Representation theory
Math 797RT
Representation Theory
Notes by Patrick Lei,
May 2020
Lectures by Ivan Mirkovic, Fall 2018
University of Massachusetts Amherst
Disclaimer
These notes are a May 2020 transcription of handwritten notes taken during lecture in Fall 2018. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style (omit lengthy computations, use category theory) and that of the instructor. If you find any errors, please contact me at plei@umass.edu.
Additional Disclaimers
- 1.
Ivan was in China for a week, and I missed one of the two makeup lectures. If there are any gaps in the material, this is probably a big reason why.
- 2.
I did not take notes for the final lecture. I believe only the classification of semisimple Lie algebras was given.
- 3.
The course has been redesigned since I took it, so these notes may not be useful. In fact, I’m told that there is no set syllabus for this course, so it may continue to change in the future.
- 4.
Ivan has notes at https://people.math.umass.edu/~mirkovic/C.Graduate/TopicsCourses/7.RepresentationTheory/8.pdf, but his notes are much more extensive and in depth than the lectures. My notes are a more faithful representation of what actually occurred during lecture.
- 5.
Ivan’s lectures assumed knowledge far beyond the stated prerequisites of the course, which were just graduate algebra. It may be clear that I did not understand what was going on half the time.
1 Groups
1.1 Overview
We begin with an overview of various aspects of representation theory.
- 1.
First, the principle that any subject should be organized by its symmetries (harmonic analysis). We want to spot symmetries in a problem, and this is related to physics and QFT.
- 2.
We want to consider symmetries related to subtle actions of space. For example, for a scheme and sufficiently nice group scheme , we can consider the quotient stack . The projection from is a principle -bundle, and in fact, we can do this for , which is a point. Also, we will need the idea of the fiber product later.
- 3.
Groups are used to describe objects that are locally trivial but not globally (for example cohomology classes).
- 4.
Groups can be recovered from their categories of representations via Tannaka-Krein duality by considering automorphisms of the forgetful functor .
- 5.
The most important class of groups is the compact Lie groups.
Recall the notion of a group action on a set . If we view for some category , then we want our group to act by automorphisms in . Therefore we can define topological group actions on topological spaces and actions of groups on vector spaces.
Lemma 1.1.
For a group and vector space , then an action of on is equivalent to a morphism .
Example 1.2.
If acts on a set , then the space of functions on is a representation of .
We can generalize the above example to where has additional structure, where we can consider functions in various categories. For example, if we consider acting on the upper half plane , then we can consider the space of holomorphic functions on .
1.2 Category of Representations
Lemma 1.3.
- 1.
Representations of on vector spaces over a field form a category . Note that morphisms are
- 2.
is an abelian category.
Recall that an abelian category is a category with the following:
- •
A zero object . Here, just take the zero representation.
- •
A direct sum (coproduct) operation . Here, take .
- •
Subobjects, which in this case are -invariant subspaces.
- •
Quotients by subobjects. This is obvious.
- •
has all (co)kernels and (co)images, and images and coimages are isomorphic.
Some examples of abelian categories are categories of modules over a ring, but in fact is equivalent to .
- •
- 3.
is a monoidal category. For module categories, this is only true in general for commutative rings. For two representations of , we can define an action on by .
Recall that a monoidal structure on a category is an associative and unital functor . The unit in is with the trivial action.
- 4.
is a closed monoidal category. This means that there is an internal Hom functor
that is the right adjoint to the tensor product.
Examples 1.4.
We give some examples of closed monoidal categories.
- 1.
In the category of sets, we see that
by the maps
- 2.
In the category of vector spaces, we again note that
by the same currying morphism.
We can check that the internal Hom in the category of representations is simply the vector space Hom. To see that this is a representation, define the action .
Recall that a right adjoint to a functor satisfies
as sets in a manner that is natural in . Being a closed monoidal category means that is a left adjoint to .
1.3 Representations of Finite Groups
First we consider invariants of , which correspond to morphisms from the trivial representation.
Lemma 1.5.
We have the two identities and .
