
In February 1904 the readers of the Strand Magazine followed Sherlock Holmes and Dr. Watson onto a sodden moor in the north of England. A ten-year-old boy, the only son of the Duke of Holdernesse, had vanished from his boarding school in the night. So had the German master, and so had the German master’s bicycle. On a black ribbon of path between two bogs, Holmes found what he was looking for: the mark of a tyre.
The story is “The Adventure of the Priory School”, and Conan Doyle later ranked it tenth among his twelve favourite Holmes stories. Holmes announces that he knows forty-two different tyre impressions. This one is a Dunlop with a patch on the outer cover; the missing master rode on Palmers, which leave long stripes. Then he goes one step further and tells Watson which way the rider was travelling. That single step has kept mathematicians busy for more than a century, because Holmes was guessing, and because the tracks really do hold the answer.
Lower Gill Moor, 1904
What Holmes said
Holmes declares that the track was made by a rider going away from the school. Watson, for once, pushes back: “Or towards it?” The reply is one of the most confident paragraphs in the canon.
“No, no, my dear Watson. The more deeply sunk impression is, of course, the hind wheel, upon which the weight rests. You perceive several places where it has passed across and obliterated the more shallow mark of the front one. It was undoubtedly heading away from the school.”
Sherlock Holmes, The Adventure of the Priory School
Read it twice and the trouble shows. The first sentence is sound. A rider sits nearly over the back wheel, so the rear tyre does press deeper. The second sentence is true as well: the rear wheel follows the front one and rolls over its mark wherever the two tracks cross. But neither fact has a direction in it. Turn the bicycle round and ride it back along the same path, and the rear wheel is still the heavy one, and it still comes second, and it still flattens the front track at every crossing. Holmes has worked out which wheel is which. He has not worked out which way it was going.
Readers noticed, and they wrote in. Conan Doyle answered them years later in his memoirs, with more grace than his detective usually managed:
“I had so many remonstrances upon this point, varying from pity to anger, that I took out my bicycle and tried.”
He found that his correspondents were right. The crossings look the same whichever way the machine is heading. Doyle then offered a rescue: on an undulating moor a wheel digs in deeper going uphill than coming down, so a careful eye could still read the direction, and “Holmes was justified of his wisdom after all.” Perhaps. But that is a clue about hills and effort. The better clue was lying in the shape of the two curves the whole time, and it works on perfectly flat ground.
The model
A bicycle is two points and a promise
Strip a bicycle down to what the mud sees. There are two points of contact, one under each wheel. The frame holds them a fixed distance apart, the wheelbase, which on an ordinary bicycle is a little over a metre. The front wheel can be turned to point anywhere. The rear wheel cannot. It is bolted into the frame, so it can only roll straight along the line of the frame, and it never slides sideways.
That is the promise, and everything follows from it. At every instant the rear wheel is moving directly towards the spot where the front wheel is touching the ground. Said in the language of curves: the frame of the bicycle always lies along the tangent to the rear track, and the front wheel sits on that tangent exactly one wheelbase ahead.
The front wheel does the steering and the rear wheel is dragged after it. Stretch the wheelbase and watch the rear track calm down.
The picture already settles the question Holmes answered by weight. You do not need to measure how deep the ruts are. The rear wheel cuts every corner, so its track is the smoother of the two, weaving inside the swings of the front track. For gentle wobbles of wavelength ridden on a wheelbase , the amount of smoothing has a tidy form:
On a normal bicycle a lazy weave that repeats every six metres reaches the rear wheel at about two thirds of its size. A nervous three-metre wobble comes through at about 40 percent. An engineer would call the bicycle a low-pass filter: it lets slow turns through and irons out fast ones. So the busier track is the front wheel, and the calmer one is the back.
The method
The tangent test
Now the direction. Pick any point on the rear track and draw the tangent line there. The line runs off both ways, and each way it eventually meets the front track. One of those two meetings is where the front wheel actually was at that moment. If the bicycle was heading to the right, the meeting on the right is a wheelbase away. Pick another point and the meeting on the right is again one wheelbase away, and again, at every point you try, because a bicycle cannot change its length mid-ride.
Look the wrong way along the tangent and nothing holds the distance fixed. It comes out long at one point and short at the next. That is the whole test: the true direction is the one in which the tangent segments all come out the same length.
One caution for anyone trying this with a stick. Where the tracks weave across each other, a tangent can cross the front track more than once. You are not looking for the first crossing. You are looking for a crossing that sits at the same distance every time, and only the true direction has one.
Slide along the tracks. The green segment never changes, so this bicycle was riding to the right. The amber one cannot make up its mind.
If you ever find a pair of tracks in snow, wet sand or a muddy lane, the recipe is short:
- Find the rear track. It is the smoother curve, on the inside of every bend, and usually the deeper one.
- At three or four points on it, lay a stick along the track so that it just kisses the curve. That is your tangent.
- Follow each tangent in the same direction until it hits the front track, and compare the distances.
- If they match, and come to about a metre, the bicycle went that way. If they do not, it went the other way.
The puzzle entered the mathematical literature through a geometry course at Princeton in the 1980s and gave its name to a well-loved problem book, Which Way Did the Bicycle Go? by Joseph Konhauser, Dan Velleman and Stan Wagon (1996). George Hart later filmed the test on real tracks for the Simons Foundation, which is how I first met it.
