[dune-pdelab] some problems in dune

Robert Jendrny Robert.Jendrny at math.tu-dortmund.de
Fri Oct 31 10:17:28 CET 2014


Hey,

ok, here you get the output for CG_AMG_SSOR (first line: Vertices, 2nd 
line: elements, 3rd line: convergence rate, 4th line: solveriterations)
This solver is the fastest in time, but the iteration number "explodes" ;-)

In "theory" i learned that the iteration number should be nearly 
constant ... but this I learned for GMG and not for AMG ... But they 
should behave in the same way

Robert


On 10/30/2014 08:25 PM, Rebecca de Cuveland wrote:
> Hey,
>
> could you also give us the output of your code? Sometimes the calculation depend on the installed software on ones computer…
> Thanks,
>
> Rebecca
> Am 30.10.2014 um 15:18 schrieb Robert Jendrny <Robert.Jendrny at math.tu-dortmund.de>:
>
>> Hi again,
>>
>> maybe the fully code would be more helpful ;-)
>>
>> All the files which end with "_orig.hh" are the original ones from the course which I changed in the files with missing '_orig'
>>
>> Cya
>> Robert
>>
>> On 10/30/2014 03:14 PM, Robert Jendrny wrote:
>>> Hi Rebecca,
>>>
>>> thank you for your answer ...
>>>
>>> I am working with the code from pdelab exercise2, example02 (Dune course this year - 2014)
>>> I hope the attached files are enough
>>>
>>> The most important file is example02_Q1.hh, there everything is calculated
>>> I want to solve Poisson (-Laplace(u) = 1) on unit square with AMG
>>>
>>> For Q1 I nearly get independent numbers of iterations for mesh size h, h/2, h/4,.... which is fine
>>>
>>> But If I change Q1 to Q2, the numbers of iterations "explode" if the mesh gets finer and finer
>>>
>>> Maybe someone can fix it ;-)
>>>
>>> Bye,
>>> Robert
>>>
>>> On 10/30/2014 03:00 PM, Rebecca de Cuveland wrote:
>>>> Hey Robert,
>>>> perhaps you can show us your program and output? (verbosity 3 or 4 for AMG and 1 for linear solver)
>>>> AMG is quite sensible for the used options. I only use Q1, but as far as I know Q2 should also work...
>>>> Bye,
>>>> Rebecca
>>>>
>>>
>>
>> -- 
>> Robert Jendrny
>> B.Sc. Technomathematik
>>
>> TU Dortmund, Fakultät für Mathematik
>> LS III: Angewandte Mathematik und Numerik
>> M 1020, Vogelpothsweg 87, 44227 Dortmund
>> robert.jendrny at math.tu-dortmund.de
>>
>> <pdelab-exercise2-sol.zip>


-- 
Robert Jendrny
B.Sc. Technomathematik

TU Dortmund, Fakultät für Mathematik
LS III: Angewandte Mathematik und Numerik
M 1020, Vogelpothsweg 87, 44227 Dortmund
robert.jendrny at math.tu-dortmund.de

-------------- next part --------------
./example02 5
This is a sequential program.
Unit-Square: 1, bench1: 2 .... choice: 1
=================== Q2 - Elements ===================
LS_AMG_SOR: 1, BICGSTAB_AMG_SSOR: 2, BICGSTAB_SSOR: 3, BICGSTAB_JAC: 4, CG_JAC: 5, LS_AMG_SSOR: 6, BiCGSTAB_AMG_SOR: 7, CG_AMG_SSOR: 8, CG_SSOR: 9 .... choice: 8
Level input: 6
Used levels:7
=== solving (reduction: 1e-06) Level 0 has 4225 unknowns, 4225 unknowns per proc (procs=1)
aggregating finished.
Level 1 has 973 unknowns, 973 unknowns per proc (procs=1)
aggregating finished.
Level 2 has 231 unknowns, 231 unknowns per proc (procs=1)
aggregating finished.
Level 3 has 57 unknowns, 57 unknowns per proc (procs=1)
aggregating finished.
Level 4 has 13 unknowns, 13 unknowns per proc (procs=1)
aggregating finished.
Level 5 has 3 unknowns, 3 unknowns per proc (procs=1)
aggregating finished.
Level 6 has 1 unknowns, 1 unknowns per proc (procs=1)
operator complexity: 1.1729
Building hierarchy of 7 levels (inclusive coarse solver) took 0.401939 seconds.

