Taniyama-Shimura 4: The Conjecture

We’ve done a lot of work so far just to try to define the terms in the Taniyama-Shimura conjecture, but today we should finally make it. Our last piece of information is to write down what the L-function of a modular form is. Since I don’t want to build a whole bunch of theory needed to define the special class of modular forms we’ll be considering, I’ll just say that we actually need to restrict our definition of “modular form” to “normalized cuspidal Hecke eigenform”. I’ll point out exactly why we need this, but it doesn’t change anything in the conjecture except that every elliptic curve actually corresponds to an even nicer type of modular form.

Let {f\in S_k(\Gamma_0(N))} be a weight {k} cusp form with {q}-expansion {\displaystyle f=\sum_{n=1}^\infty a_n q^n}. Since this is an analytic function on the disk, we have the tools and theorems of complex analysis at our disposal. We can perform something called the Mellin transform. It is just a standard integral transform given by the formula \displaystyle {\Lambda (s) = \int_0^\infty f(it)t^s\frac{dt}{t}}.

After some computation you find that this transformed function is a product of really nice functions. We get \displaystyle {\Lambda (s)=\frac{N^{s/2}}{(2\pi)^s}\Gamma(s)L(f,s)}, where {\Gamma(s)} is the usual Gamma function. Now if you actually went through and worked this out you would find out that {L(f,s)} has a really nice form in terms of the Fourier coefficients. The so-called L-series associated to the Mellin transform is given by

\displaystyle \displaystyle L(f,s)=\sum_{n=1}^\infty \frac{a_n}{n^s}.

If your eyes glazed over for the Mellin transform talk, then just think of the L-function of the modular form as taking all of its Fourier coefficients and throwing them in the numerator of this series to make a new function. A quick remark is that if all the {a_n} are {1} (this won’t happen) we recover the Riemann zeta function. Thus you could think of the L-function we get as some sort of generalization of the zeta function. If you’ve been through some elementary number theory you have probably even seen a proof that \displaystyle {\sum_{n=1}^\infty \frac{1}{n^s}=\prod \frac{1}{1-p^{-s}}} where the product is over all primes called an Euler product. Now in general if I hand you a sequence of integers {a_n} that has some reasonable growth condition, then {\sum_{n=1}^\infty \frac{a_n}{n^s}} will be a nice convergent series, probably with an analytic continuation to the plane. The tricky part is to figure out what types of sequences allow this Euler product decomposition.

This is where we have to use that {f} was of this special form. In the theory of modular forms there is something called Atkin-Lehner theory which tells us that the {a_n} for a cusp form of this special type actually satisfy some nice relations such as {a_{nm}=a_na_m} when {(m,n)=1}. These relations are precisely the ones needed to conclude that there is a nice Euler product expansion and it is given by

\displaystyle \displaystyle L(f,s)=\prod_{p|N}(*)\prod_{p\nmid N} \frac{1}{1-a_pp^{-s}+p^{k-1-2s}}.

We say that a variety is modular if {L(X,s)} coincides with {L(f,s)} up to finitely many primes for some {f\in S_k(\Gamma_0(N))}. We’ve been ignoring the technicalities of dealing with the primes of bad reduction and the primes that divide the level (a surprisingly hard problem to determine when these are the same set!), but now we see that for the definition of a variety being modular this doesn’t even matter. There are other subtleties in defining all of this for when the variety does not have {2}-dimensional middle cohomology, but again for our immediate purposes you can trust that people have made the suitable adjustments.

Now we see the truly shocking results of Taniyama-Shimura. We take this incredibly symmetric analytic object (so symmetric it is surprising any exist at all) and we take this completely algebraic variety defined over {\mathbb{Q}} and the conjecture claims that we can always find one of these symmetric things that match up with this action on the cohomology. Wiles and Taylor are often credited with proving it in 1994, but it wasn’t actually proved until 2003 by Breuil, Conrad, Diamond, and Taylor. This was the elliptic curve case.

Just last year Gouvea and Yui proved that all rigid Calabi-Yau threefolds are modular. It is a conjecture that all Calabi-Yau varieties over {\mathbb{Q}} should be modular, so this includes K3 surfaces. It might seem weird that K3 surfaces haven’t been proven but the threefold case has been. This just has to do with those technicalities of what to do if the middle cohomology is bigger than 2-dimensional, which it always is. There you have it. The famous Taniyama-Shimura conjecture which led to a proof of Fermat’s Last Theorem.


