[Trilinos-Users] [EXTERNAL] Re: AztecOO convergence using Sundance

[Roger Pawlowski]

Andrea,

The simplest things to try are:

1. Increase the overlap or fill in values for the ifpack preconditioner.

2. ILU doesn't scale, so you could switch to the ML package for 
algebraic multilevel preconditioning (assuming you unknowns are uniform 
across your grid) or use physics-based preconditioning as Kevin 
suggested.  The Teko package is a toolset for building physics-based 
preconditioners in Trilinos.

3. Loosen the Krylov solver tolerance.  1e-10 seems excessive for 
navier-stokes.  We typically run using 1e-4 on the non-dimensionalized 
equations.

Roger

On 03/12/2012 11:09 PM, Kevin Long wrote:
>
> Andrea,
>
> You'll have trouble solving the linear subproblem with ILU 
> preconditioning. You might talk to Vicki Howle, who has implemented 
> physics-based preconditioners for NS equations using Sundance.
>
> Kevin
>
>
> On 03/12/2012 08:31 PM, Andrea Penza wrote:
>> Dear experts,
>>
>> I'm trying since days to achieve convergence solving Navier-Stokes 
>> equations using Aztec solver. Particularly I'm intended to use a 
>> GMRES method and ILU preconditioner (I think this should be suit 
>> right to my problem), and the current solver parameters are the 
>> following:
>>
>> <ParameterList>
>> <ParameterList name="NOX Solver">
>> <Parameter name="Nonlinear Solver"type="string"value="Line Search 
>> Based"/>
>> <ParameterList name="Line Search">
>> <Parameter name="Method" type="string" value="More'-Thuente"/>
>> </ParameterList>
>> <ParameterList name="Status Test">
>> <Parameter name="Max Iterations" type="int" value="20"/>
>> <Parameter name="Tolerance" type="double" value="1e-8"/>
>> </ParameterList>
>> <ParameterList name="Printing">
>> <ParameterList name="Output Information">
>> <Parameter name="Error" type="bool" value="true"/>
>> <Parameter name="Warning" type="bool" value="true"/>
>> <Parameter name="Outer Iteration" type="bool" value="true"/>
>> <Parameter name="Inner Iteration" type="bool" value="true"/>
>> <Parameter name="Parameters" type="bool" value="false"/>
>> <Parameter name="Details" type="bool" value="false"/>
>> <Parameter name="Linear Solver Details" type="bool" value="true"/>
>> <Parameter name="Outer Iteration StatusTest"type="bool"value="true"/>
>> <Parameter name="Test Details" type="bool" value="false"/>
>> <Parameter name="Stepper Iteration" type="bool" value="false"/>
>> <Parameter name="Stepper Details" type="bool" value="false"/>
>> <Parameter name="Stepper Parameters" type="bool" value="false"/>
>> <Parameter name="Debug" type="bool" value="false"/>
>> </ParameterList>
>> </ParameterList>
>> <ParameterList name="Linear Solver">
>> <Parameter name="Max Iterations" type="int" value="1000"/>
>> <Parameter name="Method" type="string" value="GMRES"/>
>> <ParameterList name="Preconditioner">
>> <Parameter name="Type" type="string" value="Ifpack"/>
>> <Parameter name="Prec Type" type="string" value="ILU"/>
>> <Parameter name="Overlap" type="int" value="1"/>
>> <ParameterList name="Ifpack Settings">
>> <Parameter name="fact: level-of-fill" type="int" value="2"/>
>> </ParameterList>
>> </ParameterList>
>> <Parameter name="Tolerance" type="double" value="1e-10"/>
>> <Parameter name="Type" type="string" value="Aztec"/>
>> <Parameter name="Verbosity" type="int" value="0"/>
>> </ParameterList>
>> </ParameterList>
>> </ParameterList>
>>
>>
>>
>> This brings to message like:
>>
>> WARNING: NOX::Direction::Newton::compute() - Linear solve failed to 
>> achieve convergence - using the step anyway since "Rescue Bad Newton 
>> Solve" is true
>>
>> Notice that the same problem was solved without problems using 
>> NOX-AMESOS solver, but this did not suit fine to run in parallel as I 
>> need.
>> So now I'm investigating iterative methods in order to run parallel 
>> jobs. Maybe I'm using some parameter in a wrong way, has anyone idea 
>> about something could improve this issue?
>> Thanks in advance for your support.
>> Kind regards,
>>
>> Andrea
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>
>
> -- 
> Kevin Long
> Associate Professor
> Department of Mathematics and Statistics
> Texas Tech University
>
> "The end of fear is the beginning of wisdom"
> - Bertrand Russell
>
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20120313/34a99e47/attachment.html