AMBER Archive (2009)

Subject: Re: [AMBER] R.E.D. too small charges

From: FyD (fyd_at_q4md-forcefieldtools.org)
Date: Fri Nov 13 2009 - 07:06:51 CST


Dear Andrei,

Such a problem has been reported several times - with differences more
or less stricking. When one accounters such charge differences, two
approaches are possible: either one does not want to see the problems
& all is fine, or one tries to understand what's going on, & one can
spend some times ;-)

- For RESP charges & using the default Connolly algo. in MEP
computation, we observed a charge incertainty of up to 0.07 e when
using the same minimum represented by different molecular orientations.
- ESP charges also present charge incertainty upon molecular
orientation (even larger _?) & this is not only related to buried
atoms...
- If you do not use an identical conformation the charge differences
are even larger. The presence or absence of hydrogen bond(s) in the
selected conformation also impact charge values.
- Charge equivalencing for chemically equivalent atom or for atoms the
user consider chemically equivalent also affect charge values.
- and the worst, program bugs & human errors strongly affect charge values ;-)

Thus, in the end: how big are the differences ? impossible to answer.
The sum of all differences can be quite big: What to do to solve this
problem ? My personal opinion is that you made the good choice when
you decided to use the R.E.D. tools ;-) This means that when you will
use R.E.D., you will NOT be able to reproduce GLYCAM or AMBER charge
values - simply because they are not reproducible. But the charge
values You will generate for Your molecules when using R.E.D. will be
reproducible by other users if You decide to follow several
recommended points. If you decide to submit your data in R.E.DD.B.
your data will be reviewed before registration. The goal here is not
to be selective, but simply to correct errors carried by users, to
improve charge derivation procedure and to discuss with users about
their approaches. Finally, you might try R.E.D. Server as well, so
that you can test GAMESS-US, PC-GAMESS & Gaussian. You can send P2N
input files to get private assistance as well...

You can also send me a .tar.gz file of your whole R.E.D. run so that I
can have a look (if you wish).

regards, Francois

Now, few specific comments below:

> I am trying to reproduce the charges of alpha-D-glucopyranose form:
> Restrained electrostatic potential atomic partial charges for
> condensed-phase simulations of carbohydrates
> R. J. Woods and R. Chappelle
> Journal of Molecular Structure: THEOCHEM
> Volume 527, Issues 1-3, 4 August 2000, Pages 149-156
>
> using GAMESS-US and R.E.D. III.3
> The problem I face is that I obtain too small charges for all the
> heavy atoms in the molecule compared to the ones in the paper above.
> I use the same approach as described in the paper:
> Charges I obtain Charges from Woods & Chappelle
> 1 O1 -0.3074 **** -0.6887
> 2 HO1 0.3037 **** 0.4772
> 3 C1 -0.0319 **** 0.2552
> 4 H1 0.1292 **** 0.1186
> 5 C2 -0.0307 **** 0.2474
> 6 H2 0.1281 **** 0.0978
> 7 O2 -0.3825 **** -0.6640
> 8 H2O 0.2603 **** 0.4281
> 9 C3 -0.0266 **** 0.1484
> 10 H3 0.1228 **** 0.1188
> 11 O3 -0.3477 **** -0.6732
> 12 H3O 0.2783 **** 0.4357
> 13 C4 -0.0385 **** 0.0492
> 14 H4 0.1298 **** 0.0754
> 15 O4 -0.3830 **** -0.6607
> 16 H4O 0.2614 **** 0.0754
> 17 C5 -0.0329 **** 0.2047
> 18 H5 0.0963 **** 0.1295
> 19 O5 -0.1371 **** -0.4966
> 20 C6 -0.0139 **** 0.2543
> 21 H61 0.0984 **** -0.0080
> 22 H62 0.0984 **** 0.0003
> 23 O6 -0.5796 **** -0.6805
> 24 H6O 0.4052 **** 0.4266
>
> I understand that charges depend on conformation but the differences
> are simply too big.

too big ? you never know...

However, for O1 -0.3074 & -0.6887
              O5 -0.1371 & -0.4966, the diff. are quite big, yes...

> This problem also appear on all monosaccarides that I had tried to
> compute the charges for and even for smaller molecules.
> I am fairly new to charge derivation and I tried to solve this for two
> weeks without any success. It is obvious that I do a mistake somewhere.

May-be not...

Let's compare for instance, the H6' & H6" charge values:
You have equivalenced these charge values: this is correct.
I do not understand why they are not equivalenced by Woods & Chappelle
   21 H61 0.0984 **** -0.0080
   22 H62 0.0984 **** 0.0003

> GAMESS installation is working properly as it passed all the tests.
> Also the structures minimized within REDIII charge derivation process
> are OK as tested with another QM software at the same level of
> description.
> I think the things go wrong somehere in the RESP procedure but I can't
> seee where.
> I am using RESP-C2 charge derivation method with qwt= 0.0100 and CHELPG
>
> Here is the GAMESS input header:
>
> ! Single point to get MEP - Input generated by R.E.D.-III
> !
> $CONTRL ICHARG=0 MULT=1 RUNTYP=ENERGY MOLPLT=.T.
> MPLEVL=0 UNITS=ANGS MAXIT=200 EXETYP=RUN
> SCFTYP=RHF
> COORD=UNIQUE $END
> $SCF DIRSCF=.T. CONV=1.0E-06 $END
> $SYSTEM TIMLIM=5000 MWORDS=32 MEMDDI=0 $END
> $BASIS GBASIS=N31 NGAUSS=6 NDFUNC=1 $END
> $GUESS GUESS=HUCKEL $END
> ! CHELPG/CONNOLLY CHARGES
> $ELPOT IEPOT=1 WHERE=PDC OUTPUT=BOTH $END
> $PDC PTSEL=CHELPG CONSTR=NONE $END
> $DATA
>
> and,
>
> Here is the RESP input header:
>
> RESP-C2 project. RESP input generated by R.E.D.
> &cntrl
> ioutopt=1, iqopt=1, nmol=1, ihfree=1, irstrnt=1, qwt= 0.0100
> &end

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber