If your bike had square wheels, it would ride perfectly smoothly, with the axle never rising or falling, on one special road: a row of identical bumps, each shaped like a hanging chain turned upside down (an inverted catenary, \( y = -a\cosh(x/a) \)). On a flat road it would thump to pieces. It can also never turn, which is why the round wheel won.
- The matching road is a row of inverted catenaries, \( y = -a\cosh(x/a) \), each 1.763a long.
- At walking pace the corners need a friction coefficient of about μ = 1, the limit of a road tyre.
- The grip needed falls to about 0.2 near 20 to 40 km/h: the sweet spot (this post's own analysis, below).
- At motorcycle scale, the wheel's own inertia runs it out of grip again at 50 to 70 km/h.
- A 118 km road would need about 209,000 bumps, and the bike could never steer.
Working the physics through shows the bike needs a friction coefficient close to 1 at low speed, gets easier to ride between roughly 20 and 50 km/h, and runs out of grip again around 50 to 70 km/h at motorcycle scale.
This is not a joke answer, and Macalester College in Minnesota has the proof. In 1997 the mathematician Stan Wagon and bicycle mechanic Loren Kellen built a square-wheeled tricycle that rolls over a road of inverted catenaries, the shape of the St. Louis Gateway Arch. The college's president took the first ride.
The original wore out, and a replacement in sturdier materials went into service in April 2004 on the lower level of the Olin-Rice Science Center, where anyone can ride it. Riders say it is so smooth you can't tell the wheels aren't round. Its odometer logged just over 15 miles between September 2004 and September 2005, all of it in a straight line.
Macalester College: A Square Wheel, or The Ultimate Flat TireThe idea goes back to G.B. Robison's 1960 paper on rockers and rollers, and in April 2026 Wagon and Antonín Slavík published a new derivation of the road-and-wheel equations. Below, the road is derived from scratch, then the dynamics are solved to find the speed limit, which no popular account gives.
Why does a square wheel need a special road?
A square wheel needs a bumpy road because the distance from its centre to its edge changes as it turns, from \( a \) at the middle of a side to \( a\sqrt2 \) at a corner. On flat ground the axle rises and falls by 41% of the half-side every quarter-turn.
Calculated in this post from the geometry of a square: its centre-to-edge distance runs from a to a√2, so the rise is (√2 − 1)a = 0.414a. The wheel-and-road equations are in Slavík and Wagon, 2026.
The fix is to move the ground instead: build the road so it rises exactly where the wheel's radius shrinks and falls where it grows. Two conditions define such a road, and they are the whole problem:
- The axle stays level. Put the axle on the line \( y = 0 \). The contact point must then be at depth \( y = -r(\theta) \), where \( r(\theta) \) is the wheel's radius in the direction pointing down.
- No slipping. If the axle moves horizontally, the contact point must sit straight below it (it is the instantaneous centre of rotation), and the wheel's rim must unroll along the road at the same rate it advances: \( dx = r(\theta)\, d\theta \). This is Robison's condition, which Slavík and Wagon re-derive from the no-slip assumption alone.
How do you derive the catenary road for a square?
For a square of half-side \( a \), measured from the middle of the bottom side, the distance from the centre to the side at angle \( \theta \) is \( r(\theta) = a\sec\theta \), valid for \( -\pi/4 \le \theta \le \pi/4 \) until a corner arrives. The derivation, step by step:
- Write the no-slip condition.
\[ \frac{dx}{d\theta} = r(\theta) = a\sec\theta \]
- Integrate. The integral of \( \sec\theta \) is the inverse Gudermannian function:
\[ x = a\ln(\sec\theta + \tan\theta) \quad\Longleftrightarrow\quad \theta = \operatorname{gd}\!\left(\frac{x}{a}\right) \]
- Use the Gudermannian identity. For \( \theta = \operatorname{gd}(u) \), \( \sec\theta = \cosh u \). So the depth of the road is
\[ y(x) = -r = -a\sec\theta = -a\cosh\!\left(\frac{x}{a}\right) \]an upside-down catenary, the same curve a chain makes when it hangs between two nails.
- Find where each bump ends. The corner arrives at \( \theta = \pi/4 \), so \( x = a\ln(\sqrt2 + 1) = a\sinh^{-1}(1) = 0.8814\,a \). One bump spans \( 1.7627\,a \), and its height from trough to crest is \( a(\sqrt2 - 1) = 0.414\,a \). (Worked out here from the no-slip condition used by Slavík and Wagon, 2026.)
