Math 249B.

Root systems for split classical groups

1. Introduction
The exceptional Lie groups were discovered by searching for groups which would realize
the 5 exceptional reduced irreducible root systems. But the 4 infinite families of “classical”
reduced irreducible root systems An (n ≥ 1), Bn (n ≥ 2), Cn (n ≥ 3), and Dn (n ≥ 4) arise
from explicit split connected semisimple groups.
In this handout, we work out the root systems for the split connected semisimple groups
SLn+1 (n ≥ 1), SO2n+1 (n ≥ 2), Sp2n (n ≥ 2), and SO2n (n ≥ 3) over an arbitrary field k.
This will yield the root system of type An , Bn , Cn , and Dn respectively. In the course of
doing these calculations, we will determine:
(i) an explicit borus (T, B),
(ii) the associated system of positive roots Φ(B, T ) and its root basis ∆,
(ii) the coroot associated to each (positive) root,
(iv) an explicit Weyl-invariant positive-definite symmetric bilinear form on X(T )Q .
Our work will also prove that the special orthogonal and symplectic groups really are
semisimple (granting that they are smooth and connected, as was proved in the previous
course). Exercises 1.6.16 and 1.6.15 in the Luminy SGA3 notes provide an alternative (and
in some respects more efficient/conceptual) approach to proving the reductivity and deter-
mining the root systems for special orthogonal and symplectic groups.

2. Type A
Consider G = SLn+1 with n ≥ 1, and let T be the diagonal split maximal k-torus. We have
seen Example 2.2 of the “Root datum” handout that X := X(T ) is naturally identified with
Zn+1 /Z (quotient by the diagonally embedded Z), so V := XQ = Qn+1 /Q ' (Qn+1 )Σ=0 . We
equip this final hyperplane inside Qn+1 with the restriction of the standard inner product
on Qn+1 . Let e1 , . . . , en+1 denote the images in X of the standard basis of Zn+1 .
We know that the root groups are given by the copies of Ga corresponding to off-diagonal
matrix entries, so Φ consists of the n2 − n characters ei − ej (i 6= j), sending a diagonal t ∈ T
to ti /tj ; these do not span X over Z (indeed, their span has index n + 1, corresponding to
the fact that ZG = µn+1 has order n + 1).
For i < j, the pair of opposite roots ±(ei − ej ) have associated root groups generating the
SL2 ⊂ G supported in the matrix entries indexed by i and/or j (using 1’s elsewhere on the
diagonal and 0’s everywhere else). For i < j, this latter description gives the coroot formula
(ei − ej )∨ = e∗i − e∗j : Gm → T
carrying y to diag(1, . . . , 1, y, . . . , 1/y, 1, . . . , 1) where the only diagonal entries distinct from
1 are y in the ii-position and 1/y in the jj-position. These coroots span the lattice (Zn+1 )Σ=0
naturally dual to X, affirming that G is simply connected.
Using the coroot formula, the reflection rei −ej on V = XQ (defined in terms of coroots) is
easily verified to be the usual orthogonal reflection through ei − ej with respect to the chosen
inner product on V (induced by the standard one on Qn+1 ). Hence, the inner product we
are considering on V is invariant under all of these reflections, so invariant under the entire
1
2

Weyl group. Once we verify that Φ is irreducible we will therefore know that we have found
the Weyl-invariant inner product that is unique up to a positive scaling factor.
For i < j we have
ei − ej = (ei − ei+1 ) + · · · + (ej−1 − ej ),
so the set ∆ = {ai = ei − ei+1 }1≤i≤n of n positive roots generate the rest under repeated
additions. We know that G is connected reductive with finite center, so it is semisimple, and
its rank is n, so ∆ of size n must be a root basis.
Under the chosen Weyl-invariant dot product we have ai · aj 6= 0 for i < j precisely when
j = i + 1, and all squared root lengths ai · ai have the same value (namely, 2). Thus, Dyn(Φ)
is the connected An -diagram (establishing that Φ is irreducible):

◦ ◦ ◦ ◦ ··· ◦
a1 a2 a3 a4 an

Note in particular that

2 cos(∠(ai , ai+1 )) = hai , a∨i+1 i = −1,
so the angle between ai and ai+1 is 2π/3. For n = 2 this recovers the hexagonal picture for
the root system A2 equipped with its Weyl-invariant Euclidean structure that is unique up
to scaling.