===================Statistics for CG_AMG_SSOR===================
4225
1024
0.493498
21


./example02 6
This is a sequential program.
Unit-Square: 1, bench1: 2 .... choice: 1
=================== Q2 - Elements ===================
LS_AMG_SOR: 1, BICGSTAB_AMG_SSOR: 2, BICGSTAB_SSOR: 3, BICGSTAB_JAC: 4, CG_JAC: 5, LS_AMG_SSOR: 6, BiCGSTAB_AMG_SOR: 7, CG_AMG_SSOR: 8, CG_SSOR: 9 .... choice: 8
Level input: 7
Used levels:8
=== solving (reduction: 1e-06) Level 0 has 16641 unknowns, 16641 unknowns per proc (procs=1)
aggregating finished.
Level 1 has 3990 unknowns, 3990 unknowns per proc (procs=1)
aggregating finished.
Level 2 has 944 unknowns, 944 unknowns per proc (procs=1)
aggregating finished.
Level 3 has 227 unknowns, 227 unknowns per proc (procs=1)
aggregating finished.
Level 4 has 55 unknowns, 55 unknowns per proc (procs=1)
aggregating finished.
Level 5 has 13 unknowns, 13 unknowns per proc (procs=1)
aggregating finished.
Level 6 has 3 unknowns, 3 unknowns per proc (procs=1)
aggregating finished.
Level 7 has 1 unknowns, 1 unknowns per proc (procs=1)
operator complexity: 1.17578
Building hierarchy of 8 levels (inclusive coarse solver) took 1.62975 seconds.

===================Statistics for CG_AMG_SSOR===================
16641
4096
0.688844
38


./example02 7
This is a sequential program.
Unit-Square: 1, bench1: 2 .... choice: 1
=================== Q2 - Elements ===================
LS_AMG_SOR: 1, BICGSTAB_AMG_SSOR: 2, BICGSTAB_SSOR: 3, BICGSTAB_JAC: 4, CG_JAC: 5, LS_AMG_SSOR: 6, BiCGSTAB_AMG_SOR: 7, CG_AMG_SSOR: 8, CG_SSOR: 9 .... choice: 8
Level input: 8
Used levels:9
=== solving (reduction: 1e-06) Level 0 has 66049 unknowns, 66049 unknowns per proc (procs=1)
aggregating finished.
Level 1 has 16166 unknowns, 16166 unknowns per proc (procs=1)
aggregating finished.
Level 2 has 3833 unknowns, 3833 unknowns per proc (procs=1)
aggregating finished.
Level 3 has 919 unknowns, 919 unknowns per proc (procs=1)
aggregating finished.
Level 4 has 219 unknowns, 219 unknowns per proc (procs=1)
aggregating finished.
Level 5 has 53 unknowns, 53 unknowns per proc (procs=1)
aggregating finished.
Level 6 has 13 unknowns, 13 unknowns per proc (procs=1)
aggregating finished.
Level 7 has 3 unknowns, 3 unknowns per proc (procs=1)
aggregating finished.
Level 8 has 1 unknowns, 1 unknowns per proc (procs=1)
operator complexity: 1.17622
Building hierarchy of 9 levels (inclusive coarse solver) took 6.606 seconds.

===================Statistics for CG_AMG_SSOR===================
66049
16384
0.822805
71


More information about the dune-pdelab mailing list