Chapter 1 Counting from nothing

18 minute read

Illustration for Chapter 1, Counting from nothing

1.1The crisis of naive intuition

Before we can speak of spacetime, energy, or subatomic particles, we need a language capable of expressing identity, plurality, and change. For most of human history, mathematicians treated numbers and geometric shapes as self-evident features of the world. Counting sheep or measuring a field required no philosophical defense. Yet by the late 19th century, the rapid growth of calculus and abstract algebra forced mathematicians to ask what numbers actually are.

In the 1870s and 1880s, the German mathematician Georg Cantor proposed a bold answer. He sought to ground all of mathematics in a single primitive notion, which he called a set. A set is simply a collection of distinct objects conceived of as a single whole. The objects inside a set are called its elements or members.

Early set theorists relied on an appealingly simple rule known as unrestricted comprehension. This rule states that any clearly stated property automatically defines a valid set consisting of every object that satisfies that property. If you can name a condition—such as being a prime number or being a planet—then the collection of everything meeting that condition forms a set.

For two decades, this intuitive foundation seemed unshakable. Mathematicians used it to build elaborate towers of arithmetic and analysis. Then, on June 16, 1902, the British philosopher and logician Bertrand Russell wrote a short letter to his German colleague Gottlob Frege. Frege was just about to publish the second volume of his life's work on the logical foundations of arithmetic. Russell's letter contained a single question that brought Frege's entire system crashing down.

Most sets we encounter in everyday thought are not members of themselves. The set of all stars is not itself a glowing sphere of plasma, so it does not belong to itself. On the other hand, one might imagine the set of all abstract concepts, which is itself an abstract concept and would therefore belong to itself. Russell asked what happens when we consider the property of not belonging to oneself.

If unrestricted comprehension holds, there must exist a Russell set whose members are all those sets that do not contain themselves as members. Now ask a simple question about this Russell set: does it belong to itself? Testing whether the Russell set satisfies its own entry requirement produces a devastating logical short circuit:

R∈R  ⟺  R∉RR \in R \iff R \notin R

If the Russell set belongs to itself, then by its own definition it cannot belong to itself. Conversely, if it does not belong to itself, then it meets the exact entry condition and must belong to itself. In classical logic, every statement is either true or false and never both, so a statement equivalent to its own negation is a contradiction. Under the logical principle of explosion, a formal system that admits even a single contradiction can prove any statement whatsoever. A mathematics that can prove both that two plus two equals four and that two plus two equals five is completely useless.

To rescue mathematics without abandoning the infinite worlds Cantor had discovered, Ernst Zermelo and Abraham Fraenkel rebuilt set theory from the ground up. With later refinements by Thoralf Skolem and John von Neumann, their framework became known as Zermelo-Fraenkel set theory with the axiom of choice, or ZFC. They realized that Russell's paradox arose because unrestricted comprehension allowed mathematicians to conjure sets of unrestricted size out of thin air. In ZFC, sets cannot be summoned merely by stating an arbitrary property. Instead, they must be built step by step from a single starting point using 10 strictly regulated rules of construction.

In the universe of ZFC, every mathematical object under discussion is a set, and the entire theory is governed by equality and set membership. The first rule is the axiom of extensionality, which defines what it means for two sets to be identical. Two sets are equal if and only if they contain the exact same members, regardless of how those sets were described or constructed.

Extensionality tells us how to compare sets, but it does not guarantee that any sets exist. That job falls to the second rule, the axiom of the empty set, which asserts that there exists a set containing no elements at all. Because extensionality judges sets solely by their members, there can be only one empty set. If two sets both have zero members, they share the same members vacuously and must be identical. Mathematicians denote this unique, featureless void by the symbol ∅\emptyset.

