The Line Cannot Do Its Own Arithmetic

Contents

1. The only tools you are given
2. What the line reaches, exactly
3. One‑half already costs you a dimension
4. The round trip that fetches √2
5. The sheet of paper
6. Where the picture is wrong
7. The plane is not enough either
8. Exercises

1 · The only tools you are given

Here is a line. It runs forever in both directions — that is what the two arrowheads mean. Someone has marked two points on it and called them 0 and 1. You have a straightedge and a compass. There is no second axis. There is no y, no i, no up. Dimensions arrive one at a time, and this one has not finished arriving.

Now build. Every point you construct must land on the line, because there is nowhere else for a point to be. The straightedge can only be laid along the line it already has. The compass may be centered on a point you have built, opened to the distance between two points you have built, and swung — and where it crosses the line, you may mark.

Those are the rules. The question is simple: which numbers can you reach?

Most students answer “all of them, eventually” or “the rationals, at least.” Both are wrong, and the true answer is much smaller and much stranger than either.

2 · What the line reaches, exactly

Start with the straightedge. It contributes nothing. A straightedge produces new points by intersecting two lines, and in this world there is only one line. It never crosses itself. Set the straightedge down; you will not need it again.

That leaves the compass. Suppose you have already built points a, b, and c. You open the compass to the distance |a − b|, plant it at c, and swing. A circle centered on the line meets that line in exactly two places:

That is the entire generative machinery. And now the argument writes itself in two directions.

Nothing but integers gets in. Induct on the number of construction steps. Your starting points, 0 and 1, are integers. If every point built so far is an integer, then a, b, and c are integers, so |a − b| is an integer, so c ± |a − b| is an integer. The set of constructible points is closed under the only operation available, and it starts inside ℤ. It can never leave.

And every integer does get in. From {0, 1} the distance 1 is available, so 1 + 1 = 2, then 3, and downward 0 − 1 = −1, and so on forever in both directions. Both inclusions hold, so the answer is not “at most the integers.” It is the integers, exactly.

Figure 01

Everything a compass confined to the line can find

•  −2
•  −1
•  0
•  1
•  2
•  3
•  ⅓
•  ½
•  √2
•  π

• Reachable without leaving the line
• Not reachable at any number of steps

Look at what that costs. The reachable set is closed under addition and subtraction and nothing else. It is a group, not a field. There is no multiplication, no division, no reciprocal. A line with nowhere to go cannot even halve itself.

One‑half is not a hard number to construct. It is an impossible one — if you are honest about staying on the line.

3 · One‑half already costs you a dimension

You know how to bisect a segment. Open the compass past halfway, swing an arc from each endpoint, and connect the two crossings. Notice where those crossings are. They are above and below the line. The construction works by building a point in the plane and using it to come back down.

The same is true of thirds, and of every rational. Euclid’s method — Book VI, Proposition 9 — draws an auxiliary ray from one endpoint at some nonzero angle, steps off n equal segments along it, joins the last one to the far endpoint, and draws parallels back. The whole construction is justified by the intercept theorem, and the intercept theorem is a statement about similar triangles. Triangles are not a convenience here. They are load‑bearing, and they are two‑dimensional.

So the rationals — the numbers nobody doubts belong on the number line, the ones we teach children to place on it before they can multiply — are already imports. Every one of them was fetched from somewhere else.

4 · The round trip that fetches √2

Now √2, which is worse in two independent ways.

First, it is not a ratio at all. The classical argument: suppose √2 = p/q in lowest terms. Then p² = 2q², so p² is even, so p is even, so write p = 2k. Then 4k² = 2q², so q² = 2k², so q is even too — contradicting “lowest terms.” No fraction does it. No decimal expansion terminates or repeats.

Second, and this is the part worth sitting with: the construction that does produce √2 exactly is the diagonal of a unit square. Not an approximation — a compass swung along that diagonal lands on the point with zero error. But a square has four sides at right angles. There is no square on a line. To get √2 you must leave, build in a dimension you were not given, and carry a length back down.

Figure 02

Where √2 actually comes from

Step 1 — on the line

You have 0 and 1. Available: the integers. That is all.

↑  leave the line  ↑

Step 2 — in the plane

Erect a perpendicular at 1. Complete the unit square. Draw its diagonal. Its length is exactly √2, by a theorem that requires a right angle to state.

• perpendicular — needs 2D
• square — needs 2D
• right angle — needs 2D
• Pythagoras — needs 2D

↓  swing the length back down  ↓

Step 3 — on the line again

Center the compass at 0, open it to the diagonal, and mark. That mark is √2, precisely. It is also a souvenir: nothing on the line produced it.

5 · The sheet of paper

That is the honest phenomenology of the situation, and it is why the number line is such a persistent source of student unease. When you write √2 on a line and put a dot next to it, you are reporting a result, not performing a construction. The dot is testimony from somewhere you did not go.

