, ,

The ground field

k

is algebraically closed and of characteristic zero. Let

g

be a reductive algebraic Lie algebra over

k

and

σ

an involutory automorphism of

g

. Then

g = g_{0} \oplus g_{1}

is the direct sum of

σ

-eigenspaces. Here

g_{0}

is a reductive subalgebra and

g_{1}

is a

g_{0}

-module.

Let

G

be the adjoint group of

g

and

G_{0} \subset G

a connected subgroup with

L i e G_{0} = g_{0}

. The commuting variety of

(g, g_{0})

is the following set:

ℭ (g_{1}) = {(x, y) \in g_{1} \times g_{1} | [x, y] = 0} .

The problem whether

ℭ (g_{1})

is irreducible was considered by Panyushev [5] , [6] and Sabourin-Yu [8] , [9] . Suppose

g

is simple. Then the known results are

$∙$ if the rank of the symmetric pair $(g, g_{0})$ is equal to the semisimple rank of $g$ (called the maximal rank case), then the corresponding commuting variety is irreducible, [5] ;
$∙$ if the rank of $(g, g_{0})$ equals $1$ , then $ℭ (g_{1})$ is irreducible only in one case, namely, $(s o_{m + 1}, s o_{m})$ , [6] , [8] ;
$∙$ for $(s l_{2 n}, s p_{2 n})$ and $(E_{6}, F_{4})$ the corresponding commuting variety is irreducible, [6] ;
$∙$ if $(g, g_{0}) = (s o_{2 + m}, s o_{2} \oplus s o_{m})$ , then $ℭ (g_{1})$ is irreducible, [9] .

For all other symmetric pairs the problem is open. Here we extend the result of [9] to all symmetric pairs

(s o_{n + m}, s o_{n} \oplus s o_{m})

. The scheme of the proof is similar to that of [9] . But as it often happens, the argument in a general situation is shorter and simpler, than in a particular case.

1 Commuting variety

The commuting variety

ℭ (g) = {(x, y) \in g \times g | [x, y] = 0}

of a reductive Lie algebra

g

was considered by Richardson in [7] , where he shows that

ℭ (g)

coincides with the closure of

G (t \times t)

for any Cartan subalgebra

t \subset g

, and is therefore irreducible. It is not yet known whether

ℭ (g)

is normal or whether the ideal of

ℭ (g)

is generated by quadrics.

Let

(g, g_{0})

be a symmetric pair and

c \subset g_{1}

a maximal abelian subspace consisting of semisimple elements. Any such subspace is called a Cartan subspace of

g_{1}

. All Cartan subspaces are

G_{0}

-conjugate, see [3] . The dimension of

c

is called the rank of the symmetric pair

(g, g_{0})

. It is well-known that

ℭ_{0} = \bar{G_{0} (c \times c)}

is the unique irreducible component of

ℭ (g_{1})

of maximal dimension, see [5,Sect.3] . Here

dim ℭ (g_{1}) = dim g_{1} + dim c

. It follows that

ℭ (g_{1})

is irreducible if and only if

ℭ (g_{1}) = ℭ_{0}

The irreducibility problem for

ℭ (g_{1})

was first considered by Panyushev in [5] . As was mentioned above,

ℭ (g_{1})

is not always irreducible. On the other hand, in some particular cases one can say more about properties of

ℭ (g_{1})

. If

(g, g_{0})

is a symmetric pair of maximal rank, then

ℭ (g_{1})

is a normal complete intersection and the ideal of

ℭ (g_{1})

k [g_{1} \times g_{1}]

is generated by quadrics, see [5] .

Let

n

be a non-negative integer. The set

g_{1}^{(n)} = {ξ \in g_{1} | dim G_{0} ξ = n}

is locally closed. Irreducible components of

g_{1}^{(n)}

are called

G_{0}

-sheets of

g_{1}

. The following statement is well-known for the specialists, but we include a proof here for the sake of completeness.

