Recent Changes - Search:
Main /

InterpretingTLVMieResults

The results of TLVMie simulations submitted with the tlvmie-simulation software package are output to compute_dir/jobname/EMD/csvdata.

The following files contain results which may be of interest to the user:

results*.txt - Contains summarized results of the simulation. The last line contains the predicted viscosity and the (estimated) uncertainty (σ) of that prediction.

visc_*.html - Contains a plot of the Green-Kubo integrals of each simulation replicate (gray lines), the average Green-Kubo integral across all replicates (blue line), and the Green-Kubo integral of several bootstrapped selections (orange lines). The DIPPR line (green) simply represents the value predicted by DIPPR's Equation 101 whose A, B, C, and D parameters are defined in tlvmie-simulation/resources/compound_lookup.csv

hist_*.png - Contains a histogram of the bootstrapped statistics across the simulation replicates. The results of the bootstrapping analysis are fit to a normal distribution, where the σ is an estimation of the uncertainty of the viscosity prediction.

The TLVMie simulation tool determines a predicted viscosity as outlined in previous work (10.1016/j.fluid.2022.113522 and 10.1016/j.fluid.2022.113681):

  1. Density is determined from the equilibrated system volume of the N replicate NPT simulations. This density is used to set the volume of the NVT (EMD or viscosity) portion.
  2. N replicate NVT simulations (sometimes referred to as EMD in the code) are run and pressure tensor data is output at regular intervals
  3. After completion of these simulations, the pressure tensor data is compressed and then processed to calculate the autocorrelation function over the whole simulation.
  4. The autocorrelation function is integrated and multiplied by several constants to produce the Green-Kubo integral for each simulation replicate.
  5. The average Green-Kubo integral is calculated by averaging over the Green-Kubo integals of all simulation replicates.
  6. The average Green-Kubo integral is fit to a double exponential function, weighted by the inverse of the standard deviations of the replicate GK integrals.
  7. The predicted viscosity is determined as the long-time limit of this double exponential function fit.
  8. Uncertainty is estimated by bootstrapping the statistics by selecting (with replacement) several thousand subsets of the simulation replicates and finding an average GK integral and predicted viscosity for each subset. These subset predicted viscosities are plotted in a histogram (hist*.png) and fit to a normal distribution to estimate uncertainty.
  9. Predicted viscosity and uncertainty are reported at the bottom of results*.txt

These processes are all automated in the tlvmie-simulation software package presented in this section.