Fig. 3: Atomistic DMPC lipids flip bilayers through water-filled nanopores.

a Time-average of \({\xi }_{{\mbox{P}}}\) during the MC chains, comparing sampler starting from the tunnel mechanism, (dry, \({\varPi }_{{\mbox{T}}}\), red), with those starting with a pore (wet, \({\varPi }_{{\mbox{P}}}\), blue). b Efficiency \(\eta\) measured by difference of expected and generated TPs,\(\,\Delta n={n}_{\exp }-{n}_{{\mbox{gen}}}\), and by the simulation time \({T}_{{\mbox{TP}}}\) of new transition events compared to the total simulation time \({T}_{{\mbox{all}}}\). c Accuracy of committor models correlating the vertical lipid displacement \(z\) to bead-to-bead distances \({{{\mathbf{\Delta }}}{{\bf{r}}}}^{{\mbox{all}}}\), training on all data (bright), compared to only \({\varPi }_{{\mbox{P}}}\) (dark). Boxes show median and \(25/75\)^th percentile of 800 bootstraps drawn from a total of 3500 (1418 in \({\varPi }_{{\mbox{P}}}\)) MC steps; whiskers show \(2.5/97.5\)^th percentile. Random committor assignments: rand. d–h Distribution of committor estimates for a given feature, comparing \(z\) (d, e) and a linear combination of distances (f, g), \({{{\mathbf{\Delta }}}{{\bf{r}}}}^{{\mbox{all}}}\cdot {{{\bf{v}}}}_{{{\rm{r}}}}\), stratified to the \({\varPi }_{{\mbox{P}}}\) (d, f) and \({\varPi }_{{\mbox{T}}}\) (e, g) data, and the pore coordinate \({\xi }_{{\mbox{P}}}\) (h). i Projection of the TPE onto \(z\) and \({\xi }_{{\mbox{P}}}\). Black iso-lines show the committor averaged over \(5000\) nearest-neighbors (\(0.6\)% of all data). Representative configurations are shown in the four side-panels I–IV.