About the Validation Effort

The development of the TraPPE force field began in the mid 1990’s with Ilja Siepmann’s move to the University of Minnesota. The initial target was a united-atom model for linear alkanes (TraPPE 1). Since that time, TraPPE has been extended to include several different families, ranging from the widely used TraPPE-United Atom force field to newer extensions like TraPPE-Coarse Grain. During the same time span, simulation methods and computing power have progressively improved. We believe it is of interest to the TraPPE user community to validate the accuracy of early TraPPE models and to provide more precise simulation data using optimized simulation protocols, larger system sizes, and additional temperatures.

Toward that end, the TraPPE validation effort consists of creating a database of new simulation data (vapor-liquid coexistence densities, vapor pressures, and critical properties) for each model developed prior to 2011. The validation simulations adhere to the following standards:

• Several (8 or more) separate simulations will span state points along the vapor-liquid coexistence curve for a given molecule.
• Data for each state point will be calculated using coupled-decoupled configurational-bias Monte Carlo simulations in the NVT-Gibbs ensemble with the averages and standard errors of the mean estimated from 8-16 independent trajectories.
• At least four simulations will be carried out at temperatures higher than 0.9 T_crit, which will then be used to estimate the critical point.
• The lowest-temperature simulation will fall below the normal boiling point.
• The number of particles, N, will be set to maintain a liquid-phase box length larger than 32 Å, though a larger number of particles may be used for T > 0.9 T_crit. (A strict minimum of N = 200 will also be enforced.)
• The system volume will be adjusted to yield on average about 10-20% of the molecules in the vapor phase for T < 0.9 T_crit and approach an even phase ratio as T_crit is approached.
• The length of the simulations will be adjusted with the goal of achieving relative standard errors of the mean (RSEM) less than the following target values:
  • liquid densities, RSEM < 0.5% for T < 0.9 T_crit
  • vapor pressures, RSEM < 2% for N < 0.9 T_crit
  • critical temperature, RSEM < 1%
  • critical pressure, RSEM < 5%
• Density histogram analysis will be used to determine liquid and vapor densities at temperatures near the critical point where box identity switches may occur. For more details see: M. Dinpajooh, P. Bai, D. A. Allan, and J. I. Siepmann 'Accurate and precise determination of critical properties from Gibbs ensemble Monte Carlo simulations,' J. Chem. Phys. 143, 114113 (2015).
• Best practices will be used for setting move probabilities for optimal efficiency. For more details, please see: A. D. Cortes-Morales, I. G. Economou, C. J. Peters, and J. I. Siepmann 'Influence of simulation protocols on the efficiency of Gibbs ensemble Monte Carlo simulations,' Mol. Simul. 39, 1135-1142 (2013).

As TraPPE models are validated, the results and specific details about the simulation set-up will appear as part of the regular TraPPE webpage, in the Simulation Data section for a given model. To see the current results from the validation effort, please choose from the list of validated models in the search box above.