# How do we find the starting_ns_eigenvalue for QE calculations?

When performing DFT+U calculation, the ground state might get stuck in a local minimum. In such cases we need to provide starting_ns_eigenvalue to help the calculation reach the desired ground state. How do we find these starting eigenvalues? Can anybody explain this matter?

In this enter link description here reference video, after the first scf iteration, it shows the Hubbard occupation as

Tr[ns(  3)] (up, down, total) =   5.00000  1.00000  6.00000
Atomic magnetic moment for atom   3 =   4.00000
SPIN  1
eigenvalues:
1.000  1.000  1.000  1.000  1.000
.......
SPIN  2
eigenvalues:
0.200  0.200  0.200  0.200  0.200
eigenvectors (columns):


and they change only the 5th eigenvalue of that atom to 1 in spin 2.

but in my case, one atomic type (Fe) shows

 Tr[ns(  1)] (up, down, total) =   5.00000  1.00000  6.00000
Atomic magnetic moment for atom   1 =   4.00000
SPIN  1
eigenvalues:
1.000  1.000  1.000  1.000  1.000
SPIN  2
eigenvalues:
0.200  0.200  0.200  0.200  0.200


and for other atomic type (Ti)

 Atomic magnetic moment for atom   3 =  -0.02149
SPIN  1
eigenvalues:
0.269  0.269  0.309  0.314  0.314
......
SPIN  2
eigenvalues:
0.285  0.285  0.296  0.315  0.315


I need to know, which eigenvalue needs to change here

• Did you take a look at this video? Here Iurii Timrov explained why we need starting_ns_eigenvalue and how to choose them. The corresponding slide can be found here - see exercise 2 and 3 (page 22-40) Commented Feb 9 at 11:55
• @AbdulMuhaymin, I'm following that video, but I want to know how they select that eigenvalue is it from a pseudopotential file? Commented Feb 9 at 13:06
• What do you mean by 'select the eigenvalue'? Did you get how the occupation matrix has formed? such as why the matrix has that dimension, why we have two such matrices, etc. If you could state what you understood and what you want to know in a narrower scope, maybe I can write an answer. Basically, this is occupation matrix controlling. This paper might be an interesting read, especially the intro and method section to understand what you achieve using starting_ns_eigenvalue. Commented Feb 9 at 13:39
• @Prasad if you want to know more about occupation matrices, please post a new question. Commented Feb 10 at 4:41
• @AbdulMuhaymin thanks for the given reference. I have updated my question with relevant details can you please answer that? Commented Feb 10 at 5:40