Why does a curling stone curl?

Nobody is certain. Four mechanisms are live in the physics literature: front–back friction asymmetry, pivot–slide, scratch-guide, and a velocity-dependent left–right asymmetry. Each reproduces some of what a real stone does and none has been shown to reproduce all of it. A 2013 paper in Tribology Letters showed that no redistribution of friction under the stone, however extreme, can account for the observed motion, which ruled out the simplest family of explanations.

Which way does a curling stone curl?

In the direction it is turning: a stone rotating clockwise, seen from above, bends to the right. This is the opposite of what most sliding objects do — a tumbler slid upside-down across a table with a clockwise spin bends to the left — and that reversal is the whole puzzle.

Does spinning a curling stone harder make it curl more?

Almost not at all. Across the normal range of two to five rotations over the length of the sheet, total curl is close to flat in rotation rate. It falls away only for a stone that is barely turning, which is why a stone set moving by a collision runs nearly straight. Any theory of curling has to reproduce this, and it is the fact that most of them find hardest.

How far does a curling stone curl?

Published physics measurements put a draw at roughly 0.75 to 1.5 metres of deflection, around a metre being typical. Working ice technicians report five to six feet on championship-prepared sheets, which is considerably more. No published study has measured a championship sheet, so the disagreement is unresolved rather than settled in either direction.

When does a curling stone do most of its curling?

At the end. A stone lays down roughly 15% of its total curl in the first half of its slide and about half of it in the last five metres. This is why a come-around only works if the weight is right: a stone thrown heavy arrives before the curl does.

CurlTi.me/Learn/Why stones curl

Why a curling stone curls

Spin a tumbler across a table and it bends one way. Spin a curling stone down a sheet of ice and it bends the other. Physicists have been arguing about this since 1924, and — despite what most articles about it will tell you — they have not finished.

Last updated 2026-08-18

What actually happens#

A curler releases the stone with a slow rotation — two to five full turns over the length of the sheet is normal, and more than that is a fault rather than a tactic. The stone travels almost straight for most of its journey and then, over the last few metres, slides sideways by about a metre and stops. Which way it goes is decided entirely by which way it is turning.

≈1 msideways deflection on a draw, in published measurements
2–5full rotations over the length of the sheet
15%of the total curl laid down in the first half of the slide
1924the first paper. Still not settled.

Curlers name the two rotations after the way the hand turns, and the names are the wrong way round from every other sport's: the in-turn for a right-handed curler is clockwise and bends the stone to the right, and the out-turn is anticlockwise and bends it left. See the shots for what each is used for.

Why that is the wrong way round#

Take an ordinary drinking glass, turn it upside-down, and slide it across a table with a clockwise spin. It bends left. This is not an obscure result — it is what nearly every spinning object slid across a surface does, and the reason is straightforward.

A decelerating object tips very slightly forward, so the front of its contact presses hardest and the friction there is the strongest anywhere under it. Friction acts against the way that part of the object is moving. On something spinning clockwise, the front of the contact is sweeping to the right — so the friction on it pushes left, and because that contact is the loaded one, its push wins. The object drifts left: opposite to the spin.

Same object, same spin, opposite outcome. The tumbler does what the textbook says. The stone does not.

A curling stone, given the same clockwise spin, bends right. Not slightly less to the left — the other way. Something about a granite ring on pebbled ice reverses the sign of the whole effect, and a hundred years of papers have been about what.

The puzzle in one sentence

The question is not why a curling stone bends. It is why it bends in the direction of its rotation when almost nothing else does.

The four facts any theory must fit#

This is what makes the problem hard. Producing a sideways force of about the right size in about the right direction is easy, and many models manage it. Producing all four of the following at once is what nothing has convincingly done.

