Molcas Forum

eklahn

Member

Registered: 2020-12-07

Posts: 4

Convergence problems in RASSCF module

Dear all,

I'm new to MOLCAS and trying to run a RASSCF calculation with the following input in OpenMolcas. My goal is to calculate a number of triplet states for a complex with Ho(III) in it to use together with quintet and singlet states for RASSI/SINGLE_ANISO further down the line. The quintet and singlet calculations are working fine.

----------------------

> export MOLCAS_MOLDEN=ON
&RASSCF
Spin
3
Inactive
329
Nactel
10 0 0
Ras2
7
CiRoot
117 117 1
Alter
7
1 309 330 
1 310 331
1 311 332 
1 312 333
1 313 334 
1 314 335
1 315 336
ITERations
200 100
Levshft
0.5
&grid_it
FileOrb=$CurrDir/$Project.RasOrb
ALL
End of Input

----------------------

I get the following error under "Wave function control section" in the log-file, and then the module is stopped with an "/rc=_RC_INTERNAL_ERROR_".

----------------------

                                        RASSCF iterations: Energy and convergence statistics
                                         ----------------------------------------------------
 
      Iter CI   SX   CI       RASSCF       Energy    max ROT     max BLB   max BLB  Level Ln srch  Step   QN   Walltime
          iter iter root      energy       change     param      element    value   shift minimum  type update hh:mm:ss
        No convergence in the CI section: MAXJT will be increased to 
                    24
 !!! WARNING !!!
 NANs encountered
 
  The numbers in the array will now be checked.
  There are                     28  elements.
  Element nr.                     1  is                      NaN
 ...

----------------------

The program then proceeds to print all 28 elements as "NaN".
Does anyone have a good explanation for this error, and how I can fix i? I see something about "convergence in the CI section", and I am wondering whether this has to do with the value of "Levshft" in the input? In the documentation, it says to increase this value if the calculation diverges.

Thanks for your time in responding to this question!
Best regards,
Emil

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: Convergence problems in RASSCF module

Dear Emil,

I've had something similar once, with the damaged integral files.

Make sure they are fine (e.g. scf must run smoothly). Try a different disk or a different machine.

In my case, it was a memory problem, but a disk problem can do this too (I guess).
If it gets down to memory testing, this is the tool: https://www.techpowerup.com/memtest64/
In linux, memtester command can also help.

Best regards.
Andrew

eklahn

Member

Registered: 2020-12-07

Posts: 4

Re: Convergence problems in RASSCF module

Dear Andrew,

thanks for your comment. I hope I can draw a little further on your expertise.
Do you remember which files specifically that were affected in your case?

I'm also not sure about the comment "scf must run smoothly", maybe you could elaborate that for me? The problem occurs as I try to run RASSCF, which to my understanding is the exact program that is trying to perform the SCF-procedure, is it not?

Best,
Emil

eklahn

Member

Registered: 2020-12-07

Posts: 4

Re: Convergence problems in RASSCF module

Dear all,

this problem resurfaced for me, so I'm still trying to solve what may have happened here. Any new suggestions or elaborations on the first answer would be very welcome.

Best,
Emil

nikolay

Member

From: Stuttgart

Registered: 2016-03-21

Posts: 54

Re: Convergence problems in RASSCF module

I have the same problem. For one particular case (also different spin with large number of roots), it was solved by running on different machine with slightly different configuration. But for other case the calculation fails on both machines.
Unfortunately, the second calc is quite large.

We should generate a minimalistic input file that shows the error to debug it.

eklahn

Member

Registered: 2020-12-07

Posts: 4

Re: Convergence problems in RASSCF module

Dear all,

thanks for the sympathy. I discovered that I might have been an error on my part.

I was trying to make a RASSCF-calculation on triplet states in Ho(III). If I do the calculation (and observe what others before me have done), I see that there should be 210 triplet states of Ho(III). I was trying to make the calculation with 117 triplet states.

I can't be certain that this is what caused my problem, but now with 210 triplet states the calculation finished without a problem.

Best regards,
Emil