Reductive Groups Part II: Over More General Fields

In Reductive Groups Part I: Over Algebraically Closed Fields we learned about how reductive groups over algebraically closed fields are classified by their root datum, and how the based root datum helps us understand their automorphisms. In this post, we consider reductive groups over more general, not necessarily algebraically closed fields. We will discuss how they can be classified, and also define the Langlands dual of a reductive group, which will allow us to state the local Langlands correspondence (see The Local Langlands Correspondence for General Linear Groups) for certain groups other than just the general linear group.

Let F be a field. We will say an algebraic group G over F is a reductive group if G_{\overline{F}}, the base change of G to the algebraic closure \overline{F}, is a reductive group. Similarly, we say that G is a torus if the base change G_{\overline{F}} is a torus. This means that after base change to the algebraic closure it becomes isomorphic to the product of copies of the multiplicative group \mathbb{G}_{m}. However, if over F it is already isomorphic to the product of copies of \mathbb{G}_{m}, without the need for a base change, then we say that it is a split torus. If a reductive group G contains a maximal split torus, we say that G is split. We note that a “maximal split torus” is different from a “split maximal torus”!

The classification of split reductive groups is the same as that of reductive groups over algebraically closed fields – they are classified by their root datum. As such they will provide us with the first step towards classifying reductive groups over more general fields.

If G is a reductive group over F, a form of G is some other reductive group G' over F such that after base change to the algebraic closure \overline{F}, G_{\overline{F}} and G_{\overline{F}} are isomorphic. It happens that any reductive group is a form of a split reductive group. This follows from the fact that any abstract root datum is the root datum associated to some reductive group and some split maximal torus contained in it.

The forms of a split group are classified using Galois cohomology. Suppose we have an isomorphism f:G_{\overline{F}}\simeq G_{\overline{F}}. The Galois group \mathrm{Gal}(\overline{F}/F) (henceforth shortened to just \mathrm{Gal}_{F}) acts on the isomorphism f by conjugation, giving rise to another isomorphism ^{\sigma}f:G_{\overline{F}}\simeq G_{\overline{F}}'. Composing this with the inverse of f we get an automorphism f^{-1}\circ^{\sigma}f of G_{\overline{F}}. This automorphism is an example of a 1cocycle in Galois cohomology.

More generally, in Galois cohomology, for some group M with a Galois action (for instance in our case M=\mathrm{Aut}(G)_{\overline{F}})), a 1-cocycle is a homomorphism \varphi:\mathrm{Gal}_{F}\to M such that \varphi(\sigma\tau)=\varphi(\sigma)\cdot^{\sigma}\varphi(\tau). Two 1-cocycles \varphi, \psi are cohomologous if there is an element m\in M such that \psi(\sigma)=m^{-1}\varphi(\sigma)^{\sigma}m. The set of 1-cocycles, modulo those which are cohomologous, is denoted H^{1}(\mathrm{Gal}_{F},M).

By the above construction there is a map between the set of isomorphism classes of forms of G and the Galois cohomology group H^{1}(\mathrm{Gal}_{F},\mathrm{Aut}(G_{\overline{F}})). This map actually happens to be a bijection!

Let BR be a based root datum corresponding to G together with a pinning. We have mentioned in Reductive Groups Part I: Over Algebraically Closed Fields the group of automorphisms of BR, the pinned automorphisms of G, and the outer automorphisms of G are all isomorphic to each other. We have the following exact sequence

\displaystyle 0\to\mathrm{Inn}(G_{\overline{F}})\to\mathrm{Aut}(G_{\overline{F}})\to\mathrm{Out}(G_{\overline{F}})\to 0

and when we are provided the additional data of a pinning this gives us a splitting of the exact sequence (i.e. a way to decompose the middle term into a semidirect product of the other two terms).

When the pinning is defined over F, we obtain a homomorphism

\displaystyle H^{1}(\mathrm{Gal}_{F},\mathrm{Aut}(G_{\overline{F}}))\to H^{1}( \mathrm{Gal}_{F}, \mathrm{Out}(G_{\overline{F}}))

where H^{1}(\mathrm{Gal}_{F}, \mathrm{Out}(G_{\overline{F}})) is in bijection with the set of conjugacy classes of group homomorphisms \mu:\mathrm{Gal}_{F}\to \mathrm{Out}(G_{\overline{F}}). But we have said earlier that H^{1}(\mathrm{Gal}_{F},\mathrm{Aut}(G_{\overline{F}})) is in bijection with the set of isomorphism classes of forms of G. Therefore, any form of G gives us such a homomorphism \mu:\mathrm{Gal}_{F}\to \mathrm{Out}(G_{\overline{F}}).

We say that a reductive group is quasi-split if it contains a Borel subgroup. Split reductive groups are automatically quasi-split.

An inner form of a reductive group G is another reductive group G' related by an isomorphism f:G_{\overline{F}}\simeq G_{\overline{F}}' such that the composition f^{-1}\circ^{\sigma}f is an inner automorphism of G_{\overline{F}}.

Once we have a split group G, and given the data of a pinning, we can now use any morphism \mu:\mathrm{Gal}_{F}\to\mathrm{Out}(G) together with the given pinning to obtain an element of H^{1}(\mathrm{Gal}_{F},\mathrm{Aut}(G_{\overline{F}})), which in turn will give us a quasi-split form of G. Now it happens that any reductive group G has a unique quasi-split inner form!

Therefore, in summary, the classification of reductive groups over general fields proceeds in the following three steps:

  1. Classify the split reductive groups using the root datum.
  2. Classify the quasi-split forms using the homomorphisms \mathrm{Gal}_{F}\to\mathrm{Out}(G).
  3. Classify the inner forms of the quasi-split forms.

Let us now discuss the Langlands dual (also known as the L-group) of a reductive group. Since every abstract root datum corresponds to some reductive group G (say, over a field F), we can interchange the roots and coroots and get another reductive group \widehat{G}, which we refer to as the dual group of G.

The Langlands dual of G is the group (an honest to goodness group, not an algebraic group) given by the semidirect product \widehat{G}(\mathbb{C})\rtimes \mathrm{Gal}_{F}. In order to construct this semidirect product we need an action of \mathrm{Gal}_{F} on \widehat{G}(\mathbb{C}), and in this case this action is via its action on the based root datum of \widehat{G} together with a Borel subgroup B\subseteq G, which is the same as a pinned automorphism of \widehat{G}. We denote the Langlands dual of G by ^{L}G.

Let us recall that in The Local Langlands Correspondence for General Linear Groups we stated the local Langlands correspondence, in the case of \mathrm{GL}_{n}(F) where F is a local field, as a correspondence between the irreducible admissible representations of \mathrm{GL}_{n}(F) (over \mathbb{C}) and the F-semisimple Weil-Deligne representations of the Weil group W_{F} of F.

With the definition of the Langlands dual in hand, we can now state the local Langlands correspondence more generally, not just for \mathrm{GL}_{n}(F), and in this case it will not even be a one-to-one correspondence between irreducible admissible representations and Weil-Deligne representations anymore!

First, we will need the notion of a Langlands parameter, also called an L-parameter, which takes the place of the F-semisimple Weil-Deligne representation. It is defined to be a continuous homomorphism W_{F}\times \mathrm{SL}_{2}(\mathbb{C})\to ^{L}G such that, as a homomorphism from W_{F} to ^{L}G, it is semisimple, the composition W_{F}\to^{L}G\to\mathrm{Gal}_{F} is just the usual inclusion of W_{F} into \mathrm{Gal}_{F}, and as a function of \mathrm{SL}_{2}(\mathbb{C}) to \widehat{G}(\mathbb{C}) it comes from a morphism of algebraic groups from \mathrm{SL}_{2} to \widehat{G}.

And now for the statement: The local Langlands correspondence states that, for a reductive group G over a local field F, the irreducible admissible representations of G(F) are partitioned into a finite disjoint union of sets, called L-packets, labeled by (equivalence classes of, where the equivalence is given by conjugation by elements of \widehat{G}(\mathbb{C})) L-parameters. In other words, letting \mathrm{Irr}_{G} be the set of isomorphism classes of irreducible admissible representations of G, and letting \Phi be the set of equivalence classes of L-parameters, we have

\mathrm{Irr}_{G}=\coprod_{\phi\in\Phi}\Pi_{\phi}

where \Pi_{\phi} is the L-packet, a set of irreducible admissible representations of G(F). In the case that F is p-adic and G=\mathrm{GL}_{n}, each of these L-packets have only one element and this reduces to the one-to-one correspondence which we saw in The Local Langlands Correspondence for General Linear Groups.

The L-group and L-parameters are also expected to play a part in the global Langlands correspondence (in the case of function fields over a finite field, the construction of L-parameters was developed by Vincent Lafforgue using excursion operators). There is also much fascinating theory connecting the representations of the L-group to the geometry of a certain geometric object constructed from the original reductive group called the affine Grassmannian. We will discuss more of these topics in the future.

References:

Reductive group on Wikipedia

Root datum on Wikipedia

Inner form on Wikipedia

Langlands dual group on Wikipedia

Topics in automorphic forms (notes by Chao Li from a course by Jack Thorne)

Representation theory and number theory (notes by Chao Li from a course by Benedict Gross)

Algebraic groups by J. S. Milne

Formal Schemes

In Galois Deformation Rings we discussed the dual numbers as well as the concept of deformations. The dual numbers are numbers with an additional “tangent direction” or a “derivative” – we can further take into account higher order derivatives to consider deformations, which leads us to the concept of deformations.

In this post, we will consider spaces related to deformations, called formal schemes. Let us begin with a motivating example. Consider a field k. From algebraic geometry, we know that the underlying topological space of the scheme \mathrm{Spec}(k) is just a single point. What about the dual numbers, which is the ring k[x]/(x^{2}). What is the underlying topological space of the scheme \mathrm{Spec}(k[x]/(x^{2}))? It turns out it is also just the point!

