Posted Mon, 19 Jun 2017 12:20:31 GMT by Nedi Noseda
When trying do run the simulation I get the following error:

        PCG has a matrix with 43th diagonal element <= 0.
It could maybe be an elevation problem but how can I retrieve this 43th diagonal?
Posted Mon, 19 Jun 2017 12:28:58 GMT by Björn Kaiser
The diagonal in the matrix must not be <= 0. There are several possible reasons for negative entries along the principal diagonal of the matrix.
1.  Strong contrasts in material properties of adjacent elements. Try to refine the mesh horizontally/vertically.
2.  Topological errors of the mesh as triggered inconsistent input data (e.g. overlapping elements/element edges, wrong numeration of the nodes).
Posted Mon, 19 Jun 2017 12:35:37 GMT by Nedi Noseda
I tried to correct the elevations in the 3D Layer configuration but it doesn't work.

I initially had 4 layers and then removed 3 of them, just having two slices with about 50 m vertical distance between them. In this configuration I'm 100% sure that I don't have any overlapping in elevations.

The problem remains.
My question is if there's a way to manually correct this problem in the 43th diagonal?
Posted Mon, 19 Jun 2017 14:18:12 GMT by Björn Kaiser
You cannot manually correct the entry. The only way is to change the settings as suggested above.
Posted Tue, 20 Jun 2017 11:06:09 GMT by Nedi Noseda
Just a short reply that could be useful to other users: the problem was in the data regionalization method (conductivity): the kriging created negative values of hydraulic conductivity; by setting "logarithmic true" it doesn't create any mode negative values.
Posted Wed, 21 Jun 2017 11:45:43 GMT by Björn Kaiser
Great, thanks.
Posted Tue, 12 Nov 2019 06:00:56 GMT by Razi Sadath P V Senior Research Fellow
What is the reason for SAMG: Row(s) with negative diagonal found! ERROR
How can i rectify that?
Posted Mon, 25 Nov 2019 11:27:35 GMT by Peter Schätzl
The most likely reason is an instability somewhere in the model. Check for oszillations in the process variables and/or large error values (by adding the error norm history to the User Data distributions).

You must be signed in to post in this forum.