1 >> Kristin Lauter: Okay. So we're very pleased to have Ken Ribet here visiting us today from UC Berkeley, and he'll speak to us about -- uh-oh, password required -- modularity of two-dimensional reducible Galois representations. Thank you. >> Kenneth Ribet: Thank you very much for the introduction and thank you very much for the invitation. I've known Kristin for a long time and I've known Microsoft Research for a very long time, and I've sent graduate students here as interns and corresponded with some number of you over the years. This is my very first visit to Microsoft, and it's wonderful. So I really like it. I wondered what to talk about here, and I realized there's a very interesting subject that grew out of a question that I was asked by a mathematician in Barcelona. And this really has to do with some very classical objects that are probably familiar to many of you in the audience and have just served as a wonderful source of very profound questions for -I wrote at least a hundred years. It's kind of a benchmark. But a lot of the arithmetic of these objects has been going on roughly for 40 years. I think that's another benchmark, as I'll explain. So I'm talking about modular forms in a very classical sense, so these are functions, F of Z, and the number Z is just a complex number with positive imaginary part. These are holomorphic functions. And these functions, the way that they are usually discussed arithmetically is through their Fourier series representations. These functions have some large group of symmetries, some group of functional equations that they satisfy. And one of the simplest functional equations is just in variance under integer translation. And so it's very natural to regard these as functions of the exponential, either the 2 pi IZ and then represent them as some power series, and in addition to the holomorphicity on the upper half plane, there are a number of growth conditions that I don't discuss specifically or explicitly that are imposed on these functions that imply that these functions can be written as Fourier series beginning with a constant term and then positive powers of either the 2 pi I times the natural variable. And these Fourier series for us are basically the objects that we study, so even though these are holomorphic functions, from another perspective, all they are is just formal power series. And the coefficients that occur a priori can be any complex numbers, but the types of functions that we're looking at -- I used the word arithmetical a moment ago -- really have the property that the numbers that occur are in the most interesting cases ordinary integers and slightly more generally they're algebraic integers. Not only that, but if they're algebraic integers as opposed to rational integers, all of them as N varies will lie inside of some fixed number field. So it's not as though as you generate more and more of these coefficients you increase the size of the field that you need to contain them; they're all contained in some finite extension of the rational number field, they're all going to be algebraic integers. And typically the field in which they'll be contained in the examples I discussed will be totally real. That means that all of the conjugates of all of the numbers in this field will be real numbers as opposed to complex numbers. So it's very restrictive and kind of a very good place to work. 2 Here are some of the equations that may satisfy some of these functional equations, as I call them. Basically aside from some automorphy factor, what's going on is that the functions have to be invariant under fractional linear transformations coming from some large subgroup. It will be a subgroup of finite index in the group that I write, SL(2, Z), it's just the group of integer matrices with determinant 1. In addition to finite index, they actually lie in a congruent subgroup, which in this case is a stronger condition. It just means that they will -- a congruent subgroup is a subgroup that contains all of the matrices congruent to the identity modulo some number, usually called positive n -- positive capital N, rather. That number is kind of the level of the group. The number K that appears in automorphy factor CZ plus D to the K is called the weight of the modular form, and because I am going to be keeping things very simple, the weight will be a positive even integer in all examples. And aside from one example that I give, basically the weight is going to be 2. So you have a very simple equation: F of AZ plus B over CZ plus D is CZ plus D squared, F of Z. And that means, equivalently, if you think about calculus, that if you write F of Z and multiply it by a formal differential, DZ, that thing's going to be actually invariant under the group. There are different standard congruent subgroups. The deepest one that people use usually after some conjugation consists of matrices ABCD, where C is divisible by some integer capital N, that's the level, as I called it, and A and D are congruent to 1 modulo that same N. If you do that you get a slightly smaller group than the one that I've written down. It's called gamma one of N. And then in addition to the types of considerations that we have, there is a Dirichlet character, modulo N, that plays a role. And here I'm just taking the character to be the identity for simplicity. And we get a bigger group called gamma zero of N. It's one of the standard subgroups of SL(2, Z). And not only will I take N to be a positive integer, I'm really chickening out here and letting N be square free, so that means it's just a product of some number of primes; for example, it could be 1, it could be a prime. And in my examples, the worst that it will actually be is a product of two different primes. So it's not a very complicated situation. >>: [inaudible]. >> Kenneth Ribet: Yeah. Did I write something wrong? I'm sorry. That's -- have my laser pointer here. Thank you for the misprint. That's C is zero, not N. Okay. Thank you, Pierre. Good. Okay. Let's see what I have on this slide. In addition to being modular forms, which I said implies satisfying some growth conditions, we actually want these forms to go to zero along the boundary. And that means, for example, that when you go back -- I don't want to make you too dizzy -- and look at the Fourier series in particular this condition will mean that the series begin with a constant multiple of Q or some higher power of Q, and the constant term is actually zero. And in favorable situations, there is a theory of Eisenstein series that shows that any 3 modular form can be written as a sum of some fairly standard series that you write down a priori and a cusp form. The cusp forms are much wilder. You don't know them kind of a priori. And so by looking at cusp