You are on page 1of 29

Midwest Studies in Philosophy, XXXI (2007)

The Cosmic Ensemble1:


Reflections on the Nature–Mathematics
Symbiosis
JOSEPH ALMOG

M athematics spellbinds us (philosophers). We sing its praises—“it is pure


poetry,” says Kant—but at the same time, we treat it as a delphic stranger—
though mathematics asserts nothing but truths, we have no idea about what, viz.
what are the objects making the truths true. One problem here is as old as the hills,
and it simmers in the exchanges between Theaetetus and Plato. It is encapsulated
in the following impossibility thesis, which they both adhered to, the impossibility of
unity thesis:

(I) We cannot integrate in a single unitary realm (“nature”) the physical


entities of the cosmos and those of mathematics.

The claim does have an oracular feel to it. We all believe (I) is true, but at the same
time we cannot see—for mathematics pervades our ordinary lives—how (I) could
be true. And so, we are led to ask: where does the impossibility come from?
A trenchant diagnosis—one I believe isolates the ur-problem—is offered in
a dialog (across time) between Kurt Gödel and Charles Hermite. In closing his
1951 Gibbs lecture On Some Basic Theorems in the Foundations of Mathematics,
Kurt Gödel quotes from the eminent nineteenth-century French number theorist:

1. This is part I of a duo of papers. Part II will appear elsewhere. Before the reader plunges
in, I would urge him to have a look at the last, autobiographical note (note 35). This may explain
(if not attenuate) the hardships likely to accrue from the reading. For the philosopher, the work
may well seem overly technical, and for the mathematically inclined overly philosophical. I
dedicate the paper to the unforgettable teacher, the late Serge Lang.

© 2007 Copyright The Authors


Journal compilation © 2007 Blackwell Publishing, Inc.

344
The Cosmic Ensemble 345

There exists, unless I’m mistaken, an entire world consisting of the totality of
mathematical truths, which is accessible to us only through our intelligence
just as there exists the world of physical realities; each one is independent of
us, both of them divinely created.

So runs Gödel’s main text. In a footnote, he tells us that Hermite’s passage con-
tinues as follows:

. . . and appear different only because of the weakness of our mind; but for a
more powerful intelligence, they’re one and the same thing, whose synthesis
is partially revealed in that marvelous correspondence between abstract
mathematics on the one hand and astronomy and all branches of physics on
the other.

About this, Gödel comments,

So here, Hermite seems to turn towards Aristotelian realism. However, he


does so only figuratively, since Platonism remains the only conception under-
standable to the human mind.2

I am driven by Hermite’s integrative instinct (“they are one and the same thing”)
but with one amendment, on which perhaps everything turns. Hermite fears that in
apprehending truths, we—unlike Him—are subordinated to an epistemic dualism,
an inevitable split between the senses and the intellect.This epistemic dualism soon
breeds a metaphysical dualism of two realms of being (“worlds”). My amendment
is that in the pertinent respect, we are just like Him. Apprehension of the nature–
mathematics symbiosis (which I prefer to “synthesis” for a reason yet to emerge) is
not reserved to the sense-free unembodied pure intellects of angelic and divine
minds. Even our “weak minds” can and indeed must—for mathematics is some-
thing we live by—achieve cognition of the integrated ensemble.

SYNOPSIS OF PARTS I–II—HOW TO UNDO THE DUALISMS?

The project below—and I hope this does not sound overly naïve or
megalomaniac—is to consider manners of untying a complex knot we philosophers
have bound ourselves by. I say “we philosophers” because the knot binds not those
who practice mathematics but those who analyze-theorize the practice. This split
between the one who lives the phenomenon and the one who theorizes about it is
reminiscent of another such self-inflicted dualism philosophical thinking-and-
analysis has nurtured—the bond René Descartes called the mind–body union.

2. The quotations are from the last page (p. 323) of Gödel’s paper, “Some Basic Theorems in
the Foundations of Mathematics,” in The Collected Works of Kurt Gödel, vol. III, Oxford Univer-
sity Press, 1995. I should like it noted that where the English translator makes Hermite say “an
entire world consisting of the totality of mathematical truths,” the French uses the pregnant term
“ensemble,” as in “tout un monde qui est l’ensemble de verités mathematiques.”
346 Joseph Almog

Here too, once we are bound, we seek understanding by analysis of the knot.
And analysis leads us either to (i) sublimation—a segregated realm totally inde-
pendent of (immortal) minds or to (ii) reduction—all there is is purely material-
nature, with minds gone by the board. In tow, we shackle ourselves again with
an (I)-like impossibility result that precludes any third way of (mind–body)
integration-without-reduction.
Reflection on the mind–body interface suggests the same order of causes as
engendering the “crisis.” Each of us is first struck by the fundamentally different
ways of knowing, in the first person, (1) that I am thinking versus (2) that I have two
hands. Soon, we find ourselves segregating two correlative realms of being, one a
sublimated realm of immaterial (and destined for eternity and immortality) minds,
the other the mundane realm of material bodies. In turn, we proclaim that the only
genuine knowledge to be had—propositions we see clearly and distinctly—is of the
sublimated realm—for example, that I am thinking. It is now incumbent on us to
derive from such transparent premises, by stepwise deductions, in effect, a stepwise
reduction, our mundane knowledge, for example, that I have two hands.
It is against this background of separatist sublimation and a doctrine of
deductive–reductive foundationalism that Descartes offered Princess Elizabeth his
alternative model—symbiotic interdependence of (read with the hyphens) mind-
and-body, familiar to us all from humdrum existence. Descartes encapsulates it to
the princess in this one sentence:

Finally, it is by relying on life and ordinary conversations, and by abstaining


from meditating and studying things that exercise the imagination, that we
learn how to conceive the union of mind and body . . . the notion of the union
that each of us has inside himself without philosophizing: that he(she) is a
single person that has together a body and thought, which by nature are such
that the thought can move the body and feel the accidents that happen to it.3

I would like to pursue a similar defusing method vis-à-vis mathematical dualism. It


is thus that I distinguish below between (1) a pre-philosophical (“sans philoso-
pher”) mathematics as practiced in situ and (2) mathematics as studied in vitro by
reconstructive philosophers.
The in vitro studies are dominated, right from the outset, as in the Gödel–
Hermite exchange, by epistemological dualism. And twice over. First, we segregate
between our knowledge of the physical world (“astronomy”)—knowledge a
posteriori—and that of mathematics—knowledge purely a priori. Next, inside the
sublimated domain of the a priori, to secure all of mundane practiced mathematical
truths the mark of clear and distinct apriority, we launch an intra-mathematical

3. See The Philosophical Writings of René Descartes, edited by J. Cottingham et al., vol. III,
letter to Princess Elizabeth, June 28, 1643. In the French original: “Et enfin, c’est en usant seulement
de la vie et des conversations ordinaires, et en s’abstenant de méditer et d’étudier aux choses qui
exercent l’imagination, qu’on apprend à concevoir l’union de l’âme et du corps . . . la notion de
l’union que chacun éprouve toujours en soi-même sans philosopher; à savoir qu’il est une seule
personne, qui a ensemble un corps et une pensée, lesquels sont de telle nature que cette pensée peut
mouvoir le corps, et sentir les accidents qui lui arrivent.”
The Cosmic Ensemble 347

2- The Higher Realm

Logic (Set Theory) [A priori]

Reduction Reduction

Arithmetic Analysis Topology Geometry

1- The Lower Realm

Cosmic Nature [A posteriori]

Figure 1. Separatist Dualism.

cleansing program. We are to recall here Frege’s regulative ideal of “deduction


without gaps” or the Zermelo–Gödel deductions of all of practiced mathematics in
the universal characteristic of set theory. Such reconstructive programs are meant
to unravel the many interrelated subject matters of mathematics as all logic/set
theory in disguise. And therein—in this “deep meaning” exposing reductions—lies
the key to the transparent a priori knowledge of mathematics. However we actually
came to know this or that theorem of Diophantine geometry or algebraic topology,
however we discovered it, its ultimate justification, now with its real logical/set
theoretic meaning unraveled, lies in a deduction of that “deep meaning” from
self-evident a priori truths. We are led to Figure 1.
So much for the various segregations in the picture of the in vitro philosopher
of mathematics. In contrast, the mathematics practiced in situ, the mathematics we
live by, is a mathematics without borders. To echo Descartes’ above reciprocity
theme to the Princess, each of us feels, sans philosopher, that there is a single nature
with both mathematics and physical things hanging together so that each depends
on the other. We might call it, following Descartes, l’union vécu, the mathematics–
nature union lived and experienced.4
This lived union leads us to pursue a single-realm mathematics–nature inte-
gration. And thrice over. First, ur-structures of mathematics are grounded in—and
essentially so, viz. owing their existence to—the infrastructure of cosmic nature
(see, e.g., on Riemann surfaces in the next paragraph). Second, various branches of
mathematics (e.g., see below for the cases of number theory, complex analysis,
algebraic geometry) are strongly interdependent. Third and last, mathematical
logic (set theory) is not a universal characteristic into which all of mathematics is
to be reduced; logic and set theory are rather just more mathematics. To invert the
reductionist’s motif—mathematics is not a branch of set theory (logic); set theory
(logic) is a branch of mathematics. In all, Figure 2 emerges.
4. The French adjective “vécu” connotes both the English “lived” and “experienced.” I will
harp both on the living-it and the experiencing of it.
348 Joseph Almog

Logic

Set theory

Generation Generation

Algebraic Geometry ↔ Analysis ↔ Topology

Generation Generation Generation

Complex Plane (Riemann Surfaces)

Generation

Cosmic Nature (Spatio-Temporal Infrastructure)

Figure 2. Integrative Unification.