Once we possess the empty set, the next three axioms allow us to assemble larger collections. The axiom of pairing states that for any two sets, there exists a pair set whose only members are those two sets. When the two starting sets are identical, extensionality collapses the pair into a singleton set having just one member. Next, the axiom of union states that for any collection of sets, there exists a unified set containing every element of every member of the collection. The axiom of power set then provides a much more powerful growth engine. For any set, it guarantees the existence of a power set whose members are all the subsets of that set.

To filter sets without reviving Russell's paradox, Zermelo replaced unrestricted comprehension with the axiom schema of specification. Under specification, a logical property can only be used to carve a smaller subset out of an already existing set. You cannot ask for the set of all things in the cosmos that do not contain themselves. You can only take a pre-existing set and collect the members inside it that do not contain themselves.

This single restriction completely disarms Russell's paradox. If we carve out the subset of non-self-containing members from an existing set, that subset simply turns out not to be a member of the starting set. Rather than creating a logical contradiction, Russell's reasoning now proves a profound theorem: no set can contain every set, and a universal set of all sets cannot exist.

The axioms introduced so far allow us to build finite sets of any size, but they cannot carry us into the infinite. To cross that threshold, ZFC introduces the axiom of infinity. It asserts that there exists a completed infinite set containing the empty set and closed under forming the union of each member with its own singleton.

Three final principles complete the 10 axioms of ZFC. The axiom schema of replacement states that applying a well-defined rule to every member of an existing set produces another valid set. The axiom of regularity, also called foundation, stipulates that every non-empty set contains at least one member that shares no elements with the set itself. Regularity forbids infinite downward chains of membership and ensures that no set can ever be a member of itself. Finally, the axiom of choice states that given any collection of non-empty, non-overlapping sets, there exists a choice set containing one representative from each member of the collection. Together, these 10 rules form a self-contained logical engine capable of generating all of mathematics from the void.

1.2Building numbers from the empty set

How can a universe containing nothing but the empty set give birth to counting? In the 1920s, the young Hungarian-American mathematician John von Neumann showed that numbers do not need to be assumed as mystical primitives. Instead, every natural number can be built directly out of the empty set by identifying each number with the set of all numbers that come before it.

Von Neumann began at the absolute beginning by defining zero to be the empty set itself. Zero is literally the set with zero elements inside it. To get the next number, we apply a successor operation. It takes any number and forms a new set containing all the elements of that number together with the number itself. Applying this successor step to zero produces the number one, which is the singleton set containing zero as its sole member. Applying the successor step again yields two, three, and every subsequent whole number in an unbroken chain:

0≡∅,1≡{0},2≡{0,1},3≡{0,1,2}0 \equiv \emptyset, \quad 1 \equiv \{0\}, \quad 2 \equiv \{0, 1\}, \quad 3 \equiv \{0, 1, 2\}

Von Neumann's construction achieves a quiet miracle of economy. Out of pure nothingness, merely by placing braces around what we already have, it builds an infinite staircase of distinct mathematical objects. Better still, the number of elements inside the set representing any whole number is literally that number. Zero contains no elements, one contains the single element zero, two contains zero and one, and three contains zero, one, and two. Even numerical order emerges for free without a new definition. Saying that one number is smaller than another is the exact same thing as saying that the first set is a member of the second.

To gather all of these individual counting numbers into the single completed set of natural numbers N\mathbb{N}, we turn to the axiom of infinity. That axiom guarantees the existence of at least one inductive set that contains zero and contains the successor of every element inside it. Using specification, we define the set of natural numbers as the intersection of all inductive sets. That intersection retains only the minimal chain of whole numbers built from zero.

In 1889, the Italian mathematician Giuseppe Peano had formulated five famous axioms that any valid system of natural numbers must obey. In von Neumann's construction, all five of Peano's rules are no longer unproven assumptions, but provable consequences of set theory. First, zero is a natural number because the empty set belongs to every inductive set. Second, the successor of every natural number is also a natural number. Third, zero is never the successor of any natural number, because the successor of any set always contains that set as a member and can never be empty.

