The Recurrence

The phenomenon is old and well documented. In 1922 the sociologists William Ogburn and Dorothy Thomas published a list of one hundred and forty-eight cases of the same discovery made independently by two or more people; in 1961 Robert Merton sharpened the observation into a thesis — that all scientific discoveries are in principle multiples, including those that on the surface appear to be singletons. The roster is famous. The calculus (Newton and Leibniz). Natural selection (Darwin and Wallace — who independently arrived at it and mailed Darwin the essay that forced the Origin into print). The conservation of energy (Mayer, Joule, Helmholtz, Colding, and others, all within a decade). Oxygen (Scheele and Priestley). The telephone (Bell and Gray, who filed competing patent claims on the same February day in 1876). Logarithms, sunspots, the periodic table's gaps — the same idea, arriving from several hands at once.

Merton's explanation is sociological, and it has become the orthodoxy: a discovery is a function of the state of its field. When the prior results, the instruments, and the questions are all in place, the next step becomes available — and is then taken, independently, by whoever happens to be standing at the frontier. The phrase, much mocked but hard to improve, is that the times were right. The individual, on this view, is to a degree replaceable: had Newton never lived, Leibniz; had Darwin never lived, Wallace. The discovery was going to be made, because the field had made it makeable.

From this Merton drew the deflation the theory is known for: the romantic image of the solitary genius wresting a truth from nowhere is, mostly, a myth of attribution. Science has a momentum of its own, and the names attached to its turning points are partly accidents of who reached the available step first. And there is a corollary Merton pressed, which this volume will turn on its head: if multiples are the norm, then it is the singletons — the discoveries made only once, by one person, and apparently by no one else — that are the genuine puzzle, the cases a theory of discovery must work to explain. Hold that inversion. It returns, exactly, in the fourth section.

Both phenomena answer to the same description — found independently, more than once — and the description hides a difference that is the whole of this volume. Merton's multiples occur on a shared frontier: a common body of prior work, a field whose momentum carries everyone to the same edge, and communication, however indirect — Darwin and Wallace corresponded; Newton and Leibniz had read the same predecessors and worked the same open problems. The diagonal's recurrences occur across sealed fields. Cantor in set theory, Post and Gödel in the theory of proof, Turing in the theory of computation shared no frontier, no common prior work, and no path of transmission whatever. Merton's mechanism — social ripeness, the field being ready — is not merely weak here. It is absent. Whatever the diagonal's recurrence is, it is not a Mertonian multiple.

The diagnostic can be made geometric, and once it is, the argument becomes hard to evade. Give every pair of discoveries two coordinates: their historical distance — how far apart their authors stood in time, in field, in any channel of contact — and their logical distance — how related the results are, from identical at one extreme to unrelated at the other. In the ordinary run of science these two are correlated: you tend to discover what your neighbours are discovering, so things found close together are usually logically close, and things found far apart are usually unrelated. That correlation is the diagonal line below. Merton's multiples sit near its bottom-left corner — close in both senses. The merely analogous results, the shared-engine fixed-point theorems of analysis, sit a little up and to the left — found within one connected community, genuinely similar but not the same. And then there is the diagonal's own family.

That is the argument in one picture. Transmission needs proximity; the diagonal's instances have none, yet they are not similar but the same — proven the same, later, by Lawvere. No social process puts a point in that corner. Only a structure beneath all three fields, forcing the same construction on each as it matures, can. The recurrence is not the field ripening. It is the terrain showing through.

Naming the two kinds is not pedantry; it changes what each is evidence of. A Mertonian multiple is evidence about a field — that its prior work, its instruments, its questions had reached a state in which the next step was available. It tells you about a social and cognitive condition. A structural multiple is evidence about an object — that there is a thing, sitting beneath the fields, real enough and forced enough that any sufficiently developed approach runs into it. It tells you about what exists independently of any field's state. The two readings face in opposite directions: one toward the community, one past it.

This is where the diagonal's recurrence inverts the orthodoxy. Merton deployed multiples against the decisive individual and for the determining society: the genius is replaceable, the field is sovereign. The diagonal's structural multiples make the contrary move — they point past the society as well. No state of set theory explains why the theory of proof and the theory of computation, which knew nothing of it or of each other, independently produced the same arrow. A shared social condition cannot be the cause, because there was no sharing. Only a structure common to all three — forcing the same construction on each, at the threshold each crossed alone — accounts for the convergence. Merton deflated the genius to elevate the field; the diagonal's recurrence deflates the field, in turn, to reveal the object. The lone genius and the sovereign field are both, here, beside the point. What the convergence is about is the thing being converged upon.

