Superalgebras
Shun-Jen Cheng
Weiqiang Wang
NSC.
NSF.
Preface ix
iii
iv Contents
Bibliography 267
viii Contents
Index 273
Preface
Lie algebras, Lie groups, and their representation theories are parts of the math-
ematical language to describe symmetries, and they have played a central role in
modern mathematics. An early motivation of studying Lie superalgebras as a gen-
eralization of Lie algebras came from supersymmetry in mathematical physics.
Ever since a Cartan-Killing type classification of finite dimensional complex Lie
superalgebra was obtained by Kac [K1] in 1977, the theory of Lie superalgebras
has established itself as a prominent subject in modern mathematics. An inde-
pendent classification of the finite-dimensional complex simple Lie superalgebras
whose even subalgebras are reductive (called the basic Lie superalgebras) was
given by Scheunert, Nahm, and Rittenberg in [SNR].
The goal of this book is a systematic account of the structure and representation
theory of finite dimensional complex Lie superalgebras of classical type. The book
intends to serve as a rigorous introduction to representation theory of Lie super-
algebras on one hand, and on the other hand, it covers a new approach developed
in the past few years toward understanding the Bernstein-Gelfand-Gelfand cate-
gory for classical Lie superalgebras. In spite of much interest in representations
of Lie superalgebras stimulated by mathematical physics, these basic topics have
not been treated in depth in a book form before. The reason seems to be that the
representation theory of Lie superalgebras is dramatically different from complex
semisimple Lie algebras, and a systematic yet accessible approach toward the basic
problem of finding irreducible characters for Lie superalgebras was not available
in a great generality until very recently.
We are consciously aware that there is an enormous literature and numerous
partial results for Lie superalgebras, and it is not our intention to make this book
an encyclopedia. Rather, we treat in depth the representation theory of the three
ix
x Preface
most important classes of Lie superalgebras, namely, the general linear Lie super-
algebras gl(m|n), the ortho-symplectic Lie superalgebras osp(m|2n), and the queer
Lie superalgebras q(n). A prototype of this book is the lecture notes [CW5] by the
authors. The presentation in this book is organized around three dualities with a
unifying theme of determining irreducible characters:
There are two superalgebra generalizations of Schur duality. The first one,
due to Sergeev [Sv1] and independently Berele-Regev [BeR], is an interplay be-
tween the general linear Lie superalgebra gl(m|n) and the symmetric group which
incorporates the trivial and sign modules in a unified framework. The irreducible
polynomial characters of gl(m|n) arising this way are given by the super Schur
polynomials. The second one, called Sergeev duality, is an interplay between the
queer Lie superalgebra q(n) and a twisted hyperoctahedral group algebra. The
Schur Q-functions and related combinatorics of shifted tableaux appear naturally
in the description of the irreducible polynomial characters of q(n).
It has been observed that much of the classical invariant theory for the polyno-
mial algebra has a parallel theory for the exterior algebra as well. The First Funda-
mental Theorem (FFT) for both polynomial invariants and skew invariants for clas-
sical groups admits natural reformulation and extension in the theory of Howe’s
reductive dual pairs [H1, H2]. Lie superalgebras allow an elegant and uniform
treatment of Howe duality on the polynomial and exterior algebras (cf. Cheng-
Wang [CW1]). For the general linear Lie groups, Schur duality, Howe duality
and FFT are equivalent. Unlike Schur duality, Howe duality treats classical Lie
groups and (super)algebras beyond type A equally well. The Howe dualities allow
us to determine the character formulas for the irreducible modules appearing in the
dualities.
The third duality, super duality, has a completely different flavor. It views
representation theories of Lie superalgebras and Lie algebras as two sides of the
same coin, and it is an unexpected and rather powerful approach developed in the
past few years by the authors and their collaborators, culminating in Cheng-Lam-
Wang [CLW]. The super duality approach allows one to overcome in a conceptual
way various major obstacles in the super representation theory, via an equivalence
of module categories of Lie algebras and Lie superalgebras.
Schur, Howe, and super dualities provide approaches to the irreducible charac-
ter problem in increasing generality and sophistication. Schur and Howe dualities
only offer a solution to the irreducible character problem for modules in some
semisimple subcategories. On the other hand, super duality provides a concep-
tual solution to the long-standing irreducible character problem in fairly general
Preface xi
BGG categories (including all finite-dimensional modules) over classical Lie su-
peralgebras in terms of the usual Kazhdan-Lusztig polynomials of classical Lie al-
gebras. Totally different approaches to the irreducible character problem of finite-
dimensional gl(m|n)-modules have been developed Serganova [Sva] and Brundan
[Br1].
The book is largely self-contained, and should be accessible to graduate stu-
dents and non-experts as well. Besides assuming basic knowledge of entry-level
graduate algebra, the only other prerequisite is a one-semester course in the the-
ory of finite-dimensional semisimple Lie algebras. For example, either the book
by Humphreys or the first half of the book by Carter on semisimple Lie algebras is
more than sufficient. Some familiarity with symmetric functions can be useful, and
Appendix A provides a quick summary for our purpose. It is possible that super
experts may also benefit from the book, as several “folklore” results are clarified
and occasionally corrected in great detail here, sometimes with new proofs. The
proofs of some of these results can be at times rather difficult to trace or read in the
literature (not merely because they might be in a different language).
Here is a broad outline of the book chapter by chapter. Each chapter ends with
some historical notes and references. Though we have tried to attribute the main
results accurately and fairly, we apologize beforehand for any unintended omission
and mistakes.
Chapter 1 starts with defining various classes of Lie superalgebras. For the
basic Lie superalgebras, we introduce the invariant bilinear forms, root systems,
fundamental systems, and Weyl groups. Positive systems and fundamental systems
for basic Lie superalgebras are not conjugate under the Weyl group, and the notion
of odd reflections is introduced to relate non-conjugate positive systems. The PBW
theorem for the universal enveloping algebra of a Lie superalgebra is formulated,
and highest weight theory for basic Lie superalgebras and q(n) is developed.
In Chapter 2, we focus on Lie superalgebras of types gl, osp and q. We classify
their finite-dimensional simple modules using odd reflection techniques. We then
formulate and establish precisely the image of Harish-Chandra homomorphism and
linkage principle. We end with a Young diagramatic description of the extremal
weights in the simple polynomial gl(m|n)-modules and finite-dimensional simple
osp(m|2n)-modules. It takes considerably more effort to formulate and prove these
results for Lie superalgebras in contrast to semisimple Lie algebras, because of the
existence of non-conjugate Borel subalgebras and the limited role of Weyl groups
for Lie superalgebras.
Schur duality for Lie superalgebras is developed in Chapter 3. We start with
some results on the structure of associative superalgebras including the super vari-
ants of Wedderburn theorem, Schur’s lemma, and double centralizer property. The
Schur-Sergeev duality for gl(m|n) is proved, and it provides a classification of ir-
reducible polynomial gl(m|n)-modules. As a consequence, the characters of the
xii Preface
Acknowledgment. The book project started with the lecture notes [CW5]
of the authors, which was an expanded written account for a series of lectures
delivered by the second-named author in the summer school at East China Normal
University, Shanghai, in July 2009. In a graduate course at University of Virginia in
Spring 2010, the second-named author lectured on what became a large portion of
Chapters 3, 4 and 5 of the book. The materials of Chapter 3 on Schur duality have
been used by the second-named author in the winter school in Taipei in December
2010. Part of the materials of the first three chapters have also been used by the
first-named author in a lecture series in Shanghai in March 2011, and then in a
lecture series by both authors in a workshop in Tehran in May 2011. We thank the
participants in all these occasions for their helpful suggestions and feedbacks.
Shun-Jen Cheng
Weiqiang Wang
Chapter 1
We start by introducing the basic notions and definitions in the theory of Lie super-
algebras, such as basic Lie superalgebras and queer Lie superalgebras, Cartan sub-
algebras, Borel subalgebras, root systems, positive and fundamental systems. We
formulate the main structure results for the basic Lie superalgebras and the queer
Lie superalgebras. We describe in detail the structures of Lie superalgebras of type
gl, osp and q. A distinguished feature for Lie superalgebras is that Borel subal-
gebras, positive systems, or fundamental systems of a simple finite-dimensional
Lie superalgebra may not be conjugate under the action of the corresponding Weyl
group; instead, they are shown to be related to each other by real and odd reflec-
tions. A highest weight theory is developed for Lie superalgebras. We describe
how fundamental systems are related and how highest weights are transformed by
an odd reflection.
1
2 1. Lie superalgebra ABC
superalgebras of classical type, namely, the queer and the periplectic Lie super-
algebras, along with the three exceptional ones, are then described. The finite-
dimensional Cartan type Lie superalgebras are then realized explicitly as subal-
gebras of the Lie superalgebra of polynomial vector fields on a purely odd di-
mensional superspace. The section ends with Kac’s classification theorem of the
finite-dimensional simple Lie superalgebras over C, which we state without proof.
For a Lie superalgebra g = g0̄ ⊕ g1̄ , the restriction of the adjoint homomor-
phism ad |g0̄ : g0̄ → End(g1̄ ) is a homomorphism of Lie algebras, namely, g1̄ is a
g0̄ -module under the adjoint action.
Remark 1.5. To a Lie superalgebra g = g0̄ ⊕ g1̄ we associate the following data:
(1) A Lie algebra g0̄ .
(2) A g0̄ -module g1̄ induced by the adjoint action.
(3) A g0̄ -homomorphism S2 (g1̄ ) → g0̄ induced by the Lie bracket.
(4) The condition coming from Definition 1.2(2) with a, b, c ∈ g1̄ .
Conversely, the above data determine a Lie superalgebra structure on g0̄ ⊕ g1̄ .
1.1.2. The general and special linear Lie superalgebras. Let V = V0̄ ⊕ V1̄ be
a vector superspace so that End(V ) is an associative superalgebra. As in Ex-
ample 1.4, End(V ), equipped with the supercommutator, forms a Lie superalge-
bra, called the general linear Lie superalgebra and is denoted by gl(V ). When
V = Cm|n we also write gl(m|n) for gl(V ).
Choose bases for V0̄ and V1̄ that combine to a homogeneous basis for V . We
will make it a convention to parameterize such a basis by the set
(1.2) I(m|n) = {1, . . . , m; 1, . . . , n}
with total order
(1.3) 1 < . . . < m < 0 < 1 < . . . < n.
Here 0 is inserted for notational convenience later on. The elementary matrices
are accordingly denoted by Ei j , with i, j ∈ I(m|n). With respect to such an ordered
basis, End(V ) and gl(V ) can be realized as (m + n) × (m + n) complex matrices of
the block form
( )
a b
(1.4) g= ,
c d
where a, b, c, and d are respectively m × m, m × n, n × m, and n × n matrices. The
even subalgebra gl(V )0̄ consists of matrices of the form (1.4) with b = c = 0, while
the odd subspace gl(V )1̄ consists of those with a = d = 0. In particular, gl(V )0̄ ∼
=
1.1. Lie superalgebras: Definition and Examples 5
gl(m) ⊕ gl(n), and as a gl(V )0̄ -module, gl(V )1̄ is self-dual and is isomorphic to
(Cm ⊗ Cn∗ ) ⊕ (Cm∗ ⊗ Cn ). Here and below, Cn∗ denotes the dual space of Cn .
For each element g ∈ gl(m|n) of the form (1.4) we define the supertrace
str(g) := tr(a) − tr(d),
where tr(x) denotes the trace of the square matrix x. One checks that
str([g, g′ ]) = 0, for g, g′ ∈ gl(m|n).
Thus, the subspace
sl(m|n) := {g ∈ gl(m|n) | str(g) = 0}
is a subalgebra of gl(m|n), called the special linear Lie superalgebra. One verifies
directly that [gl(m|n), gl(m|n)] = sl(m|n). Furthermore, sl(m|n) ∼ = sl(n|m), and
when m ̸= n, sl(m|n) is simple. When m = n, sl(m|m) contains a non-trivial center
generated by the identity matrix Im|m . For m ≥ 2, sl(m|m)/CIm|m is simple.
Example 1.6. Let g = gl(1|1) and consider the following basis for g:
( ) ( ) ( ) ( )
0 1 0 0 1 0 0 0
e= , f= , E1̄.1̄ = , E11 = .
0 0 1 0 0 0 0 1
Set h := E1̄,1̄ + E11 = I1|1 . Then h is central, [e, f ] = h, and sl(1|1) has a basis
{e, h, f }.
B(·, ·) : V ×V −→ V
One checks that osp(V ) is a Lie superalgebra, called the ortho-symplectic Lie
superalgebra, that is, osp(V ) is the subalgebra of gl(V ) that preserves a non-
degenerate supersymmetric bilinear form. Its even subalgebra is isomorphic to
so(V0̄ ) ⊕ sp(V1̄ ), a direct sum of the orthogonal Lie algebra on V0̄ and the symplec-
tic Lie algebra on V1̄ . When V = Cℓ|2m , we write osp(ℓ|2m) for osp(V ). Note that
when ℓ or m is zero, the ortho-symplectic Lie superalgebra reduces to classical Lie
algebra sp(2m) or so(ℓ), respectively.
Similarly, we define the Lie superalgebra spo(V ) as the subalgebra of gl(V )
that preserves a non-degenerate skew-supersymmetric bilinear form on V (here
dimV0̄ has to be even). When V = C2m|ℓ , we write spo(2m|ℓ) for spo(V ).
A supersymmetric bilinear form B on V induces a skew-supersymmetric bilin-
ear form C(·, ·) on Π(V ) defined by C(Π(v), Π(v′ )) := B(v, v′ ), for v, v′ ∈ V , where
Π is the parity reversing functor defined in Section 1.1.1. Define an isomorphism of
Lie superalgebras from End(V ) to End(ΠV ) by sending T to ΠT Π−1 . One checks
that the restriction of the map to osp(V ) sends osp(V ) to spo(ΠV ). It follows that
osp(ℓ|2m) ∼ = spo(2m|ℓ) as Lie superalgebras.
