A steam engine needs a piston to move dead straight, and for eighty years the best a machine of pivoted bars could manage was almost. Then someone turned a circle inside out.
James Watt had a problem that sounds like it shouldn't be one. His double-acting engine of 1784 pushed and pulled a piston, so the piston rod had to be tied rigidly to the great rocking beam overhead — but the beam swings on an arc, and the rod must go straight up and down. Tie a straight thing to a thing that swings and something has to give: the rod binds, the gland leaks, the engine wears itself out.
Watt's answer was a linkage — a little parallelogram of jointed bars — that took the beam's arc and converted it into motion that was very nearly a straight line. He was inordinately proud of it. Late in life, at seventy-two, he wrote to his son:
I am more proud of the parallel motion than of any other invention I have ever made.James Watt to his son, 1808
And to his partner Boulton, the week he worked it out, he described exactly what it does and — to his great credit — exactly what it does not:
The convexities of the arches, lying in contrary directions, there is a certain point in the connecting-lever, which has very little sensible variation from a straight line.James Watt to Matthew Boulton, 11 September 1784
Very little sensible variation. Not none. Watt's tracer does not draw a straight line; it draws a long thin figure-eight — a curve geometers later named Watt's curve — and rides the nearly-flat sliver where the two loops cross. Good enough to run the Industrial Revolution on. Not good enough to be true. And nobody could yet say whether true was even possible.
Here is Watt's linkage as a working mechanism. Two equal arms pivot at the fixed dots; a coupling bar joins their ends; the pencil rides the middle of that coupler. Drive the crank — drag it, or press Run — and the pencil traces its path. Near the centre it looks perfectly straight. It isn't. Crank the stroke slider wider and the bow blooms into view; the dashed line is where a true straight line would lie.
Pin the stroke down small and Watt is straight to about 1 part in 4,000 — astonishing engineering, and exactly the order of accuracy his engines ran on. Open it up and you see the truth: an osculating curve, hugging the line for a moment and then peeling away. The question that hung over the next eighty years was whether the peeling-away was a fact of geometry or just a failure of cleverness.
It is not an obvious question, and it is not a small one. You are asking: can a finite arrangement of rigid bars, pinned at the ends and turning on fixed pivots, constrain one point to move on a perfectly straight line? Every bar wants to swing in a circle. A straight line is a circle of infinite radius. Can you build infinity out of a box of finite arcs?
The best minds of the century improved the approximation and assumed that was the game. Pafnuty Chebyshev studied Watt's linkage around 1850 and asked the sharpest possible version — what is the best a linkage can do? — and in chasing the bar lengths that minimise the worst-case error he was driven to a wholly new piece of mathematics: the theory of best uniform approximation, and the Chebyshev polynomials that still bear his name. A failure to draw a straight line gave us a cornerstone of numerical analysis. (His 1854 memoir carried the wonderfully deadpan title "Théorie des mécanismes connus sous le nom de parallélogrammes.")
But better and better approximations are not a yes. The yes came from outside the academy.
In 1864 a French army engineer, Charles-Nicolas Peaucellier, sent a short note to a mathematics journal claiming the thing everyone had stopped expecting: an exact straight-line linkage. Seven years later a young Russian, Lippman Lipkin, arrived at the same device independently. It is so simple it looks like a trick, and the trick is a beautiful old idea — inversion in a circle.
Here is the cell. A fixed pivot O. Two long equal bars from O hold two corners of a rhombus of four equal short bars. The rhombus's other two corners are the driver D and the pencil B. That is the whole inventory: seven bars. And no matter how you flex it, the pencil and the driver obey one iron law:
OB · OD = a² − b²
— the product of the two distances from O is a fixed constant, set by the bar lengths alone (a the long bars, b the rhombus side). That is inversion in a circle of radius √(a²−b²): every point of the plane has a partner, near for far and far for near, and B is forever D's partner. Drive D and B is dragged to its mirror through the glass.
Now the masterstroke. Inversion has a property a schoolchild can check and an engineer can exploit: it sends any circle passing through the centre to a straight line. So pin the driver to a circle that runs through O — one more bar, a crank of the right length does it — and the pencil, its inverted partner, is forced onto a perfect straight line. Not nearly. Exactly.
The line it draws sits at x = −(a²−b²)/2c — a closed form you can read off the bar lengths before you build it — and the pencil sweeps the whole of that line as the crank turns. Turn the magnifier to maximum: the trail does not thicken. There is nothing to magnify. The straightness is not a measurement that came out small; it is an identity, true bar-length by bar-length, and the only error is the arithmetic of the machine drawing it.
