## Navier Stokes using Mumps

Numerical methods and mathematical models of Elmer
mark smith
Posts: 207
Joined: 26 Aug 2009, 18:20
Location: Peterborough, England

### Navier Stokes using Mumps

Hi All,
I've finally updated to the Ubuntu distribution so as I can use the Mumps solver, can someone send me a minimal solver setting for N-S using Mumps for use in my sif files please?
There are various sif's in the examples using Mumps but they all seem to include iterative solvers as well or they are commented out ?
Best regards
Mark

raback
Posts: 3404
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

### Re: Navier Stokes using Mumps

Hi Mark

Code: Select all

``````Linear System Solver = direct
Linear System Direct Method = mumps
``````
-Peter

mark smith
Posts: 207
Joined: 26 Aug 2009, 18:20
Location: Peterborough, England

### Re: Navier Stokes using Mumps

Hi Peter,
Surely this is not enough as it produces lots of of warnings about Nonlinear System Convergence Tolerance not found etc? and the output is just the boundary condition values.

So what values would you advise for

Nonlinear System Convergence Tolerance =
Nonlinear System Max Iterations =
Nonlinear System Newton After Iterations =
Nonlinear System Newton After Tolerance =
Nonlinear System Relaxation Factor =

And any thing else that might be required?

Best regards
Mark

raback
Posts: 3404
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

### Re: Navier Stokes using Mumps

Hi Mark

The solution strategies are divided into "linear system" and "nonlinear system". The choices for MUMPS only deal with the linear system while the missing keywords are related to nonlinear system.

Nonlinear system settings are quite problem specific. One would typically start without relaxation and turning on Newton early on. Newton's convergence rate is faster than that of Picard (the other method) but the radius of convergence is smaller. Hence you might try to resolve convergence problems to some limit trying with relaxation (down to ~0.5, or maybe even 0.3) and keeping Picard on longer.

The convergence of nonlinear cases may also be aided by shorter timesteps. Only flows with Re<~100 have steady state solutions. When trying to find steady state solution for problems where it does not exist often leads to toggling between different attractors.

-Peter