Definition 1.6.
A representation is irreducible if and has no nontrivial subrepresentations.
Then define to be the set of isomorphism classes of irreducible representations of . We will use the following strategy to understand :
- 1.
Find all irreducible representations;
- 2.
Study each irreducible representation;
- 3.
For any interesting representation, decompose it into irreducibles.
For a finite group , consider the set of functions . Then we have two subrepresentations and .
Lemma 1.7.
If the characteristic of does not divide , then .
Proof.
The dimensions match up, so the only constant satisfying is by assumption. □
Example 1.8.
If , then are irreducible, but they are actually the same subrepresentation of over .
Definition 1.9.
A representation is semisimple if it is a direct sum of irreducibles.
Theorem 1.10 (Schur Lemma).
Let be two irreducible finite-dimensional representations of .
- 1.
If , then is an isomorphism;
- 2.
is a finite-dimensional division algebra;
- 3.
If , then .
Proof.
- 1.
Note that is a subrepresentation and is irreducible, so . Similarly, the image is nonzero and is a subrepresentation of , so it must be all of .
- 2.
Any nonzero endomorphism must be an isomorphism.
- 3.
If is nonzero, it has an eigenvalue with eigenvector . Then is a nonzero subrepresentation, so it must be everything.
□
Corollary 1.11.
- 1.
is semisimple if and only if it can be written as a direct sum
- 2.
Let be irreducible. Then .
Proof.
- 1.
Note that if is a direct sum of irreducibles, then we have
- 2.
Hom is an additive functor, so just use Schur’s Lemma.
□
Corollary 1.12.
If is algebraically closed, then for semisimple and irreducible.
Theorem 1.13.
Let be abelian and be algebraically closed. Then all finite-dimensional irreducible representations of are one-dimensional.
Proof.
Let and . Then has an eigenvalue with eigenvector . Therefore, is a subrepresentation because is abelian, so all subspaces are invariant and thus . □
This tells us that . This is known as the group of characters1 of .
Lemma 1.14.
Subrepresentations of semisimple representations are semisimple.
Proof.
Let be a subrepresentation of an irreducible representation. Then has an irreducible subrepresentation and thus , so there exists such that , so . Then , so we have an exact sequence
Therefore , so . □
Theorem 1.15.
- 1.
Sums of semisimple representations are semisimple.
- 2.
is semisimple iff and only if for every subrepresentation , there is another subrepresentation such that .
Proof.
We will only prove the second part (the first part is obvious). Note that if the condition holds, then choose an irreducible subrepresentation , apply the condition, and use induction on the dimension.
Now suppose is semisimple. Then apply Lemma 1.14 to and we see that there is a complement. □
Theorem 1.16.
Let be a final group and . Then every has a -invariant Hermitian inner product and is semisimple.
Proof.
To construct an invariant inner product , just take any Hermitian inner product and define
If we have an inner product, let be a subrepresentation. Then is also a subrepresentation, and clearly , so is semisimple by Theorem 1.15. □
Theorem 1.17 (Maschke).
If , then all finite-dimensional representatons of are semisimple.
We will now study in detail the characters of .
Lemma 1.18.
- 1.
is a group with operation the multiplication of functions.
- 2.
because .
- 3.
.
If is abelian, then we call the group of characters the Pontryagin dual of .
Examples 1.19.
- 1.
Note that because .
- 2.
We can see that .
- 3.
If is algebraically closed, and , then .
Note that by the categorical properties of the product. Consider the functor given by . Note that there is a natural2 morphism given by for .
Lemma 1.20 (Pontryagin Duality).
If is an algebraically closed field of characteristic , then is an equivalence of categories. In fact, the natural transformation defined above is an isomorphism.
Proof.
This is obviously true for finite cyclic groups, and then use the structure theorem for finite abelian groups and the properties of the product. □
Remark 1.21.
If , then the result extends to finite abelian group schemes. If we take locally compact abelian groups and , then this result still holds.
Now we will study representations of products of groups. Consider the box product
with the operation the tensor product with acting from the left and acting from the right.