3. Type B
Fix n ≥ 2. Let V = k 2n+1 with the non-degenerate quadratic form
q = x0 x2n + · · · + xn−1 xn+1 + x2n = (x0 , . . . , xn−1 )J(xn+1 , . . . , x2n )t + x2n
where J is the anti-diagonal n × n matrix with 1’s along the anti-diagonal. Denote by
{e0 , . . . , e2n } the standard basis of V . The defect space V ⊥ vanishes if char(k) 6= 2 and is
the line ken if char(k) = 2. A subspace V 0 ⊂ V is called isotropic if q|V 0 = 0.
Define G to be the affine k-group scheme SO2n+1 := SO(q) = O(q) ∩ SL(V ). From the
first course we know that this is smooth and connected with dimension n(2n + 1). We will
prove below that it is semisimple with trivial center and has root system Bn . This will entail
some hard work, but with this experience under our belts the cases of types C and D will
be smooth sailing.
The split diagonal torus
T = {diag(t1 , . . . , tn , 1, t−1 −1
n , . . . , t1 )} ⊂ G

is easily verified (using weight space considerations for the standard representation on k 2n+1 )
to satisfy ZG (T ) = T , so T is a maximal torus. The choice of indexing of the variables
appearing in q will ensure that an appropriate “upper triangular” subgroup will be a Borel
subgroup containing T .
To discover a Borel subgroup containing T , we briefly digress to discuss some related “flag
varieties”. The isotropic subspace
Wstd := span(e0 , . . . , en−1 )
 3

of dimension n is certainly maximal as such since q is non-degenerate and q|ken 6= 0 in

characteristic 2. By similar reasons, every isotropic subspace of V has vanishing intersection
with the defect space (a non-tautology only in characteristic 2).
Let Y be the scheme (built from Grassmannians) representing the functor of pairs (W, F )
consisting of a rank-n isotropic subbundle W of V and a full flag F of W . The scheme Y
inherits properness from Grassmannians and G naturally acts on Y . We claim that G acts
transitively on Y . For this purpose, it suffices to show that G(k) acts transitively on the set
of n-dimensional isotropic subspaces of Vk because the G-stabilizer of Wstd acts transitively
on the variety of its full flags (since the isotropicity of Wstd provides points of G = SO(q)
preserving Wstd through whatever linear automorphism we wish:
g ~0
 
0n
M (g) :=  0 1 0 
0n ~0 J(g ) J
t −1

for g ∈ GL(Wstd ) = GLn ).

To verify the transitivity of the G(k)-action on the set of n-dimensional isotropic subspaces
of Vk we now recall an important fact:
Theorem 3.1 (Witt Extension Theorem). Let (W, Q) be any (possibly degenerate!) finite-
dimensional quadratic space over a field K. If W 0 , W 00 ⊂ W are subspaces and W 0 ∩ W ⊥ =
0 = W 00 ∩ W ⊥ then any isometry W 0 ' W 00 extends to an element of O(W, Q)(K).
Proof. For a proof which (unlike most references) permits characteristic 2 and imposes no
non-degeneracy conditions on Q or parity conditions on dim W , see Theorem 8.3 in the book
“The Algebraic and Geometric Theory of Quadratic Forms”. The proof proceeds by induc-
tion on dim W , and such induction generally cannot preserve non-degeneracy hypotheses, so
for the purposes of the proof it is a virtue that we impose no non-degeneracy (nor dimension
parity) conditions on (W, Q). 
All isotropic subspaces W of V with a given common dimension are isometric since q|W = 0
for all such W , so if W, W 0 ⊂ V are isotropic with dim W ≤ dim W 0 then O(q)(k) carries
W into W 0 (and onto W 0 when dim W = dim W 0 ). Hence, all maximal isotropic subspaces
of V have the same dimension, so that dimension must be n (as Wstd is maximal isotropic),
and O(q)(k) acts transitively on the set of these. But oddness of dim V implies that O(q) =
µ2 × SO(q) as group schemes, where µ2 acts on V through scaling, so the transitivity of
O(q)(k) on the set of maximal isotropic subspaces implies the same for SO(q)(k) = G(k).
Applying this over k then gives the desired transitivity of the G-action on Y .
It follows that for the standard full flag Fstd = {ke0 ⊂ · · · ⊂ Wstd } in Wstd , the G-stabilizer
B of (Wstd , Fstd ) satisfies G/B ' Y , so G/B is proper. Explicitly, if we define
u ~0
   
