One of the greatest problems people who are getting into hardcore maths face, one of the blocks, is how *arbitrary* the axioms look. Axioms are the fundamental truths of a mathematical system, the things we do not prove. Whenever people are introduced to axioms, though, the most common feeling is that we should be able to somehow *prove* all of them and end up with no unproved theorems, no fundamental truths.

Now here’s why this is wrong.

What’s a natural number? Depending on your level of mathematical sophistication, you might not even remember that there’s such a difference. There are numbers that are natural? What are those? And the answer is that the natural numbers are… well, the natural ones. The ones we first think of, as a species, when we think of numbers; the ones that are sort of already in our brain when we’re born.

They’re You know, the counting numbers.

Let’s add to the mix because is a Very Important Number.

Now, suppose you have a baby mathematician AI. It’s super intelligent, but it has no knowledge. You can communicate with it, to an extent, but it doesn’t know numbers. If you show it scribbles like , it will just stare at you blankly. If you try to point to apples on a table it will tilt its head and think you’re crazy. How do you explain to this super mathematician what numbers *actually* are, beyond your innate intuition?

Axioms are logical sentences. But they’re not just any logical sentences: they’re logical sentences that pinpoint certain concepts from conceptspace. What does this mean?

Picture some imaginary space where all possible ideas and concepts lie. It’s vastly huge, it’s really unimaginable, but you know that somewhere there you’ll find the natural numbers. They’re a concept, right? So they must be in there somewhere. Now, how do you *find* the natural numbers there? How do you pinpoint them?

And that’s what axioms do. Axioms are sentences that are true of the things you’re trying to find, and of *nothing else*. That is, if a thing is a natural number, then the Natural Number Axioms are true of it; if a thing is not a natural number, then at least one of those axioms isn’t. So what *are* the Natural Number Axioms? What do you tell a baby mathematician AI so that it can find them in conceptspace?

**A0.** is a natural number.

This is our first axiom. If a thing is a sequence with all natural numbers, then there has to be a in there somewhere. So right now, our natural number set is just .

Then, we define equality.

**A1.** Reflexivity: If is a natural number, then .

**A2.** Symmetry: If and are natural numbers and , then .

**A3.** Transitivity: If , , and are natural numbers, , and , then .

Not even *equality* is a logical concept we take for granted. The three axioms above are well-known equivalence relation axioms: every binary relation that obeys these three axioms is known as an equivalence relation. But there’s another equality axiom that’s exclusive of natural numbers:

**A4.** If is a natural number and then is also a natural number.

These four axioms don’t really change what our set of natural numbers is. Any set that contains a number called “zero” so far is a natural number set.

Next, we define the successorship operation, Sx.

**A5.** If is a natural number, is also a natural number.

**A6.** If and are natural numbers and , then .

**A7.** There is no natural number such that .

The fifth axiom defines the successorship operation and creates the first way of finding new natural numbers. Every other way – addition, multiplication, exponentiation – will be found via recursive application of that operation. To understand what it means: is just the number . is the number which is – you guessed! – . So this is an operation that takes a natural number and finds the next one. But this is not enough. *Yes* our regular natural numbers, obey this axiom, of course. But a set where, say, does too. This would be a set that has a loop: ‘s successor is , but ‘s successor is also , and there are only these two numbers.

So the sixth axiom is there to prevent loops: if two numbers have the same successor, then they’re the same number. But it’s still not enough. With it, we have that if (that is, if ‘s successor is also ), then it must be the case that (that is, ‘s successor is , which means that ), and nothing prevents this from being the case yet.

And finally the seventh axiom is one that makes special: is not the successor of *any* number. There are no numbers before it. (“What?! What about ?!” Chill, young padawan, chill. Everything in time. For now, does not exist.) Now this axiom, with the previous one, guarantees that there can be no loops, since every loop would imply at some point that has a predecessor, and this is impossible. So our set of natural numbers now is something like .

You might be thinking that this is enough to pin down the natural numbers uniquely. This is true of every natural number, yes? So let’s put this to the test. Take a set that goes like this: . The numbers that have a after them are greater than *any* regular natural number. They’re also infinite in both directions. What axioms is this set violating?

is in it, so A0 is ok. We can just assume that Axioms 1 through 4 are also obeyed by the set. Axiom 5 is also obeyed: every number’s successor is a natural number, yes? So are axioms 6 and 7, there are no loops there, has no predecessor ( is *not* the same as , it’s not special, and it’s in fact infinite, so it can have predecessors). So, this weird set with those bizarre starry numbers obeys all axioms so far! We need the final and most important axiom: the axiom of mathematical induction.

**A8.** Induction: If is a set that contains the number and has the property “If is in then is also in ,” then contains all natural numbers.

*Now* we’re done. Let’s say the set is just , the set of all natural numbers. is certainly in it, and Axiom 5 means every number that’s reachable via successorship from is also in it. Therefore, we have finally pinned down uniquely the set using those axioms.

Now, if you followed this construction closely, you’ll have seen that without the axioms, *lots* of other concepts could also fit the concept of natural number! That’s what axioms *do*, they slowly constrain conceptspace until they’ve squeezed every other concept out.

A historical example of the importance of axioms is Euclidean Geometry. Euclid (and everyone else) had an intuitive understanding of what geometry is and how it should behave. So he made four postulates that he thought were necessary, and a fifth, the Parallel Postulate, which he and many other people throughout the ages felt should be derivable from the first four postulates. The first four postulates defined the mathematical entities like “point” and “line,” and the fifth said that two parallel lines never intercept. From the wordless, intuitive concept of point and line, that sounds pretty obvious, so Euclid always felt that we should be able to somehow prove that fifth postulate without having it as an axiom. That is, he wanted to *prove* its theoremhood, instead of assuming it.

