Summary of "Anomalous PM3 Hydrogen Interactions"
Our thanks to the many individuals who responded to our concerns
on the PM3 parameterization. What follows is a transcript of the
responses. We hope you had a wonderful holiday season.
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
Original Query: (From: cburkhart (- at -) goodyear.com)
> Dear netters,
>
> We have a query regarding anomalously stable close contacts between
> nonbonded hydrogen pairs when using PM3.
>
> We would be interested in any comments regarding a problem we
> have encountered when using the semi-empirical method PM3. The
> simplest example of this problem is the formation of a "stable"
> "dimer" of methane. The geometry minimizes to a straight line
> between C-H ... H-C bonds in the dimer. This dimer possesses
> an apparent stabilization energy of about 2kcal/mole, with an
> H ... H separation distance of 1.7 Angstroms.
>
> We are concerned, as this effect is showing up in a host of crowded
> molecules--for example, t-butyl cyclohexane, or even in all-gauche
> conformation of n-pentane! We are not aware of any published
> discussion of this problem, but it is leading to what may appear
> to be erroneous geometries and heats of formation in many cases.
>
> Is this problem an inherent weakness of PM3? Are there satisfactory
> workarounds? Any references to discussions of this problem?
>
> Thanks in advance. Send all responses to cburkhart (- at -) goodyear.com,
> and we will summarize for the net...
>
> Sincerely,
>
> Eilert Ofstead and Craig Burkhart
> Goodyear Research
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: Christopher J Cramer <cramer (- at -) maroon.tc.umn.edu>
Craig,
Your problem was discussed by Csonka in J. Comput. Chem. 14 (1993) 895.
Not much in the way of solutions was offered, however.
CJC
--
Christopher J. Cramer
University of Minnesota
Department of Chemistry
207 Pleasant St. SE
Minneapolis, MN 55455-0431
(612) 624-0859
cramer (- at -) maroon.tc.umn.edu
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: metzger (- at -) quinn.medc.umn.edu (Thomas Metzger)
Dr. Burkhart,
Hello. My name is Tom Metzger and I am a graduate student in the Dept. of
Medicinal chemistry at the Univ. of Minnesota. I have done some work on
the various methods by which molecular mechanical, semi-empirical and ab initio
packages capture dispersion forces. In particular, the methane dimer was
chosen as a model system. The head-to-head or straight line geometry is the
preferred geometry by PM3 and and it is by far the most attractive.
Of all methods I tested (MM2, AMBER, MNDO, AM1, PM3, and Gaussian 90)
only PM3 gave this geometry. I believe the cause of this excess attraction
lies in modifications to the core-core repulsion term. It was found in MNDO
that there was excessive core-core repulsion just outside bonding distances.
To compensate for this, "attractive" gaussian terms were added to AM1
and PM3.
PM3 has two gaussians per atom and AM1 has between two and four. I would refer
you to the Journal of Computer Aided Molecular Design, Vol. 4 (1990) pp. 1-105.
It is an article by J.J.P. Stewart entitled MOPAC: A Semiempirical Molecular
Orbital Program. I am not sure of what can be done to get around this.
I have worked on a manuscript on my work on this topic but it has yet to
be submitted for publication and is not my thesis project and so I am not
sure when, if ever, I will get back to it. At any rate, I found it interesting
that you seem to have found the same result. I have found no discussion
of this particular topic in the literature. I gave a talk on this research
at the ACS meeting in Denver in April and since then have not worked on it.
I'd be glad to answer any further questions regarding this matter.
I would appreciate it if you'd pass along any comments you get regarding PM3
and this phenomenon. My address is metzger (- at -) quinn.medc.umn.edu. Thank
you.
Tom Metzger
metzger (- at -) quinn.medc.umn.edu
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: tom (- at -) sgih.roc.wayne.edu (Tom Wiese)
Craig,
I have run into this PMS Hydrogen problem when optimizing Estrogens. With
MOPAC 5 and 6 (PM3, Precise keywords) The C18 methyl h's are too close to
say an 11 beta OH hydrogen. In fact, when an OH is put on the 11 beta of
estrogen, the orientation of the 11OH hydrogen is always optimized to be
very (too) close to the C18 methyl, no mater which of the 3 possible 11B OH
orientations are used as starting conformations.
