[molpro-user] Linear dependent basis

[Glen Jenness]

John,
I've encountered this issue before, and have done one of the 3 options below.

1)  The simplest way around this is to simply set gthresh,THROVL=1.d-9
in your input file.  However, this can lead to either slow or
non-convergent behavior, so be warned!

The next two I'm not to sure about for extrapolating to the CBS limit, but
are things you can do:

2)  Modify which atoms have your diffuse functions (i.e. your augmented
functions).

3)  Modify the diffuse function exponents themselves.  In my opinion, this
is by far the more riskier methods, but it will allow you to have your
full AVQZ basis set and should in principal allow for your extrapolation
(I make no guarantees however!).   Sometimes, multiplying these exponents
by 1.1 or 1.3 might get rid of your linear dependency, but if you go to
high you might as well not have diffuse functions.

Hope this helps!
Glen

> Hi all,
>
> I'm trying to run single reference ab initio calculations on
> transition metal complexes and build up enough calculations to
> extrapolate to the CBS limit.
>
> My molecule is the PdCl4 dianion, and I'm using the Stuttgart Pd
> pseudopotential and the (aug)-cc-pvNZ basis sets on all other valence
> electrons. All are taken from the EMSL basis set library. I get SCF
> convergence with cc-pvDZ, aug-cc-pvDZ, cc-pvTZ, and aug-cc-pvTZ basis
> sets, but at cc-pvQZ I run into an error saying:
>
> ------------------------
>  Eigenvalues of metric
>
>          1 0.212E-08 0.406E-05 0.681E-05 0.681E-05 0.835E-05 0.102E-04
> 0.103E-04 0.106E-04
>
>  ERROR: BASIS LINEARLY DEPENDENT OR WRONG S
>
>  The current threshold for the smallest tolerated eigenvalue of S is
> 1.0D-08
> ------------------------
>
> I spent some time searching the user-list and google, and I couldn't
> find a solution I needed.  What I gather is that since my geometries
> are reasonable, the jobs run fine with smaller basis sets, and since I
> only have one eigenvalue below the threshold, this error probably just
> comes from the fact my QZ basis is (not surprisingly) linearly
> dependent and should be scaled down one or a few basis functions.
>
> Pardon the naive question, but does someone know the "right" way to
> modify this basis set so it is not linearly dependent, but that it is
> also of desired quality for a meaningful CBS extrapolation? Is it as
> straightforward as starting at the QZ basis and removing individual
> diffuse functions until the basis set is not linearly dependent, or is
> there a higher art to it?
>
> My input file is:
> ***Pd_4Cl
>
> cartesian
>
> geomtyp=xyz
> orient,mass
> geometry={
> 5
> Pd_4Cl calc
> Pd   -0.213992    0.027380   -0.117721
> Cl   -1.943705   -1.515957    0.437439
> Cl    1.304488   -1.784426   -0.423439
> Cl    1.515859    1.570547   -0.672946
> Cl   -1.732472    1.839247    0.188003
> }
> set,charge=-2
>
> <BASIS>
>
> hf
>
> In my <BASIS> block, I've been using a script to pasting the explicit
> basis sets from the EMSL website into the input file. I don't want%2