Residence-Time Approach in Diffusivity Estimation

Updated 9 July 2026

Residence-Time Approach is a transport-analysis method that quantifies how long trajectories remain within defined spatial intervals to infer local diffusivity.
It estimates position-dependent diffusion by measuring first-exit times from finite intervals, avoiding the need for noisy time-correlation integrations.
The method is optimized for drift-free ABF simulations where carefully chosen interval widths balance statistical robustness and local constancy.

The Residence-Time Approach (RTA) is a transport-analysis methodology that infers dynamical properties from the time trajectories remain within prescribed states or spatial regions. In the molecular-simulation formulation introduced for position-dependent diffusivity estimation, the method uses first-exit statistics from finite intervals along a transport coordinate $z$ to recover a local diffusivity profile $D(z)$ from biased molecular dynamics trajectories. Its defining regime is one in which the reduced dynamics along $z$ are approximately drift-free and locally diffusive, as realized with adaptive biasing force (ABF) simulations; under those conditions, the average time spent in an interval before exit is directly related to the local diffusion coefficient (Thomas et al., 2 Apr 2026). Within a wider scientific literature, closely related residence-time constructions also appear in random-walk theory, lattice gases, hydrology, heliospheric transport, and stochastic flight processes, but the recent RTA is distinctive in using interval mean first-exit times as a direct diffusivity estimator rather than correlation-function integrals or harmonically restrained fluctuations (Thomas et al., 2 Apr 2026).

1. Definition and governing idea

In the RTA, the transport coordinate is partitioned into finite intervals

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$

For each interval, one measures the first-exit time

$T=\inf\{t\ge 0: z(t)\notin \Omega\},$

that is, the time required for a trajectory that starts inside the interval to leave it for the first time. The corresponding mean first-exit time (MFET) for a starting point $z_0\in(a,b)$ is

$\tau(z_0)\equiv \langle T\mid z(0)=z_0\rangle.$

If the interval is sufficiently small that $D(z)$ is approximately constant and the effective drift is negligible, then the interval-averaged MFET becomes a direct proxy for the local diffusivity (Thomas et al., 2 Apr 2026).

The central identity is the residence-time formula

$\tau_r=\frac{L^2}{12D},$

obtained by averaging the one-dimensional drift-free MFET over uniformly distributed starting points in an interval of width $L$ . This leads to the estimator

$D(z)$ 0

where $D(z)$ 1 is the measured mean residence time in interval $D(z)$ 2 (Thomas et al., 2 Apr 2026).

The method is therefore a first-passage-time estimator of diffusivity. Unlike fluctuation-based approaches, it does not require dedicated harmonically restrained simulations or numerical integration of noisy time-correlation functions. The observable is instead the directly measured time a trajectory remains inside a finite spatial bin before escape (Thomas et al., 2 Apr 2026).

2. Mathematical formulation

The projected dynamics along $D(z)$ 3 are described in the paper by the Smoluchowski equation

$D(z)$ 4

with $D(z)$ 5 the PMF and $D(z)$ 6 the position-dependent diffusivity. In the drift-free regime, the reduced equation simplifies to

$D(z)$ 7

For an interval $D(z)$ 8, if $D(z)$ 9 is treated as constant within the interval, the MFET solves

$z$ 0

with solution

$z$ 1

Under stationary flat-PMF sampling, the starting-point density is uniform, $z$ 2, and averaging over $z$ 3 gives

$z$ 4

This relation is the formal basis of the RTA diffusivity estimator (Thomas et al., 2 Apr 2026).

Three approximations are explicit in the formulation. First, the diffusivity must be approximately constant within each interval, $z$ 5. Second, the effective drift must be negligible, so that residence times are governed by diffusion in a nearly flat free-energy landscape rather than by drift-diffusion competition. Third, the projected coordinate must behave approximately Markovianly over the interval width and over the timescale of the first-exit process (Thomas et al., 2 Apr 2026).

