One rack and every wheel

Pick any two involute spur gears of the same module and pressure angle, at any tooth counts, and they will mesh. That is such a familiar fact that it reads as a convention — a standard somebody agreed, like thread pitches — and it is not. It is a theorem about a generating process, and the process is a straight edge rolling on a circle.

Why one tool is enough #

The rack has a module, a pressure angle and a tooth proportion, and nothing else. When it is rolled on a wheel of pitch radius $r$, the only number that changes is how fast it advances — $r$ millimetres per radian — and the flank that comes out is the involute of $r\cos\alpha$.

So every wheel the rack cuts has a base circle proportional to its own pitch radius, with the same constant of proportionality $\cos\alpha$. Put two of them together at the sum of their pitch radii and the common tangent to the two base circles makes the angle $\alpha$ with the common tangent to the pitch circles, passes through the pitch point, and is therefore a valid line of action for the pair. They are conjugate, and the ratio is the ratio of the tooth counts.

Measured over five wheels cut by one rack — 12, 17, 24, 40 and 60 teeth — every generated point lies on the involute of its own base circle to $1.4\times10^{-14}$ mm. The tool did not know how many teeth it was cutting, and could not have: the tooth count appears in the process only through the rolling rate.

That is interchangeability, and it is a property of the generation rather than of the involute. A cycloidal system does not have it. Its flanks are traced by a describing circle rolling between two specific pitch circles, so a wheel is cut for the mate it will run with, and a wheel cut for one pinion is not conjugate to a different one. Clockmakers accepted that; a factory could not.

The claim, stated so it could fail #

“Any two wheels of the same module mesh” is the sort of statement that is easy to nod at and hard to test, so it is worth putting in a form that a measurement could refute.

Take the rack. Generate the flank on wheel A and, separately, on wheel B. Assemble the two wheels at the sum of their pitch radii and ask whether there is a contact at every position of the input, with the common normal through the pitch point and a constant ratio. If the generation on A were subtly different from the generation on B — a rack whose flank angle varied with the rolling rate, say, or a tooth thickness that drifted — the pair would fail that test even though each wheel had been cut correctly.

It does not fail. The flanks are involutes of $r_A\cos\alpha$ and $r_B\cos\alpha$ to fourteen places, and any two involutes whose base circles are in the ratio of their pitch radii are conjugate at that centre distance. The chain of reasoning has one weak link and it is the measured one: that the generated curve is the involute of that particular circle, on every wheel. Everything else follows without arithmetic.

What actually has to match #

Two wheels mesh if the rack that cut them was the same. Reading that carefully gives the list of things that have to agree, and it is shorter than it looks.

The module — the tooth size — must match, because it is the rack’s pitch and two racks of different pitch do not lay teeth in the same places.

The pressure angle must match, because it is the slope of the rack’s flank.

The addendum proportion must be compatible, though not identical: it decides where each flank ends, and therefore the contact ratio and whether either wheel’s tip reaches past the other’s base circle.

The tooth counts need not match anything. Any two work.

The centre distance need not be the sum of the pitch radii. The pair is conjugate at any distance, as the centre-distance measurement shows; what changes is backlash and contact ratio.

That is why a gear standard is a description of a rack. Standardising the tool standardises every wheel that any workshop will ever cut with it, without anybody having to standardise a curve or a wheel.

An accident of history that was not one #

It is worth noticing how much of the modern arrangement follows from the decision to standardise the tool, because the alternative was live for a long time.

Through the nineteenth century gears were made to fit their mates. A wheel and its pinion were cut together, matched, and stayed together; a replacement was made against the surviving half. That is a perfectly workable system for a mill or a clock, and it makes every gear a bespoke part with an identity.

Interchangeable manufacture needs the opposite: a part defined by a specification rather than by its neighbour. The involute system supplies it, and it supplies it through the process — because a wheel cut by the standard rack is conjugate to every other wheel cut by the standard rack, whether or not the two were ever in the same building.