Inversion of radius k about O sends a point at distance d to distance k²/d along the same ray. Take a circle through O with the far end of its diameter at distance m. A ray from O hits the circle at distance d = m·cos θ (Thales: the angle in the semicircle is right). Its image lands at k²/d = k²/(m·cos θ) — and a point whose distance along a ray is (const)/cos θ is precisely a point whose perpendicular distance from O is the constant: a straight line, perpendicular to that diameter, at distance k²/m. The circle-through-the-centre unrolls into a line. That is the entire secret of the machine.
And the cell's own law OB·OD = a²−b² falls straight out of the rhombus: drop both diagonals, let them cross at M; then OB·OD = (OM−MB)(OM+MB) = OM² − MB², and since OM² = a² − AM² and MB² = b² − AM², the AM² cancels and you are left with a² − b². No calculus. Just the Pythagorean theorem, twice.
Peaucellier's note sank almost without trace until 1874, when J. J. Sylvester — one of the great Victorian mathematicians — carried a working model to a lecture at the Royal Institution and the room went up. The English mathematical world discovered linkages overnight. Two years later a young barrister-mathematician, A. B. Kempe, gave a public lecture with a title that is still the best in the subject: How to Draw a Straight Line.
Kempe went further than a straight line. He proved — the argument had gaps later mathematicians had to mend, but the theorem stands — that linkages are universal: for any curve you can write as a polynomial equation, there is some arrangement of bars whose pencil traces exactly that curve. A circle, an ellipse, a figure-eight, the outline of the letters of your name. The popular gloss is true to the spirit: there is a linkage that signs your signature. The straight line was not a finish line. It was the first proof that a box of rigid bars is, secretly, a machine for drawing anything.
One honesty owed to you. Peaucellier's line is exact in the same sense a theorem is exact: given ideal rigid bars and frictionless pin-joints, the pencil is on the line, full stop, and the live cell above confirms it to the last bit of floating-point. A physical Peaucellier linkage still wobbles — its joints have play, its bars flex, its holes were drilled by someone. But the wobble is now an engineering tax on a perfect idea, not a flaw in the idea. Watt's bow is in the geometry; you cannot polish it out. Peaucellier's wobble is in the steel; better steel narrows it without limit. That is the whole difference between approximate and exact, and it took eighty years and a complete theory of approximation to find the far side of it.
Watt drew a near-miss and changed the world with it. Peaucellier drew the real thing and almost nobody noticed for a decade. Both are here, both run, and every number under them was checked before it was shown.
Everything above is computed, not asserted. Both linkages are simulated from their bar-length constraints alone — each joint placed by intersecting the circles its bars allow — so the inversion law and the straightness are not assumed; they are observed to emerge, and then measured. The offline verifier research/peaucellier/verify.mjs runs every check below; the figures here are pulled from its output.
Watt's bow vs. stroke — the cubic sharpening, straight from the verifier. Open the stroke and the error climbs as its cube; close it and Watt grows arbitrarily straight, but never exact.
| stroke | bow off straight | straightness |
|---|
The history. Watt's parallel motion: patented 28 April 1784; the Boulton letter (11 Sep 1784) and the 1808 letter to his son are quoted verbatim from the record. His tracer's path is Watt's curve, a figure-eight (the lemniscate of Bernoulli for the crossed-square proportions). Peaucellier announced his cell in 1864 (a brief note in the Nouvelles Annales de Mathématiques); Lipkin reached it independently in 1871. Sylvester lectured on it at the Royal Institution in January 1874; Kempe's How to Draw a Straight Line followed (lecture 1876, published 1877). Kempe's universality theorem (1876) — a linkage exists for any bounded piece of any real algebraic plane curve — had flaws in its original proof, repaired by later workers (rigorous modern treatments include Kapovich–Millson, 2002, and Abbott–Demaine–others). Chebyshev's study of linkage approximation (memoir of 1854) is the acknowledged origin of his theory of best uniform approximation and the Chebyshev polynomials.
The "1 part in 4,000." This is computed for this page's Watt proportions (rocker 1.6, coupler 3.2, pivots at ±1, tracer at the node of Watt's curve) at a representative working stroke — see the table. It happens to land at the order of magnitude historically cited for engine practice; it is not a claim about the exact geometry of any surviving Watt engine, whose refined "parallel motion" added a pantograph parallelogram to fold the linkage compactly.
"Exact." Peaucellier's deviation here is ~10⁻¹³ — floating-point round-off, i.e. zero to the precision carried. Exactness is a statement about ideal rigid bars and pin-joints; real linkages deviate for reasons of manufacture, not of principle (§V).