What Is a Number
Begin from a question any student of mathematics meets today: what are the natural numbers?
Before the end of the nineteenth century, the question had two kinds of ready-made answer. The philosopher's: numbers are abstractions of the mind — the "three" drawn out of three stones, three sheep, three days. The set theorist's: a number is a particular kind of set — 0 is the empty set, 1 is the set of the empty set, 2 wraps on one more layer, and so it goes on. The second kind of answer was seriously constructed in the late nineteenth century, and more than once. In Zermelo's scheme, 0 is the empty set, 1 is the set containing only the empty set, 2 the set containing only that one — one layer inside another; in von Neumann's, 0 is again the empty set, but 1 is the set containing 0, and 2 the set containing 0 and 1 — each number takes into itself all the numbers before it. The "2" of the two schemes is a completely different set. Which is the real 2?
Modern mathematics answers: the question has no meaning. The natural numbers are not any particular set but any sequence satisfying the Peano axioms — there is an initial element, there is a successor operation, the principle of induction holds. Zermelo's sequence satisfies them, von Neumann's satisfies them, a string of milestones counted off from the peak of the Brocken satisfies them too, so long as they are ordered rightly; they are isomorphic to one another, and in the sense of isomorphism they are one and the same thing. As for "number in itself" — the naked object left over when every structure has been stripped away — modern mathematics declines to speak. The objects have left the stage; what remains is the relations between structures and structures.
This exit did not happen overnight, and no legislature decreed it. It is the silent legacy left by the battle of foundations that ended without victor: since "what the foundation is" could not be adjudicated, mathematicians little by little learned to study "how the foundation works". This chapter tells the three steps of this silent revolution, and then performs the core transcription of Part Three and closes it.
The First Step: Axioms Become Implicit Definitions
The first step was taken in geometry.
In 1889 Peano, in his Arithmetices principia, gave the axioms of the natural numbers: initial element, successor, induction — the very criterion with which this chapter opened is of their issuing. In the same years, a more systematic project opened in geometry: in 1899 Hilbert published his Foundations of Geometry, carrying the axiomatization Euclid had begun and non-Euclidean geometry had forced forward to a wholly new degree of strictness.
The novelty of the book lies not in the content of the axioms but in their status. Euclid's "definitions" tried to say what the objects are: a point has no part, a line has length and no breadth — saying it amounts to saying nothing, for no one can reason with "having no part". Hilbert went the opposite way: point, line, and plane are not defined; what they are is fixed by the axioms as a whole. The transmitted reminiscence is his remark that those words might just as well be replaced by "tables, chairs, and beer mugs" — with the axioms standing unchanged, not one theorem would alter by a letter. This is implicit definition: the axioms no longer state truths about pre-existing objects; they stipulate the relations a class of objects must satisfy in order to enter this system; "point" thereby becomes anything that plays the role of point under the axioms. The foreshadowing planted in Chapter Eight is redeemed here: Pasch and Hilbert itemized Euclid's smuggled imports — superposition, order — promoting, one by one, relations taken on default for two thousand years into explicit axioms. The way a building is perfected is to replace every invisible brick with a visible one.
Once the status changed, the rank of the questions changed too. Since the axioms were no longer self-evident, two new questions became legitimate: is this set of axioms consistent (might it derive a contradiction)? and is each axiom independent (can it be derived from the others — the two-thousand-year open case of Chapter Eight is now a standard item on an axiomatic system's checklist)? Hilbert gave a new-style answer to both questions: the model. Interpret point, line, and plane in the analytic geometry of the real numbers, and the geometric axioms all translate into arithmetical propositions — the consistency of geometry is thereby reduced to the consistency of arithmetic; construct another model in which some axiom deliberately fails, and that axiom's independence is proved. This craft descends directly from the translation dictionaries of Beltrami, Klein, and Poincaré — the tool invented in Chapter Nine to rescue non-Euclidean geometry was generalized by Hilbert into the standard equipment of the axiomatic method. The axiom passed from "self-evident truth" (the bill for the misreading of Chapter Eight) to "implicit definition" — the stipulation of a set of relations; and the validity of a mathematical proposition therewith passed from "agreeing with the objects" to "holding in the structure". This is the methodologization of "necessity does not cross the boundary of the snapshot" (Chapter Nine): to exchange one set of axioms is to exchange one snapshot, and within each snapshot the necessity stands undiminished to a hair.
The Second Step: Structure Becomes the Hub
The second step was taken between the two wars, and those who took it were a group of young Frenchmen writing under a single pseudonym.
