A universal scaling law for active diffusion in complex media Authors: Qun Zhang, Yuxin Tian, Xue Zhang, Xiaoting Yu, Hongwei Zhu, Ning Zheng, Luhui Ning, Ran Ni, Mingcheng Yang, Peng Liu Published: 2026 ABSTRACT Understanding how active particles transport in structurally heterogeneous environments is a fundamental and challenging problem, with relevance to biological and synthetic microswimmers in tissues and porous media. Here, using granular experiments and computer simulations, we investigate the long-time diffusion of active tracers in quasi-two-dimensional heterogeneous media. We show that diffusion-structure relations established for passive systems fail to describe active transport across different activity levels. To resolve this, we formulate a modified diffusion-structure relation by incorporating the dimensionless persistence length Q = vdτr/dt, which captures the activity-induced extension of the effective interaction range. The proposed relation yields a consistent collapse within both experimental and simulation datasets across active and passive tracers, diverse environmental structures, and propulsion mechanisms. Our results thus provide a universal predictive framework for transport in non-equilibrium heterogeneous systems. Introduction Active matter systems consist of self-driven units that convert stored or environmental energy into mechanical motion, including self-propelled particles1–5, contractile active systems such as actomyosin networks6,7, breathing particles8,9, and growing particles10,11. Biological entities include bacteria12,13, protozoa14,15, and spermatozoa16,17, whose propulsion is powered by molecular motor-driven flagella or cilia18,19. Synthetic microswimmers, by contrast, are typically actuated by chemical reactions20,21 or external fields such as light22,23, electromagnetic24,25, or acoustic stimuli26,27. Their long-time transport properties show promise in a wide range of applications, from targeted delivery of drugs, biomarkers, and contrast agents28–30, to the design of smart materials and microdevices31,32. In idealized structureless environments, an isolated active particle exhibits persistent ballistic motion at short times and transitions to an enhanced diffusion at long times due to stochastic reorientation33,34. In two-dimensional systems, the corresponding long-time diffusion coefficient is given by the well-established relation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{a}^{0}={D}_{t}^{0}+{v}_{d}^{2}{\tau }_{r}/2$$\end{document}Da0=Dt0+vd2τr/2, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{t}^{0}$$\end{document}Dt0 denotes that of the passive counterpart, vd the propulsion speed, and τr the rotational relaxation time33,35. This description applies to an isolated active particle and does not account for the role of environmental structures, which introduce a motility-dependent effective interaction range. In practice, however, active particles typically move through complex environments such as biological tissues, porous media, and microfluidic networks. Considerable efforts have therefore been devoted to elucidating active transport in both homogeneous and heterogeneous media33,36–38. Complex background structures are known to profoundly alter active dynamics, giving rise to phenomena such as subdiffusion and trapping39,40, dynamic freezing41,42, and chirality-dependent sorting43,44. These observations raise a fundamental question: how does the structure of complex environments govern active diffusion? For passive systems, extensive studies have established a universal diffusion-structure scaling that quantitatively links long-time normal diffusivity to the two-body structural entropy in both dynamic and quenched media45–49. With appropriate normalization, the diffusion coefficient D* collapses over several orders of magnitude onto a single master curve, demonstrating robust universality across diverse environments. In contrast, analogous diffusion-structure relations for active systems remain largely unexplored, except for the recent simulation study on dense active glass formers50. Experimental investigations that directly link transport to structural properties in heterogeneous active systems are also lacking. Whether a unified diffusion-structure relation exists for active probes in external heterogeneous environments remains an open question. In this work, combining granular experiments and particle-based simulations, we investigate the diffusion of active particles in various spatially heterogeneous environments, where obstacles are arranged in disordered structures