forms, in a sense I am looking at the interesting part, F to the D composition. Another way to say this in weight two is that when you have a weight-two modular form and you multiply it by a formal differential, DZ, it's not the same D that appears in the matrix, that formal differential is invariant by the fractional linear transformations. So it's really a differential on the quotient on the upper-half plane by gamma zero of N requiring that the form be a cusp form is the requirement that the differential extends to the boundary and doesn't blow up so it becomes a differential on the natural compact identified quotient, which is usually called X zero of N. It's a standard modular curve. This modular curve is I'm going to be say in a little bit begins life in this description as a Riemann surface. It's just an object over the complex numbers. But there is a theory that is very often associated with Shimura of canonical models, which shows that the modular curve actually has a standard description as an algebraic curve over the rational number field. And not only that, but there is a well-defined way that you can take this curve and you can start looking at it over finite fields, reduce it modulo of prime, P, for prime's P, not dividing N at least. That theory began with Aguzza [phonetic] in the 1950s. And perhaps the final thing that I'm going to ask about these -- well, not really final, but I want to constrain things to really get at the arithmetic by requiring some recursion conditions among the coefficients of the Fourier series, and these conditions are really simply encapsulated by a single formula that are these standard Hecke operators in certain cases were initially defined by Mordell that operate on spaces of modular forms. And we're going to ask that these modular forms by eigenvectors for all of these Hecke operators. So that means that if you hit F with teese [phonetic] of N and hitting it with teese of N is traditionally written with a right action, you get a constant multiple of the form. And if you read, for example, Shimura's book on automorphic representations -- I think that's what it's called -- you find that things can be normalized so that the modular form is really wearing its eigenvectors -- eigenvalues, rather, on its sleeve. What will happen is that the first coefficient, the coefficient of Q in the Fourier series is necessarily going to be nonzero in -- if you have this formula, otherwise all of the coefficients will be zero. You want a nonzero eigenvector. And once the first coefficient is nonzero, you can scale the form by it, just divide by that nonzero number, make the first coefficient be 1. And then what happens is that the Nth coefficient becomes equal to the Nth eigenvalue. And so you have this very beautiful situation where hitting the modular form by the Nth Hecke operator just means that the Nth coefficient comes out as a factor. And this implies the Hecke operators are set up so that they're multiplicative in the sense of functions in number theory. That means that the coefficient ace [phonetic] of NM, if N and M are two integers that are relatively prime, is just the product of a respective coefficients for N and for M. And not only that, being an eigenvector for the teese of P means that you can express by some simple formulas A sub-P to the K, for P being a 4 prime and K some higher power. That's not the weight. I'm sorry. That's just a random power of P. You can express all the P power index coefficients simply in terms of the Pth coefficient. So, in other words, when you have a situation where the form satisfies this condition of being an eigenvector for the Hecke operators, all of the information about its Fourier coefficients is already carried by the prime-indexed coefficients, ace of P. And therefore what you're interested in basically is the packet of eigenvalues associated to this form. And what we're interested in is packets that are occurring for the first time as you look at the different possible levels. So you can imagine that there's a packet that occurs for level 11 that will occur for level 22 and level 33 and for all multiples of 11 if you get some packet that occurs at level 77, that hadn't occurred at level 7 or level 11, then you're very interested in it and you say that you have a new form; otherwise, you kind of ignore it except when you're really at the first level where it occurs for the first time. There's a little technical point to mention which is that as you go from level 11 to level 77, say, what happens is that you perturb the eigenvalue for the prime 7, in my example. And so what we want to do is really look at these packets for all but finitely many primes and require that the packet is new in that sense, that they're -- that something is old if it's occurred before even where you take out a couple of exceptions. And that's the definition. This comes from an article in the late 1960s I think by Atkin and Lehner. And they call forms that are normalized in this way new forms. And new forms are really the object of study in the subject. Nowadays, if you do Langlands' theory, you say that the new forms are really the same as the automorphic repetitions of the adelic group, GL(2), with some condition involving the component infinity. That condition is exactly what determines the weight. And here we're taking a very classical perspective. We're really looking at these things as functions on the upper-half plane or as Fourier series. And now I said already that the case that mostly interests me is the case K equal 2. On the other hand, there's another case where the level is 1, but you let the weight grow. And that's a case that many of you have heard of before. If you have the full modular group SL(2, Z), you start looking for cusp forms of different weights for weight two and weight four and weight six. I said the weights were even. There's nothing doing, there are no nonzero forms. And the first time you get an interesting new form is case K equals 12. And what happens when you have K equal 12 is you get the beautiful Ramanujan Tau-function. And this really was studied by Ramanujan in 1916 or so where he discovered empirically a number of amazing facts about the coefficients of a certain formal power series that I've written out there. It's the product of 1 minus Q to the N raised to the 24th power. You shift it by multiplying the whole thing by Q. The coefficients of that power series satisfies, as Ramanujan saw, these amazing multiplicativity properties. For example, tau of a product of relatively prime integers is the product of a tau values. And tau of a prime power can be traced back to tau of the relevant prime. 