Algebraic Symmetries I Just as we can factor z 3 1 = (z 1)(z 2 + z + 1), we can factor z 5 1 = (z 1)(z 4 + z 3 + z 2 + z + 1) It follows that each of the four

four numbers zi = cos(2k/5) + i sin(2k/5), satises the equation z4 + z3 + z2 + z + 1 = 0 Thus our algebraic interpretation of the ve vertices of the regular pentagon as the ve fth roots of unity has destroyed the ve-fold symmetry. We have distinguished one vertex, placing it at the point 1=1+0i. So we now have to look for a dierent kind of symmetry, that among the four remaining vertices, or better the four remaining roots. There is a one obvious symmetry, that which interchanges z1 and z4 as well as z2 and z3 . This is an algebraic as well as an geometric symmetry because it is just a matter of replacing each of the numbers by its complex conjugate a + bi a bi and (a+bi)(c+di) = acbd+(ad+bc)i acbd(ad+bc)i = (abi)(cdi) I take it as obvious that the complex conjugate of the sum or the dierence of two complex numbers is the sum or the dierence of their complex conjugates. (a + bi) + (c + di) = (a + c) + (b + d)i (a + bi) (c + d)i = (a bi) + (c di) In other words the operation of complex conjugation that interchanges z1 and z4 as well as z2 and z3 z1 z4 , z2 z3 k = 1, 2, 3, 4

is like reection in a mirror. All arithmetic properties are faithfully preserved. 1

AB

A B = AB

Small remarks It is sometimes useful to recall that the relation

is the same as z4 + z3 + z2 + z1 + z0 = 0

Why is, for example, z1 z2 = z3 ? (cos() + i sin())(cos() + i sin() = cos( + ) + i sin( + ) Take = 2/5 and = 4/5. Then + = 6/5 and this becomes z1 z2 = z3 One shows in the same way that, for example, z2 z2 = z4 . Observe also that, along the same lines,
7 z1 = (cos(2/5) + i sin(2/5))7 = cos(14/5) + i sin(14/5)

and cos(14/5) + i sin(14/5) = cos(4/5) + i sin(4/5) = z2 This is because the angle 14/5 is equal to 2 + 4/5 and the cosine and sine do not change when 2 is added to or subtracted from an angle. Indeed, in some respects, the angle itself does not change! (Note: this statement is correct, but calls for some reection!)

Algebraic Symmetries II This is because i is just a symbol that stands for the square root of i and i is then introduced and dened by the condition that i + (i) = 0 But i is also just a symbol and can be taken as the primary symbol. Then i is a second symbol that functions as (i). Even if i is taken to have some meaning beyond that of a mere symbol, it cannot have a dierent meaning than i, so that the two have to be regarded as perfectly interchangeable.

Are there other symmetries of this kind? Whether there are other symmetries of this kind aecting all complex numbers is not a question for us, but we can ask whether there are symmetries of this kind aecting just z1 , z2 , z3 and z4 . Before we do, we make use of the symmetry at hand. Since z1 z1 = z2 z1 z2 = z3 , z1 z3 = z4 , z1 z4 = 1, z2 z2 = z4 , z2 z3 = 1 and so on, and since in addition 1 = z1 z2 z3 z4 the numbers az1 + bz2 + cz3 + dz4 , where a, b, c and d are arbitrary ordinary fractions form a collection closed under addition, subtraction, multiplication, and even as it turns out division. The numbers that are equal to their own reections can be singled out. These are the numbers a(z1 + z4 ) + b(z2 + z3 )

The distinguishing characteristic of i is that i2 = 1 but this characteristic is shared by i. 4

The appearance of

5.

Let w be the number z1 + z4 . It is equal to its own reection. So is its square. Thus w 2 = a(z1 +z4 )+b(z2 +z3 ) = (ab)(z1 +z4 )+b(z1 +z2 +z3 +z4 ) = (ab)w b Thus w satises a quadratic equation w 2 + cw + d = 0, We calculate this equation
2 2 w 2 = (z1 +z4 )2 = z1 +z1 z4 +z4 z1 +z4 = z2 +1+1+z3 = 21z1 z4 = 1w

