Flag Varieties

Luca Schaffler
December 2, 2013

1 A brief review about Grassmannians

Fix two integers k, n with 0 ≤ k ≤ n. The Grassmannian Gr(k, n) is defined (set theoret-
ically) as the collection of all k-dimensional vector subspaces of Cn . The Grassmannian
Gr(k, n) is more than just a set as we are going to explain now. It’s possible to define a
map: ^k 
ϕ : Gr(k, n) → P Cn
in the following way. Given Λ ∈ Gr(k, n), we can take a basis of Λ, say {v1 , . . . , vk }. Then
define:
ϕ(Λ) := [v1 ∧ . . . ∧ vk ].
This map doesn’t depend on the V choice
 of the basis and it’s injective. Moreover its image
k n
is a Zariski closed subset of P C (see [J, Lecture 6]). From this map, which is called
the Plücker embedding, Gr(k, n) inherits the structure of a projective algebraic variety.

About this structure of algebraic variety, we will just remark that:

dim(Gr(k, n)) = k(n − k).
Proving this fact is an easy exercise: dimension k subspaces of Cn correspond to C-linear
surjections Cn → Cn−k . To conclude we observe that the linear space of (n−k)×n complex
matrices of maximum rank depends on k(n − k) parameters.

2 Flag varieties: a geometric description

Definition 1. Let V be an finite dimensional complex vector space. A flag in V is a
strictly increasing sequence of vector subspaces:
{0} ( Λ1 ( . . . ( Λ` ( V.
The signature of the flag is defined to be the sequence (dim(Λ1 ), . . . , dim(Λ` ), dim(V )).

1
Now take a1 , . . . , a` , n integers with 0 < a1 < . . . < a` < n. Define F(a1 , . . . , a` ; n) to be the
set of all possible flags in Cn with signature (a1 , . . . , a` , n). Observe that F(a1 , . . . , a` ; n) is
contained in Gr(a1 , n) × . . . × Gr(a` , n) and, in the case ` = 1, F(a1 ; n) = Gr(a1 , n).

As in the case of the Grassmannian variety, also F(a1 , . . . , a` ; n) has the structure of a
projective variety.

Proposition 1. F(a1 , . . . , a` ; n) is a Zariski closed subset of Gr(a1 , n) × . . . × Gr(a` , n).

Proof. We already know this for ` = 1. Assume this is true for ` = 2. For any 1 ≤ i < j ≤ `,
let πij be the restriction to F(a1 , . . . , a` ; n) of the projection Gr(a1 , n) × . . . × Gr(a` , n) →
Gr(ai , n) × Gr(aj , n). Then:
\
F(a1 , . . . , a` ; n) = πij−1 (F(ai , aj ; n)),
1≤i<j≤n

and we are done.

The ` = 2 case is stated in [J, Lecture 8] as an exercise.

Now that we know that F(a1 , . . . , a` ; n) has the structure of a variety, we will call it a flag
variety. We can compute the dimension of this variety recursively as follows: let π1 be
the restriction to F(a1 , . . . , a` ; n) of the first projection map Gr(a1 , n) × . . . × Gr(a` , n) →
Gr(a1 , n). π1 is obviously surjective and the generic fiber is isomorphic to F(a2 −a1 , . . . , a` −
a1 ; n − a1 ). Therefore, using what we know about the dimension of the Grassmannian
variety:

dim(F(a1 , . . . , a` ; n)) = dim(Gr(a1 , n)) + dim(F(a2 − a1 , . . . , a` − a1 ; n − a1 )) =

= a1 (n − a1 ) + dim(F(a2 − a1 , . . . , a` − a1 ; n − a1 )) = . . .
`
X
... = (ai − a1 − . . . − ai−1 )(n − ai ).
i=1

3 Flag varieties and algebraic groups

3.1 Quotients of algebraic groups
Here we give a second description of flag varieties by means of the theory of algebraic
groups. Our base field will be the complex numbers. We recall that an algebraic group is an
affine group scheme. A group scheme G is a scheme together with morphisms m : G×C G →

2
G, i : G → G and  : Spec(C) → G that satisfy some commutative diagrams which reflect
the group axioms. For more details about schemes and group schemes see [Ha, Chapter
II] and [MFK, Chapter 0].