as well as ordered square and hexagonal crystal lattices. We fabricate 3D-printed polylactide particles featuring tilted legs that allow the activity tuning via the inclination angle51,52. In simulations, self-propulsion is modeled using active Brownian motion33,53, run-and-tumble dynamics54,55, and active Ornstein-Uhlenbeck processes56,57, both overdamped and underdamped. We show that diffusion-structure relations established for passive systems fail to collapse active transport across different activity levels in heterogeneous environments. By incorporating activity through a dimensionless persistence length Q, we recover a consistent collapse of the normalized diffusion coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{ * }$$\end{document}Dmod* as a function of the two-body structural entropy S248,58, independent of the self-propulsion mechanism or the medium structure. This finding reveals a general active diffusion-structure relation and provides a predictive framework for long-time active diffusion in spatially heterogeneous environments. Results Experimental system and tracer dynamics Disk-like tracers were placed in a 54-cm-diameter circular vessel mounted on an electromagnetic shaker [Fig. 1(a)]. As shown by the short-time [Fig. 1(b)] and long-time trajectories [Fig. 1(c)], passive tracers undergo random motion, whereas active tracers exhibit persistent directed motion at short times that gradually randomizes at longer times, consistent with the dynamics of persistent active Brownian motion [Supplementary Movies 1 and 2]. In simulations, active tracers are propelled with a speed vd, leading to a long-time effective diffusivity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{a}^{0}={D}_{t}^{0}+{v}_{d}^{2}{\tau }_{r}/2$$\end{document}Da0=Dt0+vd2τr/2 (measured in a free background without obstacles)32, and reorient via three representative models: active Brownian particles (ABPs), run-and-tumble particles (RTPs), and active Ornstein-Uhlenbeck particles (AOUPs), which capture the stochastic dynamics of synthetic Janus colloids33,53, E. coli-like swimmers54,55, and particles driven by active baths56,57, respectively. Each model was implemented in both overdamped and underdamped forms, labeled as OABP, ORTP, OAOUP and UABP, URTP, UAOUP, respectively. Representative long-time trajectories are shown in Fig. 1(d).Fig. 1Experimental setup and tracer dynamics in complex environments.a Schematic of the quasi-2D electromagnetic shaker setup. The inset highlights 3D-printed tracer particles with symmetric (passive) and asymmetric (active) legs; propulsion speed vd is tuned via the leg inclination angle. b Short-time trajectories (0–0.7 s) in a disordered background, where the upper path corresponds to an active tracer (vd = 4.08 cm/s) and the lower path to a passive particle undergoing passive diffusion. c Long-time trajectories (0–70 s) in a hexagonal lattice: left, passive particle; right, active tracer (same as in (b)). d Simulated long-time trajectories of OABP, OAOUP, and ORTP tracers in a hexagonal lattice over 5 × 105 steps. In (b–d), trajectory color indicates time, from blue (start) to yellow (end). Experimental setup and tracer dynamics in complex environments. a Schematic of the quasi-2D electromagnetic shaker setup. The inset highlights 3D-printed tracer particles with symmetric (passive) and asymmetric (active) legs; propulsion speed vd is tuned via the leg inclination angle. b Short-time trajectories (0–0.7 s) in a disordered background, where the upper path corresponds to an active tracer (vd = 4.08 cm/s) and the lower path to a passive particle undergoing passive diffusion. c Long-time trajectories (0–70 s) in a hexagonal lattice: left, passive particle; right, active tracer (same as in (b)). d Simulated long-time trajectories of OABP, OAOUP, and ORTP tracers in a hexagonal lattice over 5 × 105 steps. In (b–d), trajectory color indicates time, from blue (start) to yellow (end). Diffusion and structural correlations In both experiments and simulations, the long-time dynamics of active tracers are characterized by the diffusion coefficient Da, extracted from the mean square displacements (MSDs) defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \Delta {{\bf{r}}}^{2}(t)\rangle=\langle {[{\bf{r}}({t}_{0}+t)-{\bf{r}}({t}_{0})]}^{2}\rangle$$\end{document}⟨Δr2(t)⟩=⟨[r(t0+t)−r(t0)]2⟩, where 〈 ⋅ 〉 denotes the ensemble average. Figure 2(a) shows log-log plots of MSDs for tracers with different activities, characterized by the Péclet-like parameter Q = vdτr/dt (with dt the tracer diameter)59, obtained experimentally