The area that is not in the equation

Friction is the one force in statics with an inequality where the others have an equation, and the inequality is $F \le \mu N$. Everything in this collection that slides, grips, jams, holds a bearing still or lets one go is decided by those four symbols, and one thing that ought to be in them is missing: there is no area anywhere in the expression.

That absence is not a simplification made to keep the arithmetic short. It is the finding, it was measured before it was explained, and the explanation — when it eventually arrived, two hundred and fifty years later — turned out to say something about contact that changes how every number on this page should be read.

What the law says, and what it does not #

The statement has two halves and both are experimental.

Friction is proportional to the normal force. Press twice as hard and the contact will resist twice as much before it lets go.

Friction is independent of the apparent area of contact. Lay the same block on its long face or its end, and the force needed to start it moving is the same.

A third clause is usually added — that the sliding force is independent of speed — and it is the weakest of the three, being roughly true over a narrow range and visibly false outside it. The two that matter are the first pair, and the second is the one that reads as though it must be wrong.

It reads that way because every other capacity in this collection is a stress times an area. A bolt in shear, a weld, a bearing pad, a strut: all of them carry a force that scales with how much material is presented to the job. Friction does not, and asking why is not pedantry. A rule with no area in it can be applied to a contact of any size, which is exactly what a designer does when the same coefficient is used for a 200 mm bearing plate and a 30 m raft, and it is worth knowing whether that is a licence the physics actually grants.

The paradox, stated so that it has to be answered #

Suppose friction were a shear strength $\tau$ acting over the contact area $A$. Then $F = \tau A$, and doubling the area would double the capacity. Nothing observed does that.

Now suppose instead that the resistance depends on the contact pressure, $\sigma = N/A$, in some way. Doubling the area halves the pressure. If the resistance per unit area happened to be proportional to the pressure, the two changes would cancel exactly and the total would not move — which is the observation.

That is the whole of the answer in outline: the shear resistance per unit area must be proportional to the normal pressure, and the question is why a material would behave that way, since no material’s shear strength depends on how hard it is being squeezed in anything like that proportion.

The resolution is that the area in $\sigma = N/A$ is not the area anybody draws.

Real contact against apparent contact #

Two nominally flat surfaces are not flat. Machined steel is rough at a scale of a micron or so, concrete at a scale of a millimetre, and when they are laid together they touch only where a high point on one meets a high point on the other. The load goes through those junctions and through nothing else.

Each junction is small, so the pressure on it is enormous — large enough that the material there yields. The junction flattens until it is big enough to carry its share at the material’s own indentation hardness $H$, which is the pressure at which it stops flattening. Summed over all of them, the real area of contact is

$$A_{\text{real}} = \frac{N}{H}$$

and it is a property of the load, not of the block. It grows in exact proportion to $N$, which is the missing proportionality.

Sliding then means shearing those junctions. If the junction material shears at $\tau$, the force required is

$$F = \tau A_{\text{real}} = \frac{\tau}{H} N$$

so that

$$\mu = \frac{\tau}{H}$$

The coefficient of friction is a ratio of two material properties, both of which are strengths, and the apparent area has vanished because it was never carrying anything. It has vanished for the same reason that a factor of two on the block’s footprint changes nothing: doubling the drawn area does not double the number of junctions carrying load; it halves the pressure on each, they flatten less, and the real area comes out the same.

The numbers are worth having. For steel on steel, $\tau$ is of order 200 N/mm² and $H$ of order 1,500 N/mm², giving $\mu \approx 0.13$ for clean surfaces — which is close to the measured value for lubricated steel and well below the 0.5 or so a rusty contact gives, because rust adds junctions that are not being sheared at the parent material’s strength. The real area at a modest load is a few thousandths of the apparent one.

Which free body produced the number #

The free body is the panel in the first figure, cut on the plane of contact, and the cut is the part of it worth insisting on.

Everything below the cut — the surface, its roughness, its asperities, its rust — is outside. What crosses the cut is one distributed traction, and the only two things statics is allowed to ask about it are its resultant’s magnitude and its direction. Resolving that resultant into a component normal to the plane and a component along it gives $N = W\cos\alpha = 44.5$ kN and $F = W\sin\alpha = 18.0$ kN, and the whole of the friction law is a statement about the angle between them.

