, , ,

Let

V

be a vector space endowed with a symplectic form. It is well known that the adjoint representation of

Sp (V)

is equivalent to the symmetric square of

V

, i.e., to the coinvariants of

Z / 2 Z

acting by cyclic permutation on

V \otimes V

. This allows us to endow

(V^{\otimes 2})_{Z / 2 Z}

with the structure of a (symplectic) Lie algebra. In this note, we show that this is just the degree

0

part of a larger object: the graded vector space

^{\infty} \oplus_{n = 0} (V^{\otimes n + 2})_{Z / (n + 2) Z}

has a natural structure of Lie algebra determined by the symplectic form on

V

. In the special case that

dim V = 2

, for every associative algebra

A

, this algebra acts infinitesimally on fibers of the commutator map

[,] : A^{2} \to A

This paper developed from our interest in the fibers of the commutator map on the Lie algebra

s l_{n}

. After we wrote it, B. Tsygan called our attention to a paper of M. Kontsevich [K] which arrived at similar conclusions, coming from a rather different (and more sophisticated) point of view. Kontsevich considered the tensor algebra of a symplectic space

V

over

Q

and looked at the Lie algebra of derivations preserving its symplectic form. He asserted that this algebra is naturally isomorphic to

^{\infty} \oplus_{n = 0} (V^{\otimes n + 2})^{Z / (n + 2) Z} .

Of course in characteristic zero, there is a natural isomorphism between invariants and coinvariants for any finite group, which is not the case in characteristic

p

. By contrast, we give a characteristic free formula for the Lie bracket which is naturally defined on the coinvariant space. In addition, we indicate a dihedral theory which may be useful in analyzing the commutator fibers for

s o_{n}

and

s p_{2 n}

2 The Lie Algebra

Let

K

be a field and

V

a vector space over

K

equipped with an alternating bilinear form

〈, 〉

. For any

l

Z / l Z

acts on

V^{\otimes l}

in the obvious way and we form the graded

K

-vector space

L (V) =^{\infty} \oplus_{l = 2} (V^{\otimes l})_{Z / l Z} .

where

(V^{\otimes l})_{Z / l Z}

has degree

l - 2

. We now define a Lie bracket on

L (V)

. To do this, let us first introduce some additional notation. Fix a basis

ℬ

V

. Given integers

i

and

j

such that

1 \leq i, j \leq l

, we set

D_{i, j} (α_{1} \otimes \dots \otimes α_{l}) = {\begin{matrix} α_{i + 1} \otimes \dots \otimes α_{j - 1} & if i + 1 \leq j - 1, \\ 0 & if i = j - 1, \\ α_{i + 1} \otimes \dots \otimes α_{l} \otimes α_{1} \otimes \dots \otimes α_{j - 1} & if i \geq j, \end{matrix}

where

α_{i} \in ℬ

. Now if

α = α_{1} \otimes \dots \otimes α_{l}

and

β = β_{1} \otimes \dots \otimes β_{m}

are tensor products of basis elements, we define

[α, β] = \sum_{i = 1}^{l} \sum_{j = 1}^{m} 〈 α_{i}, β_{j} 〉 \bar{D_{i, i} (α) \otimes D_{j, j} (β)} .

and extend by linearity. (Here

\bar{γ}

denotes the class of

γ \in V^{\otimes n}

(V^{\otimes n})_{Z / n Z}

.) Note that if

σ

denotes the canonical generator of

Z / l Z

then

D_{i, i} (σ^{r} (α)) = {\begin{matrix} D_{r + i, r + i} (α) & if i \leq l - r \\ D_{r + i - l, r + i - l} (α) & if i > l - r . \end{matrix}

Thus

[,] : (V^{\otimes l})_{Z / l Z} \times (V^{\otimes m})_{Z / m Z} ⟶ (V^{\otimes (l + m - 2)})_{Z / (l + m - 2) Z}

is well-defined. Moreover, it does not depend on the choice of basis

ℬ

For each

x \in ℬ

, we also define

D_{x} : L (V) \to L (V)

D_{x} (α) = \sum_{i = 1}^{l} δ_{α_{i}, x} \bar{D_{i, i} (α)},

where

δ

is the Kronecker delta and

α

is as before. We extend

D_{x}

to a (well-defined) endomorphism of

L (V)

Theorem 1. The above bracket makes

L (V)

into a Lie algebra. Moreover each

D_{x}

is a derivation.