So as far as the underlying topological spaces go, there is no difference between \mathrm{Spec}(k) and \mathrm{Spec}(k[x]/(x^{2})) – they are both just points. This is because they both have one prime ideal. For k this is the ideal (0), which is also its only ideal that is not itself. For k[x]/(x^{2}), its one prime ideal is the ideal (x); note that the ideal (0) in k[x]/(x^{2}) is not prime, which is related to this ring not being an integral domain. However a scheme is more than just its underlying topological space, one also has the data of its structure sheaf, i.e. its ring of regular functions, and in this regard \mathrm{Spec}(k) and \mathrm{Spec}(k[x]/(x^{2})) are different.

We sometimes consider \mathrm{Spec}(k[x]/(x^{2})) as a “thickening” of \mathrm{Spec}(k) – they are both just the point, but the functions on \mathrm{Spec}(k[x]/(x^{2})) have derivatives, as if there were tangent directions on the point that is the underlying space of \mathrm{Spec}(k[x]/(x^{2})) on which one can move “infinitesimally”, unlike on the point that is the underlying space of \mathrm{Spec}(k).

Just like in our discussion in Galois Deformation Rings, we may want to consider not only the “first-order derivatives” which appear in the dual numbers but also “higher-order derivatives”; we may even want to consider all of them together. This amounts to taking the inverse limit \varprojlim_{n}k[x]/(x^{n}), which is the formal power series ring k[[x]]. However, if we take \mathrm{Spec}(k[[x]]), we will see that it actually has two points, a “generic point” (corresponding to the ideal (0), which is prime because k[[x]] is an integral domain) and a “special point” (corresponding to the ideal (x), which is also prime and furthermore the lone maximal ideal), unlike \mathrm{Spec}(k) or \mathrm{Spec}(k[x]/(x^{2})) (or more generally \mathrm{Spec}(k[x])/(x^{n})), for any n, justified by similar reasons to the preceding argument).

This is where formal schemes come in – a formal scheme can express the “thickening” of some other scheme, with all the “higher-order derivatives”, where the underlying topological space is the same as that of the original scheme but the structure sheaf might be different, to reflect this “thickening”.

A topological ring is a ring A equipped with a topology such that the usual ring operations are continuous with respect to this topology. In this post we will mostly consider the I-adic topology, for some ideal I called the ideal of definition. This topology is generated by a basis consisting of sets of the form a+I^{n}, for a in A.

An example of a topological ring with the I-adic topology is the formal power series ring k[[x]] which we have discussed, together with the ideal of definition (x). Another example is the p-adic integers \mathbb{Z}_{p}, together with the ideal of definition (p). We note that all these examples are complete with respect to the I-adic topology.

More generally for higher dimension one can take, say, an affine variety cut out by some polynomial equation, say y^{2}=x^{3}, and consider the ring k[x,y]/(y^{2}-x^{3}). Note that \mathrm{Spec}(k[x,y]/(y^{2}-x^{3})) is an affine variety. Now we can form a topological ring complete with respect to the I-adic topology, by taking the completion with respect to the ideal I=(y^{2}-x^{3}), i.e. the inverse limit of the diagram

\displaystyle k[x,y]/(y^{2}-x^{3})\leftarrow  k[x,y]/(y^{2}-x^{3})^{2} \leftarrow  k[x,y]/(y^{2}-x^{3})^{3} \leftarrow\ldots

The formation of this ring is an important step in describing the “thickening” of the affine variety \mathrm{Spec}(k[x,y]/(y^{2}-x^{3})), but as said above, it cannot just be done by taking the “spectrum”. Therefore we introduce the idea of a formal spectrum.

Let A be a Noetherian topological ring and let I be an ideal of definition. In addition, let A be complete with respect to the I-adic topology. We define the formal spectrum of A, denoted \mathrm{Spf}(A), to be the pair (X,\mathcal{O}_{X}), where X is the underlying topological space of \mathrm{Spec}(A/I), and the structure sheaf \mathcal{O}_{X} is defined by setting \mathcal{O}_{X}(D_{f}) to be the I-adic completion of A[1/f], for D_{f} the distinguished open set corresponding to f. Applied to the examples earlier, this gives us what we want – a sort of “thickening” of some affine scheme, with an underlying topological space the same as that of the original scheme but with a structure sheaf of functions with “higher-order derivatives”.

More generally, to include the “non-affine” case a formal scheme is a topologically ringed space, i.e. a pair (X,\mathcal{O}_{X}) where X is a topological space and \mathcal{O}_{X} is a sheaf of topological rings, such that for any point x of X there is an open neighborhood of x which is isomorphic as a topologically ringed space to \mathrm{Spf}(A) for some Noetherian topological ring A.

Aside from being useful in deformation theory, formal schemes are also related to rigid analytic spaces (see also Rigid Analytic Spaces). For certain types of formal schemes (“locally formally of finite type over \mathrm{Spf}(\mathbb{Z}_{p})“) one can assign (functorially) a rigid analytic space. For example, this functor will assign to the formal scheme \mathrm{Spf}(\mathbb{Z}_{p}[[x]]) the open unit disc (the interior of the closed unit disc in Rigid Analytic Spaces). This functor is called the generic fiber functor, which is an odd name, because \mathrm{Spf}(\mathbb{Z}_{p}[[x]]) has no “generic points”! However, there is a way to make this name make more sense using the language of adic spaces, which also subsumes the theory of both formal schemes and rigid analytic spaces, and also provides a natural home for the perfectoid spaces we hinted at in Perfectoid Fields. The theory of adic spaces will hopefully be discussed in some future post on this blog.

References:

Formal Scheme on Wikipedia

Berkeley Lectures on p-adic Geometry by Peter Scholze and Jared Weinstein

Modular Curves at Infinite Level by Jared Weinstein

Adic Spaces by Jared Weinstein

Algebraic Geometry by Robin Hartshorne

p-adic Hodge Theory: An Overview

In The Theory of Motives we discussed the notion of a Weil cohomology, and mentioned four “classical” examples, the singular (also known as Betti) cohomology, the de Rham cohomology, the \ell-adic cohomology, and the crystalline cohomology.

Cohomology theories may be thought of as a way to study geometric objects using linear algebra, by associating vector spaces (or more generally, modules or abelian groups) to such a geometric object. But the four Weil cohomology theories above actually give more than just a vector space:

  • The singular cohomology has an action of complex conjugation.
  • The de Rham cohomology has a Hodge filtration.
  • The \ell-adic cohomology has an action of the Galois group.
  • The crystalline cohomology has an action of Frobenius (and a Hodge filtration as well).

There are relations between these different cohomologies. For example, for a smooth projective variety X over the complex numbers \mathbb{C}, the singular cohomology of the corresponding complex analytic manifold X(\mathbb{C}), with complex coefficients (this can be obtained from singular cohomology with integral coefficients by tensoring with \mathbb{C}) and the de Rham cohomology are isomorphic:

\displaystyle H_{\mathrm{sing}}^{k}(X(\mathbb{C}),\mathbb{Z})\otimes_{\mathbb{Z}}\mathbb{C}=H_{\mathrm{dR}}^{k}(X)

The roots of this idea go back to de Rham’s work on complex manifolds, where chains in singular homology (which is dual to singular cohomology, see also Homology and Cohomology) can be paired with the differential forms of de Rham cohomology (see also Differential Forms), simply by integrating the differential forms along these chains. By the machinery developed by Alexander Grothendieck, this can be ported over into the world of algebraic geometry.

Again borrowing from the world of complex manifolds, the machinery of Hodge theory gives us the following Hodge decomposition (see also Shimura Varieties):

\displaystyle H_{\mathrm{sing}}^{k}(X(\mathbb{C}),\mathbb{Z})\otimes_{\mathbb{Z}}\mathbb{C}=\bigoplus_{i+j=k} H^{i}(X,\Omega_{X/\mathbb{C}}^{j})

Now again for the case of smooth projective varieties over the complex numbers , \ell-adic cohomology also has such an isomorphism with singular cohomology – but this time if it has \ell-adic coefficients (i.e. in \mathbb{Q}_{\ell}).

\displaystyle H_{\mathrm{sing}}^{k}(X(\mathbb{C}),\mathbb{Z})\otimes_{\mathbb{Z}}\mathbb{Q}_{\ell}\simeq H_{\mathrm{et}}^{k}(X,\mathbb{Q}_{\ell})

Such isomorphisms are also known as comparison isomorphisms (or comparison theorems).

More generally, if we have a field B into which we can embed both \mathbb{Q}_{\ell} and \mathbb{C} (for instance \mathbb{C}), we obtain the following comparison theorem:

\displaystyle  H_{\mathrm{et}}^{k}(X,\mathbb{Q}_{\ell}) \otimes_{\mathbb{Q_{\ell}}} B\simeq  H_{\mathrm{dR}}^{k}(X) \otimes_{\mathbb{C}} B

Here is a very interesting thing that these comparison theorems can give us. Let X be a modular curve. Then the Hodge decomposition for the first cohomology gives us

\displaystyle H_{\mathrm{sing}}^{1}(X(\mathbb{C}),\mathbb{Z})\otimes_{\mathbb{Z}}\mathbb{C}=H^{1}(X,\Omega_{X/\mathbb{C}}^{0})\oplus H^{0}(X,\Omega_{X/\mathbb{C}}^{1})

But the H^{0}(X,\Omega_{X/\mathbb{C}}^{1}) is the cusp forms of weight 2 as per the discussion in Modular Forms (see also Galois Representations Coming From Weight 2 Eigenforms). By the results of Hodge theory, the other summand H^{1}(X,\Omega_{X/\mathbb{C}}^{0}) is just the complex conjugate of H^{0}(X,\Omega_{X/\mathbb{C}}^{1}). But we now also have a comparison with etale cohomology, which has a Galois representation! For this the modular form must lie in the cohomology with \mathbb{Q} coefficients, which happens if it is a Hecke eigenform whose Hecke eigenvalues are in \mathbb{Q}. So one of the great things that these comparison theorems gives us is this way of relating modular forms and Galois representations.

The comparison isomorphisms above work for smooth projective varieties over the complex numbers, but let us now go to the p-adic world, and let us consider smooth projective varieties over the p-adic numbers.

