QuantumATK Forum
QuantumATK => General Questions and Answers => Topic started by: bjnagare on February 7, 2010, 15:39
-
Hello ATK users,
My system with 290 carbon atoms is converging well. But when I am using initial_scaled_spin=1 on the all the carbon atoms, I am getting an error;
ATKError: inverse(DZMatrix const &) : Could not LU factorize!
How one can solve this problem?
It may be noted that as the system is converging well without any spin, so there is no question of atomic overlap.
Please help me to understand this problem?
Thanking you
Regards,
Nagare
-
What constraint are you using?
If you are using the density matrix constraint, you will get this error if you lose all the charge from the central region.
-
Dear Sir,
Thank you for reply.
I am not using DensityMatrix constraint. The charges in the central region is not zero.
For your clarification, here is part of input file and output file is shown IN THE ATTACHMENT.
Please guide to proceed further.
Regards
Nagare
-
Does the error come right when the two-probe run starts or is after some time?
Try alternative setting the initial spin to 0.9999 instead of 1.000. Then the GGA potential will be have much better.
A wild short is that you will not see this in LDA.
-
Dear Sir,
The error comes when the two probe calculations starts. I have tried with spin=0.9999 instead of 1, even though I am getting same error.
Thank you
Regards,
Nagare
-
Do you see it in LDA?
-
Dear Sir,
Using LDA.PZ with dzp, I am getting same error but after 134 scf iterations of two probe calculation. Pl. help to proceed.
Regards
Nagare
-
Since I dont have the geometry, I can not check it myself.
However this is the way I would approach it.
Step one step back :)
- Perform a calculation on the electrode as simple bulk system setting the initial scaled spin to 1.0
- From the converged bulk calculation, find out what final scaled spin is
- To make sure you got the right value for this, try starting the bulk calculation again with this value, and see that you obtain the
same result as before. - Go back to your twoprobe calculation. Instead of using an initial scaled spin of 1.0 ( or 0.9999 ) as I suggest using the value you got
from our little pre-studie.
If this does not solve the problem we might have to get medieval on it 8)
Joke aside, judging from what you write, I am sure that this is a matter of convergence. Spin systems are simply much harder, but once
you get the right side of them, they start to behave better. Once I struggled with Fe-MgO-Fe junctions, but now I have gotten the hold of them ;)
-
Thank you sir for your kind info.
I will follow ur suggestion and try to check it as per procedure. I will reply soon.
Thanks once again,
regards,
Nagare