Lemma 1.22.
Let . Then there exist and such that .
Proof.
Let . Then consider the morphism . We claim this is a representation of . To see this, define , which commutes with the action of on . Clearly, evaluation is surjective, so we can consider for some representation of . Therefore we can write
for some (if is reducible, then so is the entire representation). □
Lemma 1.23.
If and , then .
Proof.
Consider an irreducible subrepresentation . Then
so and . □
Theorem 1.24.
The box product gives an isomorphism .
Now let be a representation of and consider the matrix coefficient map
This is equivariant with respect to , where . Also, the matrix coefficient map is an embedding because are both irreducible.
Corollary 1.25.
Any finite group has finitely many irreducible representations.
Theorem 1.26.
Let be an algebraically closed field of characteristic . Then
as a representation of
Proof.
We compute that as representations of ,
by Frobenius reciprocity. □
It is not hard to see that an explicit isomorphism can be given by , which is the same thing as evaluation.
1.4 Characters
From now on, we will always work over the complex numbers. It turns out that each irreducible representation of gives us one conjugation-invariant function , the trace. Denote the character of a representation by .
Lemma 1.27.
- 1.
;
- 2.
.
These results follow from linear algebra of direct sums and tensor products.
2 Representation Theory of the Symmetric Group
Now we will construct irreducible representations of the symmetric group. Because irreducible representations correspond to conjugacy classes, there are bijections between the following:
- 1.
Irreducible representations of ;
- 2.
Conjugacy classes of ;
- 3.
Partitions of ;
- 4.
Young diagrams with squares;
- 5.
Conjugacy classes of nilpotent matrices of size ;
- 6.
Finite (length ) subschemes of .
We want to construct an irreducible representation for each partition . For a partition , write . We will write , where is the trivial representation. Similarly, define , where is the sign representation.
Theorem 2.1.
If are conjugate partitions, then there exists a unique irreducible representation that is common to and .
Now, we will consider the dominance order on the set of partitions of . We say that if , , and so on.
We may also consider a more geometric notion. Let be the cone of nilpotent matrices. Then every nilpotent matrix has a Jordan canonical form that corresponds to a shift operator (shift left on a Young diagram ). Then is a stratified space with strata given by orbits corresponding to partitions . Then we say that if .
Lemma 2.2.
The dominance and closure orders coincide.
2.1 Geometry of
The correspondence between irreducible representations of and orbits of nilpotent matrices under conjugation by , due to Springer, is the origin of geometric representation theory. This also extends to other reductive algebraic groups. For this, we will need to do some geometry.
First recall the Grassmannian which parameterizes dimension subspaces of . Then we can consider generalized flag varieties , which parameterizes flags of subspaces with dimension . Finally, we have the flag variety
Lemma 2.3.
These spaces are all homogeneous spaces of which are smooth projective varieties.
Now recall that the nilpotent cone is a disjoint union of smooth pieces. By definition of the dominance order, we know that
However, note that itself is a singular space, so we should be able to extract information from the singularities.
Definition 2.4.
Let be a singular variety. Then a resolution of is a proper birational map from a smooth variety .
Example 2.5.
We can resolve the variety by taking the disjoint union of two lines.
If we consider , then this is just a quadric cone in , so we can resolve by blowing up the origin to get .
Lemma 2.6.
, where is the subgroup of upper-triangular matrices, also called the standard Borel. In addition, has a natural identification with the space of Borel subgroups of .
Remark 2.7.
Note that is a nonabelian reductive group. has a maximal Borel , and we call the subgroup of with s on the diagonal . Finally, we write , and this is called the standard Cartan, or the maximal torus, because .
2.2 Springer Theory for
We want to resolve in general. To do this, we will consider . Then define
so we have a natural morphism that forgets the flag. Similarly, define
Again, there is a map . Write .
Lemma 2.8.
Let be a regular semisimple matrix. In other words, has distinct eigenvalues. Then and is a torsor for .
Here are some examples of torsors:
- 1.
The set of bases of is a torsor for .
- 2.
