The rope that squares a corner

A loop of rope with twelve knots in it, spaced equally, is a complete instrument for building a right angle. Hold the first knot, the fourth and the eighth, pull until the rope is taut, and the corner at the first knot is square — not approximately, not for practical purposes, but exactly.

Nothing here measures an angle. There is no protractor, no set square, no plumb line and no straight edge. There is a length divided into equal parts, which is the easiest thing in the world to make — fold a rope in half, in half again, in three — and out of it comes the one angle that every other construction in flat geometry is built on.

The theorem this rests on is not the one proved by rearrangement. That one says: given a right angle between sides of $a$ and $b$, the third side is $\sqrt{a^2+b^2}$. What the rope uses is the sentence with its ends swapped: given three sides with $a^2+b^2=c^2$, the angle between the first two is right.

Those are different statements. One takes an angle and produces a length; the other takes three lengths and produces an angle. Neither follows from the other by logic alone, and the second is the one worth having.

What a converse is, and why it is not free #

A statement of the form if P then Q says nothing whatever about what happens when Q holds. If it is raining then the pavement is wet is a fact about rain; the pavement is also wet after a street cleaner has passed. The converse of a true statement is a separate claim, sometimes true and sometimes not, and the only way to know which is to prove it.

Plenty of geometry’s converses are false. Every square has four equal sides, and a rhombus has four equal sides without being a square. Every equilateral triangle is isosceles, and the converse fails at once. Two triangles with equal areas need not be congruent. The habit of assuming a converse is one of the more reliable ways to arrive at a wrong answer while feeling entirely secure.

So the rope needs an argument, and there is a good one.

The argument runs in three steps, and the middle one is where the original theorem gets used.

Start with the triangle that actually exists: sides $a$, $b$, $c$, with $a^2+b^2=c^2$, and an unknown angle between the first two. Now build a second triangle, on a clean patch of ground, with a right angle between two sides of the same lengths $a$ and $b$. The forward theorem — the one already proved — says the third side of that constructed triangle is $\sqrt{a^2+b^2}$, which by assumption is $c$.

Both triangles now have sides $a$, $b$, $c$. Side-side-side congruence says they are the same triangle. Corresponding angles of congruent triangles are equal. The angle in the first, between $a$ and $b$, therefore equals the angle in the second between $a$ and $b$, which was built to be right.

That is the whole proof, and it is worth noticing how it works: it does not examine the given triangle at all. It manufactures a second one that is known to be right, and then shows the two cannot be told apart. The given corner is square by an alibi.

The rope that does not close square #

A figure that only ever draws the case that works is not evidence about the case that does not. The same rope with a different knot count closes into a triangle too — every triple of lengths that satisfies the triangle inequality does — and the corner comes out wherever the arithmetic sends it.

Nothing about the rope objects. It has no way to object: it is a length divided into equal parts, and it will make whatever triangle it is asked for. What decides the corner is the arithmetic, and the picture merely reports the verdict.

That is the part worth holding on to about instruments generally. The rope does not know the theorem. It embodies a consequence of the theorem, and it does so only when the numbers chosen are the right ones. A builder using a 4-5-7 rope in the belief that any three numbers will do would produce a wall twelve degrees out of true and would have followed the procedure faithfully throughout.

What a mistake is worth, in degrees #

The interesting engineering question is not whether the exact triple works. It is how much a small error costs — because a real rope stretches, real knots have width, and the ground is never quite flat.

The slope of that curve at the right angle has a clean closed form. Differentiating the law of cosines at $\theta=90°$ gives

$$\frac{d\theta}{dc} = \frac{c}{ab\sin\theta} = \frac{c}{ab},$$

which for the 3-4-5 rope is $5/12$ of a radian per unit of $c$, or about $24°$ per knot. One part in a hundred wrong along the rope — a centimetre in a metre — is a fifth of a degree out of square, which over a ten-metre wall is a displacement of about three centimetres at the far end.

That is the honest answer to why the method survived: it is good, not perfect, and it degrades gently. A construction whose error grew faster than its input would have been abandoned. This one loses a fifth of a degree for a one-percent mistake, which is comfortably better than a builder can hold a set square by eye.

The same curve reads in the other direction too, and this is the direction the law of cosines lives in: the corner angle is a continuous, strictly increasing function of the third side, so every angle between $0°$ and $180°$ is produced by exactly one length $c$. The right angle is not special to the geometry — it is special because it is the one whose $c$ satisfies a clean equation in the other two.

Which ropes are worth carrying #

A rope is useful only if the triple is small enough to knot and remember. The 3-4-5 needs twelve knots; the next few are longer.

The primitive triples in order of perimeter are 3-4-5 at twelve knots, 5-12-13 at thirty, 8-15-17 at forty, 7-24-25 at fifty-six, and 20-21-29 at seventy. Only the first is convenient. The second is already a rope five times as long for a triangle that is worse shaped for laying out a rectangular room — long and thin, so the far corner is set by a short leg, which magnifies whatever error the rope carries.

Those triples are not a scattered list. Every one of them comes from a rational point on a circle, and the parametrisation produces them all exactly once, so the sequence of usable ropes is completely known and completely described. There is no hunting to be done.

Doubling works too — 6-8-10 is a rope of twenty-four knots making the same triangle at twice the size — and for a builder that is the useful direction. A 3-4-5 rope of twelve knots can be laid out with each knot at a metre and produce a triangle three metres by four, which is the scale a room actually needs.

What it costs #

Two things, and both are the reason the rope is not the modern method.

