Skip to main content
. Author manuscript; available in PMC: 2020 Jun 9.
Published in final edited form as: J Comput Aided Mol Des. 2020 Jan 27;34(5):601–633. doi: 10.1007/s10822-020-00290-5

Figure 3. Comparison of bidirectional and unidirectional free energy estimators of the same nonequilibrium work switching data.

Figure 3.

Average free energy estimates obtained by different estimators from the same nonequilibrium work data collected for CB8-G3 (left), OA-G3 (center), and OA-G6 (right) as a function of the number of energy/force evaluations. The average and the 95% t-based confidence interval (shaded areas) are computed from the 5 replicate calculations. BAR and BAR-long correspond to the GROMACS/NS-DS/SB and GROMACS/NS-DS/SB-long submissions in Figure 2, and utilize the bidirectional Bennett acceptance ratio estimator based on the Crooks fluctuation theorem [101]. Jarzynski-Forward/Reverse are the free energy estimates computed through unidirectional estimators derived from the Jarzynski equality using only the nonequilibrium work values accumulated in the forward/reverse direction respectively. The Gaussian-Forward/Reverse trajectories are based on the Crooks fluctuation theorem and the assumption of normality of the forward/reverse nonequilibrium work distribution, as described in [102]. Unidirectional estimators can introduce significant instabilities and bias in the estimates.