2 atoms optimization failed
Just Got Here
10:04:45 AM PDT - Thu, Mar 24th 2016
Dear NWChem developers,

When I trying optimize interatomic distant for Al 2 atoms claster starting from different points I get 2 different extrema: staring from 1.5 A gives interatomic distant 2.563 A with minimal Tot DFT E= -484.755378 (DFT b3lyp/6-31G**) while starting from initial distant 5 A gives interatomic distant 3.062 A with Tot DFT E=-484.767366227. Changing the theory level does not help: the two extrema remain.

As I found out scanning the PES the problem arises because of the brake on the curve at the distant of approximately 2.91 A and two extrema at the left and at the right from the break. This type of PES takes place only if one would scan it by means of making single points energy calculations starting each time from initial guess. But if instead of it you will load vectors from the previous point of calculation the PES becomes smooth with only one extremum.

It looks like standard optimization algorithm at each step of optimization starts SCF from initial guess and does not use vectors from the previous step. How it can be changed?

Thank you.

Forum Regular
9:05:23 AM PDT - Fri, Mar 25th 2016
For an optimization, the code will read the MOs from the previous step as the initial guess for the current step (unless for some reason it cannot reuse the previous MOs, e.g. the number of linearly dependent MOs changed from one step to the next). There is currently no option that can be set to change this behavior.

Just Got Here
2:03:52 AM PDT - Sat, Mar 26th 2016
Quote:Sean Mar 25th 4:05 pm
For an optimization, the code will read the MOs from the previous step as the initial guess for the current step (unless for some reason it cannot reuse the previous MOs, e.g. the number of linearly dependent MOs changed from one step to the next). There is currently no option that can be set to change this behavior.


So, in this particular case it looks like it can't reuse previous MO's in some point on the curve for the reason of linear dependency and forced to start from "scratch" so to speak. How to check it?

One more question. Do two extrema on this curve of Al2-cluster have any physical meaning or it is only another numerical artifact? I prone to treat it as a result of crossing Pi and Sigma terms at one point. As a result one minimum belongs to Pi term and another one belongs to Sigma term. What do you think about it?

Forum Regular
10:01:11 AM PDT - Mon, Mar 28th 2016
If the code is using the previous orbitals as the starting point for the current SCF, the line

Loading old vectors from job with title :

will appear in the output just after the printing of the general information for the DFT calculation and just before the printing of the information during the iterations of the SCF. If the code cannot load the previous orbitals for whatever reason, you should see the line

Load of old vectors failed. Forcing atomic density guess

after the line about loading the old vectors.
I am no expert on Al2 so I cannot really answer your other question. If I were you I would first be checking if there is any instability in the SCF solution, e.g. is there a lower energy unrestricted solution (you would need to make sure that your initial guess breaks the spin symmetry, otherwise you will just converge back to the restricted solution)?


Forum >> NWChem's corner >> General Topics