Hi:
Surely we can learn much from these confabulations. However, any time
I am somewhat surprised about such concentration of thoughts about
Gaussian. Is anything in NWChem inferior to Gaussian as far as basic
ab initio or DFT are concerned? If so, I could understand, however I
would like to see were computations with NWChem are at fault. NWChem
is provided on academic free license for nearly any machine, without
any restriction about the number of nodes/processors or whatsoever
else, and with code that you are allowed to compile at your will. Is
the same true for Gaussian? Probably Gaussian offers ready-to-use
plots, with (I suspect largely illusory) things as simulated IR
spectra. However, I could use with NWChem the functional M05-2X at a
time that Gaussian was at B3LYP (which failed to perform correctly on
my molecules). But I know that Francois is sensible to all that and
promised to extend to NWChem his important contributions. Hope 2009
will be the right year.
Cheers
francesco pietra
On Fri, Dec 12, 2008 at 5:00 AM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Francois
>
>> > is is also recommended to use iop(6/41=10) on top, which would add
>> > 10 concentric layers of points for each atom. Is this option
>> > recommended or is it a carryover?
>>
>> If you are interested in deriving RESP charges as in regular AMBER
>> force fields DO NOT USE IOp(6/33=2,6/41=10,6/42=17). This set of
>> keywords presents indeed some advantages in some cases [the tutorial
>> should explain why these keywords are used instead of IOp(6/33=2):
>> this is really confusing for new users] but should NOT be used if your
>> goal is to follow what has been done. The reference is IOp(6/33=2) &
>> not IOp(6/33=2,6/41=10,6/42=17) !
>
> Which reference is this specified in? I could never find a definitive
> document that specified exactly what options were used for Gaussian fits.
>
> I think these options were found by the students that wrote this tutorial
> from the attached webpage. I have no idea what the history of this page is.
> However, do you know if these options actually make a measurable difference?
> Sure they might give slightly different charges but then my experience with
> RESP is that it is only marginally reproducible at best anyway. It seems to
> me that, for example, upping the number of points per layer and the number
> of layers should simply result in the charges generated converging to a
> specific value as a function of the number of points used. If one gets
> different charges at the default settings then the default options are
> surely not large enough to obtain convergence in the results and will just
> lead to large error bars.
>
> I guess people should read the disclaimer at the top of all the tutorials:
>
> "These tutorials are meant to provide illustrative examples of how to use
> the AMBER software suite to carry out simulations that can be run on a
> simple workstation in a reasonable period of time. They do not necessarily
> provide the optimal choice of parameters or methods for the particular
> application area."
>
> The main purpose of the tutorials here is to take people through the steps
> involved in a way that lets them appreciate (and hopefully think!!!) about
> what is going on and avoid use of any black box approaches as much as
> possible. They don't attempt to discuss the complexities of MD simulations
> which make MD such a dynamic area of research.
>
> Note that Antechamber (from the Antechamber source) uses:
>
> #HF/6-31G* SCF=tight Test Pop=MK iop(6/33=2) iop(6/42=6) opt
>
> Which includes the 6/42 option although with a value of 6 instead of 17.
>
>> See for instance a simple Gaussian input to compute a MEP (Connolly
>> surface) as defined in AMBER force field .
>> http://q4md-forcefieldtools.org/Tutorial/3-Charges-Gaussian-A-QMRA/JOB2-
>> gau_m1-1-1.com (GFInput GFPrint are even not
>> required)
>>
>> SCF=Tight is not useful for single point/MEP computation (SCF=Tight is
>> only required in the geometry optimization step).
>
> I'm not sure the SCF Tight comment is strictly true here. Note Gaussian (at
> least in Gaussian 98) uses a very loose SCF convergence criteria for single
> point SCF calculations(something like 10^-4 if I remember, I don't have the
> manual to hand) which will certainly affect the results. I have not checked
> to what extent it effects the final RESP charges though. It certainly can't
> hurt to include SCF=Tight - it will just up the computational requirements.
>
>> I would also suggest you to use HF/6-31G* (or HF/6-31G**, Duan et al.
>> FF) in the geometry optimization step - if you do want to rigorously
>> follow what has been done (& not MP2/6-31G* or B3LYP/6-31G* as it is
>> described in this tutorial).
>
> Are you sure HF/6-31G* was used for geometry optimization in FF94 and
> subsequently FF99, 99SB etc? The original FF94 JACS paper is awful when it
> comes to describing exactly what was done but I find it hard to believe that
> HF/6-31G* was used for the optimization. Note Cornell et al. JACS 1993, 115,
> 9620 talks about MP3/6-31+G**//HF/6-31G* and MP2/6-31G*//HF/6-31G* and MM2
> but never mentions HF/6-31G*//HF/6-31G*. I don't have the FF03 paper to hand
> to check what Yong did (no internet connection at the moment). I am
> intrigued to know where you get the recommendation for HF/6-31G* geometry
> optimization from.
>
>> > To get a feeling of whether this procedure is correct I had a look
>> > at /usr/local/amber9/examples/resp_charge_fit/water. I created a
>> > water molecule in molden and applied the same procedure (this time
>> > with and without option iop(6/41=10)). The obtained charge are
>> > basically identical:
>> > O1 -0.81327
>> > H1 0.40664
>> > H2 0.40664
>
> As I thought - upping the Gaussian precision does not impact the results.
>
>> I looked at your problem & computed RESP or ESP charges for water using
>> HF/6-31G*//HF/6-31G* (Cornell at al. FF, 1994-...)
>> or the olds:
>> HF/STO-3G//HF/6-31G* (Weiner at al. FF, 1984/1986)
>> HF/STO-3G//HF/STO-3G (Weiner at al. FF, 1984/1986)
>
>> This means the data available .
>> /usr/local/amber10/examples/resp_charge_fit/water/ have been generated
>> using HF/STO-3G//HF/STO-3G !
>> I guess they come from Amber... 4/3 ?. Do not use that as a reference !
>
> So if you use HF/6-31G*//HF-6-31G* can you reproduce the charges in the FF94
> paper?
>
> All the force field parameter derivations, like bond lengths and angles were
> done using MP2/6-31G* (same for GAFF) so it makes sense that they would use
> the same for charges, although there is the discussion that HF/6-31G*
> overestimates the gas phase dipole moment which is beneficial for solution
> phase simulations but from what I can see the discussion was never very
> clear what geometry this referred to.
>
> Perhaps someone should volunteer to provide a tutorial on the AMBER website
> that goes through a step by step process for reproducing the FF94 / FF99 /
> FF99SB and FF03 parameters (charges, VDW and valence terms) for say Alanine.
> I think this would really help clear up lots and lots of confusion that
> exists in the "black art" of force field derivation.
>
> Just my 3c.
>
> All the best
> Ross
>
>
> /\
> \/
> |\oss Walker
>
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> | http://www.rosswalker.co.uk | PGP Key available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
>
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
to majordomo.scripps.edu
Received on Sun Dec 14 2008 - 01:08:13 PST