I discussed this with James Stewart about 3 years ago and sent him out
files. His comment at that time was that this was just the result of the
In Vacu simulation and mabe just a result of smaller hydrogens in PM3. In
oterh words not to worry.
We have been cautious of PM3 ever since and use MMP2 for all the Opts. we
can and use PM3 for Charges. This seems to work best for getting our
compounds ready for CoMFA QSAR. In some cases, we do use PM3 for geometry,
but only when MMP2 is lacking. PM3 and MMP2 both give steroid structures
that are very very close to X-Ray where AM1, MNDO and Tripos give very,
very poor steroid structures. The only deviation we have seen with the PM3
and the steroids structures is a described above for a single of our 50
analogues.
The comparison we did with X-Ray was with 12 estrogens, by the way.
Also, PM3 charges for the steroids are closer to the Gausian charges we
compaired with.
Call or email if you need more info on our case.
Please email if you find other coments/discussion.
tom
##################################################
Tom Wiese
Department of Biochemistry
Wayne State University School of Medicine
540 E. Canfield
Detroit, MI
48201
Phone: (313) 577-5605
FAX: (313) 577-2765
email: tom (- at -) sgih.roc.wayne.edu
##################################################
##################################################
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: zheng (- at -) violet.berkeley.edu
Hi, Craig:
This problem was reported in QCPE Newletter in 1991 (sorry I forget which
issue it was). The calculations were done on methane dimer. It is a weakness
of PM3 and AM1 does not have this problem.
Happy Holiday!
Yajun Zheng
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: zheng (- at -) violet.berkeley.edu
The exact reference is QCPE BULLETIN, vol. 11, number 1, page 5-6, 1991.
It was a brief communication by V. Bub, J. Messinger, and N. Heuser. I donot
know whether they have any other publication on this subject. Hope this is
useful to you.
Yajun
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: Andy Holder <AHOLDER (- at -) VAX1.UMKC.EDU>
Dear Dr. Burkhart,
I'm not sure that these results are unexpected with PM3. Dr. Stewart
has recently "recanted" on his PM3 method and has more or less gone
back
to doing things in the way that the Dewar group always did, i.e. one
element at a time with great attention being paid to the chemistry of
the systems involved in the parameterization. I recently did a
comparative study using PM3, AM1, and Professor Dewar's new SAM1 method.
These results were presented at the ACS meeting in Chicago. Basically,
SAM1 does the best job at handling these sorts of association/H-bonding
interactions. This is likely due to the better theoretical basis that
underlies the AM1 procedures. In any case, I would try AM1 and/or SAM1
for these systems.
Most problems of this sort are caused by spurious Gaussian functions
that were added to the elemental parameter sets to take care of other
problems. This is less of a problem with AM1 and SAM1 because these
functions are VERY carefully placed, whereas in PM3 they are largely
added at the beginning and become a large part of the method's
chemistry. I'll be happy to discuss this with you further if you like.
This area is of interest to me.
Andy Holder
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
DR. ANDREW HOLDER
Assistant Professor of Computational/Organic Chemistry
Department of Chemistry || BITNET Addr: AHOLDER (- at -)
UMKCVAX1
University of Missouri - Kansas City || Internet Addr: aholder (- at -)
vax1.umkc.edu
Spencer Chemistry, Room 315 || Phone Number: (816) 235-2293
Kansas City, Missouri 64110 || FAX Number: (816) 235-1717
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: jstewart (- at -) fai.com (Dr. James Stewart)
With chagrin I have learned that PM3 is not good at modeling non-H
bonded hydrogen bonds. The main fault appears to be that the H- -X
distance is too small. The secondary fault, the the energy is too
high, also appears to be valid.