In the mid-1930s a collective of mathematicians from the École normale supérieure in Paris formed, determined to rewrite the textbooks of the whole of mathematics according to the new conception. They signed themselves with one pseudonym: Nicolas Bourbaki. The project ran for half a century; the Éléments de mathématique ran to volume upon volume — but what truly stood was the map of mathematics behind the skeleton of the books. In Bourbaki's depiction, mathematics is not a tree classified by objects (geometry, arithmetic, algebra, each upon its bough) but a building composed of structures: there are three mother structures — algebraic structures (groups, rings, fields: families of elements that can be composed and inverted), order structures (partial order, total order: families of elements that can be compared), topological structures (open sets, limits: families of elements that can be approached) — which combine with one another to beget more complex structures. One and the same structure shows its face everywhere: the language of group theory writes with equal penetration the symmetries of geometry, the solutions of equations, the classification of crystals, and the spectra of particle physics; topology governs both the connectedness of surfaces and the convergence of function spaces. The unity of mathematics lies not in objects — objects differ from one another — but in structure: mathematics keeps meeting the same structure again in its different corners, and the meeting is no coincidence, because what mathematics studies was structure all along.
On RC's reading, Bourbaki's conception of structure is a handsome re-convergence after the battle of foundations: none of the three snapshots (symbol system, mental construction, kingdom of objects) had prevailed, and the community went on to rewrite "what is mathematics" as "what are the working patterns that recur within mathematics" — which is in tune with the position RC's General Outline takes in Section 2.1, discussing processual completeness: the completeness of cognition should ultimately and only inhere in the capacity of the process to iterate, not in the degree of completion of its conclusions. What Bourbaki wrote is not the ultimate ontology of mathematics but mathematics' iterative working surface.
The Third Step: Arrows before Points
The third step is the most thorough: structure itself was brought under relation.
In 1945 Eilenberg and Mac Lane published in the Transactions of the American Mathematical Society "General Theory of Natural Equivalences" — a plainly titled paper since recognized as the birth certificate of category theory. Its origin was quite concrete: in algebraic topology, the homology groups computed by different methods for one and the same space always turned out isomorphic, and mathematicians called these isomorphisms "natural" — but what does "natural" mean? To say it strictly, the two had first to build parts more basic still: categories (a class of objects and the arrows between them — morphisms), functors (correspondences between categories), natural transformations (correspondences between functors). They set out to define "the natural correspondence from A to B", and found that they had first to take the whole system of arrows between A and B as an object of study.
What they built proved far larger than its motive. The principle of category theory can be put in one sentence: what an object is is determined by its relations to all the other objects. In the category of sets, to ask whether two sets are isomorphic, one may look only at the arrows; the mark of a singleton is that from it there is exactly one arrow to any set — the property "single point" has been translated into a property of a bundle of arrows. Yoneda's lemma later wrote this intuition into a theorem: every object is uniquely determined by its relations to the whole category. To know an object, one need not — and cannot — lift its lid and look inside: walk all the roads that lead to it, and it is known through and through. The verdict of Chapter Three here reaches its strongest form inside mathematics: relations precede objects, paths precede destinations. And it falls in astonishingly with the program of Chapter Three — mathematics is the self-closure and expansion of convergence paths: what closure closes into (objects) was never the center of study; the center is the paths themselves (structures, arrows, transformations). One may say that category theory is the contemporary cashing of that program: mathematics at last recognized the object of its study as its own method.
The Choice of Axioms Comes to the Front of the Stage
The revolution had one further, parallel front, which carried the conclusion of Chapter Nine farther still: the continuum hypothesis.
At the end of the nineteenth century Cantor asked: between the reals and the naturals, does there exist an infinity of intermediate size? The continuum hypothesis says: no. It is simple, natural, and either true or false — Chapters Six and Seven said that set theory is the foundation of contemporary mathematics, and this question is the most conspicuous loose stone in that foundation. The history of two thousand years of attempts upon the fifth postulate was re-enacted here, with the period shortened by one order of magnitude: for decades no one could prove it, and no one could refute it.
