DRIVE

Active Brownian Particles — What Is Self-Propulsion?

An ordinary colloidal particle in water is passive: it jiggles randomly because thermal fluctuations kick it, but it has no preferred direction of motion. An active Brownian particle is different — it continuously converts stored energy (fuel, light, chemical gradient) into directed self-propulsion. It moves at speed v₀ along a heading θ, which drifts by rotational diffusion at rate D_r.

Biological examples are everywhere: E. coli propels itself at ~30 μm/s with flagella and tumbles to change direction; sperm cells swim at ~100 μm/s; Janus particles (half-coated platinum beads in H₂O₂) self-propel at 1–10 μm/s via asymmetric catalysis. All of them share the same minimal physics: directed motility plus orientational noise.

Equations of motion (overdamped Langevin dynamics, 2D):

  dx/dt = v₀ cos θ + Fₓ(repulsion)
  dy/dt = v₀ sin θ + Fᵧ(repulsion)
  dθ/dt = ξ(t)

where ξ(t) is Gaussian white noise: ⟨ξ(t)ξ(t')⟩ = 2Dᵣ δ(t−t')

Persistence length (average distance before direction randomizes):
  ℓₚ = v₀ / Dᵣ

At v₀ = 3.2, Dᵣ = 0.016:  ℓₚ = 200 px  (≈14 disk diameters)
At v₀ = 0.8, Dᵣ = 0.080:  ℓₚ =  10 px  (≈1 disk diameter, nearly passive)

The Mechanism — Why Clustering Needs No Attraction

The clustering in Drive is not driven by any force between particles. It emerges from a simple feedback loop that Tailleur & Cates (2008) and Fily & Marchetti (2012) identified as the core of MIPS:

Step 1 — Slowdown in crowded regions. When a self-propelled disk collides with a neighbour, steric (contact) forces push them apart, but both particles continue trying to self-propel into each other. The net result is that the disks’ effective translational speed decreases in regions where many neighbours are present — not because of any explicit slowdown rule, but because collisions dissipate directed momentum.

Step 2 — Accumulation in slow regions. A particle moving slower spends longer inside a given volume. The local particle density ρ in a region of effective speed v_eff(ρ) is governed by the exact continuity equation: ρ × v_eff(ρ) = constant flux. Where speed is low, density is high — the same reason traffic jams form on motorways.

Step 3 — Positive feedback. High density → slow speed → even higher density. The system is spinodally unstable: any small density fluctuation amplifies instead of damping. The instability criterion is that d/dρ [ρ v_eff(ρ)] < 0 — the product of density and speed must decrease with density. Past this point, the uniform gas is no longer a stable state, and a macroscopic dense cluster nucleates.

MIPS instability criterion (Cates & Tailleur, Annual Review 2015):

  The uniform phase is unstable when:
    ∂/∂ρ [ ρ v_eff(ρ) ] < 0

  i.e. the effective current decreases with increasing density.

  For active hard disks, v_eff(ρ) ≈ v₀ (1 − a·ρ)  where a depends
  on contact geometry. Substituting:
    ρ v_eff(ρ) ≈ v₀ ρ (1 − aρ)

  This has maximum at ρ* = 1/(2a). Above ρ*, the phase is unstable.

  Packing fraction of maximum: φ* = πR²ρ* ≈ 0.25–0.35 for hard disks.
  In Drive (φ ≈ 0.25): the system sits near this maximum — MIPS-accessible
  with sufficient activity v₀.

The Phase Diagram — Activity vs Density

MIPS has a 2D phase diagram in the space of packing fraction φ and (dimensionless) Péclet number Pe = v₀ / (Dᵣ σ), where σ = 2R is the particle diameter. Pe measures how many diameters a particle travels before its heading randomizes.

Below a critical Pe_c ≈ 40 (at intermediate φ), the system is uniform gas at all densities. Above Pe_c, a coexistence region opens between a dilute gas phase φ_gas and a dense liquid phase φ_liq. The coexistence region widens as Pe increases — at high activity the dilute gas becomes nearly empty and the dense cluster approaches close-packing. The transition has the character of a first-order equilibrium phase transition, including nucleation, coexistence, and surface tension between the two phases.

Approximate phase boundary (Redner, Hagan, Baskaran 2013):

  Pe_c(φ) ≈ 40 + 80 φ   for φ ∈ [0.10, 0.45]

  Pe = v₀ σ / Dᵣ  = v₀ × 14 / Dᵣ

Drive presets:
  Gas Phase:    v₀=0.8, Dᵣ=0.08 → Pe =  140, φ=0.25 → inside stable region
  Critical:     v₀=2.2, Dᵣ=0.03 → Pe = 1027, φ=0.25 → onset of coexistence
  MIPS Cluster: v₀=3.8, Dᵣ=0.012→ Pe = 4433, φ=0.25 → deep inside coexistence

Note: Drive uses σ = 14px, L = 500px, N = 400 → φ = N·πR²/L² ≈ 0.246.

Why MIPS Is Not Equilibrium Phase Separation

An equilibrium liquid-gas transition (like water boiling) also shows phase coexistence — a dense liquid phase and a dilute vapor phase can coexist at the right temperature and pressure. But MIPS is fundamentally different in three ways:

1. No attractive forces. In water, the gas condenses because molecules attract each other via van der Waals forces. In MIPS, the inter-particle potential is purely repulsive (hard-core or Weeks-Chandler-Andersen). The “cohesion” is kinetic, not energetic: slow particles accumulate by a purely hydrodynamic mechanism. A passive system with the same interactions and density would remain a uniform gas forever.