So the reason a gear can be ordered from a catalogue is the same reason a rack cuts every tooth count: one tool, one geometry, one definition. The bearing, the fastener and the pipe thread arrived at interchangeability by agreeing dimensions. Gearing arrived at it by agreeing a motion, and it is the more elegant route because the dimensions then follow.

Shift is the same tool, held out #

The one wheel the standard rack is awkward with is a small one, because below about seventeen teeth the cutter’s corner comes back through the flank. The cure keeps the tool and moves it.

Hold the rack $x\,m$ further from the blank’s centre while rolling it exactly as before. The pitch line still rolls on the pitch circle, so the rolling rate is unchanged, and the flank is still the involute of the same base circle — a shift does not change the base circle, because the base circle depends on the pitch radius and the pressure angle and nothing else. What changes is which part of that involute ends up on the tooth: the tooth becomes thicker at the root and thinner at the tip, and the corner no longer reaches past the point where the involute begins.

The smallest shift that clears the undercut, bisected from the generated geometry:

The right-hand column is $1 - z\sin^2\alpha/2$, computed by a different part of this site for its own purposes; the left is a bisection on where the cutter’s tip line crosses the line of action. They agree to $3.6\times10^{-13}$ of a module, and neither knows about the other.

What shift costs #

Nothing in geometry is free, and the teeth field has measured the trades. Three of them matter here because they are consequences of the tool having moved rather than of the tooth having changed.

The tip gets thinner. On an eleven-tooth wheel shifted by the minimum that cures its undercut, the tooth thickness at the tip falls from 0.606 mm to 0.240 mm. Push the shift further and the tip comes to a point; a pointed tooth cannot be hardened without burning the edge, so there is an upper limit on shift that is a manufacturing statement rather than a geometric one.

The pair no longer runs at the sum of its pitch radii. Two wheels shifted outward have thicker teeth at the pitch circle than the standard spacing allows, so they must be separated. For a 13 : 31 pair with the pinion shifted by 0.4, the operating centre distance rises from 22 to 22.377 mm and the operating pressure angle from 20° to 22.50°. The ratio does not move: it is still 2.3846, because the ratio is the base circles and the base circles have not changed.

Shifts can be compensated. Shift one wheel out by 0.4 and the other in by 0.4 and the operating centre distance is exactly the standard 22 mm again, with the pressure angle back at 20°. That is how a small pinion is saved from undercutting inside a gearbox whose centres were fixed before anybody counted teeth — and it is the reason a drawing quotes shifts in pairs.

Why the module has to match, and what a module is #

The one number that cannot be negotiated deserves a paragraph, because it is defined in three equivalent ways and the equivalence is the interesting part.

The module is the pitch diameter divided by the tooth count — millimetres of diameter per tooth. It is also the tooth pitch divided by $\pi$, measured round the pitch circle. And it is a dimension of the rack: the rack’s tooth pitch is $\pi m$, and its tooth stands $m$ proud of its pitch line.

Those three are the same number because of the rolling. The rack advances one tooth pitch while the wheel turns through one tooth’s worth of angle, which is what rolling without slipping means when both bodies carry teeth. So the module is simultaneously a property of a wheel, of a pair, and of a tool — and standardising it standardises all three at once.

Two wheels of different modules do not mesh, and the failure is immediate rather than subtle: their teeth are different sizes, so a tooth of one arrives where a tooth of the other already is. There is no approximate version and no centre distance that fixes it.

The rack is a wheel with no centre #

There is one more consequence of the generation, and it is the reason a rack appears in this argument at all rather than being a special case bolted on.

A rack is the limit of a wheel whose pitch radius has gone to infinity: its pitch circle is a straight line, its base circle is a straight line, and its involute — the involute of a line — is a straight line at the pressure angle. Everything in the standard is the wheel that this limit produces.

So the sentence “one rack cuts every wheel” and the sentence “a rack meshes with every wheel” are the same sentence. The cutter is a gear, the workpiece is a gear, cutting is meshing with the material removed, and the reason a hobbed wheel meshes with everything else the hob has ever cut is that all of them mesh with the hob.

