[EternalSeekerX]

Hey everyone,

So bare with me as I am new and learning openfoam. I have been doing a few tutorials and validation cases and I notice that with my 6 core laptop, simpleFOAM solves way faster than the other two solvers (even for the same geometry). Im not complaining but I am wondering if i am doing something wrong xD

[Tobermory]

Nope - that's totally as expected, for most steady state problems.

[joegi.geo]

Well, your question is a little bit vague. That depends on what you want to do.


If you want to reach an steady behavior (like flat time series of forces), steady solvers might be faster (but not necessary). I can show you a gazillion cases where unsteady solvers are faster, but those cases are the exception rather than the rule.



Generally speaking, steady solver are faster (if you use the optimal under relaxation factors). However, they are not as accurate as unsteady solvers.


Rule of thumb:


Use steady solvers if you are sure that you are expecting a steady behavior; otherwise, use unsteady solvers or local time stepping (pseudo transient).

[EternalSeekerX]

Originally Posted by joegi.geo

Well, your question is a little bit vague. That depends on what you want to do.


If you want to reach an steady behavior (like flat time series of forces), steady solvers might be faster (but not necessary). I can show you a gazillion cases where unsteady solvers are faster, but those cases are the exception rather than the rule.



Generally speaking, steady solver are faster (if you use the optimal under relaxation factors). However, they are not as accurate as unsteady solvers.


Rule of thumb:


Use steady solvers if you are sure that you are expecting a steady behavior; otherwise, use unsteady solvers or local time stepping (pseudo transient).

I see, I brought it up because for example for a 3D Square Cylinder the RANS version using simpleFoam runs pretty fast (I assumed because it using a 1 second time step) while the LES version using pimpleFOAM took 40hrs. I assume its because of time step as well since LES case used really small time steps (which makes sense for vortex capturing).

[Bloerb]

In CFD you have to solve for pressure p and velocity U but can't do so easily within one big matrix (nonlinear system and two variables). SIMPLE/PISO/PIMPLE are segregated solution techniques for such PDEs. Hence you split the system into two equation systems in a segregated solver. You need to solve this set of two systems in a loop to get an accurate solution of the intial system.
Solve A -> solution into B -> solve B -> solution into A --> ... --> until A and B fulfill the original system.... This is the PISO loop.

• The SIMPLE algorithm uses only one pass through the loop per time step and keeps the solution stable by relaxation.
• The PISO algorithm uses many loops per time step until convergence (these are the nCorrectors) and does not need relaxation to not blow up
• The PIMPLE algorithm is SIMPLE+PISO. So the first loops are SIMPLE within in a time step and the last is PISO. (Hence relaxation factors are prescribed differently for the last loop (e.g U and UFinal))

The benefit of relaxation is that you have no limit in time step. You can choose every value you want. SIMPLE gets away with unlimited time steps without even properly solving each time step due to relaxation, but looses time accuracy due to it (meaning you can't resolve changes in time correctly like vorticies etc). Hence it is ideal for steady state problems. PISO on the other hand does not use relaxation and hence has a limit on the time step. Which means, in order for the solution to not blow up, you need a small enough time step to capture the solution and keep errors small, and enough corrector loops per time step. PIMPLE allows for bigger time steps by combining the two. In theory allowing bigger time steps and time accuracy due to the mixed approach.


As already mentioned this does not always equate to one being faster than the other but it usually does. Relaxation means limiting the change in your solution. Hence your solution does not blow up as easily. It caps the change that is possible. This means many many solution cycles until you reach the steady state. But they are easier to solve for. For some problems this relaxation means immensely slowing the solution down, since the change itself might already be small. One example are diffusion heavy low Reynolds number flows. Here a steady state forms nearly instantaneously in real time. Hence PISO can do just fine, since the increased time step due to relaxation looses its benefit.