Front page › Forums › General › stopping criteria
This topic contains 3 replies, has 2 voices, and was last updated by Dimitar 2 years, 4 months ago.
I am analyzing the effect of the stopping criteria on GMRES
for that I differentiate between:
1) residual< epsilon
2) residual< epsilon *||b||
3) residual/r0< epsilon
In Paralution condition 1 is absolute and 3 is relative. I don’t find the setting for the second condition.
Is there a thought behind omitting condition 2?
PS: in petsc they have 1 and 2 missing the third.
(2) and (3) are the same for initial x = 0.
For x != 0, you can specify (3) which is the relative stopping criteria as you mention.
If you want to have (2) for x != 0, you need to modify a bit the IterationControl class.
thanks for your answer. I enforced 2) by taking 1) and multiplying by the norm of b.
In a steady CFD simulation I keep calling the GMRES for every flow iteration and I hope x0(i+1) is set to the last x(i) or is it always x0=0?
In all iterative solvers you need to pass A, b, x. The x you pass is the initial guess (your x0). We do not modify the initial vector nor set it to zero.
You must be logged in to reply to this topic.
PARALUTION 1.1.0 - 25th of Jan 2016