in a hexagonal lattice with lattice constant L = 3 cm and background packing fraction ϕ = 0.09. Passive tracers undergo normal diffusion (〈Δr2(t)〉 ∝ tα, α = 1), while active tracers exhibit a crossover from the short-time superdiffusion (α > 1) to the long-time normal diffusion (α = 1). The long-time diffusion coefficient Da is extracted by fitting the MSDs with 〈Δr2(t)〉 = 4Dat. Figure 2(c) presents simulated MSDs for tracers with different activities Q in a hexagonal lattice at slightly higher packing fraction ϕ = 0.10, with Da obtained using the same fitting procedure.Fig. 2Diffusion and structural correlations of active tracers.a MSDs as a function of delay time t and (b) the pair correlation function g(r) for passive (Q = 0) and active tracers with Q = 0.29 and 1.90, measured experimentally in a hexagonal lattice with lattice constant L = 3 cm and background packing fraction ϕ = 0.09. Corresponding simulation results of (c) MSDs and (d) g(r) for tracers with activity Q = 0, 1.0, 3.0 and 6.0 in a hexagonal lattice at ϕ = 0.10. Source data are provided as a Source Data file. Diffusion and structural correlations of active tracers. a MSDs as a function of delay time t and (b) the pair correlation function g(r) for passive (Q = 0) and active tracers with Q = 0.29 and 1.90, measured experimentally in a hexagonal lattice with lattice constant L = 3 cm and background packing fraction ϕ = 0.09. Corresponding simulation results of (c) MSDs and (d) g(r) for tracers with activity Q = 0, 1.0, 3.0 and 6.0 in a hexagonal lattice at ϕ = 0.10. Source data are provided as a Source Data file. The structure of the system is characterized by the two-body structural entropy S2, which approximates the excess entropy, i.e., the entropy difference between a correlated system and an ideal gas under identical conditions. While the full excess entropy includes higher-order correlations, the two-body contribution S2 accounts for more than 85% of the total excess entropy across a broad density range, and has been widely used as an approximation in complex systems60. For a tracer in a matrix, S2 is computed from the pair correlation function g(r) between the tracer and background particles48,58: 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{2}=-\pi \rho {\int }_{0}^{\infty }\{g(r)ln[g(r)]-[g(r)-1]\}rdr,$$\end{document}S2=−πρ∫0∞{g(r)ln[g(r)]−[g(r)−1]}rdr, where ρ is the number density of background matrix particles. Given the low tracer concentration, tracer-tracer interactions are normally negligible, and only tracer-matrix correlations are considered. The function g(r), defined as the probability of finding a matrix particle at distance r from a tracer, is measured by 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g(r)=\frac{1}{2\pi r\rho }\left\langle \frac{1}{{N}_{t}}\mathop{\sum }\limits_{i=1}^{{N}_{t}}\mathop{\sum }\limits_{j=1}^{{N}_{l}}\delta (r-{r}_{ij})\right\rangle,$$\end{document}g(r)=12πrρ1Nt∑i=1Nt∑j=1Nlδ(r−rij), where Nt and Nl are the numbers of tracers and matrix particles, respectively, and rij is the distance between tracer i and matrix particle j. Experimentally measured g(r) in a hexagonal lattice is shown in Fig. 2(b). As the activity Q increases from 0 (passive) to 1.90 (active), the first peak of g(r), increases, indicating enhanced tracer-matrix correlations. The periodic peaks in g(r), arising from the crystalline order of the matrix, are similarly observed in simulations across different Q [Fig. 2(d)]. In both experiments and simulations, the two-body structural entropy S2 varies from sample to sample [Supplementary Note 1]. For disordered backgrounds, S2 decreases with increasing obstacle number density ρ, whereas for crystalline backgrounds, S2 depends on the lattice constant L. To sample a broad range of S2, we prepare multiple backgrounds by tuning ρ in disordered backgrounds and varying L in square and hexagonal lattices. Limits of diffusion-structure scaling In the following, the long-time diffusion coefficients of both passive (Dt) and active (Da) tracers are normalized as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{ * }=D{\Gamma }^{-1}{d}_{t}^{-2}$$\end{document}D*=DΓ−1dt−2, following the normalization widely used in previous studies45,47–49. The Enskog collision frequency Γ between the tracer and immobile matrix particles in 2D is given by refs. 48,613\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma=2{r}_{p}\rho g({r}_{p})\sqrt{\pi {k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}/m},$$\end{document}Γ=2rpρg(rp)πkBTeff/m, where rp is the position of the first peak in g(r) and m is the reduced