[molpro-user] Error in atomic MRCI calculation
Peterson, Kirk
kipeters at wsu.edu
Thu Oct 17 18:21:37 BST 2013
Hi Jacek,
yes, your energies agree with mine but definitely not Lorenzo's.
best,
-Kirk
On Oct 17, 2013, at 10:09 AM, Jacek Antoni Klos <jklos at umd.edu> wrote:
> Hi Kirk,
>
> For the first simple Ar++ input on a single CPU I get:
>
> RESULTS FOR STATE 1.1
> =====================
>
> Coefficient of reference function: C(0) = 0.97976698 (fixed) 0.97976698 (relaxed) 0.97976698 (rotated)
>
> Energy contributions of configuration classes
>
> CLASS SQ.NORM ECORR1 ECORR2
> +++++++++++++++++++++++++++++++++++++++++++++++++++
> Internals 0.00000000 0.00000000 0.00000000
> Singles 0.02017686 -0.04486530 -0.04523828
> Pairs 0.02155131 -0.06715187 -0.06677889
> Total 1.04172817 -0.11201717 -0.11201717
> +++++++++++++++++++++++++++++++++++++++++++++++++++
>
>
> Reference energy -525.28328607
> Nuclear energy 0.00000000
> Kinetic energy 525.13810372
> One electron energy -709.40672453
> Two electron energy 184.01142129
> Virial quotient -1.00048977
> Correlation energy -0.11201717
> !MRCI STATE 1.1 Energy -525.395303234373
>
> !MRCI STATE 1.1 Dipole moment 0.00000000 0.00000000 0.00000000
> Dipole moment /Debye 0.00000000 0.00000000 0.00000000
>
> Cluster corrected energies -525.39997751 (Davidson, fixed reference)
> Cluster corrected energies -525.39997751 (Davidson, relaxed reference)
> Cluster corrected energies -525.39997751 (Davidson, rotated reference)
>
> Cluster corrected energies -525.39763935 (Pople, fixed reference)
> Cluster corrected energies -525.39763935 (Pople, relaxed reference)
> Cluster corrected energies -525.39763935 (Pople, rotated reference)
>
>
> For the run with 4 CPUS the energies are:
>
> RESULTS FOR STATE 1.1
> =====================
>
> Maximum overlap with reference state 2
>
> Coefficient of reference function: C(0) = 0.97976697 (fixed) 0.97976697 (relaxed) 0.97976697 (rotated)
>
> Energy contributions of configuration classes
>
> CLASS SQ.NORM ECORR1 ECORR2
> +++++++++++++++++++++++++++++++++++++++++++++++++++
> Internals Infinity NaN NaN
> Singles Infinity -Infinity NaN
> Pairs Infinity Infinity NaN
> Total Infinity NaN NaN
> +++++++++++++++++++++++++++++++++++++++++++++++++++
>
>
> Reference energy -525.28328607
> Nuclear energy 0.00000000
> Kinetic energy 525.13810092
> One electron energy -709.40672185
> Two electron energy 184.01141862
> Virial quotient -1.00048978
> Correlation energy -0.11201717
> !MRCI STATE 1.1 Energy -525.395303234293
>
> !MRCI STATE 1.1 Dipole moment 0.00000000 0.00000000 0.00000000
> Dipole moment /Debye 0.00000000 0.00000000 0.00000000
>
> Cluster corrected energies -Infinity (Davidson, fixed reference)
> Cluster corrected energies -525.39997751 (Davidson, relaxed reference)
> Cluster corrected energies -525.39997751 (Davidson, rotated reference)
>
> Cluster corrected energies -525.28328607 (Pople, fixed reference)
> Cluster corrected energies -525.39763935 (Pople, relaxed reference)
> Cluster corrected energies -525.39763935 (Pople, rotated reference)
>
>
> Looks like fixed reference energies are destroyed in the parallel run. I also get energies that are different then those reported in Lorenzo's post. How
> they compare to your run?
>
> Jacek
>
>
> On Oct 17, 2013, at 12:55 PM, "Peterson, Kirk" <kipeters at wsu.edu>
> wrote:
>
>> Jacek,
>>
>> what energies did you get for the simple Ar++ case? I got very different ones than Lorenzo (by more than 1 Eh). Also, as in your case no infinities in serial but lots of infinities with multiple cores (on my Mac). This is with a pretty recent build of the developer's version.
>>
>> best,
>>
>> -Kirk
>>
>> On Oct 17, 2013, at 8:53 AM, Jacek Antoni Klos <jklos at umd.edu> wrote:
>>
>>> Hi Lorenzo,
>>>
>>> Yes, with this larger input the NaNs and infinities appear for single and parallel execution, both on Mac and PC-Linux platforms with 2012.1/intel/mkl.
>>> Let me note, that these infinities appear only for the part of your input where you do 9-state MRCI. I think
>>> you need to extend active space in this case, try to put OCC,11 and the infinities will disappear.