It was observed by John Tate (and later explored by Gerd Faltings) that the p-adic cohomology (i.e. the etale cohomology of a smooth projective variety over \mathbb{Q}_{p}, or more generally some other p-adic field, with p-adic coefficients, distinguishing it from \ell-adic cohomology where another prime \ell different from p must be brought in) can have a decomposition akin to the Hodge decomposition, after tensoring it with the p-adic complex numbers (this is the completion of the algebraic closure of the p-adic numbers):

\displaystyle H^{k}(X_{\overline{\mathbb{Q}}_{p}},\mathbb{Q}_{p})\otimes_{\mathbb{Q}_{p}}\mathbb{C}_{p}=\bigoplus_{i+j=k} H^{i}(X,\Omega_{X/\mathbb{Q}}^{j})\otimes_{\mathbb{Q}}\mathbb{C}_{p}(-j)

The p-adic complex numbers here play the role of the complex numbers in the singular cohomology case above or the \ell-adic numbers in the \ell-adic case.

The ideas conjectured by Tate, and later completed by Faltings, was but the prototype of what is now known as p-adic Hodge theory. In its modern form, p-adic Hodge theory concerns comparison isomorphisms between different Weil cohomology theories on smooth projective varieties over the p-adic numbers. However, the role played by the complex numbers, \ell-adic numbers (for the complex case), and p-adic complex numbers (for the p-adic case) must now be played by much more complicated objects called period rings, which were developed by Jean-Marc Fontaine. We will discuss the construction of the period rings at the end of this post, but first let us see how they work.

Let X be a smooth projective variety over \mathbb{Q}_{p} (or more generally some other p-adic field). Let H_{\mathrm{dR}}^{i}(X) and H_{\mathrm{et}}^{i}(X_{\overline{\mathbb{Q}}_{p}},\mathbb{Q}_{p}) be its de Rham cohomology and the p-adic etale cohomology of its base change to the algebraic closure \overline{\mathbb{Q}}_{p} respectively. The comparison isomorphism at the center of p-adic Hodge theory is the following:

\displaystyle H_{\mathrm{dR}}^{i}(X)\otimes_{\mathbb{Q}_{p}} B_{\mathrm{dR}}=H_{\mathrm{et}}^{i}(X_{\overline{\mathbb{Q}}_{p}},\mathbb{Q}_{p})\otimes_{\mathbb{Q}_{p}} B_{\mathrm{dR}}

The object denoted B_{\mathrm{dR}} here is the aforementioned period ring. It is equipped with both a Galois action and a filtration akin to the Hodge filtration. More than just that isomorphism above, we also have a way of obtaining the de Rham cohomology if we are given the p-adic etale cohomology, simply by taking the part that is invariant under the Galois action:

\displaystyle  \displaystyle H_{\mathrm{dR}}^{i}(X)=(H_{\mathrm{et}}^{i}(X_{\overline{\mathbb{Q}}_{p}},\mathbb{Q}_{p})\otimes_{\mathbb{Q}_{p}} B_{\mathrm{dR}})^{\mathrm{Gal}_{\mathbb{Q}_{p}}}

To go the other way, i.e. to recover the p-adic etale cohomology from the de Rham cohomology, we will need a different kind of period ring. This period ring is B_{\mathrm{cris}}, which aside from having a Galois action and a filtration also has an action of Frobenius. Aside from providing us the same isomorphism between de Rham and p-adic etale cohomology upon tensoring, it also provides us with a solution to our earlier problem (as long as X has a smooth proper integral model) as follows:

\displaystyle H_{\mathrm{et}}^{i}(X_{\overline{\mathbb{Q}}_{p}},\mathbb{Q}_{p})= \mathrm{Fil}^{0}(H_{\mathrm{dR}}^{i}(X)\otimes_{\mathbb{Q}_{p}} B_{\mathrm{cris}})^{\varphi=1}

This idea can be further abstracted – since etale cohomology provides Galois representations, we can just take some p-adic Galois representation instead, without caring whether it comes from etale cohomology or not, and tensor it with a period ring, then take Galois invariants. For instance let V be some p-adic Galois representation. Then we can take the tensor product

V_{\mathrm{dR}}=(V\otimes B_{\mathrm{dR}})^{\mathrm{Gal}_{\mathbb{Q}_{p}}}

If the dimension of V_{\mathrm{dR}} is equal to the dimension of V, then we say that the Galois representation V is de Rham. Similarly we can tensor with B_{\mathrm{cris}}:

V_{\mathrm{cris}}=(V\otimes B_{\mathrm{cris}})^{\mathrm{Gal}_{\mathbb{Q}_{p}}}

If its V_{\mathrm{cris}} is equal to the dimension of V , we say that V is crystalline.

The idea of these “de Rham” and “crystalline” Galois representations is that if they come from the corresponding cohomologies then they will have these properties. But does the converse hold? If they are “de Rham” and “crystalline” does that mean that they come from the corresponding cohomologies (i.e. they “come from geometry”)? This is roughly the content of the Fontaine-Mazur conjecture.

Now let us say a few things about the construction of these period rings. These constructions make use of the concepts we discussed in Perfectoid Fields. We start with the ring A_{\mathrm{inf}}(\mathcal{O}_{\mathbb{C}_{p}}), which, as we recall from Perfectoid Fields, is the ring of Witt vectors of the tilt of \mathcal{O}_{\mathbb{C}_{p}}. By inverting p and taking the completion with respect to the canonical map \theta:  A_{\mathrm{inf}}(\mathcal{O}_{\mathbb{C}_{p}}) \to\mathcal{O}_{\mathbb{C}_{p}}, we obtain a ring which we suggestively denote by B_{\mathrm{dR}}^{+}.

There is a special element t of B_{\mathrm{dR}}^{+} which we think of as the logarithm of the element (1, \zeta^{1/p},\zeta^{1/p},\ldots). Upon inverting this element t, we obtain the field B_{\mathrm{dR}}.

The field B_{\mathrm{dR}} is equipped with a Galois action, carried over from the fields involved in its construction, and a filtration, given by \mathrm{Fil}^{i}B_{\mathrm{dR}}=t^{i}B_{\mathrm{dR}}.

To construct B_{\mathrm{cris}}, we once again start with A_{\mathrm{inf}}(\mathcal{O}_{\mathbb{C}_{p}}) and invert p. However, to have a Frobenius, instead of completing with respect to the kernel of the map \theta, we take a generator of this kernel (which we shall denote by \omega). Then we denote by B_{\mathrm{cris}}^{+} the ring formed by all the power series of the form \sum_{n=0}^{\infty} a_{n}\omega^{n}/n! where the a_{n}‘s are elements of A_{\mathrm{inf}}(\mathcal{O}_{\mathbb{C}_{p}})[1/p] which converge as n\to\infty, under the topology of A_{\mathrm{inf}}(\mathcal{O}_{\mathbb{C}_{p}})[1/p] (which is not the p-adic topology!). Once again there will be an element t like before; we invert t to obtain B_{\mathrm{cris}}.

There is yet another period ring called B_{\mathrm{st}}, where the subscript stands for semistable; in addition to a Galois action, filtration, and Frobenius, it has a monodromy operator. Since this is less extensively discussed in introductory literature, we follow this lead and leave this topic, and the many other wonderful topics related to p-adic Hodge theory, to future posts on this blog.

References:

p-adic Hodge theory on Wikipedia

de Rham cohomology on Wikipedia

Hodge theory on Wikipedia

An invitation to p-adic Hodge theory, or: How I learned to stop worrying and love Fontaine by Alex Youcis

Seminar on the Fargues-Fontaine curve (Lecture 1 – Overview) by Jacob Lurie

An introduction to the theory of p-adic representations by Laurent Berger

Reciprocity laws and Galois representations: recent breakthroughs Jared Weinstein

Reductive Groups Part I: Over Algebraically Closed Fields

In many posts on this blog we have talked about elliptic curves, which are examples of algebraic groups (which is itself a specific instance of a group scheme) – a variety whose points form a group. As varieties, elliptic curves (and abelian varieties in general) are projective. There are also algebraic groups which are affine, and in this post we will discuss a special class of affine algebraic groups, called reductive groups, which as we shall see are related to many familiar objects, and are well-studied. In particular, in the case when they are defined over algebraically closed fields we will discuss their classification in terms of root datum. We will also discuss how this root datum helps us understand the automorphisms of such a reductive group.

An example of a reductive group is \mathrm{GL}_{n}; let’s assume that this is a variety defined over some field F. If R is some F-algebra, then the R-valued points of \mathrm{GL}_{n} (in the “functor of points” point of view) is the group \mathrm{GL}_{n}(R) of n\times n matrices with nonzero determinant. Geometrically, we may think of the nonzero determinant condition as the polynomial equation that cuts out the variety \mathrm{GL}_{n}.

Linear algebraic groups are smooth closed algebraic subgroups of \mathrm{GL}_{n}, and they have their own “representation theory”, a “representation” in this context being a morphism from some linear algebraic group G to the algebraic group \mathrm{GL}(V), for some vector space V over some field E. The algebraic group \mathrm{GL}(V) is the algebraic group whose R-valued points give the group of linear transformations of the E-vector space R\otimes V.

A linear algebraic group is a reductive group if it is geometrically connected and every representation is semisimple (a direct product of irreducible representations).

We also denote the reductive group \mathrm{GL}_{1} by \mathbb{G}_{m}. A torus is a reductive group which is isomorphic to a product of copies of \mathbb{G}_m. A torus contained in a reductive group G is called maximal if it is not contained in some strictly larger torus contained in G.

Let T be a maximal torus of the reductive group G. The Weyl group (G,T) is the quotient N(T)/Z(T) where N(T) is the normalizer of T in G (the subgroup consisting of all elements g in G such that for any element t in T gtg^{-1} is an element of T) and Z(T) is the centralizer of T in G (the subgroup consisting of all elements in G that commute with all the elements of T).

Now let us discuss the classification of reductive groups, for which we will need the concept of roots and root datum.

