[Molpro-user] how to avoid SCF restart in a CCSD(T) scan and why does the SCF converge to a wrong state?

[Kirk Peterson]

Dear Fedor,

You are correct in that as Gerald reports, the start,atdens works  
fine, but unfortunately even with 2008.2 taking the guess
from the previous geometry's orbitals does give reliable results.

best regards,

Kirk

On Sep 15, 2008, at 8:21 AM, Fedor Goumans wrote:

> Dear Kirk,
> I think Gerald meant that in 2008 it works fine with a rohf guess from
> atomic densities. Since I do not have access to 2008 I cannot test
> whether the scf-restart works ok there or not for a scan with a
> triplet rohf wavefunction.
> But since the rohf is not the most expensive part - a scf guess from
> atomic densities works fine now for me.
> Thanks again,
> Best,
> Fedor
>
>
> 2008/9/15 Kirk Peterson <kipeters at wsu.edu>:
>> Dear Fedor,
>>
>> glad to hear that things worked out.  I believe the SCF guess  
>> routines have
>> been improved in the newest
>> version, 2008.1.  I think that Gerald mentioned that you input  
>> worked under
>> these conditions?
>>
>> About memory, asking for 4GB on a 4GB node won't hurt too much  
>> unless that
>> then leaves much less
>> for the system to buffer I/O with.
>>
>> best regards,
>>
>> Kirk
>>
>> On Sep 15, 2008, at 1:28 AM, Fedor Goumans wrote:
>>
>>> Dear Kirk,
>>> Thank you for that. That worked - although still not sure why the  
>>> rohf
>>> guess does not seem to work properly in such a scan mode.
>>> Oner further question: I have noticed that I do not need as much
>>> memory, but is there any problem in requesting 450 MW on a node that
>>> has 4GB available?
>>> with kind regards,
>>> Fedor
>>>
>>> 2008/9/12 Kirk Peterson <kipeters at wsu.edu>:
>>>>
>>>> Dear Fedor,
>>>>
>>>> which version of Molpro are you using?
>>>>
>>>> In any event the following suggestions may help:
>>>>
>>>> 1) not sure what the "restart,0" line is supposed to do. Restart  
>>>> cards
>>>> must
>>>> always come directly after any memory cards. You don't need it  
>>>> anyway
>>>> 2) your "wf" command should come in as part of the "rhf" command:
>>>>  rhf;wf,24,1,2    (all enclosed by braces if this is 2006.1 and  
>>>> above)
>>>> 3) to make the rhf use the bare hamiltonian guess every time,  
>>>> change this
>>>> to
>>>> rhf;wf,24,1,2;start,h0
>>>>
>>>> -Kirk
>>>>
>>>> PS - BTW, 450 mwords of memory (about 4GB) for this job is  
>>>> probably at
>>>> least
>>>> 350 mwords more than you need.
>>>>
>>>> On Sep 12, 2008, at 5:32 AM, Fedor Goumans wrote:
>>>>
>>>>> Dear all,
>>>>> Just getting started with molpro I wanted to use to scan PESs.
>>>>> It turns out that for this particular coordinate (see input  
>>>>> below),
>>>>> CCSD(T) always crashes in the 2nd step, with this message:
>>>>> UNREASONABLE NORM. CALCULATION STOPPED
>>>>> In the CCSD bit of the programme.
>>>>>
>>>>> It looks like in the second step the RHF reference wavefunction
>>>>> converges to a wrong state (e.g. for the input below -187.2009 H  
>>>>> vs.
>>>>> -187.5115 H for a 'fresh start' on the same geometry) giving  
>>>>> rise to
>>>>> this problem, because when I scan the coordinate by hand, the  
>>>>> RHF is
>>>>> not restarted and the CCSD converges just fine (also the RHF  
>>>>> energy is
>>>>> lower). I am surprised that this happens even at these small step
>>>>> sizes of 0.05A. I've tried to turn off restart as in the example
>>>>> below, but the rhf is restarted every time in the second step even
>>>>> though restart,0 is in the do-loop.
>>>>>
>>>>> Any ideas about why the SCF converges to a wrong state?
>>>>> Any ideas how to switch of RHF SCF restart in a PES scan?
>>>>>
>>>>> Thanks in advance for your help.
>>>>>
>>>>> Fedor Goumans
>>>>>
>>>>> Example input for scan:
>>>>>
>>>>> ***,O+CO long range complex
>>>>> memory,450,m
>>>>> DR=0.05
>>>>> do i=1,40
>>>>> restart,0
>>>>> r2=2.0+i*dr
>>>>> r1=1.13599229
>>>>> a1=120.0
>>>>> geometry={Y
>>>>> angstrom;
>>>>> O1;
>>>>> C,O1,r1;
>>>>> O2,C,r2,O1,a1;
>>>>> }
>>>>> basis=cc-pVDZ
>>>>> wf,22,1,2 ! A' surface
>>>>> rhf
>>>>> rhf-uccsd(t)
>>>>> enddo
>>>>> ---
>>>>> _______________________________________________
>>>>> Molpro-user mailing list
>>>>> Molpro-user at molpro.net
>>>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>>>
>>>>
>>
>>