Lemma 1. Let

S

be a

G_{0}

-sheet of

g_{1}

containing semisimple elements. Suppose for each semisimple

h \in S

we have

({h} \times (g_{1})_{h}) \subset ℭ_{0}

. Then

({x} \times (g_{1})_{x}) \subset ℭ_{0}

for each

x \in S

Proof. Let $x \in S$ . Since $S$ contains semisimple elements, they form a dense subset. Therefore, we can find a morphism $γ : k \to S$ such that $γ (0) = x$ and $γ (t)$ is semisimple for each $t \neq 0$ .
Then $(g_{1})_{x} = {lim}_{t \to 0} (g_{1})_{γ (t)}$ , where the limit is taken in an appropriate Grassmannian. For each $y \in (g_{1})_{x}$ , we can define elements $y (t) \in (g_{1})_{γ (t)}$ such that $y = {lim}_{t \to 0} y (t)$ . Since $(x, y) = {lim}_{t \to 0} (γ (t), y (t))$ and $(γ (t), y (t)) \in ℭ_{0}$ for each $t \neq 0$ , we conclude that $(x, y) \in ℭ_{0}$ . □

2 Semisimple and nilpotent elements in $g_{1}$

In this section

g = s o_{n + m}

g_{0} = s o_{n} \oplus s o_{m}

. Let

V = k^{n + m}

be a vector space of the defining representation of

g

. Then we have a

G_{0}

-invariant decomposition

V = V_{a} \oplus V_{b}

, where

V_{a} = k^{n}

V_{b} = k^{m}

, and

g_{1} \sim = k^{n} \otimes k^{m}

as a

G_{0}

-module. Denote by

(,)

the non-degenerate symmetric

g

-invariant bilinear form on

V

Let

g_{h}

be the centraliser of an element

h \in g_{1}

. Then

σ

induces the symmetric decomposition

(g_{h}) = (g_{0})_{h} \oplus (g_{1})_{h}

, where

(g_{0})_{h}

is the centraliser of

h

g_{0}

. First we describe centralisers

g_{h}

and

(g_{0})_{h}

of semisimple elements

h \in g_{1}

Lemma 2. Let

h \in g_{1}

be a semisimple element. Then the symmetric pair

(g_{h}, (g_{0})_{h})

is a direct sum

(\oplus_{i = 1}^{r} (g l_{k_{i}}, s o_{k_{i}})) \oplus (s o_{n + m - 2 k}, s o_{n - k} \oplus s o_{m - k})

, where

k = \sum_{i} k_{i}

Proof. Recall several well-known facts about semisimple elements of $s o_{n + m}$ . Let $v_{λ} \in V$ be an eigenvector of $h$ such that $h \cdot v_{λ} = λ v_{λ}$ and $λ \neq 0$ . Since $h$ preserves the symmetric form $(,)$ , we have $(v_{λ}, v_{λ}) = 0$ . Also if $h \cdot v = λ v$ , $h \cdot w = μ w$ , then $(v, w) \neq 0$ only if $λ = - μ$ .
Let ${\pm λ_{i}, 0 | i = 1, \dots, r}$ be the set of the eigenvalues of $h$ . Then there is an orthogonal $h$ -invariant decomposition $V = (V_{λ_{1}} \oplus V_{- λ_{1}}) \oplus \dots \oplus (V_{λ_{r}} \oplus V_{- λ_{r}}) \oplus V_{0} .$ Here each $V_{λ_{i}}$ is an isotropic subspace, $(V_{λ_{i}}, V_{\pm λ_{j}}) = 0$ if $λ_{i} \neq \pm λ_{j}$ and $(V_{0}, V_{\pm λ_{i}}) = 0$ for each $λ_{i}$ . Therefore $g_{h} \subset (\oplus_{i = 1}^{r} s o (V_{λ_{i}} \oplus V_{- λ_{i}})) \oplus s o (V_{0})$ . More precisely, if $dim V_{λ_{i}} = k_{i}$ and $k = \sum_{i} k_{i}$ , then $g_{h} = (\oplus_{i} g l_{k_{i}}) \oplus s o_{n + m - 2 k}$ .
Now it remains to describe $(g_{0})_{h} = (g_{h})^{σ}$ . We may assume that $σ$ is a conjugation by a diagonal matrix $A \in O_{n + m}$ such that $A |_{V_{a}} = - i d$ and $A |_{V_{b}} = i d$ . Since $σ (h) = - h$ , we have $A \cdot V_{λ_{i}} = V_{- λ_{i}}$ and $A \cdot V_{0} = V_{0}$ . Moreover, $A$ determines a non-degenerate symmetric form $(,)_{A}$ on each $V_{λ_{i}}$ by the formula $(v, w)_{A} = (v, A \cdot w)$ . Therefore, each $s o (V_{λ_{i}} \oplus V_{- λ_{i}})$ is $σ$ -invariant, $(s o (V_{λ_{i}} \oplus V_{- λ_{i}}))^{σ} \sim = s o_{k_{i}} \oplus s o_{k_{i}}$ , and $(g l_{k_{i}})^{σ} = s o_{k_{i}}$ . Finally, the restriction $A |_{V_{0}}$ has signature $(n - k, m - k)$ . Thus $(s o_{n + m - 2 k})^{σ} = s o_{n - k} \oplus s o_{m - k}$ . □