We now give an explicit matrix realization of the ortho-symplectic Lie super-
algebra. To this end, we first observe that the supertranspose (1.5) of a matrix in
the block form (1.4), is equal to
( )st ( )
a b at ct
= ,
c d −bt dt
Proof. Take an element g ∈ gl(2m|ℓ) that lies in spo(2m|ℓ). Thus, J2m|ℓ g+gst J2m|ℓ =
0 and hence
(1.9) gst Jst2m|ℓ + Jst2m|ℓ (gst )st = 0.
Observe that Jst2m|ℓ J2m|ℓ = I2m|ℓ so that if we multiply (1.9) on the left and on the
right by J2m|ℓ , we get the identity
This gives the identity J2m|ℓ τ(g) + τ(g)st J2m|ℓ = 0, and so τ(g) ∈ spo(2m|ℓ).
8 1. Lie superalgebra ABC
1.1.4. The queer Lie superalgebras. Let V = V0̄ ⊕V1̄ be a vector superspace with
dimV0̄ = dimV1̄ . Choose P ∈ End(V )1̄ such that P2 = In|n . The subspace
q(V ) = {T ∈ End(V ) | [T, P] = 0}
is a subalgebra of gl(V ), called the queer Lie superalgebra. (Different choices of
P give rise to isomorphic queer Lie superalgebras.) If V = Cn|n , then q(V ) is also
denoted by q(n).
In order to give an explicit realization of q(n) in terms of matrices, let us take
P to be the 2n × 2n matrix
( )
√ 0 In
(1.10) P := −1 .
−In 0
√
(The factor −1 makes no difference here, and it is introduced for later conve-
nience.) Then, for g ∈ gl(n|n) of the form (1.4) we have g ∈ q(n) if and only if
gP − (−1)|g| Pg = 0, and in turn, if and only if
( )
a b
(1.11) g= ,
b a
where a, b are arbitrary complex n × n matrices. Thus we have q(n)0̄ ∼ = gl(n) as Lie
algebras, and q(n) ∼
1̄ = gl(n) as the adjoint q(n) -module. A linear basis for q(n)
0̄
consists of the following elements
Eei j := Eī j¯ + Ei j , E i j := Ei j¯ + Eī j , 1 ≤ i, j ≤ n.
The derived superalgebra [q(n), q(n)] consists of matrices of the form (1.11),
with a ∈ gl(n) and b ∈ sl(n), and so it contains a one-dimensional center generated
by the identity matrix In|n . The quotient superalgebra [q(n), q(n)]/CIn|n has even
part isomorphic to sl(n) and odd part isomorphic to the adjoint module, and one
can show that it is simple for n ≥ 3. For n = 2, the odd part of the quotient Lie
superalgebra is an abelian ideal, since the adjoint module of sl(2) does not appear
in the symmetric square of the adjoint module.
Remark 1.9. Consider the subspace e
q(n) of gl(n|n) consisting of elements that
commute with P. That is,
e
q(n) := {g ∈ gl(n|n)|gP − Pg = 0}.
In matrix form e
q(n) consists of the following n|n-block matrices:
( )
a b
(1.12) ,
−b a
where a and b are arbitrary n × n matrices. One checks that e q(n) is closed under
the Lie bracket and hence is a subalgebra of gl(n|n). Indeed e q(n) is isomorphic to
q(n), since the map τ in (1.6) sends q(n) to e
q(n), and vice versa. Thus, (1.12) gives
another realization of the queer Lie superalgebra.
1.1. Lie superalgebras: Definition and Examples 9
1.1.5. The periplectic and exceptional Lie superalgebras. Let us describe more
examples of Lie superalgebras.
The periplectic Lie superalgebras. Let V = V0̄ ⊕ V1̄ be a vector superspace
with dimV0̄ = dimV1̄ . Let C(·, ·) be a non-degenerate odd symmetric bilinear form
on V . One checks that the subspace of gl(V ) preserving C is closed under the Lie
bracket and hence is a Lie subalgebra of gl(V ). This superalgebra is called the
periplectic Lie superalgebra and will be denoted by p(V ). (Different choices of
C give rise to isomorphic periplectic Lie superalgebras.) In the case V = Cn|n , p(V )
is also denoted by p(n).
To write down an explicit matrix realization of p(n) as subalgebra of gl(n|n),
let us take the 2n × 2n matrix
( )
0 In
(1.13) P := ,
In 0
which determines an odd symmetric bilinear form C on Cn|n . Then, g ∈ p(n) if and
only if gst P + Pg = 0. It follows that
{( ) }
a b
(1.14) p(n) = , where b is symmetric and c is skew-symmetric .
c −at
We have
(1.15) p(n)0̄ ∼
= gl(n), p(n)1̄ ∼
= S2 (Cn ) ⊕ ∧2 (Cn∗ ).
For n ≥ 3, the
( derived)superalgebra [p(n), p(n)] is simple, and it consists of matrices
a b
of the form , with tr(a) = 0, bt = b, and ct = −c.
c −at
Remark 1.10. One checks that the Lie subalgebra e p(n) of gl(n|n) preserving the
non-degenerate odd skew-symmetric( bilinear
) form corresponding to P in (1.10)
a b
consists of matrices of the form with bt = −b and ct = c. The superal-
c −at
gebra e
p(n) is isomorphic to p(n).
The exceptional Lie superalgebra D(2, 1, α). We take three copies of the Lie
algebra sl(2) denoted by gi (i = 1, 2, 3), and we associate to each gi a copy of the
standard sl(2)-module Vi .
Clearly as gi -modules we have an isomorphism S2 (Vi ) ∼
= gi . By Schur’s Lemma
we may associate a non-zero scalar αi ∈ C to any such an isomorphism. Now con-
sider g = g0̄ ⊕ g1̄ , where g0̄ ∼
= g1 ⊕ g2 ⊕ g3 , and g1̄ is the irreducible g0̄ -module
V1 ⊗ V2 ⊗ V3 . We can associate three non-zero complex numbers αi , i = 1, 2, 3,
to any surjective g0̄ -homomorphism from S2 (g1̄ ) to g0̄ . Thus our vector super-
space g satisfies Conditions (1)–(3) of Remark 1.5. It is easy to see that Con-
dition (4) of Remark 1.5 is equivalent to ∑3i=1 αi = 0. Thus we obtain a Lie su-
peralgebra g(α1 , α2 , α3 ) depending on three non-zero parameters αi , i = 1, 2, 3.
10 1. Lie superalgebra ABC
D(2, 1, α) ∼
= D(2, 1, −(1 + α)−1 α) ∼
= D(2, 1, −1 − α−1 ) ∼
= D(2, 1, −(1 + α)−1 ).
Thus, any orbit of (C \ {0, −1}) /S3 consists of six points, except for the orbit
corresponding to the three points α = 1, −2, − 12 , and the orbit corresponding to the
√
two points α = − 12 ± − 34 . Finally, note that D(2, 1, 1) ∼= osp(4|2).
The exceptional Lie superalgebra F(4). There is a simple Lie superalgebra
F(4) with F(4)0̄ ∼
= sl(2) ⊕ so(7). The odd part, as an F(4)0̄ -module, is isomorphic
the tensor product of the standard sl(2)-module and the simple so(7)-spin module.
Hence, dim F(4) = (24|16).
The exceptional Lie superalgebra G(3). There is a simple Lie superalgebra
G(3) with G(3)0̄ ∼= sl(2) ⊕ G2 . The odd part as a G(3)0̄ -module is isomorphic to
the tensor product of the standard sl(2)-module and the fundamental 7-dimensional
G2 -module. Hence, dim G(3) = (17|14).
1.1.6. The Cartan series. In this subsection, we describe the Cartan series of
finite-dimensional simple Lie superalgebras without proof. This subsection is in-
cluded for the sake of completeness of classification of finite-dimensional simple
Lie superalgebras, and will not be used in the rest of the book.
Lie superalgebra W (n). Let ∧(n) be the exterior algebra in n indeterminates
ξ1 , ξ2 , . . . , ξn , with a Z2 -gradation given by setting |ξi | = 1̄, for all i. We have
ξi ξ j = −ξ j ξi , for all i, j, and in particular, ξ2i = 0, for all i.
By general construction in Example 1.4, we have a finite-dimensional Lie su-
peralgebra of derivations on ∧(n), which will be denoted by W (n).
∂
For i = 1, . . . , n, a derivation ∂ξi
: ∧(n) → ∧(n) of degree 1̄ is uniquely deter-
mined by
∂
(ξ j ) = δi j , j = 1, . . . , n.
∂ξi
1.1. Lie superalgebras: Definition and Examples 11
The Z-gradation is compatible with the super structure on W (n), that is, W (n)s =
⊕ j≡s mod 2W (n) j , for s ∈ Z2 . The 0th degree component W (n)0 is a Lie algebra iso-
morphic to gl(n), and each W (n) j is a gl(n)-module isomorphic to ∧ j+1 (Cn )⊗Cn∗ .
In particular, when n = 2, we have W (2)0 ∼ = gl(2), W (2)−1 ∼
= C2∗ and W (2)1 ∼= C2
as gl(2)-modules. Hence, we have isomorphisms of Lie superalgebras W (2) ∼ =
∼
osp(2|2) = sl(2|1).
The Lie superalgebra W (n) is simple, for n ≥ 2. Moreover, W (n) contains
the following three series of simple Lie superalgebras as subalgebras that we shall
describe.
Lie superalgebra S(n). The first series is the super-analogue of the Lie algebra
of divergence vector fields given by
{ n ∂ n
∂ }
S(n) := ∑ f j ∈ W (n) | ∑ ( f j) = 0 .
j=1 ∂ξ j j=1 ∂ξ j
The Lie superalgebra S(n) is a Z-graded subalgebra of W (n) and we have S(n) =
⊕n−2
j=−1 S(n) j . The Lie algebra S(n)0 is isomorphic to sl(n), and the jth degree
component S(n) j is isomorphic to the top irreducible summand of the sl(n)-module
∧ j+1 (Cn ) ⊗ Cn∗ . The Lie superalgebra S(n) is simple, for n ≥ 3.
e
Lie superalgebra S(n). Let n be even so that ω = 1 + ξ1 ξ2 · · · ξn is a Z2 -
homogeneous invertible element in ∧(n). Consider the subspace of W (n) given
by
{ n }
e := ∑ f j ∂ | ∑ ∂ (ω f j ) = 0 .
n
S(n)
j=1 ∂ξ j j=1 ∂ξ j
The Poisson bracket makes ∧(n) into a Lie superalgebra, which we will denote by
e
H(n). e
Now putting deg f := k − 2, for f ∈ ∧(n)k , H(n) becomes a Z-graded Lie
e
superalgebra. The superalgebra H(n) is not simple, as it has center C1. However,
e
the derived superalgebra of H(n)/C1, which we denote by H(n), is simple, for
⊕n−3
n ≥ 4. Moreover, H(n) = j=−1 H(n) j is a graded Lie superalgebra. The 0th
degree component is a Lie algebra isomorphic to so(n). As an so(n)-module we
have H(n) j ∼= ∧ j+2 (Cn ), for −1 ≤ j ≤ n − 3. Finally, H(n) can be viewed as a
∂f ∂
subalgebra of W (n), since the assignment f 7→ (−1)| f | ∑nj=1 ∂ξ j ∂ξ j
for f ∈ ∧(n),
gives rise to an embedding of Lie superalgebras from H(n) into W (n).
1.1.7. The classification theorem. The following theorem of Kac [K1] gives a
classification of finite-dimensional complex simple Lie superalgebras. Note the
following isomorphisms of simple Lie superalgebras of low ranks:
osp(2|2) ∼
= sl(2|1) ∼
= W (2), sl(2|2)/CI2|2 ∼
= H(4), [p(3), p(3)] ∼
= S(3).
Theorem 1.11. The following is a complete list of pairwise non-isomorphic finite-
dimensional simple Lie superalgebras over C.
(1) A finite-dimensional simple Lie algebra in the Killing-Cartan list.
(2) sl(m|n), for m > n ≥ 1 (excluding (m, n) = (2, 1)); sl(m|m)/CIm|m , for
m ≥ 3; spo(2m|n), for m, n ≥ 1.
( )
(3) D(2, 1, α), for α ∈ C \ {0, ±1, −2, − 12 } /S3 ; F(4); and G(3).
(4) [p(n), p(n)] and [q(n), q(n)]/CIn|n , for n ≥ 3.
e
(5) W (n), for n ≥ 3; S(n), for n ≥ 3; S(2n), for n ≥ 2; and H(n), for n ≥ 4.
1.2. Structures of classical Lie superalgebras 13
The Lie superalgebras g = g0̄ ⊕ g1̄ in Theorem 1.11(1)-(4) have the property
that g0̄ is a reductive Lie algebra and that the adjoint g0̄ -module g1̄ is semisimple.
To distinguish them from the Cartan series of Theorem 1.11(5), we introduce the
following terminology.
Definition 1.12. The Lie superalgebras g = g0̄ ⊕ g1̄ in Theorem 1.11(2)-(4) are
called classical. Classical Lie superalgebras in Theorem 1.11(2)-(3) are called
basic. The Lie superalgebra gl(m|n) for m, n ≥ 1 is also declared to be basic.
Remark 1.13. Let us comment on the simplicity of the Lie superalgebras in The-
orem 1.11. It is not difficult to check directly the simplicity of the sl series,
spo(2m|2), and those in (4). A Lie superalgebra g in the remaining cases in (2)
and (3), with the exception of spo(2m|2), satisfies the properties that the adjoint
g0̄ -module g1̄ is irreducible and faithful, and [g1̄ , g1̄ ] = g0̄ . The simplicity of such
a g follows easily from these properties. For the Z-graded Cartan type Lie superal-
⊕
gebras g = j≥−1 gi in (5), we first note that they are all transitive (i.e., [g−1 , x] = 0
implies that x = 0, for x ∈ g j and j ≥ 0,) and irreducible (i.e., the g0 -module g−1 is
irreducible). Furthermore, we have [g−1 , g1 ] = g0 , [g0 , g1 ] = g1 , and g j is generated
by g1 , for j ≥ 1. From these properties simplicity follows. Finally, simplicity of
e can be proved with a bit of extra work using the explicit realization given in
S(n)
(1.16).