- Check the corner fits. At the trough, the road slopes down at \( \tan^{-1}(\sinh 0.8814) = 45° \) on each side, a 90° V. The square's corner is also 90°, so it nests exactly in the valley and hands over to the next side without a bump. (Worked out here from the no-slip condition used by Slavík and Wagon, 2026.)
The same recipe works for other regular polygons: Peterson's Science News account of Wagon's work notes that pentagons and hexagons roll on shorter, flatter catenary pieces, and in the limit of infinitely many sides the road becomes flat. Triangles fail, because as an equilateral triangle rolls over one arch its corner strikes the next one.
The authors revisit Robison's 1960 equations linking a wheel's shape to the road it rolls smoothly on, such as the square on the catenary. Robison's derivation leaned on assumptions about what rolling means that are not obvious; this paper derives the road equations from one assumption only, that the wheel rolls without slipping. That makes the method general enough to handle wheels that are continuous but nowhere differentiable. It is 11 pages with 19 figures, revised to version 3 in May 2026. The limit: it is pure kinematics, so it says nothing about the forces, friction or speeds worked out in this post.
Read the paper (free)
Does the ride feel smooth, and what does the rider feel?
The ride is smooth vertically because the axle never moves up or down, but the wheel's spin speed pulses four times per revolution, and that energy has to come and go somewhere. Since the contact point is straight below the axle at depth \( a\cosh(s/a) \), where \( s \) is the axle's horizontal offset from the nearest bump crest, the wheel's angular velocity at a steady road speed \( v \) is:
So the wheel spins fastest over a crest and \( \sqrt2 \) times slower at a corner. Its rotational kinetic energy \( \tfrac12 I\omega^2 \) swings by a factor of two every quarter-turn. Macalester's own description of Wagon's trike catches the feel: forward motion is almost linear with pedalling, but not quite.
At motorcycle scale the swing is large. For a 12 kg square wheel of half-side 0.32 m, treated as a flat plate with \( I = \tfrac23 m a^2 = 0.82 \) kg·m², the peak rate of energy flowing in and out of the wheel works out to \( 0.385\, I v^3 / a^3 \).
With these example inputs, that is about 5.6 kW at 30 km/h and 45 kW at 60 km/h, more than twice an NS200's rated output (24.5 PS, about 18 kW, in Bajaj's manual), sloshing back and forth through the tyre, chain and frame dozens of times a second.
Calculated in this post from 0.385 I v³ / a³ with the example inputs I = 0.82 kg·m² and a = 0.32 m (a 12 kg wheel): 5.6 kW at 30 km/h and 45 kW at 60 km/h.
What is the speed limit for square wheels?
The speed limit is set by friction, not by bouncing: a square wheel on its catenary road never loses contact, but the grip it needs along the road surface sits right at the limit of rubber (μ ≈ 1) at walking pace near the corners, and passes it again above roughly 50 to 70 km/h just off the crests. This is my own analysis; here it is in full.
Assumptions and derivation
Assumptions. The axle carries a load \( W \) (bike, rider and wheel weight) and moves at constant speed \( v \); the frame is heavy enough to hold that speed. The wheel has moment of inertia \( I \). Let \( c = \cosh(s/a) \), \( \sigma = \sinh(s/a) \), and define the inertia force \( K = I v^2 / a^3 \).
- Torque needed. The wheel must be given angular acceleration \( \alpha \), and the only force with a lever arm about the axle is the contact force, acting a distance \( d = ac \) below it. So its horizontal part is \( F_x = I\alpha / d = K\sigma / c^3 \).
- Resolve along the slope. The road at the contact point slopes at \( \tan\phi = \sigma \), so \( \cos\phi = 1/c \). Splitting the contact force into a normal part \( N \) and a friction part \( f \), and using vertical balance \( F_y = W \):
\[ N - f\sigma = Wc, \qquad N\sigma + f = \frac{K\sigma}{c^2} \]
- Solve the pair.
\[ N = \frac{W}{c} + \frac{K\sigma^2}{c^4} > 0, \qquad f = \sigma\left(\frac{K}{c^4} - \frac{W}{c}\right) \]\( N \) is always positive, so the wheel never lifts off. The grip needed is \( \mu_{\text{req}} = |f|/N \).
