Mumps solver

camphilli · Post by **camphilli** » Tue Apr 22, 2008 1:02 pm

I am trying to run a code developed in OpenSees in OpenSeesSP, as you told us in the workshop we need to change the system (in my case system SparseSPD) to Mumps. Reading the manual it says that due to copyright restrictions we have to install the system solver, I look in internet and the webpage which provide the solver http://www.enseeiht.fr/lima/apo/MUMPS/ does not exist anymore.

Do you know where I can find the solver?

Thanks for your help

Camilo Phillips

rjaeger · Post by **rjaeger** » Tue Apr 22, 2008 5:41 pm

Hi,

Looking at the source code here: http://opensees.berkeley.edu/cgi-bin/cv ... SOE/mumps/ it looks like MUMPS is included in OpenSees already. Have you tried using MUMPS yet? It looks like it should work.

camphilli · Post by **camphilli** » Tue Apr 22, 2008 7:47 pm

I tried and it is not working (Error -9) it seems something is missing, thanks for the reply

Post by **fmk** » Wed Apr 23, 2008 5:48 am

try using
system Mumps -ICNTL14 50

the manual states or should state that you need the MUMPS software for building the executables. It is included in the binary releases.

camphilli · Post by **camphilli** » Wed Apr 23, 2008 8:51 am

Thanks Frank for your answer

I tried to use: system Mumps -ICNTL14 50 and I obtained the following results:

WARNING MumpsParallelSolver::solve (void)- Error -9 returned in substitution dumps()
WARNING ModifiedNewton::solveCurrentStep()-the LinearSysOfEqn failed in solve()
DirectIntegrationAnalysis::analyse() – the Algorithm failed at time 5000
OpenSees > analyze failed, returned: -3 error flag

Then the program starts to run the seismic motion until there is no convergence and then stops. With the other system (SPD) the program was able to continue running changing the time step when the result was non-convergence.

There is something else I need to change in the code to run the Mumps solver?

Thanks

Camilo Phillips

Post by **fmk** » Wed Apr 23, 2008 9:56 am

try increasing the -ICNTL14 to 100 .. it will work
with the other solver you specified, the solving is just not done in parallel, all the data is sent to P0 .. the only thing being done in parallel is element state determination.

ps. if increasing ICNTL14 fails can you send me the file fmckenna AT berkeley DOT edu