As we shall see, the Weyl groups play a less vital though still important role
in determining central characters and the linkage principle for Lie superalgebras,
which is somewhat different from the theory of semisimple Lie algebras.
The following structure theorem shows that the basic Lie superalgebras behave
similarly to semisimple Lie algebras.
Theorem 1.15. Let g be a basic Lie superalgebra with a Cartan subalgebra h.
(1) We have a root space decomposition of g with respect to h:
⊕
g = h⊕ gα , and g0 = h.
α∈Φ
(2) dim gα = 1, for α ∈ Φ. (Fix now some nonzero eα ∈ gα .)
(3) [gα , gβ ] ⊆ gα+β , for α, β, α + β ∈ Φ.
(4) Φ, Φ0̄ and Φ1̄ are invariant under the action of the Weyl group W on h∗ .
(5) There exists a non-degenerate even invariant supersymmetric bilinear
form (·, ·) on g.
(6) (gα , gβ ) = 0 unless α = −β ∈ Φ.
(7) The restriction of the bilinear form (·, ·) on h × h is non-degenerate and
W -invariant.
(8) [eα , e−α ] = (eα , e−α )hα , where hα is the coroot determined by (hα , h) =
α(h) for h ∈ h.
(9) Φ = −Φ, Φ0̄ = −Φ0̄ , and Φ1̄ = −Φ1̄ .
(10) Let α ∈ Φ. Then, kα ∈ Φ for k ̸= ±1 if and only if α ∈ Φ1̄ and (α, α) ̸= 0;
in this case, k = ±2.
Proof. For g = gl(m|n) or osp(2m|ℓ), we will describe explicitly the root systems
and trace forms below in this section, and so the theorem in these cases follows
by inspection. The theorem for the other infinite series basic Lie superalgebras
sl(m|n) follows by some easy modification from the gl(m|n) case.
Part (4) follows by the invariance, under the action of the Weyl group W , of
the sets of weights for the adjoint g0̄ -modules g0̄ and g1̄ , respectively.
Since the remaining three exceptional Lie superalgebras D(2, 1, α), G(3) and
F(4) will not be studied in detail in the book, we will be sketchy. Most of the
items of the theorem, with an exception of (5), again follow by inspection from
the constructions of these superalgebras and standard arguments as for simple Lie
1.2. Structures of classical Lie superalgebras 15
algebras (with the help of (5)). As the exceptional Lie superalgebras are simple,
any nonzero invariant supersymmetric bilinear form must be non-degenerate. For
G(3) and F(4), the Killing form is such a nonzero even form. A direct and ad hoc
construction of a nonzero invariant bilinear form on D(2, 1, α) is possible.
Φi = {α ∈ Φ | gα ⊆ gi }, i ∈ Z2 .
Remark 1.16. One uniform approach to establish Theorem 1.15 is as follows (see
[K1, Proposition 2.5.3]). One follows the standard construction of contragredi-
ent (i.e., Kac-Moody) (super)algebras (see Carter [Car, Chapter 14]) to show that
any Lie superalgebra in Theorem 1.15 is a Kac-Moody superalgebra (or rather
the quotient by its possibly nontrivial center) associated to some generalized Car-
tan matrix with Z2 -grading. These (quotients of) Kac-Moody superalgebras carry
non-degenerate even invariant supersymmetric bilinear forms, by a standard Kac-
Moody type argument (see Carter [Car, Chapter 16]).
1.2.2. Invariant bilinear forms for gl and osp. In contrast to the semisimple Lie
algebras, the Killing form for a basic Lie superalgebra may be zero, and even when
it is nonzero, it may not be positive definite on the real vector space spanned by
Φ. In this subsection, we give a down-to-earth description of an invariant non-
degenerate even supersymmetric bilinear form for Lie superalgebras of type gl and
osp.
The supertrace str on the general linear Lie superalgebra gives rise to a non-
degenerate supersymmetric bilinear form
form on h:
1 if 1 ≤ i = j ≤ m,
(Eii , E j j ) = −1 if 1 ≤ i = j ≤ n,
0 if i ̸= j,
where i, j ∈ I(m|n). We recall here that I(m|n) is defined in (1.2). Denote by
{δi , ε j }i, j the basis of h∗ dual to {Ei i , E j j }i, j , where 1 ≤ i ≤ m and 1 ≤ j ≤ n.
Using the bilinear form (·, ·) we can identify δi with (Ei,i , ·) and ε j with −(E j j , ·).
Whenever it is convenient we also use the notation
(1.18) εi := δi , for 1 ≤ i ≤ m.
1.2.3. Root system and Weyl group for gl(m|n). Let g = gl(m|n) and h be the
Cartan subalgebra of diagonal matrices. Its root system Φ = Φ0̄ ∪ Φ1̄ is given by
specifying
Φ0̄ = {εi − ε j | i ̸= j ∈ I(m|n), i, j > 0 or i, j < 0},
Φ1̄ = {±(εi − ε j ) | i, j ∈ I(m|n), i < 0 < j}.
Observe that Ei j is a root vector corresponding to the root εi −ε j , for i ̸= j ∈ I(m|n).
The Weyl group of gl(m|n), which is by definition the Weyl group of g0̄ =
gl(m) ⊕ gl(n), is isomorphic to Sm × Sn , where we recall that Sn denotes the
symmetric group of n letters.
1.2.4. Root system and Weyl group for spo(2m|2n + 1). Now we describe the
root system for the ortho-symplectic Lie superalgebra spo(2m|2n + 1), which is
defined in matrix form (1.8) in Section 1.1.3. Recall that the rows and columns
of the matrices are indexed by I(2m|2n + 1). The subalgebra h of g of diagonal
matrices has a basis given by
Hi := Ei,i − Em+i,m+i , 1 ≤ i ≤ m,
H j := E j j − En+ j,n+ j , 1 ≤ j ≤ n,
1.2. Structures of classical Lie superalgebras 17
and it is a Cartan subalgebra for spo(2m|2n + 1). With respect to h, the root system
Φ = Φ0̄ ∪ Φ1̄ for spo(2m|2n + 1) is
{±δi ± δ j , ±2δ p , ±εk ± εl , ±εq } ∪ {±δ p ± εq , ±δ p },
where 1 ≤ i < j ≤ m, 1 ≤ k < l ≤ n, 1 ≤ p ≤ m, 1 ≤ q ≤ n.
A root vector is a nonzero vector in gα for α ∈ Φ, and will be denoted by
eα . The root vectors for spo(2m|2n + 1) can be computed explicitly as follows
(1 ≤ i ̸= j ≤ m, 1 ≤ k ̸= l ≤ n):
eεk = E2n+1,k+n − Ek,2n+1 , e−εk = E2n+1,k − Ek+n,2n+1 ,
(1.20) e2δi = Ei,i+m , e−2δi = Ei+m,i ,
(1.21) eδi +δ j = Ei, j+m + E j,i+m , e−δi −δ j = E j+m,i + Ei+m, j ,
(1.22) eδi −δ j = Ei, j − E j+m,i+m , eεk −εl = Ekl − El+n,k+n ,
(1.23) eεk +εl = Ek,l+n − El,k+n , e−εk −εl = Ek+n,l − El+n,k ,
The Weyl group of spo(2m|2n + 1), which is by definition the Weyl group of
g0̄ = sp(2m) ⊕ so(2n + 1), is isomorphic to (Zm
2 o Sm ) × (Z2 o Sn ).
n
1.2.5. Root system and Weyl group for spo(2m|2n). Let g = spo(2m|2n). The
abelian Lie subalgebra h spanned by {Hi , H j |1 ≤ i ≤ m, 1 ≤ j ≤ n} is a Cartan
subalgebra for spo(2m|2n). Let {δi , ε j | 1 ≤ i ≤ m, 1 ≤ j ≤ n} be the corresponding
dual basis. Again, when it is convenient, we will use the notation εī = δi , for
1 ≤ i ≤ m. With respect to h, the root system Φ = Φ0̄ ∪ Φ1̄ is given by
{±δi ± δ j , ±2δ p , ±εk ± εl } ∪ {±δ p ± εq },
where 1 ≤ i < j ≤ m, 1 ≤ k < l ≤ n, 1 ≤ p ≤ m, 1 ≤ q ≤ n.
The root vectors for spo(2m|2n) are given by (1.20)–(1.25).
The Weyl group of spo(2m|2n), which is by definition the Weyl group of g0̄ =
sp(2m) ⊕ so(2n), is an index 2 subgroup of (Zm2 o Sm ) × (Z2 o Sn ), with only
n
1.2.6. Root system and odd invariant form for q(n). In this subsection let g be
the queer Lie superalgebra q(n), see Section 1.1.4. As the case of q(n) is different
from the basic Lie superalgebra case, we will describe altogether its Cartan sub-
algebras, root systems, positive systems, Borel subalgebras, and invariant bilinear
form.
18 1. Lie superalgebra ABC
(
)
a b
Recall that q(n) can be realized as matrices in the n|n block form
b a
indexed by I(n|n). The subalgebra consisting of matrices with a, b being diago-
nal (which we refer to as “block diagonal matrices”) will be called the standard
Cartan subalgebra. We define a Cartan subalgebra h to be any subalgebra con-
jugate by some element in GL(n) to the standard Cartan subalgebra. Note that h
is self-normalizing in g and h is nilpotent, justifying the terminology of a Cartain
subalgebra. However, h = h0̄ ⊕h1̄ is not abelian, since [h0̄ , h] = 0, and [h1̄ , h1̄ ] = h0̄ .
Now fix h = h0̄ ⊕ h1̄ to be the standard Cartan subalgebra. The vectors
Hi := Eī ī + Eii , i = 1, . . . , n,
form a basis for h0̄ , while the vectors
H i := Eī i + Ei ī , i = 1, . . . , n,
form a basis for h1̄ . We let {εi | i = 1, . . . , n} denote the corresponding dual basis
⊕
in h∗0̄ . With respect to h0̄ , we have the root space decomposition g = h ⊕ α∈Φ gα
with root system Φ = Φ0̄ ∪Φ1̄ , where both Φ0̄ and Φ1̄ coincide with the root system
for gl(n):
Φ0̄ = Φ1̄ = {εi − ε j | 1 ≤ i ̸= j ≤ n}.
We have dimC gα = 1, for each α ∈ Φ, and gα ⊆ gi , for α ∈ Φi and i ∈ Z2 . The
Weyl group of q(n) is identified with the symmetric group Sn .
( )
a b
The matrices with a, b being upper triangular (which we refer to as
b a
“block upper triangular matrices”) form a solvable subalgebra b, which will be
called the standard Borel subalgebra of g. The positive system corresponding to
the Borel subalgebra b is Φ+ = Φ+ 0̄
∪ Φ+
1̄
, where Φ+
0̄
= Φ+
1̄
= {εi − ε j | 1 ≤ i < j ≤
n}. Let Φ− = −Φ+ and so Φ = Φ+ ∪ Φ− . Let
⊕ ⊕
n+ = gα , n− = gα .
α∈Φ+ α∈Φ−
1.3.1. Positive systems and fundamental systems. Let Φ be a root system for a
basic Lie superalgebra and let E be the real vector space spanned by Φ. We have
E ⊗R C = h∗ , for g ̸= gl(m|n). For g = gl(m|n) the space E ⊗R C is a subspace of
h∗ of codimension one.
A total ordering ≥ on E below is always assumed to be compatible with the
real vector space structure, that is, v ≥ w and v′ ≥ w′ imply that v + v′ ≥ w + w′ ,
−w ≥ −v, and cv ≥ cw for c ∈ R>0 .
A positive system Φ+ is a subset of Φ consisting precisely of all those roots
α ∈ Φ satisfying α > 0 for some total ordering of E. Given a positive system
Φ+ , we define the fundamental system Π ⊂ Φ+ to be the set of α ∈ Φ+ which
cannot be written as a sum of two roots in Φ+ . We refer to elements in Φ+ as
positive roots and elements in Π as simple roots. Similarly, we denote by Φ− the
− −
corresponding set of negative roots. Set Φ+i = Φ ∩ Φi and Φi = Φ ∩ Φi , for
+
−
i ∈ Z2 . By Theorem 1.15(9), we have Φ− = −Φ+ and Φi = −Φi , for i ∈ Z2 .
+
Then, we have
Φ+ = Φ+0̄
∪ Φ+
1̄
.
Recall Φ̄1̄ from (1.17). Associated to a positive system Φ+ , we let
(1.27) Φ̄+
1̄
:= Φ̄1̄ ∩ Φ+ .
Proposition 1.17. Let g be a basic Lie superalgebra. There is a one-to-one corre-
spondence between the set of positive systems and the set of fundamental systems.
The Weyl group of g acts naturally on the set of the positive systems (respectively,
fundamental systems).
Proof. It follows by definition that the fundamental system exists and is unique for
a given positive system. A positive root, if not simple, can be written as a sum of
two positive roots. Continuing this way, any positive root is a Z+ -linear combina-
tion of simple roots, and hence a positive system is determined by its fundamental
system.
20 1. Lie superalgebra ABC
We define
⊕ ⊕
n+ = gα , n− = gα .