TWO SYMBIOTIC THESES

As I start drawing the symbiotic nature–mathematics picture, let me articulate


the two guiding ideas I am after. The first concerns the mathematics/nature union
proper, the second our apprehension thereof. First, at the level of existence and
truth (“metaphysics”), I argue that mathematics exists in nature—not in any
other realm—and the material cosmos has already in it all the structures of
mathematics. In the case of the specific ur-mathematical kind of structure I men-
tioned (Riemann surfaces), I submit the following symbiotic thesis: no material
cosmos without the complex plane and no complex plane without the material
cosmos.
I beg the professional philosophical reader not to rush and translate this
immediately into questions about ontological categories (“Is JA saying that the
complex numbers are objects? Or properties (modes) of material objects?
Or . . .”). Putting questions of objecthood/properties on the back burner, I want
us to linger on just this claim of essential interdependence: no material cosmos
without engendering mathematics and no mathematical structures without the
engendering infrastructure of nature. I will encapsulate this conception of math-
ematical notions and truths as engendered by cosmic structure as generative
mathematics.
The second and perhaps more critical tenet I would like to propose concerns
our cognition (thinking of) mathematical notions and, eventually, mathematical
epistemology (our knowledge and justification of mathematical truths). The thesis
I am after is this: Our apprehension of mathematical notions is inseparable from
cosmic contact. And twice over: (1) Our mathematical cognition and knowledge is
The Cosmic Ensemble 349

inextricably linked to our cognition and knowledge of the material cosmos and (2)
our cognition and knowledge of one mathematical domain (e.g., below, Diophan-
tine equations) is inextricably linked to our cognition and knowledge of other
mathematical domains, for example, complex analytic and group theoretic facts.
Without lived human experience—of (1) the infrastructure of the cosmos and (2)
other findings by fellow human mathematicians across human history, we could not
cognize and know the mathematics we do cognize and do know. I will characterize
this thesis as evolutionary mathematical cognition.
Our second thesis should not be hurriedly forced into the philosophically
standard exclusionary vocabulary of a priori/a posteriori knowledge (cognition).
The notion of a priori/a posteriori is multiply ambiguous. Let us remind ourselves
of, for example, three famous glosses thereof—Descartes’ “known by the light of
(our) nature,” “Kant’s “known independently of sense experience,” and Frege’s
“knowledge justified by reasoning through absolutely general propositions.” On
the first notion, mathematical cognition, as I will pursue it, may be taken as a priori,
for it does emanate from our very cognitive nature, prior to and independent of any
specific sense impressions we might have during our lifetime. But this is only to
stress that the cognition is strictly dependent on our having the nature we have,
itself a product of our living in nature and cognizing it so that cosmic contact with
nature’s infrastructure is built into our cognitive system.
Moving on to the second gloss, it may well be that our mathematical cogni-
tion is a priori in being quite independent of sensory experiences we are about to
have in our lifetime, although I will argue against this much as well—where we are
in history, what communications impinge on us, what notions have emerged by the
time we live and think, will very much condition what cognitions we can attain and
what knowledge we can come to have, let alone actually have.
Finally, relating to our third gloss, our mathematical knowledge is now dis-
tinctly not a priori for it involves a deeply specific subject matter. Key to the case
I will discuss are fine-tuned generalizations about elliptic curves and modular
forms, about finite abelian groups-generating curves, and about analytic functions
on the complex plane, none of which is resolvable to purely general logical or set
theoretic propositions.

I. THE COSMIC ENSEMBLE: INTRA-MATHEMATICAL SYMBIOSIS

The Roots of Dualism

Mathematical dualism (as early as Plato and Theaetetus and all the way to the
above Gödel–Hermite) is inextricably linked with mind–body dualism—it is our
conception of how we are cognitively structured and consequently how we appre-
hend truths (e.g., as opposed to Him) that feeds into our segregating apart math-
ematics and material (physical, cosmic) nature.
So let us linger, if only for a moment, on our own all-too-human dualism of
mind and body. We tend to teach our students the metaphysical dualism first, the
separation of two disjoint realms of existence—immaterial minds versus material
bodies. We then add on, as derivative, remarks on a consequent epistemological
350 Joseph Almog

dualism—we know the bodies—our own or others’—by the senses, but we know
our own minds by sheer intellectual introspection (and not by the senses).5
So goes our common teaching. Earlier reflection on this hard case convinced
me that the true chain of reasons in the human dualism case runs the other way
round. The primal dualism (in both senses of the adjective—temporal priority and
fundamentality) is an epistemic dualism—the way we know our own psychological
states strikes us as totally—categorically—different from the way we know our
body (and other bodies). To accommodate the split in modes of knowledge, we go
on to posit correspondingly disjoint realms of objects to match the segregated
epistemological channels—a realm of bodies, to be known by the senses, versus a
realm of (our own) minds, to be known by the sense-free intellect.6
In a similar vein, I see the chain of philosophical reasons running the other
way round with this new case of fundamental dualism—the mathematics versus
nature segregation. Again, we teach the dualism as starting from an “obvious”
bifurcation of ontological realms—the intra-cosmic versus the a-cosmic (or “pla-
tonic”). We then go on to append matching cognizing channels, the senses (with
which to know the cosmic) versus the intellect (with which to know the “other” or
“platonic” realm). But, I would like to propose, it is rather the segregation apart of
the cognitive channels that breeds the subsequent duality of realms of being.
The driving force—and I feel we are here driven rather than doing the
driving—behind our flight to a dualism of nature versus mathematics is a human-
bound syndrome of desire to transcend what Hermite describes as our “weak
condition.” In confronting mathematics, we often speak of striving for under-
standing but we are driven by a need—to control. And so, we sublimate the
desire into an epistemic regulative ideal. Be that as it may—a desire or an
ideal—we seek to transcend our (speaking with Hermite) “weak (human) con-
dition” and attain a more perfect form of knowledge, a replica of His form of
knowledge (control). This leads us to isolate the senses as the impeding shackles.
And it is thus that soon enough an epistemic dualism of senses versus intellect is
engendered.
Recent technical philosophy—by which I understand the turning of the
theory of justification into a technical subject with its own internal architecture—
transformed the quest for purely intellectual knowledge by introducing a technical
correlate notion—a priori knowledge. We transform the primal ur-desire for
perfect knowledge into a technical program—show that each theorem of

5. I study mind–body Cartesian Dualism in my monograph What Am I?, Oxford University


Press, 2002.
6. I said we will only linger for a moment on our human dualism but let me just note that on
my reading of Descartes’ articulation (to the princess) of his “symbiotic way out,” what he urges
is a rethinking of our common idea (ironically often called “Cartesian”) that we (1) cognize
ourselves and (2) know truths about ourselves, by the sheer intellect (recall the item to be known
is the full self, the human being Joseph Almog, not just this or that mental abstraction from JA).
In a similar vein, the path followed below questions whether sheer reliance on the intellect is the
way we (1) cognize the notions and (2) know (justify) the truths of mathematics. Another way of
setting our question is this: Does the notion of cosmos-interaction-free (“sheer”) intellect make
sense?
The Cosmic Ensemble 351

mathematical practice is justified a priori. This was Frege’s pronounced ideal and it
has become our regulative ideal.7

Why Foundationalism?

There is something of a puzzle in the automatic adoption of this regulative ideal.


The puzzle is this. We adopt the ideal in spite of our exposure as high school or
undergraduate (or higher yet) students to real practiced mathematics where justi-
fication (including proof-giving) does not actually proceed according to founda-
tionalist models. The experience each of us has with mathematical practice is
relegated to the level of the anecdotal—yes, indeed, that is how “they” (the

7. Where is this ideal coming from? I have no genuine acquaintance of ancient Greek texts
to speculate so far back (although I have my doubts about repeated philosophical references to
Euclid and, later, to Archimedes. Both seem to me rather experimental and anti-foundationalist in
their practice, as one would expect problem-solvers like them to be. About problem-solving Greek
mathematics, see also R. P. Langlands, The Practice of Mathematics, lectures given at the Institute
of Advanced Study [and available in electronic form] 2000).
In postmedieval times, it is often said that a paradigm of this blueprint for perfect knowledge
had been provided by the writings of the great rationalists of the seventeenth century, for example,
Descartes and Spinoza (I cannot speak of Leibniz’s work because I have not read it). On this
textbook rationalist model, it is said that knowledge of the self’s mental life, knowledge of God (if
such there be), and knowledge of morals, all domains where perfect knowledge had been deemed
possible, are to be based in the intellect only.And it is said that for the rationalists, mathematics had
continually served as the role model (e.g., both Descartes and Spinoza presented important ideas
of theirs in “geometric” [axiomatic-seeming] form).
I must confess that I have not found this segregative rationalism in the seventeenth-century
figures (although interested in our ratio they surely were). Ironically, at least on my own reading,
the role-model thinkers of this allegedly purely intellectual methodology, Descartes and Spinoza,
were, each in his way, deeply committed naturalists, rather of the kind of Hermite, averse to any
multiplication of realms. For both Descartes and Spinoza, nature is one. They were also both
“unromantic” about the source of our quest for perfect knowledge, tracing it back to our existential
situation as mortally endangered beings prone to doubt and desire (a beautiful concise “Freudian”
analysis is provided by Descartes late in Meditation III comparing the plight of man to God).
I linger on this epistemological naturalism—unified in its approach to nature (including
mathematics)—in a forthcoming book called Cogito? (Oxford University Press, 2007).
I am not a historian of philosophy, but on my own subjective sense, encounters with dualistic
epistemology multiply only in writings posterior to Kant (and as emphasized below and in Part II,
Kant himself, with his beguiling focus on the form of spatio-temporal intuition, is not a paradigm
of this dualism nor of foundationalism vis-à-vis mathematics). On my reading, the most shining
paradigm of the dualism is Frege, for example, in his (in my view, deeply unjustified) critique of
Mill early in the Grundlagen, his insistence on a priori justification (“deduction without gaps”) for
each theorem and his crafting of a metaphysics of multiple realms to serve the dualistic episte-
mology. Although Mill has been made the whipping boy of introduction classes in the philosophy
of mathematics, I view his (true, not Frege-caricatured) practice-based and nonrevisionist view of
evidence accumulation as vindicated by the type of considerations developed below.
After I completed the present study and presented it orally in Europe, it was pointed out to
me by the number theorist Michael Harris that similar observations about Mill are made by Ian
Hacking in his “What Some Philosophers Have Learned from Mathematics,” British Academy
Lectures 2000. Hacking’s paper pursues—in ways rather different from my own—other interesting
anti-foundationalist themes arising from mathematical practice.
352 Joseph Almog

