Skip to the story
mariuscomper.uk Română

The roads a brainless slime mould would build between Romania's cities

Physarum polycephalum is a single yellow cell with no neurons. In 2010, placed where Tokyo is, with oats where the surrounding cities are, it grew a network that resembled the rail lines. Here the rule found then grows over Romania's terrain, between its 41 county capitals, and is set against what engineers built.

Move your mouse across the screen: the model follows it like food.Tap the screen: the model grows towards the spot, as towards food.

The network grown by the model between Romania's 41 county capitals, over the country's relief

The organism

A cell that builds its own roads

Physarum polycephalum lives on rotting wood and wet leaves. It is called a slime mould, but it is a myxomycete: not a plant, not a fungus, not an animal. While it feeds, it is a single large cell spread out as a network of yellow tubes, without a single neuron.

Inside the tubes the cell's contents flow one way, then back, about once every 100 seconds. Researchers have observed that tubes carrying a lot of flow thicken, while the others thin out.

A yellow network of Physarum polycephalum tubes on a fallen log
Physarum polycephalum on fallen wood. Photo: frankenstoen, Wikimedia Commons, CC BY 2.5.

What moves on the screen is a model, not the organism: more than a hundred thousand particles following the trails the others leave, after Jeff Jones's model (2010).

The maze, 2000

The shortest path, found without a brain

Toshiyuki Nakagaki and colleagues filled an agar maze with pieces of Physarum, which merged into a single cell. Then they put oats at the entrance and the exit. The branches in the dead ends withdrew and, in most trials, one thick tube remained along the shortest path.

The maze had four routes. In the first half, the tube went through the option about 22% shorter every time; in the second, where the options differ by only 2%, it used each about equally often.

On screen runs the model Atsushi Tero and colleagues wrote in 2007 for this experiment, on a maze drawn for this page: a tube thickens when a lot flows through it and thins when little does.

Tokyo, 2010

One cell grew a network resembling the trains around Tokyo

In 2010, Tero and Nakagaki's team placed the organism on a map of the Tokyo region, at the capital, with oats at 36 points representing the region's cities. In some trials, mountains and lakes were lit, because Physarum shies away from light. Within about a day, what remained was a network that resembled the region's railways.

Physarum polycephalum grown around an oat flake in a Petri dish
Physarum around an oat flake in a Petri dish. Photo: Tim Tim, Wikimedia Commons, CC BY-SA 4.0.

The rail network was 1.8 times longer than the shortest network linking the cities; the organism's networks were 1.75 times on average, over 21 trials. The trains coped better with a break, though: 4% of their links, if cut, would isolate part of the network, against 14% for the organism grown under the light mask and 20% for the one grown without it.

The researchers wrote the rule as an equation: a tube thickens according to how much fluid passes through it. On screen, the equation runs on the Kantō plain, with oats at the 36 largest places at least 12 km apart; the paper does not list the experiment's exact points.

Romania

The same rule, across the Carpathians

The oats sit at Romania's 41 county capitals, Bucharest included, all with the same helping. At the start the country is covered by a fine web of about 22,000 tubes, each about 6 km long.

At each step a city chosen at random sends fluid to another city chosen at random, and every tube thickens or thins according to how much flowed through it.

The mountains take the place of the light in Tokyo: in the version on screen, a tube that crosses 1,500 m costs four times as much as one on the plain.

  • the model’s tubes
  • county capitals

Growth

Unused tubes withdraw

As you scroll, the map runs through the model's 15,000 steps. Out of the fine web, 642 tubes remain, 3,840 km in all, linking all 41 cities.

The dots flowing in the tubes show the fluid moving back and forth, as in the organism, only much faster.

The mountains

Where it crosses the mountains

The lighter areas are those the model avoids. Between Transylvania and the south, the network crosses the Southern Carpathians, in all nine runs, between Sibiu and Râmnicu Vâlcea: beside the railway there and close to the Sibiu–Pitești motorway, partly open and the rest under construction. Only two runs add a second crossing, between Drobeta-Turnu Severin and Deva. Between Brașov and Ploiești it never crosses.

Moldavia, the north-east of the country, joins the rest from the south-east, through Focșani–Buzău and Galați–Brăila, in every run. Across the Eastern Carpathians, the most frequent crossing, between Bacău and Sfântu Gheorghe, appears in five of nine.

Railways

83% of the network runs beside a railway

Blue: main and branch railway lines, from OpenStreetMap. 83% of the length of the model's network lies within 10 km of one of them.

Read that number with care: Romania has many lines, and 59% of the country's land lies within 10 km of a railway.

  • the model’s tubes
  • railways

The test

The shortest possible network stays just as close to the trains

White: the shortest network linking the 41 cities, drawn over the same terrain. It is 2,770 km long, and 86% of it lies within 10 km of a railway, against 83% for the model. With no mountains at all, the model still reaches 76%.

So the resemblance comes mostly from where the cities are: the railways link the same cities, and a network joining them by short routes passes beside the lines. The mountains add only a few percentage points.

What the model adds is loops. Its network is 1.39 times the minimum length. In the minimal one any broken link splits the country in two; on the map, only 6% of the links between junctions would, and across the nine runs between 6% and 21%.