mass, defined as m = 2mtml/(mt + ml), with mt and ml the masses of the tracer and background particles, respectively. Since matrix particles are fixed, they effectively form an infinite-mass rigid background, and we take the limit ml → ∞, yielding m ≈ 2mt. For passive tracers, the effective temperature \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{{\mathrm{eff}}}^{p}$$\end{document}Teffp is extracted from the average kinetic energy of tracers in the obstacle-free vibrated system62,63. For active tracers, it is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}^{a}={k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}^{p}({D}_{a}^{0}/{D}_{t}^{0})$$\end{document}kBTeffa=kBTeffp(Da0/Dt0), accounting for their enhanced long-time diffusivity under identical translational damping, where both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{a}^{0}$$\end{document}Da0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{t}^{0}$$\end{document}Dt0 are measured in the same obstacle-free background. Figure 3 (a) plots the normalized diffusion coefficients D* as a function of the two-body structural entropy S2 for tracers with varying activity in disordered and ordered backgrounds. The experimental data do not collapse onto a single curve; instead, each activity level follows its own trend. Moreover, the data at all activity levels deviate significantly from the Dzugutov-like scaling45,50, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{ * }=\beta {e}^{\alpha {S}_{2}}$$\end{document}D*=βeαS2, with β and α being fitting parameters. Particularly in the regime of small − S2 where structural correlations are weak, pronounced discrepancies between the experimental results and the Dzugutov-scaling are observed [Supplementary Note 2], indicating that the empirical exponential form becomes inadequate as it does not explicitly account for the underlying binary collisional dynamics.Fig. 3Breakdown and recovery of diffusion-structure scaling in experiments.a Experimentally measured dimensionless diffusion coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{*}=D{\Gamma }^{-1}{d}_{t}^{-2}$$\end{document}D*=DΓ−1dt−2 plotted against the two-body structural entropy S2 for passive and active tracers with different activities Q in disordered and ordered backgrounds. Solid lines represent fits to the original scaling relation of Eq. (4). b The same data replotted using the activity-modified normalized \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{*}=D{\left[\Gamma (1+Q){d}_{t}^{2}\right]}^{-1}$$\end{document}Dmod*=DΓ(1+Q)dt2−1. All data collapse onto a single master curve, with the solid violet line representing the fit to Eq. (5), demonstrating a unified scaling behavior across both passive and active tracers. Open, half-filled, and solid symbols correspond to disordered, square-lattice, and hexagonal-lattice backgrounds, respectively. Source data are provided as a Source Data file. Breakdown and recovery of diffusion-structure scaling in experiments. a Experimentally measured dimensionless diffusion coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{*}=D{\Gamma }^{-1}{d}_{t}^{-2}$$\end{document}D*=DΓ−1dt−2 plotted against the two-body structural entropy S2 for passive and active tracers with different activities Q in disordered and ordered backgrounds. Solid lines represent fits to the original scaling relation of Eq. (4). b The same data replotted using the activity-modified normalized \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{*}=D{\left[\Gamma (1+Q){d}_{t}^{2}\right]}^{-1}$$\end{document}Dmod*=DΓ(1+Q)dt2−1. All data collapse onto a single master curve, with the solid violet line representing the fit to Eq. (5), demonstrating a unified scaling behavior across both passive and active tracers. Open, half-filled, and solid symbols correspond to disordered, square-lattice, and hexagonal-lattice backgrounds, respectively. Source data are provided as a Source Data file. To elucidate the origin of these deviations, we decomposed the total friction experienced by the active tracer into three contributions: the effective Brownian friction ξS arising from the substrate, binary collisional friction ξB, and structural friction from local constraints ξρρ. The long-time diffusion coefficient can then be expressed as D = kBTeff/(ξS + ξB + ξρρ). Following the previously established dynamics-structure scaling relation, normalization by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma {d}_{t}^{2}$$\end{document}Γdt2 yields the scaling form48,494\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{ * }=\frac{{k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}}{\Gamma {d}_{t}^{2}{\xi }_{B}}\frac{1}{A(1-{S}_{2})+{\xi }_{S}/{\xi }_{B}},$$\end{document}D*=kBTeffΓdt2ξB1A(1−S2)+ξS/ξB, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\xi }_{B}=2{r}_{p}\rho g({r}_{p})\sqrt{\pi {k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}m}$$\end{document}ξB=2rpρg(rp)πkBTeffm64,65, and A is a fitting parameter. Since active and passive tracers interact with the same substrate and have nearly identical contact areas, the substrate friction coefficient, ξS, is assumed to be identical for both. It is obtained from the Einstein relation for passive diffusion in the absence of obstacles, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\xi }_{S}={k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}^{p}/{D}_{t}^{0}$$\end{document}ξS=kBTeffp/Dt0, an assumption commonly adopted in granular systems66,67. Interestingly, each activity level follows its own dynamics-structure scaling described by Eq. (4), demonstrating that long-time diffusivity is largely controlled by the two-body structural entropy S2 of the environment. However, D* systematically increases with tracer activity even at fixed S2, and fitting Eq. (4) requires activity-dependent values for the fitting parameter A. This suggests that although the underlying coupling between structural entropy and long-time dynamics persists in active systems and Eq. (4) remains valid within subsets of fixed activity, it fails to quantitatively capture the universal nature of nonequilibrium dynamics. Diffusion-structure scaling with persistence In particular, the tracer diameter dt reflects only the tracer’s geometric size but neglects the directional persistence unique to active motion. Unlike passive tracers, active particles typically travel a distance vdτr before reorientation, which effectively extends their interaction range along the propulsion direction from dt to dt + vdτr, thereby increasing the effective interaction area from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{t}^{2}$$\end{document}dt2 to (dt + vdτr)dt. To quantify this persistence effect, we introduce the dimensionless persistence length Q = vdτr/dt, a Péclet-like activity parameter widely used in active systems59, which increases with both propulsion velocity and reorientation time. This parameter captures the directional memory associated with persistent self-propulsion, enhancing tracer-obstacle coupling beyond purely thermal models. Incorporating this activity-induced length scale, we define a modified dimensionless diffusion coefficient, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{ * }=D{\left[\Gamma (1+Q){d}_{t}^{2}\right]}^{-1}$$\end{document}Dmod*=DΓ(1+Q)dt2−1, which smoothly recovers the passive limit as Q = 0. Substituting this into Eq. (4) leads to the generalized scaling form: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{ * }=\frac{{k}_{{\mathrm{B}}}{T}_{{\mathrm{eff}}}}{\Gamma (1+Q){d}_{t}^{2}{\xi }_{B}}\frac{1}{A(1-{S}_{2})+{\xi }_{S}/{\xi }_{B}}.$$\end{document}Dmod*=kBTeffΓ(1+Q)dt2ξB1A(1−S2)+ξS/ξB. The self-propulsion speed vd is extracted from particle trajectories, and the rotational relaxation time τr is obtained from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{a}^{0}={D}_{t}^{0}+{v}_{d}^{2}{\tau }_{r}/2$$\end{document}Da0=Dt0+vd2τr/2, using the long-time diffusion coefficient of the active (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{a}^{0}$$\end{document}Da0) and corresponding passive (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{t}^{0}$$\end{document}Dt0) tracers in the absence of obstacles32 [Supplementary Note 3]. As shown in Fig. 3(b), all experimental data—spanning varying propulsion speeds and matrix structures—collapse onto a single master curve, in quantitative agreement with the modified diffusion-structure relation Eq. (5) with a single fitting parameter A = 1. This indicates that the relation properly captures the underlying physical mechanism rather than merely reflecting a case-specific fitting, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{ * }$$\end{document}Dmod* is entirely determined by S2 and Q, with no discontinuities observed across the transition from passive to active diffusion. This establishes a unified scaling framework for long-time diffusion in heterogeneous media, bridging equilibrium and nonequilibrium systems via two experimentally accessible parameters. Universality across active-particle models Next, we further test the universality of the scaling relation through simulations of active colloidal tracers with diverse reorientation