Main

Our system consists of large populations of colloids capable of self-propulsion and of sensing the orientation of their neighbours solely by means of hydrodynamic and electrostatic mechanisms. We take advantage of an overlooked electrohydrodynamic phenomenon known as Quincke rotation18,19 (Fig. 1a). When an electric field, E0, is applied to an insulating sphere immersed in a conducting fluid, above a critical field amplitude, EQ, the charge distribution at the sphere’s surface is unstable to infinitesimal fluctuations. This spontaneous symmetry breaking results in a net electrostatic torque, which causes the sphere to rotate at a constant speed around a random direction transverse to E0 (ref. 18). We exploit this instability to engineer self-propelled colloidal rollers. We use poly(methyl methacrylate) beads of radius a = 2.4 μm diluted in an hexadecane solution filling the gap between two conducting glass slides. Once the particles have sedimented on the bottom electrode, we apply a homogeneous electric field, and indeed observe them to start rolling at high speed (Fig. 1a). Isolated rollers, as we refer to these particles, travel in random directions (Fig. 1b). Their velocity, v0, is set by E0 and scales as (Fig. 1c and Supplementary Methods).

Figure 1: Single-roller dynamics.
figure 1

a, Sketch of the Quincke rotation and of the self-propulsion mechanisms of a colloidal roller characterized by its electric polarization, P, and superposition of ten successive snapshots of colloidal rollers. Time interval, 5.6 ms; scale bar, 50 μm. b, Probability distribution of the velocity vector (v||, v⊥) for isolated rollers: v|| corresponds to the projection of the velocity on the direction tangent to the racetrack (Fig. 2); v⊥ is normal to v||. The probability distribution involves more than 1.4 × 105 measurements of instantaneous speed. c, Roller velocity, v0, plotted versus the field amplitude, E0. Inset, versus . The black dots represent the maximum of the probability distribution. Error bars, 1 s.d.

PowerPoint slide

Full size image

To study the emergence of collective unidirectional motion, we electrically confine the roller populations in racetracks periodic in the curvilinear coordinate, s, that measures position along the track. Their width is 500 μm < W < 5 mm (Fig. 2a and Methods). During a typical 10-min-long experiment, millions of rollers travel over distances as large as 105–106 particle radii, which makes it possible to investigate exceptionally large-scale dynamics. At low area fraction, ϕ0, the rollers form an isotropic gaseous phase. They all move at the same velocity in random directions, as would an isolated particle (Fig. 2b and Supplementary Video 1). On increasing ϕ0 above a critical value, ϕc, we observe a clear transition to collective motion. A macroscopic fraction of the rollers self-organizes and its constituents cruise coherently in the same direction (Fig. 2c, d and Supplementary Videos 2–4). More quantitatively, we define a polarization order parameter, Π0, as the modulus of the time and ensemble average of the particle-velocity orientation. This parameter increases sharply with ϕ0 and has a slope discontinuity at ϕc = 3 × 10−3, revealing the strongly collective nature of the transition (Fig. 2e). Remarkably, ϕc is a material constant: it is independent of the electric field amplitude.

Figure 2: Transition to directed collective motion.
figure 2

a, Dark-field pictures of a roller population that spontaneously forms a macroscopic band propagating along the racetrack. E0/EQ = 1.39, ϕ0 = 10−2. Scale bar, 5 mm. b–d Close-up views. The arrows correspond to the roller displacement between two subsequent video frames (180 frames s−1). b, Isotropic gas. ϕ0 = 6 × 10−4. c, Propagating band. ϕ0 = 10−2. d, Homogeneous polar liquid. ϕ0 = 1.8 × 10−1. Scale bar, 500 μm. e, Modulus of the average polarization, Π0, plotted versus the area fraction, ϕ0. Collective motion occurs as ϕ0 exceeds ϕc = 3 × 10−3. ϕc is independent of E0. Error bars, 1 s.d. e|| (or e⊥) is the unit vector oriented along the tangent (or the normal) of the racetrack confinement.

PowerPoint slide

Full size image

For area fractions higher than but close to ϕc, small density excitations nucleate from an unstable isotropic state and propagate in random directions. After complex collisions and coalescence events, the system phase separates to form a single macroscopic band that propagates at a constant velocity, cband, through an isotropic gaseous phase (Figs 2a, c and 3a and Supplementary Videos 2). No stationary state involving more than a single band was observed even in the largest systems (10-cm long). The velocity cband is found to be very close to the single-particle velocity, v0, at the front of the band. The bands are coupled to a net particle flux: they are colloidal flocks travelling through an isotropic phase. Their density profile is strongly asymmetric, unlike the slender bands observed in dense motility assays7. This marked asymmetry is akin to that found in one-dimensional, agent-based models20. It might be promoted by the high aspect ratio of the confinement. The local area fraction at curvilinear coordinate s and time t, ϕ(s, t), increases sharply with s and then decays exponentially to a constant value, ϕ∞, which is very close to the critical volume fraction, ϕc (Fig. 3b). This behaviour is similar to that found in numerical simulations of the celebrated two-dimensional Vicsek model21,22. Remarkably, the bands have no intrinsic scale and their length, Lband, is set by particle-number conservation only. This result is readily inferred from Fig. 3c, which shows that the bands span a fraction of the racetrack that merely increases with ϕ0 regardless of the overall curvilinear length, L.