Taniyama-Shimura 3: L-Series

For today, we assume our {d}-dimensional variety {X/\mathbb{Q}} has the property that its middle etale cohomology is 2-dimensional. It won’t hurt if you want to just think that {X} is an elliptic curve. We will first define the L-series via the Galois representation that we constructed last time. Fix {p} a prime not equal to {\ell} and of good reduction for {X}. Let {M=\overline{\mathbb{Q}}^{\ker \rho_X}}. By definition the representation factors through {{Gal} (M/\mathbb{Q})}. For {\frak{p}} a prime lying over {p} the decomposition group {D_{\frak{p}}} surjects onto {{Gal} (\overline{\mathbf{F}}_p/\mathbf{F}_p)} with kernel {I_{\frak{p}}}. One of the subtleties we’ll jump over to save time is that {\rho_X} acts trivially on {I_{\frak{p}}} (it follows from the good reduction assumption), so we can lift the generator of {{Gal} (\overline{\mathbf{F}}_p/\mathbf{F}_p)} to get a conjugacy class {{Frob}_p} whose image under {\rho_X} has well-defined trace and determinant.

We define

\displaystyle L(X,s) =(*)\prod_{p \ good} \frac{1}{1-{tr}(\rho_X({Frob}_p))p^{-s}+\det(\rho_X({Frob}_p))p^{-2s}}

where {(*)} is a product of terms at the bad primes. Note that since this is a two-dimensional representation basic linear algebra tells us that the product is over the simpler expression {(\det(I-\rho_X({Frob}_p)p^{-s}))^{-1}}.

If you don’t like all this Galois representation stuff, we can describe this L-series without reference to the Galois representation at all. In order to ease notation we will denote the reduction of {X} at a fixed good prime {p} by {Y:=X_{\mathbf{F}_p}} and base changing to the algebraic closure {\overline{Y}:=X_{\overline{\mathbf{F}_p}}}. To simplify notation let {k=\mathbf{F}_p}.

We have several natural Frobenius actions on {\overline{Y}}. The first we will call the absolute Frobenius which we will denote {F_{ab}:\overline{Y}\rightarrow \overline{Y}}. This is the identity on the topological space and the {p}-th power map on the structure sheaf. On affine patches {{Spec} A\rightarrow {Spec} \overline{k}} the map is the one induced by {a\mapsto a^p} on {A}. We can check directly that the map on topological spaces is the identity. For any prime ideal {\frak{q}\in{Spec} A} the contraction {\frak{q}^c=\{a\in A: a^p\in \frak{q}\}=\{a\in A: a\in\frak{q}\}=\frak{q}} by the property of {\frak{q}} being prime. This map translates in the language of schemes to {(id, F): (\overline{Y}, \mathcal{O}_{\overline{Y}})\rightarrow (\overline{Y}, \mathcal{O}_{\overline{Y}})} where {F} is raising sections of the sheaf to the {p}-th power.

Note that the absolute Frobenius is not a map of {\overline{Y}} over {\overline{k}}. The map is also not the pullback despite making the same commutative diagram.

Let the standard structure map be {\phi: \overline{Y}\rightarrow {Spec}\overline{k}} and define {\overline{Y}^{(p)}=\overline{Y}\otimes\overline{k}} to be the pullback of Frobenius acting on the base field. Since we have a commutative diagram we get by the universal property of a pullback diagram some map {F_r: \overline{Y}\rightarrow \overline{Y}^{(p)}} called the relative Frobenius. We define the arithmetic Frobenius to be the projection on the first factor {F_{ar}:\overline{Y}^{(p)}\rightarrow \overline{Y}}. A nice exercise to see if you understand these would be to write down a big commuting diagram that relates all these. Due to wordpress constraints, I won’t actually do this.