The observation	Why it is awkward
Direction. The stone curls the way it is turning.	Rules out every model that works the way a sliding tumbler works, which is most of the intuitive ones.
Magnitude. About one metre over a thirty-metre slide — a curl ratio around 1:30.	Small enough that a mechanism has to be subtle, large enough that it cannot be a rounding error in something else.
Flat in spin. A stone turning twice curls about as far as one turning five times.	Brutal. Almost any force that depends on rotation gives curl that grows with rotation. This one does not.
Back-loaded. Roughly 15% of the curl in the first half; about half of it in the last five metres.	Rules out models whose sideways force is a constant fraction of the forward force, which give a parabola.

The shape of a real stone’s path against the parabola the early pivot–slide model predicts, normalised to the same total curl. A curler feels this difference as the stone “taking” late.

The third of these deserves emphasis, because it is counter-intuitive and it is the reason the handle on this site is a two-way switch rather than a dial. Turning the stone harder does not bend it further. It is a direction control, not a magnitude control, and every curler learns this by being told rather than by working it out.

What is actually touching the ice#

Before the theories, the geometry, because the arguments are all about a region most people picture wrongly.

A curling stone is a granite disc about 284 mm across and 19.0 kg in mass, and it does not rest on its face. Its underside is hollowed, and it stands on a narrow annulus — the running band — about 130 mm in diameter and between 6 and 13 millimetres wide. That is a contact area of roughly 26 cm², about the footprint of a coffee cup.

And it does not even get that, because curling ice is not flat. It is sprayed with droplets that freeze into pebble, and the stone rides on the tops of them. The real contact at any instant is a scatter of points, each a fraction of a square millimetre, that changes continuously as the stone moves. All of the physics on this page happens in those few dozen points.

The underside of a stone, to scale. The white ring is the running band; the red marks are roughly what is actually touching pebbled ice at any one moment.

Why this matters for the argument

Several of the theories below only work if the contact is discrete — a sequence of separate touches rather than a continuous smear. That is not a convenient assumption invented to save a model; it is what pebbled ice physically is, and it is the reason curling ice is pebbled in the first place. See how curling ice is made.

The four surviving theories#

Every one of these has published support, and every one has a published objection that has not been answered to everyone's satisfaction. They are listed in the order they arrived.

Front–back friction asymmetry Shegelski and others, 1996 onward; Denny 1998, 2002; Maeno 2010, 2014
A sliding stone leans very slightly on one edge, so the running band presses harder on the ice at one side of the contact than the other. Make the grip stronger at the back than at the front and the resulting torque drags the stone sideways in the direction it is turning. It is the tidiest story, it gets the direction right, and it dominated the literature for twenty years.

Against: The asymmetry it requires is enormous — far more than a stone weighing 19 kg on a flat floor could plausibly produce — and a stone that leaned that hard on its trailing edge would show it in other measurements. Murata’s 2022 tracking found the opposite lean, if anything, which would curl the stone the wrong way.
Pivot–slide Shegelski and Lozowski, 2016 onward; Mancini and de Schoulepnikoff 2019
Forget smooth sliding. The band catches on a high point, the stone pivots about that single contact for a few milliseconds, the contact breaks, and the stone slides on to the next one. Each pivot rotates the velocity vector by a hair. Thousands of them add up. Because a pivot displaces the stone by roughly the same amount regardless of how fast it is spinning, the model naturally predicts that curl barely depends on rotation rate — which is the observation that most theories struggle with.

Against: The early version predicted a parabolic path, and real stones are far more back-loaded than that. The 2024 revision moves the pivots onto individual microscopic asperities rather than whole pebbles, which fixes some of this and adds parameters that are hard to measure independently.
Scratch-guide Nyberg and the Uppsala group 2013; Penner 2019; Kameda and others 2020
The underside of a stone is not polished — it is deliberately roughened. The leading part of the band scores fine grooves into the tops of the pebbles, angled by the rotation, and the trailing part of the band arrives a fraction of a second later and is steered along them, the way a skate blade follows its own cut. Kameda measured those grooves: about 1 µm deep and 40 µm wide.

