Dear ADF users,

In response to Alessandro Bagno's E-mail below, I would like to add a
few more suggestions and comments on diffuse basis and fit functions.

- The basis sets with diffuse functions in the ET directory can be used
  for ZORA calculations as well. In general, for atoms up to Kr,
  all-electron basis sets for nonrelativistic calculations can be used
  for ZORA calculations as well.

- The standard diffuse basis sets provided with ADF are all limited to
      * light elements up to Kr
      * quite large basis sets (based on TZ2P or larger)
  There can certainly be situations where diffuse functions are useful
  for smaller basis sets (e.g. DZP) or for heavier elements.

Some suggestions for adding diffuse basis and fit functions to existing
ADF basis sets:

We will assume that one wishes to add diffuse functions to the file
$ADFRESOURCES/DZ/Pt.4d for some reason.

The first question to ask is: for what l-values are diffuse functions
most important? Here chemical intuition should help you. For example,
for hydrogen a diffuse P function should be important, whereas for
carbon, a diffuse s and d would be expected to be important.

In absence of such arguments, one can simply add diffuse functions
for all l-values, which seems to be a good default choice.

The second question is: what n-value and exponent should I choose?
We would suggest to locate the most diffuse basis function in the
file for the selected l-value(s).

For the sake of this example, we will assume that (arbitrarily)
additional diffuse P and D functions are desired. The most diffuse
functions currently in the basis are indicated below.

We would recommend to continue in an even-tempered fashion with
the same nl-values. The factor in the even-tempered series should
typically be somewhere between 1.5 and 2.0. The value 1.7 is usually
a quite good default for the basis.

In the example below, this would lead to the additional diffuse
functions

5D 1.19 (and if a second diffuse D is needed, then also 5D 0.70)
6P 1.15 (exponent = 1.95 / 1.70)

BASIS
1S 56.000
2S 36.100
3S 20.700
4S 12.200
2P 39.350
3P 17.450
4P 12.400
3D 21.750
4D 10.950

5S 6.800
5S 4.550
6S 2.850
6S 1.550
5P 5.850
5P 3.650
5D 4.300
5D 2.030 <------- most diffuse 5D basis function in original basis
4F 15.700
4F 8.600
4F 4.700
6P 1.950 <------- most diffuse 6P basis function in original basis
END

As a next step, also the fit should be adapted (the CORE and DESCRIPTION
sections do not have to be modified). The fit should be able
to describe the products of any two basis functions. If more diffuse
basis functions are added, also the fit needs to become more diffuse.

We recommend to locate the most diffuse fit functions for ALL l-values,
and add one or two additional diffuse fit functions in the same
even-tempered manner as described above. However, for the fit a smaller
factor between exponents should be used. A factor of 1.5 should be
a reasonable value. Alternatively, one could build on the already available
fit functions. For example, in the Pt file, the most diffuse S fit functions
are

11S 2.95
11S 2.48

This implies a smaller value than 1.5 and one might choose to continue
with the smaller factor.

However, we would recommend to add the following diffuse S fit functions
(using a factor of 1.5):

11S 1.65
11S 1.10

For all calculations with such modified files, one should be very careful
to check the output for signs of linearly dependent basis or fit sets, and
for large fit errors. It is advisable to perform calculations with different
similarly constructed basis and fit sets to ensure oneself of the
reliability of the results. It is further recommendable to use the
DEPENDENCY keyword when using diffuse basis or fit functions.

If diffuse functions are used on several close-lying atomic centers,
the risk of linear dependent basis sets increases, as does the need
for the DEPENDENCY keyword.

In summary, if carefully performed by an experienced ADF user, it is
not so hard to add diffuse functions to an existing atomicdata file.
However, one should be aware that diffuse functions lead to an increased
risk for numerical problems, and one should always check the results
carefully.

Best regards,
Stan van Gisbergen

=================== Original message from A. Bagno ======================

Hello ADFers,
I also have a problem related to using diffuse functions. I would like
to use a basis including diffuse functions, but in the directories where
such basis sets are defined (.../Special/Vdiff; .../ET...) it is always
stated that they are non-relativistic. Since I am forced to use ZORA
bases for my problem, is there a way out?
Also, is there a special reason for adding diffuse functions only to
very large basis sets like QZ3P etc.? This may be a limitation for
fairly big molecules.

Many thanks
Alessandro Bagno (alessandro.bagno_at_unipd.it)

------------------------------------------------------------------------
Dr. S.J.A. van Gisbergen SCIENTIFIC COMPUTING & MODELLING NV
Phone: +31-20-4447626 Vrije Universiteit, Theoretical Chemistry
Fax: +31-20-4447629 De Boelelaan 1083
E-mail: vangisbergen_at_scm.com 1081 HV Amsterdam, The Netherlands
                               http://www.scm.com
------------------------------------------------------------------------
Received on 2002-09-09 19:06:01