0n 1n −2Jv L
M (u) = ~0 t 1 ~0 t  , h(v, L) := ~0 t 1 vt 
0n ~0 J(ut )−1 J 0n 0 1n
for upper triangular unipotent u ∈ GLn , v ∈ k n , and L ∈ Matn then
t
B = T n {M (u) n h(v, L) | v 0 (Lt J + vv t )v 0 = 0 for all v 0 ∈ k n }
4

with Lt J generally not symmetric. For instance, with v = 0 the condition on L = (`ij ) ∈
Matn is that it is “alternating with respect to the anti-diagonal”: its anti-diagonal vanishes
and it is skew-symmetric relative to flipping across the anti-diagonal (i.e., `ij = 0 when
i + j = n + 1 and `i,n+1−j = −`j,n+1−i when i + j 6= n + 1).
The defining condition on pairs (v, L) is linear in L for a given v, so the scheme-theoretic
description of B is smooth and connected (of dimension n2 + n due to the description in
terms of T , u ∈ Un , v ∈ k n , and L), so properness of G/B implies that B is parabolic.
Explicitly, B has a composition series with solvable successive quotients given by: the space
of L ∈ Matn that are alternating with respect to the anti-diagonal, k n (corresponding to v),
Un (corresponding to u), and T , so B is solvable and hence is a Borel subgroup of G!
The definition of B has an evident analogue B − by reflecting conditions across the main
diagonal, and B ∩ B − = T is a torus, proving that G is connected reductive. It also follows
from dimension considerations that the size of a positive system of roots Φ+ = Φ(B, T ) is
n2 , so there are n2 roots supported in the Lie algebra of the unipotent radical of B; we seek
to find these roots.
There are (n2 − n)/2 such roots given by ti /tj for i < j, namely from the standard root
groups contained in Un (setting v and L to vanish), and n more given by ti (1 ≤ i ≤ n)
through the T -action on the coordinate vn+1−i of v ∈ k n using the evident T -equivariant
subquotient Lie(k n ) of Lie(B) (parameterized by v ∈ k n ). Finally, we get (n2 − n)/2 more
roots ti tj for unordered pairs of distinct i, j ∈ {1, . . . , n} with i + j 6= n + 1, using as the root
line the coordinate `i,n+1−j = −`j,n+1−i in the Lie algebra of the space of L ∈ Matn that are
alternating with respect to anti-diagonal flip.
The collection of n such positive roots
∆ := {ai = ti /ti+1 }1≤i≤n−1 ∪ {an = tn }
is easily checked to generate the rest of Φ(B, T ) under repeated additions, and the connected
semisimple group G has rank dim T = n, so these n roots constitute a root basis. By
inspection, ∆ spans X(T ) = Zn with standard basis {e1 , . . . , en } relative to which ai =
ei − ei+1 for 1 ≤ i ≤ n − 1 and an = en . In particular, ZG = 1.
To compute the coroot a∨i we need to determine Gai = hUai , U−ai i (e.g., is this SL2 or
PGL2 ?). For 1 ≤ i < n we have Gai = SL2 inside SL(Wstd ) = SLn using the ith and (i + 1)th
rows and columns, so a∨i = e∗i − e∗i+1 inside X∗ (T ). The case i = n requires some more work
involving how the coroots for a split reductive pair are defined. We have
Gan := hUan , U−an i = SO(xn−1 xn+1 + x2n ) = Aut(sl2 , det)
where sl2 consists of matrices of the form
 
xn xn−1
.
xn+1 −xn
The natural isomorphism PGL2 ' Gan = SO3 induced by the det-invariant GL2 -conjugation
on Mat2 ⊃ sl2 satisfies
 