Against: It predicts that curl should depend strongly on how rough the stone’s underside is, which is testable and partly confirmed — but it also makes curl a property of the stone more than of the ice, and ice technicians who can turn a straight sheet into a swingy one with the same stones find that hard to accept.
Velocity-dependent left–right asymmetry Murata 2022
Friction on ice depends on sliding speed. On a rotating stone, one side of the running band is moving forward faster than the other, so the two sides sit at different points on the friction curve. Combine that with contact that happens at discrete points rather than continuously, and the stone swings about whichever contact is on the slow, high-friction side.

Against: The newest of the four and the least tested. It also needs the contact to be genuinely discrete — a continuous left–right difference cannot do it, which is the 1937 objection that has outlived everyone who made it.

The three marked with a green edge are the ones with active support in the literature as of 2026. The first is included because most popular explanations of curling still give it as the answer.

The experiment that ruled out the obvious answer#

The most useful single paper in this field is a negative result, and it is worth knowing about even if you read nothing else.

In 2013, Nyberg, Hogmark and Jacobson at Uppsala did something obvious that nobody had done: instead of proposing yet another distribution of friction under the stone and checking whether it produced a curl, they computed the trajectory for a wide range of possible distributions — including physically absurd ones — and asked whether any of them reproduced the motion of a real stone.

None of them did. No redistribution of friction under the running band, however extreme, produces the path a curling stone actually takes.

This is a strong result and it reframes everything before it. The mechanism cannot be simply "the friction is stronger over here". Something about the contact has to be doing work that a smeared-out friction map cannot represent — which is precisely why the surviving theories are all, in one way or another, about the contact being made of separate events.

Where it stands in 2026#

Two lines of work are live, and they are not obviously incompatible.

The measurement side#

Jiro Murata's 2022 study tracked a real stone at Karuizawa frame by frame with a resolution of about 47 micrometres, which is a different order of evidence from anything before it. He reports a left–right friction asymmetry that arises from the speed dependence of friction rather than from any lean, concludes that the stone swings about discrete contact points on its slower side, and — pointedly — finds no evidence for the large front–back asymmetry the older models need. The paper describes itself as solving a "mystery of the century", which is a bold framing that the field has not yet collectively accepted.

The modelling side#

The pivot–slide programme has kept refining. The 2024 asperity-based version moves the pivots from whole pebbles down to individual microscopic high points, which addresses the path-shape objection. The cost is parameters that cannot currently be measured independently, so the model fits well without that fit being strong evidence.

Meanwhile a 2025 paper proposes getting the sideways force out of friction that is not directionally biased at all, and a 2024 review situates the whole problem on the standard tribological picture of how friction varies with speed. A reasonable summary of the state of the art is: the contact is discrete, the speed dependence of friction matters, and the remaining argument is about which of those is doing most of the work.

If somebody tells you it is solved

They may be right, but ask which mechanism. Popular articles about curling physics tend to state one theory as fact, and which one depends almost entirely on the year the article was written. The literature has not converged, and it is more interesting that it has not.

Why any of this matters to a curler#

It is tempting to file this as a curiosity. It is not, for three reasons.

Directional sweeping#

If curl comes from scratches or from discrete contacts, then a brush head that polishes the ice unevenly — or leaves its own directional texture — can steer a stone rather than just make it go further. That is not hypothetical: it is exactly what happened in 2015, when a new broom head turned out to let sweepers move stones sideways in a way that the sport decided was not curling any more. The whole story is on brooms, brushes and Broomgate, and the reason the equipment rules are now as prescriptive as they are is that nobody could say from first principles how much steering was too much.

Ice that swings, and ice that does not#

Ice technicians can make a sheet curl four feet or six, deliberately, with the same stones. That is a strong practical argument that the ice matters as much as the stone — and it sits awkwardly with the scratch-guide result that curl is set mainly by the stone's own roughness. Somebody is measuring something different from somebody else. See ice speed and what to expect from a sheet.

Prediction#

Every simulator, every shot-analysis tool and every broadcast graphic has to commit to a curl model. Ours is described below. None of them can claim to be derived from first principles, because the first principles are the thing in dispute.