These conditions delimit the method’s validity. If substantial residual drift remains, the diffusion-only MFET equation no longer applies, and the estimator is biased. If intervals are too wide, local constancy of $z$ 6 and drift suppression both degrade. If intervals are too narrow, sampling noise increases. The paper treats interval width as a tuning parameter balancing resolution against statistical robustness (Thomas et al., 2 Apr 2026).

3. Realization in biased molecular dynamics

The RTA was formulated for biased molecular dynamics trajectories produced by ABF. In that setting, ABF applies an on-the-fly bias potential $z$ 7 such that

$z$ 8

so that the effective free-energy profile

$z$ 9

becomes nearly flat across the sampled region. ABF is therefore not used to compute $\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 0 directly; it is used to create the drift-suppressed regime in which the residence-time identity is applicable (Thomas et al., 2 Apr 2026).

The operational workflow is explicit. ABF mean forces were accumulated in $\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 1 bins, and the bias was ramped gradually until enough force samples had been collected. Residence-time analysis was performed only after the PMF had converged sufficiently, judged from the time evolution of the PMF. Once the PMF was stationary and symmetric within uncertainty, residual drift was treated as negligible on the scale of the residence-time intervals (Thomas et al., 2 Apr 2026).

For the residence-time analysis itself, intervals of width

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 2

were used in all systems. First-exit times were computed from contiguous trajectory segments that stayed inside an interval. For each starting frame in such a segment, the remaining time to exit was assigned as one sample. The mean residence time in interval $\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 3 was then estimated by

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 4

Because these samples are correlated, uncertainties were estimated by blocking analysis using the Flyvbjerg–Petersen approach automated via Jonsson’s procedure (Thomas et al., 2 Apr 2026).

The inferred diffusivity profiles can be coupled to permeability calculations through the inhomogeneous solubility–diffusion relation

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 5

This places the RTA within the standard PMF-plus-diffusivity framework for membrane permeation, but with a different route to the diffusivity input (Thomas et al., 2 Apr 2026).

4. Relation to fluctuation-based diffusivity estimators

The paper contrasts the RTA with fluctuation-based methods derived from equilibrium dynamics in harmonically restrained simulations. Two named reference classes are VACF-based methods and PACF-based methods. In the notation used there, these depend on correlation functions such as

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 6

with representative formulas including

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 7

for the VACF route and

$\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 8

for the PACF route (Thomas et al., 2 Apr 2026).

The RTA differs methodologically in several specific ways. It does not require harmonic restraint simulations, does not require selecting restraint strengths, does not depend on numerical integration or extrapolation of noisy correlation functions, and does not rely on zero-frequency or long-time-tail behavior. The transport observable is instead a finite-interval escape time, which the paper describes as direct and physically transparent (Thomas et al., 2 Apr 2026).

This difference is not purely algorithmic. A conceptual distinction is also implied. Fluctuation-based methods infer transport from local equilibrium fluctuations around restrained positions, whereas the RTA infers transport from finite-interval first-passage behavior under approximately flattened free energy. A plausible implication is that the RTA is especially natural when ABF trajectories already exist and one seeks a position-dependent transport coefficient without launching a separate family of restrained simulations (Thomas et al., 2 Apr 2026).

At the same time, the article’s formulation does not claim universal superiority. In the POPC and stratum-corneum systems, the different diffusivity estimators agreed in bulk water but diverged in membrane interiors. The paper also notes that no single diffusivity profile may perfectly reproduce all lag-time regimes if the projected coordinate retains memory effects (Thomas et al., 2 Apr 2026).

5. Validation and application domains

The RTA was assessed in three systems of increasing structural complexity: oxygen diffusion across a hexadecane slab, water permeation across a POPC bilayer, and permeation of water plus selected volatile organic compounds through a model stratum-corneum membrane (Thomas et al., 2 Apr 2026).

For oxygen diffusion across the hexadecane/water slab, the test is described as the cleanest because bulk reference diffusivities are available. The ABF PMF was shown to converge and become stationary. The resulting RTA diffusivity profile displayed plateau regions in the bulk water and hexadecane phases, and the inferred bulk diffusivities agreed with independently determined MSD-based bulk diffusion coefficients within statistical uncertainty. This serves as direct validation that the residence-time identity recovers correct diffusion constants in a case with an external reference (Thomas et al., 2 Apr 2026).