>>>
>>>
>>> Best
>>> Jacek
>>>
>>>
>>>
>>> On Oct 17, 2013, at 11:29 AM, Lorenzo Lodi <l.lodi at ucl.ac.uk> wrote:
>>>
>>>> On 17/10/13 16:05, Jacek Antoni Klos wrote:
>>>>> Hi Lorenzo,
>>>>>
>>>>> I ran your input and I don't observe any infinities if I use single CPU.
>>>>>
>>>>> However when I ran this input with 2 or more CPUS in parallel, there are infinities in the output as you
>>>>> observe. I used 2012.1.
>>>>
>>>> Thank you for your feedback, Jacek! I ran my input again with versions
>>>> 2012 and 2010 compiled with gcc or intel compiler and I have infinities
>>>> in all cases (actually, the molpro2012/intel combination gives me a
>>>> segmentation fault error at CASSCF but this is not related with this
>>>> specific input, anyway).
>>>> I attached a larger input which also has these infinities and NaNs (runs
>>>> in ~1 second).
>>>>
>>>> Lorenzo
>>>>
>>>>
>>>>>
>>>>> On Oct 17, 2013, at 9:34 AM, Lorenzo Lodi <l.lodi at ucl.ac.uk>
>>>>> wrote:
>>>>>
>>>>>> Consider the following minimalist calculation for the 3P ground term of
>>>>>> Ar++:
>>>>>>
>>>>>> *** Calculation for Ar++
>>>>>>
>>>>>> basis={default, cc-pVDZ}
>>>>>> NOSYM
>>>>>> geometry={Ar}
>>>>>>
>>>>>> {CASSCF; OCC, 9; CLOSED,5; FROZEN,0; WF, elec=16, sym=1,spin=2;state,3;}
>>>>>> {MRCI ; OCC, 9; CLOSED,5; CORE,5; WF, elec=16, sym=1,spin=2;state,3; }
>>>>>>
>>>>>> CASSCF runs without problems and gives E=-526.817022986694.
>>>>>> MRCI also runs and gives a MRCI energy E=-526.92881183. However, in many
>>>>>> places I get `infinities', e.g.:
>>>>>>
>>>>>>
>>>>>>
>>>>>> Energy contributions of singly external configurations for state 1
>>>>>> ==================================================================
>>>>>>
>>>>>> NUMBER NORM ECORR1 OCCUPATION
>>>>>>
>>>>>> 1 0.00000000 -Infinity
>>>>>> 1022
>>>>>> 2 0.00000000 +Infinity
>>>>>> 0122
>>>>>> 5 0.00236458 -Infinity
>>>>>> 1220
>>>>>>
>>>>>>
>>>>>> Energy contributions of configuration classes
>>>>>>
>>>>>> CLASS SQ.NORM ECORR1 ECORR2
>>>>>> +++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>> Internals +Infinity NaN NaN
>>>>>> Singles +Infinity -Infinity NaN
>>>>>> Pairs +Infinity +Infinity NaN
>>>>>> Total +Infinity NaN NaN
>>>>>> +++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>
>>>>>>
>>>>>> Reference energy -526.81702299
>>>>>> Nuclear energy 0.00000000
>>>>>> Kinetic energy 525.51796052
>>>>>> One electron energy -711.13604804
>>>>>> Two electron energy 184.20723620
>>>>>> Virial quotient -1.00268469
>>>>>> Correlation energy -0.11178885
>>>>>> !MRCI STATE 1.1 Energy -526.928811837479
>>>>>>
>>>>>> !MRCI STATE 1.1 Dipole moment 0.00000000 0.00000000
>>>>>> 0.00000000
>>>>>> Dipole moment /Debye 0.00000000 0.00000000
>>>>>> 0.00000000
>>>>>>
>>>>>> Cluster corrected energies -Infinity (Davidson, fixed
>>>>>> reference)
>>>>>> Cluster corrected energies -526.93344888 (Davidson, relaxed
>>>>>> reference)
>>>>>> Cluster corrected energies -526.93344888 (Davidson, rotated
>>>>>> reference)
>>>>>>
>>>>>>
>>>>>>
>>>>>> If I do the MRCI calculation for one state only I obtain slightly larger
>>>>>> energies (E(MRCI)=-526.928597650037) but no `infinities' in the output.
>>>>>> What causes this problem? Is it because I'm running without point-group
>>>>>> symmetry?
>>>>>>
>>>>>> Any input would be appreciated :)
>>>>>>
>>>>>> Regards,
>>>>>> Lorenzo
>>>>>>
>>>>>> _______________________________________________
>>>>>> Molpro-user mailing list
>>>>>> Molpro-user at molpro.net
>>>>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>>>>
>>>>
>>>> <argon_plus_plus.inp>
>>>
>>> _______________________________________________
>>> Molpro-user mailing list
>>> Molpro-user at molpro.net
>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>
>
More information about the Molpro-user
mailing list