What this simulator does, and why#

We had to pick something, and it is worth being explicit that what we picked is an engineering choice fitted to observed behaviour, not a position in the physics argument.

The engine gives a stone a sideways velocity of

v_lateral = (β / 2) · v · ln(v₀ / v)

where v₀ is the release speed, v is the current speed, and β = 0.151. The logarithm is what makes the curl back-loaded: while the stone is quick, ln(v₀/v) is near zero and the path is nearly straight; as the stone slows the term grows without bound and the stone slides sideways. Integrated over a draw to the button this produces about a metre of deflection, which is the middle of the published range.

Alongside it runs the friction law the rest of the simulation rests on:

μ(v) = 0.00662 + 0.00231 / √v

Friction rises as the stone slows. That is the correction that makes real curling timings reproducible — a constant coefficient cannot give a correct hog-to-hog split and a correct total distance at the same time — and it is the same speed dependence that Murata's account of the curl mechanism turns on. The two are consistent, which is reassuring without being evidence.

Three deliberate consequences, each of which is a claim you can check by playing:

Curl is flat in rotation rate across the normal range, so the handle is a two-way switch. It falls off only for a stone barely turning.
A struck stone runs straight. No spin is transferred in a collision — the standard modelling choice, and there is no published measurement of stone-to-stone spin transfer to do better with — so a stone set moving by a hit has no rotation and therefore no curl.
How much a sheet swings is a setting, not a constant. The published measurements and the ice technicians disagree, so rather than pick a side the game ships both: the calibration sheet curls the measured amount and the championship preset curls 60% more.

The full literature, annotated Throw one and watch it take

Sources#

Every claim on this page is traceable to one of these. The full annotated bibliography, including the papers that argue with each other, is on the research library page.

Calculated trajectories of curling stones sliding under asymmetrical friction: validation of published models 2013The most destructive paper in the field. It computes the trajectory for a wide range of possible friction distributions under the stone and finds that none of them, however extreme, reproduces what a real stone does. A surviving theory has to be about something other than merely where the friction sits.
Study of curling mechanism by precision kinematic measurements of curling stone’s motion 2022Tracks a real stone frame by frame to sub-millimetre precision. Reports a left–right friction asymmetry arising from the speed dependence of friction, concludes that swinging about discrete slow-side contact points is the dominant mechanism, and finds no evidence for the large front–back asymmetry the older models require.
Pivot–slide model of the motion of a curling rock 2016The stone catches on a pebble, pivots briefly about that contact, then slides on. Reproduces observed curl distances and — crucially — reproduces the weak dependence of curl on how fast the stone is turning.
The asymmetrical friction mechanism that puts the curl in the curling stone 2013The Uppsala group’s scratch-guide result. Their demonstration is the memorable one: score the ice deliberately, and the scratches steer the stone.
The importance of the surface roughness and running band area on the bottom of a stone for the curling phenomenon 2020Measures the scratches: roughly 1 µm deep and 40 µm wide, cut into the pebble tops by the running band. Finds that curl distance is set mainly by the roughness and area of that band — which is a claim about the stone, not about the ice.
An examination of studies related to the sport of curling: a scoping review 2024The right place to start on any curling question that is not the curl mechanism. It maps the whole research landscape — delivery, sweeping, wheelchair curling, strategy, psychology, injury, facilities — and is open access.
The sports science of curling: a practical review 2009What sweeping actually does to the ice and to the sweeper, written for coaches rather than for physicists. Open access, and still the clearest short account of why sweeping works at all.
An analysis of curling using a three-dimensional Markov model 2019Builds win probability from eighteen years of Canadian championship data and then uses it to answer strategy questions. Its most-quoted conclusion is that a team with hammer facing the choice between taking one and blanking should usually blank.
The slippery science of Olympic curling: we still don’t know how it works The Conversation, 2022A readable summary of the tumbler-versus-stone contrast by researchers who are not partisans of any one model.