Vicsek Lab: Initialization
Module 2: Synchronization and Collective Behaviour
In this lab you implement the Vicsek model (Vicsek et al. 1995), validate it, and animate it. The rules are in The Vicsek Model.
The core of this lab is five pages: this one, Update Rule, Order Parameter, Validation and Animation. Vectorized Update, Interactive Animation and Predator Avoidance are optional extensions. Practice: Collective Motion closes the session.
To reproduce the main plot of \(\langle v_a \rangle\) against \(\eta\):
- Initialize random positions and random headings.
- At each time step, find the neighbors within radius \(r\) (with periodic distances), compute the average direction with
atan2of the sums of sines and cosines, add noise uniform in \([-\eta/2, \eta/2]\), and update the positions with the new headings. - Discard the transient (burn-in), then time-average \(v_a(t)\).
The common pitfalls are the wrong noise convention, forgetting the periodic distances and averaging angles incorrectly. See the checklist in The Vicsek Model.
Why not solve_ivp?
In Module 1 you used ODE solvers like scipy.integrate.solve_ivp() for continuous-time systems. The Vicsek model is a discrete-time, agent-based model. At each step all particles update their heading and position from their neighbours and some noise. There is no differential equation to integrate, so you implement it with a loop over time steps.
Prediction
Before you code, write down your expectations.
- With noise \(\eta = 0\) and a large interaction radius, what is the order parameter after many steps?
- With \(\eta = 2\pi\), what is it for \(N = 100\)?
- At fixed noise, does increasing the density help or hurt alignment?
With no noise and every particle seeing every other, all headings converge to one direction and \(v_a \to 1\). With \(\eta = 2\pi\) every heading is random at every step, so \(v_a\) is of order \(1/\sqrt{N} = 0.1\). Higher density means more neighbours inside the radius, which averages out the noise and helps alignment (Vicsek et al. 1995).
Initial State
Create the initial positions and headings. Positions are uniform on the box \([0, L]^2\) and headings are uniform on \([0, 2\pi)\).
def initialize_particles(num_boids, box_size):
# Random initial theta, shape (N)
theta = # TODO
# Random initial x, y - shape (2, N)
xy = # TODO
return xy, thetanp.random.uniform(low, high, size) accepts a tuple as size. Use size=(2, num_boids) for the positions, so that xy[0] holds the x-coordinates and xy[1] the y-coordinates.
def initialize_particles(num_boids, box_size):
theta = np.random.uniform(0, 2 * np.pi, num_boids)
xy = np.random.uniform(0, box_size, (2, num_boids))
return xy, thetaTry out your function now. Vicsek et al. (1995) used between \(N = 40\) and \(N = 10000\) particles and box sizes between \(L = 3\) and \(L = 50\). The more particles you have, the slower the simulation will be, so start with a small number like \(N=100\) and a box size of \(L=10\).
Figure 1 shows what you are building: after a few steps, nearby particles already point in similar directions.