Street network topology

Why can two neighbourhoods with the same length of road move traffic so differently?

Take two neighbourhoods with the same number of houses and, near enough, the same total kilometres of pavement. One of them absorbs the morning rush with nothing worse than a wait at a few lights. The other seizes up every weekday at one particular junction, and a broken-down lorry there can immobilise a thousand households.

If road quantity were the thing that mattered, this could not happen. So quantity is evidently not the thing that matters. But then what is left? The asphalt is the same, the cars are the same, the trips are the same. The only remaining difference is how the pieces are joined — and it is worth asking why something so abstract should have such physical consequences.

Begin with where a street system's capacity actually sits. A lane of road can carry a great many vehicles per hour when nothing interrupts them; what interrupts them is junctions, where streams must be separated in time. So capacity is metered at the nodes, not stored in the links. Notice the immediate consequence: adding kilometres of street without adding places where routes cross adds almost no capacity at all. You can lay a great deal of road and gain nothing.

Now the structural question. Consider two ways of joining a fixed set of streets. In one, streets cross each other repeatedly, so the network is full of closed circuits. In the other — the branching, cul-de-sac pattern of most postwar suburbs — there are no circuits: lanes feed courts, courts feed a loop, the loop feeds a collector, the collector feeds one arterial. Between any two points there is exactly one path.

Sit with that last sentence, because everything follows from it. If there is exactly one path, then the traffic on any link is forced to be the sum of all the traffic originating beyond it. Not likely to be, not usually — forced. Nobody chose to concentrate a whole district onto one junction; the geometry did it, and no amount of goodwill or clever signal timing can undo it. Where circuits exist, the same trips have several roughly equal routes and spread themselves across them.

This is measurable in a pleasingly plain way. A branching network with no circuits has one fewer link than it has nodes, so its ratio of links to nodes falls at or below one; a well-connected gridded fabric runs nearer one and a half. Two neighbourhoods with identical road length can sit on either side of that line.

Two more consequences drop out. The first is trip length. With one path available, the path you get is often a long way round: your neighbour across a fence may be a two-kilometre drive away. So the same activities consume more vehicle-kilometres, and journeys that would have been a short walk become drives, which quietly adds traffic that the layout then has to carry.

The second is failure. Ask what a blocked street costs in each fabric. In a network of circuits, the answer is a detour of one block. In a branching one, a blockage on a collector severs everything beyond it — not delays it, severs it — and the same is true of a burst water main, a fallen tree or a parade. Redundancy is not a luxury feature here; it is the same property as capacity, seen from another angle.

Before we make the grid the hero, though, the honest counter-case. A four-way intersection has around thirty-two points where vehicle paths can conflict; a T-junction has about nine. Branching layouts were promoted for exactly that reason — they keep through-traffic and its conflicts out of the streets where children live, and they genuinely do. Grids buy their redundancy with more junctions, which means more stopping and more places to be hit. The interesting design question is therefore not which is better but whether the two goods can be separated, and they can: several cities now make a fabric that is fully connected for walking and cycling while deliberately interrupted for cars, so through-movement is filtered by mode rather than blocked for everyone.

The misconception worth naming is that congestion is a volume problem to be solved with width. If the fabric is a tree, widening the trunk raises the ceiling but leaves the summation intact, so the relief lasts until traffic grows into it — and every branch still has exactly one way out. The structural fix is to add circuits: a connection between two branches that previously met only at the collector removes a whole class of trips from it.

Two refinements. Connectivity is not the same as density of streets, though they correlate: what counts is how many independent routes exist between real origins and destinations, so a few well-placed links can matter more than a great deal of extra pavement. And topology sets the possibilities rather than the outcome — land uses, transit, parking prices and where the jobs are all determine how heavily any of it is used.

Hold onto the general form of the argument, because it is not really about roads. The same material, arranged with circuits or without, gives different capacity, different trip lengths and different failure behaviour, and all three follow from the arrangement alone. That is what it means to say structure determines function.

• Why adding a link can occasionally make everyone slower, and what Braess's paradox does and does not show about real cities.
• How retrofitting connections into an already-built branching suburb is done, and why it is so much harder than laying it out that way.
• Whether space syntax measures of a street's integration predict where the shops end up.

Every term the collection defines is gathered in the glossary.