Alex Granovsky
gran@classic.chem.msu.su
you are welcome.
Indeed the linear dependence was not very strong. I think the most
important change I introduced into your input files is:
$contrl inttyp=hondo
The ALTTRF integral transformation code which is used for
(semi)direct Method=1 MP2 and direct (X)MCQDPT, had been designed
back in 1997, before the fastints/gencon code was developed.
Due to this reason, ALTTRF uses old Pople-Hehre and Hondo integral
packages. While Hondo code is very precise, somewhat faster Pople-
Hehre code is not very precise, especially for diffuse functions.
Any run that uses diffuse functions should be run with Pople-Hehre
code completely disabled i.e. with inttyp=hondo
Best regards,
Alex
On Thu Oct 27 '16 3:16pm, Ole_ wrote
------------------------------------
>Dear Alex,
>thank you so much for the great support you are giving on this forum!
>I noticed the quasilinear dependence, but it seemed small to me, so I
>did not expect such severe consequences to the XMCQDPT2 step.
>Thanks again and best regards,
>Ole
>
>
>On Wed Oct 26 '16 4:45pm, Alex Granovsky wrote
>----------------------------------------------
>>Dear Ole,
>>Unphysical results you got are caused by a quasilinear dependence
>>in the atomic basis set you are using, as indicated in the output
>>files. Perturbation theory calculations are typically very sensitive
>>to such dependences in ill-defined basis sets. To get correct results,
>>you need to increase the overall precision of all stages involved
>>into xmcqdpt2 computations. Attached is rar archive with corrected
>>input files and corresponding outputs from both Firefly v. 8.1.1 and
>>Firefly v. 8.2.0.
>>Hope this helps.
>>Kind regards,
>>Alex Granovsky