Instead, we’ll do an example. Let {Y={Spec} k[t]} (recall that {k=\mathbf{F}_p}). This means that {\overline{Y}={Spec} \overline{k} [t]={Spec} (k[t]\otimes_k \overline{k})}. The descriptions in terms of the ring homomorphism that induces the map on the spectra are as follows. The absolute Frobenius is still just {f\mapsto f^p}. The relative Frobenius is {F_{ab}\otimes id}. Since the absolute raises elements of {k[t]} to the {p}, everything in {k} is fixed by this map, and on {\overline{k}} it is defined to be fixed. This means that the relative Frobenius only alters the {t} by {t\mapsto t^p}. This is sometimes referred to as “raising coordinates to the {p}-th power”. The arithemtic Frobenius does nothing to the {k[t]} part, but raises the {\overline{k}} coefficients to the {p}, so {\sum a_nt^n\mapsto \sum a_n^p t^n}. Likewise, the geometric Frobenius takes the {p}-th root of the coefficients.

Straightforward (but non-trivial) computations also give that the map that the absolute Frobenius induces on the \'{e}tale site is trivial. If we look at our diagram we see that {F_{ab}=F_{ar}\circ F_r}. Since the induced map on cohomology is contravariant this gives {F^*_r\circ F^*_{ar}=id}. This means that on cohomology {F^*_r=(F^*_{ar})^{-1}=F^*_{ge}} by definition of the geometric Frobenius.

Now the smooth, proper base change theorem for \'{e}tale cohomology tells us that {H^d_{\textit{\'{e}t}}(\overline{Y}, \mathbb{Q}_\ell)\simeq H^d_{\textit{\'{e}t}}(\overline{X}, \mathbb{Q}_\ell)} which is two-dimensional. Since the {F_r} action here is a linear operator on a vector space it makes sense to take the trace and determinant. We can define the L-series without use of the Galois representation as:

\displaystyle L(X,s)=(*)\prod_{p \ good} \frac{1}{1-{tr}(F_r^*)p^{-s}+\det(F_r^*)p^{-2s}}

where again the {(*)} is a product of terms involving primes of bad reduction. Since there are only finitely many this is irrelevant for the definition of modularity. Of course we could have defined this without all the different Frobenius actions (we only used the relative one), but now we can get to the punchline. These two L-series are actually the same.
We just sketched above that the action of {F_r} and {F_{ge}} were the same on the \'{e}tale site. But {F_{ge}=1\times {Frob}_p^{-1}} where {{Frob}_p} is the canonical generator of {{Gal} (\overline{k}/k)}. We have a surjection {{Gal} (\overline{\mathbb{Q}}/\mathbb{Q})\rightarrow {Gal} (\overline{k}/k)} and if we consider {{Frob}_p} a lift of this element, then by the functoriality and equivariant isomorphisms above we get that {{tr}(\rho_X({Frob}_p))={tr}(\rho_X'({Frob}_p^{-1}))={tr}(F_r^*)}. The determinant term turns out to always be {p^3} since it can be checked to be the third power of the {\ell}-adic cyclotomic character in both cases. Thus the two L-series are the same. This also tells us the representation is odd.

Note that they appear to be off by an inverse, but we actually took the contragredient representation of the one that acts on {H^d_{\textit{\'{e}t}}(\overline{X}, \mathbb{Q}_\ell)}, so the inverse corrects for this and they are actually the same.

This got a little technical at parts, so the one thing to take away from this post is that to any {X/\mathbb{Q}} with two-dimensional middle cohomology we can produce some function which is just defined in terms of the trace and determinant of certain operators on the cohomology. This is called the L-series and will be crucial in the definition of modularity.

Taniyama-Shimura 2: Galois Representations

Fix some proper variety {X/\mathbb{Q}}. Our goal today will seem very strange, but it is to explain how to get a continuous representation of the absolute Galois group of {\mathbb{Q}} from this data. I’m going to assume familiarity with etale cohomology, since describing Taniyama-Shimura is already going to take a bit of work. To avoid excessive notation, all cohomology in this post (including the higher direct image functors) are done on the etale site.

For those that are intimately familiar with etale cohomology, we’ll do the quick way first. I’ll describe a more hands on approach afterwards. Let {\pi: X\rightarrow \mathrm{Spec} \mathbb{Q}} be the structure morphism. Fix an algebraic closure {v: \mathrm{Spec} \overline{\mathbb{Q}}\rightarrow \mathrm{Spec}\mathbb{Q}} (i.e. a geometric point of the base). We’ll denote the base change of {X} with respect to this morphism {\overline{X}}. Suppose the dimension of {X} is {n}.