Peano's fourth rule states that two distinct numbers can never share the same successor. In von Neumann's system, this follows directly from the axiom of regularity. If two distinct numbers had the same successor set, each would have to be a member of the other, creating a forbidden two-step membership loop that violates regularity.

Peano's fifth and most celebrated rule is the principle of mathematical induction. Suppose a subset of the natural numbers contains zero, and whenever it contains a number it also contains that number's successor. Induction states that this subset must be the entire set of natural numbers. In our set-theoretic construction, induction holds automatically because any such subset is an inductive set, and the natural numbers were defined as the smallest collection contained inside every inductive set.

With the staircase of natural numbers in place, the operations of arithmetic follow from two simple recursive rules. Addition is defined by specifying that adding zero leaves a number unchanged, while adding the successor of a number is the successor of their sum. Multiplication is defined in the exact same way as repeated addition, where multiplying by zero gives zero and multiplying by a successor adds one more copy of the number. From these recursive rules and mathematical induction, all the familiar laws of elementary arithmetic—associativity, commutativity, and distributivity—follow with complete rigor. We have learned to count and calculate starting from an empty box.

1.3From natural numbers to ratios

As useful as the natural numbers are for counting discrete objects, they fail the moment we try to keep a ledger of debts or divide a loaf of bread. Among the natural numbers, we can always add or multiply, but we cannot always subtract or divide. You can subtract two from five to get three, but subtracting five from two has no answer among the positive whole numbers.

Rather than inventing negative signs by decree, mathematicians build the integers directly out of pairs of natural numbers. First, however, we must explain how a set can encode an ordered pair with a clear first and second entry, since ordinary sets ignore order. In 1921, Kazimierz Kuratowski solved this problem with a clever definition that captures order using only nested braces, writing {{a},{a,b}}\{\{a\}, \{a, b\}\} for the ordered pair of a and b. The first entry is the lone element inside the one-element set, while the second is the other element of the two-element set (or equals the first when both coincide).

Armed with ordered pairs, we can represent any integer as the formal difference between two natural numbers. Think of an ordered pair of natural numbers as an accountant's balance sheet recording credits in the first column and debts in the second column. When credits exceed debts, the net balance is positive; when credits equal debts, the balance is zero; and when debts exceed credits, the balance is negative.

Of course, many different balance sheets represent the exact same net worth. A ledger with five credits and two debts represents the same net balance of plus three as a ledger with three credits and zero debts. Likewise, one credit and four debts gives the same minus-three balance as two credits and five debts. We can test whether two ledgers represent the same net balance without ever mentioning subtraction. We simply check whether the credits of the first plus the debts of the second equal the debts of the first plus the credits of the second.

Notice that this test uses only addition of natural numbers, which we already constructed in the previous section. It partitions the entire grid of natural-number pairs into bundles of equivalent ledgers called equivalence classes. Each integer is defined as one complete equivalence class of ledgers sharing the same net balance. When we add two integers, we simply add their credits together and add their debts together. Under these rules, the integers form what algebraists call a commutative ring. Zero is the class of balanced ledgers where credits equal debts, and every integer has an opposite obtained simply by swapping its credit and debt columns. Subtraction is now possible everywhere.

Yet division still remains blocked. You cannot divide one integer by another unless the first happens to be an exact multiple of the second. To build the rational numbers—the fractions—we apply the exact same equivalence-class trick a second time. We consider ordered pairs of integers where the second entry, the denominator, is never zero.

Just as different ledgers represent the same integer, different integer pairs such as one-over-two and two-over-four represent the same ratio. Two fraction pairs are equivalent whenever the first numerator times the second denominator equals the first denominator times the second numerator. Each rational number is defined as the complete equivalence class of integer pairs satisfying this cross-multiplication test.

In this new realm of rational numbers, every non-zero fraction has a multiplicative inverse obtained by flipping its numerator and denominator upside down. We can now add, subtract, multiply, and divide freely by any non-zero quantity. The rational numbers form a densely ordered field. Between any two distinct fractions, no matter how close together they lie, we can always find their average, and therefore infinitely many other fractions packed tightly between them.