dynamics. Three representative models—ABP, AOUP, and RTP—are considered, each implemented under both underdamped and overdamped conditions to assess the effect of inertia. Figure 4(a) plots the dimensionless diffusion coefficient D* as a function of S2 for passive and active tracers in the UABP model. Like in Fig. 3(a), data points with different activities remain dispersed and do not collapse onto a single curve. Each trend is described by the original scaling relation Eq. (4), with an activity-dependent fitting parameter A. Passive tracers, shown in black, serve as a reference for comparison. Results for the other five active motion models are presented in the Supplementary Note 4. Remarkably, when the same data are replotted using the modified normalization in Fig. 4(b), all simulation results (across reorientation mechanisms, damping regimes, and matrix structures) collapse onto a single universal curve, in excellent agreement with the modified scaling relation, Eq. (5), with a single parameter A = 8 independent of active models. The larger fitting parameter compared to the experimental value can be attributed to the fact that in macroscopic granular experiments, the Einstein relationship is assumed, while particle-obstacle collisions are inelastic, and the noise is athermal. This demonstrates that the long-time diffusion of active tracers in complex media is controlled by the two-body structural entropy S2 and the dimensionless persistence length Q = vdτr/dt, which quantifies tracer activity, independent of the underlying stochastic process.Fig. 4Universality of diffusion-structure scaling across diverse active models.a Simulation results for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{*}=D{\Gamma }^{-1}{d}_{t}^{-2}$$\end{document}D*=DΓ−1dt−2 plotted as a function of S2 for passive and active tracers in the UABP model, in both disordered and ordered backgrounds. Solid lines indicate fits to the original scaling relation (Eq. (4)). Different symbols denote tracers with different activity strength Q ranging from 0 to 6.0. b The same data replotted using the activity-modified normalization \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{*}=D{\left[\Gamma (1+Q){d}_{t}^{2}\right]}^{-1}$$\end{document}Dmod*=DΓ(1+Q)dt2−1, together with results from five additional active-motion models over the same range of Q. All data collapse onto a single master curve, the solid violet line denotes the fit to Eq. (5), demonstrating a unified scaling across passive and active tracers as well as diverse microscopic reorientation dynamics. Open, half-filled, and solid symbols correspond to disordered, square-lattice, and hexagonal-lattice backgrounds, respectively. Source data are provided as a Source Data file. Universality of diffusion-structure scaling across diverse active models. a Simulation results for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}^{*}=D{\Gamma }^{-1}{d}_{t}^{-2}$$\end{document}D*=DΓ−1dt−2 plotted as a function of S2 for passive and active tracers in the UABP model, in both disordered and ordered backgrounds. Solid lines indicate fits to the original scaling relation (Eq. (4)). Different symbols denote tracers with different activity strength Q ranging from 0 to 6.0. b The same data replotted using the activity-modified normalization \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{{\mathtt{mod}}}^{*}=D{\left[\Gamma (1+Q){d}_{t}^{2}\right]}^{-1}$$\end{document}Dmod*=DΓ(1+Q)dt2−1, together with results from five additional active-motion models over the same range of Q. All data collapse onto a single master curve, the solid violet line denotes the fit to Eq. (5), demonstrating a unified scaling across passive and active tracers as well as diverse microscopic reorientation dynamics. Open, half-filled, and solid symbols correspond to disordered, square-lattice, and hexagonal-lattice backgrounds, respectively. Source data are provided as a Source Data file. Discussion Our work identifies a diffusion-structure relation for long-time normal diffusion of active tracers in complex media by incorporating directional persistence into a theoretical framework previously applicable only to passive systems. The transport is characterized by two key quantities: the structural entropy S2 and the dimensionless persistence length Q = vdτr/dt. This scaling provides a quantitative link between microscopic structure and macroscopic transport. Importantly, the observed scaling arises from long-time dynamics and is robust to microscopic details of interactions and reorientation dynamics, as demonstrated across experiments and