α∈Φ+ α∈Φ−
Remark 1.18. The rank one subalgebra corresponding to an isotropic simple root
is isomorphic to sl(1|1), which is solvable. Therefore, if we enlarge a Borel sub-
algebra by adding the root space corresponding to a negative isotropic simple root,
then the resulting subalgebra is still solvable. Thus, a Borel subalgebra is not a
maximal solvable subalgebra for Lie superalgebras in general.
where
1 1
ρ0̄ = ∑+ α,
2 α∈Φ
ρ1̄ = ∑+ β.
2 β∈Φ
0̄ 1̄
Denote
Lemma 1.19. We have the following formulas for the Weyl vector ρ for the stan-
dard positive system Φ+ :
We shall next describe completely the positive and fundamental systems for
Lie superalgebras of type gl and osp case-by-case.
1.3. Non-conjugate positive systems and odd reflections 21
1.3.2. Positive and fundamental systems for gl(m|n). Recall the root system Φ
for gl(m|n) and the standard Cartan subalgebra h described in Section 1.2.3. The
subalgebra of upper triangular matrices is the standard Borel subalgebra of g
which contains h, and the corresponding standard positive system of Φ is given
by {εi − ε j | i, j ∈ I(m|n), i < j}. Bearing in mind εi = δi , its standard fundamental
system is
{δi − δi+1 , ε j − ε j+1 , δm − ε1 | 1 ≤ i ≤ m − 1, 1 ≤ j ≤ n − 1},
with the corresponding simple root vectors ei := Ei,i+1 , for i ∈ I(m − 1|n − 1), and
em := Em,1 . The simple coroots are h j := E j j − E j+1, j+1 , for j ∈ I(m − 1|n − 1),
and hm := Em,m + E11 . Denote fi := Ei+1,i , for i ∈ I(m − 1|n − 1), and fm := E1,m .
Then {ei , hi , fi | i ∈ I(m|n − 1)} is a set of Chevalley generators for sl(m|n).
Note that
(δi − δi+1 , δi − δi+1 ) = 2, 1 ≤ i ≤ m − 1,
(δm − ε1 , δm − ε1 ) = 0,
(ε j − ε j+1 , ε j − ε j+1 ) = −2, 1 ≤ j ≤ n − 1.
Thus δm − ε1 is an isotropic simple root. Following the usual convention for Lie
algebras, we draw the corresponding standard Dynkin diagram:
⊗
⃝ ⃝ ··· ⃝ ··· ⃝ ⃝
δ1 − δ2 δ2 − δ3 δm − ε1 ε1 − ε2 εn−2 − εn−1 εn−1 − εn
where we denote by ⃝ an even simple root α such that 12 α is not a root, and denote
⊗
by an odd isotropic simple root.
Let us classify all possible positive systems for gl(m|n), keeping in mind εī =
δi . If we ignore the parity of roots for a moment, the root system of gl(m|n) is the
same as the root system for gl(m+n), and hence by definition their positive systems
(respectively, fundamental systems) are exactly described in the same way, and
there are (m + n)! of them in total. It follows from the well-known classification
for gl(m + n) that a fundamental system for gl(m|n) consists of (m + n − 1) roots
εi1 − εi2 , εi2 − εi3 , . . . , εim+n−1 − εim+n , where {i1 , i2 , . . . , im+n } = I(m|n). Then we
restore the parity of the simple roots in a fundamental system for gl(m|n). The
corresponding Dynkin diagram is of the form
⊙ ⊙ ⊙ ⊙ ⊙ ⊙
··· ···
εi1 − εi2 εi2 − εi3 εik − εik+1 εim+n−1 − εim+n
⊙ ⊗
where is ⃝ or , depending on the corresponding simple root is even or odd.
Example 1.20. For n = m, there exists a fundamental system which consists of
only odd roots {δ1 −ε1 , ε1 −δ2 , δ2 −ε2 , . . . , δm −εm }, whose corresponding Dynkin
diagram is
22 1. Lie superalgebra ABC
⊗ ⊗ ⊗ ⊗ ⊗ ⊗
··· ···
δ1 − ε1 ε1 − δ2 δk − εk εk − δk+1 εm−1 − δm δm − εm
1.3.3. Positive and fundamental systems for spo(2m|2n + 1). Now we describe
the positive/fundamental systems and Dynkin diagrams for spo(2m|2n + 1) whose
root system is described in Section 1.2.4. The standard positive system Φ+ =
Φ+0̄
∪ Φ+
1̄
corresponding to the standard Borel subalgebra for spo(2m|2n + 1) is
{δi ± δ j , 2δ p , εk ± εl , εq } ∪ {δ p ± εq , δ p },
where 1 ≤ i < j ≤ m, 1 ≤ k < l ≤ n, 1 ≤ p ≤ m, 1 ≤ q ≤ n. The fundamental system
Π of Φ+ contains one odd simple root δm − ε1 , and it is given by
Π = {δi − δi+1 , δm − ε1 , εk − εk+1 , εn | 1 ≤ i ≤ m − 1, 1 ≤ k ≤ n − 1}.
The corresponding standard Dynkin diagram for spo(2m|2n + 1) is
⊗
⃝ ⃝ ··· ⃝ ··· ⃝=⇒⃝
δ1 − δ2 δ2 − δ3 δm − ε1 ε1 − ε2 εn−1 − εn εn
Example 1.22. Let g = osp(1|2). Then its even subalgebra g0̄ is isomorphic to
sl(2) = C⟨e, h, f ⟩, and as a g0̄ -module g1̄ is isomorphic to the 2-dimensional natural
sl(2)-module CE + CF. The simple root consists of a (unique) odd non-isotropic
root δ so that 2δ is an even root. The Dynkin diagram is a . The root vectors
E and F associated to the odd roots δ and −δ can be chosen such that [E, E] =
2e, [F, F] = −2 f , [E, F] = h.
As we already observed above, there are (at least) two Dynkin diagrams for
spo(2m|2n) of different shapes. The classification of all fundamental systems for
the root system Φ of spo(2m|2n) is divided into 2 cases below. As usual we keep
in mind εī = δi .
(1) First we classify the fundamental systems Π in Φ that do not contain
any long root (i.e., a root of the form ±2ε p ). With parity ignored, the subset
Φ̃ := Φ − {±2ε p | 1 ≤ p ≤ m} may be identified with the root system of the clas-
sical Lie algebra so(2m + 2n), whose fundamental systems are completely known
and they are acted upon simply transitively by the Weyl group of so(2m + 2n)
(which is an index 2 subgroup of Zm+n 2 o Sm+n ). Observe that the positive sys-
tem Φ+ for Φ corresponding to Π is completely determined by the positive system
Φ̃ ∩ Φ+ for Φ̃ (and vice versa). We conclude that the fundamental systems for
Φ that do not contain any long root are exactly the fundamental systems for Φ̃.
However, not every fundamental system of Φ̃ gives rise to a fundamental system
of Φ. To be precise, a fundamental system of Φ cannot contain a pair of roots
of the form {±εi ± ε p , ±εi ∓ ε p }, i ̸= p and 1 ≤ p ≤ m. It follows that there are
m+n · |W (so(2m + 2n))| such fundamental systems of Φ, and hence the number
n
of
(m+n−1)
W -conjugacy classes for g in this case is m+n · |W (so(2m + 2n))|/|W | =
n
m .
(2) Next, we classify the fundamental systems Π that contain some long root.
In this case, we consider Φ b := Φ ∪ {±2ε j , 1 ≤ j ≤ n}. With the parity ignored,
b may be identified with the root system of sp(2m + 2n), whose fundamental sys-
Φ
tems are completely described. Observe that the positive system Φ+ for Φ which
corresponds to Π is completely determined by the positive system Φ b + of Φb which
contains Φ . We conclude that the fundamental systems for Φ that contain some
+
long root are exactly the fundamental systems for Φ b whose long root is of the form
±2ε p . It follows that the number of W -conjugacy classes of such fundamental
( )
m
systems for g is m+n · |W (sp(2m + 2n))|/|W | = 2 m+n−1
n .
The εδ-sequence for spo(2m|2n) associated to a positive system is now de-
fined as follows. We can first obtain an ordered sequence of ε’s and δ’s just as
for spo(2m|2n + 1). If this sequence has an ε as its last member, then it is the
εδ-sequence. If the last member in this sequence is a δ, then the εδ-sequence is
obtained from this sequence by attaching a sign to the last ε. For example, for
spo(4|4) the εδ-sequences εδδε, εδεδ, and εδ(−ε)δ are distinct.
1.3. Non-conjugate positive systems and odd reflections 25
The fundamental and positive systems for the exceptional Lie superalgebras
G(3), F(4), D(2, 1, α) are also listed completely by Kac [K1] (with one missing
for F(4); see [WZ, 5.1]).
¿From the classification of fundamental and positive systems we can read off
the following.
Proposition 1.24. Let g be a basic Lie superalgebra, excluding sl(n|n)/CIn|n and
gl(m|n). Let h be a Cartan subalgebra of g. Then any fundamental system forms
a basis in h∗ . For g = gl(m|n), any fundamental system is linearly independent in
h∗ .
26 1. Lie superalgebra ABC
Lemma 1.25. Let g be a basic Lie superalgebra. Let Π be the fundamental system
in the positive system Φ+ , and let α be an isotropic odd root in Φ+ . Then there
exists w ∈ W such that w(α) ∈ Π.
1.3.6. Odd reflections. We have seen that the fundamental systems of a root sys-
tem Φ are not W -conjugate due to the existence of odd (simple) roots. The follow-
ing fundamental lemma will play an important role in representation theory of Lie
superalgebras.
Lemma 1.26. Let g be a basic Lie superalgebra. Let α be an isotropic odd simple
root in a positive system Φ+ . Then,
Φ+
α := {−α} ∪ Φ \{α}
+
b) will be referred to as an odd reflection (with respect to α), and will be denoted
by rα . We shall write
rα (Π) = Πα , rα (Φ+ ) = Φ+
α, rα (b) = bα .
Note that r−α rα = 1.
⊗
Here the vertical dashed lines indicate an edge when it connects two ’s and no
⊕ ⊙ ⊗
edge otherwise, means either ⃝ or a , and means either or ⃝. It can be
checked directly in these cases that Πα is also a fundamental system.
Take β ∈ Φ+ α ∩ Φ . Since Π is a fundamental system for Φ , by Proposi-
+ +
Φ+
α = {−α} ∪ Φ \{α}, we compute that
+
{
ρ − α, for α even or non-isotropic odd,
ρα =
ρ + α, for α isotropic odd.
Claim. Assume that the proposition holds for a fundamental system Π, then it
holds for the fundamental system Πα for α ∈ Π.
Let us prove the claim. First assume α ∈ Π is even or non-isotropic odd. Since
ρα = rα (ρ) and Πα = rα (Π), we have for each β ∈ Π that
1 1
(ρα , rα (β)) = (ρ, β) = (β, β) = (rα (β), rα (β)).
2 2
Now assume α ∈ Π is isotropic odd. We will check case-by-case, recalling the def-
inition of Πα from (1.30). If (β, α) = 0 for β ∈ Π, then β ∈ Πα . By the assumption
of the claim, we have
1
(ρα , β) = (ρ + α, β) = (ρ, β) = (β, β).
2
If (β, α) ̸= 0 for β ∈ Π, then β + α ∈ Πα . By the assumption of the claim, (ρ, α) =
2 (α, α) = 0, and hence
1
1 1
(ρα , β + α) = (ρ + α, β + α) = (ρ + α, β) = (β, β) + (α, β) = (β + α, β + α).
2 2
This completes the proof of the claim.
By the claim and Corollary 1.27, it suffices to prove the proposition for just
one fundamental system, e.g., the standard fundamental system for type gl and
osp. Assume that β is an even simple root or an non-isotropic odd simple root. In
either case one checks that rβ (ρ) = ρ − β. Since (·, ·) is W -invariant, we have
( )
(ρ, β) = rβ (ρ), rβ (β) = (ρ − β, −β) = −(ρ, β) + (β, β).
It follows that (ρ, β) = 12 (β, β) just as for semisimple Lie algebras.
For isotropic odd simple root β, we do a case-by-case inspection. For type gl
or osp, the equation (ρ, β) = 12 (β, β) follows directly from Lemma 1.19. For the
three exceptional cases, it can be checked directly by using a fundamental system
with exactly one isotropic simple root and the detailed description of root systems
(cf. Kac [K1]). We skip the details.
Remark 1.29. In contrast to a real reflection, an odd reflection rα with respect to
an isotropic odd root α may not be extended to a linear transformation on h∗ which
sends a simple root in Π to a simple root in Πα . For example, for spo(2m|2n)
with a fundamental system Π corresponding to the first Dynkin diagram below,
and take α = δm − εn . The odd reflection transforms Π to Πα corresponding to
the second Dynkin diagram below (with the · · · portion unchanged). A plausible
transformation would have to fix all possible δi , ε j and interchange εn and δm . But
this transformation would not send δm + εn to 2δm .
1.3. Non-conjugate positive systems and odd reflections 29
⊗
δm + εn
⊗ ⊗
··· ··· ⃝ ⇐=⃝
εn−1 − δm@
@⊗ δm − εn
εn−1 − εn εn − δm 2δm
Example 1.30. Associated with gl(1|2), we have Φ0̄ = {±(ε1 − ε2 )}, and Φ1̄ =
{±(δ1 − ε1 ), ±(δ1 − ε2 )}. There are 6 sets of simple roots, that are related by real
and odd reflections as follows. There are three conjugacy classes of Borel subalge-
bras corresponding to the three columns below, and each vertical pair corresponds
to such a conjugacy class.