At present, I've no plans to correct these faults, as Prof Walter Thiel
is developing a new method which looks very good.
Jimmy Stewart
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: "Frederick F. Ignatz-Hoover"
Dr. Burkhart;
The only information I have is the report in the MOPAC 93
Manual Chapter 9.6 Faults and Errors in PM3. Reported there is
"H- non-bonded H contacts are too short, at about 1.7 A..." as
you report, no reference given. Also reported are inaccurate
proton affinities see ref. J. L. Ozmet and A. M. Schiedekamp in
Int. J. Quant. Chem., 000:000, 1992 "Proton Affinities of
Molecules containing Nitrogen and Oxygen: Comparing Ab Initio and
Semi-Empirical methods with experiments."
I'm interested in your findings....
Dr. Fred Ignatz-Hoover
Research Spec.
Monsanto, IPG, Akron, Oh.
Phone: (216) 668-6346
Email: 557-8973 (- at -) MCIMAIL.COM
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: Hens Borkent -- CAOS/CAMM Center <BORKENT (- at -) caos.caos.kun.nl>
Dear Dr. Burkhart,
I don't have the MOPAC manual at hand (the new one, MOPAC93),
but I'm sure there is a remark by Stewart concerning this problem,
when discussing the relative merits of AM1 and PM3.
Sincerely,
Hens Borkent
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: h.rzepa (- at -) ic.ac.uk (Henry Rzepa)
This was discussed, in the QCPE Bulletin about 3 years ago. I do not
have these, but if you find the reference, I would be grateful.
It is in effect a known problem.
Dr Henry Rzepa, Dept. Chemistry, Imperial College, LONDON SW7 2AY;
rzepa (- at -) ic.ac.uk via Eudora 2.01, Tel:+44 71 225 8339, Fax:+44 71 589
3869.
>From April '94: (44) 171 584 5774, Fax: (44) 171 584 5804
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: h.rzepa (- at -) ic.ac.uk (Henry Rzepa)
Sorry; another example. We cite it in chem comm, 1993, p1337
Dr Henry Rzepa, Dept. Chemistry, Imperial College, LONDON SW7 2AY;
rzepa (- at -) ic.ac.uk via Eudora 2.01, Tel:+44 71 225 8339, Fax:+44 71 589
3869.
>From April '94: (44) 171 584 5774, Fax: (44) 171 584 5804
http://www.ch.ic.ac.uk/rzepa.html
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: dimas (- at -) hobbit.quimica.uniovi.es (Dimas Suarez Rodriguez)
Dear Eilert and Craig,
The following reference may be useful for you :
Molecular Orbital Studies of C-H....O H-Bonded Complexes.
Laszlo Turi and J.J. Dannenberg
J. Phys. Chem. 97 7899-7909 (1993)
These authors study a variety of CH..O bonded complexes and
report calculations performed by MP2 and HF ab-initio leves using
6-31G(d,p) and D95++(d,p) basis sets. Also they use AM1, PM3 and the
recently developed SAM1 methods. Comparing the semiempirical results
with ab-initio calculations, they conclude that AM1 is clearly the
best semipirical method to model this kind of electrostatic interaction,
while "PM3 is erratic in both the energies and structure, sometimes
giving what appear to be physically absurd results".
Your sincerely,
Dimas Suarez
Dpto Quimica Fisica y Analitica
Universidad de Oviedo
Asturias. Spain
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: "Eric P. Wallis" <epw (- at -) ppco.com>
Craig,
Have you tried the same calculation with the AM1 Hamiltonian? It might
give better results, since PM3 was optimized for nitrogen compounds and the
hydrogens might have stronger interactions. The AM1 Hamiltonian was
optimized for hydrocarbons. Let me know how this works.
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From: Jose Ignacio Garcia <jig (- at -) qorg.unizar.es>
Dear Dr. Burkhart,
Concerning your mail to the CCL about anomalous stable close contacts between
nonbonded hydrogen pairs, I have observed exactly the same effect when
studying conformational energy surfaces and transition structures.