Figure 3: Propagating-band state.
figure 3

a, Spatiotemporal variations of the area fraction measured along the curvilinear coordinate, s, and temporal variations of the area fraction at s = 0.8L (white dashed line), where L is the overall length of the racetrack. b, Band shape plotted versus the rescaled curvilinear coordinate, s/L, for ϕ0 = 5.3 × 10−3 (dark blue), 7.8 × 10−3 (blue), 1.0 × 10−2 (cyan) and 1.5 × 10−2 (orange). c, The rescaled band length, Lband/L, increases with ϕ0 and is independent of L (white dots, L = 28 mm; grey dots, L = 50 mm; black dots, L = 73 mm). Error bars show the estimated error associated with the measurement of Lband. d, Π(s) plotted versus 1 − ϕ∞/ϕ(s). The black dots correspond to averages over 5,000 local measurements (grey dots). The red curve is the theoretical prediction. Error bars, 1 s.d.

PowerPoint slide

Full size image

Looking now at the local polarization, we observe that the colloidal flock loses its internal coherence away from the band front as Π(s, t) decays continuously to zero along the band. Quantitatively, ϕ(s, t) and Π(s, t) are related in a universal manner irrespective of both the particle velocity and the mean volume fraction (Fig. 3d). All our data collapse on a single master curve solely parameterized by the particle fraction, ϕ∞, away from the band: . As it turns out, this relation corresponds to particle-number conservation in a system where density and polarization waves propagate steadily at a velocity v0 (ref. 22 and Supplementary Methods). This observation unambiguously demonstrates that the band state corresponds to a genuine stationary flocking phase of colloidal active matter.

On further increasing the area fraction to more than ϕ0 ≈ 2 × 10−2, transient bands eventually catch up with themselves along the periodic direction and form a homogeneous polar phase (Fig. 2d and Supplementary Video 4) in which the velocity distribution condenses on a single orientation of motion (Fig. 4a, to be contrasted with the perfectly isotropic distribution for fractions less than ϕc in Fig. 1b). Conversely, the roller positions are weakly correlated, as evidenced by the shape of the pair-distribution function, which is similar to that found in low-density molecular liquids (Fig. 4b). We also emphasize that the density fluctuations are normal at all scales (Fig. 4c). This is experimental observation of a polar-liquid phase of active matter. The existence of a polar-liquid phase was theoretically established yet had not been observed in any prior experiment involving active materials. Until now, collective motion has been found to occur in the form of patterns with marked density, orientational heterogeneities or both7,10,13,14,16. Furthermore, in contrast with the present observations, giant density fluctuations are considered to be a generic feature of the uniaxially ordered states of liquids comprising self-propelled particles2,3,17. We resolve this apparent contradiction below and quantitatively explain our experimental observations.

Figure 4: Polar-liquid state.
figure 4

a, Probability distribution of the velocity vector (v||, v⊥) in the polar-liquid state. The probability distribution involves more than 3.2 × 107 measurements of instantaneous speed. b, Pair correlation function, g, of the particle position in the polar-liquid state. c, The variance of the number of colloids, ΔN2, scales linearly with the average number of colloids, N, counted inside boxes of increasing size. E0/EQ = 1.39, ϕ0 = 9.5 × 10−2.

PowerPoint slide

Full size image

From a theoretical perspective, the main advantage offered by the rollers is that their interactions are clearly identified. We show in Supplementary Methods how to establish the equations of motion of Quincke rollers interacting through electrostatic and far-field hydrodynamic interactions. They take a compact form both for the position ri and the orientation of the ith particle:

Here makes an angle θi with the x axis, and a dot denotes a time derivative. In dilute systems, the particle interactions do not affect their propulsion speed, yet the electric field and flow field compete to align the with them. This competition results in an effective potential, Heff, for the . At leading order in a/r

where A(r) is a positive function and thus promotes the alignment of the neighbouring rollers, I is the identity matrix, is the outer product of with itself, and a dot denotes tensor contraction. Importantly, A is dominated by a hydrodynamic interaction, which arises from a hydrodynamic-rotlet singularity screened over distances of the order of the chamber height23. The function B(r) is also short ranged and accounts for a dipolar repulsion. Conversely, C(r) is long ranged and decays algebraically as r−2 owing to another hydrodynamic singularity induced by the roller motion in confinement. This singularity is referred to as a source doublet24. Neither B nor C yields any net alignment interaction. If these two terms were neglected, our model built from the actual microscopic interactions would amount to the ‘flying xy model’ introduced on phenomenological grounds in ref. 25. We emphasize that Heff is independent of v0 and E0, and that it is not specific to the Quincke mechanism. Its form could have been deduced from generic arguments based on global rotational invariance.