If we have a group scheme G acting on a scheme X, it will be very important to know how
to take the quotient of X modulo the action of G in a meaningful way. For example we
would like our quotient to be again a scheme. We give the following definition.

Definition 2. Let G be a group scheme acting on X. A categorical quotient of X by G is

a pair (Y, φ) where Y is a scheme and φ : X → Y is a morphism such that:

(1) the following diagram commutes

σ
G ×C X X
p2 φ
X Y
φ

where σ is the action morphism and p2 is the usual projection on X.

(2) if (Y 0 , φ0 ) is a second pair satisfying (1), then there exists a unique morphism ψ : Y →
Y 0 such that φ0 = ψ ◦ φ.

The categorical quotient may not exist, but if it does, it’s easy to see it is unique. Luckily,
categorical quotients exist in the case that interests us.

The case we care about is the following. Assume G is an algebraic group and let H be a
subgroup of G, i.e. a closed subscheme which is also a subgroup by looking at the group
structure of G. Then obviously we have an action of H on G. The result is

Theorem 1. The categorical quotient of G by H exists. It will be denoted by (G/H, π).

Proof. See [CH, Theorem 3.7.7].

It’s pretty natural to ask what happens if we quotient our algebraic group G by a normal
subgroup N .

Theorem 2. The categorical quotient G/N is an algebraic group.

Proof. See [CH, Corollary 3.7.4].

3
3.2 Special subgroups of an algebraic group
The goal here is to define Borel subgroups and parabolic subgroups of an algebraic group
G and state their first properties. Later we will use them to describe flag varieties.

Definition 3. A Borel subgroup B of an algebraic group G is a maximal subgroup among

the ones that are connected and solvable.

Theorem 3. The Borel subgroups of an algebraic group G are all conjugate and, given a
Borel subgroup B, G/B is projective.

Proof. See [Hu, 21.3].

Definition 4. A parabolic subgroup P of an algebraic group G is any subgroup such that

G/P is projective. Alternately, a parabolic subgroup is any subgroup containing a Borel
subgroup.

Observation 1. Borel subgroup ⇒ parabolic subgroup.

The following fact is claimed in [FH, page 384].

Theorem 4. Let B be a Borel subgroup and P be a parabolic subgroup of an algebraic

group G. Then there exists an x ∈ G such that:

B ⊆ xP x−1 .

In group theory, given a group G and a subgroup H, the normalizer of H in G is the

biggest subgroup of G containing H in which H is normal. It’s denoted by NG (H).

Theorem 5. Let B be a Borel subgroup of an algebraic group G. Then NG (B) = B.

Proof. See [Hu, 23.1, 23.2].

It follows from the previous theorem that, in general, Borel subgroups are not normal.
Moreover, we have:

Corollary 1. Let P be a parabolic subgroup of an algebraic group G. Then P = NG (P ).

In particular P is connected.

Proof. See [Hu, 23.1].

4
3.3 Flag varieties as quotients of algebraic groups
Here we will finally establish the connection between flag varieties and quotients by parabolic
subgroups. First we give a definition.

Definition 5. Let V be a finite dimensional vector space. A full flag in V is a flag with
signature (1, 2, . . . , dim(V )). Given a positive integer n, F(1, . . . , n − 1; n) is called a full
flag variety.