2. Far from thermodynamic equilibrium. MIPS violates detailed balance — the probability of the forward trajectory is not equal to the reverse. The dense cluster continuously has active particles trying to push in from the gas phase and others escaping. It is a non-equilibrium steady state, not a free-energy minimum. No Boltzmann weight, no partition function, no free energy description of the coexistence.

3. Activity is the control parameter. In equilibrium, temperature drives the transition. In MIPS, the activity Pe drives it. Turning off self-propulsion (v₀ → 0) immediately dissolves the cluster, because there is no equilibrium force binding the particles. In Drive, slide v₀ to zero and watch the cluster scatter in seconds.

Real-World Applications

MIPS is not just a theoretical curiosity — it appears throughout biology and materials science wherever self-propelled entities are dense enough:

Bacterial biofilm initiation. Pseudomonas aeruginosa and other swimming bacteria spontaneously form dense clusters when population density exceeds a threshold, even in the absence of quorum-sensing signals. The initial aggregation stage closely matches MIPS predictions: cluster size distribution follows the expected active spinodal decomposition kinetics. Biofilm-associated infections are responsible for 80% of chronic infections and are 1,000× more resistant to antibiotics than planktonic cells.

Cell clusters in epithelial tissue. Motile cells in dense epithelial sheets show spontaneous jamming and clustering transitions that have been modelled as MIPS analogues. Cell division and apoptosis can be spatially biased by the activity-driven density fluctuations. The transition from motile individual cells to jammed collective behaviour is relevant to wound healing and cancer invasion.

Synthetic Janus microswimmers. Platinum-silica Janus particles half-coated with platinum self-propel in H₂O₂ at 1–10 μm/s and show clear MIPS-like clustering above critical density and activity. These systems are tunable — the propulsion speed can be controlled by H₂O₂ concentration — making them ideal experimental tests of the theory (Palacci et al., Science 2013; Buttinoni et al., Physical Review Letters 2013).

Active granular matter. Vibrated granular disks with asymmetric shape self-propel under shaking and cluster in the same way. This provides an athermal (T=0), purely mechanical realisation of MIPS that can be filmed with a camera and analysed directly without particle tracking algorithms.

The Persistence Length and Why It Controls Everything

The key dimensionless number is the Péclet number Pe = ℓ_p/σ = v₀/(D_rσ) — the persistence length measured in particle diameters. For MIPS to occur, a particle must be able to travel at least one diameter before its heading randomizes. If ℓ_p << σ (low Pe), the particle behaves like a passive Brownian sphere — isotropic, no MIPS.

In E. coli, the persistence length during a run is ≈ 30–50 μm (typical cell size 2 μm → Pe ≈ 15–25). Dense suspensions of E. coli exhibit MIPS-like dynamic clusters visible under dark-field microscopy. In Janus microswimmers, Pe can exceed 100 and the dense-cluster phase becomes very robust.

Mean-squared displacement of an active Brownian particle (ABP):

  ⟨r²(t)⟩ = 4D_T t + 2v₀² [ t/Dᵣ − (1−e^{−Dᵣt})/Dᵣ² ]
                              ↑                      ↑
                     ballistic regime           crossover
                     (t << 1/Dᵣ)             (t ~ 1/Dᵣ = τᵣ)

Short times (t << τᵣ): ballistic, ⟨r²⟩ ≈ v₀² t²
Long times  (t >> τᵣ): diffusive, ⟨r²⟩ ≈ 4(D_T + v₀²/2Dᵣ) t

Effective diffusivity D_eff = D_T + v₀²/(2Dᵣ)
  At v₀=3.2, Dᵣ=0.016: D_eff ≈ 320 px²/step — 320× faster than pure diffusion.

This enhanced diffusivity is what drives particles to quickly sample the
coexistence region and nucleate a cluster on timescales of seconds.

Implementation

Drive simulates N = 400 active Brownian disks on a 500×500 periodic (toroidal) domain. Each frame runs 5 substeps of overdamped Langevin dynamics, followed by one local-density computation for coloring. The total per-frame cost is approximately 370,000 floating-point operations — well within a single JavaScript frame budget at 60 fps.

Algorithm per substep:

  1. Pairwise forces (O(N²), N = 400 → 79,800 pairs):
     for each pair (i, j):
       dx, dy = position difference with minimum-image periodic BC
       if |r| < σ = 2R:  fᵢ += K (σ−|r|)/|r| · r̂,  fⱼ −= same

  2. Position + heading update:
     xᵢ += (v₀ cos θᵢ + fxᵢ) · dt  (mod L)
     yᵢ += (v₀ sin θᵢ + fyᵢ) · dt  (mod L)
     θᵢ += √(2Dᵣ dt) · N(0,1)        ← Box-Muller Gaussian

  3. Local density (per frame, not per substep):
     for each pair (i, j):
       if |r| < 2.2σ:  nc[i]++, nc[j]++

Rendering:
  Pass 1 (screen blend, α = 0.18): large soft disks (radius 2.6R) → cluster glow
  Pass 2 (source-over): main disks colored by nc:
    nc = 0 → rgb(38, 180, 172) = teal
    nc ≥ 11 → rgb(245, 165, 35) = amber
  globalAlpha = 0.55 + 0.42 × (nc/11) → dense particles brighter

Periodic boundary: minimum-image convention (dx > L/2 → dx −= L).
Repulsion: harmonic spring F = K(σ − d)/d, K = 30. No cutoff needed
  since the O(N²) loop only evaluates √d when d < σ.

The two-pass rendering (glow then solid) creates the visual: dilute teal gas particles appear dim and cool, while the dense amber cluster glows from overlapping screen-blended halos — an accurate metaphor for the physics, where the cluster is the high-energy, high-density, active phase.