c = b a,

d=b

Thus w +w1=0
2

w=

1+4

Since w is a positive number, 1 + w= 2 5

In other words, w can as we know be constructed with ruler and compass. Since 1 2 = w z1 + 1 = z1 w z1 + z4 = w z1 + z1 we have z1 = Since w 2 = 1 w, this is w 3 w 1 + 5 = 2 4 Since z1 lies above the axis of abscissas, 1 + 5 +i z1 = 4
5+ 5 2

w2 4 2

5 2

5 2

Symmetries III Having found z1 , we can easily nd z4 , its complex conjugate, and we can certainly nd z2 by squaring z1 . We can also nd z2 by working with z2 + z3 rather than w. This is, however, straightforward algebra. As Descartes insisted, the algebra often turns a problem into an almost unthinking manipulation of symbols, a turn that it can indeed often take, but we prefer another direction. This is the direction taken by Gauss. Let be the number z1 . Then z2 = 2 , z3 = 3 , and z4 = 4 = 1 2 3 because (I) 4 + 3 + 2 + + 1 = 0

Our numbers ez1 + fz2 + gz3 + hz4 can also be expressed as a + b + c 2 + d 3 a = h, b = e h, c = f h, d=gh

Are these numbers all dierent? This is an important question! The answer is yes and I shall give a proof following Gauss. We rst note the consequences. Since the numbers are all dierent, we cannot have an equation a + b + c 2 + d 3 = 0 = 0 + 0 + 0 2 + 0 3 unless a = b = c = d = 0. This means that satises the equation (I) Z4 + Z3 + Z2 + Z + 1 = 0

and essentially only this equation, because if we have any other such as (II) Z 6 + AZ 5 + BZ 4 + CZ 3 + DZ 2 + EZ + F = 0

then by the process of long division we will have Z 6 +AZ 5 +BZ 4 +CZ 3+DZ 2 +EZ+F = (Z 2 +GZ+H)(Z 4 +Z 3+Z 2 +Z+1)+P Z 3+QZ 2 +RZ+S so that P 3 + Q 2 + R + S = 0 We have just seen that this implies P = Q = R = S = 0. Thus Z 6 + AZ 5 + BZ 4 + CZ 3 + DZ 2 + EZ + F = (Z 2 + GZ + H) (Z 4 + Z 3 + Z 2 + Z + 1) and (II) is a consequence of (I). 6

Symmetries IV This means that, from an abstract point of view, is simply a number that satises the equation 4 + 3 + 2 + + 1 and no other. But z2 , z3 and z4 have exactly the same property. Thus for strictly algebraic purposes we could take to be z2 rather than z1 . Thus z1 is replaced by z2 . Since all algebraic relations are to be preserved, this entails 2 2 4 3 3 6 replacing z2 = z1 by z2 = z1 = z4 and z3 = z1 by z2 = z1 = z1 . Once again, because all algebraic relations are to be respected, the number
4 2 3 z4 = z1 = 1 z1 z1 z1

is to be replaced by
2 3 4 8 3 1 z2 z2 z3 = z2 = z1 = z1 = z3

In general a number
2 3 a + b + c 2 + d 3 = a + bz1 + cz1 + dz1 = (b a)z1 + (c a)z2 + (d a)z3 az4

is replaced by
2 3 2 4 2 3 a + bz2 + cz2 + dz2 = a + bz1 + cz1 + dz1 = (a c) + (d c)z1 + (b c)z1 cz1

which is a number of the same kind. For example

Forms of algebraic symmetries The symmetry just examined can be viewed in two ways: 1) It takes the sequence {z1 , z2 , z3 , z4 } of all roots of Z4 + Z3 + Z2 + Z + 1 to a sequence formed from the same numbers but in a dierent order {z2 , z4 , z1 , z3 } 2) It takes any number (I) to a number (II) az2 + bz4 + cz1 + dz3 of the same kind. This is the kind of symmetry that was later investigated in general by Galois. We have to spend some time growing accustomed to it. Suppose we apply the symmetry twice. Then z1 z2 z4 z2 z4 z3 z3 z1 z2 z4 z3 z1 Thus applying the basic symmetry twice leads to the rst symmetry considered, complex conjugation. We apply it again. z1 z4 z3 z2 z3 z1 z3 z2 z4 z4 z1 z2 Yet again! z1 z3 z1 z2 z1 z2 z3 z4 z3 z4 z2 z4 So the symmetry when repeated four times comes back where it began. It is a four-fold symmetry. 8 az1 + bz2 + cz3 + dz4