Proof. It is easily verified that the bracket is bilinear and antisymmetric. We now prove that the Jacobi identity holds. Suppose

α = α_{1} \otimes \dots \otimes α_{l}

β = β_{1} \otimes \dots \otimes β_{m}

and

γ = γ_{1} \otimes \dots \otimes γ_{n}

; then we have

[α, [β, γ]] + [β, [γ, α]] + [γ, [α, β]] = \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq j}^{m} 〈 β_{j}, γ_{k} 〉 〈 α_{i}, β_{r} 〉 \bar{D_{i, i} (α) \otimes D_{r, j} (β) \otimes D_{k, k} (γ) \otimes D_{j, r} (β)} + \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq k}^{n} 〈 β_{j}, γ_{k} 〉 〈 α_{i}, γ_{r} 〉 \bar{D_{i, i} (α) \otimes D_{r, k} (γ) \otimes D_{j, j} (β) \otimes D_{k, r} (γ)} + \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq k}^{n} 〈 γ_{k}, α_{i} 〉 〈 β_{j}, γ_{r} 〉 \bar{D_{j, j} (β) \otimes D_{r, k} (γ) \otimes D_{i, i} (α) \otimes D_{k, r} (γ)} + \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq i}^{l} 〈 γ_{k}, α_{i} 〉 〈 β_{j}, α_{r} 〉 \bar{D_{j, j} (β) \otimes D_{r, i} (α) \otimes D_{k, k} (γ) \otimes D_{i, r} (α)} + \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq i}^{l} 〈 α_{i}, β_{j} 〉 〈 γ_{k}, α_{r} 〉 \bar{D_{k, k} (γ) \otimes D_{r, i} (α) \otimes D_{j, j} (β) \otimes D_{i, r} (α)} + \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1}^{n} \sum_{r = 1 r \neq j}^{m} 〈 α_{i}, β_{j} 〉 〈 γ_{k}, β_{r} 〉 \bar{D_{k, k} (γ) \otimes D_{r, j} (β) \otimes D_{i, i} (α) \otimes D_{j, r} (β)} .

Now note that

\bar{D_{i, i} (α) \otimes D_{r, j} (β) \otimes D_{k, k} (γ) \otimes D_{j, r} (β)} = \bar{D_{k, k} (γ) \otimes D_{j, r} (β) \otimes D_{i, i} (α) \otimes D_{r, j} (β)}

(V^{\otimes (l + m + n - 4)})_{Z / (l + m + n - 4) Z}

. Therefore, interchanging the sums over

j

and

r

in the last term, we see that it is exactly the negative of the first term (since

〈 α_{i}, β_{j} 〉 〈 γ_{k}, β_{r} 〉 = - 〈 β_{r}, γ_{k} 〉 〈 α_{i}, β_{j} 〉

). Similarly, the second term cancels the third and the fourth term cancels the fifth. The Jacobi identity now follows by linearity.

Finally, we have

\begin{matrix} D_{x} ([α, β]) & = & \sum_{i = 1}^{l} \sum_{j = 1}^{m} 〈 α_{i}, β_{j} 〉 D_{x} (\bar{D_{i, i} (α) \otimes D_{j, j} (β)}) \end{matrix}

\begin{matrix} = & \sum_{i = 1}^{l} \sum_{j = 1}^{m} 〈 α_{i}, β_{j} 〉 \sum_{k = 1 k \neq i}^{l} δ_{α_{k}, x} \bar{D_{k, i} (α) \otimes D_{j, j} (β) \otimes D_{i, k} (α)} \end{matrix}

\begin{matrix} + & \sum_{i = 1}^{l} \sum_{j = 1}^{m} 〈 α_{i}, β_{j} 〉 \sum_{k = 1 k \neq j}^{m} δ_{β_{k}, x} \bar{D_{k, j} (β) \otimes D_{i, i} (α) \otimes D_{j, k} (β)} \end{matrix}

\begin{matrix} = & [D_{x} (α), β] + [α, D_{x} (β)] \end{matrix}

□

φ : V \to W

is a linear transformation such that

〈 φ (α), φ (β) 〉_{W} = 〈 α, β 〉_{V}

for all

α, β \in V

then we have an obvious Lie algebra homomorphism

L (φ) : L (V) \to L (W)

induced by

α_{1} \otimes \dots \otimes α_{l} \mapsto φ (α_{1}) \otimes \dots \otimes φ (α_{l})