We then use a conventional Boltzmann-like kinetic-theory framework to derive the large-scale equation of motion for the density, and the polarization fields22,25. In the present case, this approximation was fully supported by the weak positional correlations in all the three phases, as exemplified in Fig. 4b. The resulting hydrodynamic equations are shown in Supplementary Methods. At the onset of collective motion, the magnitudes of the terms arising from the long-range hydrodynamic interactions are negligible. We are therefore left with equations for ϕ and Π akin to those in the model of refs 2, 3. However, we explicitly provide the functional form of the transport coefficients introduced on phenomenological grounds in ref. 2. Accordingly, we find that the competition between the polar ordering (induced by the short-range hydrodynamic interactions) and rotational diffusion yields a mean-field phase transition between an isotropic state and a macroscopically ordered state (Supplementary Methods). The phase transition occurs above a critical fraction, ϕc, that does not depend on the particle velocity (that is, on E0), in agreement with our experiments: collective motion chiefly stems from hydrodynamic interactions between the electrically powered rollers. However, at the onset of collective motion (that is, for ϕ0 > ϕc), the homogeneous polar state is linearly unstable to spatial heterogeneities. Moreover, for ϕ0 > ϕc, the compression modes are unstable eigenmodes of the isotropic state, in agreement with the emergence of density bands observed in the experiments, all starting from a homogeneous state and an isotropic velocity distribution.

We also rigorously establish a kinetic theory for the strongly polarized state reached for (Supplementary Methods). In this regime, the short-range electrostatic repulsion matters, causing the density fluctuations to relax and stabilizing the polar-liquid state. In addition, the long-range hydrodynamic interactions further stabilize the system by damping the modes of Π with non-zero divergence, that is, the splay modes that couple orientation disturbances to density fluctuations24. As a result, the giant density fluctuations26 are suppressed, in agreement with our unanticipated experimental findings (Fig. 4c and Supplementary Methods). We stress here that these long-range hydrodynamic interactions do not depend at all on the propulsion mechanism at the individual level. They solely arise from the confinement of the fluid in the z direction24. They are therefore not specific to the Quincke propulsion mechanism.

The only way to destroy the robust polar-liquid phase is to prevent it geometrically by eliminating the angular periodicity of the confinement in the curvilinear coordinate. In rectangular geometries with large enough aspect ratios, we observe that the bands never relax but rather bounce endlessly against the confining box (Supplementary Video 5). In confinement with an aspect ratio of order one, the band state is replaced by a single macroscopic vortex (Supplementary Video 6).

We have engineered large-scale populations of self-propelled particles from which collective macroscopic polar motion emerges from hydrodynamic interactions at exceptionally small densities. We believe that control over their interactions, and the ease with which they can be confined in custom geometries, will extend the present framework of active matter to include collective motion in more complex environments relevant to biological, robotic and social systems.

Methods Summary

We use commercial poly(methyl methacrylate) colloids (Thermo Scientific G0500; 2.4-μm radius), dispersed in a 0.15 mol l−1 AOT/hexadecane solution. The suspension is injected into a wide microfluidic chamber made of double-sided Scotch Tape. The tape is sandwiched between two ITO-coated glass slides (Solems, ITOSOL30; 80-nm thick). To achieve electric confinement, an additional layer of Scotch Tape including a hole with the desired geometry is added to the upper ITO-coated slide. The holes are made with a precision plotting cutter (Graphtec Robo CE 5000). The gap between the two ITO electrodes, H = 220 μm, is constant over the entire chamber. The electric field is applied by means of a voltage amplifier (Trek 606E-6). The colloids are observed with a Nikon AMZ1500 stereomicroscope (×1 magnification) equipped with a dark-field illuminator, and with a Zeiss Axiovert microscope (×10 objective) for local measurements. In both case, high-speed videos are taken with CMOS camera (Basler Ace) at frame rates between 70 and 900 frames per second. The particles are detected to a precision of one pixel by locating the intensity maxima on the experimental pictures. The particle trajectories are reconstructed using a conventional tracking code27.