Anticipating Galois and his successors No matter which of the numbers z1 , z2 , z3 , z4 we take to be, the collection of numbers (I) a + b + c 3 + d 4 , a, b, c, d all fractions

is the same. Modern mathematicians usually call the collection a eld. the sum and the product of two numbers of this sort are again numbers of the same sort. This we have seen already. I give another example. (1 + )(1 + 3 ) = 1 + + 3 + 4 = 2 Any symmetry of this collection that respect the algebraic operations will take 0 to 0 and 1 to 1. Then adding and dividing it takes any fraction a/b to a/b. Moreover any root of Z4 + Z3 + Z2 + Z + 1 = 0 will be taken to another root. Thus z1 will be taken to z1 , z2 , z3 or z4 . In other words, the symmetry will be one of the four (including the trivial symmetry!) we already have. Denote the one taking z1 to z2 by the letter . Then the repeating to obtain = we obtain the symmetry taking z1 to z4 . Repeating again, we obtain = 3 which takes z1 to z3 . Repeating again, we nd that = 4 is the trivial symmetry. Inside the collection of numbers there is a smaller collection of numbers that have a special symmetry. We met them before. They are those that are not aected by 2 , thus by complex conjugation. They are the numbers a+bw, w = z1 +z4 . We were able to construct z1 by succesive square roots, by rst singling out this special collection of numbers, nding that any number in it satised a quadratic equation with fractions as coecients, in particular (1+ 5) 2 2 , and then solving z1 z1 w + 1 = 0. that w + w 1 = 0, so that w = 2 We now apply these ideas, which I hope are clear, to the heptadecagon!

The Heptadecagon

z5 z6

z4 z3 z2

z7 z1 z8 z0 z9 z16 z10 z15 z11 z12 z14 z13

zk = cos(2k/17) + i sin(2k/17)

10

From the Disquisitiones Arithmeticae There is a famous remark from the introduction to the seventh and last chapter of the Disquisitiones that I quote here. What it anticipates is the study of the division points on elliptic curves, in the remark a special elliptic curve, a study that led over the course of the nineteenth and twentieth century to many things, especially complex multiplication and l-adic representations, that are relevant to the Shimura-Taniyama-Weil conjecture, Ceterum principia theoriae, quam exponere aggredimur, multo latius patent, quam hic extenduntur. Namque non solum ad functiones circulares, sed pari successu ad multas alias functiones transscendentes applicari possunt, e. g. ad eas, quae ab integrali dx 1 x4 pendent, praetereaque etiam ad variam congruentarium genera: sed quoniam de illis functionibus transscendentibus amplum opus peculiare paramus, de congruentibus autem in continuatione disquisitionum arithmeticarum copiose tractabitur, hoc loco solas functiones circulares considerare visum est.

11

Lecture 7

A proof by Gauss (beginning) Recall that we want to show that z1 is a root of the equation I Z4 + Z3 + Z2 + Z + 1 = 0

but of no equation of the forms Z 3 + aZ 2 + bZ + c = 0 Z 2 + aZ + b = 0 Z +a=0 (I) (II) (III)

in which a, b c and d are fractions. The impossibility of the last equation is clear because z1 is not a fraction. If it were a root of the rst, then using long division to divide (I) by Z 3 + aZ 2 + bZ + c, we would nd Z 4 + Z 3 + Z 2 + Z + 1 = (Z 3 + aZ 2 + bZ + c)(Z + d) + eZ 2 + f Z + g Substitute z1 to nd that z1 is also a root of eZ 2 + f Z + g = 0 Then, as we just observed e is not 0, unless e = f = g = 0. If e is not 0, divide by it. Thus either (IV) Z 4 + Z 3 + Z 2 + Z + 1 = (Z 3 + aZ 2 + bZ + c)(Z + d)

or z1 satises an equation of type (II). If it satises (II), then perform a long division to obtain Z 4 + Z 3 + Z 2 + Z + 1 = (Z 2 + aZ + b)(Z 2 + cZ + d) + eZ + d Since ez1 + d cannot be 0 unlesse = d = 0, we conclude that (V) Z 4 + Z 3 + Z 2 + Z + 1 = (Z 2 + aZ + b)(Z 2 + cZ + d)