For a maximal torus T in a reductive group G, the characters (homomorphisms from T to \mathbb{G}_{m}) and the cocharacters (homomorphisms from T to \mathbb{G}_{m}) will play an important role in this classification. Let us denote the characters of T by X^{*}(T), and the cocharacters of T by X_{*}(T).

Just like Lie groups, reductive groups have a Lie algebra (the tangent space to the identity), on which it acts (therefore giving a representation of the reductive group, called the adjoint representation). We may restrict to a maximal torus T contained in the reductive group G, so that the Lie algebra \mathfrak{g} of G gives a representation of T. This gives us a decomposition of \mathfrak{g} as follows:

\displaystyle \mathfrak{g}=\mathfrak{g}_{0}\oplus \bigoplus_{\alpha}\mathfrak{g}_{\alpha}

Here \mathfrak{g}_{\alpha} is the subspace of \mathfrak{g} on which T acts as a character \alpha:T\to\mathbb{G}_{m}. The nonzero characters \alpha for which \mathfrak{g}_{\alpha} is nonzero are called roots. We denote the set of roots by \Phi.

For a character \alpha, let T_{\alpha} be the connected component of the kernel of \alpha. Let G_{\alpha} be the centralizer of T_{\alpha} in G. Then the Weyl group W(G_{\alpha},T) will only have two elements, the identity and one other element, which we shall denote by s_{\alpha}. There will be a unique cocharacter \alpha^{\vee} satisfying the equation

s_{\alpha}(x)=x-\langle \alpha^{\vee},x\rangle\alpha

for all characters x:T\to\mathbb{G}_{m}. This cocharacter is called a coroot. We denote the set of coroots by \Phi^{\vee}.

The datum (\Phi, X^{*}(T), \Phi^{\vee}, X_{*}(T)) is called the root datum associated to G. This root datum is actually independent of the chosen maximal torus, which follows from all maximal tori being contained in a unique conjugacy class in G.

There is also a concept of an “abstract” root datum, a priori having seemingly nothing to do with reductive groups, just some datum (M, \Psi, M^{\vee}, \Psi^{\vee}) where M and M^{\vee} are finitely generated abelian groups, \Psi is a subset of M\setminus \lbrace 0\rbrace, and \Psi^{\vee} is a subset of M^{\vee}\setminus \lbrace 0\rbrace, and they satisfy the following axioms:

  • There is a perfect pairing \langle,\rangle:M\times M^{\vee}\to\mathbb{Z}.
  • There is a bijection between \Psi and \Psi^{\vee}.
  • For any \alpha\in \Psi, and \alpha^{\vee} its image in \Psi under the aforementioned bijection, we have \langle \alpha,\alpha^{\vee}\rangle=2.
  • For any \alpha\in \Psi, the automorphism of M given by x\mapsto \alpha-\langle x,\alpha^{\vee}\rangle\alpha preserves \alpha.
  • The subgroup of \mathrm{Aut}(M) generated by x\mapsto \alpha-\langle x,\alpha^{\vee}\rangle\alpha is finite.

Again, a priori, such a datum of finitely generated abelian groups and their subsets, satisfying these axioms, seems to have nothing to do with reductive groups. However, we have the following amazing theorem:

Any abstract root datum is the root datum associated to some reductive group.

For reductive groups over an algebraically closed field, the root datum classifies reductive groups:

Two reductive groups over an algebraically closed field have the same root datum if and only if they are isomorphic.

Let us now discuss how root datum helps us understand the automorphisms of a reductive group. For this we need to expand the information contained in the root datum.

A root basis is a subset of the roots such that any root can be expressed as a unique linear combination of the roots, where the integer coefficients are either all positive or all negative. A based root datum is given by (\Phi, X^{*}(T), S, \Phi^{\vee}, X_{*}(T), S^{\vee}), i.e. the usual root datum together with the additional datum of a root basis S.

The root datum already determines the reductive group G. What does the additional data of a root basis mean? The root basis corresponds to a Borel subgroup of G that contains our chosen maximal torus T. A Borel subgroup of G is a maximal connected solvable Zariski closed algebraic subgroup of G.

A pinning is the datum (G,T,B,\lbrace x_{\alpha}\rbrace_{\alpha\in S}) where T is a maximal torus, B is a Borel subgroup containing T, and \lbrace x_{\alpha}\rbrace_{\alpha\in S} is a basis element of \mathfrak{g}_{\alpha}. Given a pinning, a pinned automorphism of G is an automorphism of G that preserves the pinning.

An inner automorphism of a group G is one that comes from conjugation by some element; in a way they are the automorphisms that are easier to understand. The inner automorphisms form a normal subgroup \mathrm{Inn}(G) of the group of automorphisms \mathrm{Aut}(G), and the quotient \mathrm{Aut}(G)/\mathrm{Inn}(G) is called \mathrm{Out}(G). We have similar notions for algebraic groups.

Now a pinned automorphism is an automorphism, therefore has a map to \mathrm{Out}(G). A pinned automorphism also has a map to the automorphisms of the corresponding based root datum. Both of these maps are actually isomorphisms! Therefore we have a description of \mathrm{Aut}(G) as follows:

The automorphisms of G as an algebraic group are given by the semidirect product of the inner automorphisms and the automorphisms of the based root datum.

In this post we have only focused on the case of reductive groups over algebraically closed fields. Over more general fields the theory of reductive groups, for instance the classification, is more complicated. This will hopefully be tackled in future posts on this blog.

References:

Algebraic group on Wikipedia

Linear algebraic group on Wikipedia

Reductive group on Wikipedia

Root datum on Wikipedia

Topics in automorphic forms (notes by Chao Li from a course by Jack Thorne)

Representation theory and number theory (notes by Chao Li from a course by Benedict Gross)

Lectures on the geometry and modular representation theory of algebraic groups by Geordie Williamson and Joshua Ciappara

Algebraic groups by J. S. Milne

Rigid Analytic Spaces

This blog post is inspired by and follows closely an amazing talk given by Ashwin Iyengar at the “What is a…seminar?” online seminar. I hope I can do the talk, and this wonderful topic, some justice in this blog post.

One of the most fascinating and powerful things about algebraic geometry is how closely tied it is to complex analysis, despite what the word “algebraic” might lead one to think. To state one of the more simple and common examples, we have that smooth projective curves over the complex numbers \mathbb{C} are the same thing as compact Riemann surfaces. In higher dimensions we also have Chow’s theorem, which tells us that an analytic subspace of complex projective space which is topologically closed is an algebraic subvariety.

This is all encapsulated in what is known as “GAGA“, named after the foundational work “Géometrie Algébrique et Géométrie Analytique” by Jean-Pierre Serre. We refer to the references at the end of this post for the more precise statement, but for now let us think of GAGA as giving us a fully faithful functor from proper algebraic varieties over \mathbb{C} to complex analytic spaces, which gives us an equivalence of categories between their coherent sheaves.

One may now ask if we can do something similar with the p-adic numbers \mathbb{Q}_{p} (or more generally an extension K of \mathbb{Q}_{p} that is complete with respect to a valuation that extends the one on \mathbb{Q}_{p}) instead of \mathbb{C}. This leads us to the theory of rigid analytic spaces, which was originally developed by John Tate to study a p-adic version of the idea (also called “uniformization”) that elliptic curves over \mathbb{C} can be described as lattices on \mathbb{C}.

Let us start defining these rigid analytic spaces. If we simply naively try to mimic the definition of complex analytic manifolds by having these rigid analytic spaces be locally isomorphic to \mathbb{Q}_{p}^{m}, with analytic transition maps, we will run into trouble because of the peculiar geometric properties of the p-adic numbers – in particular, as a topological space, the p-adic numbers are totally disconnected!

To fix this, we cannot just use the naive way because the notion of “local” would just be too “small”, in a way. We will take a cue from algebraic geometry so that we can use the notion of a Grothendieck topology to fix what would be issues if we were to just use the topology that comes from the p-adic numbers.

The Tate algebra \mathbb{Q}_{p}\langle T_{1},\ldots,T_{n}\rangle is the algebra formed by power series in n variables that converge on the n-dimensional unit polydisc D^{n}, which is the set of all n-tuples (c_{1},\ldots,c_{n}) of elements of \mathbb{Q}_{p} that have p-adic absolute value less than or equal to 1 for all i from 1 to n.

There is another way to define the Tate algebra, using the property of power series with coefficients in p-adic numbers that it converges on the unit polydisc D^{n} if and only if its coefficients go to zero (this is not true for real numbers!). More precisely if we have a power series

\displaystyle f(T_{1},\ldots,T_{n})=\sum_{a} c_{a}T_{1}^{a_{1}}\ldots T_{n}^{a_{n}}

where c_{a}\in \mathbb{Q}_{p} and a=a_{1}+\ldots+a_{n} runs over all n-tuples of natural numbers, then f converges on the unit polydisc D^{n} if and only if \lim_{a\to 0}c_{a}=0.

The Tate algebra has many important properties, for example it is a Banach space (see also Metric, Norm, and Inner Product) with the norm of an element given by taking the biggest p-adic absolute value among its coefficients. Another property that will be very important in this post is that it satisfies a Nullstellensatz – orbits of the absolute Galois group \mathrm{Gal}(\overline{\mathbb{Q}}_{p}/\mathbb{Q}_{p}) in D^{n} correspond to maximal ideals of the Tate algebra. Explicitly, this correspondence is given by the evaluation map x\in D^{n}\mapsto \lbrace f\vert f(x)=0\rbrace.

A quotient of the Tate algebra by some ideal is referred to as an affinoid algebra. The maximal ideals of the underlying set of an affinoid algebra A will be denoted \mathrm{Max}(A), and this should be reminiscent of how we obtain the closed points of a scheme.

Once we have the underlying set \mathrm{Max}(A) , we need two other things to be able to define “affinoid” rigid analytic spaces, which we shall later “glue” to form more general rigid analytic spaces – a topology, and a structure sheaf (again this should be reminiscent of how schemes are defined).

