DHF convergence question

[Peterson, Kirk]

Dear Dirac experts,

 

I've wondered about this for a while and thought I would post the question. When running an DHF calculation and starting from a previous, converged solution from a close-lying geometry, I often see a convergence pattern like the following:

 

It.    1    -30519.55570958      3.05D+04  0.00D+00  2.24D+01              12.18518700s   LL             Tue Jun 25

It.    2    -30515.00869200     -4.55D+00 -1.27D+00  6.61D+00              12.76632774s   LL             Tue Jun 25

It.    3    -30518.51450574      3.51D+00  4.24D-01  1.79D+00   DIIS   2   12.53840066s   LL             Tue Jun 25

It.    4    -30519.37964003      8.65D-01 -2.65D-01  7.91D-01   DIIS   3   12.42622744s   LL             Tue Jun 25

It.    5    -30519.52745323      1.48D-01  1.14D-01  3.98D-01   DIIS   4   12.26571989s   LL             Tue Jun 25

It.    6    -30519.56400470      3.66D-02 -7.92D-02  1.85D-01   DIIS   5   12.21246704s   LL             Tue Jun 25

It.    7    -30519.57541010      1.14D-02  1.11D-02  9.30D-02   DIIS   6   12.10690402s   LL             Tue Jun 25

 

The converged energy of the previous step was  -30519.5757689147249  in this example.  What always concerns/confuses me is the energy of the 2^nd iteration. It is always 5-8 Hartrees (!) higher than the initial energy.  The DHF then has to work to find its way back down again.  When I start from the default sum of fitted atomic potentials, the 1^st energy of course is terrible, but it smoothly marches its way down.  For example:

 

It.    1    -19196.64529149      0.00D+00  0.00D+00  0.00D+00               0.41734800s   Atom. scrpot   Tue Jun 25

It.    2    -30516.67552549      1.13D+04  9.08D+01  6.60D+01              13.11610529s   LL             Tue Jun 25

It.    3    -30519.02252275      2.35D+00  2.25D+00  2.43D+00   DIIS   2   12.21538672s   LL             Tue Jun 25

It.    4    -30519.25045885      2.28D-01 -1.10D+00  1.38D+00   DIIS   3   12.07547632s   LL             Tue Jun 25

It.    5    -30519.30929885      5.88D-02  3.70D-01  3.37D-01   DIIS   4   11.85025974s   LL             Tue Jun 25

 

Does anyone have a rational for the first behavior?  It seems that this could/can really spoil the advantage of using converged spinors from a nearby geometry.

 

regards,

 

-Kirk

 

[Peterson, Kirk]

Dear all,

 

of course nearly as soon as I sent this I realized and tested what was going on - I was applying some level shifts. Those of course initially lead to large increases in the energy….

 

sorry for the interruption.

 

regards,

 

-Kirk

[Trond Saue]

No problem !

Sent from my iPad

On 1 Aug 2024, at 21:04, 'Peterson, Kirk' via dirac-users <dirac...@googlegroups.com> wrote:



--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dirac-users/IA0PR01MB8308E1F74070688717B203F6D6B22%40IA0PR01MB8308.prod.exchangelabs.com.