5 Ramanujan saw that this was true and I think those facts were proof for the first time by Mordell in this article where Mordell prediscovered the Hecke operators that Hecke discovered later in Humborg [phonetic], but there are also kind of amazing congruences. I guess they occur in the next slide. Here are the first few values of the tau function. They grow quite rapidly, as you can see experimentally, and they've been tabulated -- it's a great project to check your computer algebra program to see how fast and how well you can tabulate these numbers. There's an open conjecture that none of these values is ever zero, for example. That's something that people have tried unsuccessfully to attack. And as you'll see in a moment, there are a group of congruences that are known for the tau values that were noticed and in many cases proved by Ramanujan himself. A word that describes this form and the one that's at the bottom is eta products. This is related to the Dedekind eta function. Here is a very closely related example which has weight two. But if you want the weight to be two, since for SL(2, Z) there are no nonzero things for weight two, you want the capital N to be something nontrivial. And if you take N to be 11, that's going to be the first example of a nonzero cusp form on gamma zero of N and weight two. And I've written down the cusp form here. It's a related project, a product. And this was studied by Eichler and Shimura and Serre and lots of other people. If you take this product and you write it out, I just called these things ace of N. They don't have a special name like tau of N. And then as far as the congruences go, a very famous congruence that was proved by Ramanujan is modulo to the prime number of 691, tau of P is congruent to 1 plus P to the 11th. And an analogous congruence for the other modular form, the one of weight two and level 11, is simply that ace of piece congruent to 1 plus P to the 1st power, modulo 5. If you want to compare them, you'll probably notice that 11 is one less than the weight and this P could probably be written P to the first power, 1 is again one less than the weight, so they really are very parallel. And I ask the question what is the meaning of those congruences. And I'll give an answer only for the second of the two, the one that has to do with weight two, and there the answer involves elliptic curves, and the answer has been known for a very long time. And this is really roughly the 40th anniversary of a seminar that Jean-Pierre Serre gave on the Ramanujan tau-function in a Paris number theory seminar where he asked himself what is the meaning of the first congruence. And he posited the existence of some analog of elliptic curves for forms of higher weight. And that was an incredibly productive remark. It gave rise to the whole theory of motives attached to modular forms, which are analogs of the elliptic curves that are attached to certain modular forms of weight two. For more complicated examples, you need abelian varieties, if the coefficients are algebraic integers but not rational integers. And basically there's an analog for 691 of what I'm about to say for the ace of P. But historically it took a while for people to really understand the importance of having motives attached to modular forms of higher weight. And that was really tremendous advance in our understanding of the arithmetic of these Fourier series in the general case. 6 Well, now, as kind of an aside, I mostly want to focus in this talk about weight two. And there there's an elliptic curve. So if you have a new form of weight two, the fact of the matter is that it gives rise to something in relatively simple algebraic geometry. If the ace of N attached to a new form of weight two for gamma zero of N generate over the rational field some algebraic extension of Q of dimension D, D might be 17, let's say, then you get a 17 dimensional abelian variety with a big field of endomorphisms. Actually, what acts on it is the ring of integers of the field generated by the coefficients. When the coefficients are ordinary integers, the abelian variety is an elliptic curve. The elliptic curve in this case is the Jacobian of the modular curve that I wrote down before, X zero of 11. That modular curve is of genus 1. The Jacobian is there for essentially the modular curve. And you can write down the simple equations for the elliptic curve. I should have done that, and I'll say, you know, it's kind of Y squared plus Y equals X cubed minus X squared minus two further coefficients. Very easy to find them on online tables, for example, those of William Stein and John Cremona. This is an elliptic curve that we know very, very well. And the point about the relationship between the cusp form and the object and algebraic geometry, even in the general case, is that the coefficients can be seen by taking traces of Frobenius. So in the elliptic curve case, the simple thing to say is you have a Weierstrass equation, you have a cubic equation defining the elliptic curve. Take a prime number P different from 11. Remember I said before that Aguzza proved that X zero of N was kind of okay. Modulo prime's not dividing N. That's I think from around 1957. And when you reduce an elliptic curve, mod P, there's some Frobenius endomorphism that's widely used in cryptography. Lots of you know about it. But its main raison d'être in the whole subject is that it counts points. You want to know how many points your elliptic curve has mod P. If you know the coefficients of the modular form, the answer is very easy: it's 1 plus P modulo, the ace of P. And so that's one connection between the elliptic curve and the modular form. Another connection which is strongly related is that if you take a random prime number, let's call it L, and you look at the points of order L on the elliptic curve and then what you'll get is a two-dimensional vector space over the field with L elements. L is a prime number. And on that two dimensional vector space you have an action of the Frobenius coming from the field with P elements. Take the trace of the action on a two-dimensional vector space. You get a number, mod L, and that number is ace of P, modulo L. And if you have some knowledge about ace of P modulo 5, L is going to be equal to 5, then you say to yourself, oh, what's going on. Well, what's going on is that there's something special happening with the group of L division points, 5 division points for this elliptic curve. So what I can do is I can take the elliptic curve. You know this is a complex torus. If you take the kernel of multiplication by some integer, call that integer 5, do this over the complex numbers, what you'll get is the integers mod 5, a cyclic group of order of 5 taken twice. You can say that that's because you're elliptic curve is the complex plane modulo of lattice. The lattice is a free abelian group of