Skip to main content

The Fundamental Diagram

Module 6: Explorables and Discrete Models

NoteReading

The fundamental diagram is the relation between traffic density and traffic flow in the stationary state. It is the main macroscopic observable of a traffic model and the one you compare with road measurements (Nagel and Schreckenberg 1992; Chowdhury et al. 2000).

Density, flow and speed

On a ring of \(L\) cells with \(N\) cars:

\[ \rho = \frac{N}{L}, \qquad \bar v = \frac{1}{N}\sum_{n=1}^N v_n, \qquad q = \frac{1}{L}\sum_{n=1}^N v_n = \rho\, \bar v . \]

The flow \(q\) is the number of cars that pass a fixed point per time step, averaged over the ring. In simulations you average \(q\) over time after discarding a burn-in period.

Free flow and congested branches

  • Free flow. At low density cars rarely interact. Each car moves at about \(v_{\max} - p\) on average, so \(q \approx \rho\,(v_{\max} - p)\) grows linearly with \(\rho\) (Chowdhury et al. 2000).
  • Congested branch. At high density the gaps limit the speeds and \(q\) decreases with \(\rho\). At \(\rho = 1\) nothing moves.
  • Maximum flow. Between the two branches the flow reaches a maximum, the capacity of the road. Randomization lowers the capacity and moves it to lower density.

For \(p = 0\) the diagram is the exact triangle \(q = \min(\rho\, v_{\max},\ 1 - \rho)\) with its maximum at \(\rho = 1/(v_{\max} + 1)\) (Chowdhury et al. 2000).

Figure 1: Simulated fundamental diagram (ring of 400 cells, v_max = 5, 400 steps after 200 burn-in steps, random initial positions). The dashed line is the exact result for p = 0. Left: flow. Right: mean speed.

The simulation for \(p = 0\) lies on the exact triangle, which validates the code. For \(p > 0\) the peak is lower and rounded.

What the paper measured

Nagel and Schreckenberg obtained the diagram on a ring of \(10^4\) cells with \(v_{\max} = 5\) (Nagel and Schreckenberg 1992). They started from random positions at rest and discarded \(10L\) steps. Each dot in their Figure 4 averages the flow over only 100 steps, so that the scatter can be compared with detector data on real roads (their Figure 5). The line averages over \(10^6\) steps. They report that the flow changes from laminar to jammed near \(\rho \approx 0.08\), that the position and form of the maximum depend on the system size, and that this dependence is absent without randomization. The maximum flow in their open-road experiment and on the ring is about \(0.32\) to \(0.30\) cars per step. Real freeway data put the maximum at roughly twice that density, and at about \(2000\) vehicles per hour and lane, which is \(0.56\) vehicles per second.

We reproduced these features with \(p = 0.5\), a value that the text does not state and that we inferred from the mean free speed of \(4.5\) cells per step quoted in the paper. The maximum lies at \(\rho \approx 0.08\) to \(0.10\) with \(q \approx 0.32\) on a ring of \(10^4\) cells. On a ring of 100 cells the maximum is higher, about \(0.43\) at \(\rho = 0.10\), and for \(p = 0\) the flow on rings of 100, 1000 and \(10^4\) cells is identical. The figure above uses a ring of 400 cells, so for \(p > 0\) the exact height of its peak depends on that choice. The page What Nagel and Schreckenberg Did contains the code and the figure.

Explore the ring road

The figure below runs the four Nagel-Schreckenberg rules in parallel on a ring of 200 cells. Measure q at this ρ repeats the procedure behind Figure 1: a ring of 400 cells, random start at rest, 200 burn-in steps and an average over the next 400.

  1. Set \(p = 0\) and \(\rho = 0.15\). After the transient, what is the mean speed? Now set \(\rho = 0.25\): why do some cars stay stopped for ever?
  2. Keep \(\rho = 0.15\) and switch on \(p = 0.3\). Do jams appear without any bottleneck? In which direction do they move in the space-time diagram?
  3. Press Sweep all densities with \(p = 0\) and compare the dots with the dashed exact result. Then repeat with \(p = 0.3\): where is the maximum flow now, and how high is it?
  4. With \(p = 0\), change \(v_{\max}\). How does the density of maximum flow move? Compare with \(\rho^* = 1/(v_{\max} + 1)\).

Build this in Python. In the lab you write the code behind this figure: Traffic Model for the update rule and the space-time diagram, and Flow and Congestion for the fundamental diagram.

Empirical evidence: jams without bottlenecks

Sugiyama and co-workers asked 22 drivers to drive around a circular track of 230 m, starting from a uniform configuration. A stop-and-go wave formed spontaneously and travelled backwards, with no bottleneck on the track (Sugiyama et al. 2008). They interpret the jam as a collective instability of free flow that appears when the density exceeds a critical value.

This is the phenomenon that the randomization step produces in the Nagel-Schreckenberg model: jams on a homogeneous ring, triggered by fluctuations. The experiment does not test the specific rules of the model. It shows that a model of homogeneous traffic on a ring must be able to produce such jams.

NoteExercise

Use the free-flow approximation \(q \approx \rho (v_{\max} - p)\) and the congested branch \(q \approx 1 - \rho\) of the deterministic model to estimate where the maximum flow occurs for \(v_{\max} = 5\) and \(p = 0.3\). Compare with the figure.

The two lines cross at \(\rho (v_{\max} - p) = 1 - \rho\), so \(\rho^* = 1/(v_{\max} - p + 1) = 1/5.7 \approx 0.175\) and \(q^* \approx 0.82\). This is a poor estimate for the stochastic model. In simulations on a ring of 400 cells (600 steps, 200 burn-in) the flow for \(p = 0.3\) rises along the free-flow line up to \(\rho \approx 0.10\) and then stays near \(0.46\) to \(0.49\), with a maximum of about \(0.49\) near \(\rho \approx 0.12\). The sampling noise is about \(0.02\). So the maximum is at a lower density than the crossing and its height is about 40 per cent lower. The same happens for the case in the paper, \(p = 0.5\): the crossing gives \(\rho^* \approx 0.18\) and \(q^* \approx 0.82\), while the simulated maximum is \(q \approx 0.32\) near \(\rho \approx 0.08\). Jams form before the free-flow branch reaches the line \(1 - \rho\), and the congested branch lies below \(1 - \rho\), because stopped cars restart with a delay. The crossing of the two straight lines is therefore only an upper bound on the maximum flow.

References

Chowdhury, Debashish, Ludger Santen, and Andreas Schadschneider. 2000. “Statistical Physics of Vehicular Traffic and Some Related Systems.” Physics Reports 329 (4–6): 199–329. https://doi.org/10.1016/S0370-1573(99)00117-9.
Nagel, Kai, and Michael Schreckenberg. 1992. “A Cellular Automaton Model for Freeway Traffic.” Journal de Physique I 2 (12): 2221–29. https://doi.org/10.1051/jp1:1992277.
Sugiyama, Yuki, Minoru Fukui, Macoto Kikuchi, et al. 2008. “Traffic Jams Without Bottlenecks: Experimental Evidence for the Physical Mechanism of the Formation of a Jam.” New Journal of Physics 10 (3): 033001. https://doi.org/10.1088/1367-2630/10/3/033001.