ramess101
commented
6 years ago I have attached a file that addresses two of the points that we discussed.
First, @ejmaginn this article provides results using atomic and molecular shear-stress. Figure 1 and 3 show that the difference is only appreciable at very short times for Green-Kubo.
Second, @dcarls0n this article compare Green-Kubo and Einstein. They get very similar results for the two methods, see Figure 2. Not sure if there is anything we really need from this, but it might be a good starting point.
@ejmaginn @dcarls0n @jrelliottoh @drroe
Here are the meeting notes from 4/17:
Modify title so that it is more representative of the text, i.e. Best Practices for Transport Properties: v1.4 Diffusivity and Viscosity from Equilibrium Molecular Dynamics
4.1.1. Bulk up discussion about why we do not recommend NPT, why NVT is OK, and why NVE is best.
4.1.1. Provide brief discussion about barostats and thermostats
We need to get permission from journals to reproduce figures.
4.2.3. @ramess101 will clarify step 6 of bootstrap uncertainty quantification
4.3. What do we mean by "we also recommend checking the constraint tolerances"? Did either @drroe or @jrelliottoh write this statement? We need some clarification. Was this referring to bond constraints?
4.3. Where it says "we recommend that RECOMMENDATION", we will move the discussion about using the log-log MSD vs t plot and the heurestic of square root of MSD > radius of gyration. These topics are currently found in Section 5.2.2. but belong here.
5.1.1. Rename this section "Data analysis" and clarify that we recommend using the COM for computing r_i and v_i (see Table 1). Unwrapped coordinates is a secondary issue for this section but should be clarified a bit. The point is to encourage clear communication of what is being computed in Equations 1 and 2.
5.1.2. Bulk up the discussion on finite-size effects. Include the equations for how to correct for finite size effects. Discussion about non-cubic boxes should either go here or in Section 5.4 Special topics.
5.1.3. Explain that the off-diagonal components should be 0. Also, we should be more explicit that for heterogenous systems they diagonal components are not equal.
5.2.1. Discuss the trade-off between accuracy and file size
5.2.2. The heurestic and discussion of log-log plot is moved to section 4.3. Briefly refer back to that section.
6.1.2. Clarify that 5-10 fs is recommended because any more frequent would require huge amounts of file storage (step size is only 1-2 fs to begin with) while any less frequent would not capture the initial decay of autocorrelation (see Figure 10).
6.1.4. Provide the viscosity equations for three or six tensor elements. We think that we should not include these equations in Table 1 since that is too much information for a reader right at the beginning. However, we think it might belong better in the appendix to not detract from the discussion.
Appendix: in addition, we might want to include Dxx, Dyy, Pxy, etc. in the appendix.
6.3. We do not have any good examples for viscosity Einstein. @dcarls0n is going to try to find something that we could include here.
Conclusions: @dcarls0n is going to write a conclusions section.
References: We should update the Allen and Tildesley reference so that it is their 2017 publication
We do not want to recommend any package in particular, LAMMPS or GROMACS. But we should probably mention some other tools. MDtraj is one that comes to mind. I do not think OpenMM has transport analysis tools. I am not sure about NAMD or ls1 mardyn. Does VMD have this capability? At any rate, we want to make it clear that these should not be used as a black box.
@ejmaginn is going to work on manuscript for the next week. He will post additional issues/changes on GitHub. @dcarls0n will provide a conclusions section. Then @dcarls0n and @ramess101 will review the manuscript again. The goal is to have this submitted to the NIST internal review in 2-3 weeks.