rank 2. And you ask what it means for 5 times a point in that quotient to be zero, to be the identity element of the group, well, it means that the point comes from 1 over 5 times the lattice, modulo the lattice. And that's 7 something of rank 2 over the integers mod 5. So you have a module of five division points and because you know how to write multiplication or addition on the elliptic curve in terms of polynomials with integer coefficients or rational coefficients, you see that the solutions to the five division equation have coordinates that are really algebraic numbers. And these coordinates get permuted around by a Galois action, action of the group of all -- Galois group of all algebraic numbers. And you get a representation of that Galois group on this group of five division points. And the point is that what you really get is for every element of the Galois group, you get an automorphism of a vector space. You get a kind of two -- invertible two-by-two matrix over the field with five elements. And what you're saying when you look at this congruence and you look at the equation above, is that you're getting understanding about the trace of that action, the trace of that representation, because you know what happens when you take Frobenius elements in the Galois group, and there's a theorem of Chebotaryov of these Frobenius elements kind of determine everything. So what you could discern even kind of abstractly just from this congruence is the fact that this two-dimensional vector space over the field of five -- field with five elements has something very special going for it, and you could even pin down what that is. But from another point of view what you can do is just kind of examine that thing as an object for which you're amassing information and you can see that it contains a very nice subgroup on which the Galois action is trivial. This is called this cuspidal group. And it also contains another subgroup on which the Galois action is cyclotomic. That's called the Shimura subgroup, really discovered by Shimura. What is mu [phonetic] 5? Mu 5 is the group of roots of unity of order 5, consists of 1 in the -- 4 nontrivial 5th roots of unity. The action of the Galois group of Q on that cyclic group defines a little character from the Galois group of Q over to the automorphisms of this mu 5. Automorphisms of mu 5 is just the invertible integer is mod 5. It's a cyclic group of order 4. And there's a wonderful paper which I recommend to all comers. It's from 1977. It's Barry Mazur's paper called Modular Curves and the Eisenstein Ideal. And he kind of explains in general for J zero of N where N is a prime that there are special prime numbers, if N is 11 the special prime number is 5. And in general it's the divisors of N minus 1 except sometimes you have to ignore 2 and 3. And for those divisors what's happening is that you have a kernel analogous to the E of 5 and you have a cyclic group of order L. L here is 5. And also another cyclic group of order L, one of them with trivial action and one of them with cyclotomic action. And this is kind of the very first instance of this whole thing. Shimura, whom I've mentioned quite a bit here, has a beautiful article in Crella's [phonetic] journal from around also 1967, I think, where he talks about the action of the Galois group of Q on the division points of this very special elliptic curve. And he says to himself, well, this is really amazing. You know, class field theory is a study where 8 you try to explain what are the abelian extensions of number fields or local fields, and you do that in terms of the internal structure of the number fields or local fields. And there was always this question starting with Artin in the 1950s, you know, what happens if you try to look at nonabelian extensions of arithmetic fields. How can you describe them. And, you know, Artin is famous for saying make a conjecture and I'll prove it. And nowadays I think that the best answer to that conundrum is to say that we can't really describe very well the absolute Galois group of all extensions of Q or a number field or a local field, but the best we can do is try to explain what the representations of that group on various objects explain what those representations are related to. And Shimura said, well, you know, here's an example where I'm producing in general nonabelian extensions of Q and they're related to this specific modular curve, X zero of 11. This requires further study and we should try to understand what we can about these extensions and try to understand the arithmetic. And that was a kind of challenge thrown out in 1967, and there were two very prominent echoes of that. One was a beautiful article that Serre published in 1972 in French called Properties of Torsion Points of Elliptic Curves, where he said that if you take a random elliptic curve over the rational number field, one without complex multiplication, the action of the Galois group of Q on the division points is incredibly rich. The Galois representations that you get, aside from some finite fooling around, will have images that are as large as you can contemplate, they'll kind of fill up all of the GL(2)s that are receiving them. And then there's this 1977 article five years later by Mazur where he talks specifically about the arithmetic of the abelian varieties, not only elliptic curves, coming from Jacobians of modular curves. So there's something very deep and varied that kind of explains this very simple congruence; namely, something geometric -- I'm talking so much that my laptop is falling asleep; I hope you're not -- something geometric is going on that explains these simple congruences. So the way I want to write this in kind of capsule form is that you started with a modular form. The modular form was the one of level 11 and weight 12. That's F. And if you have the number 5, then there's an associated Galois representation. So the Galois representation as a module is just this E of 5. And if I think about the homomorphism that describes the action of the Galois group of Q on this module, well, I call that row F5. And all of this equation is simply to say that row F5 is a direct sum of two representations: the trivial representation, that's the Z mod 5; and the cyclotomic representation, I call it cy [phonetic]. Some people might call it epsilon or omega . Cy sub-5, the mod 5 cyclotomic character. And actually, more generally, what might happen is that this row F5 five instead of being a direct sum like that might be an extension; namely, it might be upper triangular [inaudible], not irreducible. And in one of the corners you had the trivial representation and the other you had the cyclotomic representation. 