↕ rε1 −ε2
rδ1 −ε2
↕ rε1 −ε2
rδ1 −ε1
↕ rε1 −ε2
⊗ ⊗ ⊗ ⊗
⃝ ⃝
δ1 − ε2 ε2 − ε1 ε2 − δ1 δ1 − ε1 ε2 − ε1 ε1 − δ1
⊗
⃝ ⃝ ⃝ ⃝
XX
rδ4 −ε1
δ1 − δ2 δ2 − δ3 δ3 − δ4 δ4 − ε1 ε1 − ε2 X X ⊗ ⊗ ⊗
rδ3 −ε1 ⃝ ⃝
⊗ ⊗ ⊗ δ1 − δ2 δ2 − δ3 δ3 − ε1 ε1 − δ4 δ4 − ε2
⃝ ⃝
δ1 − δ2 δ2 − ε1 ε1 − δ3 δ3 − δ4 δ4 − ε2 XX
rδ2 −ε1
X X ⊗ ⊗ ⊗
rδ1 −ε1 ⃝ ⃝
⊗ ⊗ δ1 − ε1 ε1 − δ2 δ2 − δ3 δ3 − δ4 δ4 − ε2
⃝ ⃝ ⃝
ε1 − δ1 δ1 − δ2 δ2 − δ3 δ3 − δ4 δ4 − ε2
⊗ ⊗
⃝ ⃝ ⃝
XX
rδ4 −ε2
X
ε1 − δ1 δ1 − δ2 δ2 − δ3 δ3 − δ4 δ4 − ε2 X ⊗ ⊗ ⊗
rδ3 −ε2 ⃝ ⃝
⊗ ⊗ ⊗ ε1 − δ1 δ1 − δ2 δ2 − δ3 δ3 − ε2 ε2 − δ4
⃝ ⃝
ε1 − δ1 δ1 − δ2 δ2 − ε2 ε2 − δ3 δ3 − δ4 XX
rδ2 −ε2
X X ⊗ ⊗ ⊗
rδ1 −ε2 ⃝ ⃝
⊗ ε1 − δ1 δ1 − ε2 ε2 − δ2 δ2 − δ3 δ3 − δ4
⃝ ⃝ ⃝ ⃝
ε1 − ε2 ε2 − δ1 δ1 − δ2 δ2 − δ3 δ3 − δ4
i j
Using the assumption that [g1̄ , g1̄ ] ⊆ [g0̄ , g0̄ ] , it follows from (1.31) and (1.32)
that every yk leaves Vi invariant, and hence each Vi is a g-module. Thus, the above
filtration is a composition series with one-dimensional composition factors.
Example 1.34. The Lie superalgebra g = Cz ⊕ g1̄ in Example 1.4(4) associated to
a non-degenerated symmetric bilinear form B on a nonzero space g1̄ is solvable.
But an irreducible module of g with the central element z acting as a nonzero
scalar has dimension more than one, as they are irreducible modules of the Clifford
superalgebra on (g1̄ , B). So the condition [g1̄ , g1̄ ] ⊆ [g0̄ , g0̄ ] in Lemma 1.33 cannot
be dropped.
1.4.3. Highest weight theory for basic Lie superalgebras. Let g be a basic Lie
superalgebra. Let h be the standard Cartan subalgebra and Φ be the root system.
Let b = h ⊕ n+ be a Borel subalgebra of g = b ⊕ n− and let Φ+ be the associated
positive system. The condition of Lemma 1.33 is satisfied for the solvable Lie
superalgebra b, since we have b1̄ = n1̄ , and
[b1̄ , b1̄ ] = [n1̄ , n1̄ ] ⊆ n0̄ = [h, n0̄ ] ⊆ [b0̄ , b0̄ ].
Let V be a finite-dimensional irreducible representation of g. Then by Lemma 1.33
V contains a one-dimensional b-module, which is of the form Cλ = Cvλ , for λ ∈
h∗ ∼
= (b/[b, b])∗ . That is,
hvλ = λ(h)vλ (h ∈ h), xvλ = 0 (x ∈ n+ ).
By the PBW theorem and the irreducibility of V , we obtain that V = U(n− )vλ , and
thus a weight space decomposition
⊕
(1.33) V= Vµ ,
µ∈h∗
a b-highest weight vector for V . When no confusion arises, we will simply say
highest weight by dropping b. Hence, we have established the following.
Lemma 1.36. Let L be a simple g-module and let v be a b-highest weight vector
of L of b-highest weight λ. Let α be an isotropic odd simple root.
Example 1.37. Let g = gl(4|2). Denote the sequence of Borel subalgebras cor-
responding to the fundamental systems Πi in Example 1.31 as bi , for 0 ≤ i ≤ 8,
with bst = b0 . Consider the finite-dimensional irreducible gl(4|2)-module L(λ),
where λ = a1 δ1 + a2 δ2 + a3 δ3 + a4 δ4 + b1 ε1 + b2 ε2 with a1 ≥ a2 ≥ a3 ≥ a4 ≥ 2
and b1 ≥ b2 ≥ 0. We identify λ = (a1 , a2 , a3 , a4 |b1 , b2 ). The bi -extremal weights,
34 1. Lie superalgebra ABC
1.4.4. Highest weight theory for q(n). Let g = q(n) be the queer Lie superalge-
bra. We recall several subalgebras of g from 1.2.6. Let h be the standard Car-
tan subalgebra consisting of block diagonal matrices and let b be the standard
Borel subalgebra of block upper triangular matrices of g. We have b = h ⊕ n+
and g = b ⊕ n− . The Cartan subalgebra h = h0̄ ⊕ h1̄ is a solvable but nonabelian
Lie superalgebra, since [h1̄ , h1̄ ] = h0̄ and [h0̄ , h0̄ + h1̄ ] = 0.
For λ ∈ h∗0̄ , define a symmetric bilinear form ⟨·, ·⟩λ on h1̄ by
⟨v, w⟩λ := λ([v, w]), for v, w ∈ h1̄ .
Denote by Rad⟨·, ·⟩λ the radical of the form ⟨·, ·⟩λ . Then the form ⟨·, ·⟩λ descends
to a nondegenerate symmetric bilinear form on h1̄ /Rad⟨·, ·⟩λ , and it gives rise to a
Clifford superalgebra C = Cλ := C(h1̄ /Rad⟨·, ·⟩λ ). By definition we have an iso-
morphism of superalgebras
(1.34) Cλ ∼
= U(h)/Iλ ,
where Iλ denotes the ideal of U(h) generated by Rad⟨·, ·⟩λ and a − λ(a) for a ∈ h0̄ .
Let h′1̄ ⊆ h1̄ be a maximal isotropic subspace with respect to ⟨·, ·⟩λ , and define
the Lie subalgebra h′ := h0̄ ⊕ h′1̄ . The one-dimensional h0̄ -module Cvλ , defined by
hvλ = λ(h)vλ , extends to an h′ -module by letting h′1̄ .vλ = 0. Define the induced
h-module
Wλ := Indhh′ Cvλ .
1.4. Highest weight theory 35
Recall the well-known fact that the Clifford superalgebra C admits a unique
eλ (See ?? in Chapter 3).
irreducible (Z2 -graded) module W
Lemma 1.38. For λ ∈ h∗0̄ , the h-module Wλ is isomorphic to W eλ (viewed as an
h-module via the pullback through (1.34)), and is irreducible. Furthermore, every
finite-dimensional irreducible h-module is isomorphic to Wλ , for some λ ∈ h∗0̄ .
Proof. The action of U(h) on Wλ descends to an action of U(h)/Iλ , and via (1.34)
eλ of the Clifford superalgebra
we identify Wλ with the unique irreducible module W
Cλ . Hence, the h-module Wλ is irreducible.
Suppose we are given an irreducible h-module U. Then it contains an h0̄ -
weight vector v′λ of weight λ ∈ h∗0̄ . Recall that h′ = h0̄ ⊕ h′1̄ , and consider the
( )
h′ -submodule π,U(h′ )v′λ of U such that π([h′1̄ , h′1̄ ]) = π([h′0̄ , h′0̄ ]) = 0. By ap-
plying Lemma 1.33 to Lie superalgebra π(h′ ), there exists a one-dimensional h′ -
submodule Cvλ of U(h′ )v′λ ⊆ U. By Frobenious reciprocity we have a surjective
h-homomorphism from Wλ onto U. Since Wλ is irreducible, we have U ∼ = Wλ .
Let ℓ(λ) be the number of nonzero parts in a composition λ (the notation here
is consistent with the notation for length ℓ(λ) of a partition λ). We set
{
0, if ℓ(λ) is even,
(1.35) δ(λ) =
1, if ℓ(λ) is odd.
Now for λ ∈ h∗ , recalling the notation Hi from 1.2.6, we identify λ with the
composition λ = (λ(H1 ), . . . , λ(Hn )), and hence, ℓ(λ) is equal to the dimension
of the space h1̄ /Rad⟨·|·⟩λ . We remark that the highest weight space Wλ of L(λ)
36 1. Lie superalgebra ABC
has dimension 2(ℓ(λ)+δ(λ))/2 . Note that the Clifford algebra Cλ admits an odd au-
tomorphism if and only if ℓ(λ) is odd. Hence, the h-module Wλ , or equivalently
the irreducible C-module Wλ , has an odd automorphism if and only if ℓ(λ) is an
odd integer. An automorphism of the irreducible g-module L(λ) clearly induces
an h-module automorphism of its highest weight space. Conversely, any h-module
automorphism on Wλ induces an automorphism of the g-module IndgbWλ . Since an
automorphism preserves the maximal submodule, it induces an automorphism of
the unique irreducible quotient g-module. We have proved the following.
Lemma 1.40. Let g = q(n) be the queer Lie superalgebra with a Cartan subalge-
bra h. Let L(λ) be the irreducible g-module of highest weight λ ∈ h∗0̄ . Then,
dim(Endg (L(λ)) = 2δ(λ) .
1.5. Notes
Section 1.1. The classification of finite-dimensional simple complex Lie superal-
gebras was first announced by Kac in [K4]. The detailed proof of the classifica-
tion, along with many other fundamental results on Lie superalgebras, appeared in
Kac [K1] two years later. An independent proof of the classification of the finite-
dimensional complex simple Lie superalgebras whose even subalgebras are reduc-
tive was given in the two papers by Scheunert, Nahm, and Rittenberg in [SNR]
around the same time.
Section 1.2. The structure theory of root systems, root space decompositions
and invariant bilinear forms of basic Lie superalgebras presented here is fairly stan-
dard. More details can be found in the standard references (see Kac [K1, K2] and
Scheunert [Sch]).
Section 1.3. Odd reflections was introduced by Leites, Saveliev, and Serganova
[LSS] to relate non-conjugate Borel subalgebras, fundamental and positive sys-
tems. The list of conjugacy classes of fundamental systems under the Weyl group
action for the basic Lie algebras was given by Kac [K1]. We introduce a no-
tion of εδ-sequences (which appeared in Cheng-Wang [CW5]) to facilitate the
parametrization of the conjugacy classes of fundamental systems for Lie super-
algebras of type gl and osp. The fundamental Lemma 1.26 on odd reflections is
sometimes attributed to Serganova’s 1988 thesis (cf., e.g., [Sva2]), and more gen-
erally for Kac-Moody Lie superalgebras in Kac-Wakimoto [KW, Lemma 1.2].
Section 1.4. A detailed proof of the PBW Theorem for Lie superalgebras, The-
orem 1.32, can be found in Milnor-Moore [MM, Theorem 6.20] and Ross [Ro,
Theorem 2.1]. As in the classical theory, the first step to develop a highest weight
theory for the basic and queer Lie superalgebras is to study representations of Borel
1.5. Notes 37
Finite-dimensional
modules
In this chapter, we mainly work on Lie superalgebras of type gl, osp or q. The
finite-dimensional irreducible modules for these Lie superalgebras are classified in
terms of highest weights. In contrast to semisimple Lie algebras, there is more
work to do to achieve such classifications for Lie superalgebras. We describe the
images of the Harish-Chandra homomorphisms in terms of (super)symmetric func-
tions and formulate the linkage principle on composition factors of a Verma mod-
ule. As an application, we present a Weyl-type character formula for the so-called
typical finite dimensional irreducible modules. The chapter concludes with a study
of extremal weights, i.e., highest weights with respect to (not necessarily conju-
gate) Borel subalgebras, of various finite dimensional irreducible modules.
g = n− ⊕ h ⊕ n+ .
39
40 2. Finite-dimensional modules
Proof. Following the proof of [San, Proposition 5.7], we shall make use of the
so-called Bernstein functor L0 [San, (12)] from the category of h-semisimple g-
modules to the category of g0̄ -semisimple g-modules. Actually, when restricting to
the category of h-semisimple g0̄ -module, the functor L0 is just the classical Bern-
stein functor, which is a special case of the functor P defined in Knapp and Vogan
[KV, (2.8)]. As we shall only need some standard properties of the Bernstein func-
tor, which can be found in [KV, San], we will not recall here the precise definition
which is quite involved.
An immediate consequence of the definition of the functor L0 is that every g0̄ -
composition factor of L0 (V ) is also a g0̄ -composition factor of V , counting multi-
plicity. Hence L0 takes a finitely generated h-semisimple g0̄ -module in the BGG
category to a finite-dimensional g0̄ -module. Denote by Li the ith derived functor of
L0 . Then, each Li (V ) is a finite-dimensional g-module, for any Verma g-module
V (which lies in the BGG category of g0̄ -modules). Now it is a classical result
(cf. [KV, Corollary 4.160]) that the resulting Euler characteristic when applying
L0 to a g0̄ -Verma module is given by the Weyl character formula, i.e.,
∞
∑w∈W (−1)ℓ(w) ew(λ+ρ0̄ )
∑ (−1)i ch Li (Indb Cλ ) =
g0̄
,
i=0
0̄
∏α∈Φ+ (eα/2 − e−α/2 )
0̄
g
and Li (Indb0̄ Cλ ) = 0 for i ≫ 0.
0̄
Given a short exact sequence of finite-dimensional h-semisimple g0̄ -modules
0 −→ M −→ E −→ N −→ 0.