  • shortest possible network
  • railways

Motorways

22 of the 41 county capitals have no motorway within 20 km

Red: open motorways, from OpenStreetMap on 25 October 2026; dashed red: those under construction. Even with those finished, 20 county capitals would remain more than 20 km from a motorway.

Against the motorways, the model matches less well than the shortest network, in all nine runs. It does link every city, with 3,840 km of tubes. It is not a motorway plan: it does not know how many people travel between two cities, what a viaduct costs, or where building is allowed.

  • the model’s tubes
  • open motorways
  • motorways under construction

What stays

The links that appear in every run

The model starts from a random web and picks cities at random, so no two runs give the same map. It was run nine times: three starting webs, each with the mountains weighing little, medium and a lot. The glow shows how often a tube passed each place.

41 links between cities appear in all nine runs. For 24 of them, less than 30% of the way has a motorway, open or under construction, within 10 km; for example Iași–Vaslui, Brașov–Sfântu Gheorghe and Craiova–Drobeta-Turnu Severin. The full list is below.

  • the model’s tubes
  • open motorways
  • motorways under construction

Try it

Put oats wherever you like

Here the model runs on your device, on a sparser web with tubes 12 km apart. Tap or click the map to add oats or to remove a city. The network rebuilds itself in a few seconds.

What the model shows, and what it does not

The rule that describes a brainless organism draws a network between Romania's cities that follows the valleys and runs close to the trains. The test shows this is no proof of any skill in the rule: the shortest possible network over the same terrain stays just as close to the railways, and matches the motorways better than the model does. The resemblance comes mostly from where the cities are, which the railways link anyway, and from the fact that 59% of the country lies within 10 km of a line. The mountains add a few percentage points.

What the model adds is a compromise the Tokyo team also saw: a network somewhat longer than the minimum, with loops that keep it from breaking at the first cut.

The model against three comparison networks

Each number is a match score from 0 to 100 that is high only when both parts are high: the share of the network within 10 km of a railway or motorway, and the share of the railway or motorway within 10 km of the network. For each mountain weight, the average over the three starting webs. All comparison networks are drawn over the same terrain. "Close neighbours" links two cities when no other city lies in the circle whose diameter joins them; the triangulation links every neighbour.

How much the mountains weighThe model
rail · motorway
Shortest networkClose neighboursTriangulation
no mountains5432473668397237
light mountains5636504570417639
medium mountains (the map)5837514671417639
heavy mountains5940524571417640

On the railways the score rises with density, because a denser network touches more lines; the model is third of four at every mountain weight. On the motorways, wherever the mountains carry weight, the shortest network clearly beats the rest; the model comes last, with heavy mountains almost level with the triangulation.

How far the nearest motorway is

Straight-line distance from the centre of each county capital to the nearest motorway, from OpenStreetMap (25 October 2026). A city with a motorway bypass on its edge can show as 12–15 km.

County capitalOpen motorwayWith those under construction
Drobeta-Turnu Severin134 km134 km
Botoșani124 km21 km
Suceava121 km8 km
Tulcea104 km104 km
Craiova101 km101 km
Târgu Jiu88 km87 km
Miercurea Ciuc84 km45 km
Iași78 km9 km
Bistrița72 km69 km
Baia Mare70 km68 km

The 41 links that appear in every run

For each link: the share of the model's route with a motorway (open or under construction) and a railway within 10 km. Average over the nine runs.

How it was made

The rule. The model of Tero and colleagues (Science, 2010): a web of tubes, each with a conductance D. At each step, a city chosen at random sends a fixed amount of fluid (I0 = 2) towards another city chosen at random; flow through each tube follows Poiseuille's law, and D grows with the flux Q, as dD/dt = Q^1.8/(1 + Q^1.8) − D. I0 = 2 and 1.8 are the values the paper used for the Tokyo network.

The map. A web of about 22,000 tubes between points set almost regularly 6 km apart, within Romania's outline. Each tube's length is multiplied by 1 + k × c, where c rises from 0 below 300 m to 1 above 1,500 m (Copernicus GLO-30 relief). On the map k = 3; results were checked for k = 0, 1.5 and 6 too.

The runs. 15,000 steps per run, three starting webs for each value of k. A tube belongs to the final network if D > 0.01; in the main runs (and almost the same in those without mountains) the others fall below 0.000001, so the threshold does not change the result. Variants (approximate lengths, because there the threshold matters): with I0 = 0.5 the network is 2,410 km, with I0 = 6 it is 4,960 km; with the values that gave the paper a network very similar to the Tokyo trains (I0 = 0.2, exponent 1.15), 3,070 km.

The comparison. Railways (main and branch lines) and motorways come from the OpenStreetMap extract for Romania of 25 October 2026. All networks are drawn on the same 0.8 km grid, and the match is measured with a 10 km tolerance; with 5 km the order stays the same. For reference, 59% of the country's land lies within 10 km of a main or branch railway, and 17% within 10 km of an open or under-construction motorway.

What the test is not. The real organism has never been placed on a map of Romania. Only the rule that describes it runs here. The model on the opening screen (the particles) is a different one, Jeff Jones's, and is there only as a picture.

Sources