“Ut pendet continuum flexile, sic stabit contiguum rigidum inversum.”
“As hangs the flexible line, so but inverted will stand the rigid arch.”
Robert Hooke, A Description of Helioscopes, 1675 — published as an anagram, decoded only after his death, 1705
research/catenary/facts.md); the music is synthesised from the catenary's own contour. Second companion film for a Wasteland layer.
Hang a chain from two nails. It falls into a curve so calm and certain that it looks like it must have a simple name. For most of history people thought it did: a parabola, the same curve a thrown stone traces. Galileo wrote as much in 1638. He was wrong — and the true answer took Leibniz, Huygens and Johann Bernoulli until 1691 to find, the year they finally named the curve the catenary, from the Latin catena, a chain.
This page lets you hang the chain yourself. The one below is not drawn from a formula — it is a small physics simulation: forty-eight beads, gravity, and rigid links between them. It falls and settles, exactly as a real chain does, finding the lowest-energy shape it can with no idea what that shape is supposed to be. Then we lay two curves over it: the catenary (green) and Galileo's parabola (orange). One of them lands on the beads. The other does not.
Grab either anchor and drag it. Pay out more chain with the slack control. The beads re-settle every frame; the green curve is computed, the unique catenary y = a·cosh(x/a) through your two anchors with this much chain — and it falls right onto them, because that is the curve a hanging chain is.
The catenary is fitted to the chain's own length, not tuned to match. The parabola passes through both anchors and the chain's lowest point — the fairest parabola there is — and still misses by the amount shown.
The deeper the sag, the worse the parabola does. Pull the anchors close or add slack and the gap above climbs to several percent of the sag; let the chain go nearly straight and the two curves become almost indistinguishable. That second fact is the key to Galileo.
The textbook line is that Galileo claimed the hanging chain is a parabola. Read him and it is, in Osserman's phrase, “at best half true.” In Two New Sciences (1638) he says it twice. Once in passing, as a drafting trick. But in the careful passage, right after deriving that a thrown projectile flies in a parabola, he writes that a hanging cord “closely approximates the parabola… the coincidence is more exact in proportion as the parabola is drawn with less curvature… so that using parabolas described with elevations less than 45° the chain fits its parabola almost perfectly.”
That is not a man asserting an identity. That is a man reporting a very good approximation and telling you exactly when it holds — for shallow curves. And he is right. The instrument above shows it: at a shallow sag the parabola sits almost on the chain. The error only opens up as the curve deepens. Galileo found the leading term of the answer with his eye; the catenary is what is left when you keep going.
He was wrong about the curve and right about the regime. The parabola is the catenary's shadow in the shallows.
The reason is in the arithmetic of cosh. Expanded as a series, cosh(u) = 1 + u²/2 + u⁴/24 + … A parabola is exactly the 1 + u²/2 part. So a parabola and a catenary agree on everything up to the fourth-order term; the first thing that separates them is that little u⁴/24, invisible when u is small and unmissable when it is not. Galileo's eye was good to third order.
Here is the twist that rescues Galileo's instinct entirely. A cable's shape is decided by what it carries. A chain carrying only its own weight spreads that weight evenly along its length — and that gives the catenary. But hang a heavy, flat roadway from the cable, so the weight is spread evenly along the horizontal, and the equation changes by exactly one square root — and the solution becomes a parabola.
Weight per unit of chain is constant — the links are identical. This is the free chain of Instrument I.
So a suspension bridge's main cables — the Brooklyn Bridge, the Golden Gate — are parabolas, to a very close approximation, because the deck they carry is far heavier than the cable itself. Galileo's parabola is the exact answer to a question he wasn't asking: not the free chain, but the loaded one. (Real cables are a faint blend of the two, since the cable has some weight of its own; the parabola dominates because the deck outweighs it many times over.)
Turn the catenary upside down and something quietly profound happens. A hanging chain is in pure tension — every link pulls on its neighbours, nothing pushes. Flip the curve and every relationship flips with it: an arch built on the inverted catenary is in pure compression, every stone pressing on the next, with no sideways bending to crack it. Robert Hooke saw this in 1675 and, in the fashion of the age, published it as a scrambled Latin anagram so he could claim priority without giving the secret away. It was decoded after his death: as hangs the flexible line, so but inverted will stand the rigid arch.
The most famous structure built on this principle is Eero Saarinen's Gateway Arch in St. Louis — 630 feet tall and, almost exactly, 630 feet wide. It is constantly called a catenary. It isn't, quite. Its centroid follows a published equation, worked out by the engineer Hannskarl Bandel:
D = 1 is an ordinary chain. The Arch's D = 0.690 is a flattened catenary — the curve a chain takes when it is loaded down more toward the base, just as the Arch's hollow legs taper from a 54-ft triangle at the ground to a 17-ft triangle at the top.
The number that settles it is A·B, the product of the curve's height-scale and its width-rate. For any ordinary catenary A·cosh(Bx) this product is forced to equal exactly 1. The Arch's published constants give A·B = 0.690. It is provably not a plain chain. Robert Osserman, whose 2010 analysis is the one that pins this down, calls it a flattened catenary, and notes drily that the popular term “weighted catenary” is nearly empty — it means little more than “a convex curve.” The honest description is the equation, and the equation is above.
A small, true footnote in the same spirit: for years a plaque in the Arch's tram lobby printed the inverse of this equation — and printed it with three separate errors. The monument to the catenary could not state its own curve correctly. (That error is documented independently of Osserman's paper; his analysis does not mention the plaque.)
Nothing on this page is asserted that does not pass an offline verifier first. The mathematics — the catenary ODE, the energy principle that the chain hangs lower than any equal-length rival, the parabola limit, the two load equations, and the Gateway Arch's constants — is recomputed from scratch in research/catenary/verify.mjs, which the live instruments are then checked against.
One more entry in the verification venue — a real claim, its primary source, and the check runnable in the page. Here the claim is a shape, and the correction is gentle: Galileo was less wrong than the textbooks say.
Like the coast of Britain — a length that isn't a number — and the sun's crooked clock, where “the shortest day” isn't the earliest sunset. The catenary is what hides behind a curve everyone already “knew.”
The other bridge on the site: Königsberg's seven, where walking the bridges once each turns out to be impossible. There the bridge is a problem in graph theory; here it is a problem in which curve a cable chooses.
The house signature — a claim computed in front of you, as in the fixed point. The chain above is the purest version: it doesn't read the formula, it falls into it.
y = a·cosh(x/a) is the modern form; the hyperbolic functions did not exist in 1691.y = 693.8597 − 68.7672·cosh(0.0100333x), −299.2239 ≤ x ≤ 299.2239 (ft); apex 625.0925 ft; D = A·B = 0.69; apex radius ≈ 145 ft; cross-sections 54 ft → 17 ft equilateral triangles. The equation also appears in Larson, Calculus 11th ed. §7.4. The plaque error: szcz.org (2002) and NPR (2006), independent of Osserman.y = wx²/2H; self-weight (per arc length) → catenary. Suspension-cable main spans are standardly modeled as parabolas.√(L²−v²) = 2a·sinh(h/2a)); the hanging chain is a position-based Verlet simulation with rigid distance constraints — no curve assumed. All load-bearing figures are independently verified offline by research/catenary/verify.mjs (35/35), including a free discrete chain that settles onto cosh as the link count grows.