It follows by the Euler-Poincaré principle that
∞ ∞ ∞
∑ (−1)i ch Li (M) + ∑ (−1)i ch Li (N) = ∑ (−1)i ch Li (E).
i=0 i=0 i=0
As a g0̄ -module, ∆(λ) and Indb0̄ Cλ ⊗ Λ(n∗1̄ ) have Verma flags, where a Verma
g
0̄
flag is a filtration whose sections are Verma modules. Since these two g0̄ -modules
2.1. Classification of finite-dimensional simple modules 43
also have identical characters, their Verma flags have the same Verma module mul-
tiplicities. Now the character of the b0̄ -module Cλ ⊗ Λ(n∗1̄ ) equals to eλ ∏β∈Φ+ (1 +
1̄
e−β ). Noting the W -invariance of ∏β∈Φ+ (eβ/2 + e−β/2 ), we have
1̄
∞ ∞ ( )
∑ (−1)i ch Li (∆(λ)) = ∑ (−1)i ch Li Indb0̄ Cλ ⊗ Λ(n∗1̄ )
g
0̄
i=0 i=0
( )
∑w∈W (−1)ℓ(w) w eλ+ρ0̄ ∏β∈Φ+ (1 + e−β )
1̄
=
∏α∈Φ+ (eα/2 − e−α/2 )
( )
0̄
be a half-integer Φ0̄ -dominant integral weight. Then the highest weight g-module
+
Since the Weyl group of sp(2m) is ∼ = Zn2 o Sn and it consists of the signed permu-
tations among δ1 , . . . , δm , we conclude by the assumption λm ≥ n that the weight
λ + ρ remains Φ+ 0̄
-dominant and regular. Thus the highest weight occurring in the
virtual character of finite-dimensional g-modules (2.3) is (λ + ρ) + ρ1 − ρ0 = λ,
and so L(λ) must be finite dimensional.
Conversely, suppose that L(λ) is finite dimensional, for a half-integer Φ+ 0̄
-
dominant integral weight λ. We apply the following sequence of mn odd reflections
to the standard fundamental and positive systems of g (see 1.3.3). We first apply
the n odd reflections with respect to δm − ε1 , δm − ε2 , . . . , δm − εn , then the n odd
reflections with respect to δm−1 − ε1 , . . . , δm−1 − εn etc., until we finally apply the
n odd reflections with respect to δ1 − ε1 , δ1 − ε2 , . . . , δ1 − εn . After a computation
similar to Example 1.31, the resulting fundamental system has the following form
⊗
⃝ ⃝ ··· ⃝ ··· ⃝=⇒a
ε1 − ε2 ε2 − ε3 εn − δ1 δ1 − δ2 δm−1 − δm δm
By the half-integer weight assumption on λ, the value of λ at any odd coroot be-
longs to 12 + Z (and hence is nonzero). By Lemma 1.36, after an odd reflection with
respect to an odd root α the new highest weight is λ − α, which again takes value
in 12 + Z at any odd coroot. Continuing this way, by a computation very similar
to Example 1.37 we see that the new highest weight corresponding to the above
Dynkin diagram is equal to λ − n ∑m i=1 δi + m ∑ j=1 ε j . Now this weight must take
n
non-negative integer value at h2δm by the finite dimensionality of L(λ), from which
we conclude that λm ≥ n.
2.1. Classification of finite-dimensional simple modules 45
µ
m
a Φ+
0̄
-dominant integral weight.
(1) Suppose that λ is a half-integer weight. Then L(λ) is finite dimensional if
and only if λm ≥ n.
(2) Suppose that λ is an integer weight. Then L(λ) is finite dimensional if
and only if λ is of the form µ♮ , where µ is an (m|n)-hook partition.
weight ∑kj=1 δ j , for k ≤ M. Write the conjugate partition µ′ = (ν1 , ν2 , . . .). Then
the spo(2M|2n + 1)-module ∧ν1 (C2M|2n+1 ) ⊗ ∧ν1 (C2M|2n+1 ) ⊗ · · · contains the ir-
reducible spo(2M|2n + 1)-module V of highest weight µ as a quotient, and hence
V is finite dimensional. Now, we use a sequence of odd reflections to change
the standard Borel of spo(2M|2n + 1) into a Borel subalgebra that contains as a
subalgebra the standard Borel of spo(2m|2n + 1). (Here spo(2m|2n + 1) is re-
garded as a subalgebra of spo(2M|2n + 1)). A sequence of odd reflections that
we may use for this purpose is the following: We first apply the n odd reflections
with respect to δM − ε1 , δM − ε2 , . . . , δM − εn , then the n odd reflections with re-
spect to δM−1 − ε1 , . . . , δM−1 − εn etc., until we finally apply the n odd reflections
with respect to δm+1 − ε1 , δm+1 − ε2 , . . . , δm+1 − εn . Thus, we have applied a total
of (M − m)n odd reflections. The subdiagram consisting of the first (m + n − 1)
vertices of the new Dynkin diagram is of type gl(m|n). In a similar way as in Ex-
ample 1.37, one shows that the highest weight of V with respect to the new Borel
is µ♮ . Hence by restriction we conclude that the spo(2m|2n + 1)-module L(µ♮ ) is
finite dimensional, proving the “if” direction of (2).
By Lemma 2.8, an integer Φ+ 0̄
-dominant integral highest weight for a simple
finite-dimensional g-module is necessarily of the form µ = µ1 δ1 + . . . + µm δm +
ν1 ε1 + . . . + νn εn , where (µ1 , . . . , µm ) and (ν1 , . . . , νn ) are partitions. To prove the
remaining condition ν′1 ≤ µm , it suffices to prove it in the case of m = 1 by noting
that spo(2|2n + 1) is a subalgebra of spo(2m|2n + 1). Let V be a finite-dimensional
simple spo(2|2n + 1)-module. Via the sequence of odd reflections with respect to
δ1 − ε1 , δ1 − ε2 , . . . , δ1 − εn we change the standard Borel to a Borel with an odd
non-isotropic simple root δ1 . Now if µ1 ≥ n, then there is no condition on µ and
we are done. Suppose now that µ1 = k < n, and we shall prove by contradiction
by assuming that νk+1 ≥ 1 (and so ν′1 > µ1 ). This implies that νi ≥ 1, for all
1 ≤ i ≤ k, and hence, when applying the odd reflection with respect to to δ1 − εi ,
the new highest weight is obtained from the old one by subtracting δ1 − εi , for
1 ≤ i ≤ k. This remains true also when applying the odd reflection with respect
to δ1 − εk+1 . Hence the resulting new highest weight has negative δ1 -coefficient.
But this implies that the new highest weight evaluated at the coroot h2δ1 must be
negative, contradicting the finite dimensionality of the module V . Hence µ must
satisfy the hook condition, as claimed.
For g = spo(2m|2n), a Φ+ 0̄
-dominant integral weight λ ∈ h∗ is called an integer
weight if it satisfies (DI.d-i), and it is called a half-integer weight if it satisfies
(DI.d-ii). The L(λ)’s for integer and half-integer weights λ again are quite different.
Proposition 2.13. Let g = spo(2m|2n) with n ≥ 2. Let λ = ∑m i=1 λi δi + ∑ j=1 υ j ε j
n
be a half-integer Φ0̄ -dominant integral weight. Then the highest weight g-module
+
a Φ+
0̄
-dominant integral weight.
(1) Suppose that λ is a half-integer weight. Then L(λ) is finite dimensional if
and only if λm ≥ n.
(2) Suppose that λ is an integer weight. Then L(λ) is finite dimensional if
and only if λ is of the form µ♮ or µ♮− , where µ is an (m|n)-hook partition.
Proof. Part (1) is Proposition 2.13. Since the proof for (2) is very similar to the
proof for Theorem 2.11(2), we shall only give a sketch below.
We recall from 1.3.4 that the standard Dynkin diagram of g = spo(2m|2n)
possesses a non-trivial diagram involution, which induces an involution of the Lie
superalgebra g. Twisting the module structure with this involution interchanges the
irreducible modules L(µ♮ ) and L(µ♮− ).
48 2. Finite-dimensional modules
∆(λ) = IndgbWλ .
The highest weight space of ∆(λ) can be naturally identified with Wλ of dimension
2(ℓ(λ)+δ(λ))/2 , and any nonzero vector in the highest weight space Wλ generates the
whole g-module ∆(λ). Following standard arguments for semisimple Lie algebras,
any g-submodule of ∆(λ) has a weight space decomposition, and then the Verma
module ∆(λ) contains a unique maximal proper submodule J(λ) (which is the sum
of all proper submodules). Hence, ∆(λ) has a unique irreducible quotient g-module
L(λ) = ∆(λ)/J(λ).
Assume that the g-module L(λ) is finite dimensional, and write λ = ∑ni=1 λi εi .
Since g0̄ ∼
= gl(n), the g0̄ -dominance integral condition imposes a necessary condi-
tion on λ: λi − λi+1 ∈ Z+ , for all 1 ≤ i ≤ n − 1. However, this condition on λ is not
sufficient for L(λ) to be finite dimensional, as we see in the following lemma.
2.1. Classification of finite-dimensional simple modules 49
Lemma 2.15. Let g = q(2) and λ = mε1 + mε2 ∈ h∗ with m ̸= 0. Then L(λ) is
infinite dimensional.
Proof. Recall the notations Eei j , E i j etc. from 1.1.4 and 1.2.6. Then Ee12 , Ee21 and
H1 − H2 = Ee11 − Ee22 form a standard sl(2)-triple s. Take a highest weight vector
vλ in L(λ). For λ = mε1 + mε2 , we have
To prove the lemma, it suffices to show that Ee21 vλ ̸= 0 in L(λ). Indeed, by the
sl(2) representation theory, Ee21 vλ ̸= 0 implies that the s-submodule generated by
vλ must be the infinite-dimensional Verma module of zero highest weight.
To that end, recalling H i = Eii + Eii , we compute that
Theorem 2.16. The highest weight irreducible q(n)-module L(λ) is finite dimen-
sional if and only if λ = ∑ni=1 λi εi satisfies the conditions (i)-(ii):
(i) λi − λi+1 ∈ Z+ ; (ii) λi = λi+1 implies that λi = 0, for 1 ≤ i ≤ n − 1.
Proof. For L(λ) to be finite dimensional, λ = ∑ni=1 λi εi must satisfy the usual dom-
inance integral condition for q(n)0̄ ∼
= gl(n), which is Condition (i); λ also must
satisfy Condition (ii) by applying Lemma 2.15 to various standard subalgebras of
q(n) which are isomorphic to q(2). This proves the “only if” part.
Assume that λ satisfies the conditions (i)-(ii), and assume in addition that λi ̸∈
Z. This implies that λi ̸= 0, for all i, and λ j > λ j+1 , for j = 1, . . . , n − 1. We want
to prove that L(λ) is finite dimensional. Write gε = n− +
ε + hε + nε for the standard
triangular decomposition of gε , for ε ∈ Z2 . The Verma g-module ∆(λ), regarded as
a g0̄ -module, has a g0̄ -Verma flag, and we have
( )
ch∆(λ) = 2⌊(n+1)/2⌋ ch Indh0̄ +n+ Cλ ⊗ ∧(n−
g
1̄
) ,
0̄ 0̄
where ⌊r⌋ for a real number r denotes the largest integer no greater than r. As in the
proof of Proposition 2.5 we apply the Bernstein functor L0 and its derived functor
Li (with g0̄ = gl(n)) to ∆(λ) and compute the corresponding Euler characteristic
50 2. Finite-dimensional modules
(recall Φ1̄ and Φ0̄ are identical copies of the root system for gl(n)):
( )
∑ ⌊(n+1)/2⌋
∑ −
g0̄
(−1) i
chL i (∆(λ)) =2 (−1)i
chL i Ind C
h +n+ λ
⊗ ∧(n1̄
)
0̄ 0̄
i≥0 i≥0
( )
∑w∈W (−1)ℓ(w) w eλ+ρ0̄ ∏α∈Φ+ (1 + e−α )
=2⌊(n+1)/2⌋ 1̄
∏α∈Φ+ (1 + e−α )
=2 ⌊(n+1)/2⌋ 1̄
Since the highest weight appearing in this last expression is clearly λ and the Euler
characteristic is a virtual character of finite-dimensional g-modules (see the proof
of Proposition 2.5), L(λ) is a finite-dimensional module.
Now suppose that λ satisfies the conditions (i)-(ii) of the theorem and in ad-
dition that λi ∈ Z for each i. We write λ = (λ1 , . . . , λk , 0, . . . , 0, λl , . . . , λn ) such
that λ1 > · · · > λk > 0 > λl > · · · > λn . Denote λ+ = (λ1 , . . . , λk , 0, . . . , 0) and
λ− = (0, . . . , 0, λl , . . . , λn ). By Theorem 3.46 on Sergeev duality in Chapter 3, L(µ)
is finite dimensional for µ = ∑i µi εi corresponding to a strict partition (µ1 , . . . , µn ).
Then, the dual module L(µ)∗ ∼ = L(µ∗ ) is finite dimensional, where we have denoted
∗
µ = − ∑i=1 µn+1−i εi . It follows that both L(λ+ ) and L(λ− ) are finite dimensional.
n
where we recall from Section 1.1.1 that sdimW = dimW0̄ − dimW1̄ stands for the
superdimension of a vector superspace W = W0̄ ⊕ W1̄ . If we regard e as the Euler
number, a formal expansion of eµ gives us
∞
1 [ ]
[schV ](h) = ∑ str π(h)i , ∀h ∈ g.
i=0 i!
Proposition 2.18. Let g be a basic Lie superalgebra with Cartan subalgebra[ h. Let]
(π,V ) be a finite-dimensional g-module. Then the polynomial functions str π(h)i
on h, for i > 0, can be identified with the homogeneous components of the super-
character of π, and they arise as restrictions of g-invariant polynomial functions
on g.
2.2.2. The central characters. In this subsection, we let g be a basic Lie superal-
gebra, and h be its Cartan subalgebra.