That is worth holding on to, because it explains the shape of the whole standard. A gear specification is a specification of one member of the family — the infinite one — and every finite member is defined as whatever meshes with it.

What the hob adds, and what it does not #

The rack in these figures is an idealisation of a hob, and the difference is worth stating so the argument is not read as being about a machine nobody uses.

A hob is a worm with cutting edges ground into it — a rack wrapped round a cylinder. Instead of a rack sliding, two rotations are coupled: the hob turns, the blank turns, and their ratio is set by the tooth count so that the effective rack advances at the pitch radius per radian. The teeth are cut a slice at a time as the hob is fed along the blank’s axis.

Geometrically nothing changes. The generating profile is still the rack’s straight flank, the rolling condition is still the same rolling condition, and the flank is still the involute of the same base circle. What the hob adds is helical geometry — the cutting edges lie on a helix, so a real hobbed flank has a small deviation from the planar model that depends on the hob’s lead — and a manufacturing consequence: one hob cuts continuously, so the process is fast, and the same hob cuts helical gears by tilting it.

What it does not add is a new definition. If it did, the interchangeability claim would be about a machine rather than about a geometry, and a wheel cut on a shaper would not mesh with one cut on a hobber.

Where the interchangeability stops #

Three qualifications, because “any two wheels mesh” is a claim with edges.

Different tooth proportions still mesh, but not always well. A stub-toothed wheel and a full-depth one of the same module and pressure angle are conjugate — the flanks are involutes of the same base circles — but their contact arc is set by the shorter pair of tips, and the contact ratio may fall below the useful range.

Shifted wheels mesh with unshifted ones, at a centre distance that is neither wheel’s nominal. That is fine and it has to be known: a shifted pinion dropped into a housing bored for standard centres runs with reduced backlash and a higher contact ratio, and if the shift is large enough the teeth interfere.

Internal gears are the same family with a sign. An annulus is cut by a shaper rather than a rack, its flank is the involute of a base circle, and it meshes with anything of the same module — but its own undercut and interference conditions are different, because its centre is on the other side of the contact.

None of those is an exception to the theorem. Each is a reminder that conjugacy is one property among several, and that the other properties are what decide whether a pair works.

What a workshop actually stocks #

The practical form of the theorem is worth spelling out, because it is the reason the geometry mattered commercially.

With a generating process, a workshop stocks one cutter per module and pressure angle. Everything else — every tooth count, every shift, every helix angle, every face width — is a setting on the machine. A shop with a dozen hobs can make any spur or helical gear anybody asks for within its size range.

With form cutting, the shape of the gap depends on the tooth count, so in principle a shop needs one cutter per tooth count per module. In practice it stocks a set of eight per module, each of which is approximately right for a band of tooth counts and exactly right for one of them. Wheels cut that way are approximately conjugate — the transmission error is small and real — and the compromise is invisible until somebody measures it.

That is the whole commercial argument, and it is a geometric one: the generating process makes an exact tooth for every wheel from one tool, and the alternative makes an approximate tooth from many.

The idea, in one sentence #

A standard that describes a product has to enumerate; a standard that describes a process does not.

The involute system standardises the cutter, and every wheel any workshop makes with it is compatible with every other, forever, without a list. That is a rarer property than it sounds — it needs the process to define the product uniquely, which is exactly what generating an envelope does — and it is the reason this field’s first essay puts the rack at the top of its ledger.

The rung’s claim has a form worth stating for its own sake, because it is unusual among this site’s results. Most of what is computed here is a property of a mechanism: this linkage’s transmission angle, that cam’s pressure angle, this platform’s workspace. Interchangeability is a property of a standard — a statement about which parts, made by different people at different times, will work together — and geometry is not usually asked to settle questions of that kind. Here it does, and completely: two wheels mesh if their module and pressure angle agree, because those two numbers are the tool, and the tool is the definition. So a gear standard can be written as a description of a cutter rather than as a family of curves, a workshop can stock cutters rather than gears, and a wheel cut in one century meshes with one cut in the next. A geometric property became an industrial institution, and the essay’s whole argument is that the second followed from the first rather than being laid on top of it.