Again taking a cue from how schemes are defined, we define “rational domains”, which are analogous to the distinguished Zariski open sets of (the underlying topological space of) a scheme. Given elements f_{1},\ldots,f_{r},g of the affinoid algebra A, the rational domain \displaystyle A\left(\frac{f}{g}\right) is the set of all x\in\mathrm{Max}(A) such that f_{i}(x)\leq g(x) for all 1\leq i\leq r.

The rational domains generate a topology, however this is still just the p-adic topology, and so it still does not solve the problems that we originally ran into when we tried to define rigid analytic spaces by mimicking the definition of a complex analytic manifold. The trick will be to make use of something that is more general than just a topology – a Grothendieck topology (see also More Category Theory: The Grothendieck Topos).

Let us now define the particular Grothendieck topology that we will use. Unlike other examples of a Grothendieck toplogy, the covers will involve only subsets of the space being covered (it is also referred to as a mild Grothendieck topology). Let X=\mathrm{Max}(A). A subset U of X is called an admissible open if it can be covered by rational domains \lbrace U_{i}\rbrace_{i\in I} such that for any map Y\to X where Y=\mathrm{Max}(B) for some affinoid algebra B, the covering of Y given by the inverse images of the U_{i}‘s admit a finite subcover.

If U is an admissible open covered by admissible opens \lbrace U_{i}\rbrace_{i\in I}, then this covering is called admissible if for any map Y\to X whose image is contained in U, the covering of Y given by the inverse images of the U_{i}‘s admit a finite subcover. These admissible coverings satisfy the axioms for a Grothendieck topology, which we denote G_{X}.

If A is an affinoid algebra, and f_{1},\ldots,f_{k},g are functions, we let \displaystyle A\left\langle \frac{f}{g}\right\rangle denote the ring A\langle T_{1},\ldots T_{k}\rangle/(gT_{i}-f_{i}). By associating to a rational domain \displaystyle A\left(\frac{f}{g}\right) this ring \displaystyle A\left\langle\frac{f}{g}\right\rangle, we can define a structure sheaf \cal{O}_{X} on this Grothendieck topology.

The data consisting of the set X=\mathrm{Max}(A), the Grothendieck topology G_{X}, and the structure sheaf \mathcal{O}_{X} is what makes up an affinoid rigid analytic space. Finally, just as with schemes, we define a more general rigid analytic space as the data consisting of some set X, a Grothendieck topology G_{X} and a sheaf \mathcal{O}_{X} such that locally, with respect to the Grothendieck topology G_{X}, it is isomorphic to an affinoid rigid analytic space.

Under this definition, we have in fact a version of GAGA that holds for rigid analytic spaces – a fully faithful functor from proper schemes of finite type over \mathbb{Q}_{p} to rigid analytic spaces over \mathbb{Q}_{p} that gives an equivalence of categories between their coherent sheaves.

Finally let us now look at an example. Consider the affinoid rigid analytic space obtained from the affinoid algebra \mathbb{Q}_{p}\langle T\rangle. By the Nullstellensatz the underlying set is the unit disc D. The “boundary” of this is the rational subdomain (and therefore an admissible open) \displaystyle D\left(\frac{1}{T}\right), and its complement, the “interior” is covered by rational subdomains \displaystyle D\left(\frac{T^{n}}{p}\right). With this covering the interior may also be shown to be an admissible open.

While it appears that, since we found two complementary admissible opens, we can disconnect the unit disc, we cannot actually do this in the Grothendieck topology, because the set consisting of the boundary and the interior is not an admissible open! And so in this way we see that the Grothendieck topology is the difference maker that allows us to overcome the obstacles posed by the peculiar geometry of the p-adic numbers.

Since Tate’s innovation, the idea of a p-adic or nonarchimedean geometry has blossomed with many kinds of “spaces” other than the rigid analytic spaces of Tate, for example adic spaces, a special class of which generalize perfectoid fields (see also Perfectoid Fields) to spaces, or Berkovich spaces, which are honest to goodness topological spaces instead of relying on a Grothendieck topology like rigid analytic spaces do. Such spaces will be discussed on this blog in the future.

References:

Rigid analytic space on Wikipedia

Algebraic geometry and analytic geometry on Wikipedia

Several approaches to non-archimedean geometry by Brian Conrad

Reciprocity laws and Galois representations: recent breakthroughs by Jared Weinstein

p-adic families of modular forms [after Coleman, Hida, and Mazur] by Matthew Emerton

The Local Langlands Correspondence for General Linear Groups

The Langlands program is, roughly, a part of math that bridges the representation theory of reductive groups and Galois representations. It consists of many interrelated conjectures, among which are the global and local Langlands correspondences, which may be seen as higher-dimensional analogues of global and local class field theory (see also Some Basics of Class Field Theory) in some way. Just as in class field theory, the words “global” and “local” refer to the fact that these correspondences involve global and local fields, respectively, and their absolute Galois groups.

In this post we will discuss the local correspondence (the global correspondence requires the notion of automorphic forms, which involves a considerable amount of setup, hence is postponed to future posts). We will also only discuss the case of the general linear group \mathrm{GL}_{n}(F), for F a local field, and in particular \mathrm{GL}_{1}(F) and \mathrm{GL}_{2}(F) where we can be more concrete. More general reductive groups will bring in the more complicated notions of L-packets and L-parameters, and therefore also postponed to future posts.

Let us give the statement first, and then we shall unravel what the words in the statement mean.

The local Langlands correspondence for general linear groups states that there is a one-to-one correspondence between irreducible admissible representations of \mathrm{GL}_{n}(F) (over \mathbb{C}) and F-semisimple Weil-Deligne representations of the Weil group W_{F} (also over \mathbb{C}).

Let us start with “irreducible admissible representations of \mathrm{GL}_{n}(F)“. These representations are similar to what we discussed in Representation Theory and Fourier Analysis (in fact as we shall see, many of these representations are also infinite-dimensional, and constructed somewhat similarly). Just to recall, irreducible means that the only subspaces held fixed by \mathrm{GL}_{n}(F) are 0 and the entire subspace.

Admissible means that, if we equip \mathrm{GL}_{n}(F) with the topology that comes from the p-adic topology of the field F, for any open U subgroup of \mathrm{GL}_{n}(F) the fixed vectors form a finite-dimensional subspace.

Now we look at the other side of the correspondence. We already defined what a Weil-Deligne representation is in Weil-Deligne Representations. A Weil-Deligne representation (\rho_{0},N) is F-semisimple if the representation \rho_{0} is the direct sum of irreducible representations.

In the case of \mathrm{GL}_{1}(F), the local Langlands correspondence is a restatement of local class field theory. We have that \mathrm{GL}_1(F)=F^{\times}, and the only irreducible admissible representations of \mathrm{GL}_1(F) are continuous group homomorphisms \chi:F^\times\to\mathbb{C}^{\times}.

On the other side of the correspondence we have the one-dimensional Weil-Deligne representations (\rho_{0},N) of W_F, which must have monodromy operator N=0 and must factor through the abelianization W_F^{\mathrm{ab}}.

Recall from Weil-Deligne Representations that we local class field theory gives us an isomorphism \mathrm{rec}:F^{\times}\xrightarrow{\sim}W_{F}^{\mathrm{ab}}, also known as the Artin reciprocity map. We can now describe the local Langlands correspondence explicitly. It sends \chi to the Weil-Deligne representation (\rho_{0},0), where \rho_{0} is the composition W_{F}\to W_{F}^{\mathrm{ab}}\xrightarrow{\mathrm{rec}^{-1}}F^{\times}\xrightarrow{\chi}\mathbb{C}^{\times}.

Now let us consider the case of \mathrm{GL}_{2}(F). If the residue field of F is not of characteristic 2, then the irreducible admissible representations of \mathrm{GL}_{2}(F) may be enumerated, and they fall into four types: principal series, special, one-dimensional, and supercuspidal.

Let \chi_{1},\chi_{2}:F^{\times}\to\mathbb{C}^{\times} be continuous admissible characters and let I(\chi_{1},\chi_{2}) be the vector space of functions \phi:\mathrm{GL}_{2}(F)\to\mathbb{C} such that

\displaystyle \phi \left(\begin{pmatrix}a&b\\0&d\end{pmatrix}g\right)=\chi_{1}(a)\chi_{2}(d)\Vert a/d\Vert^{1/2}\phi(g)

The group \mathrm{GL}_{2}(F) acts on the functions \phi, just as in Representation Theory and Fourier Analysis. Therefore it gives us a representation of \mathrm{GL}_{2}(F) on the vector space I(\chi_{1},\chi_{2}), which we say is in the principal series.

Now the representation I(\chi_{1},\chi_{2}) might be irreducible, in which case it is one of the things that go into our correspondence, or it might be reducible. This is decided by the Bernstein-Zelevinsky theorem, which says that I(\chi_{1},\chi_{2}) is irreducible precisely if the ration of the characters \chi_{1} and \chi_{2} is not equal to plus or minus 1.

In the case that \chi_{1}/\chi_{2}=1, then we have an exact sequence

\displaystyle 0\to\rho\to I(\chi_{1},\chi_{2})\to S(\chi_{1},\chi_{2})\to 0

where the representations S(\chi_{1},\chi_{2}) and \rho are both irreducible representations of \mathrm{GL}_{2}(F). The representation S(\chi_{1},\chi_{2}) is infinite-dimensional and is known as the special representation. The representation \rho is the one-dimensional representation and is given by \chi_{1}\Vert\cdot\Vert^{1/2}\det.

If \chi_{1}/\chi_{2}=-1 instead, then we have a “dual” exact sequence

\displaystyle 0\to S(\chi_{1},\chi_{2}) \to I(\chi_{1},\chi_{2})\to \rho\to 0

So far the irreducible admissible representations of \mathrm{GL}_{2}(F) that we have seen all arise as subquotients of I(\chi_{1},\chi_{2}). Since characters such as \chi_{1} and \chi_{2} are the irreducible admissible representations of \mathrm{GL}_{1}(F), we may consider the principal series, special, and one-dimensional representations as being built out of these more basic building blocks.