The case

dim V = 2

is of particular interest. Let

{x, y}

be a basis with

〈 x, y 〉 = 1

We identify

V^{\otimes l}

with the noncommutative homogeneous polynomials of degree

l

x

and

y

Suppose that

A

is an associative

K

-algebra considered as a Lie algebra via the usual bracket

[X, Y] = X Y - Y X

. Let

N_{l}

denote the the trace with respect to the

Z / l Z

-action on

V^{\otimes l}

; then the

N_{l}

induce a well-defined map

N : L (V) \to \oplus_{l \geq 2} (V^{\otimes l})^{Z / l Z}

. Let

K [ɛ]

denote the ring of dual numbers. If

α \in V^{\otimes l}

represents a class in

L (V)

, write

N (α) = x p_{α} - y q_{α}

and consider the vector field

F_{α} (X, Y) = (X - ɛ q_{α} (X, Y), Y - ɛ p_{α} (X, Y))

A \times A

. Note that

N (α)

p_{α}

q_{α}

, and

F_{α}

depend only on

\bar{α}

Theorem 2. The map

\bar{α} \mapsto F_{α}

is a homomorphism of Lie algebras from

L (V)

to the algebra of vector fields on

A \times A

tangent to the fibers of the commutator map.

Proof. As

σ (N_{l} (α)) = N_{l} (α)

(x p_{α} - y q_{α}) = (p_{α} x - q_{α} y)

, so

[X - ɛ q_{α} (X, Y), Y - ɛ p_{α} (X, Y)] \equiv [X, Y] (m o d ɛ) .

Thus,

F_{α}

is always tangent to the fibers of the commutator map. To see that

\bar{α} \mapsto F_{α}

is a Lie algebra homomorphism, it suffices to check the case that

α = α_{1} \otimes \dots \otimes α_{l}

and

β = β_{1} \otimes \dots \otimes β_{m}

are tensor products of basis vectors. Now,

p_{α} = \sum_{i = 1}^{l} δ_{α_{i}, x} D_{i, i} (α), q_{α} = - \sum_{i = 1}^{l} δ_{α_{i}, y} D_{i, i} (α) .

(Note that

\bar{p_{α}} = D_{x} (α)

and

\bar{q_{α}} = D_{y} (α)

.) Regarding

F_{α}

and

F_{β}

as sections

A \times A \to A [[ɛ]] \times A [[ɛ]]

of the evaluation at zero map,

\begin{matrix} F_{α} ( & F_{β} (X, Y)) = F_{α} (X + ɛ \sum_{j = 1}^{m} δ_{β_{j}, y} D_{j, j} (β) (X, Y), Y - ɛ \sum_{j = 1}^{m} δ_{β_{j}, x} D_{j, j} (β) (X, Y)) = \end{matrix}

\begin{matrix} (X & + ɛ \sum_{j = 1}^{m} δ_{β_{j}, y} D_{j, j} (β) (X, Y) \end{matrix}

\begin{matrix} + ɛ & \sum_{i = 1}^{l} δ_{α_{i}, y} D_{i, i} (α) (X + ɛ \sum_{j = 1}^{m} δ_{β_{j}, y} D_{j, j} (β) (X, Y), Y - ɛ \sum_{j = 1}^{m} δ_{β_{j}, x} D_{j, j} (β) (X, Y)), \end{matrix}

\begin{matrix} Y & - ɛ \sum_{j = 1}^{m} δ_{β_{j}, x} D_{j, j} (β) (X, Y) \end{matrix}

\begin{matrix} - ɛ & \sum_{i = 1}^{l} δ_{α_{i}, x} D_{i, i} (α) (X + ɛ \sum_{j = 1}^{m} δ_{β_{j}, y} D_{j, j} (β) (X, Y), Y - ɛ \sum_{j = 1}^{m} δ_{β_{j}, x} D_{j, j} (β) (X, Y))) . \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix}  \end{matrix}

γ = γ_{1} \otimes \dots \otimes γ_{n}

is a tensor monomial regarded as a noncommutative homogeneous polynomial, then

γ (X + ɛ X_{1}, Y + ɛ Y_{1}) \equiv γ (X, Y) + ɛ \sum_{k = 1}^{n} D_{n, k} (γ) (X, Y) (δ_{γ_{k}, x} X_{1} + δ_{γ_{k}, y} Y_{1}) D_{k, 1} (γ) (X, Y) (m o d ɛ^{2}) .