6 · Where the picture is wrong

Now the correction, because a good picture that is taken literally becomes a bad belief.

The real line is one‑dimensional. It is not secretly a plane. As a vector space over ℝ it has dimension one; its Hausdorff dimension is one; and √2 is genuinely a point on it, not a point hovering off it. Everything above is a claim about construction — about what a finite sequence of legal moves can locate — and construction is not the same thing as existence.

Which raises the obvious question: if you cannot construct the point, on what authority does it exist?

The answer is that mathematics found your objection unanswerable and responded by changing the terms. In 1872, Richard Dedekind published Stetigkeit und irrationale Zahlen and defined an irrational number as a cut — a partition of the rationals into a lower set and an upper set with no largest element below. √2 is not built. It is the name of the gap. Completeness, the property that distinguishes ℝ from ℚ, is an axiom: we assert that every such gap is occupied.

So when a student says “I can’t actually put √2 on this line,” the correct response is not that they are confused. It is that they have located the exact seam where nineteenth‑century analysis stopped constructing and started declaring — and that they should know a declaration when they see one.

What is true

Restricted to the line, straightedge and compass reach exactly ℤ. Rationals and √2 both require an off‑line construction. This is provable and not in dispute.

What is also true

The line is one‑dimensional, complete, and contains √2 as an ordinary point. The completeness axiom, not a construction, is what puts it there.

Where the metaphor breaks

“Irrationals are far off the edge of a plane” predicts that a bigger space would eventually reach them. It would not. The next section is the counterexample.

7 · The plane is not enough either

Grant yourself the whole plane. Perpendiculars, squares, circles, everything. You can now construct √2, and √(2+√3), and the golden ratio, and any number obtainable by a finite tower of square roots over ℚ. That is a lot.

It is also, provably, not everything. In 1837 Pierre Wantzel proved that no straightedge‑and‑compass construction doubles the cube or trisects a general angle — the cube root of 2 is out of reach. In 1882 Ferdinand von Lindemann proved π transcendental, which is why nobody will ever square the circle. Charles Hermite had done the same for e in 1873, and Joseph Liouville had exhibited the first transcendental numbers back in 1844.

And then the fact that reframes all of it. The constructible numbers are countable — each one is specified by a finite recipe, and there are only countably many finite recipes. Cantor showed in 1874 that ℝ is uncountable. A countable subset of the real line has Lebesgue measure zero.

Figure 03

What each dimension buys, and where the purchases stop

1. ℤ — the integers All you get on the line itself. Closed under addition and subtraction. Not a field. Countable
2. ℚ — the rationals Costs one dimension. Similar triangles and the intercept theorem buy you division. Countable
3. The constructible numbers Same dimension, more patience. Finite towers of square roots: √2, the golden ratio, the regular 17‑gon. Countable
4. The algebraic numbers Roots of polynomials with integer coefficients. Includes ∛2, which no construction reaches (Wantzel, 1837). Countable
5. ℝ — the reals Almost all of them are transcendental: π, e, and uncountably many with no name, no recipe, and no construction in any finite number of dimensions. Uncountable — the first four have measure zero inside it

Here, then, is the strongest true version of the intuition we started with. It is not that √2 is unreachable and the rest of the line is fine. It is that the numbers you can construct are a set of measure zero. Throw a dart at the number line and the probability that you hit anything you could have built, in any dimension, by any finite procedure, is exactly zero.

The line is not hiding a second dimension. It is doing something considerably more unsettling: it is almost entirely composed of points that no construction will ever touch, and it is one‑dimensional anyway.

8 · Exercises

1. Change the starting data Run the one‑line model from {0, α} for an arbitrary real α. Prove the reachable set is αℤ. What does this tell you about why the model can never produce a field?
2. Break the compass A collapsing compass forgets its opening the moment you lift it, so you may only reflect: c ± |c − b|. Show that from {0, 1} you still generate all of ℤ. Why does the restriction cost nothing here when it costs real work in the plane?
3. Audit a bisection Perform the standard perpendicular‑bisector construction on paper and mark every point that is not on the original line. Count them. That count is the price of one‑half.
4. Find the axiom Write out a Dedekind cut for √2 explicitly. Identify precisely the step at which you stop constructing an object and start asserting that one exists.
5. Argue the other side A constructivist rejects the completeness axiom as illegitimate. Steelman that position in one paragraph, then say what analysis loses if you grant it.

One last note on the history, since it usually gets told badly. The discovery of incommensurability is traditionally pinned on Hippasus of Metapontum, who legend says was drowned at sea for revealing it. Almost none of that survives contact with the sources: no ancient writer names Hippasus as the discoverer of irrationality, the accounts postdate him by centuries, and the ones that mention a drowning attribute it to a different disclosure entirely. The proof itself is traditionally printed as Proposition 117 of Book X of Euclid’s Elements and has been regarded since the early nineteenth century as a later interpolation. The mathematics is airtight; the martyrdom is not.