    1 x2 x − x2
t 0 1 x
7→ diag(t, 1, 1/t) ∈ T ∩ Gan , 7 0 1
→ x  ∈ Uan ,
0 1 0 1
0 0 1
 5

so this isomorphism carries the diagonal of PGL2 over to a∨n (GL1 ) and carries the upper tri-
angular unipotent subgroup over to the root group for an (recall the unique characterization
of root groups). Hence, the standard coroot formula t 7→ diag(t2 , 1) for PGL2 implies that
a∨n = 2e∗n .
With the coroots determined for each root in ∆, we see that Z∆∨ has index 2 inside X∗ (T )
(so Ge is a central double cover of G). Likewise, the orthogonal reflection on Qn in each ai
relative to the standard dot product on Qn = X(T )Q is directly verified to calculate the
effect of rai on ∆ (the ra ’sdefined via coroots, not via Euclidean geometry!), so the standard
dot product is Weyl-invariant. But ai · aj 6= 0 for 1 ≤ i < j ≤ n if (and only if) j = i + 1, so
the Dynkin diagram is connected. The root system is therefore irreducible, so the standard
dot product is the canonical (up to positive scaling) Euclidean structure.
By inspection of our list of all positive roots, it follows that there are n short positive roots
and n2 − n long positive roots, with ratio of square root lengths equal to 2 (long divided by
short). Hence, Dyn(Φ(G, T )) is the Bn -diagram:

◦ ◦ ◦ ◦ ··· ◦ +3 ◦
a1 a2 a3 a4 an−1 an

4. Type C
Fix n ≥ 2, and define G = Sp(ψ) where ψ is the alternating form on V = k 2n given by
the matrix  
0 1n
J= .
−1n 0
In other words, G is the functor of points g ∈ GL2n such that g t Jg = J. By the first
course, we know G is smooth and connected. We say that a subspace W ⊂ V is isotropic if
ψ|W ×W = 0.
It is easy to check that the group of diagonal matrices T ⊂ GL2n having the form
 −1 
t 0
0 t
for diagonal t ∈ GLn is contained in G, and by studying weights for the standard represen-
tation of T on k 2n it is not difficult to verify that ZG (T ) = T . Thus, T is a maximal torus
of G. Once again, we will prove G is semisimple.
To compute a Borel subgroup B of G containing T we will introduce an appropriate “flag
variety” involving maximal isotropic subspaces, as we did for odd orthogonal groups above.
This requires the following analogue of the transitivity result that we deduced for orthogonal
groups from the Witt Extension Theorem:
Theorem 4.1. All maximal isotropic subspaces W ⊂ V have dimension n, the G(k)-action
on the set of such subspaces is transitive, and if W is such a subspace then StabG(k) (W ) acts
transitively on the set of full flags in W .
This is much easier to prove than the analogue for orthogonal groups, as the structure
of symplectic spaces is much simpler than that of (possibly degenerate!) quadratic spaces.
More specifically, the proof of this Theorem requires nothing more than a review of the proof
6

of the structure theorem for symplectic spaces (over fields), so we leave this as an exercise
for the reader.
As in the odd orthogonal case, we use Grassmannians to build a proper scheme Y repre-
senting the functor of pairs (W, F ) consisting of a rank-n isotropic subbundle of V and a full
flag F of W . Letting {e1 , . . . , e2n } denote the standard basis of V = k 2n , define the maximal
isotropic subspace Wstd = span(e1 , . . . , en ), and let Fstd be its standard full flag (beginning
with the line ke1 ).
One verifies by computation that

