NaN Error on new GASCIP KRCI on DIRAC25

[E. Eduardus]

Dear DIRAC Developer,

I was trying to run KRCI calculation for CH2BrI^+ on DIRAC using GASCIP. 

Before doing full active space calculation, I was trying small correlation space first. First, I tried to include 3 occupied orbitals and 50 virtual orbitals in the active space, then increasing the occupied orbitals to 4. Here is the result of the calculations (I did all these calculations using 6 processors):

As we can see, there is definitely an improvement in convergence time on DIRAC 25 on the 3/50 active space. However, it takes more iterations to converge. However, unfortunately, this goes to NaN on 4/50 active space on DIRAC25, while it converges on DIRAC24. One thing also to note, that the default threshold on DIRAC24 is changed from 1.00D-06 to 3.00D-03 on DIRAC25.

To help debug this problem, I attached these output files:

ci24_3_50_parallel6.out : CI calculation of active space 3/50 using DIRAC24

ci25_3_50_parallel6.out : CI calculation of active space 3/50 using DIRAC25

ci24_4_50_parallel6.out : CI calculation of active space 4/50 using DIRAC24

ci25_4_50_parallel6.out : CI calculation of active space 4/50 using DIRAC25

In summary, I am happy to see the improvement of the GASCIP module on DIRAC25 as it brings a lot faster calculation. However, there might be a bug causing this NaN result, more iteration cycles needed, and convergence criteria that could be resolved.

Thank you in advance for your help.

Kind regards,

Eduardus