Denote by

N (g_{1})

the nullcone of

G_{0} : g_{1}

, i.e., the set of all nilpotent elements in

g_{1}

Recall several standard facts concerning nilpotent elements in

g l (V)

. Suppose

e \in g l (V)

is nilpotent and

m = dim K e r (e)

. Then by the theory of Jordan normal form, there are vectors

w_{1}, \dots, w_{m} \in V

and non-negative integers

d_{1}, \dots, d_{m}

such that

e^{d_{i} + 1} \cdot w_{i} = 0

and

{e^{s} \cdot w_{i} | 1 \leq i \leq m, 0 \leq s \leq d_{i}}

is a basis for

V

. Let

V_{i} \subset V

be a linear span of

{w_{i}, e \cdot w_{i}, \dots, e^{d_{i}} \cdot w_{i}}

. Then the spaces

{V_{i}}

are called the Jordan (or cyclic) spaces of the nilpotent element

e

and

V = \oplus_{i = 1}^{m} V_{i}

Lemma 3. Suppose

e \in N (g_{1})

. Then the cyclic vectors

{w_{i}}_{i = 1}^{m}

and hence the cyclic spaces

{V_{i}}

's can be chosen such that the following properties are satisfied:

(i) there is an involution $i \mapsto i^{*}$ on the set ${1, . . ., m}$ such that $d_{i} = d_{i^{*}}$ , $i = i^{*}$ if and only if $dim V_{i}$ is odd, and $(V_{i}, V_{j}) = 0$ if $i \neq j^{*}$ ;
(ii) $σ (w_{i}) = \pm w_{i}$ .

Proof. Part (i) is a standard property of the nilpotent orbits in $s o (V)$ , see, for example, [1,Sect. 5.1] or [2,Sect. 1] . Then part (ii) says that in the presence of the involution $σ$ cyclic vectors for $e \in N (g_{1})$ can be chosen to be $σ$ -eigenvectors, see [4,Prop. 2] . □

For each

e \in N (g_{1})

we choose cyclic vectors

{w_{i}}

as prescribed by Lemma 3 . Say that

e^{s} \cdot w_{i}

has type

a

σ (e^{s} \cdot w_{i}) = - e^{s} \cdot w_{i}

, i.e.,

e^{s} \cdot w_{i} \in V_{a}

; and

e^{s} \cdot w_{i}

has type

b

e^{s} \cdot w_{i} \in V_{b}

Since

σ (e) = - e

, if

e^{s} \cdot w_{i} \in V_{a}

, then

e^{s + 1} \cdot w_{i} \in V_{b}

and vice versa. Therefore each string

〈 w_{i}, e \cdot w_{i}, \dots, e^{d_{i}} \cdot w_{i} 〉

has one of the following types:

a b a \dots a b

b a b \dots b a

a b a \dots b a

b a b \dots a b

Let

e \in N (g_{1})