12

A proof by Gauss (continued) We now show that the factorizations of (IV) and (V) are impossible. We rst observe a very important fact. An equation of degree n (VI) Z n + aZ n1 + bZ n2 + . . . + d = 0

cannot have more than n roots! Suppose (VI) has a root e. Using long division, divide by Z e. The result is Z n + aZ n1 + bZ n2 + . . . + d = (Z e)(Z n1 + AZ n2 + . . . + D) + f Substitute e to see that f = 0. Now any other root e not equal to e of (VI) must be a root of (VII) Z n1 + AZ n2 + . . . + D = 0,

so that if (VI) had more than n roots, then (VII) would have more than n 1. All we have to do now is continue, working our way down to lower and lower degree until we arrive at an equation of degree one Z+ =0 that clearly has only one root.

13

General comments The proof we present is a proof that can be extended without too much additional eort to the following statement. Suppose p is a prime. Then the polynomial Z p1 + Z p2 + Z p3 + . . . Z + 1 admits no factorization. This is a statement proved by Gauss in the Disquisitiones. We will need it for p = 17. e According to Bourbakis Elments de lhistoire des mathmatiques, this e was the rst general statement of this sort about polynomials ever proved. It seems to me that in one sense, it is also the last. In the nature of things, there are very few large classes of polynomials that are all irreducible. On the other hand, if taken as a statement in the sense of Galois theory, as a statement arming that the Galois group of the equation is large, then there are indeed other statements of this sort available and perhaps many more yet to be proved. The size and nature of Galois groups is a reasonably important mathematical topic, at least for mathematicians. Since the proof in the general case is not so dierent from the proof for p = 5, we can expect a certain amount of sophistication.

14

A proof by Gauss (continued) Suppose we had a factorization (IV). Since z1 , z2 , z3 and z4 are all roots of Z4 + Z3 + Z2 + Z + 1 = 0 each of them is either a root of (VIII) or of Z +d= 0 Since the latter is impossible, because d is, in contrast to z1 , z2 , z3 and z4 , a fraction, they are all roots of (VIII) which gives (VIII) one root too many. Thus we must have (V) Z 4 + Z 3 + Z 2 + Z + 1 = (Z 2 + aZ + b)(Z 2 + cZ + d) Z 3 + aZ 2 + bZ + c = 0

The rst thing is to establish that a, b, c and d are all integral. For this we need a little number theory. Suppose that one of these numbers is not integral. Then it is of the form m , n n positive, n>1

in which m and n have no common divisor. To be explicit, we write 3 30 as 1 1 = 10 25

Thus there is some prime number p such that n = rpt , t > 0, p does not divide r. For example, if m/n = 1/10 then p could be 5 and we would have m = 1, r = 2, t = 1. We write each of a, b, c and d in this way. a= a1 , a2 pq b= b1 , b2 pr 15 c= c1 , c2 ps d= d1 , d2 pt

Suppose that at least one of the numbers q, r, s, t is positive. If, for example, q = 0 then p might divide a1 . Thus suppose that not all of a, b, c and d are integral. We multiply (IV) out to obtain Z 4 + (a + c)Z 3 + (b + ac + d)Z 2 + (bc + ad)Z + bd = 0. First of all, a + c = 1 is an integer. This is possible only if q = s, for if q is not equal to s then one of them is positive and larger than the other. The denominator of a + c would contain the factor pn if n is the larger of q and s. For example 2 3 47 2+95 + = = 3 52 5 3 52 3 52 We conclude that q = s. Looking at b + ac + d, we next conclude that either r = t or q + s is equal to one of r or t. This line of argument quickly becomes confusing. The best thing is to follow Gauss and to prove a general theorem. Oddly enough, it is an argument that can best be explained in the general case. Here is what we want to show. Suppose that the product of Z n + a1 Z n1 + a2 Z n2 + . . . + an1 Z + an and of Z m + b1 Z m1 + a2 Z b2 + . . . + bn1 Z + bm has integral coecients and that all of the numbers ai and bj are fractions (and not some more complicated kind of irrational number!). Then all of the numbers ai and bj are integers. We suppose not and choose a p that divides the denominator of at least one ai or one bj . Then we write ai = mi ni pri bj = sj tj psj

It is understood that p does not divide ni and that it does not divide tj . Starting with r1 , examine all the ri and let rk be the rst that is at least as large as all the others. Thus rk is bigger than those that came before and at least as large as those that come after. In the same way sl is to be larger than the sj that come before and at least as large as those that come after. 16

m+nkl

The product is of degree m + n. we look at the coecient of the power . It is am+nkl b0 + am+nkl1 b1 + am+nkl2 b2 + . . . + a0 bm+nkl