mathematicians) actually did things on the blackboard or how I actually solved my


homework problem . . .
Autobiographical anecdotes about one’s practice provide no substitute for
theory. And so instead of recounting how I actually—in real history—came to
know (or be convinced of) this or that mathematical truth, we now focus on how I
or we in general, in principle we like to say, ought to justify this known truth.
We distinguish the (historical) context of discovery from the (logical) context of
justification.

A Case Study: Fermat’s Last Theorem

Let me use a ubiquitous (both mathematically and philosophically) case study to


bring home the distinction—the (Frey–Serre–Ribet) Wiles–Taylor 1994 proof of
Fermat’s last theorem (FLT).
Fermat’s theorem states that:

(FLT) for integers n > 2, for all x, y, z nonzero integers x n + y n ≠ zn

This is a simple statement that middle-school students have no problem under-


standing. Of course, it had a long and famous history, in between 1637 and 1994, and
that makes it all the more a vivid and interesting case study. But my reasons for
focusing on it are different.8

8. Since FLT is our case study, I will say a word about what it says and about its actual context
of discovery. We need for its statement a few informal glosses of the pertinent notions. And
informal they will be for we are philosophers (at least I am) and we address here a highly technical
set of results in number theory and so some level of generality (and genericity) is inevitable. For
light but enlightening introductions to an amateur like the present writer, see Cox, “Introduction
to Fermat’s Last Theorem,” American Mathematical Monthly 1 (1994): 3–14 or the very elegant C.
Goldstein, “Le theoreme de Fermat,” La Recherche 263 (1994): 268–75 (in French). I here state the
bare structure of what we need. I amplify on some technical notions in the appendix.
The main tool in the Wiles–Taylor 1994 proof is the theory of elliptic curves. An elliptic curve
E over the field of rationals Q can be characterized as a set of solutions in Q of an equation of
the form:

E: Y 2 = X 3 + aX 2 + bX + c
with a, b, c integers.
The overall context of discovery of the proof is this.
(i) The conjecture was known to be reducible to prime exponents and verified for p = 3
(Euler).
(ii) G. Frey argued in 1985 that if nonzero integers a, b, c greater or equal to 1 and n greater
or equal to 5 verify an + bn = cn, then the associated elliptic curve

Y 2 = X(x − a n )(X + bn )

is semistable (for the notions of semistable and conductor, see the appendix).
(iii) A few years later, Ribet showed that an elliptic curve associated with a hypothetical
solution of FLT cannot be modular (see the appendix for Serre–Ribet work).
The Cosmic Ensemble 353

Why Not the Homier 7 + 5 = 12?

It would have been more pleasant and pedagogically better to make the likes of the
rudimentary 5 + 7 = 12 our case study. And indeed for some of the features I’d like
to bring to the fore, 5 + 7 = 12 suffices. Not least among them are the following four
features—henceforth the rudimentary quartet—true of our living by and convers-
ing about 5 + 7 = 12:
(i) not by a logical/set theoretic object—the arithmetic fact in question is not
made to hold by relations among pure sets or (Frege’s) concept-extensions.
(ii) not by logical/set theoretic meaning—what we understand and communicate
to one another in using this arithmetic sentence in language is not a set
theoretic/logical meaning.
(iii) not by cosmos-free logical/set theoretic cognition—we need intra-
cosmic contact to cognize (have in our mentality) the notions of 5, 7, 12,
and +.9

(iv) In 1993–94, Wiles (with additions by Taylor) proved a restricted form of the 1955
Taniyama–Shimura–Weil hypothesis, or in short the semistable modularity theorem (SMT)
(for semistable curves):
Every semistable elliptic curve is modular.
This proves FLT by reductio (the full modularity theorem was to be settled a few years later
[see below]).
Finally, I mention here an earlier background fact. FLT is often stated by asking for solutions
in integers. But it is equivalent to the problem of finding rational values for A = x/z and B = y/z such
that An + Bn = 1 (A, B > 0). So stated it becomes a geometric problem. We consider A, B as the
coordinates of points in the plane. We consider now the first quadrant represented by the equation
An + Bn = 1 (see figure below). For n = 2, we get the unit circle; for n → •, the unit square; and for
2 < n < •, with curves in between, we get the shaded critical region.

η
n=∞
1

n=2

1 ε
This region is filled with rational points, points such that both their coordinates are rational.
FLT now becomes a geometric statement—the curves for n > 2 cut through the region without
hitting a single rational point. The Mordell conjecture–Faltings’ theorem already restricted the
number of points hit by each curve to a mere finite one (1983). The Wiles–Taylor proof reduced it
to 0.
9. In my own “doctrines,” we need cosmic contact (with material objects) for apprehending
even the purely logical relation of “=.” I bracket away at the moment how we apprehend logical/set
theoretic notions, pretending they are given to us for free, prior to and independent of cosmic
contact.
354 Joseph Almog

(iv) not by logical/set theoretic epistemology—the justification for (evidence for,


conviction in) the statement’s truth does not come from a proof in a perti-
nent (logical/set theoretic) axiomatic system.10
Thus, in certain nontrivial ways, 5 + 7 = 12 would have sufficed to make some of our
points. Kant must have sensed something of the kind and used this simple example
to illustrate various types of irreducibility of arithmetic. Of course, in his day, set
theory was not the characteristica universalis. But as Gödel intimates (correctly in
my view), if Leibniz ever got word of it, he would have immediately adopted it as
the reductionist ur-grammar of any possible human thought and Kant would have
to tell him: no, no, 5 + 7 = 12 is no piece of set theory.
But alas, in the end, 5 + 7 = 12 does not suffice to make our case study. For
there is a cluster of fundamental features of number theoretic practice, I will call it
the sevenfold spectrum, of which 5 + 7 = 12 is not illustrative. This segregation of
characteristic marks—separating the rudimentary from the advanced—should in
itself indicate that we are about to take issue with the common foundationalist
classifications. On the foundationalist view, “practicalities notwithstanding” as we
like to say, “essentially” the two cases are on a par.
On this classical philosophical view, we take God’s point of view and arm
ourselves with logical omniscience, so that we can factor out the difference
between the practically simple reasoning to 5 + 7 = 12, as opposed to the complex
reasoning leading to FLT. And we not only arm ourselves with the ability to
follow all logical consequences of the axioms (notions, definitions, etc.) that we
actually do, as luck and history would have it, have in mind; we assume, stipulate
really, that each individual mathematician, all by him/herself, just like Him, can,
at any point in time and space, have the full repertoire of axioms, notions,
definitions, etc. from which the consequences are to be drawn; for example, it
does not matter whether our mathematician is Diophantus, Fermat (1637) or
Wiles (1994). What is viewed as critical, on this (as we like to call it) rational

10. There exist of course such proofs of the translates (“logical [set theoretic] symboliza-
tions”) of 5 + 7 = 12 both in the combinatorial sense of proof, in, for example, first order formal
axiomatization of arithmetic (e.g., first order Peano Arithmetic [PA]), and in the model theoretic
sense of full logical consequence in the categorical second order axiomatization. My point is that
the evidence, conviction, and ultimate justification for (1) the truth of 5 + 7 = 12 and (2) our belief
in it do not rest on such proofs or on its being a second-order model theoretic consequence of the
axioms.
Let me point out that from now on, when I shall speak of logical proof, I will assume as the
underlying system the full second-order logic and the pertinent (respectively, to arithmetic and set
theory) axiomatic systems of second-order PA and second-order ZF(C). I include both strictly
deductive proofs in those systems as well as semantic consequence relations (an argument in that
language that preserves truth in all models). No doubt, there are important differences between
the notion of combinatorial proof in a first-order system and the notion of combinatorial proof in
a second-order system. Stronger yet, both differ from the (nonrecursively enumerable) notion of
semantic (model theory) notion of consequence (either in the standard model of first-order PA or
in full second-order PA). Still for all the differences, I urge below that (1) the fact that 5 + 7 = 12;
(2) the meaning of this sentence when I use it; (3) the notions I cognize when I apprehend it; and
(4) my justification for my belief in it and its truth, are all given neither by such combinatorial
proofs nor by a semantic consequence-drawing from the second-order Peano-Dedekind axioms.
This much I claim below directly and primarily for FLT. But I also claim it eventually for 5 + 7 = 12.
The Cosmic Ensemble 355

reconstruction, is the following common fact: both 5 + 7 = 12 and FLT are (proof
theoretically and model theoretically) consequences of self-evident axioms. On
this common foundationalist view, we assimilate the complex case of FLT to the
simple case of 5 + 7 = 12.11

Why the More Involved Case of FLT?