This bears, as carefully as anything in the collection, on whether mathematical objects are found or made — and it is worth stating the bearing precisely, because it is strong but not a proof. If the constructions of mathematics were free human inventions — conventions we lay down, fictions we agree to — it would be a standing mystery why sealed fields, inventing freely, should invent the same one. Independent convergence across non-communicating contexts is just what one expects if a real structure is being discovered, and just what is hard to explain if there is only stipulation. So the recurrence is a genuine constraint on the philosophy of mathematics: any account must explain why the diagonal keeps being found. Realism explains it at once — there is something there. But the anti-realist is not refuted, and honesty requires saying how he escapes: he can grant that the rules of any consistent self-referential formalism force the construction, and insist that this is a fact about the constraints — consistency together with self-reference — and not about a Platonic heaven. On that reading the recurrence shows the robustness of the construction under those constraints, which the realist and the structuralist can both accept while disagreeing about what grounds them. What no one can do is wave the convergence away as coincidence. It has a structural cause; that much is common ground, and it is not small.

One word has been carrying weight throughout — forced — and it should be paid for. The diagonal is not forced causally: no physical law compels it. It is not forced psychologically: no one was made to think it. It is forced in the sense of structural inevitability — any system that crosses the self-reference threshold contains the construction as a matter of logical necessity, whether or not a mind ever notices. The discovery, then, is the recognition of something already present, which is exactly why it can be made independently: there is a determinate thing there to be found, the same for everyone who looks from far enough in. And the epistemology has a rare closing move. The claim that the recurrence is best explained by a forced object is an abduction — an inference to the best explanation. But this abduction was, decades later, confirmed deductively: Lawvere proved the object. It is an unusual thing to be able to say — that an inference to the best explanation about the structure of mathematics was afterwards verified by proving the structure. The clue did not merely persuade. It came true.

Run the test over the whole record and a sharp asymmetry appears. Every limitative result is a structural multiple. The paradox that bears Russell's name was Zermelo's first; incompleteness was anticipated by Post and re-derived, in its second half, by von Neumann; undecidability was formalized three independent ways in a single year; the undefinability of truth was Gödel's silently before it was Tarski's in print. Forced, found independently, again and again. But the unification — the recognition that all of these are one construction — is not a multiple. It is a singleton: found essentially once, by Lawvere in 1969; overlooked for the better part of three decades; recovered and made legible by Yanofsky in 2003. The pieces were struck a dozen times. The unity was struck once. Why the difference?

The answer is exact, and it is the most important thing this volume has to say. The limitative results are forced by a local threshold — self-reference within a single field — and every field crosses that threshold on its own clock, independently of every other. That is precisely why they are multiples: each field manufactures its own encounter with the diagonal as it matures, needing nothing from outside. But the unity is invisible from inside any single field. To see that the three constructions are one, you need a vantage above all three — a language in which their shared form can even be written. And no single field's development forces that vantage into being. It is a meta-level achievement, not compelled by any object-level threshold; it had to be invented, and its invention was contingent: the concept of a category arrived only in 1945, and the specific apparatus the unity requires — the cartesian closed category — only across the 1960s, in Lawvere's own hands. The unity is a singleton because its enabling condition was a new mathematical language, and a language, unlike a threshold, is built once and need not be built again.

This is exactly the puzzle Merton left standing, now solved. Recall his corollary: if multiples are the norm, the singletons are what demand explanation. Here the explanation is precise. The limitative results are multiples because their cause — the self-reference threshold — is reproducible; every rich field builds it anew. The unification is a singleton because its cause — a new language — is not reproducible; it was a particular invention, and once made it did not need remaking. The general principle has fallen out of the diagonal's history almost by itself: a discovery comes as a multiple to the degree that its enabling conditions are forced and reproducible, and as a singleton to the degree that they are contingent and unrepeatable. Most of science lives between the poles. The diagonal is the rare case that occupies both at once — a forced multiple at the object level and a contingent singleton at the meta level — which is what makes it the cleanest specimen of the whole phenomenon there is.