. There is an

s l_{2}

-triple

{e, f, h}

such that

f \in g_{1}

and

h \in g_{0}

. Recall that

e

is called even if the eigenvalues of

a d (h)

g

are even. An element

e \in N (g_{1})

is said to be

σ

-distinguished (in other notations

p

(- 1)

-distinguished) if

(g_{1})_{e}

contains no semisimple elements of

[g, g]

Lemma 4. In case

(g, g_{0}) = (s o_{n + m}, s o_{n} \oplus s o_{m})

each

σ

-distinguished element

e \in N (g_{1})

is even.

Proof. Let ${V_{i}}$ be the cyclic spaces of $e \in N (g_{1})$ chosen as prescribed by Lemma 3 . Suppose there is an even-dimensional $V_{i}$ . According to [4,Prop. 2] , if $V_{i}$ has type $a b a \dots a b$ , then $V_{i^{*}}$ has type $b a b \dots b a$ , i.e., if $σ (w_{i}) = - w_{i}$ , then $σ (w_{i^{*}}) = w_{i^{*}}$ . Let $l$ be a Levi subalgebra of $(s o (V_{i} \oplus V_{i^{*}}))_{e}$ . We may assume that $l$ is $σ$ -invariant. Then $l = s l (k w_{i} \oplus k w_{i^{*}}) \sim = s l_{2}$ . The restriction of $σ$ defines a symmetric decomposition $l = l_{0} \oplus l_{1}$ , where $l_{0} = l^{σ} = s o_{2}$ . Therefore $l_{1} = l \cap g_{1}$ contains semisimple elements. This means that $e$ is not $σ$ -distinguished. Hence, all $V_{i}$ are odd-dimensional and $e$ is even. □

3 $ℭ (g_{1})$ is irreducible

In this section we prove that

ℭ (g_{1})

is irreducible in case

(g, g_{0}) = (s o_{n + m}, s o_{n} \oplus s o_{m})

The following lemma is taken from [9] , but the proof given here is shorter. Note that this lemma is valid for any symmetric pair

(g, g_{0})

Lemma 5. Suppose

e \in N (g_{1})

is even. Then

e

belongs to a

G_{0}

-sheet containing semisimple elements.

Proof. Let $(e, f, h)$ be an $s l_{2}$ -triple such that $f \in g_{1}$ , $h \in g_{0}$ . Since $e$ is even, we have $dim g_{h} = dim g_{e}$ . Set $e (t) : = e - t^{2} f$ for $t \in k$ . If $t \neq 0$ , then $e (t)$ is semisimple and conjugated to $t h$ . Therefore $dim g_{e (t)} = dim g_{h} = dim g_{e}$ . Clearly $e (0) = e = {lim}_{t \to 0} e (t)$ and the $G_{0}$ -sheet containing $e$ contains also semisimple elements $e (t)$ . □

Suppose

(g, g_{0}) = (s o_{n + m}, s o_{n} \oplus s o_{m})

and let

c \subset g_{1}

be a Cartan subspace.

Theorem 1. The commuting variety

ℭ (g_{1})

is irreducible.