However there exist irreducible admissible representations that do not arise via this process, and they are called supercuspidal representations. For \mathrm{GL}_{2}(F) there is one kind of supercuspidal representation denoted \mathrm{BC}_{E}^{F}(\psi) for E a quadratic extension of F and \psi an admissible character \psi:E\to\mathbb{C}^{\times}.

Now we know what the irreducible admissible representations of \mathrm{GL}_{2}(F) are. The local Langlands correspondence says that they will correspond to F-semisimple Weil-Deligne representations. We can actually describe explicitly which Weil-Deligne representation each irreducible admissible representation of \mathrm{GL}_{2}(F) gets sent to!

Let \chi_{1},\chi_{2}:F^{\times}\to \mathbb{C}^{\times} be the same continuous admissible characters used to construct the irreducible representations as above, and let \rho_{1},\rho_{2} :W_{F}\to \mathbb{C}^{\times} be the corresponding representation of the Weil group given by the local Langlands correspondence for \mathrm{GL}_{1}, as discussed earlier. Then to each irreducible admissible representation of \mathrm{GL}_2(F) we associate a 2-dimensional Weil-Deligne representation as follows:

To the principal series representation I(\chi_{1},\chi_{2}) we associate the Weil-Deligne representation (\rho_{1}\oplus\rho_{2},0).

To the special representation S(\chi_{1},\chi_{1}\times\Vert\cdot\Vert), we associate the Weil-Deligne representation \left(\begin{pmatrix}\Vert\cdot\Vert\rho_{1} & 0\\0 & \rho_{1}\end{pmatrix},\begin{pmatrix} 0 & 1\\0 & 0\end{pmatrix}\right).

To the one-dimensional representation \chi_{1}\circ\det, we associate the Weil-Deligne representation \left(\begin{pmatrix}\rho_{1}\times\Vert\cdot\Vert^{1/2} & 0\\0 & \rho_{1}\times\Vert\cdot\Vert^{-1/2}\end{pmatrix},0\right).

Finally, to the supercuspidal representation \mathrm{BC}_{E}^{F}(\psi) we associate the Weil-Deligne representation (\mathrm{Ind}_{W_{E}}^{W_{F}}\sigma,0), where \sigma is the unique nontrivial element of \mathrm{Gal}(E/F).

We have been able to describe the local Langlands correspondence for \mathrm{GL}_{1}(F) and \mathrm{GL}_{2}(F) explicitly (in the latter case as long as the characteristic of the residue field of F is not 2). The local Langlands correspondence for \mathrm{GL}_{n}(F), for more general n on the other hand, was proven via geometric means – namely using the geometry of certain Shimura varieties (see also Shimura Varieties) as well as their local counterpart, the Lubin-Tate tower, which parametrizes deformations of Lubin-Tate formal group laws (see also The Lubin-Tate Formal Group Law) together with level structure.

There has been much recent work regarding the local Langlands program for groups other than \mathrm{GL}_{n}(F). For instance there is work on the local Langlands correspondence for certain symplectic groups making use of “theta lifts”, by Wee Teck Gan and Shuichiro Takeda. Very recently, there has also been work by Laurent Fargues and Peter Scholze that makes use of ideas from the geometric Langlands program. These, and more, will hopefully be discussed more here in the future.

References:

Langlands program on Wikipedia

Local Langlands conjectures on Wikipedia

Local Langlands conjecture on the nLab

MSRI Summer School on Automorphic Forms and the Langlands Program by Kevin Buzzard

Langlands Correspondence and Bezrukavnikov Equivalence by Anna Romanov and Geordie Williamson

The Local Langlands Conjecture for GL(2) by Colin Bushnell and Guy Henniart

Weil-Deligne Representations

Let F be a finite extension of the p-adic numbers \mathbb{Q}_{p}. In Galois Representations we described some continuous Galois representations of \mathrm{Gal}(\overline{F}/F), but all of them were p-adic (or rather \ell-adic, see the discussion in that post for the explanation behind the terminology). What about complex Galois representations? For instance, since the complex \ell-adic numbers (the completion of the algebraic closure of the \ell-adic numbers) are isomorphic to the complex numbers, if we fix such an isomorphism we could just base change to the complex numbers to get a complex Galois representation.

Complex Galois representations, also known as Artin representations, are in fact an interesting object of study in number theory. However, the issue is that if we require these Galois representations to be continuous, like we have required for our \ell-adic representations, we will find that they always have finite image, which also means in essence that we might as well just have been studying representations of finite Galois groups, not the absolute one as we intend to do.

To get a complex representation that will be as interesting as the p-adic ones, we have to make certain modifications. We will look at certain representations of a certain subgroup of the Galois group instead, called the Weil group, and together with some additional information in the form of a “monodromy operator“, we will have a complex representation that will in a way carry the same information as a \ell-adic representation.

Let us first define this Weil group. F be a local field and let \kappa be its residue field. The absolute Galois groups of F and \kappa fit into the following exact sequence

\displaystyle 0\to I\to \mathrm{Gal}(\overline{F}/F)\to\mathrm{Gal}(\overline{\kappa}/\kappa)\to 0

where I is the kernel of the surjective map \mathrm{Gal}(\overline{F}/F)\to\mathrm{Gal}(\overline{\kappa}/\kappa) and is called the inertia subgroup (this can be considered the “local” and also “absolute” version of the exact sequence discussed near the end of Splitting of Primes in Extensions).

The residue field \kappa is a finite field, say of some cardinality q. Finite fields have the property that they have a unique extension of degree n for every n, and the Galois groups of these extensions are cyclic of order n. As a result, the absolute Galois group \mathrm{Gal}(\overline{\kappa}/\kappa) of the residue field \kappa is isomorphic to the inverse limit \varprojlim_{n} \mathbb{Z}/n\mathbb{Z}, also known as the profinite integers and denoted \widehat{\mathbb{Z}}.

There is a special element of \mathrm{Gal}(\overline{\kappa}/\kappa) called the Frobenius, which corresponds to raising to the power of q. The powers of Frobenius give us a subgroup isomorphic to the integers \mathbb{Z} inside \mathrm{Gal}(\overline{\kappa}/\kappa) (which again is isomorphic to \widehat{\mathbb{Z}}). The inverse image of this subgroup under the surjective morphism \mathrm{Gal}(\overline{F}/F)\to\mathrm{Gal}(\overline{\kappa}/\kappa) is what is known as the Weil group of F (denoted W_{F}). Since \widehat{\mathbb{Z}} is the completion of \mathbb{Z}, the Weil group may be thought of as a kind of “decompletion” of the Galois group \mathrm{Gal}(\overline{F}/F).

It follows from local class field theory (see also Some Basics of Class Field Theory) that we have an isomorphism between the abelianization W_{F}^{\mathrm{ab}} of the Weil group and F^{\times}.

A Weil-Deligne representation is a pair (\rho_{0},N) consisting of a representation \rho_{0} of the Weil group W_{F}, together with a nilpotent operator N called the monodromy operator, which has to satisfy the property

\displaystyle \rho_{0}(\sigma)N\rho_{0}(\sigma)^{-1}=\Vert\sigma\Vert N

for all \sigma in W_{F}, where \Vert\sigma\Vert is the valuation of the element of F^{\times} corresponding to \sigma under the isomorphism given by local class field theory as mentioned above.

Grothendieck’s monodromy theorem them says that given a continuous p-adic representation \rho we can always associate to it a unique Weil-Deligne representation (\rho_{0},N) satisfying the property that, if we express an element of the absolute Galois group as \phi^{m}\sigma where \phi is a lift of Frobenius and \sigma belongs to the inertia group, then \rho(\phi^{m}\sigma)=\rho_{0}(\phi^{m}(\sigma))\mathrm{exp}(Nt(\sigma)), where t:\mathrm{Gal}(F^{\mathrm{tame}}/F^{\mathrm{ur}})\to\mathbb{Z}_{\ell}, F^{\mathrm{tame}} being the “tamely ramified” extension of F and F^{\mathrm{ur}} the unramified extension of F. The point is that, we can now associated to a p-adic Galois representation a complex representation in the form of the Weil-Deligne representation, which is the goal we stated in the beginning of this post.

It turns out that certain Weil-Deligne representations (those which are called F-semisimple) are in bijection with irreducible admissible representations of the \mathrm{GL}_{n}(F), thus linking two kinds of representations – those of Galois groups like we have discussed here, and those of reductive groups, similar to what was hinted at in Representation Theory and Fourier Analysis. This will be discussed in a future post.

References:

Weil group on Wikipedia

MSRI Summer School: Automorphic Forms and the Langlands Program (Lecture Notes) by Kevin Buzzard

Perfectoid Fields

Consider the field of p-adic numbers \mathbb{Q}_{p}. An element of \mathbb{Q}_{p} may be written in the form

\displaystyle \sum_{n=k}^{\infty}a_{n}p^{n}

with each a_{n} being an element of the finite field \mathbb{F}_{p}. Let us compare this with the field of Laurent series \mathbb{F}_{p}((t)) in one variable t over \mathbb{F}_{p}. An element of \mathbb{F}_{p}((t)) may be written in the form

\displaystyle \sum_{m=l}^{\infty}a_{m}t^{m}

We see that they look very similar, even though \mathbb{Q}_{p} is characteristic 0, and \mathbb{F}_{p}((t)) is characteristic p.

How far can we push this analogy? The fact that one is in characteristic 0, and the other is characteristic p means we cannot ask for an isomorphism of fields. However, the Fontaine-Wintenberger theorem gives us another connection between \mathbb{Q}_{p} and \mathbb{F}_{p}((t)) – if we modify them by adjoining p-power roots of p and t respectively. This theorem states that the fields \cup_{n}\mathbb{Q}_{p}(p^{1/p^{n}}) and \cup_{n}\mathbb{F}_{p}((t^{1/p^{n}})) have the same absolute Galois group! By the fundamental theorem of Galois theory, this means the category formed by their extensions will be equivalent as well.