For water permeation across a POPC bilayer, the RTA was compared with VACF- and PACF-based profiles and with propagator calculations. All three methods agreed in bulk water. In the membrane interior, the RTA membrane-center diffusivity lay between the VACF and PACF estimates. The ABF PMF was reported to converge and to be consistent with earlier umbrella-sampling results (Thomas et al., 2 Apr 2026).

For the model stratum-corneum membrane, the system was described as ordered, heterogeneous, slow-relaxing, and compositionally complex. ABF was carried out in overlapping windows, and RTA analysis used only converged window segments. For water, the RTA diffusivity profile again lay between VACF and PACF in the membrane center. For acetone and 6-MHO, RTA and PACF were very similar, while VACF systematically differed. For water, the RTA gave especially good agreement with MD-derived propagators across the examined lag-time range (Thomas et al., 2 Apr 2026).

The strongest validation device in the paper is propagator-level validation. Given $\Omega_i=[z_i,z_{i+1}), \qquad L_i=z_{i+1}-z_i.$ 9 and $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 0, one solves the Smoluchowski equation numerically to obtain the propagator

$T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 1

then compares that predicted conditional distribution with the propagator measured directly from unbiased MD. This checks the joint consistency of the reduced stochastic model rather than only the plausibility of a local coefficient profile (Thomas et al., 2 Apr 2026).

6. Assumptions, limitations, and broader methodological context

The main technical limitation of the RTA is its reliance on negligible effective drift. Residual PMF gradients invalidate the drift-free MFET relation, and interval averaging then no longer isolates diffusivity alone. A second limitation is the assumption that $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 2 is approximately constant over the analysis interval. A third is approximate Markovianity of the projected coordinate. The interval width $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 3 consequently acts as a control parameter: intervals that are too narrow yield poor statistics, whereas intervals that are too wide compromise local constancy and drift suppression (Thomas et al., 2 Apr 2026).

A common interpretive issue in residence-time methods is that a numerical trajectory duration is not automatically the relevant physical average. This point is explicit in heliospheric SDE transport, where an equal-weight average of pseudo-particle exit times was argued to yield the average duration of pseudo-trajectories rather than the propagation time of the physical particles represented by those trajectories. The proposed correction weights trajectories by phase-space density, producing residence times of $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 4 days instead of $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 5 days for 6 MeV Jovian electrons (Vogt et al., 2020). A related paper further tied this weighting to adiabatic energy changes and the source spectrum, emphasizing that residence time is a probability-weighted expectation value rather than a raw simulation time (Vogt et al., 2021). This suggests a broader principle: residence-time observables inherit the sampling measure of the underlying transport model.

The phrase “residence time” also has several established meanings outside diffusivity estimation. In stochastic transport theory, it may denote the time spent in a spatial region,

$T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 6

with moments obtained by Feynman–Kac methods and related to equilibrium or collision densities (Zoia et al., 2011). In random-walk and lattice-gas settings, it may be a conditional first-passage time, such as the mean time to exit through the right boundary conditioned on reaching that boundary before the left one (Ciallella et al., 2017), or an on-site waiting time controlled by density correlations and shock motion in driven exclusion processes (Messelink et al., 2015). In hydrology, Water Residence Time is treated as the impulse response of a linear system, inferred by constrained deconvolution of rainfall and aquifer signals (Meresescu et al., 2018). In steady-state storage-throughput formulations, residence time is defined by

$T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 7

and this was extended from matter to energy to estimate about 56 days for Earth’s atmosphere and $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 8 yr for the Sun (Osacar et al., 2019).

Against this background, the molecular-simulation RTA can be viewed as one member of a broader family of residence-time methodologies. Its specific contribution is to use first-exit statistics from ABF-flattened trajectory segments to estimate $T=\inf\{t\ge 0: z(t)\notin \Omega\},$ 9 directly through