The Möbius strip with the zero section removed is a torsor for .
- 3.
Rank vector bundles over a base space are the same thing as -torsors over .
If acts on and , we can create a new -set , where the diagonal action of is . We can see that is a -bundle over , so if is a -torsor, then .
Note that the set of regular semisimple matrices is a Zariski-open subset of . We now consider the special fibers:
- 1.
Above , the fiber is simply all of .
- 2.
If corresponds to the Young diagram , then is a point.
- 3.
For , the fiber above the Young diagram is two copies of intersecting at a point.
Lemma 2.9.
As spaces, . This means that is a vector bundle over . Because is a representation of , is a representation of .
Remark 2.10.
These Associated Bundle constructions allow us to construct -equivariant vector bundles from representations of . If we consider the character that sends as a representation of , then we can define the vector bundle . Then we can consider the representation . Then each is either zero or irreducible or zero and all irreducible representations of are of this type. This is the Borel-Weil theorem.
Example 2.11.
For , then it is easy to see that .
Lemma 2.12.
As spaces, .
Lemma 2.13.
If acts on , then any -equivariant map identifies with an associated bundle .
Note that there is a natural symplectic structure on and that is a resolution. Here, over a regular nilpotent matrix, the map is an isomorphism and the map is proper because fibers is projective.
We will call the partition subregular and the partition minimal. In general, these classes exist for all semisimple Lie algebras and we can perform the Springer construction for all of them. We now guess that is the number of irreducible components of .
Remark 2.14.
Semisimple Lie algebras are the same thing as Dynkin diagrams, which are displayed below:
Theorem 2.15.
All irreducible components of Springer fibers are equidimensional.
We now guess that . However, it is not clear what the action is. For this, we will need a continuity argument. For this, we will need sheaf cohomology.
Theorem 2.16.
is paved. This means that it is a disjoint union of affine spaces.
Being paved means that it is easy to compute the cohomology with coefficients in .
2.3 Sheaves
A presheaf on a space valued in a category is a presheaf . A presheaf is a sheaf if for an open set and a cover with sections such that , then there exists a unique section such that .
Let be a morphism of spaces. Then there is a pushforward
and a pullback
There operations define functoriality for categories of sheaves.
If is an abelian category, then so is , so we may consider complexes of sheaves. Then we may consider the derived category of complexes. In this setting, we have derived versions of as well as the compactly supported and exceptional inverse image functor .
Given a sheaf and a point , the stalk is defined as the colimit
Alternatively, given the inclusion , the stalk is simply . Then we can define the support of a sheaf as the closure of the set of points with nonzero stalks.
Remark 2.17.
The derived category is not an abelian category, but it is an triangulated category. This means we have exact triangles
There is a Verdier duality1 functor such that and .
Example 2.18.
If , then and , the dualizing sheaf.
Under base change, we have the standard push-pull trick, and if is proper, then . If is a smooth morphism, then , where is the orientation sheaf.
Note that cohomology is simply the derived functor of global sections, so we will treat everything as a derived functor.
Definition 2.19.
A local system on is a locally constant sheaf, and a sheaf is constructible if there exists a stratification such that is a local system.
Remark 2.20.
All of the derived category and functoriality constructions apply in the case of constructible sheaves.
2.4 Springer Fibers Again
Returning to our representation theory, we want to produce from our setup. Therefore, we will push the constant sheaf forward from to . The resulting complex of sheaves is called the Grothendieck sheaf.
Remark 2.21.
Note that is embedded in and is embedded in the normalizer of the standard Cartan. In addition, .
If is the projection that forgets the flag, denote the Grothendieck sheaf by . Then the stalk is the same as the cohomology . We know that acts on , so we need to extend this action.
Proposition 2.22.
is obtained from its restriction to the regular semisimple matrices by a procedure called “intersection cohomology extension,” and this works because is “small,” which means that the fibers increase slowly in some sense.
Definition 2.23.
A morphism is semi-small if is generically zero and the closed set where has codimension at least . The morphism is small if it is semi-small and the set of points with has codimension at least .
