Sire and OpenMM energies 1,4 scaling factor unity

[julien]

I pasted the email below so we keep a record of these discussions on the mailing-list. Please try to post discussions on bugs/issues etc here as much as possible. 

I think I didn't noticed this because I have only compared vacuum energies computed with no cutoff so I could get exact agreement between sander, OpenMM and Sire. 

So the issue seems to be that (please correct me if I am wrong):

- 1,4 Coulombic interactions are not modified by reaction field or shifted potentials in Sire. Sounds reasonable as doing this would unbalance the torsional energies.

- Sire assumes that 1,4 interactions could never have a scaling factor of 1.  We have just broken Sire by switching to a forcefield that uses a scaling factor of 1.0 for 1,4 interactions. As a result, when loading Glycam solutes, a reaction field is incorrectly applied by Sire to compute the 1,4 coulombic energies.  It's a Sire problem, so we carry on with disagreement between OpenMM and Sire energies until we fix Sire.  

Gaetano - 

Can you confirm whether a reaction field is applied or not to compute the 1,4 coulombic energies by ''standard'' OpenMM (in NonBonded) ? If unsure ask clarification on the OpenMM forums. 

If OpenMM applies the RF to 1,4 interactions, then am I right to say that OpenMMMDintegrator() and Sire will not give the same energies?

Can you confirm that if you otherwise do not have a scale factor of 1.0 for your 1,4 interactions, the energies between Sire and OpenMM are identical when using a reaction field ? 

Julien

/Hi Julien,
/
/The introduction of the new 1-4 is giving me different energy results between Sire and OpenMM. The problem seems to be in Sire. Sire seems to behave differently if /the scaling factor is 1. I have attached an old Chris email. This means that I have to modify the force field as well. What should I do?
/
/
/>>>>>>>>>>>>
/   Hi Gaetano,
/
/I think I have found the source of your problem. Sire assumes that any intramolecular interactions that have a non-bonded scale factor that is not 1 must be purely /intramolecular, so are calculated exactly, without applying any cutoff scheme. See the IntraCLJPotential in cljpotential.cpp. There are lines of code like;
/
/                            if (cljscl.coulomb() != 1)
/                                icnrg += cljscl.coulomb() *
/                                            param0.reduced_charge * param1.reduced_charge *
/                                                one_over_r;
/                            else
/                                icnrg += param0.reduced_charge * param1.reduced_charge *
/                                            (one_over_r - one_over_Rc + one_over_Rc2*(r-Rc));
/
/cljscl.coulomb() returns the intramolecular coulomb scale factor (actually, can be any scale factor). This is normally 1 for all intramolecular pairs, but is 0 for 1-2 pairs, /0 for 1-3 pairs, and equal to the 1-4 scale factor for 1-4 pairs. Your 1-4 interaction in HOOH will have a 1-4 scale factor that is not equal to 1, so the exact coulomb /equation will be used to calculate the interaction (i.e. reaction field or electrostatic shifting will be ignored).
/
/  Best wishes,
/
/  Christopher
/
/>>>>>>>>>>

[Gaetano Calabró]

Hi Julien,

I think I didn't noticed this because I have only compared vacuum energies computed with no cutoff so I could get exact agreement between sander, OpenMM and Sire. 

So the issue seems to be that (please correct me if I am wrong):

- 1,4 Coulombic interactions are not modified by reaction field or shifted potentials in Sire. Sounds reasonable as doing this would unbalance the torsional energies.

YES, I can confirm this. All the 1-4 Coulumbic interactions are evaluated without reaction field or shifted potentials

- Sire assumes that 1,4 interactions could never have a scaling factor of 1.  We have just broken Sire by switching to a forcefield that uses a scaling factor of 1.0 for 1,4 interactions. As a result, when loading Glycam solutes, a reaction field is incorrectly applied by Sire to compute the 1,4 coulombic energies.  It's a Sire problem, so we carry on with disagreement between OpenMM and Sire energies until we fix Sire.  

YES, I can confirm this as well. 

 

Can you confirm that if you otherwise do not have a scale factor of 1.0 for your 1,4 interactions, the energies between Sire and OpenMM are identical when using a reaction field ? 

I wrote a little OpenMM program and it does not seem to apply reaction field between 14-pairs using standard Non Bonded. 

If OpenMM applies the RF to 1,4 interactions, then am I right to say that OpenMMMDintegrator() and Sire will not give the same energies?

Apparently the reaction field is not applied on the 14 using standard non-bonded. If in Sire you use reaction field and the scaling factor is 1 then on the 14 pair Sire applies the reaction field otherwise not.  

Can you confirm that if you otherwise do not have a scale factor of 1.0 for your 1,4 interactions, the energies between Sire and OpenMM are identical when using a reaction field ? 

YES, if the scaling factor is not 1 Sire and OpenMM are in agreement using or not reaction field. 

--
You received this message because you are subscribed to the Google Groups "Sire Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sire-develope...@googlegroups.com.
To post to this group, send email to sire-de...@googlegroups.com.
Visit this group at http://groups.google.com/group/sire-developers.
For more options, visit https://groups.google.com/d/optout.

[Gaetano Calabró]

I’m sorry I made a mistake to the question:

Can you confirm whether a reaction field is applied or not to compute the 1,4 coulombic energies by ''standard'' OpenMM (in NonBonded) ? If unsure ask clarification on the OpenMM forums. 

>>

I wrote a little OpenMM program and it does not seem to apply reaction field between 14-pairs using standard Non Bonded.

[julien]

Hi Gaetano, 

Commit your code changes, even if the energies do not match with Sire. Write comments in the source code so we don't forget this in the future (or someone else tries to figure out what is going on). 

We don't need the Sire energies for Sire/OpenMM free energy calculations since all gradients are computed within OpenMM. 

I will discuss separately with Chris how to modify Sire to give the expected behaviour.

Thanks, 

Julien

To unsubscribe from this group and stop receiving emails from it, send an email to sire-developers+unsubscribe@googlegroups.com.
To post to this group, send email to sire-developers@googlegroups.com.