DeFT is a linear combination of gaussian type orbitals density functional
(LCGTO-DF) program. The code has been written in standard FORTRAN and should
therefore run as is on most any computer. However, I can only personally
attest to its functionning properly on HP, IBM, and SGI workstations.
Analytical first derivatives have been implemented. Consequently, geometry
optimizations can be efficiently performed. However, vibrational analyses,
including infrared intensities, can be quite time-consuming as they require
the evaluation of second derivatives by finite differentiation of the
analytical first derivatives. Gradient corrections to the local spin density
approximation have also been implemented.
The accompanying compressed tar file includes all the DeFT source code, sample
input and output files, documentation explaining the use of the various
input options, and documentation explaining the LCGTO-DF methodology. Basis
sets for the first row elements and hydrogen are included. Basis sets
for heavier elements can be obtained from the Cray WWW site
( http://www.cray.com/apps/UNICHEM/BASIS.html ). They will have to however
be reformatted to respect the format used by DeFT for basis sets.
You will need to compile the code yourself. Be prepared for this to take
a significant amount of time.....
Jul 10, 1995 corrected sources accourding to the message.
From: email@example.com (Alain St-Amant)
Subject: CCL:Potential Problems with DeFT
firstname.lastname@example.org (Delin Shen) writes:
> We are currently installing DeFT program of Dr. St-Amant in several
>computer systems. we came cross one section of the program that we think
>might has a minor mistake:
> subroutine delphidr.f
> >>130 call delphi(ndelphi,ncenters,nptcharges,ngrid,ngridby2,natomtype,
> >>131 & iwkvec,epsilon,coord,wkvec(jf),coordptcharge,ptcharge,
> >>132 & wkvec(ja),wkvec(jb),wkvec(jc),wkvec(jd),wkvec(je),
> >>133 & wkvec(ig),wkvec(jh))
> Should it be "jg" instead of "ig"?
Sorry about that. Yes it should be "jg" instead of "ig". For those using
the code, don't worry about it. `delphi' is the module having to do with
the program's solvation model, which I never got around to finishing up.
It's very similar to the Honig/Friesner work. As it is, there's nothing
in the input deck that you can add to make this routine jump into action.
> There also some other sections of the program that don't conform standard
> Fortran. We sent Dr. St-Amant a message more than month ago about this and
> haven't got any answers yet.
Sorry about the delay in getting back. To anybody who doesn't get a reply
within a day or two, just e-mail me again.
Again, don't worry about it. Answers are in no way affected. The problem
that arose is that one of the subroutines (gradxxds1.f) had an extra
(or *bonus* :-)) argument at the end of its argument list (grad).
This argument is never used throughout the course of the subroutine. The
calls to this subroutine therefore seem to have one too few arguments. Since
the missing argument is the never used last argument, no problem. No wonder
it slipped my attention. But to make everything nice, people should remove
'grad' from the 'gradxxd1.f' argument list.
> We have reproduced all sample output. The problem we have right now
>is a test run we did with benzene optimization using a fine and random grid:
>D6h symmetry input gives nonsymmetrical result. Since it is optimization
>without conserving the symmetry, we are not expecting output with D6h
>symmetry. What bothers us is that C-H bond distances differ about 0.03 A.
This is way too big a variation. True, the grids can't maintain the symmetry
of benzene, and there will be variations in bond lengths, but nowhere near this
large. I've run it, and I get nowhere near the errors^H^H^H^H^H^H noise (:-)
you're getting. All C--C bonds are within 0.0001 Angstroms of each other and
all C--H bonds are within 0.0001 Angstroms of each other, which of course is
but a fraction of the inherent errors of DFT relative to experiment. This
is actually better than usual. Don't be shocked with something of the order
of 0.001-0.002 Angstroms. A fine grid slaps on about 3000 points per atoms.
Send me a copy of your input deck, and I'll check it for potential problems.
For those wondering why we can't maintain D6H symmetry:
Unfortunately, the grids used don't have the symmetry of the molecule
in question and symmetry is broken. We could enforce symmetry, but that
would be sweeping the problem under a rug. Randomizing the orientation
of the angular points on the radial shells is an attempt to smooth out
the noise in the total energy (at the cost of losing symmetry in systems
that actually could maintain their symmetry with the grids in question).
Hope this helps.
Nicely recovering from the coronary I had when I saw 0.03 Angstroms,
P.S. DeFT is available for free from the OSC archives. If you want, just
come over to the U. of Ottawa Chemistry WWW site and there will be a
hyperlink to the specific place in the archives where DeFT is stored.