So take a positive integer n. Let {e1 , . . . , en } be the canonical basis for Cn . Call F the full
flag {0} ( {e1 } ( {e1 , e2 } ( . . . ( {e1 , . . . , en }. By change of basis, it’s obvious that any
other possible full flag is obtained from F by the action of GLn .
But there’s something to notice: different elements of GLn can possibly give the same full
flag. Let’s explore this fact. Consider the Borel subgroup B of GLn given by the upper
triangular matrices (the fact that B is a Borel subgroup is left as an exercise). Given any
full flag it’s straightforward to verify that B leaves it invariant. In conclusion we have that
GLn /B parametrizes all the possible full flags without repetitions. In other words we have
that:
F(1, . . . , n − 1; n) ∼ = GLn /B.
More generally, any flag variety F(a1 , . . . , a` ; n) can be described as GLn modulo an ap-
propriate parabolic subgroup. More details about this can be found in [FH, page 96].
Conversely, it’s not that hard to see that any parabolic subgroup of GLn is the stabilizer
of some flag. As a conclusion we can state the following

Theorem 6. Let n be a positive integer. There is a one-to-one correspondence between

parabolic subgroups of GLn and flag varieties F(a1 , . . . , a` ; n) for all possible choices of
integers ` ≥ 1 and 0 < a1 < . . . < a` < n.

3.4 Generalization of flag varieties

We just argued that the study of flag varieties (as we meant in section 1) is equivalent to
the study of quotients of GLn by its parabolic subgroups. But what if instead of GLn we
pick any algebraic group G? From now on we will adopt a more general definition of flag
variety, which is the following.

Definition 6. A flag variety is the quotient of an algebraic group by a parabolic subgroup.

We will call it a full flag variety if we are quotienting by a Borel subgroup.

In the next section, we are going to study the cohomology of invertible sheaves on flag
varieties (with some additional hypotheses).

5
4 Cohomology of flag varieties
4.1 Some more definitions and facts
Proposition 2. Any algebraic group has a unique largest normal solvable subgroup.

Proof. See [Hu, 19.5].

Definition 7. Let G be an algebraic group. The radical of G, denoted by R(G), is defined

to be the identity component of the unique largest normal solvable subgroup.

Definition 8. Assume G is a nontrivial connected algebraic group. Then G will be called

semisimple if R(G) is trivial.

We recall now the following fact from algebraic groups (a reference will be [M, Chapter
4, Section 2]). Take an algebraic group G. Let Lie(G) be the tangent space of G at
the identity. Lie(G) is contained in the distribution algebra of G, denoted by H(G).
The product in H(G) is called the convolution product. From H(G) we have that Lie(G)
inherits the structure of a Lie algebra with the commutator bracket. We have the following
result.

Theorem 7. If G is a semisimple algebraic group then Lie(G) is a semisimple Lie algebra.

Proof. See [Hu, 13.5]. Here actually more is proved, but we won’t need more than what
we stated.

So now take a semisimple algebraic group G. Lie(G) is a semisimple Lie algebra, and we
know everything about it: take a Cartan subalgebra h, let Φ be the corresponding root
system, fix a base ∆ := {α1 , . . . , α` } and let W be the Weyl group. Moreover we can
consider all the integral weights (λ ∈ h∗ s.t. λ(αi ) ∈ Z ∀i) and the integral dominant
weights (λ such that λ(αi ) is a nonnegative integer ∀i). Let’s recall how the Weyl group
acts on integral weights. To do this, it is enough to see how reflection act on an integral
weight. So let α ∈ Φ and let λ be an integral weight. Then:

sα (λ) = λ − hλ, α∨ iα.

Define also the following action of W on the integral weights. If w ∈ W and λ is an integral
weight, let:
w • λ := w(λ + ρ) − ρ,
where ρ is half of the sum of all positive roots.

6
4.2 Invertible sheaves on a full flag variety
Consider a semisimple algebraic group G and let B be a Borel subgroup. So we have a
full flag variety G/B. Let λ be an integral weight corresponding to the Lie algebra Lie(G).
Let C be a B module under the following action. If b ∈ B and z ∈ C, set:

b · z := λ(b)z

(here there’s a subtle identification of λ with a morphism of algebraic groups B → Gm = C∗

that I intentionally skip). So we have a line bundle:

G ×B C−λ → G/B.

