Greenshields Speed–Density Relation

In 1934 Bruce Greenshields, then a graduate student, drove out to a two-lane road near Ohio State University with a camera. He photographed traffic at intervals, measured how far vehicles moved between exposures, and plotted the resulting speeds against the densities he counted off the same photographs. The points fell in a rough downward band, he drew a straight line through them, and in 1935 he presented the result to the Highway Research Board. That line is the first mathematical model of a traffic stream, and ninety years later every course in the subject still starts with it.

The model is $v = v_f(1 - k/k_j)$, and it is built entirely out of getting two endpoints right. On an empty road, $k = 0$, the model gives $v = v_f$: the free-flow speed, which is what drivers choose when nothing constrains them — a function of the geometry, the surface, the sight distance and, some way down the list, the posted limit. In a complete jam, $k = k_j$, the model gives $v = 0$: the jam density, which is one vehicle per average vehicle-plus-gap length, roughly 120 to 160 veh/km per lane for mixed traffic. Between the two, Greenshields drew the simplest thing that connects them.

And that is where the honesty has to come in, because the straight line is the model's one real claim and it is not true. A linear relation says the marginal effect of adding one more vehicle is the same on an empty road as in a jam, and any driver knows better. Real speed–density curves measured with modern equipment are concave: traffic holds close to the free-flow speed until density becomes substantial, then falls away sharply near capacity. Greenshields' data came from one two-lane road, in 1934, with pre-war vehicles, pre-war brakes and drivers who had never seen a freeway — and, as later authors have pointed out, from a fairly small number of observations, many of them at low density, with the jam-density end of the line effectively extrapolated rather than measured.

Better models exist and have for decades. Greenberg's logarithmic model fits congested flow well and misbehaves at low density. Underwood's exponential model does the reverse. The Van Aerde and Newell models handle both ends. Multi-regime models simply fit different curves to different density ranges and accept the discontinuity. None of them has displaced Greenshields in teaching, for one good reason: it is the only one simple enough that the algebra downstream stays legible. Substitute Greenshields into $q = kv$ and you get a parabola whose maximum you can find in one line. Substitute Van Aerde and you get something you solve numerically.

So use it for the shape of the argument and not for a design number. Both parameters are facility-specific inputs and neither is a constant you can look up. A rural two-lane road, an urban arterial and a freeway lane give three quite different pairs, and so do the same road wet and dry, day and night, or with and without heavy vehicles. Calibrate them from your own paired speed and density observations: regress speed on density, read $v_f$ off the intercept and $k_j$ off the crossing. And be sceptical of a jam density extrapolated from uncongested data — if every point you have sits near the free end of the line, the crossing point is far outside your data and can be wrong by a factor of two.

What the model is genuinely good for is teaching the shape of the fundamental diagram, giving closed-form expressions for capacity and critical density, and providing the tractable starting point that kinematic-wave theory — Lighthill, Whitham and Richards, a decade and a half later — was built on top of.