I follow quite a different methodology by starting with the complex case of FLT. If
any assimilation is eventually to take place (and it will later, in part II), it is rather
the other way around—even 7 + 5 = 12 presupposes more than meets the eye. But
at the moment, the main thing to take in is that when we start with FLT as a
paradigm case of number theoretic practice, the profile that emerges is categori-
cally different from that of 5 + 7 = 12.
For a contrast with the fully transparent meaning of 7 + 5 = 12, let me use an
example deeply embedded in the natural world, an example after which I would
like to model the case of FLT. In recent philosophical theories of meaning (“seman-
tics”), philosophers—I am thinking chiefly of the work of Hilary Putnam—have
pondered examples like the word “water.”12
One apparent aspect of our use of “water” is the availability of a certain
kind of “transparent and qualitative information,” at least to those who ever
interacted with water, regarding water’s appearance, for example, “tastes so and
so, looks thus and such, runs in rivers, etc.” We may call this the apparent meaning
of “water.” The apparent meaning is supposed to be available to every lay user
of the word.13
However, as pointed out by Putnam, natural science goes beyond this
apparent meaning. Modern chemistry found out that what water is to what the
liquid itself is: is hydrogen hydroxide (H2O). This was not known to Aristotle
(our analog of Diophantus). Aristotle did use the same word as we do (or a

11. As should be obvious by now, my claim that 5 + 7 = 12 does not suffice to make our
case—does not have the right profile—does not turn on its being part of a fragment of complete
arithmetic. From a certain deeply epistemological perspective centered on proof—whether com-
binatorial (as in Hilbert) or logical inhalt-involving as in Frege—the threat might seem to come
only with certain universal sentences over recursive relations, for there we get incompleteness
(recall that Gödel’s sentences can be coded as a certain kind of Diophantine problems). But to
reiterate: such unprovable (Gödel) sentences are not what I want to contrast with 7 + 5 = 12. It is
the very PA-provable FLT (and its simple Diophantine mates, see below) we will contrast with
7 + 5 = 12.
12. See Hilary Putnam, “The Meaning of Meaning,” in his Mind, Language and Reality:
Philosophical Papers, vol. 2 (Cambridge: Cambridge University Press, 1975), 215–71; and especially
Putnam, “Is Semantics Possible?,” ibid., 139–52. These insights were amplified and deepened later
to points about the environment-dependence structure of our cognition by Tyler Burge; see, for a
clear statement, Burge, “Other Bodies,” in Thought and Object, ed. Andrew Woodfield (New York:
Oxford University Press, 1982), 97–120.
13. The terminology of apparent meaning is intended to connote both uses of the adjectival
modifier—a meaning that involves surface information and a type of information that is only
apparently the real meaning (with this last being not so apparent).
356 Joseph Almog

Greek cognate), and I shall assume that he associated with it the same apparent
meaning. But the chemical fact was unavailable to him and may well still not be
known to the majority of the present users of the word, for example, when they
ask for a glass of water.
Reflection on this example suggests that there are two appearance-
transcendent layers of meaning to mark out. First, there is what-we-mean (with this
verb now taken in the active-transitive form, what stuff we target and pick out), viz.
we mean with “Water” the stuff water, H2O. I would like to call this the embodied
meaning of “water.” So we here transcend the internal, “in the head” of the lay
speaker, apparent and transparent information. We break out of our mental cages
and embed ourselves in the embodied world, in particular, in the liquid stuff out
there. And now, after this embedding, we are ready for the second step. Having
made H2O the embodied meaning of “water,” there are now structural character-
istics of this embodied meaning awaiting to be revealed. Hydrogen and oxygen
have natural stable and radioactive isotopes and water molecules of masses
roughly 18(H216O) to 22 (D218O) are expected to form. Pure water, H2O, has a
unique molecular structure. The O-H bond-lengths are 0.096 and the H-O-H angle
is 104.5. This distinct geometry calls for a complex explanation. This second layer of
structurally revelatory information about the embodied meaning I will call the fine
structure embodied meaning.
In all, we separate: (1) the apparent meaning—what’s in every lay user’s
head—(2) the embodied meaning, the basic stuff meant by the word—H2O—and
finally, (3) the fine structure embodied meaning unraveling the fine structure of
hydrogen hydroxide. Of course, many philosophers, aware of Putnam’s observa-
tions, may well want to call the apparent level the meaning of the word “water” and
contrast this with the nature or essence of the worldly stuff proper. They would
regard what I call the embodied meaning, let alone the fine structure meaning, as
not part of what we mean but part of what the stuff is. I should like it noted that this
separation of semantics from metaphysics, what-we-mean from what-the-stuff-is,
rests on turning a deaf ear to an ambiguity in the very phrase “what-we-mean,” for
at least on one reading the phrase “what-we-mean” targets the whatness of the
thing meant. After all, what we mean by “water” is . . . water.
In any event, this two-tiered separation between what we mean and what the
stuff itself is (the essence of water proper) is a version of the dualism I have been
lamenting—an us versus it duality. As the philosopher Van Quine put it (so well),
“Meaning is what essence becomes when it is divorced from the object of reference
and wedded to the word.”
Putnam himself does not so divorce meaning and essence. He famously quips
“meanings ain’t in the head” and relegates the apparent-transparent qualitative
information—what is in the lay head—to the status of mere “stereotypic informa-
tion.” And he says just the right thing about what we do mean by “water”—we
mean water (the stuff), the hydrogen hydroxide. But let me not pin too much here
on Putnam and so, in what follows, let me not hang on to this large tree. I will just
say for myself that what water means (or in the active transitive verb form, what we
mean in using the word “water”) is the embodied meaning, further unraveled at the
level of fine structure embodied meaning.
The Cosmic Ensemble 357

Two Models: Apparent versus Embodied Meaning

So now we have two purported models, “7 + 5 = 12” and (say) “water is wet” (or
better yet: “Water is a compound”). In the former, the meaning we see is the
meaning we get, the apparent meaning is the full meaning. With the latter, there is
more to the meaning than meets the eye—the apparent and the embodied mean-
ings differ. We may now try to classify new cases using the two models as our
paradigms.
In a moment, I will try to so classify FLT. But since I have harped on the
theme of dualism, let me mention an analog hard case lying at the very core of
dualism, a case that like FLT could (and did) go either way, be made to fit either of
our paradigms. Consider then my use of the first person pronoun “I.” What do I
thereby mean?
We could view me as meaning by “I” (what philosophers call) the transpar-
ently available inner self, what many have called the “Cartesian Ego,” an immate-
rial item seemingly so prominent in our first-person reflections (“meditations”)
cogito (“je pense”) and in sum (“j’existe”). For example, the leading philosopher of
language of our age, Saul Kripke, has often looked for such a “purely qualitative”
(his own phrase), epistemologically transparent meaning for the word “I.” He has
emphasized this purely qualitative meaning with his famous case of my use of “I am
in pain” to bring home that what I mean is this inner item (and the qualitative
feeling of hurt), all allegedly available prior to and independently of any allusion to
my mind’s (self, ego) embodied connections, the physical body (in particular, brain)
of Joseph Almog. Sure enough, JA has such a body. But this is a subsequent point
about what JA is (or his essence if you will), not about what I mean by “I.” We
crown here yet again a dualism between a transparent meaning and a hidden
essence.14
In contrast to this dualism, we may insist on not divorcing meaning and
essence. We may (I surely would) answer my original query, “What do I mean when
using ‘I’?” with: I mean this man—Joseph Almog, the fully embodied human being.
And of course, once we have that item of the natural order, the man, as the
embodied meaning of my use of “I,” we have, in turn, waiting to unfold a layer of
a fine structure embodied meaning. As we unravel what it is to be this man—how
it-JA—originated from a particular sperm and egg and wasn’t brought by a stork or
angels—and also unravel how this kind of thing—mankind—has emerged from
higher primates, we unravel the fine structure of the embodied meaning—what I
mean—the human being I mean—with my use of “I.”
Very well then: We can treat “I” dualistically and we can treat it unitarily. Let
us go back to our number theoretic sentences. I see us as facing a similar choice

14. This kind of account is often called “Cartesian,” but in my view it was not the one put
forward by that man, René Descartes, who championed the orthogonal picture about to be
introduced. (See my What Am I?). In any event, the adjective “Cartesian” is so embedded in
philosophical use that it seems futile to try to undo it. Saul Kripke is a paradigm of such a
“Cartesian”: See his recent “The First Person,” address delivered at the CUNY Graduate Center
Conference, Saul Kripke: Philosophy, Language and Logic, New York, 25 January 2006, available
at web.go.cuny.edu/philosophy/events/kripke.conference.html.
358 Joseph Almog

with mathematical—and our specific simple example, Diophantine—sentences. We


can think of their meaning on the apparent-meaning model of “7 + 5 = 12” or on
the model of the embedded-embodied meaning of “water” and “I.” I would like to
develop below the latter picture.