Some of these terms are purely ctive, those in which ai appears with i larger than n or bj with j larger than m. If a0 or b0 appear they are taken to be 1. For example, if n = 5, m = 3 then the coecient of Z 4 is a4 + a3 b1 + a2 b2 + a1 b3 In general, the coecient of Z k+l is supposed integral and contains the term ak bl in whose denominator prk +sl occurs. It also contains terms like ak1 bl+1 whose denominator contains at most the factor prl 1+sl or like ak+1 bl1 whose denominator contains at most the factor prk +sl 1 . Thus when we put everything over a common denominator, we will have a prk +sl in the denominator that cannot be removed. This is a contradiction. An example. 1 2 4 1 7 (Z 3 + Z 2 + Z + ) (Z 2 + Z + ) 3 9 9 3 3 Here k = 2, l = 1, so that m + n k l = 2. The coecient of Z 2 in the product is 1 8 1 21 8 3 32 17 24 1 + + = + + = + + = 33 93 9 3 27 9 27 27 27 27

17

Final step We now know that in (V) the four numbers a, b, c and d are integral. The argument we used was taken from 42 of the Disquisitones, thus from an early chapter. The next part is from 341, thus from a very late, indeed the last chapter, the one devoted to the division of the circle, thus to regular polygons. The number z1 will be a root of Z 2 + aZ + b = 0 or of Z 2 + cZ = d = 0 We can suppose that it is a root of the rst. Then z4 , its complex conjugate will also be a root, so that the roots of the second are z2 and z3 . The polynomial Z 2 + aZ + b is divisible by Z z1 Z 2 + aZ + b = (Z z1 )(Z z4 ) = Z 2 (z1 + z4 )Z + 1 = Z 2 2 cos(2/5)Z + 1 which we write as Z 2 2 cos(2/5)Z + cos2 (2/5) + sin2 (2/5) = (Z cos(2/5))2 + sin2 (2/5) In the same way, Z 2 + cZ + d = (Z cos(4/5))2 + sin2 (4/5) Thus if Z is an ordinary real number, both Z 2 + aZ + b and Z 2 + cZ + d are positive. If Z is an integer, they are integers. Since (Z 2 + aZ + b)(Z 2 + cZ + d) = Z 4 + z 3 + Z 2 + Z + 1 we have rst of all, for Z = 1, (1 + a + b)(1 + c + d) = 5 Thus one of these numbers is 1 and the other is 5. On the other hand, Z 2 + aZ + b + Z 2 + cZ + d = Z 4 (z1 + z2 + z3 + z4 )Z + 2 = 2Z 4 + 3 which is 5 when Z = 1. We conclude that 1 + 5 = 5, which is out of the question. 18

19

20

III. Neue Entdeckungen. Es ist jedem Anfnger der Geometrie bekannt, dass a verschiedene ordentliche Vielecke, namentlich das Dreyeck, Viereck, Funfzehneck, und die,welche durch wiederholte Verdoppelung der Seitenzahl eines derselben entstehen, sich geometrisch construiren lassen. So weit war man schon zu Euklids Zeit, und es scheint, man habe sich seitdem allgemein uberredet, dass das Gebiet der Elementargeometrie sich nicht weiter erstrecke: wenigstens kenne ich keinen geglckten Versuch, ihre Grenzen u auf dieser Seite zu erweiten. Desto mehr, dnkt mich, verdient die Entdeckung u Aufmerksamkeit, dass ausser jenen ordentlichen Vielecken
noch eine Menge anderer, z. B,. das Siebenzehneck, einer geometrischen Construction fhig ist. Diese Entdeckung ist a

eigentlich nur ein Corollarium einer noch nicht ganz vollendeten Theorie von grsserem Umfange, und sie soll, o sobald diese ihre Vollendung erhalten hat, dem Publicum vorgelegt werden. C. F. Gauss, a. Braunschweig, Stud. der Mathematik zu Gttingen. o Es verdient angemerkt zu werden, dass Hr. Gauss jetzt in seinem 18ten Jahr steht, und sich hier in Braunschweig mit eben so glcklichem Erfolg der Philosophie u und der classischen Litteratur als der hheren Matheo matik gewidmet hat. Den 18 April 96. E. A. W. Zimmermann, Prof.
21