Paris, the 1670s
A pocket watch and a curve called the tractrix
The mathematics under the bicycle is older than the bicycle by two hundred years. The usual story begins in Paris with Claude Perrault, physician, architect of the Louvre colonnade and brother of the man who wrote down Cinderella. Perrault laid his pocket watch on a table, took the free end of its chain, and drew that end along the straight edge of the table. The watch slid after it. What curve, he asked the mathematicians of his day, does the watch follow?
Perrault’s experiment. Then take the chain yourself: your hand is the front wheel and the watch is the rear one.
Newton looked at the curve in 1676. Leibniz and Huygens both solved it in the early 1690s, and from them it got its name: the tractrix, from the Latin trahere, to drag, the same root as tractor. The table edge is its asymptote. The watch creeps closer to the edge for ever and never arrives. With a chain of length the curve can be written down exactly:
The defining property is the one you have just watched. From any point of the tractrix, the tangent segment to the asymptote has the same length , because that segment is the chain. A bicycle whose front wheel rolls along a straight line while the rear starts off to one side does exactly what the watch does, so its rear track is a tractrix. On a winding road the front wheel follows something more complicated, and the rear track is the same idea generalised: the curve whose tangents reach the front track after a fixed distance. I used the same curve in the geometry of perfect parking, where it describes a car nosing into a space.
The tractrix has a habit of producing clean numbers. The area trapped between the whole curve and its asymptote is , exactly half the disc you could draw with the chain as radius. Spin the curve about its asymptote and you get the pseudosphere, a trumpet-shaped surface whose area is , the same as a sphere of radius . In 1868 Eugenio Beltrami showed that this surface has constant negative curvature, which made it the first model anyone could hold of non-Euclidean geometry. The same years that solved the watch problem also gave us the brachistochrone; the new calculus was being tested on every curve its inventors could find.
The fine print
When even the mathematics cannot tell
Holmes liked to say that when you have eliminated the impossible, whatever remains must be the truth. The tangent test has cases in which two possibilities remain. The dull one is a dead straight line: both wheels leave the same track and it reads the same in both directions. The beautiful one is a circle.
Ride in a perfect circle and the rear wheel traces a circle of radius while the front wheel traces a larger one around the same centre. The frame is tangent to the inner circle and reaches the outer one after a wheelbase, so by Pythagoras the outer radius is . But the picture is perfectly symmetric. Every tangent segment has length clockwise and anticlockwise alike, and no amount of geometry can say which way the rider went round. There is a small prize hidden in the ring between the two tracks:
The radius has cancelled. Ride a tight circle or a huge one, and the area between your two tracks is the same, the area of a disc whose radius is your wheelbase. About three and a half square metres on an ordinary bicycle. Mathematicians have shown that the same is true for any simple closed loop, oval or wobbly, as long as the bicycle comes back to where it began after turning round once. A nineteenth-century drawing instrument, the hatchet planimeter, measures areas on exactly this principle. It is a bicycle the size of a pencil.
There are stranger exceptions. In 2002 David Finn constructed a ride in which the rear wheel follows precisely in the path of the front one, so that a bicycle leaves a single track like a unicycle. Mark Levi and Sergei Tabachnikov later proved that such a track must wiggle more and more violently as it goes forward, so nobody will ride one for long. For every track a real cyclist is likely to leave, the tangent test gives one answer.
Back on the moor
Would it have worked for Holmes?
A real bicycle is not quite two points on a rigid stick. The front fork is raked, so the front tyre touches the ground a few centimetres behind the steering axis, and that contact point shifts a little when the rider turns or leans. Tyres have width. Mud smears. Measured tangent segments will agree to within a few centimetres, not to the millimetre. That is still plenty. In the wrong direction the lengths typically differ by half a metre or more, and the two answers are hard to confuse.
So the method was available on Lower Gill Moor. A patch of soft ground, a walking stick to lay along the hind-wheel track at three or four places, and Holmes could have answered Watson with a measurement. He would have enjoyed it, too. It is exactly his kind of clue: a thing everybody sees and nobody observes, sitting in plain view between two lines in the mud. It is also a fair answer to anyone who asks what mathematics is good for. Sometimes it is good for catching the detective out.
Sources
- Arthur Conan Doyle, “The Adventure of the Priory School”, Collier’s, 30 January 1904, and The Strand Magazine, February 1904.
- Arthur Conan Doyle, Memories and Adventures, quoted in the Arthur Conan Doyle Encyclopedia.
- Joseph Konhauser, Dan Velleman and Stan Wagon, Which Way Did the Bicycle Go? And Other Intriguing Mathematical Mysteries, Mathematical Association of America, 1996.
- George Hart, Mathematical Impressions: Bicycle Tracks, Simons Foundation, 2013.
- Robert Foote, Mark Levi and Serge Tabachnikov, Tractrices, Bicycle Tire Tracks, Hatchet Planimeters, and a 100-year-old Conjecture, American Mathematical Monthly 120, 2013.
- David Finn, Can a Bicycle Create a Unicycle Track?, College Mathematics Journal 33, 2002.
- Mark Levi and Serge Tabachnikov, On Bicycle Tire Tracks Geometry, Hatchet Planimeter, Menzin’s Conjecture and Oscillation of Unicycle Tracks, Experimental Mathematics 18, 2009.
- Tractrix, Wikipedia, for the history of Perrault, Newton, Huygens and Leibniz.





