DFSPH Pressure and Divergence Solver are Identical? about splishsplash

[janbender]

In pressureSolve it uses computeDensityAdv which calculates the exact same thing as computeDensityChange

That is not correct. computeDensityAdv computes the advected density, i.e. a preview of the density at time t+h (first equation in section 3.3). computeDensityChange computes the divergence of the velocity field (equation 6 in the paper). So the first is a density and the second the density derived wrt. t.

The computation of the solvers is the same since the matrix on the left hand side is equal, only the right hand side of the linear system changes.

[ibchristensen]

In computeDensityAdv:

densityAdv = density + h*delta;
densityAdv = max(densityAdv, density0);

In computeDensityChange:

densityAdv = density + h*densityAdv;
densityAdv = max(densityAdv, density0);
densityAdv = (densityAdv - density0) * (1.0 / h);

The value of delta and densityAdv are calculated exactly the same in both functions so the only difference is the last line in computeDensityChange. Looking at how the final value is used in pressureSolve (which uses computeDensityAdv):

const Real b_i = m_simulationData.getDensityAdv(i) - density0;
const Real ki = b_i*m_simulationData.getFactor(i);

And how factor is adjusted for the pressureSolve:
m_simulationData.getFactor(i) *= invH2;
vs how factor is adjusted for the divergenceSolve:
m_simulationData.getFactor(i) *= invH;
(this gets multiplied back out at the end of the divergenceSolve to reset the factor)
k_i ends up being the same value for both solvers.

[janbender]

Maybe there is a bug in the open-source implementation. I will check this. Unfortunately, this will take a while since currently I have not much time for the maintainance of the open-source libraries.