Theorem 2.24.
The morphism is small.
This means that acts on the entire Grothendieck sheaf , so in particular, it acts on the stalks. This tells us that each stalk is a representation of .
Lemma 2.25.
There is an isomprhism .
We conclude from our geometric construction that
- 1.
Interesting objects have geometric realizations.
- 2.
Geometric methods then give construction of irreducible representations.
- 3.
This allows us to study more efficiently.
3 Lie Algebras
3.1 Basic Notions
First, we will define manifolds over . These are topological spaces locally isomorphic to with transition functions that are (, analytic, holomphic). An example of a manifold is a Riemann surface. Vector fields are simply derivations of and the tangent space at is simply the derivations at that point.
Example 3.1.
If , then the space of vector fields is simply and the tangent spaces are .
The tangent sheaf is locally free, so it is the sheaf of sections of the tangent bundle. If is the local ring at the point and is the maximal ideal, the cotangent space is isomorphic to .
Definition 3.2.
A Lie Group is a group object in the category of manifolds.
Example 3.3.
The group is a Lie group, given by the equation in ambient space . Similarly, the group is given by , and the group is the set of invertible matrices that preserve the symplectic form. Finally, the group is the group of matrices that preserve a nondegenerate symmetric bilinear form.
Remark 3.4.
A symplectic form is a nondegenerate alternating bilinear form.
Definition 3.5.
A Lie algebra over is a vector space together with an anticommutative product that satisfies the Jacobi identity.
Example 3.6.
Any associative algebra can be made into a Lie algebra with the commutator as Lie bracket. Similarly, if is an associative algebra, then is a Lie algebra with the commutator as Lie bracket.
Example 3.7.
Let be a Lie group. Then the space is a Lie algebra and is the same as the left-invariant vector fields. We will denote this Lie algebra by .
- 1.
It is easy to see that .
- 2.
The Lie algebra of is simply the set of trace zero matrices.
- 3.
The Lie algebra is the set of matrices where .
- 4.
The orthogonal Lie algebra is the set of matrices where .
Theorem 3.8.
There is an exponential map such that
- 1.
;
- 2.
is the identity. In particular, is an isomorphism near and ;
- 3.
. This tells us that the Lie algebra commutator is an infinitesimal version of the group commutator.
Theorem 3.9.
We will see that controls .
- 1.
Any morphism of Lie groups is determined on by .
- 2.
If is a map of Lie algebras and , then for a unique map of groups .
The universal cover of any connected Lie group is still a Lie group, so we can preserve most of the information by studying simply connected Lie groups.
3.2 Representation Theory and Classification
A representation of a Lie group is a morphism and a representation of a Lie algebra is a morphism .
Lemma 3.10.
Representations of differentiate to representations of . In particular, if is simply connected, differentiation is an equivalence of categories.
A Lie algebra is simple if it has no nontrivial ideals such that . A solvable Lie algebra is a Lie algebra where the series eventially arrives at .
We will now state the classification of finite-dimensional semisimple Lie algebras over . These are simply sums of simple Lie algebras, so we simply need to classify the simple Lie algebras, which are simply given by Dynkin diagrams:
The correspondence is given by the vertices corresponding to simple roots in .
Definition 3.11.
A connected Lie group is semisimple if its Lie algebra is semisimple.
Our slogan will be that . A Kac-Moody Lie algebra is a generalization of the semisimple or affine Lie algebras. An important example of such an object is the loop group. For a real manifold , the loop space of is . For a complex manifold , then , and finally if is an algebraic variety, .
If is a Lie group, then is an infinite-dimensional version of a Lie group and . If is an algebraic group, then for any ring , where we think of and as functors.
Now we will consider the representation theory of . Note that . The torus is a complex Lie group . We will consider a maximal torus . There are also called Cartans.
Theorem 3.12.
All maximal tori are conjugate to the set of diagonal matrices.
We will denote the set of roots of in by . We will aim to describe in terms of .
Lemma 3.13.
Let be a compact Lie group. Then it has a left-invariant Haar measure that is unique up to constants and is semisimple.