Die Allgemeine Literatur-Zeitung From Meyers Enzyklopdisches Lexikon: a This review appeared in Jena from 1785 to 1803 and in Halle from 1804 to 1849 and was a leading organ of German classical and romantic literature. Goethe, Schiller and Kant were among the editors and authors. It moved to Halle as a result of an eort of the romantics to increase their control over the review and was replaced in Jena, on the initiative of Goethe by the Jenaische Allgemeine Literatur-Zeitung. Gauss Gauss, who was born in 1777, wrote the Disquisitones between 1796 and 1798, thus between his nineteenth and twenty-rst years. It did not appear until 1801. It contains a great deal in the way both of theorems and theories, the most important being: 1. Proof of law of quadratic reciprocity. The statement was already known at the time, but even the best of the eighteenth century mathematicians were unable to nd a proof. It remains a central mathematical theorem. 2. Developed the theory of binary quadratic forms. In particular, he introduced the notion of composition of quadratic forms and established its properties. Although in some respects, namely in the context of the notion of ideal number, composition has become a common working tool of all algebraists and number-theorists, Gausss theory itself is still dicult and little known. His form of the theory would appear to be that best suited to computation. 3. Cyclotomic elds. The construction of the regular heptadecagon, and, more generally, the analysis of the numbers formed from roots of unity. This is thus one of the earliest manifestations of Galois theory. Gauss presumably knew more than he included in the book, but he, apparently, published very little more on the subject. His thesis of 1799 established, in eect and for the rst time, that every polynomial equation has a root. Z n + aZ n1 + bZ n2 + . . . + d = 0 As the root may be complex and the thesis did not refer to complex numbers, the formulation of the thesis was necessarily somewhat dierent. I have already emphasized that this is a basic mathematical fact. These are all theories and results to which the contributions of Gauss are clear. In addition, he appears to have occupied himself as a very young man, even as an adolescent, 22

with other important problems and observations, for example, with the nature of geometries in which the parallel axiom of Euclid is not satised and with the arithmetic-geometric mean, which is both an elementary and an advanced topic. Here, however, the evidence is dierent. It consists of Gausss recollections as a somewhat saturnine older man, so that it appears to be dicult to disentangle what he himself discovered later, what he learned from other authors as an adolescent he had an early and extensive acquaintance with the work of various leading eighteenth century mathematicians as well as what he learned later, from what he discovered early. One could certainly spend a lot of time with Gausss Collected Works and with his correspondence, reecting on these matters. One extremely useful reference is a diary that Gauss kept between 1796 and 1814, with 146 entries that record his principal discoveries. The lemniscate function to which, as I noticed, Gauss alludes in the rst lines of chapter 7 of the Disquisitiones appears several times. I was asked what Gauss what may have meant with his allusion to the lemniscate. A brief examination of the diary and of the collected works, which contain ample editorial comments, makes perfectly clear what Gauss knew. Although it is a digression from my main purpose, it is worthwhile to tarry a little on the matter. The goal of this years lectures is after all to communicate, starting at the beginning, some genuine mathematical understanding, beyond the gee-whiz or what I refer to as the Jack Horner manner, of recent achievements in number theory, and it would be a shame, since we are now in a position to appreciate a more detailed explanation of Gausss allusion, to let slip the occasion of acquiring more concrete information. Gauss was an overwhelming presence in nineteenth century mathematics, Even though he exerted little active inuence. The German pre-eminence in mathematics as a whole, and in certain domains such as number theory in particular, that lasted throughout the nineteenth century and until the early thirties is due in good part to him, although the Prussian university system was probably also a signicant factor. I have not studied these matters. Oddly enough it appears to have been Andr Weil who was most thoroughly e imbued with the aspirations of German number theory, both through direct, personal experience as a young man during the twenties and through his studies of variou nineteenth century authors. Simplifying, for the purposes of brevity, an elaborate development in which a large number of mathematicians took part, one might say that he not only brought it intact through the war at its highest level but also was the principal source of its transformation into the theories that were nally and successfully exploited in the proof of Fermats theorem. His major contribution was, oddly enough, a set of conjectures, the Weil conjectures, now demonstrated, but by others. He is, in various comments to his papers, quite explicit about the relation of these conjectures to Gauss and the lemniscate. I quote from Weils Two lectures on number theory, past and present, an essay I recommend to your attention. In 1947, in Chicago, I felt bored and depressed and, not knowing what to do, I started reading Gausss two memoirs on biquadratic residues, which I had never read before. . . . This led me in turn to some conjectures. 23

A few lines later Weil draws attention to the very last entry in Gausss diary, an entry to which we shall come in a moment.

