[molpro-user] Extrapolation to the complete basis set limit of def2-anzvp basis sets.

Sun Oct 27 00:04:02 BST 2013

Dear Sasha,

the contraction coefficients of the basis set have to be explicitly optimized for the underlying ECP, so one can’t just “mix and match”.

best regards,

-Kirk

On Oct 26, 2013, at 3:19 PM, sasha medvedev <vinsanity305 at mail.ru<mailto:vinsanity305 at mail.ru>> wrote:

Dear Kirk Peterson.
Is it possible to combine MDF ecps and def2 basis sets?
Of course, we can write such a program in molpro, but is it acceptable?
Or mixing of ecps and basis sets is a rather controversial idea, which we should throw away immediately?
                                        Sasha Medvedev.

Пятница, 25 октября 2013, 21:02 UTC от "Peterson, Kirk" <kipeters at wsu.edu<mailto:kipeters at wsu.edu>>:

I can’t say that the dhf- versions are any more reliable, but certainly the new  ECP’s on which they are based are more accurate than the older MWB ones that the def2 sets are matched with.

best,

-Kirk

On Oct 25, 2013, at 1:51 PM, sasha medvedev <vinsanity305 at mail.ru<x-msg://179/sentmsg?mailto=mailto%3avinsanity305@mail.ru>> wrote:

Dear Kirk Peterson.
I decided that the cardinal number n (i.e. d,t,q,5,6) of a basis set should be equal to lmax. I thought, that lmax and n are identical.
Thank you for explanation of my misunderstanding, concerning the construction of correlation consistent basis sets.
----------------------
> yes, there was an update to this file by Gerald Knizia in August of 2012.  Presumably this is included in the latest production version of Molpro.  In this new file most of the previous def2- sets that used the older MWB ECPs have been modified to use the new MDF ones.  These are called dhf-

So, dhf basis sets are more reliable? (in comparison with def2 sets)
                                          Sasha Medvedev

Пятница, 25 октября 2013, 15:36 UTC от "Peterson, Kirk" <kipeters at wsu.edu<x-msg://179/sentmsg?mailto=mailto%3akipeters@wsu.edu>>:
Dear Sasha,

unfortunately what you describe below about increasing the lmax of light atoms so that they would match heavier ones would spoil the very thing that makes them systematically convergent towards the CBS limit, i.e., with each “n”, functions are added that contribute similar amounts of correlation energy. The higher angular momentum functions (f for DZ, g for TZ, etc.) are not included since they do not “belong” energetically at that level.   One could of course construct sets in this way, but it’s not clear what their convergence properties would be.  And then if you wanted to do molecules containing f-block elements, you have to increase them again….

best regards,

-Kirk

On Oct 24, 2013, at 2:24 PM, sasha medvedev <vinsanity305 at mail.ru> wrote:

Dear Mr. Grant Hill.
Thank you for a reply.

>The def2 series of basis sets was not designed to systematically approach the basis set limit...

These piece of information is important for us.
----------------------------------------------
>Think of a minimal basis for a first row element (C) and one for a transition metal (Ti),
lmax must be different simply to construct the occupied orbitals.

I get your idea about Carbon and Titanium minimal atomic basis sets. But I think, that it is possible
to make the correlation consistent basis sets in such a way, that all elements have the same lmax for a particular basis set.
For instance, one should augment the minimal Carbon basis set (lmax=p=1) with d-functions. As the result, Titanium and Carbon will
have the same value of lmax (for a particular basis set). This approach is fruitful until we take into consideration lanthanide and actinide elements
(in such case we need to go further and add f-functions to C and Ti basis sets).
----------------------------------------------
>One approach when using correlation consistent sets is to keep lmax as the largest for the system (usually the heaviest element) - meaning cardinal number + 1 in this case.

So, in case of AB molecule we should compare lmax for A and B and then choose the bigger one for extrapolation formula?
Hm, it sounds sensible. Thanks.

                           Alexander Medvedev

On, 23 Oct 2013, 16:32 +01:00 от Grant Hill <Grant.Hill at glasgow.ac.uk>:
Dear Sasha,

The def2 series of basis sets was not designed to systematically approach the basis set limit, so the standard extrapolation formulae are not really expected to work here (although I see no reason why Molpro should stop you trying - GIGO). There may be some more appropriate formulae in the literature, but I am not aware of them.

On 23 Oct 2013, at 13:17, sasha medvedev <vinsanity305 at mail.ru> wrote:

> Besides that, extrapolation formulae depend on the cardinal number n (which is the part of the basis set name - i.e. 2 for asvp, 3 for atzvp, 4 for aqzvp, and should be equal to the lmax). When we looked through output file, we noticed, that the same basis set has different lmax for different atoms:

This is entirely expected. Think of a minimal basis for a first row element (C) and one for a transition metal (Ti), lmax must be different simply to construct the occupied orbitals. One approach when using correlation consistent sets is to keep lmax as the largest for the system (usually the heaviest element) - meaning cardinal number + 1 in this case. A lack of high accuracy reference data means this isn't particularly well tested, but it _seems_ to work.

Best regards,

Grant

--
s m
_______________________________________________
Molpro-user mailing list
Molpro-user at molpro.net
http://www.molpro.net/mailman/listinfo/molpro-user

--
s m

--
s m

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20131026/6784a7c2/attachment.html>