Proof.
We need to find a left-invariant differential form. Fortunately, the tangent and cotangent bundles are trivial, so in particular, the canonical bundle is trivial.
Once we have a Haar measure, choose an inner product . Then we can write an invariant inner product by
Then we finish the proof as in the case of finite groups. □
Corollary 3.14.
Let be a connected complex Lie group with a compact real form . Then let and be the -invariant hermitian inner product on . Then is invariant under .
Proof.
Because is connected, we can transfer the problem to the Lie algebra. The condition that becomes . The other direction is true when because generates . Therefore, we can reduce the problem to because is its complexification. However, we can finally reduce from to , so we are done. □
Example 3.15.
Let , Then is compact.
Lemma 3.16.
For any finite dimensional representation of a torus,
Here .
Note that any character is given by for some , so we see that .
Because has a compact real form , this means that is a semisimple category. Then , and there is an isomorphism given by . Therefore . In particular, for , , so it is simply connected. We also have a fiber sequence as below:
Given a maximal torus , its Lie algebra acts on . If acts by , then acts by , so . We will denote these functionals by , so define
Lemma 3.17.
The eigenspaces of and are the same. In particular, differentiation is injective.
Lemma 3.18.
For all there exists a unique such that . We denote .
For , we see that , so . If we set , then we see that . Alternatively, we can differentiate at .
We will define the set of coroots by . Then it is easy to see that and are dual lattices under the pairing
| given by |
Corollary 3.19.
- 1.
Differentiation at gives an injection ;
- 2.
Differentiation at the identity gives an injection ;
- 3.
The map given by is an isomorphism.
We will now study the system of roots. Let . Then for , write
A Cartan subalgebra of a Lie algebra is a maximal toral subalgebra. Then for all , the Lie bracket is a semisimple operator.
Lemma 3.20.
If is a Cartan, then
Example 3.21.
In both and , the maximal torus is the set of diagonal matrices, and for , is the set of diagonal matrices. In , the Cartan subalgebra is the set of trace zero diagonal matrices.
Define the set of roots . This tells us that and that for , if and , then .
3.3 Representation Theory of
Lemma 3.22.
- 1.
For , ;
- 2.
The weight spaces are simply ;
- 3.
;
- 4.
For all , ;
- 5.
If , then .
Lemma 3.23.
- 1.
If but , then ;
- 2.
.
In the case , we can write a basis for :
with the identities
Lemma 3.24.
Any root defines an -subalgebra .
- 1.
There exists a Lie algebra map with , ;
- 2.
The image is a Lie subalgebra;
- 3.
The image is isomorphic to ;
- 4.
is independent of .
Proof.
The first three parts are easy. Suppose and . Then
so . □
Corollary 3.25.
We see that .
Now we will consider reflections. If and , then if , there is an involution . Then we see that
The group generated by all reflections is called the Weyl group of . For , .
Lemma 3.26.
For and , .
Lemma 3.27.
has a -invariant inner product.
The inner product is simply the standard inner product on . This restricts to , so we can compute the inner product of the indices. An inner product of corresponds to an angle of , corresponds to an angle of , and corresponds to an angle of .
Proposition 3.28.
Define the coroot .
- 1.
For all , ;
- 2.
For all , .
- 3.
spans .
We will now describe the irreducible representations of . Note that and . Thus we will see that the irreducible representations of are simply natural numbers, where
We see that increses the weight by and decreases the weight by . In addition, if , then
where . The representation given above is the only irreducible representation of with highest weight . In general, we have irreducible representations with highest weight .
Note that is a direction in , which comes from a choice of Borel, so the roots come from .
For , define the dominant cone as the set of weights which are positive with respect to all roots in .
Definition 3.29.
In , we say that if is in the -span of .
Definition 3.30.
In a representation , is primitive of weight if for some maximal weight .
Theorem 3.31.
For all , there exists a unique representation of such that
- 1.
has highest weight ;
- 2.
There exists a primitive vector of weight , where .
Proof.
We know that
Because is maximal, then . □