24

Z n + aZ n1 + bZ n2 + . . . + d = 0

25

Digression The digression at rst sight seems to demand some knowledge of the calculus, but it does not. I rst write down a formula that may be familiar to some, but not to all. No matter! Do not puzzle over the left side. It is no more than the mathematicians usual fastidious way of writing down the length of the arc from the point (1, 0) on the circle to the point (x, y) and that is what we mean by , which of course has to be measured in radians, thus in units in which the radius is 1, but that is the unit chosen. If
y

(A)
0

dt = , 1 t2

then y = sin(). If y = 1 then = /2.

(x,y)

(1,0)

We now do something similar for the lemniscate, a curve dened by the equation (x2 + y2 )2 (x2 y2 ) = 0

26

The lemniscate is the curve in the form of a bow. The point (x, y) is the point where the line through the points (0, 0) and (1, u) cut the curve. The length of the curve from (1, 0) to the point (x, y) is expressed mathematically as
u

(B)
0

dt = . 1 t4

Once again, there is no need to be troubled by the integral. It is again just a way of expressing the length of an arc of a curve. Observe that plays here the role of the angle in a circle measured in radians. Gauss wrote u =sin lemn(). If u = 1, then is some number that I call, following Gauss, /2. Thus /2 is the length of the upper loop on the right running from (1, 0) to (0, 0).

0.5

(x,y) (1,u)

x=

1 u2 1 + u2

-1

-0.5

0.5

u 1 u2 y= 1 + u2

-0.5

-1

27

Constructing a regular triangle, a regular pentagon, or a regular heptadecagon is the problem of dividing the total circumference of the circle into three, ve or seventeen arcs of equal length. We could consider the same problem for the lemniscate, taking the initial point, which is now important as it was not, because of symmetry, for the circle, to be the point (0, 1). In an entry for March 19, 1797 Gauss notes that this leads to an equation for u of degree m2 , whereas for the circle it was an equation of degree m. In the cases already considered, m was 3 or 5. We remove one easy root, u = 0 corresponding to the rst point of division. This leads to equations of degree m 1 or for a lemniscate m2 1. In a later entry, apparently for April 15, he observes there is a problem of separating the real roots of this equation from the complex. He is seeking the real roots. The corresponding equation for the circle, thus for y, has only real roots. It is the numbers x + iy that are complex. The real roots of the equation of degree m2 1 give the division points and there are m 1 of them. In an entry dated March 21, he observes implicitly (all the entries are cryptic) that these m 1 roots are numbers that can be constructed with a ruler and compass. Lemniscata geometrice in quinque partes dividitur. The young Norwegian mathematician Abel published in 1827 and 1828 proofs of the assertions implicit in Gausss remark in the Disquisitiones. For m = 3, 5 the necessary constructions had been found much earlier, by 1750, by the Italian geometer Count Fagnano The connections of the lemniscate with number theory are too ramied for us to discuss them in any more detail. The division points were important then and are important now, but so are what are called congruences modulo a prime. The two topics are very closely related. I end with the entry to which Weil alluded, the very last in the diary, dated July 9, 1814.

28

29

30

31

32

Observatio per inductionem facta gravissima theoriam residuorum biquadraticorum cum functionibus lemniscatis elegantissime nectens. Puta si a+bi est numerus primus, a-1+bi per 2+2i divisibilis, multitudo omnium solutionum congruentiae 1 xx + yy + xxyy inclusis x = , t = (a 1)2 + bb. y = i; x = i, y = , (mod a + bi)

We have progressed a little beyond this last entry in the intervening 185 years, but not so much as one might think! Without going into further detail, which would be too much of a digression, I observe that if 1 = x2 + y2 + x2 y2 and z = y(1 + x2 ) then z 2 = y2 (1+2x2 +x4 ) = y2 (1+x2 )+y2 x2 (1+x2 ) = 1x2 +x2 (1x2 ) = 1x4 .

33