Diophantine Sentences—the Sevenfold Spectrum

The embedded account of Diophantine sentences involves a cluster of character-


istic marks that I will call the sevenfold spectrum. The first batch of marks—four of
them—concern the witnessing by FLT-like cases of the phenomenon of intra-
mathematical integration; the second batch of marks reflects the phenomenon of
nature–mathematics integration. I will state the seven headers concisely, then go on
to develop each in detail.

A. Intra-Mathematical Integration

(Intra-1) apparent meaning versus embodied meaning

There is more—from inside the theory of elliptic curves—to the embodied meaning
of FLT than meets the eye—viz. than the visible meaning of the Diophantine
statement.15

(Intra-2) proof as fine structure-revelatory

There is more to the full meaning of FLT than even the just-mentioned algebraic
meaning; the proof of FLT—from the modularity theorem—is revelatory of hidden
structural layers, features of FLT’s algebraic meaning—what, in the complex plane,
is the analytic generative basis of the embodied meaning—elliptic curve-
involving—meaning. I will call this the full fine-structure embodied meaning.16

(Intra-3) no algebra (arithmetic, logic) versus geometry dualism

Frege and Kant, who disagreed much about geometry and logic, did nonetheless
agree on there being a deep difference between the types of meaning of arithmetic-
algebraic statements and those of geometry. If one’s paradigm of the former is the
likes of 5 + 7 = 12 and the paradigm of the geometric is, for example, the likes of the
Pythagorean theorem, it is natural to let a dualism emerge—this one intra-
mathematical—between the meaning (and basis of knowledge) of algebraic state-
ments and those of (Euclidean) geometry. But if our paradigm is the slightly more
complex Diophantine equations (e.g., like FLT or the problem of Pythagorean

15. As mentioned in the last, autobiographical note of this paper (note 35), after completing
the present work it was pointed out to me that the theme of looking for such a deeper complex
meaning appears (in quite a different form) both in A. Weil’s “Two Lectures on Number Theory”
and especially McMullen’s “From the Dynamics of Surfaces to Rational Points on Curves.” See the
last note of this paper (note 35).
16. See the discussion that follows below on the generation of elliptic curves from notions of
the complex plane (as reflected by the modularity theorem).
The Cosmic Ensemble 359

triples [“congruent numbers”] mentioned below), the algebra/geometry dualism


disappears, as indeed witnessed by the emergence of a domain called Diophantine
geometry. Integer and rational solutions draw upon rings and fields from which to
assemble the solutions but at the same time the system of polynomial equations
describes an algebraic variety, a geometric object. Linear and quadratic equations
in two variables express curves of genus zero and our key paradigm case—on
which much below—elliptic curves, express curves of genus 1 (emerging, in tow, in
the complex plane, as Riemann surfaces).17

(Intra-4) a web of experimental conjectures unravels the fine-structure of


the embodied meaning

As in the natural sciences, for example our “water” case above, to lay out the full
meaning of a Diophantine equation like FLT, we bring its embodied meaning
under deep structure-unraveling conjectures regarding this type of object—elliptic
curve on the rationals.18

B. Trans-Mathematical Integration

(Trans-1) existence—cosmically generated meanings

The sheer existence of the full fine-structure embodied meaning of FLT (involving
as it does, for example, Riemann surfaces) rests on the existence of the complex
plane, whose existence in turn rests on the existence of the infrastructure of the
physical universe.

(Trans-2) cognition—evolutionary emergence of notions

The notions required by the full fine-structure embodied meaning of FLT could not
have been cognized by us without (1) contact with a community of human math-
ematicians across history and (2) a natural historical order behind their emergence,
for example, Fermat could not have formulated the modularity theorem.

(Trans-3) proof-evolutionary and regressive emergence of proofs

Not only is the emergence of the notions and their cognition history-bound; even
when they have been fully cognized, the laying down of the conjectures needed for
proving FLT is evolutionary and regressive. That is, the theorem to be proved
precedes, both in time and conceptually, the hypotheses from which it is eventually
established. Conceptually, the process is regressive—the theorem precedes and
sends us looking for the conjecture (its eventual ground), in the sense that we often
search deep generalizations about the key notions (e.g., elliptic curves) from which
to derive the target theorem.
17. See our discussion below of congruent numbers for another anti algebra/geometry
dualism case study.
18. Examples discussed below concern the modularity conjecture, the Mordell conjecture
(both now theorems) and the Swinnerton-Dyer/Birch conjecture.
360 Joseph Almog

Unwinding the Sevenfold Spectrum of FLT

First, the embodied meaning of FLT is not the meaning that meets the eye. In
unraveling a “deeper” meaning I do not mean a reductive translation to some
favorite foundationalist universal characteristic. I target rather what mathematical
practice reveals the sentence to mean in humdrum mathematics. The sentence
expresses a claim in the theory of elliptic curves. This is surprising because on the
face of it (the visible meaning) xn + yn ! zn is not a cubic. But as mentioned above,
Frey has associated with a nontrivial solution of ap + bp = cp, with p equal to or
greater than 5, an elliptic curve given by the cubic Y2 = x(x - ap)(x + bp).
Let me linger on this point that I regard as the ground-zero fact of the
intra-mathematical integrative picture that is drawn below—what the sentence,
when put into “mathematical action,” really means; later and further discoveries
due to proofs of Diophantine equations like FLT, about what I called its fully
fine-structure embodied meaning (expounding deep structural connections to ana-
lytic and group theoretic features) emanate from this one ground fact—what the
sentence’s embodied meaning is.
I have spoken of Fermat and his last theorem in which we consider any old
exponent n. For this Fermat thought he had a proof. It is often speculated that he
must have believed that his correct proof, in the particular cases of n = 3 and n = 4,
somehow generalizes. About this, he was wrong. But already in the cases he did
control—x4 + y4 = z2 (which gave him the impossibility of FLT solutions for n = 4),
we can see the basic idea that the Diophantine equation’s embodied meaning calls
upon elliptic curves and their structural features. So we can see that the Diophan-
tine equation has an elliptic curve meaning quite independent of subsequent
fine-structure revelations (about which much more below) coming from eventual
proofs. These last, to analogize with our “water” paradigm above, reveal very fine
structure and thus the generative basis of the phenomenon (what engenders water
[elliptic curves on Q]). But what I focus on at the moment is the primary embodied
meaning of FLT and the fact, again in analogy with “water,” that it involves more
than the apparent meaning—calls upon elliptic curves on Q.19
Still lingering on this ground point—the embodied meaning invokes elliptic
functions—let us recall another deceptively simple case, where what meets the
eye—the apparent meaning—is not what does the mathematical action.

19. The points that follow, both historically about the man Fermat and, as it were with
hindsight, on the elliptic curve algebraic-embodied meaning of the case he did control, are inspired
by A. Weil’s discussion “Two Lectures about Number Theory—Past and Present” and his discus-
sion of Fermat’s method of infinite descent as part and parcel of assigning elliptic curves as the
meanings of the equations. I have in mind pp. 95–96, where he discusses Euler’s articulation of the
case of n = 4 and n = 3 using de facto elliptic curve articulations. Then, Weil explains why we ought
to read (with hindsight) Fermat’s infinite descent as already processing the elliptic curve meaning.
Indeed, I believe the point generalizes and even for the “easy” solvable cases of x2 + y2 = z2, we
should assign the elliptic curve embodied meaning. This is what Pythagoras talked about
(meant)—even if it took a while to see that this is what he was meaning (just as H2O is what he was
meaning in asking for a glass of water 2,500 years ago).
The Cosmic Ensemble 361

A famous problem running back to the ancient Greeks concerns the areas of
right triangles.20 It is standardly called the problem of congruent numbers. A
rational (non zero) number is congruent if it is the area of a right angle with three
rational sides. The question whether a number is congruent can be correlated with
the following problem about elliptic curves. A rational number R is the area of a
right triangle with rational sides and hypotenuse h iff (h/2)2 with R added or
subtracted are both rational squares. Thus R is a congruent number iff the follow-
ing two equations

(1) u 2 + Rv 2 = w 2
(2) u 2 − Rv 2 = z 2

have a solution in integers (u,v,w,z) with v non-zero. These two hypersurfaces in the
projective plane P3 intersect in a smooth quartic in P3 which contains the point
(1,0,1,1). The intersection is an elliptic curve over Q. The projection from (1,0,1,1)
to the plane z = 0 gives an isomorphism with a cubic plane whose forms are:

ER : y 2 = x 2 − R 2 x or also Ry 2 = x 3 − x

The points on the space curve with v = 0 correspond to the points where y = 0 and
the point at infinity on ER. When these points are reduced modulo p (for finite
p-fields offering values for solutions), it turns out the rational solution points on
ER(Q) are precisely those of finite order. As will turn out in our discussion in a
moment, a test whether ER(Q) is finite involves use of one of the deepest standing
conjectures about elliptic curves, the Swinnerton-Dyer and Birch conjecture. And
so, on my reading, when Pythagoras wondered about the seemingly homey con-
gruent number problem, his assertion concerned the deepest structural features of
elliptic curves.21

II. APPARENT VERSUS EMBODIED MEANING

The apparent meaning of “water”—its appearance—results from its embodied


meaning, the chemical structure. It is because the molecular structure is what it is
that there results a certain appearance of this liquid. It is surely not the other way
around—because something looks thusly, it has the H2O fabric. That it so looks
helps us approach the phenomenon—we track and identify the phenomenon,
communicate to others its presence, etc. by using the apparent meaning. But that
we can so identify water (using the appearance) is due to the determination
running the other way: It is the H2O structure that determines the appearance.
In like manner, it is the elliptic curve features that determine that for n
greater than 2, xn + yn ! zn. We use the Diophantine notation to track and identify
the elliptic curve. In the same vein the order of determination runs as follows—it