Therefore,

\begin{matrix} F_{α} (F_{β} & (X, Y)) - F_{β} (F_{α} (X, Y)) \end{matrix}

\begin{matrix} \equiv ɛ^{2} ( & \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1 k \neq i}^{l} δ_{α_{i}, y} (δ_{α_{k}, x} δ_{β_{j}, y} - δ_{α_{k}, y} δ_{β_{j}, x}) (D_{i, k} (α) D_{j, j} (β) D_{k, i} (α)) (X, Y), \end{matrix}

\begin{matrix} \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1 k \neq i}^{l} δ_{α_{i}, x} (- δ_{α_{k}, x} δ_{β_{j}, y} + δ_{α_{k}, y} δ_{β_{j}, x}) (D_{i, k} (α) D_{j, j} (β) D_{k, i} (α)) (X, Y)) \end{matrix}

\begin{matrix} - ɛ^{2} ( & \sum_{j = 1}^{m} \sum_{i = 1}^{l} \sum_{k = 1 k \neq j}^{m} δ_{β_{j}, y} (δ_{β_{k}, x} δ_{α_{i}, y} - δ_{β_{k}, y} δ_{α_{i}, x}) (D_{j, k} (β) D_{i, i} (α) D_{k, j} (β)) (X, Y), \end{matrix}

\begin{matrix} \sum_{j = 1}^{m} \sum_{i = 1}^{l} \sum_{k = 1 k \neq j}^{m} δ_{β_{j}, x} (- δ_{β_{k}, x} δ_{α_{i}, y} + δ_{β_{k}, y} δ_{α_{i}, x}) (D_{j, k} (β) D_{i, i} (α) D_{k, j} (β)) (X, Y)) \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix} \equiv ɛ^{2} ( & \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1 k \neq i}^{l} δ_{α_{i}, y} 〈 α_{k}, β_{j} 〉 (D_{i, k} (α) D_{j, j} (β) D_{k, i} (α)) (X, Y) \end{matrix}

\begin{matrix} - & \sum_{j = 1}^{m} \sum_{i = 1}^{l} \sum_{k = 1 k \neq j}^{m} δ_{β_{j}, y} 〈 β_{k}, α_{i} 〉 (D_{j, k} (β) D_{i, i} (α) D_{k, j} (β)) (X, Y), \end{matrix}

\begin{matrix} \sum_{i = 1}^{l} \sum_{j = 1}^{m} \sum_{k = 1 k \neq i}^{l} δ_{α_{i}, x} 〈 β_{j}, α_{k} 〉 (D_{i, k} (α) D_{j, j} (β) D_{k, i} (α)) (X, Y) \end{matrix}

\begin{matrix} - & \sum_{j = 1}^{m} \sum_{i = 1}^{l} \sum_{k = 1 k \neq j}^{m} δ_{β_{j}, x} 〈 α_{i}, β_{k} 〉 (D_{j, k} (β) D_{i, i} (α) D_{k, j} (β)) (X, Y)) \end{matrix}

\begin{matrix} \equiv ɛ^{2} & (- q_{_{[\bar{α}, \bar{β}]}} (X, Y), - p_{_{[\bar{α}, \bar{β}]}} (X, Y)) (m o d ɛ^{3}) . \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix}  \end{matrix}

□

Each

V^{\otimes l}

has a natural involution

ι

induced by

ι (α_{1} \otimes \dots \otimes α_{l}) = (- 1)^{l} α_{l} \otimes \dots \otimes α_{1} .

Extending

ι

to all of

L (V)

, we get a Lie algebra involution; to see this, observe that

\begin{matrix} ι (\bar{D_{i, i} (α) \otimes D_{j, j} (β)}) & = (- 1)^{l + m} \bar{D_{m - j + 1, m - j + 1} (ι (β)) \otimes D_{l - i + 1, l - i + 1} (ι (α))} \end{matrix}

\begin{matrix} = (- 1)^{l + m} \bar{D_{l - i + 1, l - i + 1} (ι (α)) \otimes D_{m - j + 1, m - j + 1} (ι (β))} \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix}  \end{matrix}

and since

(ι (α))_{l - i + 1} = α_{i}

(ι (β))_{m - j + 1} = β_{j}

and

(- 1)^{l + m - 2} = (- 1)^{l + m}

, it follows that

[ι (α), ι (β)] = ι ([α, β])

. We will also need to consider the unsigned involution

I

induced by

I (α_{1} \otimes \dots \otimes α_{l}) = α_{l} \otimes \dots \otimes α_{1}

Denote by

P_{+} (V)

and

P_{-} (V)