9 For me, for this purpose, I'd say that's just as good. Kind of look at situations like that and say that the mod 5 representation attached to the modular form is this direct sum. So we just kind of semisimplify. If you have something upper triangular, just look at the diagonal. Okay. So there are the words that repeat what I just said. This is a semisimple representation of the direct sum of the trivial one and the cyclotomic character, it comes from this form of level 11. And the question is why 11? You know, why is it important that the number be 11? You might say what is the kind of true level of this representation, the direct sum of the mod 5 cyclotomic representation and the trivial representation? And one answer -- well, I'll just stick by my slide so I don't do too much digression anymore -- is that one of the things that Mazur shows is that if you start with a level capital N, then aside from 2 and 3, which I said were special, the primes that play the role of 5 are the divisors of N minus 1. So the point about 11 is just simply that it's congruent to 1 mod 5. So if you have another such prime, for example, 101, then you'll find by taking one of the modular forms, one of the new forms of that level and one of the primes dividing five in the ring of integers of the field containing the coefficients, you will find the congruence like the one that we had a couple of slides ago, ace of P congruent to 1 plus P mod 5. So to say it in a different way, this trivial-looking direct sum, the trivial character and the cyclotomic character, will occur as -- from a new form of weight two and level N anytime N is congruent to 1 mod 5. Okay. And this is just some slightly technical point, which I already said; namely, if you had a form whose coefficients were, let's say, you know, in the ring of integers of Q to the square root of 101, then what would happen perhaps, and in the case it does, is that 5 would split into different prime ideals. And what you do is you take the different reductions of your modular form, one of them might satisfy the congruence, another one doesn't. A perfectly good example is instead of 5 you could take 11 and then you take L to be 23. In that case there is a unique new form up to Galois conjugation of weight two and level N, and the coefficients are not rational integers, they're integers in I believe the ring of integers of Q to the square root of 5. And in that ring the prime 11 splits as a product of two different maximal ideals. These maximal ideals live their own independent lives. Modulo 1 of them, the ace of P, will be congruent to 1 plus P. Modulo the other, there's no such congruence whatsoever. So when I say that there's a new form that is doing this thing, mod L, that just is shorthand for saying that it's doing it modulo 1 of the primes dividing L, but probably not more than 1. And here's the thing that I've now said twice, which is that the key congruence in this example is N congruent to 1 mod L. 10 And earlier in discussing this with you, I mentioned Eisenstein series. If you have an arbitrary weight-two modular form on gamma zero of N, you can write it as a multiple of this very well-mannered series, which I'll discuss in a moment, the sum of a multiple of that and a cusp form, the kind of thing that goes to zero at infinity. And this Eisenstein series, it has a 1 minus N in it floating around, and then aside from that the major term in this N minus 1 over 24 is the second Bernoulli number, which is minus the 6, I think. And then there's also a 4 in the denominator that's twice the weight. There's all this numerology of the thing. And the coefficients of the Fourier series, the Nth coefficient is the sum of the positive divisors of N, except you throw out divisors that are divisible by the prime capital N. So, for example, if you take N to be a random prime number P, the coefficient is 1 plus P. And if you take it to be N, the coefficient is 1. And the point what's really behind -- kind of drives Mazur's paper -- remember, Mazur's paper has the word "Eisenstein" as part of the title -- is that if you take this cusp form, this Eisenstein series modulo L, L is a prime different from 2 and 3, if L happens to divide N minus 1, then when you look at this Fourier series, it kind of starts with zero. So it looks as though it is a cusp form. And what happens is that in that case you can really produce a cusp form that's congruent to it. That's how you get the kind of congruence that we've been discussing. And the converse is that if you have a cusp form which is congruent to the nonconstant part of the Fourier series attached to the Eisenstein series, then there's a theory, a so-called Q expansion principle that forces the constant term to be zero module L as well. And that's how you get that L divides N minus 1. So you really have this very tight theory. You know exactly when it's true that the representation 1 plus cy 5 comes from a modular form of level N; namely, N has to be congruent to 1 mod 5. And in that description 5 can be replaced by any prime number greater or equal to 5 in fact. And although I'm sure -- I haven't looked at Ramanujan's paper, at least lately, one way in which you can interpret what happens modulo 691 is some similar circumstance; namely, you can start writing down Eisenstein series. In the case of level 1 and weight 12, the Eisenstein series begins with a constant -- trivial constant multiple of the 12th Bernoulli number, which has a 691 in the numerator. And it's exactly that numerator of the Bernoulli number that drives the congruence that was discovered by Ramanujan. And there are articles by Serre and Swinnerton-Dyer, again, from around 1972, where they kind of used the Galois representations attached to modular forms. Remember, I mentioned something about motives before. Well, even though motives took a while to get going, almost immediately after Serre suggested the Galois representations for these higher-weight modular forms might exist, Deligne constructed them. And using Deligne's Galois representations, Serre and Swinnerton-Dyer really kind of ordered in a sense of bringing order to. You know, they kind of -- they cleaned up the whole subject. They explained why Ramanujan and his successors could prove the 11 congruences that they did. And they also show that aside from one case where there was a new congruence that hadn't been found by the classical people, there were no more congruences to be found, had a way of showing that the Galois representations in their case had large images. So, for example, they couldn't be upper triangular. They did for modular forms the same kind of thing that Serre had been doing for elliptic curves, if that makes sense. And now there's a little bit of larger context, which I guess I won't go -- well, I should say a little bit about it. So I already