Choosing that cut is the whole of the skill, and it is the choice that makes the area disappear: a cut taken through the plane of contact converts everything happening in the roughness into two numbers, and a cut taken anywhere else would have to describe the roughness.

That is why the cone in the first figure is the right picture and a pair of numbers is not. The contact can deliver a reaction in any direction within arctan $\mu$ of the normal, and no direction outside it. Equilibrium exists exactly when the reaction the rest of the free body demands lies inside that cone. Here the demanded reaction leans 22.0° against a cone of 31.0°, so it does.

Nothing in that statement mentions how big the cut is. The free body has a face and the face has an area, and the area is used to convert the traction into a resultant and then never appears again.

What the model leaves undecided #

Being a bound rather than an equation, the law leaves a range rather than an answer, and the range is not small.

On the flat the panel is in equilibrium under any longitudinal force between $-\mu W$ and $+\mu W$, and the friction force adjusts to whatever is required. The band is 57.6 kN wide, which is more than the panel weighs. A structure with friction in it does not have one state; it has a set of them, and which one it is in depends on how it got there.

The band narrows as the slope steepens, because the demand rises toward the capacity from one side, and it closes at 31.0° where the two meet. That is the angle of repose, and it is the one configuration in which friction is determinate — the point at which the ladder in the previous rung on this anchor has exactly one answer, and every gentler ladder has a range.

The exception that shows the mechanism #

If the coefficient really is $\tau/H$, then anything that changes either strength changes the coefficient, and there is one contact in ordinary structural use where that happens visibly.

PTFE creeps. Its junctions do not stop flattening at a fixed hardness the way a metal’s do, so the real area grows faster than in proportion to the load and the coefficient falls as the pressure rises. The standard fit is $\mu = 1.2/(10 + \sigma)$ with $\sigma$ in N/mm², and it is a description of a material whose $H$ is not constant.

The consequence for a designer is the reading in the figure, and it inverts the usual instinct. The largest force is at full load and is what the pier is sized for. The largest nuisance is at light load, where 46 kN of friction is nearly half of the wind the bearing was installed to let past — a roller that is not a roller, delivering its worst proportional restraint exactly when nothing is checking.

So the constancy of $\mu$ is a property of contacts whose junctions reach a definite hardness, which most structural contacts do and one important one does not. That is what an explanation buys over a correlation: it says where the correlation will fail before anybody measures it failing.

Pushing harder makes it worse, up to a limit that is a direction #

The same arithmetic run in a different direction produces the one result in this subject that looks like a trick and is not.

A force applied at an angle into the plane contributes $P\cos\beta$ to the drive and $P\sin\beta$ to the normal force, so the capacity gains $\mu P \sin\beta$. When $\mu\tan\beta > 1$ the capacity gains faster than the demand, and the block cannot be moved at all — not by a large force, not by any force.

The boundary is $\beta = 59.0°$ for $\mu = 0.6$, which is $90° - \arctan\mu$. This is self-locking seen from the outside: the wedge and the screw jack are devices built so that the driving force arrives inside that cone, and the reason they hold with the spanner removed is the same reason nothing here can be pushed loose. The condition contains an angle and a coefficient and nothing else — no weight, no size, no strength.

The coefficient inside a structural check #

Once the coefficient enters a real check it stops being a curiosity and starts deciding which of two failures happens.

Two failures come out of one free body, and only one of them contains $\mu$. Overturning is decided by weight and geometry alone; sliding is decided by weight and the coefficient. Because their factors fall at different rates with height, which one governs is a property of the shape rather than of the material, and it changes as the body gets taller.

The same split governs the ground under a footing. Bearing capacity is a mechanism in the soil whose whole strength is a friction angle plus a cohesion, and a foundation is checked against sliding on its underside with a coefficient that is a property of the concrete-to-soil interface rather than of either material — a pair of surfaces again, and one of them cast against the other.

That gives the coefficient an unusual role. It does not merely scale a capacity; it decides which capacity is being checked, and a value that is wrong by a factor of two can move the governing mode rather than the answer.

Two coefficients, and why the difference is a stability question #

One thing the two laws leave out is that the force needed to start a contact moving is usually larger than the force needed to keep it moving, and the gap is not a detail of the measurement.