We now let F denote the completion of \cup_{n}\mathbb{Q}_{p}(p^{1/p^{n}}), and we let F^{\flat} suggestively denote the completion of \cup_{n}\mathbb{F}_{p}((t^{1/p^{n}})). Completing these fields does not change their absolute Galois groups, so the absolute Galois groups of F and F^{\flat} remain isomorphic. We say that the characteristic p field F^{\flat} is the tilt of the characteristic 0 field F, and that F^{\flat} is an untilt of F (note the subtle change in our choice of article – untilts are not unique).

In this post, we will explore these kinds of fields – which are called perfectoid fields – and the process of tilting and untilting that bridges the world of characteristic 0 and characteristic p. After Fontaine and Wintenberger came up with their famous theorem their ideas have since been developed into even more general and even more powerful theories of perfectoid rings and perfectoid spaces – but we will leave these to future posts. For now we concentrate on the case of fields.

First let us look at a much more primitive example of bridging the world of characteristic 0 and characteristic p. Consider \mathbb{Q}_{p} (characteristic 0). It has a ring of integers \mathbb{Z}_{p}, whose residue field is \mathbb{F}_{p} (characteristic p). To got the other way, starting from \mathbb{F}_{p} we can take its ring of Witt vectors, which is \mathbb{Z}_{p}. Then we take its field of fractions which is \mathbb{Q}_{p}.

More generally, there is a correspondence between characteristic 0 discretely valued complete fields whose uniformizer is p and characteristic p fields which are perfect, i.e. for which the Frobenius morphism is bijective, and the way to go from one category to the other is as in the previous paragraph.

This is a template for “bridging the world of characteristic 0 and characteristic p“. However, we may want more, something like the Fontaine-Wintenberger theorem where the characteristic 0 object and the characteristic p object have isomorphic absolute Galois groups. We will be tweaking this basic bridge in order to create something like Fontaine-Winterger theorem, and these tweaks will lead us to the notion of a perfectoid field. However, we already have isolated one property that we want from such a “perfectoid” field:

The first property that we want from a perfectoid field is that it has to be nonarchimedean. This allows us to have a “ring of integers” that serves as an intermediary object between the two worlds, as we have seen above.

Now let us concentrate on the Fontaine-Wintenberger theorem. To understand this phenomenon better, we need to make use of a version of the fundamental theorem of Galois theory, which allows us to think in terms of extensions of fields instead of their Galois groups. More properly, we want an equivalence of categories between the “Galois categories” of certain extensions of these “base” fields and this will be the property of these base fields being perfectoid. Now the problem is that the extensions that we are considering may not fit into the primitive correspondence we stated above – for example the corresponding characteristic p object may not be perfect, i.e. the Frobenius morphism may not be surjective.

The fix to this is a kind of “perfection”, which is the tilting functor we mentioned earlier. Let R be a ring. The tilt of R, denoted R^{\flat} is defined to be the inverse limit

\displaystyle R^{\flat}=\varprojlim_{x\mapsto x^{p}}R/pR

In other words, an element x of R^{\flat} is an infinite sequence of elements (x_{0},x_{1},x_{2},\ldots) of the quotient R/pR such that x_{1}\cong x_{0}^{p}\mod p, x_{2}\cong x_{1}^{p}\mod p, and so on. We want R^{\flat} to be a ring, so we define it to have componentwise multiplication, i.e.

\displaystyle (xy)_{i}=x_{i}y_{i}

However the addition is going to be more complicated. We define it, for each component, as follows:

\displaystyle (x+y)_{i}=\lim_{n\to\infty}(x_{i+n}+y_{i+n})^{p^n}

At this point we take the opportunity to define another important concept in the theory of perfectoid fields (and rings). Let W be the Witt vector functor (see also The Field with One Element). Then we give the Witt vectors of the tilt of R, W(R^{\flat}), a special name. We will refer to this ring as A_{\mathrm{inf}}(R). It will make an appearance again later. For now we note that there is going to be a canonical map \theta: A_{\mathrm{inf}}(R)\to R.

As we can see, we have defined the tilt of an arbitrary ring. This is not exclusive to the ones which are “perfectoid” whatever the definition of “perfectoid” may be (we will come to this later of course). Again what makes perfectoid fields (such as our earlier examples) special though, is that if F is a perfectoid field of characteristic 0, then F and its tilt F^{\flat} will have isomorphic absolute Galois groups. This will actually follow from the following statement (together with some technicalities involving fiber functors and so on):

There is an equivalence of categories between the category of finite etale algebras over a perfectoid field F and the category of finite etale algebras over its tilt F^{\flat}.

This in turn will follow from the following two statements:

  1. Finite extensions of perfectoid fields are perfectoid.
  2. There is an equivalence of categories between the category of perfectoid extensions of a perfectoid field F and the category of perfectoid extensions over its tilt F^{\flat}.

This equivalence of categories is given by tilting a perfectoid extension over F. This will actually give us a perfectoid extension over F^{\flat}. However, we need a functor that goes in the other direction, a “quasi-inverse” that when composed with tilting gives us back our original perfectoid extension over F (or at least something isomorphic to it, this is what the “quasi-” part means). However, we also said in an earlier paragraph that the “untilt” of a characteristic p field may not be unique (two different untilts may also not be isomorphic). How do we approach this problem?

We recall again the ring A_{\mathrm{inf}}(R) defined earlier as the ring of Witt vectors of the tilt of R, and we recall that it has a canonical map \theta:A_{\mathrm{inf}}(R)\to R. If we know this map, and if we know that it is surjective, then we can recover R simply by quotienting out by the kernel of the map \theta!

The problem is that (aside from not knowing whether it is in fact surjective or not) is that we only know this map if we know that R^{\flat} was obtained as the tilt of R. If we were simply handed some characteristic p field for instance we would not be able to know this map.

However, note that we are interested in an equivalence of categories between the category of perfectoid extensions over the field F and the corresponding category over its tilt F^{\flat}. By specifying these “bases” F and F^{\flat}, it is in fact enough to specify unique untilts! In other words, if we have say just some perfectoid field A, we cannot determine a unique untilt for it, but if we say in addition that it is a perfectoid extension over F, and we are looking for the unique untilt of it over F^{\flat}, we can in fact find it, as long as the map \theta is surjective.

So now how do we guarantee that \theta is surjective? This brings us to our second property, which is that the Frobenius morphism from \overline{R} to itself must be surjective. This is actually the origin of the word “perfectoid”; since as above a field for which the Frobenius morphism is bijective is called perfect; hence, requiring it to be surjective is a relaxation of this condition. This condition guarantees that the map \phi:A_{\mathrm{inf}}(R)\to R is going to be surjective.

The final property that we want from a perfectoid field is that its valuation must be non-discretely valued. The reason for this is that we want to consider infinitely ramified extensions of \mathbb{Q}_{p}. The two previous conditions that we want can only be found in unramified (discretely valued) or infinitely ramified (non-discretely valued) of \mathbb{Q}_{p}. We have already seen above that if we only look at the ones which are unramified then our corresponding characteristic p objects will be limited to perfect \mathbb{F}_{p}-algebras, and this is not enough to give us the Fontaine-Wintenberger theorem. Therefore we will want infinitely ramified extension of \mathbb{Q}_{p}, and these are non-discretely valued.

These three properties are enough to give us the Fontaine-Wintenberger theorem. To summarize – a perfectoid field is a complete, nonarchimedean field F such that the Frobenius morphism from \mathcal{O}_{F}/\mathfrak{p} to itself is surjective and such that its valuation is non-discretely valued.

We have only attempted to motivate the definition of a perfectoid field in this post, and barely gone into any sort of detail. For that one can only recommend the excellent post by Alex Youcis on his blog The Fontaine-Wintenberger Theorem: Going Full Tilt, which inspired this post, but barely does it any justice.

Aside from the Fontaine-Wintenberger theorem, the concepts we have described here – the idea behind “perfectoid”, the equivalence of categories of perfectoid extensions that gives rise to the Fontaine-Wintenberger theorem, the idea of tilting and untilting which bridges the worlds of characteristic 0 and characteristic p, the ring A_{\mathrm{inf}}(R), and so on, have found much application in many areas of math, from the aforementioned perfectoid rings and perfectoid spaces, to p-adic Hodge theory, and to many others.

References:

Perfectoid Space on Wikipedia

What is…a Perfectoid Space? by Bhargav Bhatt

Berkeley Lectures on p-adic Geometry by Peter Scholze and Jared Weinstein

Reciprocity Laws and Galois Representations: Recent Breakthroughs by Jared Weinstein

The Fontaine-Wintenberger Theorem: Going Full Tilt by Alex Youcis

Galois Representations Coming From Weight 2 Eigenforms

In Galois Representations we mentioned briefly that Galois representations can be obtained from modular forms. In this post we elaborate more on this construction, in the case that the modular form is a weight 2 eigenform (a weight 2 cusp form that is a simultaneous eigenfunction for all Hecke operators not dividing the level N). This specific case is also known as the Shimura construction, after Goro Shimura.

Let f be a weight 2 Hecke eigenform, of some level \Gamma_{0}(N) (this also works with other level structures). We want to construct a p-adic Galois representation associated to this Hecke eigenform, such that the two are going to be related in the following manner. For every prime \ell not dividing N and not equal to p, the characteristic polynomial of the image of the Frobenius element associated to \ell under this Galois representation will be of the form

\displaystyle x^{2}-a_{\ell}x+\ell\chi(\ell)

where a_{\ell} is the eigenvalue of the Hecke operator T_{\ell} and \chi is a Dirichlet character associated to another kind of Hecke operator called the diamond operator \langle \ell\rangle. This diamond operator acts on the argument of the modular form by an upper triangular element of \mathrm{SL}_{2}(\mathbb{Z}) whose bottom right entry is \ell mod N. This action is the same as the action of a Dirichlet character \chi:\mathbb{Z}/N\mathbb{Z}\to\mathbb{C}^{\times}. The above polynomial is also known as the Hecke polynomial.