1.4Completing the continuum

Because the rational numbers are packed so densely along the number line, the ancient followers of Pythagoras believed that every length in nature is a ratio of two whole numbers. Around the 5th century BCE, a Pythagorean mathematician—traditionally said to be Hippasus of Metapontum—made a discovery that shattered that creed. Draw a simple square whose sides have length one. By the Pythagorean theorem, the square of its diagonal equals one squared plus one squared, which is two. Yet no fraction of whole numbers can ever square to two.

The proof is a jewel of elementary reasoning. Suppose a fraction reduced to lowest terms—so its numerator and denominator are not both even—squared to two. Then the square of the numerator would equal twice the square of the denominator, making the numerator's square even, which forces the numerator itself to be an even number. But if the numerator is a multiple of two, its square is a multiple of four, which forces the denominator's square—and thus the denominator itself—to be even as well. That contradicts our assumption that the fraction was already reduced to lowest terms.

This discovery reveals a profound geometric flaw in the rational numbers. Even though fractions are everywhere dense, the rational number line is perforated by infinitely many microscopic pinholes. Consider the set of all rational numbers whose squares are strictly less than two. This collection is bounded above among the rationals, yet it possesses no least upper bound within the rational numbers. If you slide along the rational line from below the square root of two to above it, you pass right through an empty pinhole.

Without a seamless continuum, calculus falls apart. A continuous curve could pass from negative values to positive values right through a rational pinhole without ever crossing zero. That would destroy the intermediate value theorem and make differential equations impossible to formulate. Physics requires a number line with zero gaps.

In 1872, the German mathematician Richard Dedekind found a way to seal every gap in the rational line at once using nothing but sets of fractions. Dedekind realized that even when an irrational point like the square root of two is missing from the rationals, it still cleanly slices the rational numbers into two halves. On the left lie all the fractions smaller than the target spot, and on the right lie all the fractions larger than it. We can therefore define any real number—rational or irrational—by identifying it with the leftward ray of all rational numbers strictly less than it.

Formally, a Dedekind cut is a subset of the rational numbers that satisfies three simple conditions. First, the cut is neither empty nor the entire set of rationals. Second, the cut is closed downward, meaning that if a fraction belongs to the cut, every smaller fraction also belongs to it. Third, the cut has no greatest element, meaning that for every fraction inside the cut, there is always a slightly larger fraction inside the cut as well.

Because every Dedekind cut is simply a subset of the rational numbers, it is a member of the power set of the rationals. Using the axiom of power set and specification, we define the set of real numbers R\mathbb{R} as the collection of all Dedekind cuts. Ordering two real numbers reduces once again to ordinary set inclusion, and adding two cuts is defined by adding all pairs of their rational members.

Why does Dedekind's construction guarantee that the real line has no holes left in it? Suppose we have any non-empty collection of real numbers that is bounded above. To find the least upper bound of that collection, we simply take the set-theoretic union of all the Dedekind cuts in the collection. Each cut in the collection is a downward-closed ray with no greatest element. Their union automatically inherits all three properties of a valid Dedekind cut, and forms the exact least upper bound of the collection. By turning points on a line into rays of sets, the axiom of union seals the continuum into a complete ordered field.

One final algebraic obstacle remains before our number system is ready for quantum physics. Whenever you square a positive or negative real number, the result is never negative, so no real number squares to minus one. In quantum mechanics, nature relies on wave amplitudes that rotate smoothly through phase angles. Describing those continuous rotations requires a number system where every polynomial equation can be solved.