20. I follow here S. Lang’s Diophantine Geometry, 135–37. Thanks to slow and patient dis-
cussions by the author.
21. On the Swinnerton-Dyer and Birch Conjecture, see below.
362 Joseph Almog

is because the Frey elliptic curve could not have a certain feature (be modular and
semistable at once) that the visible Diophantine claim cannot be true. The use of
“because” here is determinative: The impossibility of the Frey curve makes a
solution for the equation impossible.
We could say of course that the association of the curve with the Diophantine
sentence is accidental, merely representational or instrumental, in the way we can
associate with a given person a social security number. It would be silly to suggest
that this social security number makes the person what he is.
And that is exactly what is not the case with the elliptic curve “association”
with FLT. We don’t just so “represent” the Diophantine claim. Here the elliptic
curve structure makes the Diophantine claim be what it is, in the way the molecular
structure of hydrogen/oxygen combination makes water be what it is. Our word
“water” primarily tracks the underlying structure. The diophantine notation pri-
marily tracks the elliptic curve structure.22

Two Corollaries of Embodied Meanings

In the case of “water” the primality of the embodied meaning—we mean by


“water” H2O—leads to two fundamental characteristics. The first concerns
the metaphysics of the phenomenon, the second our epistemology, the kind of
knowledge we can have of the phenomenon. The metaphysical corollary induces
the epistemological result. And so I start with it.
The metaphysical corollary concerns the constitutional unity underlying all
natural phenomena. I spoke of “water” (and water) but of course I could have
spoken of “salt” and “gold”; in the same vein, we could have considered “whale”
and “elm”; and, in turn, “river” and “planet” and “black hole”; and indeed, I have
spoken of that most surprising of all natural elements, mankind, the embodied
meaning of all our uses of “I.” What is at first blush so striking, what is so apparent,
is the staggering diversity of natural phenomena; what later blushes reveal is that
the diversity is underlined by symbiotic codependencies, not least among them a
deeply unitarian fabric that goes into the composition of all the aforementioned
phenomena. A black hole is many tons of underlying atoms, so is a water-pond and,
in the end (and that’s why there is necessarily an end), I just am, as Tom Nagel puts
it coldly, a hundred and sixty-five pounds of atoms. Let me call this lesson—from
reflection on the embedded-embodied meanings of “water” and “black hole” and
“I”—the (constitutional) unity behind the diversity.
The epistemological thesis emanates from this non-apparent but nonetheless
primal constitutional unity. It concerns our inevitably trial-and-error, conjectural
and evolutionary apprehension of what phenomena like water and salt, whales and
trees, black holes and men are. The hidden natures (and in turn, the induced
nontransparency of the nature-imbued meanings) make for a slow growth of
knowledge, generation after generation, building on cumulative trials and errors of

22. Like remarks apply to the Diophantine notation and the elliptic curve structure in the
case of the Congruent number problem. See below.
The Cosmic Ensemble 363

previous ones. Our current apprehension of what water is was not available to
Parmenides or Aristotle or Descartes or Newton. Let us call this second thesis the
evolutionary character of natural cognition.23

Diophantine Statements: the Unity behind the Diversity


With the water model in mind, let us have a closer look at statements like FLT and
the congruent number (CN) (Pythagorean) problem and what they teach us about
the meanings of Diophantine statements.24
First, let us look at the metaphysical front and the problem of constitutional
unity/diversity. Philosophy often ponders with some angst the alleged rift between
algebra and geometry. It is even said to students that the Greeks primed the
geometric and defined the algebraic in terms of it. It is then suggested Descartes
inverted everything upside down and primed the algebraic, reducing things geo-
metrical to the language of polynomial equations. For myself, I think a simple
reflection on cases like FLT and CN exposes the rift as illusory. There never was
any rift—not in ancient Greece, not in Descartes, and not anywhere else. Geometry
and algebra are rather like two sides of the same piece of paper—generated
simultaneously and interwoven in one another.
We may put the symbiosis by speaking, but only for a moment, in the
language of Kronecker about God the creator, (“God created the integers, all the
rest is the work of man”): even for God, there was to be no algebra without
geometry nor geometry without algebra. Of course, as mentioned, 7 + 5 = 12 may
not quite bring this home. But FLT does.25
As already mentioned above, when FLT was introduced into our discussion
(see note 8 above), critical to Wiles’s proof of FLT was the (semistable) modularity
theorem: (SMT) Every semistable elliptic curve over Q is modular.26
Reflection on the many forms of this theorem brings out just how interde-
pendent are the following notions—the algebraic idea of elliptic curve E, the
analytic idea of L-function associated with (the growth of solutions mod p of) E,

23. By using the adjective “evolutionary” I target two separate features: the fact that the very
notions to be cognized (e.g., hydrogen atom) are becoming available to cognizing agents in a slow
cumulative historical process; furthermore, the fact that our knowledge (of truths about them) is
cumulative. We may articulate the two features by recalling the two senses of “know” (in French,
German, etc.), for example, “kennen” versus “wissen.” Both our kennen of the notions, our
acquaintance with them, and our wissen of the truths, are cumulative.
24. I here follow two illustrative cases discussed by S. Lang, Survey of Diophantine Geometry,
130–39. As mentioned, I return to this theme of intra-mathematical integration in Part II when I
discuss a different type of example resting on interconnections between Riemann surfaces, Dessins
d’enfant and algebraic curves (but see below, for a bit of ankle, the remark about Belyi’s theorem).
25. One does not feel comfortable speaking for God but I should add that if what I am about
to develop below makes sense, He was further constrained (than making Geometry-Algebra in
one fell swoop)—He had to create cosmic space-time to engender geometry-algebra.
26. For the restriction to semistable curves (Frey’s curve is of the kind), see the appendix.
(SMT) had to be combined with a result of Serre–Ribet to yield FLT. See the appendix. The many
forms of the modularity theorem (by which I now mean the full modularity theorem proved in full
a few years after Wiles) are reviewed in detail (and their interconnections laid out) in Fred
Diamond and Jerry Shurman, A First Course in Modular Forms (Springer, 2004).
364 Joseph Almog

the hyperbolic geometry notion of modular curve, and the group theoretic notion
of Galois representation associated with E (as with the modular curve). We have
anything but an algebra–geometry rift here; we are rather exposed to the many
lives of E. An elliptic curve leads a double life as both algebraic and geometric.
I explain.
The key conceptual (or “philosophical”) fact in reading the modularity
theorem is not the bidirectional correlation of elliptic curves over Q and modular
forms. This is the mathematical ground fact; without it we wouldn’t have a further
conceptual question I am about to raise. The conceptual query is this: In between
the elliptic curve and modular form, what, if anything, engenders what? Why are we
so lucky to have such a deep correlation? Is it a cosmic accident or is there a
mechanism explaining the correlation as a generative relation?27

From Riemann Surfaces to Elliptic Curves

On my understanding of the theorem, all elliptic curves with rational invariants are
generated from modular curves by appropriate holomorphic maps. The generation
reveals both the modular curve and the engendered elliptic curve to be forms of
compact Riemann surfaces. So, to linger for a moment at this elementary under-
graduate “I am puzzled” level, if one wonders (as I at least often did) why a curve
and a surface are so intimately correlated (let alone said by some to be the same),
my speculative diagnostic is this: the ground notion is that of a (compact) Riemann
surface. We generate from it the elliptic curve by preserving in the generation
process—by the use of certain types of maps—a bunch of critical properties.
In so speaking of the generative order, we must of course keep our discovery-
order and the generative order apart. We may well discover the chemical (and
fine-structure chemical) structure of H2O late in the game, having entered the
investigation by way of the apparent meaning of “water.” But in the generative order
it is the fine structure chemistry that engenders the H2O molecules, which engender,
in turn, the water-appearance. I would like to pursue a similar tack of explanation in
our case.
There are various manners of explaining this generation process, different
languages with which to close the maps on the Riemann surfaces under group
actions that preserve the key structure—for example, the just mentioned detailed
analysis of modularity (see note 27). A First Course in Modular Forms reviews at
least five different languages-methods of describing the structure of such genera-
tive maps. My own preferred explanation—the one that I grasp best—rests on a
sixth language, the language of dessin d’enfant, in my view the most vivid
(“topological”—what Grothendieck describes as the one employing the most