Recall that g is equipped with an even non-degenerate invariant supersymmet-
ric bilinear form (see Section 1.2.2). Furthermore, the bilinear form (·, ·) on g
allows us to identify g with its dual g∗ , and h with h∗ so that we have correspond-
ing isomorphisms S(g)g ∼ = S(g∗ )g and S(h) ∼= S(h∗ ). Via such identifications, the
∗ ∗
restriction map S(g ) → S(h ) and the induced map η : S(g) → S(h) are homomor-
phisms of algebras.
Let g = n− ⊕h⊕n+ be a triangular decomposition. By the PBW Theorem 1.32,
we have U(g) = U(h) ⊕ (n−U(g) +U(g)n+ ), and we denote by ϕ : U(g) → U(h)
the projection associated to this direct sum decomposition. Since [h, n± ] ⊆ n± ,
by considering ad h(z) = 0 for any element z ∈ Z(g), we conclude that z affords a
unique expression of the form
(2.5) z = hz + ∑ n− +
i hi ni , for hz , hi ∈ U(h), n± ± ±
i ∈ n U(n ).
i
52 2. Finite-dimensional modules
Hence Z(g) consists of only even elements. The restriction of ϕ to Z(g), denoted
again by ϕ, defines an algebra homomorphism from Z(g) to U(h), which sends z
to hz .
As usual, we freely identify S(h) with the algebra of polynomial functions on
h∗ . Extending the standard pairing ⟨·, ·⟩ : h∗ × h → C, we can make sense of ⟨λ, p⟩
for λ ∈ h∗ and p ∈ S(h) by regarding it as the value at λ of the polynomial function
p on h∗ . For λ ∈ h∗ , define a linear map χλ : Z(g) → C by
χλ (z) = ⟨λ, ϕ(z)⟩.
Lemma 2.19. Let g be a basic Lie superalgebra. An element z ∈ Z(g) acts as the
scalar χλ (z) on any highest weight g-module V (λ) of highest weight λ.
Proof. Our proof here is phrased so that it makes sense for q(n) later on as well.
Let z ∈ Z(g). The highest weight space V (λ)λ is an irreducible h-module,
which is one-dimensional for basic g (but may fail to be one-dimensional in the
q(n) case). The even element z commutes with h, and hence by Schur’s Lemma, z
acts as a scalar ξ(z) on V (λ)λ . Now let vλ be a highest weight vector of V (λ), and
write an arbitrary vector u ∈ V (λ) as u = xvλ for x ∈ U(h + n− ). Since z is central,
we have zu = xzvλ = ξ(z)u. It follows by (2.5) that
( )
zvλ = hz + ∑ n− i hi ni vλ = ⟨λ, hz ⟩vλ = χλ (z)vλ .
+
i
Hence ξ(z) = χλ (z). The lemma is proved.
Proof. The character of the Verma module ∆(µ) of highest weight µ ∈ h∗ is given
by ch∆(µ) = eµ+ρ D−1 , where
∏β∈Φ+ (eβ/2 − e−β/2 )
0̄
D := .
∏ α∈Φ+ (eα/2 + e−α/2 )
1̄
Since the character of a module equals the sum of the characters of its compo-
sition factors, we have
(2.6) ch∆(λ) = ∑ bµλ chL(µ),
µ
where bµλ ∈ Z+ and bλλ = 1. Since ∆(λ) is a highest weight module, the µ’s in the
summation need to satisfy λ − µ ∈ ∑α∈Φ+ Z+ α and also χλ = χµ by Lemma 2.19.
We choose a total order ≥ on the set λ − ∑α Z+ α with the property that ν ≥ µ
if ν − µ ∈ ∑α∈Φ+ Z+ α. Then the ordered sets {ch∆(µ)}µ,≥ and {chL(µ)}µ,≥ are
two ordered bases for the same space. Now (2.6) says that the matrix expressing
{ch∆(µ)}µ,≥ in terms of {chL(µ)}µ,≥ is upper triangular with 1 along the diagonal,
and hence we have
chL(λ) = ∑ aµλ ch∆(µ),
µ
Assume for now that λ ∈ h∗ is chosen so that the irreducible g-module L(λ)
is finite dimensional. Thus L(λ) is a semisimple g0̄ -module, and so chL(λ) is W -
invariant. On the other hand, D is W -anti-invariant by Theorem 1.15(4). Thus the
right-hand side of (2.8) is W -anti-invariant, and hence can be written as
(2.9) ∑ aµλ ∑ (−1)ℓ(w) ew(µ+ρ) ,
µ∈X w∈W
where X consists of Φ+
0̄
-dominant integral weights such that aµλ ̸= 0. We compute
that aw(λ+ρ)−ρ,λ = ±aλλ = ±1, and hence by (2.7) we have χλ = χw(λ+ρ)−ρ , for all
w ∈ W.
Now (1) as a polynomial identity on λ ∈ h∗ follows from the following claim.
Claim. S := {λ ∈ h∗ | L(λ) is finite dimensional} is Zariski dense in h∗ .
Indeed, given a Φ+ 0̄
-dominant integral weight µ ∈ h∗ , the induced module
g 0
Indg0̄ L (µ) is finite dimensional and has a highest weight µ+2ρ1 . It is clear that the
set of such weights µ + 2ρ1 associated to all Φ+ 0̄
-dominant integral µ ∈ h∗ (which
∗
is a subset of S) is Zariski dense in h . Hence the claim follows.
54 2. Finite-dimensional modules
By (1), we have that ⟨λ + ρ, hc(z)⟩ = ⟨w(λ + ρ), hc(z)⟩, for all w ∈ W and z ∈
Z(g). Thus hc(z) ∈ S(h)W , and equivalently, hc∗ (Z(g)) ⊆ S(h∗ )W by definition of
hc∗ .
2.2.4. Invariant polynomials for gl and osp. Suppose that g is of type gl or osp.
The goal of this subsection is to establish Proposition 2.21 which partially describes
the image Im(η∗ ) via supersymmetric polynomials, see Appendix A.2.1 (The word
“partially” can and will be removed subsequently). The description of this image
plays an important role in the study of the Harsih-Chandra map later on.
Associated to a positive system Φ+ , we have the subset Φ̄+
1̄
of positive isotropic
odd roots defined in (1.27). Introduce the following element
(2.10) P := ∏ α ∈ S(h∗ ).
α∈Φ̄+
1̄
Here and further, we shall identify S(h∗ ) = C[δ, ε], where we set
δ = {δ1 , . . . , δm }, ε = {ε, . . . , εn }.
Hence an element of S(h∗ )W is precisely a polynomial which is symmetric in δ and
symmetric in ε. Recalling from A.2.1 the definition of supersymmetric polynomi-
als, we introduce the following subalgebra of S(h∗ )W :
(2.12) S(h∗ )W
sup = {supersymmetric polynomials in variables δ, ε}.
Invariant polynomials for g = spo(2m|2n + 1). Recall the Weyl group W for
g = spo(2m|2n + 1), which is by definition the same as for sp(2m) ⊕ so(2n + 1), is
2 o Sm ) × (Z2 o Sn ),
WB := (Zm n
(2.13)
and hence an element f ∈ S(h∗ )W = C[δ, ε]W is exactly a polynomial which is
symmetric among δ2i ’s and symmetric among ε2j ’s. We introduce the following
subalgebra of S(h∗ )W :
(2.14) S(h∗ )W
sup = {supersymmetric polynomials in δi , ε j , 1 ≤ i ≤ m, 1 ≤ j ≤ n}.
2 2
(For other positive systems, P might differ from the above formula by an irrelevant
sign.) By (A.28), we have that f P ∈ S(h∗ )W ∗ W
sup for any f ∈ S(h ) .
polynomials in δi , ε j , 1 ≤ i ≤ m, 1 ≤ j ≤ n.
2 2
(For other positive systems, P might differ by a sign from the above formula.)
Since Q is an integral polynomial in e±δi and e±ε j that is W -invariant, Q is actually
a virtual g0̄ -character. By Remark 2.6, D1̄ Q is a virtual supercharacter of finite-
dimensional g-modules. Hence every homogeneous term of D1̄ Q lies in Im(η∗ ) by
Proposition 2.18, and in particular so does the lowest degree term of D1̄ Q, i.e.,
ε1 ε2 · · · εn Pg ∈ Im(η∗ ).
Proof. Fix a positive system Φ+ . We may assume that the isotropic odd root α ∈
Φ+ (otherwise, use −α to replace α).
First suppose that α is an isotropic simple root in Φ+ . Then (ρ, α) = 12 (α, α) =
0, and hence (λ, α) = 0 by assumption. Let vλ be a highest weight vector in the
Verma module ∆(λ). Then,
eα fα vλ = [eα , fα ]vλ = hα vλ = (λ, α)vλ = 0.
Also, we have eγ fα vλ = 0 for other simple roots γ by weight considerations. Hence,
fα vλ is a singular vector in ∆(λ). This gives rise to a non-trivial g-module homo-
morphism ∆(λ − α) → ∆(λ), whence χλ−α = χλ .
Now suppose α is any isotropic root in Φ+ . By Lemma 1.25, there exists
w ∈ W such that β = w(α) is simple. Then (α, α) = (β, β) = 0 by the W -invariance
of (·, ·). By Proposition 2.20, we have
χλ−α = χw(λ−α+ρ)−ρ = χµ ,
where we have denoted µ := w(λ + ρ) − ρ − β and used w(α) = β. Now by the
W -invariance of (·, ·) again, we compute that
(µ + ρ, β) = (w(λ + ρ), w(α)) − (β, β) = (λ + ρ, α) − 0 = 0,
Thus, we can apply the special case established in the preceding paragraph to the
weight µ and simple root β to obtain that
χλ−α = χµ = χµ+β = χw(λ+ρ)−ρ = χλ ,
where we have used Proposition 2.20 again in the last equality. This implies that
χλ = χλ−tα , for any t ∈ Z+ and any isotropic odd root α with (λ + ρ, α) = 0. Since
Z+ is Zariski dense in C, we conclude that the polynomial identity χλ = χλ+tα
holds for all t ∈ C.
where S(h∗ )W
sup is defined in (2.12), (2.14), and (2.15), respectively.
Proof. By Propositions 2.21 and 2.23, we have a map hc∗ : Z(g) → Im(η∗ ), and
that hc∗ (Z(g)) ⊆ S(h∗ )W
sup ⊆ Im(η∗ ). So it suffices to show that hc∗ (Z(g)) = Im(η∗ )
in order to complete the proof. The commutative diagram (where gr denotes the
associated graded)
gr γ−1
(2.16) gr Z(g) / S(g)g
H ∼
HH=
HH
HH η∗
HH
gr hc∗ $
S(h∗ )
implies that gr hc∗ : gr Z(g) → Im(η∗ ) is surjective.
Now the surjectivity of the map hc∗ : Z(g) → Im(η∗ ) follows by a standard
filtered algebra argument. Indeed, consider hc∗ (Z(g)) and Im(η∗ ) as filtered spaces:
C= ∼ hc (Z(g))0 ⊆ hc (Z(g))1 ⊆ · · · , C∼= Im(η∗ )0 ⊆ Im(η∗ )1 ⊆ · · · .
∗ ∗
2.2. Harish-Chandra homomorphism and linkage 59
We shall prove by induction on k that hc∗ (Z(g))k = Im(η∗ )k , for all k. Clearly we
have hc∗ (Z(g))0 = Im(η∗ )0 . Now assume that hc∗ (Z(g))k−1 = Im(η∗ )k−1 . Let us
take a = η∗ (x) ∈ Im(η∗ )k for x ∈ S(g)g . Then, a = gr (hc∗ γ)(x), and
γ−1
(2.17) Z(g) / S(g)g
∼
= t
tt
tt η
hc∗
ttt η∗
zt ∼
S(h∗ )W o
=
sup S(h)W sup
θ
Remark 2.25. The surjective homomorphism hc : Z(g) → S(h)W sup is actually injec-
tive as well, and hence all the homomorphisms in the diagram (2.17) are isomor-
phisms. Actually Sergeev [Sv3, Corollary 1.1] showed that η : S(g)g → S(h) is
injective, and the injectivity of hc follows from this by a filtered algebra argument.
Except for its intrinsic value, the injectivity does not seem to play any crucial role
in the representation theory of g.
The following corollary is immediate from Proposition 2.21 and Theorem 2.24,
and it will be used later on.
Corollary 2.26. Let g be one of the Lie superalgebras gl(m|n), spo(2m|2n + 1), or
spo(2m|2n). Then, for any f ∈ S(h)W , we have f P ∈ hc∗ (Z(g)).
2.2.6. Linkage for gl and osp. In this subsection, we let g denote the Lie su-
peralgebra gl(m|n), spo(2m|2n), or spo(2m|2n + 1). Using the description of the
images of Harish-Chandra homomorphisms, we obtain a necessary and sufficient
condition for two central characters of g to be equal.
Let h be its standard Cartan subalgebra of diagonal matrices. Let {δi , ε j |1 ≤
i ≤ m, 1 ≤ j ≤ n} be the standard basis of h∗ . Let Φ+ be the standard positive
system of the root system Φ for g. The following lemma is proved by a direct
computation.