multiple active-particle models. This theoretical framework highlights the key role of directional persistence in active diffusion within complex environments away from the dense ergodic-breaking regime, while intrinsically non-diffusive states lie beyond its scope. The newly established scaling relation opens several promising directions for future research, including tests of its applicability in thermal systems such as colloidal suspensions, comparisons of fitted parameters between thermal and athermal environments, and systematic investigations of how hydrodynamic interactions affect the scaling behavior. Extending this framework to living active systems, such as motile bacteria and ciliated microorganisms, may further bridge nonequilibrium statistical physics and biological transport processes. Methods Experiment As shown in Fig. 1(a), fixed polymethyl methacrylate (PMMA) cylinders with diameter dl = 1 cm and height h = 2 cm served as immobile obstacles, arranged in three types of backgrounds: disordered configurations with obstacle number densities 0.005-0.025 cm−2, and square and hexagonal lattices with lattice constants L = 3-8 cm. Disordered backgrounds were generated using a random placement algorithm with a minimum inter-obstacle separation exceeding 2 cm, ensuring that tracers move freely through the obstacle-obstacle gaps. Distinct disordered backgrounds were obtained by varying the obstacle number density. Vertical vibrations were applied with a magnitude of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Z=\zeta \sin (2\pi ft)$$\end{document}Z=ζsin(2πft) and frequency f = 80 Hz. The dimensionless acceleration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widetilde{\Gamma }}_{vib}=\zeta {(2\pi f)}^{2}/g=3.0$$\end{document}Γ~vib=ζ(2πf)2/g=3.0, with g the gravitational acceleration, maintaining stable quasi-2D dynamics. Tracers with diameter dt = 1.5 cm were 3D-printed from polylactide, equipped with either symmetric (passive) or asymmetric (active) legs [inset, Fig. 1(a)]51,52. For active tracers, the asymmetric leg geometry rectifies vertical vibrations into persistent horizontal motion. Propulsion speeds vd = 2.04, 2.72, 3.40, and 4.08 cm/s were tuned via the leg inclination angle68. Images of the samples were recorded for 12-15 min at 100 frames/s under both obstacle-free and obstacle-present conditions in each measurement. For each tracer activity, more than 30 independent measurements were performed across disorder, square-lattice and hexagonal-lattice backgrounds. In total, over 160 experimental measurements were conducted for five tracer activities, yielding independent estimates of the long-time diffusion coefficient D and the corresponding two-body structural entropy S2. Particle trajectories were extracted using particle tracking techniques69, excluding regions within 1/8 of the vessel diameter to avoid boundary effects. All physical quantities reported, including diffusion coefficients and structural entropies, were evaluated only after the system was confirmed to have reached a steady state [Supplementary Note 5]. Simulation Numerical simulations were performed in a 2D periodic box containing 1600 immobile matrix particles arranged in disordered, square, and hexagonal configurations, matching the experiments. Under the standard reduced simulation units, the matrix particle radius, tracer mass, and thermal energy kBT are taken as the units of length, mass, and energy, respectively. The tracer diameter is set to dt = 0.25dl (the matrix-particle diameter dl = 2.0) in all models. To examine tracer-size dependence, we systematically varied the tracer-to-matrix size ratio R = dt/dl in the OABP model (R = 0.25, 0.5, 0.75, 1.0, 1.25, and 1.5) while keeping dl fixed. The corresponding results [Supplementary Note 6] show that the fitting parameter A remains unchanged for all R, demonstrating that the proposed scaling relation is independent of tracer size, further confirming its universality. Tracer-matrix interactions were modeled using a truncated and shifted purely repulsive generalized Lennard-Jones potential, corresponding to a Weeks-Chandler-Andersen (WCA)-type interaction70, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U(r)=4\epsilon \left[{\left(\frac{\sigma }{r}\right)}^{2n}-{\left(\frac{\sigma }{r}\right)}^{n}\right]+\epsilon$$\end{document}U(r)=4ϵσr2n−σrn+ϵ, r < 21/nσ, with interaction diameter σ = (dt + dl)/2, interaction strength ϵ = kBT = 1.0, and stiffness parameter n = 12. To verify robustness, additional simulations of the OABP model were performed using the same generalized form with n = 