit of gratitude i'm fairly having a blast with the splishsplash code, very well organized and so easy to modify.

I have a case where I push some particles into a receptacle but i don't want to push them too hard as this gives the pressure solver a hard time. So, to avoid this, I'm looking to confirm some impressions about volume mapping and boundary detection:

• On volume map resolution: part A has dimension X spanning from min_x to max_x. when I apply resolution to the SDF map, will that chop up into cells with width (max_x - min_x)/resolution? And if so, is the grid spaced from the volumetric center or do the cell nodes align with the bounding box around the geometry at say, (min_x, min_y, min_z)? There is some empty space surrounding the actual walls, I'm looking to make the fit a bit better without going too wild with the cell count.

• For a second part B with half the size, will resolution lead to half the cell width of the volumes?

• is a particle allowed to live anywhere between the volume-map (so not the actual geometry) of two parts, given both are half a particle radius apart, or do I need to leave at least two or three particle widths, i.e. should I be able to squeeze a row of particles through a narrow channel?

Thanks in advance,

Twan

[janbender]

1. The domain of the grid is a bit larger than the bounding box and is aligned with the enlarged bounding box. You find the code here:

SPlisHSPlasH/Simulator/SimulatorBase.cpp

Line 2585 in 08f9fc8

domain.max() += (8.0*supportRadius + tolerance) * Eigen::Vector3d::Ones();

Btw. the domain is printed when starting the simulation.

2. It won't be exactly the half cell size since the bounding box is extended a bit (see 1)

3. The boundary representation uses a smoothing kernel. The boundary is smoothed out. You can put a particle exactly next to the boundary but probably it will be pushed away a bit due to the kernel influence.

[twanverweij]

Dear Jan,

Many thanks, from semi-learned trial and error I find that a full particle-diameter offset captures the geometries very accurately without going all out on the SDF map resolution; geometry dimension / particle radius = res gives good / efficient results.