[gmx-users] Force-field checking options

[Mark Abraham]

On 3/08/2011 9:20 PM, Justin A. Lemkul wrote:
>
>
> mcgrath wrote:
>> Hi Mark.
>>
>>> Uh, they're available wherever you were getting energy break-downs 
>>> from (e.g. in the .log file, or via g_energy on the .edr file). You 
>>> can't get a break-down for each interaction, however. Simplify your 
>>> system to probe things here. Do zero-step MD (not EM) without 
>>> constraints, to evaluate the energy of a single conformation, and 
>>> compare that with your other software. Complex things are complex to 
>>> compare. :-) Reduce the complexity.
>>
>> I must be missing something really obvious, then, because this is where
>> I'm getting the energies from in the log file.
>>
>>    Energies (kJ/mol)
>>             U-B    Proper Dih.  Improper Dih.          LJ-14
>> Coulomb-14
>>     2.42228e+02    1.23963e+02    6.15742e-02    1.43130e+02   -5.01705e
>> +03
>>         LJ (SR)   Coulomb (SR)      Potential    Kinetic En.   Total
>> Energy
>>     8.30424e+01   -4.63865e+02   -4.88849e+03    1.05098e+02   -4.78339e
>> +03
>>     Temperature Pressure (bar)   Constr. rmsd
>>     3.00959e+02   -7.08408e+02    6.28488e-06
>>
>> I don't see any bend energies there.  No bonds, either, but that's fine,
>> because I was constraining them (constraints all-bonds).  Removing that
>> constraint gives a bond energy in that section, but still no bend
>> energy.
>>
>
> All angles in CHARMM27 use Urey-Bradley type angles, so the angle 
> bending energies are the U-B term.

Different forcefields/programs use the label "Urey-Bradley" to refer to 
either the sum of the harmonic angle potential and the harmonic 1-3 
potential (e.g. CHARMM27 in GROMACS), or just the latter (e.g. CHARMM27 
in CHARMM). So mileage does vary.

> -Justin
>
>> Agree completely about simplification.  The first system I quoted for
>> Justin was for two molecules (ATP + Mg), which is the smallest system
>> that I see the problem for (as I mentioned, using 29 or 25000 waters
>> gave a difference of only 1%, which is explainable by slight rounding of
>> the coordinates...at least, possibly).

You didn't say how you were measuring the energies, but if it was MD 
averages, then your averages are over different ensembles, even from the 
same starting point with notionally the same algorithm.

>>   He then asked for the breakdown
>> for the big system, so I gave him that, too.
>>
>>> ...and the magnitude of its energy contribution is available.
>>
>> Also agreed; that's how I concluded that its contribution was only a
>> small part, and using only ATP there appears to be no CMAP (see the
>> above energies).  But, even on this small system, I haven't been able to
>> hunt down the differences.  The parameter files have the same values for
>> the nonbonded parameters (converting between the GROMACS and CHARMM
>> format) and charges, so I'm not seeing any simple solution.  Unless
>> someone has an idea to save me, I guess I'm stuck with looking at
>> individual contributions for this system.
>>
>> Sigh.  Some days this field isn't so much fun.  :)  Especially since the
>> bug will probably be something stupid I did in setting it all up.

Probably. Getting the output from grompp -pp, and then choosing atom(s) 
to have zero charge, and/or zero VDW parameters is a good way to 
trouble-shoot, assuming your other software can be massaged to replicate 
the effect. Before that, check that the energy of a single amino-acid 
can be replicated.

