Date: Wed, 04 Jul 2001 11:50:19 -0400
From: "Bayly, Christopher"
Subject: RE: CCl4 ESP Grid problem with jaguar | 010628a (fwd)

I have checked Jaguar vs Gaussian92 for the same gridpoints around the same
geometry of methanol; the value of the electrostatic potential (ESP) at each
gridpoint is very similar but not identical. With a range of -0.52 to 0.036
for ESP values, the deviations between jaguar and Gaussian span a range of
-0.00015 to 0.00015, with higher deviations associated with higher values of
the ESP. The unrestrained fitted charges are pretty similar:
atom g98 jag
---- ------ -----
O -.6680 -.6657
HO .4233 .4224
C .1955 .1941
(methyl H's not listed)
I think this level of charge variability is not significant for simulations,
so I would use Gaussian or Jaguar ESPs interchangeably for the same grid of
points.
--- --- ---- ---- -- ------

So why were the CCL4 charges so different between Gaussian and Jaguar? I
agree with Dale Braden's point that the grid differences are important;
personally I think that it is the most important factor for the variability.
CCl4 is very non-polar and it has a point-group symmetry Th, for which the
lowest non-vanishing moment is maybe octapole or hexadecapole. These are
worst-case scenarios for sampling differences (i.e. different grids) and
numerical instabilities; even a modest difference in radial sampling could
lead to significant variability in the fitted charges. The high symmetry
would also exacerbate the the effects of subtle differences in geometry.
Were RESP restraints applied? Not trying to flog my own work here, but it
was developed to address exactly this kind of variability. Hope this helps,

Christopher

-----Original Message-----
From: jim caldwell
Sent: Tuesday, July 03, 2001 3:38 PM
To: amber@heimdal.compchem.ucsf.edu
Subject: Re: CCl4 ESP Grid problem with jaguar | 010628a (fwd)



For those of you generating RESP charges, let message below
be a cautionary tale. Each of the 3 common QM codes (Gaussian,
GAMESS, Jaguar) yield different Electrostatic Potentials. It
may be possible to coax them into giving the same but that is
not the default behavior.

Also note that ALL the Kollman/Cornell/AMBER charges have been
and will be generated with Gaussian.

best,
jim


---------- Forwarded message ----------
Date: Tue, 3 Jul 2001 13:26:41 -0300 (BRT)
From: Jones de Andrade
To: amber-request@cgl.ucsf.edu
Subject: Re: CCl4 ESP Grid problem with jaguar | 010628a (fwd)

Hello,

Well, this is just a forwarded Schroedinger Support' message,
concerning about some noticed differences in the resp calculated charges
(in this case, for CCl4 molecule) with jaguar esp grid. The point was that
they really differ from the ones obtained with gaussian and, as far as I
know, from the ones that can be obtained with gamess. Only to inform,
maybe it could help someone else.

Sorry for taking your time,

Sincerally yours,

Jones de Andrade
Date: Mon, 2 Jul 2001 10:53:48 -0700 (PDT)
From: "Help at Schrodinger, Inc"
Subject: Re: CCl4 ESP Grid problem with jaguar | 010628a

Hi,

I don't see anything obviously wrong in Jaguar's ESP charges. As I
mentioned before, they will certainly differ a little from those reported
by Gaussian.

Jaguar gives the following atomic charges after fitting to the ESP:

Atom C1 Cl2 Cl3 Cl4 Cl5
Charge -.24133 .06007 .06035 .06054 .06038

Gaussian gives:

1 C -0.433819
2 Cl 0.107895
3 Cl 0.107895
4 Cl 0.108053
5 Cl 0.109976

But when you are comparing the results from two different programs, you
must make sure that both programs use the same methods and the same
accuracy, tolerances, etc. For fitting to the ESP, Jaguar used 12414 grid
points; Gaussian used 776. Gaussian used Merz-Kollman radii for the atoms:

Merz-Kollman atomic radii used.
Atom Element Radius
1 6 1.50
2 17 1.70
3 17 1.70
4 17 1.70
5 17 1.70

but Jaguar does not (numbers in 'vdw2' column):

atom vdw vdw2 cov mass grid daf charge basis
C1 1.949 1.900 .770 12.000 Y Y Y 6-31g*
Cl2 1.958 1.974 .990 34.969 Y Y Y 6-31g*
Cl3 1.958 1.974 .990 34.969 Y Y Y 6-31g*
Cl4 1.958 1.974 .990 34.969 Y Y Y 6-31g*
Cl5 1.958 1.974 .990 34.969 Y Y Y 6-31g*

The geometry was constrained to C2v symmetry in the Gaussian optimization,
but symmetry was turned off entirely in the Jaguar optimization, so the
final geometries are slightly different. Finally, Jaguar's pseudospectral
method will yield a slightly different wave function from Gaussian's, so
the ESP at a particular point in space will be different. I think these
differences between Jaguar and Gaussian account for the discrepancy between
the ESP charges calculated by the two programs.

Sincerely,

Dale Braden


On Fri, 29 Jun 2001, Jones de Andrade wrote:

> Hello,
>
> Jus writing in order to send you the input and output files of the
> similar simulations in gaussian. The esp grid aren't similar. So, if you
> can find an exponential for this fact, I would be really glad in
> knowing... ;-)))