An angle that does not care where it stands

Draw a circle. Mark two points on it and call them $A$ and $B$. Now pick any third point on the longer of the two arcs between them, join it to both, and measure the angle.

Move the third point. Measure again.

The angle does not change. Not approximately, not for a while — it is the same number everywhere on that arc, and the arc is infinite in points. A quantity that ought to depend on where the observer stands turns out not to, and the reason is worth the essay.

What is actually being claimed #

Two claims travel together, and separating them makes both easier.

The first is the constancy: every point of the major arc sees the chord $AB$ under the same angle. The second is the value: that angle is exactly half the angle the same chord subtends at the centre. The second implies the first, since the centre does not move, and it is the one worth proving.

It is easy to underrate how odd this is. The three points form a triangle, and almost nothing about a triangle survives moving one of its vertices — the side lengths change, the area changes, the other two angles change. This one angle does not, and the constraint that holds it still is nothing more than the requirement that all three vertices lie on one circle.

The proof is two isosceles triangles #

Join the apex to the centre and continue the line to the far side of the circle. That single extra line finishes the argument.

Look at the left half. Two of its sides are radii of the circle, so they have the same length, so the triangle is isosceles and its two base angles are equal. Call that shared angle $\theta_1$. The angle at the centre in that half is the exterior angle of the other two, and an exterior angle equals the sum of the two interior angles it is not adjacent to — so it is $2\theta_1$.

The right half is the same argument with different letters: base angles $\theta_2$, centre angle $2\theta_2$.

Now add. The angle at the apex is $\theta_1 + \theta_2$. The angle at the centre is $2\theta_1 + 2\theta_2$. One is half the other, and neither the position of the apex nor the size of the circle entered anywhere.

That is the whole proof, and its shape is worth noticing: it does not compute the angle at all. It never asks what $\theta_1$ or $\theta_2$ actually is. It establishes a relation between two quantities and lets both of them float, which is why the result holds for every apex rather than for a particular one — the same move as proving a dissection works without ever measuring the triangle.

The case with a name #

Push the chord until it passes through the centre. Now it is a diameter, the angle at the centre is a straight angle, and half of a straight angle is a right angle.

Every point of the circle sees a diameter at a right angle. This is Thales’ theorem, traditionally the oldest theorem in Greek mathematics to be attributed to a named person, and it is a special case of a result Thales is not usually credited with.

The converse matters more than the theorem. If a triangle has a right angle, its hypotenuse is a diameter of the circle through its three vertices, and the centre of that circle is the midpoint of the hypotenuse. So the set of all right-angled triangles on a given hypotenuse is exactly the set of triangles with the third vertex on a particular circle — a family of shapes described entirely by a curve.

That reformulation is what makes the theorem useful rather than merely tidy. It converts a condition on angles into a condition on position, and positions are drawable.

There is a second consequence, and it is the one that gets used. The midpoint of the hypotenuse is equidistant from all three vertices, because it is the centre of that circle. So the median to the hypotenuse of a right-angled triangle is exactly half the hypotenuse — a fact that looks like it needs the Pythagorean relation and needs nothing but the circle.

A locus, which is the useful way to say it #

The general statement is best read backwards. Instead of “the angle is constant on the arc”, say: the set of points from which a fixed segment subtends a given angle is an arc of a circle.

That is a locus — a shape defined by a property rather than by a construction — and it answers questions no amount of algebra makes obvious. Where should a camera stand so that a wall of known length fills exactly $40°$ of its view? Anywhere on a particular arc. Where should a footballer stand so that the goal subtends the largest possible angle? At the point where the arc through the goalposts is tangent to the run of play, since the angle grows as the arc shrinks and the tangent point is where the smallest arc still reaches.

Both are the same picture, and both are questions about all positions at once — the kind of question a locus is for. The alternative is to write the angle as a function of two coordinates and hunt for its level sets, which produces a correct answer and no understanding at all.

Four points, and the condition for a circle #

Take four points on the circle instead of three and join them in order.

Opposite angles add to $180°$. The reason is the theorem again: two opposite corners stand on the same diagonal from opposite sides, so between them they account for the whole circle at the centre — $360°$ — and each is half of its share.

The converse is the part with teeth. If a quadrilateral’s opposite angles add to $180°$, its four corners lie on a circle. That is a test: an angle condition that detects a geometric coincidence. Four points in the plane generically do not lie on a circle — three always do, and the fourth has one degree of freedom to be wrong in — so a criterion that recognises the exceptional case is worth having.

It is the criterion underneath the Delaunay triangulation, where the deciding question about four points is whether the fourth lies inside, on, or outside the circle through the other three, and the whole structure is defined by that circle staying empty. The empty-circumcircle rule and the cyclic-quadrilateral rule are the same statement seen from two directions.

The independence from shape is the point here as it was for the triangle. Both quadrilaterals above have four different corner angles and the same two sums, because the sums are properties of the circle rather than of the quadrilateral drawn on it. That is the recurring pattern of this whole family — a quantity that survives moving the thing it was measured on.

What it costs to compute #

The theorem is free to state and cheap to use, and it is worth being explicit about what “cheap” means, because the computational question is where a great many pretty results quietly stop being useful.