The first thing that we will need is the identification of the weight 2 cusp forms with the holomorphic differentials on the modular curve (as mentioned in Modular Forms in the case of \mathbb{SL}_{2}(\mathbb{Z}), although this is can be done more generally).

The second thing that we will need is the Jacobian. One can think of the Jacobian as the space given by the equivalence classes of all path integrals on a curve (in general we can do this for any algebraic curve, not just modular curves), where two path integrals are to be considered equivalent if they differ by integration along a loop. Since path integration can be considered as a linear functional from holomorphic differentials to the complex numbers, we consider such path integrals as the dual space to the space of holomorphic differentials. However, the loops we wanted to quotient out by can also be expressed as elements of the homology group of the curve (see also Homology and Cohomology)!

Therefore we now define the Jacobian of a curve X as

\displaystyle J(\Gamma)=\Omega^{\vee}/H_{1}(X,\mathbb{Z})

where \Omega denotes the holomorphic differentials on X. The notation \Omega^{\vee} denotes the dual to \Omega, since as we said the path integrals form the dual to the holomorphic differentials. The Jacobian can also described in other ways – for instance it is also the connected component of the Picard group (see also Divisors and the Picard Group), and the connection to the description given here is an important classical theorem called the Abel-Jacobi theorem.

The Jacobian is a higher-dimensional complex torus, and actually more is true – it is also an abelian variety, i.e. a projective variety whose points form a group (and hence a generalization of elliptic curves). Note that every complex torus is an elliptic curve, but this is not true in higher dimensions – only certain special kinds of higher dimensional complex tori (namely those with a polarization) are abelian varieties. In this vein the Jacobian of a curve has yet another description – it is “universal” among abelian varieties in that, if there is a morphism from a curve to any abelian variety, it can be expressed as a morphism from the curve to its Jacobian, followed by a morphism to that other abelian variety.

Now we go back to the case of modular curves. Denoting by S_{2}(\Gamma_{0}(N)) the space of cusp forms of weight two for the level structure \Gamma_{0}(N), which as discussed above is isomorphic to the space of holomorphic differentials on the corresponding modular curve X(\Gamma_{0}(N)), we can now define the Jacobian J(\Gamma_{0}(N)) as

\displaystyle J(\Gamma_{0}(N))=S_{2}(\Gamma_{0}(N))^{\vee}/H_{1}(X,\mathbb{Z})

The third ingredient that we need is a certain ideal of the Hecke algebra (the ring of endomorphisms of S_{2}(\Gamma_{0}(N)) generated by the actions of the Hecke operators and diamond operators) corresponding to the weight 2 Hecke eigenform f (let us denote this ideal by \mathbb{I}_{f}) that we want to obtain our Galois representation from. This ideal \mathbb{I}_{f}) is defined to be the one generated by all elements of the Hecke algebra whose eigenvalue when acting on f is zero.

Since the Hecke operators and diamond operators act on the Jacobian (we can see this this way – since the Jacobian is the quotient of the linear functionals on S_{2}(\Gamma_{0}(N)), the action is obtained by first applying the Hecke operator or diamond operator to the weight 2 eigenform, then applying the linear functional), we can use the ideal \mathbb{I}_{f} to cut down a quotient of the Jacobian which is another abelian variety A_{f}:

\displaystyle A_{f}=J(\Gamma_{0}(N))/\mathbb{I}_{f}J(\Gamma_{0}(N))

Finally, we can take the Tate module of A_{f}, and this will give us precisely the Galois representation that we want. The abelian variety A_{f} will have dimension equal to the degree of the number field generated by the eigenvalues of the Hecke operators.

If the eigenvalues are all rational, then A_{f} will actually be an elliptic curve – in other words, given an eigenform of weight 2 whose Hecke eigenvalues are all rational, we can always use it to construct an elliptic curve! This also gives us a map from the modular curve X(\Gamma_{0}(N)) to this elliptic curve, called a modular parametrization. The resulting elliptic curve will have the property that its L-function, built from point counts when it is reduced modulo primes, is the same as the L-function of the modular form which is built from its Fourier coefficients! This is because the Frobenius and the Fourier coefficients (which are also the eigenvalues of the Hecke operators) are related, as discussed above. The question of whether, given an elliptic curve, it comes from a modular form in this way, is another restatement of the question of modularity. The affirmative answer to this question, at least for certain elliptic curves over \mathbb{Q}, led to the proof of Fermat’s Last Theorem.

This theory, which is only very roughly sketched here, is just a very special case – one can also obtain, for instance, Galois representations from modular forms which are not of weight 2. We leave this for the future.

References:

Jacobian variety on Wikipedia

Abel-Jacobi map on Wikipedia

Modularity theorem on Wikipedia

Course on Mazur’s Theorem Lecture 10: Jacobians by Andrew Snowden

Course on Mazur’s Theorem Lecture 17: Eichler-Shimura by Andrew Snowden

A First Course in Modular Forms by Fred Diamond and Jerry Shurman

Representation Theory and Fourier Analysis

In Some Basics of Fourier Analysis we introduced some of the basic ideas in Fourier analysis, which is ubiquitous in many parts of both pure and applied math. In this post we look at these same ideas from a different point of view, that of representation theory.

Representation theory is a way of studying group theory by turning it into linear algebra, which in many cases is more familiar to us and easier to study.

A (linear) representation is just a group homomorphism from some group G we’re interested in, to the group of linear transformations of some vector space. If the vector space has some finite dimension n, the group of its linear transformations can be expressed as the group of n \times n matrices with nonzero determinant, also known as \mathrm{GL}_n(k) (k here is the field of scalars of our vector space).

In this post, we will focus on infinite-dimensional representation theory. In other words, we will be looking at homomorphisms of a group G to the group of linear transformations of an infinite-dimensional vector space.

“Infinite-dimensional vector spaces” shouldn’t scare us – in fact many of us encounter them in basic math. Functions are examples of such. After all, vectors are merely things we can scale and add to form linear combinations. Functions satisfy that too. That being said, if we are dealing with infinity we will often need to make use of the tools of analysis. Hence functional analysis is often referred to as “infinite-dimensional linear algebra” (see also Metric, Norm, and Inner Product).

Just as a vector v has components v_i indexed by i, a function f has values f(x) indexed by x. If we are working over uncountable things, instead of summation we may use integration.

We will also focus on unitary representations in this post. This means that the linear transformations are further required to preserve a complex inner product (which takes the form of an integral) on the vector space. To facilitate this, our functions must be square-integrable.

Consider the group of real numbers \mathbb{R} (under addition). We want to use representation theory to study this group. For our purposes we want the square-integrable functions on some quotient of \mathbb{R} as our vector space. It comes with an action of \mathbb{R}, by translation. In other words, an element a of \mathbb{R} acts on our function f(x) by sending it to the new function f(x+a).

So what is this quotient of \mathbb{R} that our functions will live on? For now let us choose the integers \mathbb{Z}. The quotient \mathbb{R}/\mathbb{Z} is the circle, and functions on it are periodic functions.

To recap: We have a representation of the group \mathbb{R} (the real line under addition) as linear transformations (also called linear operators) of the vector space of square-integrable functions on the circle.

In representation theory, we will often decompose a representation into a direct sum of irreducible representations. Irreducible means it contains no “subrepresentation” on a smaller vector space. The irreducible representations are the “building blocks” of other representations, so it is quite helpful to study them.

How do we decompose our representation into irreducible representations? Consider the representation of \mathbb{R} on the vector space \mathbb{C} (the complex numbers) where a real number a acts by multiplying a complex number z by e^{2\pi i k a}, for k an integer. This representation is irreducible.

If this looks familiar, this is just the Fourier series expansion for a periodic function. So a Fourier series expansion is just an expression of the decomposition of the representation of R into irreducible representations!

What if we chose a different vector space instead? It might have been the more straightforward choice to represent \mathbb{R} via functions on \mathbb{R} itself instead of on the circle \mathbb{R}/\mathbb{Z}. That may be true, but in this case our decomposition into irreducibles is not countable! The irreducible representations into which this other representation decomposes is the one where a real number a acts on \mathbb{C} by multiplication by e^{2 \pi i k a} where k is now a real number, not necessarily an integer. So it’s not indexed by a countable set.

This should also look familiar to those who know Fourier analysis: This is the Fourier transform of a square-integrable function on \mathbb{R}.

So now we can see that concepts in Fourier analysis can also be phrased in terms of representations. Important theorems like the Plancherel theorem, for example, also may be understood as an isomorphism between the representations we gave and other representations on functions of the indices. We also have the Poisson summation in Fourier analysis. In representation theory this is an equality obtained from calculating the trace in two ways, as a sum over representations and as a sum over conjugacy classes.

Now we see how Fourier analysis is related to the infinite-dimensional representation theory of the group \mathbb{R} (one can also see this as the infinite-dimensional representation theory of the circle, i.e. the group \mathbb{R}/\mathbb{Z} – the article “Harmonic Analysis and Group Representations” by James Arthur discusses this point of view). What if we consider other groups instead, like, say, \mathrm{GL}_n(\mathbb{R}) or \mathrm{SL}_n(\mathbb{R}) (or \mathbb{R} can be replaced by other rings even)?

Things get more complicated, for example the group may not be abelian. Since we used integration so much, we also need an analogue for it. So we need to know much about group theory and analysis and everything in between for this.

These questions have been much explored for the kinds of groups called “reductive”, which are closely related to Lie groups. They include the examples of \mathrm{GL}_n(\mathbb{R}) and \mathrm{SL}_n(\mathbb{R}) earlier, as well as certain other groups we have discussed in previous posts such as the orthogonal and unitary (see also Rotations in Three Dimensions). There is a theory for these groups analogous to what I have discussed in this post, and hopefully this will be discussed more in future blog posts here.

References:

Representation theory on Wikipedia

Representation of a Lie group on Wikipedia

Fourier analysis on Wikipedia

Harmonic analysis on Wikipedia

Plancherel theorem on Wikipedia

Poisson summation formula on Wikipedia

An Introduction to the Trace Formula by James Arthur

Harmonic Analysis and Group Representations by James Arthur