[petsc-users] Slepc doesn't converge for some EV problems

[Jose E. Roman]

El 05/06/2013, a las 07:52, Sonya Blade escribió:

> Dear All,
> 
> Sometimes I experience that Slepc is not converging for the requested 
> eigenvalues and simply resturns 0. I do take into consideration of setting
> the requested eigenvalues from command line options rather then setting them 
> manually at the body of program. 
> 
> I observe this when I'm trying to solve this moderately large matrices e.g. 192x192 
> for small ones e.g. 48x48 there is not problem other than Slepc deviates from 
> requested to found EVs. For example if I request 7 eigenvalues to be found Slepc finds 
> 9 or 10, what could be the problem of this malfunctioning?
> 
> Your help will be appreciated,
> 
> Regards, 		 	   		  

192 is very small. SLEPc is intended for large sizes (10,000 or more). For such small matrices, you would rather use LAPACK.

It is normal that the solver returns more eigenvalues than requested. This is not a malfunctioning. 

If you get zero converged eigenvalues, try changing the parameters (e.g. ncv) or using a different strategy (e.g. shift-and-invert with a reasonable target).

Jose