Proof. Recall that $ℭ_{0} = \bar{G_{0} (c \times c)}$ . Following the original proof of Richardson [7] (see also [6,Sect. 2] ), we show by induction on $dim c$ that $ℭ (g_{1}) = ℭ_{0}$ . The base of induction is the rank $1$ case $(s o_{n + 1}, s o_{n})$ , where the irreducibility of $ℭ (g_{1})$ is proved in [6] , [8] . Let $(x, y) \in ℭ (g_{1})$ .
(1) Suppose there is a semisimple element $h \in g_{1}$ such that $[h, x] = [h, y] = 0$ . This assumption is automatically satisfied if either $x$ or $y$ is semisimple. Moreover, if $x$ (or $y$ ) is not nilpotent and $x = x_{s} + x_{n}$ is the Jordan decomposition, then $x_{s} \in g_{1}$ and $[x_{s}, x] = 0$ , $[x_{s}, y] = 0$ .
Consider the symmetric pair $(g_{h}, (g_{0})_{h})$ . Replacing $c$ by a conjugated Cartan subspace, we may assume that $h \in c$ . Then $c$ is a Cartan subspace of $(g_{1})_{h}$ . Also, $x, y \in (g_{1})_{h}$ by the assumption. By Lemma 2 , $(g_{h}, (g_{0})_{h}) = (\oplus_{i = 1}^{r} (g l_{k_{i}}, s o_{k_{i}})) \oplus (s o_{n + m - 2 k}, s o_{n - k} \oplus s o_{m - k})$ .
Note that each $(g l_{k_{i}}, s o_{k_{i}}))$ is a symmetric pair of maximal rank, hence, the corresponding commuting variety is irreducible, see [5,(3.5)(1)] . Clearly, the commuting variety corresponding to a direct sum of symmetric pairs is a direct product of the commuting varieties corresponding to the summands. Therefore, using the inductive hypothesis, we conclude that $ℭ ((g_{1})_{h})$ is irreducible. Thus $(x, y) \in \bar{(G_{0})_{h} (c \times c)}$ and, hence, $(x, y) \in ℭ_{0}$ .
(2) It remains to consider pairs of commuting nilpotent elements. Suppose first that there is a semisimple element $h \in g_{1}$ such that $[x, h] = 0$ . Then $(x, (1 - t) y + t h) \in ℭ (g_{1})$ for each $t \in k$ and $(1 - t) y + t h$ is nilpotent only for a finite number of $t$ 's. Therefore, by part (1), one has $(x, (1 - t) y + t h) \in ℭ_{0}$ for almost all $t$ . Since $y = {lim}_{t \to 0} (1 - t) y + t h$ , we get $(x, y) \in ℭ_{0}$ .
(3) Now we may assume that both $x$ and $y$ are $σ$ -distinguished nilpotent elements.
According to Lemma 4 , $x$ is even. Then, by Lemma 5 , $x$ belongs to a $G_{0}$ -sheet containing semisimple elements. According to part (1), the assumptions of Lemma 1 are satisfied and it follows that $(x, y) \in ℭ_{0}$ . □

References

Collingwood D. and McGovern W., Nilpotent orbits in semisimple Lie algebras, Mathematics Series, Van Nostrand Reinhold, 1993.
Jantzen J.C., Nilpotent orbits in representation theory, Lie theory, Progr. in Math., 228, 1–211, Birkhäuser Boston 2004.
Kostant B. and Rallis S., Orbits and representations associated with symmetric spaces, Amer. J. Math. 93(1971), 753–809.
Ohta T., Classification of admissible nilpotent orbits in the classical real Lie algebras, J. of Algebra 136(1991), 290–333.
Panyushev D., The Jacobian modules of a representation of a Lie algebra and geometry of commuting varieties, Compositio Math. 94(1994), 181–199.
Panyushev D., On the irreducibility of commuting varieties associated with involutions of simple Lie algebras, Functional Analysis and its application, 38(2004), 38–44.
Richardson R., Commuting varieties of semisimple Lie algebras and algebraic groups, Compositio Math. 38 (1979), 311–327.
Sabourin H. and Yu R.W.T., Sur l'irréductibilité de la variété commutante d'une paire symétrique réductive de rang 1, Bull. Sci. Math. 126(2002), 143–150.
Sabourin H. and Yu R.W.T., On the irreducibility of the commuting variety of a symmetric pair associated to a parabolic subalgebra with abelian unipotent radical, preprint, arXiv:math.RT/0407354.

Oksana Yakimova