To be discovered and to be registered — to enter the record under a name — are two different events, and a discovery must clear four gates to pass from the first to the second: it must be published, it must be understood, it must be attributed, and it must be taken up. Each gate can fail on its own. The result is that the canonical record is not the set of discoveries; it is a lossy projection of them, and the projection has identifiable failure modes — the same five the Annals catalogues: a finding can be an anticipation (found, written, unpublished), a simultaneity (found at once, and ceded), a withholding (found, and deliberately not printed), a misreading (found, and taken for something smaller), or an overlooking (found, published, and ignored). The record remembers; it just remembers selectively, and along these particular seams.

The eponymic shadow of this gap has a name of its own. In 1980 the statistician Stephen Stigler proposed Stigler's law of eponymy — that no scientific discovery is named after its original discoverer — and, with perfect self-awareness, attributed the law to Robert Merton, making “Stigler's law” an instance of itself. Hubble's law was Lemaître's two years earlier; the Pythagorean theorem was Babylonian; in mathematics the pattern is so routine that the physicist John Baez gave it a corollary — anything named after Professor X was first found by Professor Y, for some Y not equal to X. The diagonal's own history is a parade of the law. Select a result to see the name it carries set against the hand that reached it first.

Here is the inference the standard literature stops short of. For a structural multiple, the very forcing of the object makes the attribution maximally arbitrary — and not by bad luck but by necessity. Because the object is forced, it keeps being found; because it keeps being found, the question “who found it” collapses into “who happened to publish, in a venue that registered, at a moment the field could hear it” — which is pure contingency. The more necessary the object, the more accidental the name. So the two registers the collection keeps distinguishing — the logical, where the object is one and forced, and the historical, where the credit is scattered and contingent — are not merely different at the point of attribution. They are inversely related there: maximal structural necessity produces maximal nominal contingency. The diagonal is necessary. “Cantor's,” “Gödel's,” “Turing's” are accidents of publication — and the more surely the construction was going to be found, the more surely the names on it are arbitrary.

There is a last turn, and it is the one the collection's whole subject demands. The diagonal is the construction a system performs when it turns on itself. And the history of the diagonal turns on itself, repeatedly, at ascending levels. At the object level, the construction is forced, and so is found again and again — the structural multiple of the first four sections. At the meta level, the field's eventual recognition that it keeps finding one construction — the unification — is the field referring to itself, naming its own recurring move; the unification is a self-reference one storey up. And at the meta-meta level, the recognition repeats the fate of the thing recognized: Lawvere's unifying paper was itself overlooked and re-found — the discovery of the unity underwent, in its own life, the found, lost, found career of the unity it described. Even Stigler's law obeys this: a principle about misattribution, itself deliberately misattributed.

This is rigorous, not a flourish, and it is worth being exact about why. The pattern repeats at every level because self-reference is scale-invariant. The collection's one thesis is that wherever a domain becomes able to refer to itself, the diagonal appears in it. Now: “the history of mathematics reflecting on its own recurring constructions” is a domain that refers to itself — it is mathematics naming its own moves, one level up. So the diagonal must appear there too; and it does, three times over — as the forced recurrence of the construction, as the self-referential unification, as the self-instancing law of misattribution. The historical recurrence is not an analogy to the recursion theorem. It is the recursion theorem's phenomenon, occurring at the level of history, because history — once it reflects on itself — satisfies the construction's single premise. This is where the volume's title pays its debt. The construction that is recurrence — the fixed point, the positive face, the engine that returns a thing to itself — recurs, at every level, because recurrence is simply what self-reference does once it can.

And so there is no outside, including for this page. To study the recurrence of the diagonal is to perform, one storey higher again, the very recognition the diagonal forces — and to discover that the recognition has, of course, been reached before, in pieces, by others: Merton on the sociology of multiples, Stigler on the drift of names, Lawvere on the unity itself. This volume is one more structural multiple of an idea that was always going to be found, assembled late, in retrospect, out of parts that were lying in separate fields. It could hardly be otherwise. A construction that forces its own rediscovery forces, at the next level, the rediscovery of that — and the writing of it down, by whatever hand happens to be near, under whatever name the record will end up keeping.