(ut )−1 mu
  
B := StabG (Wstd , Fstd ) = T n | m ∈ Symn , u ∈ Un
0 u

(where Symn ⊂ Matn is the subspace of symmetric matrices). This B has a composition
series with successive quotients Symn , Un , and T , so it is smooth and connected of dimension
n2 + n and solvable. By properness of G/B = Y it follows that B is a Borel subgroup
(containing T ). The analogous subgroup B − using the “lower triangular constructions” (i.e.,
the “opposite” maximal isotropic subspace and its standard flag beginning at ke2n ; in effect,
transporting the construction of B for V ∗ via the duality ψ) is also a Borel subgroup, and
B ∩ B − = T is a torus, so this proves that G is reductive.
The n2 positive root groups and associated roots for (B, T ) can be worked out much more
easily than for the odd orthogonal case: we get n(n − 1)/2 roots ti /tj for 1 ≤ i < j ≤ n
from the ij-entry of Un , and n(n + 1)/2 roots (ti tj )−1 for 1 ≤ i ≤ j ≤ n from the common
ij and ji matrix entries in the symmetric m ∈ Symn . In particular, Φ clearly spans X(T )Q ,
so the connected reductive G is semisimple. Under the evident identification of X(T ) with
Zn with standard basis denoted e1 , . . . , en (projection onto matrix entries for t ∈ T ), the
positive roots are ei − ej for i < j and −(ei + ej ) for i ≤ j. The set of n positive roots

∆ = {an−i = ti /ti+1 }i<n ∪ {an = t−2

1 }

generates the rest under repeated additions, so since G is semisimple with rank dim T = n
it follows that ∆ is the root basis for Φ(B, T ).
It is easy to verify that Ga := hUa , U−a i = SL2 (identification using standard matrix
coordinates on G ⊂ GLn ) for all a ∈ Φ+ , with this identification carrying T ∩ Ga over to
the diagonal of SL2 and carrying Ua over to the upper triangular unipotent subgroup of
SL2 , so we see that (ei ± ej )∨ = e∗i ± e∗j for i < j (and likewise for the negative; recall that
(−a)∨ = −a∨ for any element a in a root system), and that (2ei )∨ = e∗i . In particular, ∆∨
spans X∗ (T ), so G is simply connected.
Now that the coroots have been determined, as in the odd orthogonal case we easily
check that the standard dot product on X(T )Q = Qn computes the reflection ra (defined
via coroots) via Euclidean reflection in a. In particular, the standard dot product is Weyl-
invariant. By direct calculation we find once again that for 1 ≤ i < j ≤ n, ai · aj 6= 0 if and
only if j = i + 1, so the Dynkin diagram is connected and hence Φ(G, T ) is irreducible. It
follows that the Weyl-invariant standard dot product provides the canonical (up to scaling)
Euclidean structure, so we can use it to compute root lengths. This yields the Cn -diagram:
 7

◦ ◦ ◦ ◦ ··· ◦ ks ◦
a1 a2 a3 a4 an−1 an

Note that there are n long positive roots t±2

i and n2 − n short positive roots, and the long
ones lie in 2X(T ). By inspection of the classification of reduced irreducible root systems,
simply connected type Cn for n ≥ 1 (where C1 := A1 is treated separately) are the only
semisimple root data (X, Φ, X ∨ , Φ∨ ) for which there is a root that is divisible (in fact, by 2)
in X.

5. Type D
Fix n ≥ 3, and let G = SO(q) for the quadratic form
q = x1 x2n + · · · + xn xn+1
on V = k 2n . By the previous course, this is smooth and connected. This also makes sense
for n = 2, but SO4 = SL2 ×µ2 SL2 (so the Dynkin diagram is the reducible A1 × A1 ).
Our treatment of this case will be a simplified version of the work already done in the odd
orthogonal case; in effect, everything goes similarly except that various matrices no longer
have a “middle column/row” (and the effect on the Dynkin diagram will be to split the long
vertex into two arms associated to roots with the same length as the rest).
The split torus T ⊂ GL2n consisting of points
diag(t1 , . . . , tn , t−1 −1
n , . . . , t1 )

is obviously contained in G, and as in the odd orthogonal case we see that ZG (T ) = T , so T