Let {\ell} be a prime. We consider the constructible sheaf {R^n\pi_*(\mathbb{Z}/\ell^m)}. Now we have an equivalence of categories between these sheaves and continuous {G=Gal(\overline{\mathbb{Q}}/\mathbb{Q})}-modules by taking the stalk at our geometric point. Thus {R^n\pi_*(\mathbb{Z}/\ell^m)_v\simeq H^n(\overline{X}, \mathbb{Z}/\ell^m)} has a continuous action of {G} on it, and hence we get a continuous representation {\rho_{X,m}: G\rightarrow Aut(H^n(\overline{X}, \mathbb{Z}/\ell^m)\simeq GL_d(\mathbb{Z}/\ell^m)}. These all form a compatible family and hence we can take the inverse limit and tensor with {\mathbb{Q}_\ell} to get what is known as an {\ell}-adic Galois representation {\rho_X: G\rightarrow GL_d(\mathbb{Q}_\ell)}. For a technicality that will come up later, we will abuse notation and now relabel {\rho_X} to be the dual (or contragredient) representation.

If you aren’t comfortable with etale cohomology, then you can just use it as a black box cohomology theory to get the same thing as follows. First take the base change {\overline{X}\rightarrow \mathrm{Spec} \overline{\mathbb{Q}}}. Given any element of the Galois group {\sigma \in G} we get an automorphism of {\overline{\mathbb{Q}}}. Thus we can fill in the diagram:

{\begin{matrix} \overline{X} & \stackrel{\sigma}{\rightarrow} & \overline{X} \\ \downarrow & & \downarrow \\ \mathrm{Spec} \overline{\mathbb{Q}} & \stackrel{\sigma}{\rightarrow} & \mathrm{Spec} \overline{\mathbb{Q}} \end{matrix}}

Since {\sigma} was an automorphism, then only thing you have to believe about cohomology is that you then get an isomorphism via pullback {H^n(\overline{X}, \mathbb{Q}_\ell)\stackrel{\sigma^*}{\rightarrow} H^n(\overline{X}, \mathbb{Q}_\ell)}. Thus we get a continuous group homomorphism {G\rightarrow Aut(H^n(\overline{X}, \mathbb{Q}_\ell))} as before. Again, we’ll actually use the dual of this in the future.

To return to an elliptic curve {E} over {\mathbb{Q}}, we know that these are just tori, and hence the first Betti number is {2}. In this case we get that our Galois representation {\rho_E: G\rightarrow GL_2(\mathbb{Q}_\ell)}. If you’ve seen Taniyama-Shimura explained before, this should look familiar. This turns out to be exactly the same representation as the one you get from the Galois action on the Tate module. But the definition of the Tate module requires a group law, and hence the ability to get such a representation doesn’t generalize to all varieties in the way that using middle \ell-adic cohomology does. This is the standard modern approach to defining modularity for other types of varieties.

Taniyama-Shimura 1

It’s time to return to plan A. I started this year by saying I’d post on some fundamental ideas in arithmetic geometry. The local system thing is hard to get motivated about, since the way I was going to use it in my research seems irrelevant at the moment. My other option was to blog some stuff about class field theory, since there is a reading group on the topic that I belong to this quarter.

The first goal of this new series is to understand the statement of the famous Taniyama-Shimura conjecture that led to the proof of Fermat’s Last Theorem. A lot of people can probably mumble something about the conjecture if they have any experience in algebraic/arithmetic geoemtry or any of the number theory type fields, but most people probably can’t say anything precise about what the conjecture says (I’ll continue to call it a “conjecture” even though it has been proved).

The statement of the conjecture is that every elliptic curve over {\mathbb{Q}} is modular. Simple enough, but to unravel what it means to be modular we are going to have to take many posts just for the definition. If you’ve seen this explained before, it might still be interesting to read this series because I’m going to set up the machinery in a slightly different (but equivalent) way so that it will generalize to varieties other than elliptic curves in the future.

