VMD-L Mailing List
From: Daniel Fellner (dfel694_at_aucklanduni.ac.nz)
Date: Wed Jun 10 2020 - 07:01:37 CDT
- Next message: Rameswar Bhattacharjee: "Radial distribution function calculation"
- Previous message: yjcoshc: "Re: error: can't read "w" no such variable."
- In reply to: JC Gumbart: "Re: FFTK: Charge Optimisation partial charges"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Thanks, that's really helpful!
I managed to get most of these to converge near zero by increasing the
water shift range from -0.4 / 0.4 to -2.0 / 2.0
Should be much easier to get these to reasonable values now.
*Daniel Fellner BSc(Hons)*
PhD Candidate
School of Chemical Sciences
University of Auckland
Ph +64211605326
On Wed, 10 Jun 2020 at 23:44, JC Gumbart <gumbart_at_physics.gatech.edu> wrote:
> Please keep vmd-l in your replies.
>
> The COLP is described in the source paper - I recommend reading it
> thoroughly: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3874408/
>
> Best,
> JC
>
> On Jun 10, 2020, at 7:38 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz>
> wrote:
>
> Is this via COLP? I was wondering if there was some kind of guide or
> manual for interpreting that. I'm unsure about what 'delta or obj' means or
> what to look for in terms of convergence. And why most distance plots are
> blank.
>
>
> *Daniel Fellner BSc(Hons)*
> PhD Candidate
> School of Chemical Sciences
> University of Auckland
> Ph +64211605326
>
>
> On Wed, 10 Jun 2020 at 23:19, JC Gumbart <gumbart_at_physics.gatech.edu>
> wrote:
>
>> FFTK tracks the agreement between the energies in QM and MM as well as
>> the distances. You can identify problematic ones and try to adjust the
>> weighting, for example, to put more emphasis on them during the fitting.
>>
>> I think it’s okay to help the optimization by giving it a limited range
>> to explore, as long as the results are good.
>>
>> Best,
>> JC
>>
>> On Jun 10, 2020, at 6:56 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz>
>> wrote:
>>
>> I've excluded the water interactions which didn't find a minimum or were
>> affected by secondary interactions (2x symmetrical ACC-carbonyl carbon, 2x
>> symmetrical ACC-120 degree carbonyl, and 2x symmetrical DON-carbonyl
>> carbon) though most of the waters found a reasonable minimum.
>>
>> *"Are the interaction energies for the optimized charges in good
>> agreement between QM and MM?"*
>>
>> By this, do you mean do the QM water interaction energies (derived from
>> delta water-complex energy v.s. molecule in vacuo HF energy) correlate with
>> the distances? Or that the optimised charges in FFTK correlate with the HF
>> charges? Because I haven't checked the former and the latter is where I'm
>> seeing the problem. If I change the initial charges from the CGenFF charges
>> to the MK dipole charges from the MP2 SP calculation, and set a range
>> of +/-0.2 from the HF MK Dipole charges, I get more reasonable results, but
>> I fear this may be overfitting.
>>
>> Thank you.
>>
>>
>> *Daniel Fellner BSc(Hons)*
>> PhD Candidate
>> School of Chemical Sciences
>> University of Auckland
>> Ph +64211605326
>>
>>
>> On Wed, 10 Jun 2020 at 05:33, JC Gumbart <gumbart_at_physics.gatech.edu>
>> wrote:
>>
>>> FFTK is not using Mulliken charges. It’s trying to optimize the
>>> interaction energy with water molecules. As much as we might wish
>>> otherwise, FFTK can’t be treated as a black box. Look at the QM output -
>>> are all water molecules finding a minimum at a reasonable distance? Are
>>> the interaction energies for the optimized charges in good agreement
>>> between QM and MM? etc.
>>>
>>> Best,
>>> JC
>>>
>>> On Jun 9, 2020, at 12:15 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz>
>>> wrote:
>>>
>>> Hi all,
>>>
>>> I'm having some difficulty deriving appropriate partial charges. The
>>> charges don't seem to match chemical intuition and differ significantly
>>> from the CGenFF-assigned charges (when not specifically excluded) for even
>>> low penalty atom types.
>>>
>>> Upon reading the Gaussian output files, I actually only find Mulliken
>>> charges in the output. I was under the impression that Mulliken charges are
>>> not suitable for such uses. Is the Gaussian input script writer perhaps not
>>> providing the correct keywords for my version of Gaussian (09-D.01)?
>>>
>>> As an example, the input line for the MP2 calculation reads:
>>>
>>> *# MP2/6-31G* SCF=Tight Density=Current*
>>>
>>>
>>> *Daniel Fellner BSc(Hons)*
>>> PhD Candidate
>>> School of Chemical Sciences
>>> University of Auckland
>>> Ph +64211605326
>>>
>>>
>>>
>>
>
- Next message: Rameswar Bhattacharjee: "Radial distribution function calculation"
- Previous message: yjcoshc: "Re: error: can't read "w" no such variable."
- In reply to: JC Gumbart: "Re: FFTK: Charge Optimisation partial charges"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]