The mechanism follows from the same picture. Junctions that have sat still have had time to creep, oxidise across and grow, so the real area at the moment of first movement is larger than the real area during sliding. The static coefficient is therefore a function of how long the contact has been at rest, which is a strange property for a design parameter to have and is why handbooks quote it with a caveat and codes quote a single value.

The structural consequence is stick-slip. If the static value exceeds the kinetic one, a contact being pushed slowly does not slide steadily: it holds, releases, accelerates, arrests, and holds again. The force in whatever is doing the pushing therefore oscillates between two values rather than settling at one, and the movement arrives in jerks rather than as a drift. A bridge bearing releasing a slow thermal movement does it in a series of small events, each of which is a load case nobody wrote down, and the same mechanism is what makes a door hinge squeal and a brake shudder.

It is also the reason a friction problem is a stability problem rather than only an equilibrium one. A capacity that falls the instant it is exceeded is exactly the shape that turns a limit into a collapse: there is no reserve past the peak, and the energy that was being stored has somewhere to go. That is the same structure as a load that makes itself worse, arrived at from a contact rather than from a geometry.

Where the coefficient is specified rather than measured #

Every number so far has been a measurement of a surface nobody controls. There is one structural contact where the opposite is true.

A slip-resistant connection is designed by choosing a surface class — blasted, blasted and metal-sprayed, wire-brushed, untreated — and each class carries a coefficient that the preparation is required to deliver. The surface is prepared to reach a number rather than measured to report one, and both halves of $\tau/H$ are being manipulated: blasting removes the mill scale that shears easily and roughens the profile so the junctions interlock.

Past the slip the joint is a different structure. The bolts stop being clamps and become dowels, the plates start bearing on the sides of their holes, and the capacity that follows is a stress times a projected area in the ordinary way — which is the sharpest possible contrast, because the same connection changes at 275.2 kN from a mechanism with no area in it to one made of nothing else.

It is also the clearest demonstration that the area is genuinely absent. The slip resistance of that joint is $n \mu F_p$ — a count of bolts, a coefficient and a preload. The plates could be twice as wide with no change whatever, and the only thing width buys is somewhere to put more bolts.

Where the model stops #

The junctions are not all sheared at once. The derivation treats $A_{\text{real}}$ as though every junction reaches its shear strength simultaneously. In a large contact they do not, and the difference is one of the reasons a big joint slips at a slightly lower average stress than a small one — the same statistical argument that makes a big one weaker than a small one.

Adhesion is left out. Very clean, very flat surfaces in vacuum weld themselves together and the model above says nothing about it. Structural surfaces are covered in oxide and dirt, which is why the neglect is safe here and not in general.

The coefficient is quoted to two figures and known to about one. Handbook values scatter by a factor of two between sources for the same nominal pair, because they are properties of a surface state that no specification fully controls. A design that turns on the difference between 0.5 and 0.6 is a design that turns on nothing.

Nothing here is a statement about how much a contact moves. The friction law bounds a force and says nothing about the displacement needed to reach the bound — which in a bolted joint is a fraction of a millimetre and in a bearing is several. The band in the third figure is a set of forces, not a set of positions.

And the whole of it is a rigid-body statement. The block does not deform, the surface does not deform, and the pressure distribution across the contact is never asked about — which matters the moment the contact is long enough to be flexible, where the ends slip while the middle has not.

The ladder from here #

Later rungs on this anchor: friction in three dimensions, where the bound stops being a pair of inequalities and becomes a cone, and the admissible set stops being an interval. The order of loading, and structures whose friction forces depend on how they were assembled rather than on what they carry. Friction dampers, where the bound is put into a structure deliberately in order to cap the force a member can attract. Static against kinetic coefficients as a stability question rather than a materials one. And the limit theorems for frictional systems, which are the general statement of why a problem with an inequality in it has a set of answers and not one.

Coulomb reached the memoir that carries his name in 1785, and the two laws in it were already a century old: Amontons published them in 1699, and Leonardo had both of them in a notebook around 1493 and told nobody. The mechanism waited until Bowden and Tabor’s work in the 1940s, which means the profession used the area-independence of friction for two and a half centuries without being able to say why an area that is obviously there should be absent from the arithmetic.