, respectively, the

+ 1

and

- 1

eigenspaces for

ι

Then of course

P_{+} (V)

is a Lie subalgebra of

L (V)

and

[P_{-} (V), P_{-} (V)] \subset P_{+} (V)

Similarly, we let

P_{l, +}

and

P_{l, -}

denote the

+ 1

and

- 1

eigenspaces for

ι

V^{\otimes l}

. To simplify the notation, we will assume that

V

is finite-dimensional, so that

ℬ = {x_{1}, . . ., x_{r}}

Lemma. Suppose

α \in (V^{\otimes l})^{Z / l Z}

and write

α = \sum x_{i} \otimes p_{i}

, where the

p_{i} \in V^{\otimes l - 1}

. Then

α \in P_{l, +}

(resp.,

P_{l, -}

) if and only if

p_{1}, . . ., p_{r} \in P_{l - 1, -}

(resp.,

P_{l - 1, +})

Proof. This is clear, as $ι (α) = - \sum ι (p_{i}) \otimes x_{i}$ and $σ (α) = \sum p_{i} \otimes x_{i}$ . □

As an application of these definitions, we turn to the case of a Lie algebra

g

defined over

K

and identify

V^{\otimes l}

with noncommutative polynomials of degree

l

x_{1}, . . ., x_{r}

. Our first observation is that if

g

is either

s o_{n}

s p_{n}

and

p = p (x_{1}, . . ., x_{r})

is in

\oplus_{l} P_{l, -}

then for all

X_{1}, . . ., X_{r} \in g

p (X_{1}, . . ., X_{r}) \in g

. Indeed write

p = \sum p_{i}

, where each

p_{i}

is a monomial. Then if

g = s o_{n}

\begin{matrix} p (X_{1}, . . ., X_{r})^{t} & = \sum p_{i} (X_{1}, . . ., X_{r})^{t} = \sum I (p_{i}) (X_{1}^{t}, . . ., X_{r}^{t}) \end{matrix}

\begin{matrix} = \sum (- 1)^{deg p_{i}} I (p_{i}) (X_{1}, . . ., X_{r}) = ι (p) (X_{1}, . . ., X_{r}) \end{matrix}

\begin{matrix} = - p (X_{1}, . . ., X_{r}) . \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix}  \end{matrix}

g = s p_{n}

let

J

denote the matrix of the nondegenerate alternating form which defines

g

. Then

\begin{matrix} J p (X_{1}, . . ., X_{r}) & = \sum J p_{i} (X_{1}, . . ., X_{r}) = \sum (- 1)^{deg p_{i}} p_{i} (X_{1}^{t}, . . ., X_{r}^{t}) J \end{matrix}

\begin{matrix} = (ι (p) (X_{1}, . . ., X_{r}))^{t} J = - p (X_{1}, . . ., X_{r})^{t} J . \end{matrix}

\begin{matrix}  \end{matrix}

\begin{matrix}  \end{matrix}

With

dim V = 2

and

g

of the above type (in particular,

g \subset g l_{n}

) we consider the above map

\bar{α} \mapsto F_{α}

from

L (V)

to the algebra of vector fields on

g l_{n} \times g l_{n}

Proposition. The image of

P_{+} (V)

under this map consists of vector fields on

g \times g

tangent to the fibers of the commutator map

g \times g \to g

Proof. If $α \in V^{\otimes l}$ then $ι (σ (α)) = σ^{- 1} (ι (α))$ , which implies that $N (P_{+} (V)) \subset \oplus_{l} P_{l, +}$ . Moreover if $α \in P_{+} (V)$ and $N (α) = x p_{α} - y q_{α}$ then $p_{α}, q_{α} \in \oplus_{l} P_{l, -}$ by the Lemma. Now we are done, as $p_{α} (X, Y), q_{α} (X, Y) \in g$ whenever $X, Y \in g$ by the preceding discussion. □

References

M. Kontsevich, Formal (non)-commutative symplectic geometry, The Gelfand Mathematical Seminars, 1990–1992 (L. Corwin, I. Gelfand, J. Lepowsky, eds.), Birkhauser, 1993, pp. 173–187.

Department of Mathematics, Northwestern University, 2033 Sheridan Road, Evanston, IL 60208 E-mail address : ekushnir@math.northwestern.edu Department of Mathematics, Indiana University, Bloomington, IN 47405 E-mail address : larsen@math.indiana.edu

Eugene Kushnirsky

Michael Larsen