6 [Supplementary Note 7], which corresponds to the standard 12-6 WCA potential. Matrix packing fractions were varied to access a wide range of structural entropies. Translational and rotational thermal noises were included as independent Gaussian-distributed forces and torques following the fluctuation-dissipation relation. The persistence length vdτr/dt reached up to 6, significantly exceeding the experimental maximum of 1.9 limited by vibrational constraints. We simulate active tracers under both underdamped and overdamped conditions using three representative models: active Brownian particles (ABPs)33,53, run-and-tumble particles (RTPs)54,55, and active Ornstein-Uhlenbeck particles (AOUPs)56,57. For ABPs, the translational dynamics depend on the damping regime. In the underdamped case, tracer motion obeys the Langevin equation, 6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m\dot{\boldsymbol{v}}={{\bf{F}}}_{r}+{{\bf{F}}}_{d}+{{\bf{F}}}_{G}-\gamma {\boldsymbol{v}},$$\end{document}mv°=Fr+Fd+FG−γv, where m = 1 is the tracer mass, γ = 100 is the translational friction coefficient, Fr is the tracer-matrix interaction force, Fd = γvd is the self-propelling force, and FG is a Gaussian stochastic force satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle {{\bf{F}}}_{G}\right\rangle=0$$\end{document}FG=0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle {{\rm{F}}}_{G}^{\alpha }(t){{\rm{F}}}_{G}^{\beta }({t}^{{\prime} })\right\rangle=2{k}_{{\mathrm{B}}}T\gamma {\delta }_{\alpha \beta }\delta (t-{t}^{{\prime} })$$\end{document}FGα(t)FGβ(t′)=2kBTγδαβδ(t−t′). The propulsion speed is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{v}}}_{d}={v}_{d}\cos \theta {\widehat{{\boldsymbol{e}}}}_{x}+{v}_{d}\sin \theta {\widehat{{\boldsymbol{e}}}}_{y}$$\end{document}vd=vdcosθe^x+vdsinθe^y, where θ is the orientation angle, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{{\boldsymbol{e}}}}_{x}$$\end{document}e^x and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{{\boldsymbol{e}}}}_{y}$$\end{document}e^y are the unit vectors along the x and y directions, respectively. The rotational dynamics are governed by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\ddot{\theta }={T}_{G}^{\theta }-{\gamma }_{r}\dot{\theta}$$\end{document}Iθ¨=TGθ−γrθ°, with moment of inertia \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I=\frac{1}{8}m{d}_{t}^{2}$$\end{document}I=18mdt2, the rotational friction coefficient γr, and Gaussian stochastic torque \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{G}^{\theta }$$\end{document}TGθ satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle {T}_{G}^{\theta }(t)\right\rangle=0$$\end{document}TGθ(t)=0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\langle {T}_{G}^{\theta }(t){T}_{G}^{\theta }({t}^{{\prime} })\right\rangle=2{k}_{{\mathrm{B}}}T{\gamma }_{r}\delta (t-{t}^{{\prime} })$$\end{document}TGθ(t)TGθ(t′)=2kBTγrδ(t−t′). In the overdamped case, the equations reduce to γv = Fr + Fd + FG and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{r}\dot{\theta}={T}_{G}^{\theta }$$\end{document}γrθ°=TGθ. The rotational relaxation time is defined as τr = γr/(kBT) and is tuned by varying γr. For RTPs, the translational dynamics are identical to those of ABPs, but particles undergo instantaneous random tumbles, with orientations reset at an average time interval τr. For AOUPs, the translational dynamics are the same as for ABPs in both underdamped and overdamped conditions, while the self-propulsion speed vd evolves according to an Ornstein-Uhlenbeck processes, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }_{r}\dot{{\boldsymbol{v}}}_{d}=-{{\boldsymbol{v}}}_{d}+\sqrt{2C}{\boldsymbol{\eta }}$$\end{document}τrv°d=−vd+2Cη, where C is the noise strength, and η denotes Gaussian white noise. Simulations were performed with a time step \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t=1{0}^{-3}\times {d}_{l}\sqrt{m/\epsilon }$$\end{document}Δt=10−3×dlm/ϵ. Each data point was averaged over eight independent trajectories, with 5 × 105 steps to relax the initialized system followed by 1.2 × 109 steps for data collection. For each active-motion model (OABP, ORTP, OAOUP, UABP, URTP, and UAOUP), more than 70 independent data points were obtained across different backgrounds, yielding over 440 simulation data points in total. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data Supplementary Movie 1 Supplementary Movie 2 Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data Supplementary Movie 1 Supplementary Movie 2 Transparent Peer Review file Source data Source Data Source Data