Call L(λ) the corresponding invertible sheaf in Pic(G/B). The important statement here
is that any invertible sheaf on G/B can be obtained in this way (this fact is stated in [L,
page 1]).

4.3 Borel-Bott-Weyl theorem

Recall the following important fact from Lie algebras. Let g be a semisimple Lie algebra and
let λ be a dominant weight. Then there is a unique (up to isomorphism) finite dimensional
irreducible g-module V (λ) which has highest weight λ (see [EW, Theorem 15.5]). Now we
are ready to state our main theorem.

Borel-Bott-Weil theorem. Let G be a semisimple algebraic group, B ⊆ G a Borel

subgroup and let λ be an integral dominant weight. Let w be an element of the Weyl group
W. Then:
V (λ)∗ if p = `(w),

H (G/B, L(w • λ)) ∼
p
=
0 otherwise.

Proof. See [K, Theorem 7]. Recall that `(w) is the length of w, i.e. the minimal amount
of reflections whose composition is w.
Reality check: If w and λ are given in such a way that L(w • λ) ∼ = OG/B (the structure
sheaf over G/B), then we expect the BBW theorem to give us H (G/B, L(w • λ)) ∼
0
=C
since G/B is projective (here we just care about isomorphism as complex vector spaces).
Let’s assume w = id. Therefore we need to find λ such that L(λ) ∼ = OG/B . This happens
when λ = 0. Therefore the BBW theorem tells us that H 0 (G/B, OG/B ) ∼ = V (0)∗ . Therefore
we need to check that V (0)∗ is isomorphic to C as complex vector spaces. But C as a
Lie(G)-module with the trivial representation is definitely irreducible with only weight 0.
By uniqueness of V (0), we argue that V (0) ∼
= C ⇒ H 0 (G/B, OG/B ) ∼ = V (0)∗ ∼
= C∗ ∼= C.

Acknowledgements: I am grateful to Will Hardesty for all his explanations and sugges-
tions.

7
References
[CH] T. Crespo, Z. Hajto. Algebraic Groups and Differential Galois Theory, Volume 122
of Graduate studies in mathematics, American Mathematical Soc., 2011.

[EW] K. Erdmann, M. Wildon. Introduction to Lie Algebras, Springer undergraduate

mathematics series, 2006.

[FH] W. Fulton, J. Harris. Representation Theory: A First Course, Volume 129 of Grad-
uate texts in mathematics: Readings in mathematics, Springer study edition, 1991.

[Ha] R. Hartshorne. Algebraic Geometry, Springer-Verlag, New York, 1977, Graduate

Texts in Mathematics, 52.

[Hu] J. Humphreys. Linear Algebraic Groups, Volume 21 of Graduate Texts in Mathemat-

ics, Springer, 1975.

[J] J. Harris. Algebraic Geometry: A First Course, Springer-Verlag, New York, 1992, Grad-
uate Texts in Mathematics, 133.

[K] S. Kumar. Geometry of Schubert varieties and Demazure character formula, lectures
by Shrawan Kumar during April, 2011 Hausdorff Research Institute for Mathematics
Bonn, Germany notes written by Brandyn Lee.

[L] Z. Lin. The structure of cohomology of line bundles on the flag varieties for some groups
of rank 2, Ph.D. Dissertation, U. Massachusetts, 1989.

[M] S. Mukai. An Introduction to Invariants and Moduli, Volume 81 of Cambridge Studies

in Advanced Mathematics, Cambridge University Press, 2003.

[MFK] D. Mumford, J. Fogarty, F. Kirwan. Geometric Invariant Theory, Springer, 1994,

Vol. 34 of Ergebnisse der Mathematik und ihrer Grenzgebiete: a series of modern
surveys in mathematics.