Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Re^2: Linear dependence in the basis set, spherical basis sets, numerical instability of RHF and MP4(D)

Jiri Wiesner

Dear all,
I can only say that the modified inputs work well. The energies are nearly identical with the reference Gaussian computation. These are the previous numbers from the mis-configured calculation MP4(SDTQ)/6-311+G(2d,p) (all conformers differ only by rotation of a hydrogen atom on carboxylic group):
1.st_conformer(-180)   -722.3570145010
2.nd_conformer( -90)   -722.3375461923
3.rd_conformer(   0)   -722.3486605886
4.th_conformer(  90)   -730.0598741517
The 4th conformer was the one with the most notable difference, but all these values were actually wrong.

Gaussian computation:
1.st_conformer(-180)   -722.33563490
2.nd_conformer( -90)   -722.31558620
3.rd_conformer(   0)   -722.32690404
4.th_conformer(  90)   -722.31571941

Firefly computation with correct input:
1.st_conformer(-180)   -722.3356348335
2.nd_conformer( -90)   -722.3155861255
3.rd_conformer(   0)   -722.3269039741
4.th_conformer(  90)   -722.3157193300

I want to say a word of thanks to Alex Granovsky for his help.


On Sun Mar 14 '10 4:50pm, Jiri Wiesner wrote
>Dear Jiri,

>this is actually not the bug but simply the program misuse. In the outputs,
>you can find the warnings on quasi-linear dependence. There are also
>some suggestions printed on how to proceed. I've modified your input
>file to be suitable quasi-linear dependence cases, and performed MP4(SDQ)
>calculation just to show you that there are no any problems with code.
>The archive is attached.

>Note the following:

>1. The 6-311 basis set is spherical so you need to use $contrl d5=.t. $end
>2. It is very well known that MP4(D) contribution is the most sensitive part
>wrt. to orbitals quality and round off errors.  
>3. You need to use inttyp=hondo, nconv=6 or 7, and icut at least 10 or 11
>in the case of partial linear dependence.
>These things has been already discussed on the forum, perhaps in the somewhat
>different contexts, so it could be the good idea if you raise this question again.
>Also, there are some recommendation on the post-SCF calculations and integral
>code selection in the html documentation on the Firefly server.

>On Sun Mar 14 '10 4:47pm, Jiri Wiesner wrote
>>Dear Alex Granovsky,
>>I'm using Firefly for MP4 computation of an energetic torsional profile
>>of hydrogen in the carboxyl group of aminoacid glutamate. I found that
>>MP4(D) contribution in MP4 energy of the Glu clinal conformer is
>>unrealistically low making energy of its conformer several thousand of
>>kcal/mol lower than the energy of the trans confomer (I have a reference
>>computation in Gaussian for this case). Please see the attached output
>>files. I tried to computate the clinal conformer twice with different
>>parallel set-up and on different machines with Firefly version 7.1.G.
>>Please note also that the difference between the energies is cca 0.5
>>Jiri Wiesner

[ Previous ] [ Next ] [ Index ]           Sun Mar 14 '10 5:14pm
[ Reply ] [ Edit ] [ Delete ]           This message read 802 times