[wildfire230]

Hi all,


I have been a long-time user of the icoFoam solver for my transient flow problems. However, I have now run into a case which requires several million grid cells for spatial convergence, and tens of thousands of time steps using the icoFoam solver due to the time step limitation. I am wondering if I should switch to the pimpleFoam solver for my case in order to take advantage of the higher Courant numbers that can be used, but I am unclear about when exactly it makes sense to do use pimpleFoam vs. icoFoam. For example, pimpleFoam can often use tens or hundreds of additional iterations per time step, so it is not obvious to me that the overall calculation will actually proceed faster, even if you can use a 10x larger time step.


Are there any general guidelines on when to pick one or the other, especially regarding accuracy and the overall speed of the simulation?


Thank

[RobertHB]

TL;DR: Bold text at the end of my post.

Originally Posted by wildfire230

[...]I am wondering if I should switch to the pimpleFoam solver for my case in order to take advantage of the higher Courant numbers that can be used, but I am unclear about when exactly it makes sense to do use pimpleFoam vs. icoFoam.

PimpleFoam is a combination of the transient solver PISO and the steady-state solver SIMPLE. If you use an empty fvSolute PIMPLE { } entry the solver behaves like pisoFoam. If you then use turbulenceModel laminar; in your turbulenceProperties you have a transient, laminar solver. Like icoFoam. As such, there is nothing in icoFoam that pimpleFoam can't do. The benefit of pimpleFoam is the combination of PISO and SIMPLE. Which leads to your next question:

Originally Posted by wildfire230

[...]For example, pimpleFoam can often use tens or hundreds of additional iterations per time step, so it is not obvious to me that the overall calculation will actually proceed faster, even if you can use a 10x larger time step.

The number of iterations is controlled by two keywords:

PIMPLE
{ 
    nOuterCorrectors             20;
    nCorrectors                 2;
  }

Both values are equal to one (1.0) if the solver behaves like PISO. When you change nOuterCorrectors to any value above one, PIMPLE becomes active and spends nOuterCorrectors iterations (here 20) on finding a steady-state solution for the timestep. Then procedes with the next timestep (like a transient solver) and, again, spends 20 iterations on finding a stable solution. nCorrectors adds further stability but at the cost of more iteration. In General you will have at timestep dt, nOuterCorrectors iterations, each including nCorrectors iterations of pressure correction. In my example for a timestep dt, 20 nOuterCorrectors and 40 nCorrectors iterations.
Combined with relaxationFactors from the SIMPLE algorithm, which restrict the change of a variable from one nOuterCorrectors iteration to the next, you can add a lot of stability to your simulation. And in the same time increase you CFL-number. From my experience a CLF-number of 4.0 works without problems. I tried up to CLF 20.0, but that gets a bit unstable at times.

To return to you original question:

Originally Posted by wildfire230

[...]but I am unclear about when exactly it makes sense to do use pimpleFoam vs. icoFoam.

When your prior simulation is so complex that it eats up alot of computational time or the solution diverges quickly. Then, and if you can live with some averaging of your results, a pimpleFoam simulation may be a good choice.


Further reading:
[cfd-online] pimpleFoam vs simpleFoam vs pisoFoam vs icoFoam?
[Holzmann-cfd.de] Mathematics, Numerics, Derivations and OpenFOAM - For an introduction to pimpleFoam see chapter 11.4.

[akidess]

Originally Posted by RobertHB

If you use an empty fvSolute PIMPLE { } entry the solver behaves like pisoFoam.

Both values are equal to one (1.0) if the solver behaves like PISO.

This is not strictly true, for PISO you can and should also use iterations (Issa recommended up to 3).

Further reading:
R. I. Issa, Solution of the Implicitly Discretized Fluid Flow Equations by Operator-Splitting, Journal of Computational Physics, 62, pp 40-65, 1985

[RobertHB]

Originally Posted by akidess

This is not strictly true, for PISO you can and should also use iterations (Issa recommended up to 3).

Further reading:
R. I. Issa, Solution of the Implicitly Discretized Fluid Flow Equations by Operator-Splitting, Journal of Computational Physics, 62, pp 40-65, 1985

Thanks for the literature source. I wasn't aware of the fact that pisoFoam should be run with multiple correctors.

[akidess]

Originally Posted by RobertHB

I wasn't aware of the fact that pisoFoam should be run with multiple correctors.

This is also done in the pisoFoam tutorials by the way.