Mark

>>
>> Thanks!
>>
>>                                      Cheers, Matt
>>
>> On 3/08/2011 3:45 PM, Mark Abraham wrote:
>>> On 3/08/2011 3:34 PM, mcgrath wrote:
>>>> Hi Justin, thanks for the response.  For the big system, the
>> dihederal
>>>> is still fine, while now the improper, U-B, and nonbonded (CP2K
>> groups
>>>> all nonbonded, 1-4, and real space Coulomb into the same term) are
>> off
>>>> by a factor of 2-3.  Is there any way to print the energy due to
>> angles
>>>> in GROMACS?
>>> Uh, they're available wherever you were getting energy break-downs 
>>> from (e.g. in the .log file, or via g_energy on the .edr file). You 
>>> can't get a break-down for each interaction, however. Simplify your 
>>> system to probe things here. Do zero-step MD (not EM) without 
>>> constraints, to evaluate the energy of a single conformation, and 
>>> compare that with your other software. Complex things are complex to 
>>> compare. :-) Reduce the complexity.
>>>
>>>>    The nonbonded interactions dominate (for GROMACS, the short
>>>> range Coulomb term is an order of magnitude bigger than everything
>>>> else).  I guess it's possible my Ewald parameters are off, though
>> that
>>>> wouldn't affect the SR term.  My box is 85x85x105 and I use a grid of
>>>> 88x88x108, along with a 4th order interpolation for both simulations,
>>>> along with an ewald_rtol of 1e-5.  Playing with these doesn't seem to
>>>> change the numbers much, nor does switching between PME, SPME, or
>>>> regular Ewald (for CP2K).  rlist=rvdw=rcoulomb=1.2.  Is there a
>>>> parameter that I'm missing?  I've tried playing around with vdwtype
>> and
>>>> coulombtype (I believe CP2K truncates and shifts nonbond and
>> Coulombic
>>>> realspace potentials), but nothing had a significant effect.
>>>>
>>>> Thanks for the gmxdump tip.  I'll start looking at the parameters for
>>>> the smaller system, and hopefully I can hunt down where the
>> differences
>>>> are popping up. CP2K doesn't have CMAP implemented, but that only a
>>>> small part according to GROMACS.
>>> ...and the magnitude of its energy contribution is available.
>>
>> ...and trivial to turn off in pdb2gmx.
>>
>> Mark
>>> Mark
>>>
>>>>                            Cheers, Matt
>>>>
>>>>>> -------- Forwarded Message --------
>>>>>> From: Justin A. Lemkul<jalemkul at vt.edu>
>>>>>> Reply-to: jalemkul at vt.edu, Discussion list for GROMACS users
>>>>>> <gmx-users at gromacs.org>
>>>>>> To: Discussion list for GROMACS users<gmx-users at gromacs.org>
>>>>>> Subject: Re: [gmx-users] Force-field checking options
>>>>>> Date: Wed, 03 Aug 2011 00:12:35 -0400
>>>>>>
>>>>>>
>>>>>> mcgrath wrote:
>>>>>>> Hi.  I'm fairly new to GROMACS, and I've been using it to run some
>>>>>>> classical MD simulations.  In order to be sure that I'm using the
>>>>>>> correct force field (I had to add a molecule to CHARMM27), I'm 
>>>>>>> comparing
>>>>>>> it to another simulation code that I know well, CP2K (using
>> GROMACS
>>>>>>> 4.5.4 and a recent CVS version of CP2K).  Sadly, they are giving
>> me
>>>>>>> energy differences of about a factor of 2 for a 75000 atom 
>>>>>>> protein+water
>>>>>>> system.  As far as I can tell, I'm using the same PME parameters,
>> and
>>>>>>> there's not a big change in energy when I change those, anyway.
>> I've
>>>>>>> been able to confirm that it's not a global problem (computing the
>>>>>>> energy of only the 25,000 TIP3P waters give a result to within
>> 1%,
>>>>>>> which
>>>>>>> is not perfect, but better...I seem the same 1% difference if I 
>>>>>>> only use
>>>>>>> 29 waters).  For a smaller system (using the molecule I added),
>> the
>>>>>>> total energy is incorrect, but the torsion and improper torsions
>> are
>>>>>>> good to within 1%.  So it looks like my topology is perhaps being
>>>>>>> incorrectly specified, or the parameters for them.
>>>>>> Which energy terms show the discrepancy in the case of the 
>>>>>> twofold difference?
>>>>>>
>>>>>>> What I would like to do is get GROMACS to print out all the 
>>>>>>> charges (the
>>>>>>> electrostatics/nonbonded are also different) that it is actually 
>>>>>>> using,
>>>>>>> as well as the force field parameters being used.  By using mdrun
>> -v
>>>>>>> -debug I get some of that, but lines like
>>>>>>>
>>>>>> You don't need mdrun -debug for this.  The topology is static, so 
>>>>>> run gmxdump on
>>>>>> your .tpr file.  This will map all parameters to each atom.
>>>>>>
>>>>>> -Justin
>>>>>>
>>>>>>> c6= 1.48497790e-03, c12= 6.58807176e-06
>>>>>>> c6= 3.78914556e-04, c12= 4.08982345e-07
>>>>>>> c6= 2.02168059e-03, c12= 4.42785949e-06
>>>>>>> c6= 1.71635114e-03, c12= 4.54480232e-06
>>>>>>> c6= 2.60732602e-03, c12= 6.42257237e-06
>>>>>>> c6= 3.75109637e-04, c12= 2.77185705e-07
>>>>>>> c6= 1.98187144e-03, c12= 5.24788538e-06
>>>>>>> c6= 6.18919948e-05, c12= 7.54611396e-09
>>>>>>> c6= 1.73354731e-03, c12= 5.36550624e-06
>>>>>>> c6= 4.41942073e-04, c12= 4.93156790e-07
>>>>>>> c6= 6.79507386e-03, c12= 2.55060841e-05
>>>>>>> c6= 1.61714898e-03, c12= 3.18964840e-06
>>>>>>> c6= 2.48678902e-04, c12= 2.13336705e-07
>>>>>>>
>>>>>>> are somewhat unclear.  They are obviously non-bonded terms, but
>>>>>>> corresponding to which atoms?  The same goes for the bond angles
>> and
>>>>>>> torsions printed later.  The worst-case scenario is that I have
>> to
>>>>>>> poke
>>>>>>> around in the source files, which I would to avoid so I'm hoping 
>>>>>>> there
>>>>>>> is some documentation or more switches I can flip somewhere.
>> Thanks!
>>>>>>>                     Cheers, Matt
>>>>>>>
>>
>