Num Krylov vector

Altair Forum User

Hi 

I am Working on a Thermal Analysis problem.If i increase the Num Krylov vector from the default value.

 

Will there be change in results/emoticons/default_rolleyes.gif' title=':rolleyes:' />

or 

It will lead to faster convergence/emoticons/default_smile.png' srcset='/emoticons/smile@2x.png 2x' title=':)' width='20' />

 

What will be the optimum value of Krylov vector for Thermal Problems??

Altair Forum User

Hi Dinesh,

The krylov vector is used in the linear equation solver (GMRES), wherein you need to compute a base vector for the krylov subspace, in which the Euclidian norm of the residuum is minimized.

The problem is, that you need all base vectors from previous iterations to compute the base vector for the current iteration, which is an expensive operation for solving.

 

As the costs and memory increase linear with the number of iterations, it’s usual to drop the computed base and restart the iteration with the current approximation.

 

The number after which the iteration gets restarted is the number of krylov vectors you define in AcuSolve. And this number determines how much memory you need for your computation with AcuSolve.

 

Regarding convergence, a krylov vector which is too small can affect convergence negatively (in difficult cases), but at a certain level a higher krylov vector won’t improve the convergence anymore.

 

If the problem had difficulty converging, this will/may help.  If it is converging well, it does not help at all, and perhaps makes the solution a little more expensive.

 

As a general rule, use 10 and perhaps 20-40 for the Flow thermal coupled system (Natural convection).

 

Please refer the AcuSolve manual for more details as well.

 

Regards,

kamlesh