×
About Getting Out
A room, one door and a crowd that wants to leave. The surprise: if everyone hurries hard, the room usually takes longer to empty, not shorter. And a pillar standing a little in front of the door, which looks like it should only get in the way, can make it empty faster. Both come out of a simple model of people as pushing, sliding discs. Faster-is-slower has since been reported in experiments with people, sheep and grains. The obstacle has been tested with sheep at a barn door and with grains leaving a silo, and helped in both when it was not too close.
THE ROOM : a 15 m by 15 m room seen from above, drawn like a fire-safety plan, with people as round counters. Set how many people, how wide the door is and how hard they hurry (the speed each person would like to go), tick Pillar and drag it about, then press RUN . Show pushes draws a red line between any two people pressed together hard, so you can see the arches that lock across the doorway. Longest gap is the longest time nobody went through the door. The chart counts how many go through the door each second, in 5-second steps; every finished run is written in the list underneath, so you can compare. A setting change puts everyone back in the room, ready.
FASTER IS SLOWER : runs the room again and again at hurries from a stroll to a sprint and plots how long everyone takes to get out, with and without the pillar (door and pillar as set on THE ROOM). The page opens with a sweep it made earlier with the same model; RUN SWEEP makes a new one.
TWO DOORS : the room is full of smoke and has two 0.9 m doors, one on each side. Nobody can see a door until they are within 2 m of it (the dashed circles). Until then each person mixes two things: their own search (walking a wandering line and turning back off the walls) and the average direction wanted by the people within 5 m of them who are still searching too. People who have seen a door are left out of that average. The Herding slider sets the mix. Counters turn green once they have seen a door. With full herding the crowd moves as one and usually most of it queues at one door; RUN SWEEP tries every mix several times and plots how long it takes for 9 in 10 people to get out.
WHY : the arch, the forces in the model, what the pillar does, and the sheep and grain that do the same thing.
THE STORY
In 2000 Dirk Helbing, Illés Farkas and Tamás Vicsek published "Simulating dynamical features of escape panic" in Nature . They treated people as discs with a mass of 80 kg that each try to reach their own wanted speed, keep a little distance from others and from walls, and, when bodies touch, push back and rub sideways like grains. Their simulations showed that above a desired speed of about 1.5 m/s the doorway clogs with arch-shaped jams that break in bursts, so a nervous crowd leaves more slowly: they called it "faster is slower". They also found that a column placed asymmetrically in front of an exit improved the outflow, and that in smoke a mix of searching alone and following others found the exits best.
Experiments came later. In 2011 a group at the University of Navarra in Spain, led by Iker Zuriguel, found that an obstacle a little above the outlet of a silo made grains clog far less often. In 2015 the same group (Garcimartín, Zuriguel and others) filmed a flock of sheep going through a narrow door into a barn: clogs came and went just as in the simulations, and an obstacle placed in front of the door helped. In the same year Pastor and colleagues reported faster-is-slower in three systems: people leaving a room through a narrow door, sheep entering a barn, and grains on a vibrated slope going through a gap. In 2016 the Navarra team tried the obstacle in different places in front of the sheep's door: it helped in two places and made things worse when it was too close, with clogs forming between the obstacle and the door post.
The effect only matters at bottlenecks, when people are pressed together. It is one reason building rules set how wide exits must be and how many there must be for the number of people a room may hold: in a crush the danger is the pressure in the crowd, not the speed.
HOW IT WORKS
Each person is a disc 0.5 to 0.7 m across with a mass of 80 kg. Each step (2 thousandths of a second) every person feels: their own drive, which brings their velocity towards their wanted speed in the direction of the door within half a second; a "keep your distance" push from everyone nearby and from the walls, 2,000 N times e to the power of minus (gap ÷ 0.08 m), so it fades within a few tens of centimetres; and, where two bodies overlap, a squeeze of 120,000 N per metre of overlap and a sliding friction of 240,000 times the overlap times how fast they slide past each other. These are the numbers in the 2000 paper. Nobody gets hurt in this model.
To find the way, the page works out the walking distance to the door from every 10 cm square of the floor (Dijkstra's shortest-path method), counting squares close to a wall or the pillar as a little longer, so routes keep some room from them. Each person heads straight down the slope of those distances: round the pillar by the easier side, and through a narrow squeeze only when it is really shorter. Each person's wanted speed is the setting give or take 5%, and everyone gets a small random jostle, bigger for a crowd that is pushing harder, so no stand-off lasts for ever. The same crowd number always gives the same run. Pushes between people more than 1.25 m apart (under about 2 N) are left out to save time.
The Expected line is missing on purpose: there is no formula for this. The curves come from running the model.
SIMPLIFICATIONS
Real people look ahead, step sideways, turn their shoulders and help each other; here they are round, flat and always aim straight for the door. The model was built to show what can happen, not to predict a real building: its numbers are typical of the published model, not measurements of any room. Some details are the page's own: the route-finding; the door posts' keep-your-distance push only centres people in the doorway, and a wall or the pillar never holds back someone nobody is pressing on (otherwise a lone, slow walker could be pinned for good beside a door post); and the random jostle, which the paper's equation of motion includes only as an option. How much faster or slower a room gets depends on the room, the door and the crowd. A pillar is not a fix to copy: in the wrong place it makes things worse, in the model and in the sheep experiments.
Smoke in TWO DOORS follows the idea in the paper, not its exact rules: there, people who came within 2 m of a door counted as out, and its doors were 1.5 m wide. Here they still have to walk through, and the doors are narrower so queues show. The difference between a little herding and none is small next to the luck of each run; with full herding the doors usually come out lopsided, though not in every run. Later research (see Haghani and Ronchi, below) found that real crowds in emergencies rarely panic or stampede: people mostly help each other and follow familiar routes. Herding here just means following the people around you.
FURTHER WATCHING AND READING