[molpro-user] Numerical Freq calculations using F12-CCSD(T)

[Werner Győrffy]

Dear John,

Basis sets: Based on my experience, large uncontracted basis set used as 
an RI basis set in F12 numerical frequency calculations may cause basis 
set linear dependency problems. The following basis sets might be 
alright for your example:

basis={
default,O=aug-cc-pwCVQZ,S=aug-cc-pwCVQZ,H=aug-cc-pVQZ
set,df
default,S=aug-cc-pwCVQZ/MP2FIT,O=aug-cc-pwCVQZ/MP2FIT,H=aug-cc-pVQZ/MP2FIT
set,jk
default,S=aug-cc-pV5Z/JKFIT,O=aug-cc-pV5Z/JKFIT,H=aug-cc-pVQZ/JKFIT
set,ri
default,S=aug-cc-pV5Z/JKFIT,O=aug-cc-pV5Z/JKFIT,H=aug-cc-pVQZ/JKFIT
}

Note that in the explicit specifications of basis functions, the default 
contraction scheme can be used by adding a "c" at the end of the line, 
for example,

s,S,aug-cc-pV5Z/JKFIT;c
p,S,aug-cc-pV5Z/JKFIT;c
...


F12 methods for properties: The CCSD(T)-F12b/3*C method is a good choice 
for calculation of molecular properties. You can use the * approximation 
without significant loss of accuracy.


Numerical frequency calculations: To calculate frequencies without 
symmetry, and without using symmetry-unique displacements, use

{freq,numerical,symm=no,coord=3n}

but the number of single point calculations required may increase 
significantly compared to a calculation using symmetry.

Regards,

Werner.

On 01/18/2017 01:13 AM, John Travers wrote:
> Dear Werner
>
> Thank you for the suggestions! I guess it is related to the choice of
> the basis. Since I used a aug-cc-pcwVQZ basis, there is no corresponding
> JK fit ones. I used an aug-cc-pV5Z JKFIT (see attached input). Do you
> have other suggestions for the choice of ri, df, and jk basis for the
> aug-cc-pcwVQZ orbital basis? BTW, the molpro version I used is 2012.1.
> Thanks again!
>
> Best wishes
>
> John
>
>
> On Tuesday, January 17, 2017 8:13 AM, Werner Győrffy
> <gyorffy at theochem.uni-stuttgart.de> wrote:
>
>
> Dear John,
>
> There might be some problem with the choice of the basis sets you use.
> It is recommended to use the optri basis sets, for example, for
> basis=avtz use ccsd(t)-f12b,ri_basis=avtz/optri. Could you please
> provide an input for your calculation failed?
>
> Regards,
>
> Werner.
>
> On 01/16/2017 05:41 PM, John Travers wrote:
>> Hi
>>
>> I was trying to do a numerical frequency calculation using f12-ccsd(t)
>> in c1 symmetry for a small molecule. The program always produces some
>> warning messages like : "Poor overlap with frozen basis set. May have
>> deleted the wrong vectors!
>>  ? The problem occurs in SvdConstructOrthoBasis"
>>
>> When checked the detailed output,  I also found some possible numerical
>> errors, something looks like :
>>
>>  Construction of CABS:
>>  Pseudo-inverse stability              NaN
>>  Smallest eigenvalue of S        -3.71E-13  (threshold= 1.00E-08)
>>  Ratio eigmin/eigmax              -3.71E-13  (threshold= 1.00E-09)
>>  Smallest eigenvalue of S kept    -3.71E-13  (threshold=-3.71E-13, 0
>> functions deleted, 658 kept)
>>  Maximum overlap with frozen deleted vectors: (nDel=  44)
>>    Vector:        1      2      3      4      5      6      7
>>    8      9      10      11      12      13      14      15      16
>>  17      18      51      25      67      44      36      35      43
>>  32      40      38      33      39      55      52      34      22
>>  37      21      42      50      20      41      74      19      53
>>  54      26      23    231      28      24      95      93      27
>> 158      75      61    123      68    108      31      63    163
>> 121    203    152      62      29    157      45      94      72
>> 97      77      65    204    103      30    115    126      66
>> 250      73    131    101    162    116      91    200    122
>>    Overlap:      NaN    NaN    NaN    NaN    NaN    NaN    NaN
>>  NaN    NaN    NaN    NaN    NaN    NaN    NaN    NaN    NaN
>>  NaN    NaN  1.436  1.248  1.102  1.000  0.999  0.997  0.996
>> 0.994  0.985  0.983  0.973  0.964  0.962  0.912  0.896  0.871
>> 0.858  0.853  0.827  0.825  0.742  0.596  0.594  0.537  0.523
>> 0.485  0.465  0.386  0.375  0.373  0.346  0.327  0.321  0.310
>> 0.307  0.268  0.259  0.257  0.249  0.247  0.216  0.204  0.194
>> 0.191  0.183  0.166  0.164  0.159  0.135  0.134  0.130  0.126
>> 0.118  0.114  0.110  0.100  0.095  0.093  0.092  0.092  0.091
>> 0.089  0.080  0.079  0.079  0.078  0.076  0.075  0.072  0.070
>>
>>
>> My questions are related to these "NaN" and warnings. Do these matter
>> and how to avoid this? Thanks!
>>
>> Best wishes
>>
>>
>> John
>
>>
>>
>>
>>
>>
>> _______________________________________________
>> Molpro-user mailing list
>> Molpro-user at molpro.net <mailto:Molpro-user at molpro.net>
>> http://www.molpro.net/mailman/listinfo/molpro-user
>>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net <mailto:Molpro-user at molpro.net>
> http://www.molpro.net/mailman/listinfo/molpro-user
>
>
>