is maximal in G. In fact, since there is no “middle 1” in the description of points of T , one
can verify that even ZO(q) (T ) is equal to T .
We will again use the Witt Extension Theorem to find a Borel k-subgroup of G containing
T . Let Y be the proper flag scheme defined as in the odd orthogonal case, so for the natural
action of O(q) on Y we see that O(q)(k 0 ) acts transitively on Y (k 0 ) for every field k 0 /k.
(Beware that G does not act transitively on Y . This is due to the disconnectedness of O(q)
being inherited by Y , as we will see below, and so is a fundamental distinction from the odd
orthogonal case.) Define the standard maximal isotropic subspace
Wstd = span(e1 , . . . , en ) ⊂ k 2n = V
and its standard full flag Fstd beginning with ke1 . Define B similarly to the odd orthogonal
case but with no “middle row/column” contribution (so no intervention of v ∈ k n !).
One verifies by similar (but easier) calculations to the odd orthogonal case that
StabO(q) (Wstd , Fstd ) = B
as schemes, and that B is smooth, connected, and solvable. In particular, by quotient consid-
erations for transitive actions by smooth groups we conclude that (O(q)/B)k has underlying
reduced scheme Yk (in fact, O(q) is smooth in all characteristics, being an extension of Z/2Z
by SO(q), but we do not need this). Thus, O(q)/B is proper, so G/B is proper. Hence, B
is parabolic and therefore is a Borel subgroup of G.
8

The positive system of roots Φ+ = Φ(B, T ) comes out exactly as in the odd orthogonal
case except that we do not get ti ’s as roots (due to the absence of the middle column/row in
the present calculations). We again build another Borel subgroup B − satisfying B ∩B − = T ,
so G is reductive. The positive roots span X(T )Q , so G is semisimple. Hence, Φ+ has a root
basis of size n. We verify without difficulty that now the root basis is given by
∆ = {ai = ti /ti+1 }1≤i<n ∪ {an = tn−1 tn }.
Using the usual identification X(T ) = Zn with standard basis {e1 , . . . , en }, we have ai =
ei − ei+1 for 1 ≤ i < n and an = en−1 + en . We need to compute the associated coroots a∨i .
As in the odd orthogonal case we have the same identification Gai = SL2 for i < n, yielding
again that a∨i = e∗i − e∗i+1 for i < n. The computation of Gan turns out differently: it is not
PGL2 as in the odd orthogonal case, but rather is SL2 in a specific manner. Namely, we use
the standard isomorphism
SO(xn−1 xn+2 + xn xn+1 ) = SO4 = Aut(gl2 , det) ' SL2 ×µ2 SL2
to identify Gan with one of these SL2 -factors, and by keeping track of where T ∩ Gan and
Uan go under this identification we obtain
a∨n = e∗n−1 + e∗n .
As in the odd orthogonal case, the standard dot product on X(T )Q = Qn computes the
reflections rai as the Euclidean reflection through ai , so this dot product is Weyl-invariant.
Moreover, now ai · ai+1 = −1 for i < n − 1 whereas an−1 · an = 0 but an−2 · an = −1
(note that an−2 makes sense because n ≥ 3), so the Dynkin diagram is connected. Hence,
Φ(G, T ) is irreducible, so the standard dot product is the unique Weyl-invariant Euclidean
structure up to scaling. By inspection the root lengths for all ai ’s coincide, and we arrive at
the Dn -diagram:

◦
an−1
◦ ◦ ◦ ◦ ··· ◦ ◦
a1 a2 a3 a4 an−3 an−2
◦
an
∨
The determination of ∆ and ∆ yields that the root lattice Z∆ has index 2 inside X(T )
and that the coroot lattice Z∆∨ has index 2 inside X∗ (T ). Hence, ZG has order 2 (so the
evident central µ2 ⊂ G coincides with ZG ) and the simply connected central cover Ge is a
degree-2 cover of G = SO2n .
Remark 5.1. A rather special situation, called triality, arises for the case n = 4: the diagram
for D4 has automorphism group S3 . This is the only reduced irreducible root system whose
diagram has nontrivial automorphisms beyond a single involution. The corresponding group
SO8 then has “more symmetry” than typical special orthogonal groups. This was seen classi-
cal via constructions involving octonion algebras. (Beware however, that the automorphism
group of an octonion algebra much smaller than SO8 : in fact, it is connected semisimple
of type G2 , as explained in 2.3.5 and 2.4.5 of the book “Octonions, Jordan algebras, and
exceptional groups”.)