Consistency, Independence, and Models

Three questions you must ask any axiom list

In the previous guide you watched Hilbert's axioms patch every crack Euclid left open — the missing betweenness, the unstated 'the two circles actually meet', the silent appeals to a picture. But a tidy list of rules is not automatically a good list of rules. Before we trust it, three questions deserve honest answers, and they are surprisingly different from one another.

First: do the axioms quietly contradict one another? An axiom system is consistent if you can never derive both a statement and its negation from it. An inconsistent system is the worst kind of disaster, because from a contradiction you can prove anything — every theorem and its opposite, all at once — so the whole edifice says nothing. This is the property called consistency, and it is non-negotiable.

Second: is each axiom actually pulling its weight? An axiom is independent of the others if it cannot be proved from them — if dropping it genuinely loses information. The famous candidate, of course, is the parallel postulate: for two thousand years people suspected it was a hidden theorem in disguise, secretly provable from the rest. Showing it is truly independent was the whole drama. Third, underlying both: is there any world at all where every axiom comes true? That world is a model, and as we will see, it is the single tool that settles the other two questions.

A model: giving the undefined words a home

Recall from the very first rung that 'point', 'line', and 'plane' are undefined words — the axioms only describe how they behave, never what they are. A model takes that freedom seriously. It chooses concrete objects to play the role of each undefined word, then checks that every axiom comes out true under that interpretation. If it does, the model is a living witness that the axioms can all hold together at once.

Here is the smallest, most concrete example. Forget geometry's usual pictures and build a tiny finite world from three objects we will call A, B, C. Declare that the 'points' are exactly A, B, C, and that the 'lines' are the three pairs {A,B}, {A,C}, {B,C}. Now read the incidence axioms — 'through any two points there is exactly one line', 'every line has at least two points', 'there exist three points not all on one line' — and check each by hand against this little table.

points : A, B, C
lines  : {A,B}   {A,C}   {B,C}

through A and B  ->  exactly {A,B}      OK
through A and C  ->  exactly {A,C}      OK
through B and C  ->  exactly {B,C}      OK
each line has 2 points                  OK
A, B, C not all on one line             OK

Every incidence axiom checks out, so this three-point triangle of dots is a perfectly legal geometry. That tiny success carries a big payload: because a model exists, the incidence axioms cannot secretly contradict one another. This is the engine of the whole subject — a model proves consistency. You do not argue abstractly that no contradiction can ever appear; you exhibit one world where all the axioms are plainly true at the same time, and a true statement and its negation can never both be true in the same world.

Relative consistency: standing on arithmetic

The three-point model was so small you could verify it by eye. Full Euclidean geometry has infinitely many points, so we cannot lay them all on a table. The classic move instead is to build a model out of numbers: let a 'point' be an ordered pair of real numbers (x, y), let a 'line' be the solution set of an equation a x + b y = c, and define distance by the familiar |AB| from the Pythagorean relation. One can then check that every one of Hilbert's axioms becomes a true statement about real numbers.

This is honest but it comes with a caveat worth stating plainly. The coordinate model does not prove geometry is consistent in some absolute, unconditional sense — that turns out to be impossible to establish from inside. What it proves is relative consistency: if the arithmetic of the real numbers is free of contradiction, then so is Euclidean geometry. We have not removed all doubt; we have transferred geometry's trustworthiness onto the more elementary trust we already place in numbers. That is exactly how modern foundations work, and pretending otherwise would be the falsehood we promised never to tell.

Independence: changing one axiom, building two worlds

Models do double duty: the same idea that proves consistency also proves independence. To show an axiom P is independent of the others, you do not search forever for a missing proof. Instead you build two models. In one, all the other axioms hold and P holds. In the other, all the other axioms still hold but P fails. If both worlds are legal, then the others cannot force P either way — so P genuinely adds new information and earns its place on the list.

Now apply this to the prize specimen. Strip the parallel axiom away and keep only what remains; that common core is neutral geometry, the subject of the next guide. For the axiom P, use Playfair's axiom — 'through a point not on a line there is exactly one parallel'. The ordinary coordinate plane is a model where neutral geometry holds and Playfair holds: one parallel, as you always expected. The question is whether a second, equally legal model exists where neutral geometry still holds but Playfair fails.

It does. The Poincare disk model reinterprets the words: a 'point' is a point inside a fixed circle, and a 'line' is an arc of a circle meeting the boundary at right angles. In that disk, all the neutral axioms still come out true — but through a point not on a given line you can draw infinitely many non-intersecting 'lines'. Playfair fails, dramatically. This is the world of hyperbolic geometry, and its mere existence is the proof, finally, that the parallel postulate is independent: it could not have been a theorem all along, or no such model could exist.

Categoricity: one shape, or many?

One last question rounds out the picture. We saw that the incidence axioms alone are satisfied by a three-point triangle — but also by the infinite coordinate plane, and by countless other worlds of every size. An axiom system that allows wildly different-shaped models is weak: it pins down very little. An axiom system is categorical when, in effect, it has essentially one model — any two worlds satisfying it are structurally identical, mere relabellings of each other. This is the property called categoricity.

Here is the crucial contrast. The full set of Hilbert's axioms, including its continuity assumptions, is categorical: every model is the ordinary plane in disguise, so 'Euclidean geometry' really names a single shape. But neutral geometry — Hilbert minus the parallel axiom — is not categorical, and that is the whole point. The flat coordinate plane and the curved hyperbolic disk are both models of it, yet they are profoundly different worlds. The non-categoricity is exactly the room in which the parallel postulate gets to make a choice.

So the discovery of non-Euclidean geometry did not break Euclid, and it is a myth to say it did. It revealed that the neutral core is silent about parallels, leaving a genuine fork in the road. Add 'exactly one parallel' and you walk into the Euclidean plane; add 'many parallels' and you walk into the hyperbolic disk; each is true to its own axioms, each backed by an honest model. Consistency, independence, and categoricity are the three lenses that let you see the fork clearly — and choose with open eyes.