We’ll first define modular forms. A modular form of weight {k} and level {N} is an element of the vector space {M_k(\Gamma_0(N))} which consists of holomorphic functions on the upper half plane {f:\mathcal{H}\rightarrow \mathbf{C}} satisfying the additional transformation property that

\displaystyle \displaystyle f \left(\frac{az+b}{cz+d}\right)=(cz+d)^kf(z)

for all matrices {\left(\begin{matrix} a & b \\ c & d \end{matrix}\right)\in SL_2(\mathbf{Z})} such that {c\equiv 0 \mod N} (plus something else that we’ll get to shortly).

This is an analytic object if there ever was one. If this is the first time you’ve seen this, then the thing to pay attention to is that these depend on a choice of weight, {k}, and level, {N}. To get a feel for the level, note that it becomes “easier” to satisfy this transformation law as the level increases, because the amount of matrices we have to check is less. For example, when {N=1} this says our {f} has to behave nicely under every single linear fractional transformation that sends the upper half plane to the upper half plane. One might reasonably guess that {0} is the only holomorphic function with this property. More on this later. The weight is a little harder to get a feel for.

The map {z\mapsto e^{2\pi i z}} is a holomorphic map from the upper half plane onto the punctured unit disk. Note that {e^{2\pi i z}\rightarrow 0} as {z} tends to infinity along the imaginary axis. We can compose with this map and consider our modular form to be a holomorphic function on the punctured disk. This is well-defined because if {e^{2\pi i z}=e^{2\pi i w}}, then {z} and {w} differ by an integer and {f(z+n)=f\left(\left(\begin{matrix}1 & n \\ 0 & 1 \end{matrix}\right)\cdot z\right)=(1)^kf(z)=f(z)}.

We say {f} extends to be holomorphic at infinity if there is a holomorphic extension to the whole disk. We require modular forms to have this property. Thus a modular form has a Fourier expansion called a {q}-expansion denoted

\displaystyle \displaystyle f=\sum_{n=0}^\infty a_nq^n \ \text{where} \ q=e^{2\pi i z}

(note that a Fourier series in general involves negative powers, but these would give a pole at infinity). The cusp forms are the subspace denoted S_k(\Gamma_0(N)) of the modular forms that vanish at all cusps. To define cusp, just think of the extended upper half plane as {\mathbf{H}\cup \mathbb{P}^1_{\mathbb{Q}}}. We stick all the rational numbers along the real line in and also throw in a point at infinity. In practice, we only have to check holomorphic extension across finitely many of these cusps because due to the transformation law we only need to pick on cusp in each equivalence class under the action of the matrix group. When for instance N=1 again, all we have to check is that f vanishes at infinity, or upon composing to the disk we get that a_0=0.

Do any of these things exist? Well, as we’ve already noted, for small N it seems very hard to satisfy these properties. In fact, our guess was right, \dim_\mathbb{C} S_k(\Gamma_0(1))=0 for 1<k<12. So until we bump the weight up to 12, we actually only have the 0 function satisfying our properties. For weight 12, there is only one up to scalar multiple. This doesn't look good, but actually when we allow the level to grow we get a lot (even of low weight). But before next time, just ponder how severe the symmetry condition we are imposing is. Somehow every elliptic curve is closely related to one of these which is why the result is so surprising.

Now we have our basic analytic object of the conjecture. The next several posts will go back to the algebraic side of things. Depending on how much detail I decide to give to define the terms in the Taniyama-Shimura conjecture could take anywhere from 4 to 8 or so posts, just to give you an idea of how long you have to hold out for the statement.

2012 Blogging

It’s the start of a new year, so I’m going to start up something new here. My research interests have recently tended towards arithmetic geometry, so my plan for 2012 is to write some basics of algebraic number theory and arithmetic geometry that I use a lot. I’ll try to avoid redoing some of the stuff that has already been done at Climbing Mount Bourbaki.

I’d like to explain what modular forms are and what some of their basic properties are. I may detour a little into Galois representations at some point. I definitely want to talk about L-functions of varieties and what it means for a variety to be modular. This may lead to some discussion about Fermat’s Last Theorem and the Taniyama-Shimura conjecture. Scattered throughout I’ll probably have to cover some more classical algebraic number theory.

If you’re interested in related topics just post a comment and maybe I’ll get to it. Maybe in a few weeks I’ll scratch this whole idea and do something else. Who knows?