have in some sense; namely, in general, if you have a new form, which for this discussion we can think of as just having weight two, all the coefficients rely on some interger ring. For example, for that 23 example it would be the ring of integers of Q to the square root of 5. And you can start reducing the Fourier coefficients modulo different maximal ideals. So these maximal ideals, they can't be called L, L is a prime number, so we call them lambda. And then associated by Deligne, or in this case, for weight two it's especially easy, one could say it's really kind of Shimura, in fact, for every lambda there's a two-dimensional Galois representation. It's just what happens is you take an abelian variety attached to the modular form, it has an action of this interger ring O. That's explained in chapter 7 of Shimura's book that was published in the early 1970s. And you take the lambda division points of this abelian variety, instead of taking the five division points of the initial elliptic curve, Galois acts on it and you get these automorphisms of the vector space, which are really two-by-two invertible matrices. And if you look at the literature in the subject, some of which is by me, the main point is that when you take these representations, you fix F and you start looking at the different lambda, the main point is that for most of them the Galois action is rich. As I just said, in 1972 Serre and Swinnerton-Dyer were showing kind of representations, filled up the group of matrices, and that would impede congruences. What we show is that this happens for all by finitely many cases. So there just aren't that many congruences to be found. But, on the other hand, it does occur that you can have a congruence every now and then. And so maybe we should be spending more of our time thinking about the reducible representations that come from these modular forms. And as I've already said, if you have a reducible representation that means an upper triangular matrix, just look at the diagonal. So you semisimplify. And in the simple case that I've described -- namely, square free level -- it's quite easy to show that the two characters that occur on the diagonal of the matrix up to permutation will be the trivial character and this mod L cyclotomic character. L is the prime that lambda divides. And since we're semisimplifying up to permutation doesn't matter. So it's just the -- reducible case means that you're getting this 1 plus cy cybele [phonetic]. And now what's the question? The question, if we back up a little bit, is to take any two-dimensional Galois representation and ask ourselves does it come from a modular form. And, if so, what kind of modular form's given. So the simplest case would be take the two-dimensional representation here and you say does it come from cusp forms. Well, the answer is it does. And we know, for example, that if you take a weight-two cusp form, you look at the weight-two cusp forms of level N where N is congruent to 1 mod L, you'll get these representations. But you can ask what else? You know, what if 12 the level is not prime, what if it's a product of several primes and so on. What's the condition that guarantees that you get this from some new form. And if we go back now and ask the question in generality where the representation might be irreducible as well as reducible, then we're really in the landscape of other conjectures that Serre made around 1987. Serre published an article in the Duke journal in which he made what are called "Serre's Conjectures," but they kind of started with correspondence with Tate as early as 1972 that kind of took off from this -- these articles by Serre and Swinnerton-Dyer that I mentioned; namely, they try to ask is there a converse to Deligne's construction. That was kind of the question. Given a modular form, Deligne makes a Galois representation, can you go backwards. You know, if you start with a Galois representation, does it come from a modular form. And the irreducible case seemed to be the essential one. It's a very hard question. And I have been going around in the last couple of years explaining a method by Khare and Wintenberger in which they basically have resolved the conjectures; namely, they use a lot of technical algebraic geometry, much of which is due to Mark Kisin from the University of Chicago. And they kind of throw all kinds of different things into the pot, everything in the kitchen. And they really come to the conclusion that if you have an irreducible two-dimensional Galois representation that satisfies some very tiny necessary conditions that come from modular forms, that would be satisfied if the determinant is an odd power of cy L. Odd is essential. They show that the thing actually comes from modular forms, some modular forms. And then there's a lot of literature. A buzz phrase for that is level lowering, that shows that if the thing comes from any modular form whatsoever, you kind of understand the different weights and the different levels of different modular forms that give that same representation. And what I'm asking for is really some recipes to whether or not you can do that sort of thing in the reducible case. In the reducible case, the questions are a little bit different because you know going in that the thing does come from some modular form, it comes -- you can kind of make these Eisenstein series and then make cusp forms if you like that are congruent to them. But then you ask, you know, can you describe all levels and all weights that give that given representation. And all of a sudden the question becomes interesting. If somebody just gave you this 1 plus cy L, you might say that the natural level is 1. I kind of mentioned that here and might have said that before. Because there's kind of -there's no prime that naturally comes into the picture. If you have 5 being L, why 11? Why 101? Well, okay, there's a congruence. But those primes don't seem to be essentially connected to the initial representation. But we do have Mazur's answer. If you look at prime level, if things are congruent to 1 mod N, mod L, and we know there are lots of primes congruent to 1 mod L, but as I was trying to say just a moment ago, these different primes have nothing to do with each other. 