Something always stays put

Take a map of the country, crumple it into a ball, and drop it anywhere within the country’s borders. At least one point of the crumpled map lies directly above the point of ground it represents.

Stir a cup of coffee — slowly, quickly, in any pattern at all — and once it has stopped moving, some point of the liquid is exactly where it began.

Both statements sound like conjuring, and both are instances of one theorem: any continuous map of a disc into itself leaves some point where it was. It is Brouwer’s fixed point theorem, from 1911, and the one-dimensional version can be seen at a glance.

The one-dimensional case is a crossing #

Take a continuous function $f$ sending $[0,1]$ into $[0,1]$. A fixed point is an $x$ with $f(x) = x$ — which is to say, a point where the graph meets the diagonal.

Now the argument, and it is three lines.

At the left end, $f(0)$ is somewhere in $[0,1]$, so $f(0) \ge 0$. The graph starts on or above the diagonal.

At the right end, $f(1) \le 1$. The graph ends on or below it.

A continuous curve that starts above a line and finishes below it must cross. So there is a point where $f(x) = x$.

That is the whole proof, and it is the intermediate value theorem applied to $g(x) = f(x) - x$, which is non-negative at $0$ and non-positive at $1$ and therefore zero somewhere between.

The two hypotheses are both load-bearing and both easy to lose sight of.

The map must send the interval into itself. Drop that and there is nothing: $f(x) = x + 1$ is perfectly continuous and moves everything. The theorem is about a space mapped into itself, and the generator enforces it — it checks a thousand points and refuses to draw a map that escapes.

The map must be continuous. Drop that and it also fails, and the counterexample is instructive. Send everything in the lower half to $0.9$ and everything in the upper half to $0.1$. Every point is moved, the map stays inside the interval, and the graph jumps across the diagonal without touching it. A jump is exactly how a curve gets from one side of a line to the other without crossing.

The second hypothesis is doing more work than it looks, and it is the one that connects this to the rest of analysis. Continuity is what licenses the intermediate value theorem, and the intermediate value theorem is not a fact about pictures — a curve on the rational numbers alone can cross from negative to positive without ever being zero, since it can step over $\sqrt2$. What rules that out is the completeness of the real numbers, the same property that makes a sequence of approximations converge to something rather than to a gap. Brouwer’s theorem in one dimension is completeness, wearing a disguise.

The crossing in each figure is located by bisection rather than by eye: the generator halves the interval two hundred times on $f(x) - x$ and then applies the map to the result, requiring it to return the same number. That is a stronger claim than the curves appear to meet, and it is the claim the caption makes.

Where the space matters #

Change the shape and the theorem can fail, which shows that it is a statement about the space rather than about maps in general.

Take an annulus — a disc with a hole punched out. Rotate it by any angle at all. Every point moves, the map is continuous, and it sends the annulus into itself. No fixed point.

Take a circle, the boundary alone. Same rotation, same failure. And a Möbius band, which also has a hole in the relevant sense, escapes it too.

So the theorem is not about continuous self-maps; it is about continuous self-maps of spaces with no holes. A disc works, an annulus does not, and the difference is exactly the hole — the same feature that the Euler characteristic counts and that separates a sphere from a doughnut.

The rotation makes the mechanism plain. A rotation of the disc does have a fixed point, and it is the centre; the annulus is precisely the disc with that point deleted, and deleting it is what removes the guarantee. One point is the whole difference — the same situation as a sphere with a point removed, where deleting one point turns a compact surface into a plane and takes several theorems with it.

That is what puts this result in topology rather than analysis. The relevant property of the disc is not its roundness, its size, or its smooth boundary — a square works identically, and so does any blob deformable into a disc. What matters is that it has no hole to rotate around.

Two dimensions, and the orbit picture #

In two dimensions the crossing argument has nothing to cross, and the proof becomes considerably harder — Brouwer’s own and every later one route through some topological invariant. What a picture can honestly show is a case where the fixed point is found rather than merely proved to exist.

Here the map turns the disc by $38°$ and shrinks it toward a point, and every starting point spirals into the same destination. The figure solves for that destination algebraically and then checks it — applies the map and requires the result to be the same point to machine precision — rather than reading it off the picture.

It is worth being clear that this figure is showing something stronger than Brouwer, and therefore something easier. A map that contracts distances has a unique fixed point and every orbit finds it, which is the Banach fixed point theorem — a different result, with a stronger hypothesis and a much better conclusion, and it comes with an algorithm.

Brouwer’s theorem assumes only continuity. It gives no uniqueness, no algorithm, and no orbit that converges. The coffee cup can have many still points, and iterating the stirring will not locate one.

Comparing the two disc figures makes the contraction’s character visible. The shrink factor sets how fast an orbit closes in, and the rotation sets how much it turns on the way; neither has anything to do with whether a fixed point exists, which was settled before any orbit was drawn. What the pictures illustrate is the rate, and the rate is the part Brouwer does not supply.

The proof cannot find it #

That gap is the theorem’s defining feature, and it caused Brouwer himself considerable trouble.