In the 1830s, the Irish mathematician and physicist William Rowan Hamilton showed how to build the complex numbers without invoking any mystery about "imaginary" quantities. Hamilton defined each complex number as an ordered pair of real numbers—a point on a two-dimensional plane—equipped with ordinary vector addition and a rotational rule for multiplication. Under Hamilton's rule, the vertical unit pair on the plane, called the imaginary unit, squares to minus one on the horizontal axis. By the fundamental theorem of algebra, first rigorously proved by Carl Friedrich Gauss, every non-constant polynomial equation with complex coefficients has a complete set of solutions in the complex plane. Starting from the empty set and the ZFC axioms, we have built the natural numbers, integers, rationals, real continuum, and complex numbers without smuggling in a single outside assumption.

1.5Cantor and the hierarchy of infinities

Having built both the discrete counting numbers and the continuous line of real numbers, Georg Cantor asked a question that no one had posed with mathematical precision before. Are all infinities the same size, or are some infinities genuinely larger than others?

To compare the sizes of two sets without counting them one by one, Cantor used the idea of a one-to-one correspondence, or bijection. Two sets have the same cardinality—the same size—if and only if their elements can be paired up one-to-one with nothing left over on either side. Imagine walking into a vast auditorium where every seat is occupied by one person and no one is standing. You know at a glance that the number of people equals the number of seats, without having to count either group.

For infinite sets, this pairing criterion produces counterintuitive wonders. Both the integers and the dense rational fractions can be arranged in a systematic list and paired one-to-one with the natural numbers. They all share the exact same countable infinity, which Cantor denoted by the Hebrew letter aleph with a subscript zero, ℵ0\aleph_0.

For several years, Cantor wondered whether the points on the continuous real line could also be listed in sequence. In December 1873, he proved that they cannot. In 1891, he distilled his discovery into one of the most celebrated arguments in human thought, the method of diagonalization.

In its general form, Cantor's theorem states that for any set whatsoever, its power set—the set of all its subsets—always has a strictly larger cardinality than the original set itself. It is easy to see that the power set is at least as large as the original set, since each element can be paired with its own singleton subset. The triumph of Cantor's argument is showing that no pairing can ever cover all the subsets in the power set.

Suppose someone claims to have found a complete listing that pairs every element of a set with one of its subsets. They claim that every possible subset appears somewhere on the list. For each element, we can ask a simple question: does that element belong to the subset with which it is paired? Using the axiom schema of specification, Cantor constructed a special anti-diagonal subset consisting of precisely those elements that do not belong to their own paired subset.

Because this anti-diagonal collection is a well-defined subset of the original set, it must belong to the power set. If the proposed listing truly covered the entire power set, some element in the original set would have to be paired with this anti-diagonal subset itself. Now ask whether that specific element belongs to the anti-diagonal subset. By definition, it belongs to the anti-diagonal subset if and only if it does not belong to the subset paired with it—which is the anti-diagonal subset itself.

Here Russell's paradox returns in a productive disguise. Because the anti-diagonal subset was carved out of an existing set via specification, mathematics does not collapse. Instead, the contradiction simply refutes our assumption that the list covered the entire power set. The anti-diagonal subset is guaranteed to be missing from the list, because it was engineered to differ from every listed subset at the diagonal element.

Each real number can be represented by an infinite binary expansion, which corresponds to a subset of the natural numbers. Consequently, the cardinality of the real continuum equals two raised to the power of aleph-null. By Cantor's theorem, the continuum is uncountably infinite. Even an infinitely long numbered list of real numbers will miss almost the entire number line. Better still, because Cantor's theorem applies to any set, taking the power set of the continuum repeatedly generates an endless upward tower of ever-vaster infinities:

ℵ0<2ℵ0<22ℵ0<⋯\aleph_0 < 2^{\aleph_0} < 2^{2^{\aleph_0}} < \cdots

David Hilbert declared that no one should ever expel us from the paradise that Cantor had created. Yet Cantor's diagonal trick was far more than a telescope into the transfinite. In Chapter 3, we will see how Kurt Gödel and Alan Turing turned Cantor's diagonal argument into the foundation of modern logic and computation. First, however, we must see how the real and complex numbers acquire shape, curvature, and symmetry.