27. Instead of “correlation,” we may speak technically of “parametrization” or “uniformiza-


tion,” but the conceptual puzzle remains—what, if anything, is the “hidden mechanism” that
guarantees the successful parametrization of the curves? After all, the one-one correlation works
in both directions and one may well say: Why not generate the compact Riemann surface from the
curve? I argue below why the surfaces are the basic notion and we get from them, by abstraction-
generation, the curves.
The Cosmic Ensemble 365

passe-partout notions) way of explaining the algebra–geometry symbiosis. In par-


ticular, this offers us the starkest way (known to me) of rethinking the ur-notion
of compact Riemann surface as an algebraic curve. The rethinking rests on a
bridge theorem due to Belyi.
We start with the ur-idea of a Riemann surface. Riemann proved that a
Riemann surface X is compact if and only if it is isomorphic to the Riemann surface
of an algebraic curve f(x; y) = 0 for some polynomial f(x, y) in C[x, y].Those that are
key to the generation of our notion of curve are the polynomials with coefficients
in the field of algebraic numbers. Belyi’s result shows that the Riemann surfaces
corresponding to such polynomials f(x, y) are those obtained from certain bipartite
maps. The bipartite maps are the reflection of the dessin’s actions. A bipartite map
M consists of a bipartite graph G embedded (without crossings) in a compact,
connected, oriented surface X, so that the faces (connected components of X\G)
are simply connected (thus, the dessin). One can describe M by a pair of permu-
tations g0 and g1 of its edge-set E: The vertices can be colored black or white, so that
each edge joins a black and a white vertex; the orientation of X then determines a
cyclic ordering of the edges around each black or white vertex, and these are the
disjoint cycles of g0 and g1, respectively.
In a nutshell, what I propose is that the mystery of how a rational elliptic
curve can be, in its “other life,” a Riemann surface is that it is generated from
the underlying surface.28
Let me reiterate the point just made, for it strikes me as the fundamental
point (or the basic conceptual conjecture) I am trying to make. The point was just
made using the language of bipartite maps and there are other languages to make
it (regarding “correlation” of L-functions or Galois representations, etc.) but at its
bottom it is a conceptual conjecture viz. that the “correlation” is not one of good
luck, a cosmic lucky break, but rather one in which we find that two seemingly
different kinds of objects (phenomena) are in truth two forms of the same object
(phenomenon). The correlating equation is not a “lucky match” but rather the
result of our re-thinking the very objects we are talking about in new-deeper terms.
Thus, if I may allow myself two simplified models of such “identifications” (“cor-
relations”), we may state the conjecture that E = M, where E is the ranking by
heights of the employees of a certain bank and M is their ranking by their respec-
tive salaries. This strikes as an accidental correlation (pending astounding discov-
eries about our innate disposition to favor in promotions tall people). At the other
end, we may consider identifications of the E = MC2 type, where we come to
re-understand certain familiar notions in a deeper way and see that they are not
really two “separate” categories that happen to match but that, in fact, the one is a

28. More precisely, Belyi showed that a compact Riemann surface X is defined over Q if and
only if there is a Belyi function B from X to the Riemann sphere S = C U {•}, that is, a meromor-
phic function on X which is unbranched over B\ {0, 1, •}. Of course, the “conceptual” fact just
worked out—that the surface originates the curve (not the other way around)—is just a “back-
ground” fact. The modularity theorem’s claim that modular curves parametrize the elliptic curves
on Q is a very strong claim and that is where the complex mathematics comes in to establish how
deep invariant features of surfaces control the rational curves.
366 Joseph Almog

form of the other, an engendered form emerging from the other. Of course, there
are cases which may initially pass for “cosmic accidents” but which, upon a deeper
tracing of the mathematical structure used on both sides might well turn out not
accidental at all (e.g. famously, spacings between consecutive zeros of the Riemann
zeta function align with spacings between consecutive energy levels of heavy
atoms, a seemingly accidental alignment. But it may not be so when it turns out that
both are forms of some eminently natural sequencing of eigenvalues associated
with certain types of random matrices).
The claim proposed here is that the modularity theorem rests on a correla-
tion of the E = MC2 type, a re-thinking of a familiar algebraic item—elliptic
curves—in deeper terms borrowed from complex analysis.We find ways of “param-
etrizing”, better put informally by a metaphysician like myself in terms of “gener-
ating”, the one form—elliptic curves—from the deeper form—the modular forms.
The re-thinking runs like this. We consider elliptic curves over Q (up to isomor-
phism over algebraic-numbers). And we now re-think them as compact Riemann
surfaces of genus one definable by polynomial equations with rational coefficients.
The modularity theorem then explains to us the generative basis of the elliptic
curves: for each such surface S, there is a congruence subgroup G of SL(2, Z) (Z the
integers) and non constant analytic map over the complex upper half place H,
G\H→S. It is in this map of G\H Into S that lies the generative basis of S and the
elliptic curve it encodes.29
So much then for the embedded meaning of the Diophantine statement, the
H2O of the cubic equation—microscoped, the elliptic curve turns out to be gener-
ated from a complex analytic object, a compact Riemann surface. In saying
xn + yn ! zn, we, like it or not, aware of it or not, in effect, speak of compact
Riemann surfaces in the very sense that in saying “water,” we, like it or not, aware
of it or not, in effect, speak of hydrogen hydroxide.
And there is more structure, fine structure, to our embodied meaning, just as
an H2O molecule packs a lot of fine structure underlying its hanging together.What
goes into the rational points of the elliptic curve hanging together?
I spoke of the elliptic curve and the set of its rational points as the H2O of the
Diophantine equation. We want to microscope further the basis of the structure of
these rational points. This involves a combination of various results (and conjec-
tures) about the basis of elliptic curves. Let me give two fundamental examples of
such basis-facts regarding fine-structure.
Given the curve E over Q, the group of rational points E(Q) makes an
Abelian group. A fundamental theorem regarding E(Q) due to Mordell back in
1922 shows that E(Q) is finitely generated. Thus, even if E(Q) has infinitely many
points, there is always a finite basis of points from which the full infinite set might
be reached by finitely many additions.

29. An interesting detailed working out of such a generation of elliptic curves over Q from
modular forms is offered by Barry Mazur in “Number Theory as a Gadfly”, American Mathemati-
cal Monthly 98, 1991, 593–610. He there shows, working inside the complex upper half plane H,
how to derive the elliptic curve (thought of as Riemann surface) from (Mazur’s variant) of the
holomorphic X0(N) modular forms mentioned in the appendix below.
The Cosmic Ensemble 367

Next, to understand better the group structure—and here we veer from


Mordell’s theorem toward a second structural claim about E(Q)—we need to be
able to calculate the rank of the curve, a measure of the size of the set of rational
points on E.
A key heuristic used here was to connect the absolute number of rational
point solutions to a count of solutions modulo p, at each prime p. Each such field
at p is of course finite and involves at most finitely many solutions, thus computa-
tionally a more tractable task. The heuristic idea was that an infinite absolute
solution set is witnessed in a large number of solutions mod p at each p. If we call
n(p), the number of solutions mod p, the heuristic idea was that for many primes p,
the ratio p/n(p) should be much less than 1. If so, when we take the product over
all primes p, it should be to 0. Now, a fundamental conjecture about the fine
structure of elliptic curves urges that the converse is also true: if the infinite product
is 0, the solution count should be infinite. This conjecture is called the Swinnerton-
Dyer/Birch conjecture. We may well describe it as the bridge to infinity conjecture.30
What am I reading into these fine structure results about elliptic functions?
I said earlier that the embodied meaning of an FLT-like Diophantine equa-
tion is an elliptic curve over Q. On my reading, the algebraic object has a very
complex underlying structure. We telescope this structure by unraveling the curve
to be what it really is, a Riemann surface, whereby tools of complex analysis track
the fine structure of this now revealed-to-be-complex analytic object. The elliptic
curve is no more isolated-ly algebraic (an “algebraic island”) than the modular
curve is distinctly geometric (a “geometric island”). Both are forms of Riemann
surfaces, which are neither algebraic nor geometric but algebraico–geometric or
geometric–algebraic or simply just what they are: Riemann surfaces, describable at
once by the languages of algebra (curves, Abelian groups, Galois representations
etc.), as by the languages of hyperbolic geometry and analysis.

The Epistemology of Diophatine Statements: the Evolutionary


Character of Cognition
As in the case of the fine structure of water, the fine structure of elliptic curves is
anything but apparent. We made use here of various forms of the modularity
conjecture (now theorem) for semistable elliptic curves, and, in the more general

30. I abstract here from the very complex problem of calculating such an infinite product (see
below in this note).
Harping on our theme of integration of different domains, it must be noted that the key
involvement of such infinite products over all primes leads to the activation of analytic methods of
calculating this infinite product derived from Dirichlet’s account of the density of primes in
arithmetic progressions, methods extended by Hasse to the conjecture (now after the modularity
theorem work, theorem) that there exists an analytic L-function defined for every complex
number s computing such infinite products. The Swinnerton-Dyer/Birch conjecture is that E(Q) is
infinite iff L(E,1) = 0. This would settle in particular whether there are infinitely many points on
the specific elliptic curve expressing the congruent number problem (whether there is a rational-
sided right triangle with area d).
368 Joseph Almog

case (now also proved), for any elliptic curve over Q.31 We also mentioned in
passing the fundamental conjecture (not a theorem yet, although proved for
special cases)—the bridge to infinity—of Birch and Swinnerton-Dyer.
What is striking in both cases is the conjectural-natural-science—rather than
a priori self evident—profile of the investigations.32 Both conjectures mentioned
did not arise from reflection, even in the broadest sense, on the apparent meaning
of the key notions involved or from antecedent axioms. Rather, the conjectures
were derived regressively as it were from theorems that needed proving. One
looked for generalizations about elliptic curves that would explain what was sus-
pected to be a true observation, sometimes supported by actual computer calcu-
lations of large numbers.33 In both cases, restricted forms of the conjecture were
proven first, suggesting subsequent bold generalizations.34 Through and through,
we do not have an a priori set of principles validated by inspection of the apparent
meaning from which we squeeze stronger and stronger theorems. Rather, through
and through, we have theorems awaiting proof, theorems-in-waiting. We unravel
their embodied meaning—such and such elliptic functions are claimed to have no
rational points. To understand the notion involved—elliptic function—we investi-
gate, painstakingly by trial and error, its fine structure. This sends us looking for
underlying mechanisms in a richer (if more complex) language—we understand
the H2O molecule by invoking atoms, that invoke protons and electrons (etc.), and
we look for complex mechanisms of bondage relating the atoms. In like manner, we
borrow from the language of bipartite maps, as from the language of complex
analysis and L-functions and their definability over the whole complex plane and
from the associated Galois representations, all in order to microscope the fine
structure of the elliptic curve. We then conjecture about the borrowed notions
generalizations that suggest new test cases, new observations. This is how Frey
could have contemplated a hypothesis—critical-test-case—for refutation: Look for
a hypothetical semistable curve and check to see whether it is modular (bound by
the conjecture that they all are).
If this is not the profile of investigation manifested in conjectural-
experimental natural sciences, what is?