- Walking pace (\( K \to 0 \)). \( \mu_{\text{req}} = \sigma = \tan\phi \), which reaches 1 at the corners where the slope is 45°. Risch's classroom derivation puts good road tyres on clean, dry tarmac at about \( \mu = 1 \), so a slow square wheel is right at the edge of slipping each time a corner approaches.
- Medium speed. As \( K \) grows, the inertia term cancels the slope term at the corner; at \( K = 2\sqrt2\,W \) the corner needs no friction at all. Across all positions, the worst-case \( \mu_{\text{req}} \) bottoms out near 0.2 when \( K \approx 2W \).
- High speed. When \( K \gg W \), the demand moves to just off the crests, where \( \mu_{\text{req}} \approx \tfrac12\sqrt{K/W} \). Setting that equal to \( \mu \) gives a rule-of-thumb limit:
\[ v_{\max} \approx 2\mu\sqrt{\frac{W a^3}{I}} \]
The numbers for a motorcycle
Numbers. Take the rear of a motorcycle: \( W = 0.55 \times 238 \times 9.81 + 12 \times 9.81 \approx 1{,}400 \) N, \( a = 0.32 \) m (a square about as tall as a 17-inch wheel), \( I = 0.82 \) kg·m². With these example inputs, the rule of thumb gives \( v_{\max} \approx 2 \times 1 \times \sqrt{1400 \times 0.0328 / 0.82} = 15 \) m/s, about 54 km/h.
Calculated in this post from the formula above and the example inputs (W = 1,400 N, a = 0.32 m, I = 0.82 kg·m², μ = 1, the practical tyre maximum in Risch, 2010).
Scanning every position numerically, the exact grip demand crosses 1 at about 70 km/h, so the rule of thumb is conservative. Two things move the limit:
- if the wheel's mass sits at its rim (a square frame, \( I = \tfrac43 m a^2 \)), the limit drops to about 50 km/h;
- a heavier rider raises \( W \) and the limit with it; a heavier wheel lowers it.
A square wheel never leaves the road. It runs out of grip instead, at walking pace and again at about 50 to 70 km/h.
See the data as a table
| Plate wheel, I = 0.82 kg·m² (μ) | Rim-heavy wheel, I = 1.64 kg·m² (μ) | |
|---|---|---|
| 10 | 0.907 | 0.822 |
| 20 | 0.674 | 0.439 |
| 30 | 0.39 | 0.302 |
| 40 | 0.243 | 0.651 |
| 50 | 0.49 | 0.984 |
| 60 | 0.734 | 1.298 |
| 80 | 1.192 | 1.89 |
| 100 | 1.619 | 2.456 |
| 120 | 2.029 | 3.009 |
| 140 | 2.428 | 3.553 |
Source: Author's calculation from the rigid-body model above (a = 0.32 m, wheel 12 kg, axle load 1,400 N) · Speeds in km/h. Above μ = 1 a road tyre slips (Risch: practical tyres reach about 1 on dry, clean roads). Model ignores tyre compliance and pedalling torque.
A physics-education paper that derives a vehicle's limits from free-body diagrams of the wheels: maximum acceleration and braking equal the static friction coefficient times g, the top speed in a flat turn follows from friction against centripetal force, and the "friction circle" sets what can be combined. It states that practical road tyres reach a static friction coefficient of about 1 on a dry, clean surface, that kinetic friction is about 15% lower, and works a wet-road example at 0.5. The limit: it is a teaching derivation with idealised tyres, not a measurement; real peak values vary with rubber, temperature and surface.
Read the paper (free)How big would the road be for a motorcycle?
For a square wheel with a 0.32 m half-side, every bump is 0.564 m long and 0.133 m high, so the road needs about 1,773 bumps per kilometre. Wagon's road had to be built to match his wheels; yours would too, which makes the road the expensive part.
| Quantity | Formula | a = 0.32 m (motorcycle) |
|---|---|---|
| Bump length | \( 2a\sinh^{-1}1 \) | 0.564 m |
| Bump height | \( a(\sqrt2 - 1) \) | 0.133 m |
| Bumps per km | 1000 / length | 1,773 |
| Bumps for 118 km (Bengaluru–Mysuru expressway length) | - | about 209,000 |
| Corner hits per second at 30 km/h | v / length | 14.8 Hz |
| Same at 60 km/h | v / length | 29.5 Hz |
Use this to check what you just read: change the wheel size, mass and load, and watch the bump count and speed limit move.
Design your square-wheel road
Could a square-wheeled motorcycle turn a corner?
