Author 
Topic 

chgad
11 Posts 
Posted  Mar 01 2019 : 11:45:10

Hello everyone,
I met the case that I have a working structure which converges and does not produce any warnings with it's current discretization. Now I wanted to refine the result by further discretization.
What happens if I further discretize is that suddenly Warnings about "Non negative Offdiagonals" or "Negative diagonal elements" arise.
In my understanding this shall not happen when I discretize further; overall the contrary should be the case (as stated in the manual: "poorly discretized conductors") which is improved by further discretization.
For a first result i can get along with the value i achieved so far but for the long term i'd like to understand what is going on there.
So, what is happening here ?
Any Help is appreciated.
Thanks in advance. 

Enrico
412 Posts 
Posted  Mar 04 2019 : 13:28:26

Can you share the input file?
There may be different reasons for this behavior. One common possible reason is in case of small offdiagonal elements with respect to diagonal ones (e.g. differences of more than threefour orders of magnitude). In this case the diagonal element may be too small to be reliable, and may oscillate between positive and negative. (This is caused by the nonexact arithmetic, and the fact that the iterative GMRES algorithm does not minimize the error of EACH element, but the residual norm of the charge vector.)
Negative diagonals should not happen, unless maybe you have a very small structure with respect to the other conductors around, again triggering numerical problems.
But, without the input file, it is difficult to say which is the ultimate issue.
Best Regards, Enrico




Topic 


