Involute gear teeth
A Socratic walk-through of involute gear teeth — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why must gear teeth be cut to one particular curve if the driven shaft is to turn at a steady speed?
A companion piece here on mechanical advantage settles the ratio: twelve teeth driving thirty-six means three input turns per output turn. That is exact — but it is an average over a whole revolution, since teeth mesh one for one and no profile can change the tally.
Which leaves a question averaging conceals. Inside a single tooth engagement, is the driven shaft turning at a constant speed, or speeding and slowing many times a second and merely arriving in the right place at the end? Cut the teeth to any curve and the count still comes out right. So why does the shape matter?
Reasoning it through
REASONING #Ask what contact between two flanks can do. Wherever two smooth surfaces touch, the only force they pass — friction aside — acts along the common normal at that point. Where that normal falls decides how much of the driving gear's motion becomes the driven gear's.
Draw the line of centres between the two shafts and mark the point where the common normal crosses it. Call it P. Since the two bodies must share the same velocity component along that normal, the ratio of their angular speeds is the inverse ratio of the distances from each centre to P — so the instantaneous ratio is fixed entirely by where P sits.
That states the whole problem. As the teeth roll, the contact point migrates and the common normal swings about; if P wanders, so does the ratio, and the driven shaft speeds and slows within every tooth. For a steady output, P must not move. That is the fundamental law of gearing, and it constrains the shape of the flanks before anything else: the common normal at the contact point must pass through a fixed point on the line of centres, at every instant of the engagement.
Which curves do that? Consider the involute of a circle — the path traced by the end of a taut string unwound from a spool. The string is the giveaway: at every position it is tangent to the spool and perpendicular to the path being traced. So the normal to an involute anywhere is a tangent to its base circle.
Put two such curves in contact, one wound from each gear's base circle. Their common normal must be tangent to both base circles at once — and there is only one such line, fixed in space whatever position the gears are in. A fixed line crosses the line of centres at a fixed point, so the ratio cannot vary. It comes out as the inverse ratio of the base radii, cut into the metal and unchangeable.
That last clause carries more weight than the theory suggests. Because the base circles belong to the gears rather than the assembly, pulling the shafts apart does not alter the ratio. It alters backlash, and it alters the pressure angle — the base radius is the pitch radius times the cosine of that angle, so with the base radius fixed, spreading the centres opens the angle up. The ratio is untouched, and a pair indifferent to a millimetre of centre error tolerates bearing wear, housing tolerance and thermal growth.
There is a second, more mundane reason involute won. Its mating rack — the shape it meshes with when one gear grows to infinite radius — has straight flanks, so one straight-sided tool rolled against a blank generates a correct involute for any tooth count at a given size and pressure angle. One hob cuts the whole family. The cycloidal profile, which also satisfies the fundamental law and survives in clockwork, has neither property — it needs its exact design centre distance, and its curve depends on the tooth count it runs against.
And the failure that was accepted? Contact is pure rolling only at the pitch point; everywhere else the flanks slide, and involute teeth meet convex against convex, concentrating contact more than the cycloid's convex-against-concave. Involute gearing buys interchangeability and tolerance of sloppy centres at the price of sliding wear, scuffing and pitting — which is why gear life is so often a lubrication question rather than a strength one.
The analogy
THE ANALOGY #Think of two people passing a rigid pole between them, each turning on the spot. The pole can only push along its length, so what matters is not where their hands are but which way it points. Let it sweep to different angles as they turn and one goes fast while the other goes slow, in a rhythm repeating every handover. Hold it at one angle and their rates stay locked.
the pole is one object held throughout, whereas a gear mesh hands off from one pair of flanks to the next while the previous pair is still in contact — and it is precisely that overlap, not the single contact, that keeps a real gear train quiet.
Clarifying the model
THE MODEL #The load-bearing claim is narrow and testable: a constant instantaneous ratio requires the common normal to pass through an unmoving pitch point, and the involute achieves it by making that normal the fixed common tangent of two base circles.
The test is routine: put an encoder on each shaft and plot output angle against the ratio-scaled input angle, the residual being the transmission error. The account predicts that a correctly cut pair shows only errors traceable to manufacturing deviation and tooth deflection, with no fluctuation intrinsic to the profile — and, more sharply, that increasing the centre distance changes backlash and pressure angle while leaving the ratio alone. The refuting observation would be a profile-intrinsic velocity ripple in an accurately made pair, or a measurable ratio change on shifting the centres.
One thing here is not physics. The 20-degree pressure angle in near-universal use is a standard, not an optimum — lower angles run more smoothly, higher angles give stronger roots — frozen where it is because interchangeability is worth more than the last few per cent of either. What is derivable is where the geometry stops working: with a full-depth tooth, the generating rack begins to undercut the flank root below a tooth count of two divided by the square of the sine of the pressure angle. At 20 degrees that is 2 / 0.117, about 17 teeth — and applied to the older 14.5-degree standard the same formula returns 32, exactly what the old handbooks print. Below that count, designers shift the profile rather than change the curve.
A picture of it
THE PICTURE #How to readR1 is the thing wanted and R2 the geometric condition it reduces to — read the arrow out of R1 as derived from. R3 to R5 are independent things a designer also wants, not consequences of R2. The two elements are the candidate curves, and the content is in which arrows are absent: both satisfy R2, so both give a steady ratio, but only the involute reaches R3 and R4, and only the cycloid R5.
What became clearer
WHAT CLEARED #The tooth count fixes the ratio only on average; the curve fixes it moment to moment. Since contact can only push along the common normal, a steady output demands that this normal never move off one fixed point on the line of centres — and the involute satisfies that by construction, its normal always being the common tangent to two base circles cut permanently into the gears. That one fact also delivers what made it universal: a ratio immune to centre-distance error, and interchangeable gears cut by one straight-sided tool. What was given up is sliding contact everywhere but the pitch point, and the wear that follows.
Where to go next
ONWARD #- How profile shift rescues a pinion with too few teeth without abandoning the involute.
Key terms
TERMS #| Term | What it means |
|---|---|
| Conjugate action | tooth flank geometry that holds the angular velocity ratio constant throughout the engagement. |
| Pitch point | the fixed point on the line of centres through which the contact normal must pass. |
| Base circle | the circle a tooth's involute is unwound from; the ratio is the inverse ratio of the two base radii. |
Every term the collection defines is gathered in the glossary.