The verdict fell in two halves. Gödel announced in 1938 and published the proof in 1940: if ZFC is consistent, then adding the continuum hypothesis to it leaves it still consistent — the hypothesis cannot be refuted. Twenty-three years later, in 1963, Cohen invented forcing and proved the other half: if ZFC is consistent, then adding the negation of the continuum hypothesis is also consistent — it cannot be proved either. The two halves join into one judgment: the continuum hypothesis is independent of ZFC — within the foundations now in force, it is neither true nor false. The first incompleteness theorem of Chapter Seven had exhibited the system's boundary with an artificially constructed sentence G; the continuum hypothesis is a natural resident of that boundary — a proposition no one designed that happens to live just outside the system's wall. Gödel's own reaction was Platonist (the last chapter told of it): the hypothesis has a determinate truth value, and ZFC simply does not yet carry enough axioms to see it clearly — his meaning was quite concrete: go and find new axioms. In the years that followed, the mathematical community seriously weighed large cardinal axioms, the axiom of determinacy — each candidate bringing its own costs and consequences, its adoption depending upon what consistent and fruitful mathematics it would deliver, not upon whether it was "self-evident".
Thus the file of Chapter Nine receives here its last entry: the choice of axioms = the choice of convergence paths — no longer a lesson confined to one corner of geometry but the daily, public working method of the mathematical community. In the battle of foundations the three schools argued over "what is the foundation of mathematics"; after the structuralist revolution, the grammar of practice became: choose a path, publish it explicitly, sound out its boundary, keep the interface for changing roads. This is exactly the form, inside mathematics, of the sustainable decision-making RC's General Outline discusses in Section 4.3 — keeping optionality present.
The Core Transcription: The Mode of Locking Becomes the Object
Now for this chapter's core transcription — the pivot of the pivot of Part Three.
What is structuralism? On the standard account: mathematics studies structures, not objects — that is what the preceding sections of this chapter have told. On RC's reading there is one turn further inward: structuralism is mathematics' self-understanding returning to "convergence paths". Isomorphism, in this book's vocabulary, is different realizations of one and the same structure of convergence — Zermelo's 2 and von Neumann's 2 are two track-layings of one and the same convergence path; to say that they "are the same number" is to say that the mode of locking is identical while the material locked is as you please. A category is the law of transformation of structures of convergence — it studies no particular building, but the structure-preserving passages between building and building: functors carry the structure of one world into another; natural transformations compare two ways of carrying. Relations taking the place of objects says: the center of gravity shifts back from "what is locked" to "how it is locked". Mathematics explicitly locks structures of convergence at the level of formal systems (Chapter Five), discovers in non-Euclidean geometry that the starting points can be exchanged (Chapter Nine), sees in the battle of foundations that there is no absolute foundation (Chapter Ten), and finally, in structuralism, makes that recognition itself into its object of study (this chapter) — what is being studied is no longer any locked mathematical object, but the mode of locking itself. The thesis of Part Three in one sentence: mathematics is the self-closure and expansion of convergence paths, and the self-awareness of contemporary mathematics is the recognition that this sentence is literally true — the object it studies is its own method.
Structuralism must also once more face Platonism, for it is often read as a mild edition of Platonism (a "kingdom of structures" in place of a "kingdom of objects"). On RC's reading, the divide lies elsewhere. Mathematical structures do have an astonishing stability across subjects — different civilizations, traditions out of all communication with one another, keep meeting the same structure again; Chapter Eight told of that terrifyingly thick account. But RC does not cash the account into ontology: the stability of structures is the output of the consensus reinforcement of Axiom A6 — all the subjects able to read a proof verifying one another, the loop running hundreds of millions of times — not the developing image of an independent kingdom. RC's General Outline says, discussing ideas and consensus in Section 2.4, that different ideas lock in different possible snapshots; the structures of mathematics are the most stable form among the results of convergence, stable enough to bear the name "necessary" — but by the file of Chapter Nine, the whole meaning of necessity is self-consistency inside the snapshot. Platonism asks "do structures really exist"; on RC's reading the question points in the wrong direction: a structure is the form taken by a mode of locking, and to ask whether it exists "in itself" is to take the snapshot for the Ground — the verdict Chapter Six passed upon "the set of all sets" applies equally to "the kingdom of all structures".
The Bill for Wigner's Question
The first puzzle the introduction laid down — why mathematics is effective for the world, which Wigner called a miracle — can now be paid its RC-style answer. The answer has four entries.
First, a common history of formation. The introduction and Chapter Three told of mathematics' origins: it differentiated out of such observational activities as counting, surveying, and the calendar. Mathematics' basic structures carry the shape of the environment of their formation — the integers for discrete things, geometry for extended land, probability for gambling and statistics — it is not a miraculous language arrived from outside; it is a language grown up out of this world. To say that the world "obeys" mathematics inverts the causality: first the world trod certain structures solid by repetition, and only then did mathematics extract them and make them explicit. The ground of the fit is the ground of common origin.