The first is that the rope is only as good as its knots. Twelve equal intervals are easy to make by folding, but a rope stretches unevenly under tension, and the knot that has been pulled hardest is the one that has moved. The construction assumes exactly equal spacing and has no way to check it, which is why the error curve above matters: it converts a defect that cannot be seen into an angle that can be measured later, when the wall is up.

The second is subtler. The rope produces one right angle, at one place. A building needs a great many of them, and each is set from the last, so errors accumulate along the chain rather than cancelling. Setting out a large rectangle by transferring a rope corner four times gives a figure whose closing error is the sum of four independent mistakes. The professional fix — measuring both diagonals of the finished rectangle and adjusting until they agree — is itself a converse argument, and a better one: a quadrilateral with equal sides in pairs and equal diagonals is a rectangle, which checks all four corners at once instead of one at a time.

The rope stretchers, and what the evidence says #

The 3-4-5 rope is routinely attributed to the Egyptians, and the attribution deserves more care than it usually gets.

What is documented is that Egypt employed harpedonaptae — rope-stretchers — who laid out temple foundations, and that Democritus, writing in the fifth century BC, said he had met nobody better at constructing figures with lines. What is not documented anywhere in the surviving Egyptian mathematical papyri is a 3-4-5 rope, or the relation between the squares, or any statement of the converse. The Rhind and Moscow papyri are full of area and volume computations and contain nothing of the kind. The knotted-rope story appears to have been proposed by a nineteenth-century historian as a plausible reconstruction and then repeated until it acquired the status of a source.

Two other traditions do have the statement in writing. The Śulba Sūtras, Indian manuals for building sacrificial altars, state the relation explicitly and list several triples including 3-4-5, 5-12-13 and 8-15-17 — and their purpose is construction, which is the converse direction. The Chinese Zhoubi Suanjing gives the 3-4-5 case with a dissection diagram that is essentially the rearrangement proof. Babylonian tablet Plimpton 322, older than either, is a list of numbers that are hard to explain except as triples, though what it was for remains argued about.

So the honest summary is that several ancient traditions knew which numbers make a right angle, at least one of them wrote the rule down as a building instruction, and the single culture the story is usually told about is the one with no direct evidence for it. That is a common shape for a mathematical anecdote, and it is worth checking before repeating.

Where it needs a condition #

The proof used side-side-side congruence, and that is a theorem of flat geometry. It fails on a curved surface in the sense that matters here: the two triangles are still congruent, but the constructed one no longer has a right angle at the corner where it was built with one, because on a sphere the relation between the three sides and the angle is different.

Laid out on ground that curves — which all ground does, at about one part in ten million over ten metres — a 3-4-5 rope makes a corner that is not quite square. The spherical version of the theorem says the corner comes out at slightly less than $90°$, by an amount proportional to the area of the triangle. For a rope triangle six metres across on the Earth’s surface, that is about a hundred-billionth of a degree, which is why nobody has ever noticed. For a triangle six hundred kilometres across it is a tenth of a degree, which is why geodesy exists.

The condition is therefore not a technicality about proof style. It is a statement about how large a triangle may be before the flat theorem stops being the right theorem, and the answer is: much larger than a building, much smaller than a country.

What the picture cannot show #

Every figure here draws a triangle at one size, with one set of knot counts, at one place on the page. The converse is a claim about every triple of lengths satisfying the equation, of which there are infinitely many, in infinitely many sizes. No drawing can contain them.

Worse, the drawing cannot show the thing the proof turns on. The congruence step compares the given triangle to a triangle that was never built — one manufactured for the argument and discarded immediately afterwards. It exists in the proof and nowhere in the picture, and it is the entire reason the conclusion holds. The right-hand panel of the second figure is an illustration of a hypothetical, which is exactly what a diagram is worst at signalling.

What the pictures do carry is the falsifiability. The 4-5-7 rope is on the page, closing into a triangle with a corner twelve degrees out, and its presence is what makes the 3-4-5 rope evidence rather than decoration.

The ladder from here #

The forward theorem has three rungs below this one: the rearrangement, Euclid’s shearing proof, and the triples as rational points. Above it, the theorem stops being about triangles at all: in three dimensions it becomes a statement about the diagonal of a box and then about areas, and on a sphere it becomes false in a way that is more informative than its truth.

Further along, the same equation with the exponent changed is the question that took three hundred years, and the same distance formula with the exponent changed is the family of $p$-norms, where the unit circle is a diamond or a square and the shortest path between two points depends on which one is chosen. That last direction is where the theorem stops being a fact and becomes a definition — the point at which distance is something chosen rather than discovered.

The instrument is the argument #

The lasting point is about what a converse buys, and it is not symmetry.

The forward theorem is a fact about right triangles: it describes. The converse is a procedure: it builds. Nothing in the forward direction tells anyone how to make a right angle, because it assumes one is already there. The rope is the theorem in its useful direction, and the useful direction had to be proved separately.

This pattern recurs across the whole of mathematics and is worth naming. A characterisation that runs both ways turns a property into a test. Two integers are coprime exactly when the Euclidean algorithm ends at one — the forward direction is a fact and the backward one is a decision procedure. A graph has an Eulerian circuit exactly when every vertex has even degree — the forward direction is a remark and the backward one is a construction. In each case the second half is where the work goes, and in each case the second half is the half somebody can pick up and use.

Twelve knots in a rope is a right angle, and the reason it is a right angle is a proof that manufactures a triangle nobody ever sees.