Testing whether four points are concyclic is a $4 \times 4$ determinant — the in-circle predicate — and it is one of the two primitives on which computational geometry is built. It costs a couple of dozen multiplications, which is nothing. What it costs in practice is precision: the determinant is a difference of large nearly-equal quantities, and in floating point it returns the wrong sign for points that are nearly concyclic. A triangulation built on a predicate that occasionally lies produces not a slightly wrong answer but a structurally impossible one — overlapping triangles, or an infinite loop in the construction.

The resolution used everywhere is exact arithmetic on demand: evaluate cheaply, estimate the error, and re-evaluate in extended precision only when the estimate cannot rule out a sign flip. This is a general pattern for geometric predicates and it is a genuine cost, since the exact path can be a hundred times slower. It is paid because the alternative is not inaccuracy but nonsense.

Nothing in the picture suggests any of that. The figure shows four points and a circle, and the circle is either through them or not.

The same split — a statement that is trivial to make and awkward to compute — runs through most of applied geometry, and it is worth naming because the pictures on this site systematically hide it. A dissection is exhibited and then it is finished; an in-circle test has to be run, on real numbers, in a machine that does not have real numbers. The theorem’s cost is not in the theorem.

Where it needs a condition, and where it fails #

Two conditions have been quietly assumed and both matter.

The apex must be on the major arc. Move it to the minor arc — the short way round, on the same side as the centre’s smaller angle — and the angle is no longer half the central angle but half of what is left after subtracting from $360°$. So the two arcs give two different constant values, and they sum to $180°$, which is precisely the cyclic-quadrilateral statement. The generator refuses to draw an apex on the wrong arc rather than silently drawing a different theorem.

The surface must be flat. On a sphere, the angles of a triangle sum to more than $180°$ by an amount proportional to its area, and Thales’ theorem fails outright: a triangle with all three vertices on a great circle has three right angles, and a diameter is subtended at a right angle only from particular places. The whole inscribed-angle apparatus is a fact about the Euclidean plane, and the failure is the same one that breaks the Pythagorean relation on a curved surface.

That the theorem is equivalent to flatness is not a caveat to note and move past. It means the constancy of the inscribed angle is one of the things flatness is, in the same way that the sum of a triangle’s angles is.

The history, briefly, because it is misleading #

Thales is usually given the right-angle case, on the authority of Diogenes Laertius writing eight centuries later, who reports that Thales sacrificed an ox in celebration. The general inscribed-angle theorem is Euclid’s, Elements III.20, proved in the fifth century BC by essentially the argument above.

What is worth noticing is the order. The special case is remembered and the general theorem is not, though the general theorem is no harder and contains the special one as the moment the chord passes through the centre. The famous statement is famous for being surprising, and the surprise is an artefact of how much has been thrown away: a right angle is a memorable number, and “the same unknown angle everywhere on the arc” is not.

This inversion — the corollary better known than the theorem — is common enough to be worth watching for. It is the same reason the parallelogram-law formulation of the dot product is less familiar than the cosine one, and the reason a great many people meet Euler’s characteristic as a fact about polyhedra rather than as a topological invariant.

What the picture cannot show #

The draggable figure moves the apex through twenty-one positions. The theorem is about infinitely many, and no number of frames closes that gap: what the slider corroborates is that the author did not choose one lucky point, and what it cannot corroborate is the claim itself. The proof, which is four lines about isosceles triangles, does what no amount of dragging can.

There is a second gap and it is larger. Every figure here is a Euclidean plane figure, and the theorem’s real content — as argued above — is that it is equivalent to the plane being flat. A picture drawn in the plane cannot show the failure on a sphere, because it has no way to draw a sphere’s geometry from inside. The reader is asked to take the counterexample on trust, and that is the honest position: the figures establish the theorem’s shape and say nothing whatever about its scope.

Nor can any figure here show the in-circle predicate’s precision problem, which is invisible by construction — the points that break it are the ones that look exactly concyclic.

The ladder from here #

Rungs above this one: the power of a point, where the two chords through a point multiply to the same product whichever pair is chosen, and the inscribed angle is what makes the triangles similar. The tangent-chord angle, which is the theorem’s limit as one endpoint of the chord slides into the apex. Ptolemy’s theorem on a cyclic quadrilateral, and the trigonometric identities that fall out of it. The nine-point circle, which is a cyclic-quadrilateral argument run four times. The extended law of sines, where $a/\sin A = 2R$ is exactly this theorem with the constancy made numerical. And the inversive plane, where “circle or line” becomes one kind of object and the theorem becomes a statement about that one kind — the point at which it stops being about circles at all, in the way that a sphere with one point removed stops being about spheres.

Why the constancy is the interesting part #

The value — half the central angle — is the memorable clause, and it is the less important one.

What makes the theorem matter is that a measurement taken from a moving position does not move. That is a statement about invariance, and invariance is the property mathematics reaches for whenever it wants to say two situations are really the same situation. An angle that does not depend on the observer is a property of the chord and the circle rather than of the observation, and once a quantity is known to belong to the configuration rather than to the viewpoint, it can be used to classify configurations.

Every later use of the theorem runs on that. The concyclicity test works because the angle is a property of the four points. The Delaunay condition works because emptiness of a circumcircle is a property of the four points. The locus formulation works because the angle labels the arc. None of that would survive if the angle depended on where the apex happened to be, and the two isosceles triangles are the entire reason it does not.