No, not on an ordinary road, and this is the real reason round won. The catenary road only works if the wheel meets each bump square-on and in phase. Steer a few degrees and the wheel crosses the bumps at an angle, so the effective bump length it sees changes and it falls out of step; Peterson's report notes that turning the trike's wheels too far knocks them out of sync. On top of that:
- Both wheels must be in phase. The wheelbase has to be a whole number of bump lengths, or the rear wheel climbs while the front descends.
- Reversing is fine, stopping is fine. Because the axle height is constant everywhere, the bike has no preferred resting position; it rolls freely in neutral, just as a round-wheeled bike does on the flat.
- Curves need curved bumps. A circular track would need wedge-shaped catenary bumps fanning out, matched to one radius only.
- Road wear. Every corner presses into the same valley, so the valleys wear first and the fit degrades. Macalester had to rebuild its trike in 2004 when the original wore out; the replacement's ride is so smooth that riders can't tell the wheels aren't round.

Verdict: possible with changes
Square wheels work exactly, not approximately, on a road of inverted catenaries \( y = -a\cosh(x/a) \), each 1.763a long.
The physics then sets a speed window: at walking pace the corners need a friction coefficient of about 1, the sweet spot sits around 20-40 km/h where the grip needed falls to about 0.2-0.4, and the wheel's own rotational inertia pushes the demand back past μ = 1 at roughly 50-70 km/h for a motorcycle-sized square (with these example inputs). Straight-line only, on a road 209,000 bumps long from Bengaluru to Mysuru. The version that works today is a short purpose-built track, like Wagon's.
Calculated in this post from the example inputs above, with the 118 km Bengaluru–Mysuru Expressway used for the bump count (118 km / 0.564 m = about 209,000) and Risch, 2010 for the tyre limit.
Square wheels make a good companion to the real physics of round ones: how tyres grip in countersteering explained, how steering geometry keeps a bike stable in rake and trail, and how traction is managed in traction control and cornering ABS. For the first motorcycle ever built, with very round wheels, see the Daimler Reitwagen. More engineering curiosities live on the Yantra page.
What's new
- Slavík and Wagon derive the road-and-wheel equations from the no-slip condition alone arXiv, April 2026 (revised May 2026)
- A YouTuber's "square-wheeled" bike that cheats with chain tracks around each wheel Hackaday, Apr 2023
What if next?
- Could a motorcycle run on chai? Burning it fails on the first law; fermenting it gets you about a kilometre per litre.
- How many people pedalling equals one NS200? Over 40 elite riders for its power, but two can match its top speed.
- What if a motorcycle were the size of an ant, or an elephant? The square-cube law decides which one falls apart.
Questions people ask
Do square wheels actually work?
Yes, on a matching road of inverted catenary bumps. Stan Wagon's square-wheeled trike at Macalester College has been ridden by the public since 1997, rebuilt in 2004.
What shape road do you need for square wheels?
A series of inverted catenaries, \( y = -a\cosh(x/a) \), each one 1.763 times the half-side long, meeting in 90° valleys where the square's corners sit.
Why can't triangles roll on a catenary road?
A triangle's corners are too sharp: as it rolls over one arch, its leading corner hits the next arch before the side can hand over. Squares, pentagons and polygons with more sides all work.
Could a car have square wheels?
On a dedicated straight track, yes, with all wheels in phase. The road would need bumps sized to the wheels, and the same friction window applies, scaled by the car's wheel size and load.
Sources
- Slavík A., Wagon S.: A Minimalist Approach to Rolling Wheels, arXiv:2604.03749 (2026) (no-slip derivation of road equations, square on catenary)
- Macalester College: A Square Wheel, or The Ultimate Flat Tire (1997 build by Wagon and Kellen, 2004 rebuild, almost-linear forward motion)
- Peterson I.: Riding on Square Wheels, Science News (1998, updated 2004) (steering problem, triangles fail, polygons and flat-road limit)
- Su F.E. et al.: Bike with Square Wheels, Harvey Mudd Math Fun Facts (Hall and Wagon, Mathematics Magazine 65, 1992, reference)
- Risch M.: Can Physics Teaching be improved by Explanation of Tricks with a Motorcycle? arXiv:1008.5041 (2010) (tyre friction coefficient about 1, kinetic about 15% lower)
- Hackaday: Square-wheeled bike is actually an amazing tracked build (2023)
- Outlook India: Bengaluru–Mysuru Expressway, 118 km (2023) (length used for the bump count)