Second, the grammatical freedom of expansion. Mathematics' later expansion went far beyond its origins (imaginary numbers, higher dimensions, the non-Euclidean) — how explain that these also hit? Because the expansion is not arbitrary: it proceeds along the laws of combination and transformation of structures already locked — and this book here lays down its verdict: capacity grows along paths. The space of mathematical expansion is packed with possible structures, and the small handful that hit physics could hit precisely because they remain within this generative grammar: Riemann's geometry of curvature is no language fallen from the sky but a direct descendant of the generative matrix that Euclidean geometry is — that general relativity hit upon it means the world took one more step along the extension of its own grammar.
Third, the survivors' ledger. The introduction told the other face of the puzzle: the overwhelming majority of pure mathematics never meets any application at all. In RC's ledger this face is not redundancy but the key: it proves that the fit is not a universal miracle. The effective handful is retold again and again; the ineffective multitude enters no narrative — the selection effect magnifies the "miracle". Set the two ledgers side by side, and Wigner's question turns from "why does it all work" (it does not) into "why is there any that works" — and the first two entries have already answered where the "any" comes from.
Fourth, the suturing of re-convergence. Between mathematics and physics there is no once-for-all contract, only one explicit re-joining after another: Riemann 1854, Einstein 1912 to 1915, Grossmann handing over the dictionary; quantum mechanics and Hilbert space; each joining is a re-convergence — the theory rewritten in a new language, the old paths continuing in office as limiting cases (the 1919 of Chapter Three). The suturing was done afterward, not pre-established in harmony.
The four entries together: the effectiveness of mathematics for the world owes no debt to miracle — it owes the debt of a common history of formation and the wages of one re-convergence after another. The direction the introduction recorded — the rejoining of self-enclosed expansion to the world rests not on miracle but on a shared history of formation and on the suturing of one re-convergence after another — is hereby paid in full.
The Close of Part Three
Now the four chapters of Part Three can be closed.
Chapter Eight, the boundaries of Alexandria: the axiomatic method is the technique of explicitly publishing the starting point of a convergence path, the oldest specimen of the artificialization of consensus reinforcement; two thousand years of cross-subject re-verification without failure read a chosen starting point into eternal truth. Chapter Nine, the ghost of the fifth postulate: the failure of two thousand years of proofs sprang from the will to argue the starting point of one path into the only possibility — impossible in principle; the birth of non-Euclidean geometry walked the whole road of "observational divergence, system rebuilt elsewhere, re-convergence"; geometric necessity stood out as self-consistency inside the snapshot — and sixty years later, the world picked one frame from among Riemann's several geometries. Chapter Ten, the battle of the foundations: the common form of the three crises is the foundation pierced by observation into the next level down, "it has always been so" standing out as a locking never re-audited; the three foundational stances are three different snapshots, three strategies of re-convergence — no victor, and the building growing all the while. Chapter Eleven, relations taking the place of objects: structuralism shifted the center of study from the locked object back to the mode of locking itself — isomorphism is different realizations of one and the same structure of convergence, a category is the law of transformation of structures of convergence — and mathematics recognized the object of its study as its own method.
Look back at the question laid down in Chapter Three: at the cost of which cracks does a self-enclosed freedom of expansion come? The answer Part Three gives is the deepest reversal under this book's reading: irrational numbers, the infinitesimal, the paradoxes, the undecidable — this column of "cracks" is not the price of freedom; it is freedom's bookkeeping. Every crack is the available margin taking shape inside mathematics: Axiom A7 of RC's paper says that in locking determinacy, observation does not exhaust the Ground; there always remains margin not yet locked. A building without these cracks — sealed against everything, necessary in everything, self-guaranteeing even its own consistency — is precisely the building Chapter Seven proved cannot exist. The tremors of mathematics across these two hundred years, from Saccheri's thirty-three theorems to Cohen's forcing, can be restated in one sentence: margin is not mathematics' disgrace but the proof that mathematics can still re-converge. The chess player's rules are hardest, the navigator's words most fallible, and the architect — Part Three's protagonist — lies in her nature between the two: every building she raises is inwardly necessary, and her freedom lies wholly in the exchangeability of the foundations.
Part Three closes here. The joinings between mathematics and the world — Riemann and Einstein, the plates of 1919 — have pointed, again and again, toward the protagonist of Part Four: fallible convergence in dialogue with the world. Galileo's telescope still stands in the night of the introduction; now it is science's turn to take the stage.