Subject: Re: AMBER: About RESP Calculation
Date: Fri, 19 Mar 2004 18:45:22 +0100 (CET)
From: FyD

Dear Ilyas,

> 1. I have tried to use the program RED written by Mr. Dupradeau.

Program RED was written by A. Pigache and the corresponding study was carried
out in collaboration by: P. Cieplak and F.-Y. Dupradeau

> I entered
> the .pdb file and it took 3 days and still didnt end up with the
> calculation. I killed the job. Are the gaussian calculations taking
> that long?

You should check how "works" your optimization: Does it oscillate ?
A good way to check this is to apply the following shell command:

egrep "SCF Done|Step number|Maximum Force|RMS Force|Maximum
Displacement|RMS Displacement" *.log

Then, you check at which step number the four threshold criteria are the
smallest and you use the corresponding "Standard orientation" as input for a
second optimization run using this time Opt=(Tight,CalcFC) (see RED manual page
12). Once it is done you run RED as it is describe at the end of page 7 of the
RED manual in order to read the the Gaussian minimization output as input for
MEP computation. (in the future RED version GAMESS ouputs will be read as inputs
for Gaussian calculation...)

> I did not change the ATOM NAMES, and later I realized that
> in
> the sugar molecule, there is a CH2 term. Is it possible that this is
> the
> reason for this problem? (If it is a problem for the gaussian to
> calculate a particular job more than 3 days)

The atom name are modified to create automatically RESP inputs (see RED manual
page 13). That has nothing to do with Gaussian/GAMESS optimization.

Finally, it is impossible to say that 3 days are a long computation time as you
do not report which machine type you are using.

> 2. Do we have to optimize the structure of a particular molecule in
> order
> to calculate the RESP charges? For instance, can I just use the
> crystal
> structure I got from the CCD and just do the MEP calculation and then
> the RESP calculations? Is this legitimate?

I am not an expert in that... I would say that ab initio and X-ray structures
can present structural differences. Moreover, the ab initio minimization is
performed in gaz phase and I think X-ray structures are more "condensed phase"
structures even if they are "solid". So I would apply the ab initio minimization
step in the derivation of RESP /ESP charges...

> 3. In the RED program, it is talking about re-orientation. I followed
> the
> manual, and did not understand exactly what it is meant by
> reorientation.

There are at least 4 papers that describe the molecular orientation effect on
the charge values:
Spackman JCC 1996 17 1-18
Merz Jr JCC 1992 13 749-767
Woods et al. JCC 1990 11 297-310
Breneman & Wiberg JCC 1990 11 361-373

Two of them are referenced in the RED manual.

The ab initio software (Gaussian and GAMESS) can/does re-oriente the structure
also: You should check the orientation of the "Standard orientation" generated
by Gaussian: you will see that it changes during the minimization... It means we
simply added in RED a new/another re-orientation method allowing to get highly
reproducible RESP charges whatever is the ab initio software.

> In the tutorials there are examples for re-orientation, but after
> getting
> an optimized structure, why do we have to reorient the structure?

It means that using your minimized structure, if you cannot get a constant
orientation (generated by the ab initio software you use) for the Cartesian
coordinates on which is computed the MEP, you will get different set of charges
for each orientation possible. An that is THE point: GAMESS and Gaussian canNOT
generate a constant orientation for a minimum: Did you try to optimize EtOH
using GAMESS and Gaussian (see RED tutorial in the manual) ? Using both software
we got 4 sets of different charges. Moreover, we were NOT able to reproduce the
charges published by Fox & Kollman JPC B 1998 102 8079-8079. The maximum diff.
observed was .043 for one carbon atom, same problem for the DMSO... Thus, in
some cases the ab initio software can generate several orientations for the same
minimum. Moreover, the orientation generated by GAMESS and Gaussian are
different. This means that different sets of RESP/ESP charges can be potentially
generated for the same minimum.

To solve this problem, we decided to add a "general?" "common?" re-orientation
procedure for both ab initio software. It is the rigid body re-orientation that
is applied on the minimized structure just before the MEP calculation. Thus
controlling the orientation selected using this algo., we can get reproducible
RESP charges. A multi-orientation RESP fit (by analogy to the
multi-conformational RESP fit) can also be executed in RED which allows to
average the charge differences observed for one orientation over several
orientations. I think all this is more or less explained in the RED manual pages
13-14. However, it is clear it is not easy understandable without the
corresponding paper... I am working on it ;-)

> The
> structure has a clear structure when viewed in InsightII or Molden.
> So,
> isnt it legitimate to do the MEP calculation in Gaussian and then the
> RESP calculations without reorient the structure?

It is up to you. The answer is the following: Do you want that people that will
read your work will be able to reproduce the charge values you will publish ? If
you care about this, apply the re-orientation algo. available in RED otherwise
do not apply it in RED (and it was the way to derive RESP/ESP charge before
writting RED).

> 4. I tried to follow Cornell's paper to recalculate the results
> published
> in 1995. I read the RESP documents that are in the scripps website, and
> in those info files it is saying that in order the results to be
> compatible with Cornell et al. 95 force field,
> #P MP2/6-31G* pop=mk iop(6/33=2)
> needs to be used in the gaussian input file. I use one of the cytosine
> molecule used in Cornell et al. 95 paper. The gaussian gives an error
> while doing the calculations. I do not exactly remember what the error
> sentence is, but it needs more space in the harddrive, more than 20
> Gbyte
> space. Is this normal? It happened to me twice. I used

What is your cpu ? Is it a 32 bit cpu ? Did you check the size of your temporary
files. If one is bigger than 2GB, your calculation will be killed. In this case,
you have to split the temporary file into several ones, smaller. See gaussian 98
User's Reference page 17-18.

> #P HF/6-31G* pop=mk iop(6/33=2) SCF(Conver=6) Nosymm Test
> which I found in Mr. Dubradeau's RED program manual, and it gives me a
> result without any error.
> PS: The above Gaussian command was described in an email sent by Jim
> Caldwell to amber@cg1.ucsf.edu in Feb 21, 1996 with a subject "esp to
> resp".

Well if I remember well you also reported the use of other Gaussian keywords to
derive RESP charges available on:
http://signe.teokem.lu.se/%7Eulf/Methods/resp.html
and it did not work in antechamber...

To conclude, the derivation of RESP charges can be really tricky because
numerous options are possible and different charges for the same minimum can be
obtained. It is a reason also why we work on RED, to allow people to publish
reproducible charge. By this way, we can really judge were are the errors in the
protocole. I hope RED will provide a 'standard' protocole for RESP and ESP
charge derivation... The new version will incorporate multi-conformation and
multi-orientation RESP/ESP fit, extra points (lone paires) handling and other
features...

I hope this is more clear now, best regards, Francois

F.-Y. Dupradeau