In the first case, false minima are located, due to the stabilization of
structures with two hydrogens in close contact (about 1.7 A). In the second
one, transition structures are located which are more stable than expected
because the existence of nonbonded interactions between hydrogens of both
reagents (at a distance of c.a. 1.8 A). This is particularly annoying when
one is trying to predict a selectivity coming from the relative energy of
the various transition states of the reaction.
I am not a theoretical chemist, but I think that this result represent an
inherent weakness of the PM3 (which is not present, for instance in the case
of AM1). This fact can be shown by studying the energy partition of the
methane "dimer" (by means of the keyword ENPART of the MOPAC program).
The distance between hydrogens is calculated to be 1.70 A in the case of PM3
and 2.25 A in the case of AM1. Taking into account all the possible
combinations of energy//geometry, one finds:
Geometry Method
AM1 PM3
AM1 -0.0419-0.0038+0.0294=-0.0162 -0.2174-0.0245+0.1674=-0.0744
Energy
Method
PM3 -0.1195-0.0107+0.0287=-0.1014 -0.4166-0.0474+0.0908=-0.3732
All values are in eV. The bicentric energy is done as the sum of
Resonance + Exchange + Coulombic = Total
As can be seen, PM3 always leads to a high stabilizing total energy, even at
2.25 A. In the case of AM1, there is also a stabilizing interaction between
hydrogens at 1.70, but resonance and coulombic energy terms are much better
balanced, so the optimization leads to a longer distance between hydrogen
atoms and a weaker interaction between them.
To sum up, AM1 seems to be the method to choice when such type of interaction
is expected to be geometrically possible. In such cases, PM3 can lead to very
misleading conclusions!
I would like to hear the comments of some of the popes of semiempirical
methods. I expect that your summary of responses will open an interesting
debate in the net. Please, confirm me the receipt of this message.
Merry Christmas and happy new Goodyear! (sorry, it was an irresistible
temptation)
Jose Ignacio
*******************************************************************************
Dr. Jose Ignacio Garcia-Laureiro
Departamento de Quimica Organica Phone : 34-(9)76-350475
Instituto de Ciencia de Materiales de Aragon Fax : 34-(9)76-567920
C.S.I.C.-Universidad de Zaragoza e-mail: JIG (- at -)
CC.UNIZAR.ES
50009 ZARAGOZA (SPAIN) JIG (- at -)
QORG.UNIZAR.ES
*******************************************************************************
"And all this science I don't understand is just my job five days a
week..."
ELTON JOHN - Rocket man
*******************************************************************************
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
From (- at -) SEARN.SUNET.SE:SOKALSKI (- at -) PLWRTU11.BITNET Thu Dec 23
08:29:27 1993
From: Andrzej Sokalski <SOKALSKI%PLWRTU11.BITNET (- at -) SEARN.SUNET.SE>
Dear Craig,
I am not suprized by your result. In fact all ZDO based methods
like PM3, CNDO/2, INDO etc. have the same problem, neglect of
exchange term of the interaction energy (see J.Mol. Struct.THEOCHEM
234,393(1991) Fig. 1 & 2. This deficiency is frequently corrected by
scaling nuclear repulsion term, but such procedures belong to empirical
atom-atom potential category, rather than rigorous theoretical procedure
with adjustable accuracy. See ref. 22-32 in the above mentioned paper
for reports of similar problems with other ZDO methods.
Greetings - Andrzej
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
[][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][][]
Once again, our thanks and appreciation to all who participated in this
query...
--------------------------------------------------------------------------
Craig W. Burkhart, Ph.D. Senior Research Chemist
E-mail: cburkhart (- at -) goodyear.com The Goodyear Tire &
Rubber Co.
Fone: 216.796.3163 Research Center
Fax: 216.796.3304 142 Goodyear Boulevard
Akron, OH 44305
--------------------------------------------------------------------------
For a successful technology, reality must take precedence over
public relations, for Nature cannot be fooled - Feynman
--------------------------------------------------------------------------