The standard proof is by contradiction. Suppose no point is fixed. Then for every point $x$, the two points $x$ and $f(x)$ are distinct, so they determine a direction — and following that direction from $x$ out to the boundary defines a map from the disc to its own edge which leaves every boundary point alone. Such a map is called a retraction, and it does not exist: the disc cannot be continuously squashed onto its rim without tearing, because the rim has a hole and the disc does not. The invariant that detects this is the same one distinguishing the annulus — and it is a relative of the count that tells a sphere from a doughnut, since both are ways of asking whether a space has a hole in it.

Notice what that argument produces. Nothing. It shows that the assumption no fixed point exists leads to an impossible object, and stops. It does not narrow down where a fixed point is, does not say how many there are, and provides no procedure.

Brouwer came to find this unacceptable. Having proved the theorem, he spent the rest of his career developing intuitionism, a school rejecting proof by contradiction for existence claims — on the grounds that showing there must be one without producing one is not showing anything. He effectively disowned his most famous result. It is a rare case of a mathematician taking a philosophical objection to his own theorem seriously enough to rebuild the foundations around it.

The constructive versions that exist are approximate. Sperner’s lemma — a combinatorial statement about colouring the corners of a triangulated triangle — gives a proof that can be run as an algorithm and locates a point that is nearly fixed, moved by less than any specified tolerance. Getting to an exactly fixed point still needs a limit, and the limit is where the constructive content runs out.

Where it is actually used #

The theorem’s reach is much wider than the coffee cup suggests, and the most consequential application is not in mathematics.

An economic equilibrium is a set of prices at which supply matches demand. Model the situation as a map: given a set of prices, the market responds with the prices those would produce next. An equilibrium is a set of prices the map does not change — a fixed point.

Arrow and Debreu used exactly this in 1954 to prove that a general competitive equilibrium exists, which is one of the central results in economic theory. Nash’s theorem, that every finite game has an equilibrium in mixed strategies, is the same argument applied to a different map.

Both results are famously existence claims that supply no method of computation, and the reason is that they are Brouwer in a new costume. The complaint that equilibrium theory does not say how a market reaches equilibrium is not a defect of economics; it is the theorem’s non-constructive character, inherited whole.

The same shape recurs in differential equations, where a solution is a fixed point of an integral operator and Picard’s theorem is the contraction version, and in numerical methods, where Newton’s method converges to a fixed point of an iteration built from the derivative.

There is a small physical instance worth keeping, because it can be checked rather than believed. Lay a map of the room on the floor of that room, at any scale and any angle, and one point of the map lies over the point it represents. If the map is a faithful scaled copy, the transformation is a contraction — a shrink, a turn and a slide — so the fixed point is unique and can be found, by drawing the map’s image of itself repeatedly and watching the images nest. That is the Banach case again, and it is the one demonstration of this theorem anybody can actually perform.

Brouwer’s version drops the faithfulness. Crumple the map and the transformation is no longer a contraction, no nesting occurs, the fixed point may not be unique — and it is still there. The crumpling is what takes the demonstration away and leaves only the theorem.

What the picture cannot show #

The interval figure shows a crossing, and a crossing is genuinely convincing — the argument really is visible. But it shows one function, and one function is an example rather than a theorem. What makes the crossing unavoidable is the pair of end conditions, and a picture cannot indicate that the drawn curve was constrained rather than merely chosen to cross.

The disc figure is worse in a specific way. It shows orbits converging, which is a property of contractions and not of continuous maps in general. A Brouwer map with a fixed point and no convergent orbits anywhere would look like a mess of paths going nowhere, and there would be nothing to draw and nothing to mark. So the honest two-dimensional case is undrawable, and the figure that stands in for it is showing an easier theorem.

Nor can any of these show the theorem’s most-used feature: that it applies in every dimension. The disc in $n$ dimensions has the property, and $n$ can be a hundred, and the economic applications need it to be — one dimension per commodity. The figures show $n = 1$ and $n = 2$, which are the two cases where nobody needed the theorem.

The ladder from here #

Rungs above: Sperner’s lemma, drawn on a triangulated triangle, and the constructive proof it supplies. The no-retraction theorem, and the invariant that proves it. The hairy ball theorem, which says a continuous tangent field on a sphere must vanish somewhere — the reason there is always a point of no wind. Banach’s contraction principle in its own right, with the geometric convergence rate. Kakutani’s generalisation to set-valued maps, which is the version Nash’s theorem actually uses. Fixed point index and Lefschetz’s theorem, which counts fixed points rather than merely detecting one. Newton’s method as a fixed point iteration. And the Borsuk–Ulam theorem, whose corollary is that two antipodal points of the Earth have the same temperature and pressure at once.

Existence without location #

The idea worth carrying is the distinction the whole essay runs on.

There are two quite different questions about any object: does it exist? and where is it? Ordinary experience runs them together, because the usual way to establish the first is to answer the second. Mathematics separates them, and the separation is not a technicality — whole classes of argument answer one and are structurally incapable of answering the other.

The pigeonhole principle proves two people share a hair count and names neither. Brouwer proves a point stays put and locates none. Both are counting or topological arguments that work by ruling out the alternative, and ruling out the alternative never produces a witness. The angle budget that caps the regular solids at five has the same character in the negative direction: it rules out a sixth without constructing any of the five.

That is worth knowing before starting a search rather than after. A theorem guaranteeing a solution exists is not a theorem that will help find it, and treating the two as the same has cost a great deal of wasted effort — in economics most visibly, where the equilibrium was proved to exist decades before anyone asked seriously how a market might get there.