11 and 101, they don't all come down to some root that would give rise to them. In the irreducible case, there's kind of like the greatest common divisor. There's kind of -- is that what they're called? No. Least common multiple, or something, you know, 13 take the least common multiple -- there's a smallest one, and then the others are all multiples of it. For the levels in the irreducible case, it's exactly like that. There's a smallest level. It's the minimal level, and all the levels are multiples of it. But in the reducible case, that doesn't happen. And the question is what does? Now, you might ask why do I care? And the answer is that there are lots of mathematicians in Barcelona who are interested in abelian varieties and modular forms and cryptography and the relations among all these things. And one of them is someone named Luis Dieulefait, who's been at Harvard this past spring. I don't know if he's back in Barcelona. And he asked me this very question I'm discussing here. And the question is why did he do that? And the answer is that every time you have a number field -- every time you have a new form, there's an associated number field, field generated by its coefficients, and you might ask whether these number fields can be arbitrarily large in their degree. And this question -- once again, there's some article by Serre in [speaking french] where he's applying analytic number theory to modular forms. And he has some very weak result that the degrees grow in certain situations. And there's -- in Mazur's 1977 article, he has a proposition based on the remark that I'm about to make that the degrees grow if you look at odd prime level. But Dieulefait is interested in levels that are product of two primes or several primes. And he actually sent me this argument where for three or more primes using some variant of Goldbach's conjecture, the result proved by Chen in the 1970s, you could actually treat the problem. And he said, well, we could do it for a prime level and for products of three or more primes, how about products of two primes. So that was kind of interesting to me. I started looking at the product of two primes. And the point is -- I kind of don't want to speak for much longer than an hour, so I'll really slough over this, but if you have a congruence, AP congruent to 1 plus P mod lambda, and it applies for P equal 2, which will be the case if N is odd, then that congruence for very idiotic reasons forces the degree of the field to be large. And the reason is that A2 has to be smaller than 3. For example, you could say, well, if 3 is congruent to 3 mod 5, so -- and Q is not a large field, but there's something called the Riemann hypothesis for algebraic curves proved by Weil in the late 1940s that implies that an A2 is some algebraic integer and all of its absolute values, kind of stick it into the real numbers with all of its conjugates, all of those absolute values are at most 2.83. So they can't be 3. And if you use the triangle equality in some idiotic way, you'll see that the degree of the field has to be bounded from below by essentially the logarithm of the number L. So if you can make congruences modulo larger and larger L, then you can make the degree of the field go to infinity. So that's kind of Mazur's observation, and also Dieulefait's in the case when N is prime. The question is can you make arguments like that for more complex-looking levels, not just simply for prime level. And so, for example, N could be a product PQ, just -- well, here. Product of two distinct primes, suppose L is an arbitrary prime, you make L as large as you like to force the 14 degree up. Can you make the mod L representation 1 plus cy L come from some new form of level N. N can vary. But just a product of two primes. A new form of weight-two in level N. And the answer is you can do that, and I actually can tell you very quickly what the answer is. But when I do that and you have level that's a product of two primes, there are some signs that you can keep track of; namely, if you hit your new form with a Hecke operator TP, you'll get plus or minus the new form, and if you hit it with TQ, you'll also get plus or minus the new form, so there are two signs: a sign at P and a sign at Q. And there are really two essential cases: one is where the sign is plus-minus and the other is where it's plus-plus. And the reason for that is that the minus-minus case never occurs for these congruences. You can rule that out very quickly. And the minus-plus by permuting P and Q can be reduced to the case of plus-minus. So, you know, there are only two essential cases. And in the two cases, the answers are really different. The first case has a kind of very boring answer, and the second case has a much more interesting answer that you can prove theoretically and also verify empirically by looking at different tables. So it's kind of very interesting. In the case of the first situation, you want the eigenvalue for TQ to be minus 1 and the eigenvalue for TP to be 1. All you need is for the second prime to be congruent to 1 mod L. So you want to make a mod L congruence for a product P and Q and some new form of level PQ and weight two. All you have to do is choose a prime. In Mazur's thing you choose primes that are 1 mod 5. Here you choose a prime that's minus 1 mod 5. And then you multiply it by any other prime you like. And you will end up with a congruence. And you can see that empirically. The necessity of this condition is very easy to see by comparing Galois representations, and the sufficiency really requires some argument. You have to look at the way that some variance of modular curves, called Shimura curves, behave when you reduce modulo, the prime Q. One thing I can say on the closing moments of this talk, I mentioned Aguzza a couple of times by saying that you have these modular curves, they have level N, you reduce modulo of prime, not dividing N, you can kind of understand what happens. It took a very long time for mathematics to understand what should happen at primes that do divide N; namely, there was a beautiful article again from 1972 by Deligne and Rappaport where they talk about, for example, the reductions of X zero of N where N is square free at primes dividing N. And there is a -- that's already kind of the length of a small book. And a follow-up article is the length of a very thick book by Katz and Mazur about the arithmetic of modular curves. And when you look at these Jacobians, there's kind of an interesting story of Shimura curves. Mauneed [phonetic] had two very good students, one of them Cherednik, who I think is now at the University of North Carolina in Chapel Hill, and Drinfeld, who's at the University of Chicago. And they worked on their thesis. And Drinfeld -- Cherednik's, rather, was on the reductions of Shimura curves. And Mauneed said, well, you guys are both very good students, I'd like you to give seminars in which 15 you explain each other's work. And Shimura and -- sorry, Drinfeld took Cherednik's work and re-explained it in this way that kind of began a lot of the modern fine theory of the reductions of Shimura curves. And using Drinfeld's work, which was initially Cherednik's, you get a proof that Q congruent to minus 1 mod L is sufficient to make this kind of congruence. The