United-atom Mie lambda-6 viscosities

[ramess101]

We are simulating the normal and branched alkanes to determine the adequacy of the united-atom Mie lambda-6 potential to predict viscosity. Although the ultimate goal is to predict high pressure viscosities, currently we are investigating the ability to predict saturation viscosity. In order to account for uncertainties in the parameter set (epsilon/sigma) for a given lambda, we have performed a Bayesian MCMC uncertainty analysis.

The figure below shows the feasible CH3 parameter sets for different values of lambda:

This figure shows the percent deviations in liquid density and vapor pressure for the different parameter sets:

As we can see, the 15-6 or 16-6 potential appear to be the best options for predicting VLE.

We then simulate a subset of these epsilon/sigma parameter sets to determine the overall uncertainty in viscosity for a given lambda.

Here are the viscosity results for ethane (just lambda =13-16), where the error bars represent the distribution in epsilon/sigma sets:

Clearly, the viscosity is over-predicted with the UA Mie potential, and the bias increases with increasing lambda. Although the 13-6 potential is better at predicting viscosity, recall that it is significantly worse at predicting VLE than the 15-6 or 16-6. In other words, there is a significant trade-off between accurate prediction of VLE and accurate prediction of viscosity. This suggests that we need an alternative (perhaps all-atom) model to accurately predict both VLE and viscosity.

We are repeating this analysis now for propane, butane, and n-octane where we include the uncertainty in the CH2 parameters.

[mostafa-razavi]

Potoff-C12H26-N400-rc14

compare_TDE_REFPROP_sat.pdf Potoff-C12H26-N400-rc14.zip

[ramess101]

Hallelujah! This is awesome @mostafa-razavi . So this really was a cut-off issue for the larger alkanes.

@Kaiveria

Can you prepare the C22 (or whatever the next REFPROP compound is C20 maybe?) files for Mostafa to run? Specifically, we need the N400 boxsizes for saturation. This could be really useful. Mostafa can start TraPPE and we will run Potoff and TAMie once we have some cores free.

[mostafa-razavi]

Is C16 running? Does using packmol instead of Gromacs for initiating simulations help?

On Thu, Jul 19, 2018 at 1:00 PM, Richard Messerly notifications@github.com wrote:

Hallelujah! This is awesome @mostafa-razavi https://github.com/mostafa-razavi . So this really was a cut-off issue for the larger alkanes.

@Kaiveria https://github.com/Kaiveria

Can you prepare the C22 (or whatever the next REFPROP compound is C20 maybe?) files for Mostafa to run? Specifically, we need the N400 boxsizes for saturation. This could be really useful. Mostafa can start TraPPE and we will run Potoff and TAMie once we have some cores free.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ramess101/IFPSC_10/issues/1#issuecomment-406380874, or mute the thread https://github.com/notifications/unsubscribe-auth/APma4dZhDYf-8HGBFq__0jb8YtHDWh2wks5uINdmgaJpZM4TiFN9 .

[ramess101]

@mostafa-razavi

Yes, we already finished all the C16 results. They looked good.

No, we didn't need packmol. The cut-off was the real issue. If Gromacs fails to build C20/C22 @Kaiveria will let you know.

[mostafa-razavi]

@ramess101 How can the cutoff be an issue? I thought your TraPPE-C16 with rcut=1.4 nm failed, too. Right?

[ramess101]

@mostafa-razavi

TraPPE-C16 did run successfully. I think we misspoke previously that it did not work. Potoff was the only one that failed. Using a 1.0 nm cut-off molecules don't feel any force and then suddenly feel a fairly strong force when they move closer. This is very problematic for flexible molecules that will feel non-bonded within the same chain. Does that make sense?

[mostafa-razavi]

@ramess101 Hmmm... So this problem potentially exists even for rcut=1.4, but it's less likely. Right?

[ramess101]

@mostafa-razavi

Yeah, there is always a small pulse in force when you cross from zero interaction to within the cut-off, but as you get to 1.4 nm the force is significantly smaller than at 1.0 nm. We are considering including this in a discussion/supporting information section of our manuscript.

[mostafa-razavi]

@ramess101 So this could be a demonstration for one of the reasons why force-smoothed potentials (switched-force or shifted-force) might be better. Do you think we should test this with a switched-force potential function?

[ramess101]

@mostafa-razavi

It certainly is an example of where a switched-force/shifted, etc. potential would be better. I think we can test this quite easily. I am not sure we can get meaningful results before the deadline, mainly because we would need to reparameterize the switched-force to get the same viscosity as the cut-off. I think we can just mention this without proving it.

[Kaiveria]

@mostafa-razavi @ramess101 Good afternoon. I have heard that a C22H46 TraPPE run is to be started soon. I have checked the C22H46 files and have successfully started a test run with 200 molecules, though it required changing the number of tries for insert-molecules in the AlkanesViscosity.sh script:

gmx insert-molecules -ci ../../../"$Compound".gro -nmol "$Nmol" -try 2000 -box "${liquid_box[j]}" "${liquid_box[j]}" "${liquid_box[j]}" -o "$Compound"_box.gro > insertout 2>> insertout

This used to specify "-try 500" for me which was not sufficient.