31. This allows us to prove Fermat-like claims (but generalized), for example, a perfect cube
cannot be written as a sum of two relatively prime nth powers n greater or equal to 3.
32. Another such anything-but-a priori conjecture that we have not discussed about Fermat-
type equations is Mordell’s generalization of his 1922 result to curves of genus 2 and higher, proved
by Faltings in 1983. See the above-mentioned paper of McMullen “From the Dynamics of Surfaces
to Rational Points on Curves,” in the last footnote (note 35).
33. I have in mind statistical data about solutions for large primes p (working in mod p)
amassed by computers by Swinnerton-Dyer and Birch.
34. E.g., the modularity theorem was proved for elliptic curves with complex multiplication
or, by Wiles, for semistable curves. In the Swinnerton-Dyer and Birch conjecture case, Coates and
Wiles (in 1976) proved that if E is a curve with complex multiplication and L(E,1) is not 0 then E
has only a finite number of rational points, in the case of class number 1. In 1991, Rubin showed
that for elliptic curves defined over an imaginary quadratic field K with complex multiplication by
K, if the L-series of the elliptic curve was not zero at s = 1, then the p-part of the Tate–Shafarevich
group had the order predicted by the Birch and Swinnerton-Dyer conjecture, for all primes p > 7.
The Cosmic Ensemble 369

INTERMEZZO

Let us take stock and see what, if anything, has been accomplished vis-à-vis the
various “inevitable’ dualisms we started with. At the origin, there were two such
types of dualisms: intra-mathematical and trans-mathematical. In the first category,
I mentioned the late nineteenth century and thus “modern,” vertical form of
dualism between practiced mathematics (arithmetic, algebra, geometry, analysis,
etc.) and the foundations level of a universal characteristic, whether Frege’s Logic
or modern set theory. I also pointed to a more ancient, horizontal as it were,
dualism between “algebra” (and things calculational) and “geometry” (and things
imaginable-visualizable). In the trans-mathematical dimension, I mention, follow-
ing Hermite–Gödel, a metaphysical dualism of realms of existence, the material
cosmos versus a “whole world” (realm) of mathematical objects; and I connected it,
again following Hermite–Gödel, with an epistemological dualism of modes of
apprehension—unlike God, we are doomed to live with separate (and segregable)
modes of apprehension and, consequently, knowledge: the senses that inform us
about the material cosmos and our intellect, informing us about the “other,”
mathematical realm. Let us ask then: Where do we stand on this quartet of
dualisms?
I have pursued the two intra-mathematical dualisms by examining a case
study, elementary number theory, Diophantine equations. I tried to bring out how
both the foundational dualism between practice and ultimate foundations and,
now inside the practice, between the algebraic and the geometric, are illusory.
Diophantine geometry practice is as unitarian as could be. Intra-mathematical
integration shines through by looking at such simple equations as FLT or the
congruent number case.
We have not yet broached in earnest—at least not head-on—the two trans-
mathematical dualisms. But even here progress was made and we gained some
chips that await deployment. On the metaphysical front, the as it were “ontology”
of mathematics, we have isolated the primality (in both senses of “firstness” and
“fundamentality”) of the complex plane and Riemann surfaces as ur-mathematical
categories. And on the epistemological front, we took notice of the conjectural non
a priori character of investigating what I called the embodied meanings (and their
fine structures). Indeed, I analogized such investigations with the unfolding of
“water’s” embodied meaning and its fine structure.
Granted, we are not yet back with one unified world—the cosmos with
nature-cum-mathematics forming one whole. And granted again, we are not yet
free of the senses-versus-thought dualism of modes of apprehension, that is, we are
not yet back with one synthetic mode of apprehending the cosmos, human imagi-
nation. This is the task of the complementary Part II.35

35. An autobiographical word about the growth of the present project. In presenting these
ideas orally, it was pointed out to me that a central theme—(1) there is much more to the meaning
of (what is said by) a Diophantine equation than meets the eye and (2) that leads us to unravel
intra-mathematical symbiosis between different “areas” of mathematics—has been pursued in two
other works. One is the work of Andre Weil, in particular his “Two Lectures on Number Theory-
Past and Present”; see his Collected Papers, vol. III (New York: Springer, 1979), 279–302 (where
370 Joseph Almog

APPENDIX: SOME NOTIONS FROM THE FLT PROOF

Two propositions concern us here. Together they prove FLT.


The first is the weak form of the modularity conjecture used by Wiles:

(SMT) Every semistable elliptic curve over Q is modular.

The second is the Frey-Serre-Ribet conditional:

(SR) If FLT is false, (SMT) is false.

I do not discuss the proofs of either, only the notions they mention. Expansions of
the present glosses giving technical details may be found in Fernando Gouvea, “A
Marvelous Proof,” American Mathematical Monthly 101 (3, 1994): 203–22. A thor-
ough technical foundation is afforded by the aforementioned text A First Course in
Modular Forms. Here we offer an informal summary of the key notions.
Consider first (SMT). We are focused here on an elliptic curve E over Q. As
noted, by a theorem of Mordell, the group of rational points E(Q) is finitely
generated. Now, in seeking (the number of) rational solutions for E, it is natural to
work mod p. A field Fp is finite and so is the group E(Fp). Expounding the structure
of such groups (and through it that of the more elusive E(Q)) is the basic
methodology.
Now, for the reduction to mod p to work we must make sure E is still elliptic
at the Fp’s. This calls for checking the discriminant of the curve (see note 22) and
verifying it is still not zero at Fp. If at p, it is zero, p is called a bad reduction. One
also needs to check that curves that are isomorphic over Q, do not get different
reductions modulo p.
This leads to a classification of types of reductions.To code the reduction type
(there are three), the notion of a conductor N is introduced, with n(p) = o for good
reduction, n(p) = 1 for almost good (“multiplicative”) reduction and n(p) " 2 for
bad (“additive”) reduction. The overall value of N is the product, for all p, of n(p).

historically much richer examples are dissected on top of the Fermat case). The theme is already
broached beautifully (and with much pedagogical patience) in his letter of 1940 to Simone Weil
(pp. 244–55 in Vol. I of the Collected Papers, especially pp. 251–54, where he is working out his
algebraic/Riemannian “dictionary” [“Rosetta stone”]). On quite another dimension but regarding
the very statement I focus on, FLT, Curtis McMullen argues very convincingly (and with a
charming quote from Molière to cement his case) that what Fermat meant with his Diophantine
equation is a claim in “arithmetic topology” (about the dynamics of complex surfaces). See “From
Dynamics on Surfaces to Rational Points on Curves,” Bulletin (New Series) of the American
Mathematical Society 37 (1999): 119–40.
I owe many thanks, over the years, to discussions with Bill Tait, Sten Lindstrom, co-teaching
with Tony Martin and, ten years ago, to hard questions from my searching student Dominik
Sklenar (alas, now, lost to the subject). Recent co-teaching and co-thinking with John Carriero
about themes of cosmology in Descartes’ early modern science was very formative. I also owe
many thanks to the second-hand remarks of J. Oesterle and L. Schneps and to the first-hand
comments from M. Harris, R. Langlands, B. Mazur, W. Goldring, N. Goldring and especially to that
inimitable teacher Serge Lang.
The Cosmic Ensemble 371

An elliptic curve is a semistable curve if its conductor is a square free number.


In other words, its reduction is as good as could be (good reduction at almost all
primes, almost good reduction at the rest).
Now, as for modular functions and curves. A function f(z) on the upper half
plane (thinking here of H = {z = x + iy: y > 0}) is a modular function of level N (i) if
f(z) is meromorphic, even at the “cusps” (on which more in a moment) and (ii) for
all integers a,b,c,d with ad-bc = 1 and N|c, we get

f ( azcz ++ db ) = f(z)
We consider transformations z → (az + b) / (cz + d) of the upper half plane asso-
ciated to the group of matrices:

Γ 0 (N) =  {
a b
 c d
: a, b, c, d ε Z, ad-bc = 1 N c }
The group acts on H with quotient H/G0(N). There is a compact Riemann surface
X0(N) such that:

H/Γ 0 (N) = X 0 (N) − {finite set of points}

The points are the cusps. X0(N) is a modular curve of level N.


The modularity theorem now reads as follows. Consider an elliptic curve over
Q, y2 = Ax3 + Bx2 + Cx + D.The theorem asserts the existence of non-constant func-
tions f(z), g(z) of the same level N such that:

F(z)2 = Ag(z)3 + Bg(z)2 + Cg(z) + D

For more on the sense that X0(N) thus offers a parametrization (or better (Mazur):
uniformization) of the elliptic curves as well as for a genuine development of the
notions involved, see A First Course in Modular Forms. For the role of X0(2) (and
the fact that there are no cusps of weight 2 at level 2) in showing the Frey curve
could not exist (because it is semistable but not modular), see Lang Survey of
Diophantine Geometry, 130–35.

You might also like