ramess101
commented
6 years ago @mostafa-razavi
I agree that section V should not be titled finite size effects, since that is only one of the three topics you address (also flexible bonds and MD vs MC). However, this discussion should NOT go in Section VI, Simulation Details. I think we just need to change the section name to something like: Reliable Second Virial Coefficients or Reliable Low Density Results.
This whole discussion about atomic vs molecular virials could be important to include, especially if we can verify how LAMMPS and GOMC compute the virial. I think a brief statement explaining how MD and MC often use different types of virial is sufficient. This comparison of MD and MC could be another reason to include my GROMACS results for branched alkanes. Then we would show how we can get similar results for two MD codes and two MC codes.
I do not think we really need to go into finite size effects for Z and U at higher densities, unless we already have some good results that we could present here.
@mostafa-razavi I think it will be easier to have a separate issue for each of the reviewers sections
The method presented by the authors to is relying on compressibility factors (Z=PV/RT) obtained in NVT simulations from open source packages like Cassandra using Monte Carlo sampling or LAMMPS using molecular dynamics. Unfortunately the manuscript does not discuss the limitations of pressure calculations in these codes, which are key in this process. Their interpretation of possible discrepancies in the low density limit as a result of finite size effects is therefore highly questionable (see remarks 5.1 , 5.2 and 5.3).
5.1 This section is discussing mostly the uncertainty on Z = PV/RT and on the resulting virial coefficients. It would be useful that the authors discuss the uncertainties and approximations on the determination of pressure in NVT simulations, as this is undoubtedly the most delicate point when evaluating Z. It would be worth indicating whether pressure is computed from the molecular virial (as generally the case in Monte Carlo methods) or from the atomic virial (as generally done in Molecular Dynamics). This point is very important in the proposed method because the atomic virial expression requires more severe conditions to converge at low density, as shown by the authors. Also, it would be fine mentioning what cutoff is used in low density phases and whether long range corrections are used in pressure calculations.
However this discussion is not related with finite size effects, it should be placed in the section on VI. Simulation details.
5.2 It seems to this referee that Figure 14 and Figure 15 are illustrating the divergence of the pressure determination in the zero density limit when the molecule has internal degrees of freedom. This is behavior is frequent when the atomic virial route is used for pressure determination because the atomic virial expresses as PV = nkT + W , where n is the number of atoms (not the number of molecules!) and the virial term W contains contributions from internal forces and from intermolecular forces. Depending on the numerical approximations in computing intra‐ and intermolecular forces in the atomic virial, Z may not tend toward 1 in the zero density limit. This effect is not related with finite size effects, it should be discussed thoroughly and placed in another section (Simulation details).
5.3 : Figure 16 : the authors should mention whether ethane is simulated with a flexible C‐C bond or in a rigid model approximation. It is possible to introduce bond stretching in Monte Carlo codes and compute pressure from the molecular virial route. Is it the case here ? Whatever the case, it is not surprising that MC predicts better the compressibility factor Z in low density phases, because MC codes are generally using the molecular virial route in which PV = NkT + W’ where N is the number of molecules (not the number of atoms) and where the virial W’ contains contributions from intermolecular forces only. Thus Z tends toward 1 in the low density limit. If this interpretation is correct, there is no reason to use a large number of molecules. The convergence of (Z‐1)/rho does not seem to be affected by finite size effects , as suggested by the section title. It is however affected by the way pressure is evaluated in the simulation codes used by the authors.
5.4 Finite size effects on the compressibility factor Z and on the internal energy U are not discussed for dense phases. Packing effects are known to be significant below 200 molecules. As they are probably significant, they should be discussed as well.