other case is much more interesting, and I'll just tell you the answer. It's when the two signs are plus 1. In that case by comparing with Eisenstein series, just as in Mazur's case, you know that L divides N minus 1, here you know that L divides P minus 1 times Q minus 1, which means it divides either P minus 1 or Q minus 1 or both. So let's assume again permuting if necessary that it divides P minus 1. And then there's a way you can think about the problem that makes it somewhat clear that the answer I'm going to give might be right; namely, already at level P there's a new form. That's by this Mazur result that I keep quoting. And what you're trying to do is kind of promote level P into level PQ. And just as I mentioned level lowering before, this goes by the term level raising. The question is can you raise levels. And that was really the way that Dieulefait came to me with this question. He said, You're an expert on level raising. Can you raise levels also in the reducible case? And I knew that one could but had to work out the details. And the answer is the following. When can you raise the level? Well, there are kind of two cases. These are not in exclusive or if Q divides also Q minus 1 as well as P minus 1, you can raise the level. You'll get a new form of level PQ. And in the other case, L divides P minus 1 but not Q minus 1. And what you do is you look at Q modulo P and ask that it be an Lth power. That's the condition that there be level raising. And the simplest case that I can think of is when L is 5 and P is 11, and then Z mod, LZ star is a cyclic group of order 10, so the 5th powers are just plus 1 and minus 1, the two elements of order 2. And so to say there's a new form of level PQ, 11 times Q in this case, in the case when the second coefficient, when Q is not congruent to 1 mod 5, or even if it is, is to say that Q has to be congruent to plus 1 or minus 1 modulo 11. So you get a kind of very clean result. And once again what's interesting is you can go to William Stein's online tables and see empirically that you really get the level raising that the theory predicts. Okay. That's my talk. [applause] >>: Are there cases in which [inaudible] the case in which [inaudible]? >> Kenneth Ribet: Well, so this is kind of an interesting question. The question is what happens when it's not a direct sum. So, you know, you think about different parts of my brain, I have this theory where I make unramified extensions of cyclotomic fields by looking at cases where it's not a direct sum. And basically the theory says that when you have these reducible cases, if it is in a direct 16 sum, you can kind of divide by finite subgroups and make isogonies until you get some reducible representation, some indecomposable representation, and you work with that. And the thing that's not very well explored, I believe, is whether or not it ever occurs that you get a direct sum. So my theory in that direction was, well, you might get a direct sum and you might not. If you do, that's not interesting for making these extensions, so let's put ourselves in a situation where it becomes indecomposable. In Mazur's case, Mazur has this great philosophy in his paper that you just take the objects as they are, you don't make isogonies, just look at them as they are in nature. When you look at nature, you really get a direct sum. And you can ask geometrically, in my case, do you get a direct sum at the higher level. So, in other words, what I'm saying, or what you're saying, is the following. Let's take a prime -- well, take an example where P and Q are both congruence 1 mod 5. Okay. So, in other words, you would take kind of J zero 11 times 101, for example. Okay. And then what you'd want to do is you'd want to take the kernel of a maximal ideal where you're really looking at the new form, mod 5. Lambda is going to be dividing 5. And this thing -- you make it so it says it's predicted in my theory. And you know that after you semisimplify the thing it's going to be a direct sum of two groups. And now the question is what happen if you don't semisimplify, will that be a direct sum of two groups, and I suspect typically not. But I actually thought about that question for the first time this morning when I was on the plane. And I think it would be interesting to prove that it's not a direct sum. Or to see what it is. >>: [inaudible]? >> Kenneth Ribet: Well, so if L is arbitrary, you start with L. And then the conditions on P and Q are basically congruence conditions. So it's not -- they're not mysterious. It's just kind of analytic number theory of Dirichlet. >>: What was the motivation for producing these large [inaudible]? >> Kenneth Ribet: Ah. That's an interesting question. All right. Let's take this off. I'm not exactly sure how he first came to this question, but I think it came from the analytic number theory of cusp forms. He looked at Serre's paper and was trying to better some of the results. And he's also interested -- he has a colleague or -- the three of us may end up writing a joint paper. His coauthor is named Jimenez in Barcelona. And Jimenez is very strong in analytic number theory and he's also interested in variance of Chen's results and kind of proving better things in the direction of Goldbach. And it all fits within his interests. >> Kristin Lauter: In what sense was it thought that the reducible case was uninteresting? >> Kenneth Ribet: Well, that's -- yeah, why is the reducible case uninteresting? I guess somehow the reducible case, it kind of harkens back to GL(1) and kind of ordinary class field theory, and we just say, well, we know everything there is to know about that. And then you're really interested in essentially two-dimensional things, and those should be attached to modular forms. 17 You know, I mean, for example, in Serre's original conjecture, people would look, for example, at this 1 plus cy L and they'd say, well, you know, somehow morally the level of that thing is 1, there should be some kind of missing Eisenstein series of weight two gamma zero of 1. So it would have as its nonconstant coefficient just the sum of the divisors of N without a prime, and then the constant coefficient would be something like minus 1 over 24. And the problem is that this thing is not a modular form, all of this business about nonholomorphic analytic continuation of it and Hecke. And so people would say, well, you know, it's just kind of an embarrassment, but the thing doesn't exist, and morally it should exist and the natural level of this representation is 1 and we should kind of forget about it and look at the more interesting cases of the irreducible representations. But I guess I'm arguing that you shouldn't forget about it. >> Kristin Lauter: More questions? Thank you. [applause]