Train wheels have no differential, and steer themselves

One axle, two rails of unequal length

On a railway axle the two wheels are not free to spin on their own. They are pressed onto one shaft and turn together, at one rate, like the ends of a rigid barbell. The two wheels and the axle joining them make a single rigid body, a wheelset. Yet the track ahead almost never lets that wheelset roll equal distances on its two sides. On any curve the outer rail runs longer than the inner, and a car with a differential, which every road car has, would let its outer wheel turn faster to cover the extra ground. The wheelset has no differential. It has no gears at all. It rounds the curve anyway, and it does so without the wheels scrubbing sideways or climbing off the rail. The part that manages this is not a mechanism bolted on. It is the shape of the wheel at the one place it touches the steel, a shape old enough that the coned railway wheel counts among the earliest inventions in the field.¹

The tread is not flat

Look at where a wheel meets the rail. The rolling surface, the tread, is not a cylinder. It is gently tapered: a little fatter near the inner edge by the flange, a little thinner toward the outside.² The taper is slight. One common standard cuts it at one part in twenty, so across the width of the tread the outer diameter runs only about a quarter inch, six millimetres, smaller than the inner.³ This taper is the wheel's conicity, and a wheel cut this way is coned.

Because the tread is coned, the diameter that actually meets the rail depends on where along the tread the contact sits. That contact diameter has a name: the rolling radius, the radius of the circle a wheel is rolling on at any instant, which for a coned tread is not fixed. Slide the whole wheelset sideways and each wheel takes a different rolling radius. The wheel that moves toward its fat inner edge presents a larger one; its rigid twin, moving toward its thin edge, presents a smaller one. Everything below follows from that single fact: one shaft, one turning rate, and two wheels whose working diameter changes with where they sit across the rail.

How a rigid axle takes a curve

On a curve the wheelset rides a little toward the outside of the track. Now the coning pays off. The outer wheel, pushed toward its fat inner edge, rolls on a larger radius; the inner wheel, riding toward its thin edge, rolls on a smaller one. In one turn of the shared axle the outer wheel lays down more track than the inner. That is precisely what the curve asks for: more distance on the outer rail than the inner. The rigid axle produces the two different wheel speeds a differential would hand out, with no differential present, because for the length of the curve the two wheels have quietly become two different sizes.¹ The engineering explainer states the result flatly: the effective diameter of each wheel changes in a curve, and that alone solves the problem of a wheel being dragged along at the wrong speed.³ Feynman, working the same picture aloud, put it as the wheels sliding out on the track a bit so one runs on a bigger diameter and one on a smaller.²

This is the first of two jobs the cone does: on a curve, the coned wheelset acts like a differential.

One nudge, and it steers back

The second job is a different thing, and it is the one intuition usually gets wrong. Start the wheelset centered on straight, level track, rolling true. Give it one small sideways nudge: a bump, a gust, a flaw in the rail. The geometry that cornered the curve is now at work on tangent track. The wheel shoved toward its fat edge takes a larger rolling radius; its rigid twin, pulled toward its thin edge, takes a smaller one. Both are keyed to one shaft turning at one rate. So on every revolution the larger wheel is carried a little farther forward than the smaller. The axle, advancing unevenly on its two ends, turns. It yaws to point back toward the center of the track, and rolls itself there.²

Notice what is not doing this. The flange, the lip on the inner face of each wheel, is a backstop, not the steering. It comes into play in case the real mechanism fails, and it keeps the train on the rails even if a wheel lifts.²³ The work of guidance is done a few millimetres away, in the taper of the tread.

A sine wave drawn on the rail

"Steers back toward center" is not the same as "settles at center." A wheelset that yaws back toward the middle does not stop when it gets there. It arrives with its correction still turned in, overshoots, and leans the other way, where the same geometry turns it back again. Left to pure rolling, the wheelset traces a sine wave down the track, weaving from rail to rail as it rolls forward. The oscillation was noticed on the earliest railways by George Stephenson; the first analysis of it was Klingel's, in 1883.¹

Klingel found the length of one full weave, and it depends only on the geometry: the wavelength is 2π√(rb/k), where r is the rolling radius, b is the half-gauge (half the distance between the two contact points), and k is the conicity. Put real numbers in. For a wheel of half-metre radius on standard-gauge track with a one-in-twenty cone, one full side-to-side weave takes about seventeen metres of travel.⁴ That is close to the sixty-foot lengths (18.3 m) British track was once laid in, and about the length of an articulated bus. The surprise sits in what the formula leaves out: there is no speed term. The wavelength is fixed by the wheel and the rail, not by how fast the train runs. Go faster and each weave keeps its seventeen metres; you simply meet the cycles sooner, so the wobble's frequency in time climbs with speed while its length on the ground does not.⁴⁵

Here the clean picture reaches its limit. Klingel's is a kinematic result: it assumes the wheels roll without any slip at the contact, so the whole motion is settled by geometry, with no need for forces at all.¹ Under that assumption Antali, Stepan and Hogan showed in 2015 what the clean geometry actually implies, and it is not what the intuition promises. The pure geometric oscillation is neutrally stable: it neither dies away nor builds up.¹ The cone gives a displaced wheelset a real tendency to turn back toward center, but on geometry alone that tendency yields an endless, fixed-size weave, not a coming-to-rest. The coning steers; it does not damp. (Their analysis also shows Klingel's neat formula is itself an approximation, low by two or three percent for ordinary wheel and rail shapes, which is close enough to keep and worth knowing is inexact.)¹

Why the cone needs a suspension

What decides whether the weave fades or grows is the very thing Klingel set aside: the small slip where wheel meets rail. The steel is not perfectly rigid; it deforms and micro-slips under load, and the creep forces that result, which Carter added to the analysis in 1926, are what actually drain energy from the oscillation or pump energy into it.¹⁶ Below a certain speed those forces damp the weave and the wheelset rides quiet. Above a critical speed they feed it, and the fixed-size weave becomes a growing one. That runaway has a name, hunting: an intrinsic swaying that arises from the train's own forward speed together with the wheel-rail contact forces.⁷ At its worst it slams the bogies side to side hard enough to damage track and wheels, and for much of railway history it set the ceiling on how fast a train could safely run.⁵³

Where that critical speed sits is not fixed by the cone. It is a property of the whole vehicle: the axle load, the vertical stiffness of the primary suspension, the lateral damping of the secondary suspension all move it.⁷ So railways buy their speed back by design. They stiffen the constraint on how a wheelset can yaw by grouping wheelsets into bogies and adding elastic constraints and dampers, which raises the critical speed.⁵ And they reprofile wheels as the tread wears, since a worn tread raises the wheel's equivalent conicity, and higher conicity lowers the speed at which hunting begins.⁸

So the tapered wheel is not the flawless self-correcting part the story usually makes of it. It solves two problems for nothing, with no gears and no control system: it lets a rigid axle corner like a differential, and it turns a wandering axle back toward the center of the track. What it cannot do on its own is keep quiet about the second favor. The cone that steers also weaves, and taming that weave is the work a train's suspension exists to do. The next time a train rolls past, the flange catching the light is the backup. The real guidance is happening in the invisible taper, a few millimetres wide, where each wheel meets the rail.