Reading the exported file

[Peter Knowles]

As I wrote in my last message, but perhaps not loudly enough, is that
this issue is now resolved, and the coefficients are dumped to 11
decimals with patch 2000.1.molden
Peter
At Sat, 27 Oct 2001 12:24:14 -0400,
Dave Moore wrote:
> 
>  
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hello,
> 
> 	There are issues with recreating densities by simply using the
> MOLDEN output as-is, because there are precision problems with the
> basis coefficients as written by MOLPRO in the MOLDEN input file. 
> Specifically, all of the coefficients are truncated after the 6th
> decimal place, rather than the 6th significant digit.  This causes
> artifacts in densities computed using the coefficients, particularly
> when calculating difference densities.  
> 	For example, take th HF/6-311++G(3df,3pd) density for a Mg atom,
> then rotate the basis functions by 45 degrees and subtract the
> resulting density from the original.  The difference density computed
> in this manner  should be zero within the precision of the
> calculation (specified as 1e-8 using the appropriate MOLPRO
> keywords), however the difference density that is computed looks like
> a d-orbital with maximum magnitude ~1e-4.  This occurs because the
> low precision of the d-orbital coefficients in the valence orbitals
> (which have a negligible contribution to the diagonal density) causes
> low precision in the s-d cross terms in the density matrix (which are
> not negligible).  
> 	This problem would be avoided if the orbital coefficients were
> reported to the 6th significant digit, rather than the sixth decimal
> place.  It would be very useful for me to have this option when
> generating MOLDEN-style output, which I use for my difference density
> program.  Or if it is possible to get the coefficients output to 6
> sig figs in *any* ASCII output file, that would also be a help. 
> Thank you,
> 
> Dave Moore
> Chemistry Dept.
> University of North Carolina at Chapel Hill
> 
> - -----Original Message-----
> From: owner-molprouser at tcpc3.bham.ac.uk
> [mailto:owner-molprouser at tcpc3.bham.ac.uk]On Behalf Of Peter Knowles
> Sent: Saturday, October 27, 2001 10:43 AM
> To: John Kerkines
> Cc: molpro-user at tcpc3.bham.ac.uk
> Subject: Reading the exported file
> 
> 
> I think at this point we can say definitively that the EXPORT command
> should essentially be forgotten about. It was originally set up when
> orbital and density records were simple dumps of the (floating-point)
> numbers; now they contain lots of metadata as well, and any attempt
> to
> EXPORT them as floating point may well cause floating point
> exceptions
> on, for example, Alpha hardware.
> 
> The recommended alternative is to use the interface to the molden (
> http://www.caos.kun.nl/~schaft/molden/molden.html ) code (see Molpro
> manual section 12.3.1,
> http://www.tc.bham.ac.uk/molpro/molpro2000.1/doc/manual/node98.html#90
> 25
> ) in which are dumped the geometry, MO coefficients and occupation
> numbers. Provided that one always dumps natural orbitals, this should
> be enough to reconstruct the density matrix as well.
> The precision of the dumped MO coefficients has for these purposes
> been
> raised through patch 2000.1.molden
> 
> Peter
> 
> At Tue, 18 Sep 2001 17:57:18 +0300 (EET DST),
> Ioannis Kerkines wrote:
> > 
> > Hello,
> > 
> > I have used the EXPORT card to export the orbitals of the records
> > 2130.2 and 2140.2. After that, I have used the Fortran code
> > specified in the manual in order to read these two exported binary
> > files. I get two files (one for 2130.2 and one for 2140.2) with
> > lots of numbers. Does anybody know how to read this file?
> > 
> > I have spotted in various places in these last files, the nuclear
> > charges, the geometry, the coefficients of the basis functions in
> > each molecular orbital and the MO eigenvalues. There are also lots
> > of zeroes (actually there are numbers like 10**(-322) = 0 ).
> > 
> > In which lines are the above data located? And what about in the
> > record 2140.2 when doing a state-averaged CASSCF? 
> > 
> > Best Regards,
> > John Kerkines
> 
> - -- 
> Prof. Peter J. Knowles              
> Email P.J.Knowles at bham.ac.uk  Phone +44-121-414-7472  Fax
> +44-121-414-7471
> School of Chemical Sciences, University of Birmingham, Birmingham,
> B15 2TT, UK
> WWW http://www.tc.bham.ac.uk/~peterk/
> 
> -----BEGIN PGP SIGNATURE-----
> Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>
> 
> iQA/AwUBO9rfrV5cg+YcPBRAEQLDdwCglO7i7sRm28J620ARrVIL4biZE58AoJvB
> lNm9tbGUPVNd6FmdgnzLt6dn
> =9MDH
> -----END PGP SIGNATURE-----

-- 
Prof. Peter J. Knowles              
Email P.J.Knowles at bham.ac.uk  Phone +44-121-414-7472  Fax +44-121-414-7471
School of Chemical Sciences, University of Birmingham, Birmingham, B15 2TT, UK
WWW http://www.tc.bham.ac.uk/~peterk/