Turns out he couldn’t. Why? Because the wordless intuitive entities he called “point” and “line” had, imbued in them, the Parallel Postulate. But they weren’t fully defined by the other four postulates. That is, to talk about points and lines in the intuitive fashion, you need *all five postulates*, and if you discard the fifth, you get the so-called Non-Euclidean Geometries. Everything that obeys the first four is a geometry, and there you bend and twist the meaning of point and line beyond recognition, unless you also have the fifth one.

As a final consideration, I’ll comment on what exactly I mean when I say I’m a mathematical realist/platonist. I think that mathematical entities have some reality external to the humans who think them. And what *is* that reality? It’s embodied in the axioms, and in model theory: everything that obeys certain axioms *has to* obey all of their consequences. No matter what any human or other sapient being has ever thought, regardless of any person’s beliefs and theorems and ideas, it’s simply true that if two apples fall from a tree, and then another two, there will have fallen four apples.

In fact, “quantity of apples that fell from a tree” (henceforth called “quapp”) obeys all 9 of the above axioms! To see that, let’s say our successorship operation is the physical act of “an apple falling from the tree,” and equality is “comparing the number of apples that fell from two trees.” Then:

**A0.** Naturally, there can be apples on the ground, because no apples fell.

**A1.** If you compare a tree with itself, the number of apples that fell from it is the same.

**A2.** If you see that as many apples fell from tree A as from tree B, then as many apples fell from tree B as from tree A.

**A3.** If you see that the same number of apples fell from tree A and tree B, and also tree B and tree C, then the same number of apples fell from tree A and tree C.

**A4.** This is playing with words, but if apples fell from tree A, apples fell from tree B, and both are the same number, then the same number of apples fell from tree B.

**A5.** If apples fell from a tree and a further apple falls, then that new number of apples that fell is a quapp.

**A6.** If the quapp from tree A is the same as that from tree B, and they’re the successor of some other quapp, then before the last apple fell from each tree they also had the same quapp.

**A7.** There is no quapp that comes before no apples having fallen from a tree.

**A8.** The same argument we used to define the set can be used to define the set of all apples that fell from a tree.

Well, what do you know, the quantity of apples that fell from a tree is, in fact, a number! And so, everything you prove about natural numbers can also be said of quapps. And it’s in *that* sense that mathematics is external to humans, and has a separate reality.

It’s in *that* sense that I’m a platonist.

You say that you’re a realist and use quapps as an example, but quapps don’t obey all the natural number axioms. Let’s say I have a tree with 50 apples on it. Then the successor operation is not defined for 50. No more apples can possibly fall from the tree; S50 is not a quapp because S50 does not exist at all. Similarly, no structures embedded in the Universe perfectly correspond to standard Euclidean geometry (or even any of the standard noneuclidean geometries).

I believe that you said elsewhere that you are still a mathematical realist even when there is no physical model of a mathematical theory, at least not in our Universe, but from reading your post here it seems that you require a physical model. Could you please clarify your thoughts?

I was trying to convey a certain sense, and the quapps analogy was the best I could find. I could go on and say that “if we waited for more apples to sprout” or “if we had an infinite tree” but then the analogy would be less clear. Maybe another idea would be analogise something like “the stars in a Universe” since ours is theoretically spatially infinite, but once again the sense is subtler. Yes, I’m a mathematical realist even when there is no physical model of a mathematical theory in our universe. The sense I was trying to convey was that the logical consequences of a theory are set regardless of what any humans may think or believe. The process of theorem-finding is a process of

discovery, not a process of invention. Thelanguageof mathematics is of course an invention, but “2 + 2 = 4” or “the Axiom of Choice is independent of ZF” look to me much more like territory than map.Pingback: Bayes’ Theorem | An Aspiring Rationalist's Ramble

Pingback: How to prove stuff | An Aspiring Rationalist's Ramble

Is there a particular reason you go with the second-order induction axiom?

I ask because, as is well known, first-order theories (with infinite models) cannot be categorical, by the Löwenheim-Skolem theorem. The result is that first-order theories cannot constrain conceptspace until they’ve squeezed out every concept but one. It seems like you are avoiding talking about first-order theories, the most commonly used theories in mathematical logic, because it would weaken your argument about the role of theories.

Perhaps you respond by saying that that’s exactly the point, that this non-categoricity is a weakness in first-order logic and hence you went with second-order induction. However, all you’ve done is push the problem back to sets. What subsets of the natural numbers exist? This now requires some substantial set theory to say something about. All kinds of answers to this question are possible. For example, starting from any model of ZFC, we can force to get a generic extension of the model with additional subsets of

N. We can also go the other direction and look at inner models which don’t contain all of the ground universe’s subsets ofN.The upshot of all this is that it’s rather a lot to say what subsets of

Nwe have for second-order induction. This further isn’t helped by the fact that we can purposefully throw out some subsets we do have to get a model of second-order PA. For example, we could reinterpret “set” in your formulation of induction to mean “definable set” and get a structure that satisfies second-order PA.In short, there’s a lot of subtleties that you gloss over here. If we pay attention to them, we see that your thesis is on shaky ground.

Oh, I know. I’m glossing over them on purpose. This is supposed to be a particularly intuitive and simple explanation, not a rigorous one. I

willwrite on first- vs. second-order logics and the Löwenheim-Skolem Theorem and ZFC and Model Theory in general in the future. Just hold on 😛