We define a relation ∼ on h∗ by declaring
λ ∼ µ, for λ, µ ∈ h∗ ,
60 2. Finite-dimensional modules
Proof. Assume first that λ ∼ µ. Since the αi ’s are orthogonal to another, the second
condition in (2.18) implies that we can apply repeatedly Lemma 2.22 to conclude
that
χλ = χλ−c1 α1 = χλ−c1 α1 −c2 α2 = . . . = χλ−∑ℓ ca αa .
a=1
∑mi=1 µi δi + ∑ j=1 η j ε j . Recall from Theorem 2.24 that for g = gl(m|n) the poly-
n
nomials σkm,n given in (2.11) lie in hc∗ (Z(g)), for all k ∈ Z+ , while in the case of
g = spo(2m|2n) or spo(2m|2n + 1), the polynomials σkm,n lie in hc∗ (Z(g)), for all
even k ∈ Z+ . Let zkm,n be an element in Z(g) with hc∗ (zkm,n ) = σkm,n , where k is
assume to be even for type spo. Then,
m n
χλ (zkm,n ) = ∑ (λ + ρ, δi )k − ∑ (λ + ρ, ε j )k .
i=1 j=1
independent for distinct a. In the identity obtained from (2.19) by canceling the
terms corresponding to all the pairs (ia , ja ) and (i′b , jb′ ), the survived (λ + ρ, δi )’s
must match bijectively with the survived (µ + ρ, δi )’s, while the (λ + ρ, ε j )’s and
(µ + ρ, ε j )’s must match bijectively. So ℓ = r. Now we can find a pair of per-
mutations w = (σ1 , σ2 ) ∈ Sm × Sn which extends the above bijections such that
w−1 (µ + ρ) − λ − ρ is a linear combination of these mutually orthogonal isotropic
odd roots δia − ε ja , and hence λ ∼ µ.
(2) Now suppose that g = spo(2m|2n + 1). Then, for all even k ≥ 0, we have
χλ (zkm,n ) = χµ (zkm,n ).
This is equivalent to the following generating function identity:
m n m n
∑ e(λ+ρ,δ ) t − ∑ e(λ+ρ,ε ) t = ∑ e(µ+ρ,δ ) t − ∑ e(µ+ρ,ε ) t .
2 2 2 2
(2.20) i j i j
we have
τ1 σ1 w−1 (µ + ρ) − (λ + ρ)
( ) ( )
= τ1 σ1 w−1 (µ + ρ) − (λ + ρ) + τ1 σ1 (λ + ρ) − (λ + ρ)
( ) ℓ
= − 2(λ + ρ, δi1 ) + c1 (δi1 − s1 ε j1 ) + ∑ ca (δia − sa ε ja ).
a=2
Hence we have obtained a set of mutually orthogonal of isotropic odd roots, a
corresponding set of complex numbers, and the Weyl group element wσ1 τ1 ∈ W
satisfying (2.18). Thus, λ ∼ µ.
By the description of the center Z(g) in Theorem 2.24 and (2.15), we have
additional elements in Z(g) to use. In particular, we have
⟨ ⟩ ⟨ ⟩
(2.22) ε1 ε2 · · · εn P, θ−1 (λ + ρ) = ε1 ε2 · · · εn P, θ−1 (µ + ρ) .
Assume
( now that λ is typical ) (and so is µ), and(hence ℓ = 0. Then the (m ) + n)-
tuple (λ + ρ, δi )2 , (λ + ρ, ε j )2 i, j coincides with (µ + ρ, δi )2 , (µ + ρ, ε j )2 i, j , up
to a permutation in Sm × Sn . Recall P = ∏i, j (δ2i − ε2j ). Hence, ⟨P, θ−1 (λ + ρ)⟩ =
⟨P, θ−1 (µ + ρ)⟩ =
̸ 0, since λ and µ are typical. By canceling this nonzero factor in
(2.22), we obtain that
⟨ ⟩ ⟨ ⟩
ε1 ε2 · · · εn , θ−1 (λ + ρ) = ε1 ε2 · · · εn , θ−1 (µ + ρ) .
Note that in this case (2.18) reads that µ + ρ = w(λ + ρ). If (εi , λ + ρ) ̸= 0 for all
i, we conclude that w2 changes the signs for an even number of ε j ’s, and hence
w = (w1 , w2 ) lies in the Weyl group W . If (εi , λ + ρ) = 0 for some i and w ̸∈ W ,
then w′ = (w1 , w2 σ) ∈ W satisfies that µ + ρ = w′ (λ + ρ), where σ denotes the sign
change at εi .
Theorem 2.27 has the following implications. For a composition factor L(µ) in
a Verma module ∆(λ) of g, µ must satisfy the following conditions: λ − µ ∈ Z+ Φ+
and µ ∼ λ. This will be referred to as the linkage principle for Lie superalgebra g
of type gl and osp.
We fix a positive system Φ+ of the root system Φ for g. Recall the subset
+
Φ1̄ ⊆ Φ+ defined in (1.27).
Definition 2.28. The degree of atypicality of an element λ ∈ h∗ , denoted by #λ, is
+
the maximum number of mutually orthogonal roots α ∈ Φ1̄ such that (λ + ρ, α) =
0. An element λ ∈ h∗ is said to be typical (relative to Φ) if #λ = 0, and is atypical
otherwise.
Corollary 2.29. If χλ = χµ , then the degrees of atypicality for λ and µ coincide.
Proof. This can be read off from the proof of Theorem 2.27.
Corollary 2.30. A finite-dimensional spo(2m|1)-module is completely reducible.
2.2. Harish-Chandra homomorphism and linkage 63
Theorem 2.32. Let g be of type gl or osp. Let λ ∈ h∗ be a typical weight such that
L(λ) is a finite-dimensional irreducible representation of g. Then
∏α∈Φ+ (1 + e−α )
chL(λ) =
∏
1̄
(1 − e−β )
∑ (−1)ℓ(w) ew(λ+ρ)−ρ .
β∈Φ+
0̄
w∈W
where X consists of Φ+
0̄
-dominant weights such that aµλ ̸= 0.
Since λ is typical and χλ = χµ for all µ ∈ X, it follows by Lemma 2.31 that
X = λ. Recalling that aλλ = 1, we can rewrite (2.23) as
DchL(λ) = ∑ (−1)ℓ(w) ew(λ+ρ) ,
w∈W
2.3.1. The central characters for q(n). In this section, let g be the queer Lie
superalgebra q(n), and h = h0̄ + h1̄ be its (nonabelian) Cartan subalgebra. As this
subsection is parallel to that of the basic Lie superalgebra case in Section 2.2.2, we
will be brief and only point out the differences from therein.
Recall that g is equipped with a non-degenerate odd invariant supersymmetric
bilinear form (·, ·), which allows us to identify g with its dual g∗ , h with h∗ , and
h0̄ with h∗1̄ . Via such identifications, the restriction map S(g∗ ) → S(h∗ ) and the
induced map η : S(g) → S(h) are homomorphisms of algebras. Let g = n− ⊕ h ⊕
n+ be a triangular decomposition. As before we have a projection ϕ : U(g) →
U(h), and any element z ∈ Z(g) affords a unique expression of the form (2.5).
Moreover, it follows from ad h(z) = 0 that hz ∈ U(h0̄ ). Hence Z(g) consists of only
even elements. The restriction of ϕ to Z(g), also denoted by ϕ, defines an algebra
2.3. Harish-Chandra homomorphism and linkage for q(n) 65
homomorphism from Z(g) to U(h0̄ ), which sends z to hz . For λ ∈ h∗0̄ , define a linear
map χλ : Z(g) → C by χλ (z) = ⟨λ, ϕ(z)⟩.
The proof of Lemma 2.19 also works for the following lemma.
Lemma 2.35. Let g = q(n). An element z ∈ Z(g) acts as the scalar χλ (z) on any
highest weight g-module V (λ) of highest weight λ.
Recall that the Weyl group W for g = q(n) is the symmetric group Sn .
Proposition 2.37. Let g = q(n). We have χλ = χw(λ) , for all w ∈ W and λ ∈ h∗0̄ .
Also, we have hc(Z(g)) ⊆ S(h0̄ )W .
66 2. Finite-dimensional modules
Proof. By Lemma 2.36, we have a nonzero homomorphism ∆(si (λ)) → ∆(λ), for
λ ∈ h∗0̄ with λi − λi+1 ∈ N, which implies by Lemma 2.19 that χλ = χsi (λ) . Hence,
χλ = χw(λ) , for all w ∈ W and for all integral weights λ = ∑ni=1 λi εi satisfying λ1 >
. . . > λn . Since the set of such weights is Zariski dense in h∗0̄ , it follows that χλ =
χw(λ) , for all w ∈ W and λ ∈ h∗0̄ . That is, ⟨λ, hc(z)⟩ = ⟨w(λ), hc(z)⟩, for all z ∈ Z(g),
w ∈ W and λ ∈ h∗0̄ . Hence, we conclude that hc(Z(g)) ⊆ S(h0̄ )W thanks to the W -
invariance of (·, ·).
Remark 2.38. One can imitate the proof of Proposition 2.20 to give an alternative
proof of Proposition 2.37, bypassing Lemma 2.36. On the other hand, a second
proof of Proposition 2.20 can be given in the spirit of the proof of Proposition 2.37
with some extra work involving odd reflections, as the Weyl group W is not gener-
ated by the even simple reflections.
εi (H j ) = δi j .
That is, θ(Hi ) = εi , for each i. Let (π,V ) be the natural representation of g =
q(n) on the vector superspace Cn|n . We recall the odd trace operator otr : g →
C from (1.26). A straightforward super generalization of the arguments ( for
) Lie
algebras as given in Carter [Car, pp.212-213] shows that x 7→ otr π(x)2k−1 , for
k ∈ N, defines
( )a g-invariant homogeneous polynomial on g of degree k; and note
that otr π(x)2k = 0 by definition of the odd trace. Restricting to h1̄ we obtain a
homogeneous polynomial on (Πh1̄ of degree ) 2k −1. Then a direct calculation shows
that, for x ∈ h1̄ and k ≥ 1, otr π(x)2k−1 equals the value at x of the polynomial
n
p2k−1 (ε) = ∑ ε2k−1
i , k ∈ N.
i=1
We define
(2.26) P := ∏ (εi + ε j ) ∈ S(Πh∗1̄ ).
1≤i< j≤n
Proof. (1) Assume that k = 0. Note that H i and H i+1 act trivially on the highest
weight space Wλ of ∆(λ), and hence H i vλ = H i+1 vλ = 0 for any highest weight
vector vλ . Then, Eei,i+1 u = (Hi − Hi+1 )vλ = 0, and E i,i+1 u = (H i − H i+1 )vλ = 0.
Also it follows from weight consideration that Eej, j+1 u = E j, j+1 u = 0 for j ̸= i.
Hence, u is a singular vector.
(2) Assume that k ̸= 0. We again identify the highest weight space of the Verma
module ∆(λ) as the h-module Wλ . We compute that
λ([H i ± H i+1 , H i ± H i+1 ]) = 2λ(Hi + Hi+1 ) = λi + λi+1 = 0.
Hence, by Lemma 1.38, there exists a highest weight vector vλ ∈ Wλ such that
wλ := (H i − H i+1 )vλ ̸= 0 and (H i + H i+1 )vλ = 0. It follows that
(H i + H i+1 )wλ = λ(Hi − Hi+1 )vλ = 2kvλ ,
(H i − H i+1 )wλ = 0.
We compute that
Eei,i+1 Eei+1,i wλ = λ(Hi − Hi+1 )wλ = 2kwλ ,
E i,i+1 Eei+1,i wλ = (H i − H i+1 )wλ = 0,
Eei,i+1 E i+1,i vλ = (H i − H i+1 )vλ = wλ ,
E i,i+1 E i+1,i vλ = (H i + H i+1 )vλ = 0.
We summarize various maps for the q(n) case in the following diagram, which
only becomes commutative when we pass from Z(g) to its associated graded (note
that the other algebras are graded).
γ−1
(2.27) Z(g) / S(g)g
∼
=
tt
tt
hc∗ tttη η
ytt ∼ ∗
S(Πh∗ )W o
=
1̄ S(h )W
0̄
θ
Remark 2.44. Actually hc and hc∗ are injective (see Sergeev [Sv3]), and so we have
an isomorphism of algebras hc∗ : Z(g) → Γn,C . The injectivity of hc∗ does not play
any role in the representation theory of q(n) in the book.
α∨ := Hi + H j ∈ h0̄ .
λ ∼ µ, for λ, µ ∈ h∗0̄ ,
2.3. Harish-Chandra homomorphism and linkage for q(n) 69
We first observe that |Iλ | = |Iµ | (and hence ℓ = r) by dividing both sides of the
above identity by t and then taking the limit t 7→ 0. By the definition of Iλ and
Iµ , we conclude from (2.29) that there is a bijection between (λi )i∈Iλ and (µ j ) j∈Iµ .
Now we can find a permutation w ∈ Sn which extends the bijection between Iλ and
Iµ above such that w−1 µ − λ is a linear combination of the roots εia − ε ja . Thus,
λ ∼ µ.
Theorem 2.45 has the following implications. For a composition factor L(µ)
in a Verma module ∆(λ) of q(n), µ must satisfy the following conditions: λ − µ ∈
Z+ Φ+ , and µ ∼ λ. This will be referred to as the linkage principle for q(n).
Definition 2.46. Let g = q(n) and h be its standard Cartan subalgebra. The degree
of atypicality of a weight λ ∈ h∗0̄ , denoted by #λ, is the maximum number of pairs
(ia , ja ) with all ia , ja distinct, such that ⟨λ, Hia + H ja ⟩ = 0. A weight λ ∈ h∗0̄ is called
typical if #λ = 0, and is atypical otherwise.
Corollary 2.47. If χλ = χµ for λ, µ ∈ h∗0̄ , then the degrees of atypicality for λ and
µ coincide.
∏ ⟨λ, Hi + H j ⟩ ̸= 0.
1≤i< j≤n
Lemma 2.48. Let g = q(n) and λ, µ ∈ h∗0̄ with λ typical. Suppose that χλ = χµ .
Then, there exists w ∈ W such that µ = w(λ), and µ is typical.
Theorem 2.49. Let g = q(n) and let λ ∈ h∗0̄ be a typical weight such that the irre-
ducible q(n)-module L(λ) is finite dimensional. Then
ℓ(λ)+δ(λ)
λ
∏β∈Φ+ (1 + e−β )
1̄
ch∆(λ) = 2 2 e .
∏α∈Φ+ (1 − e−α )
0̄
Observe that the highest weight spaces of L(µ) and ∆(µ), for every µ ∈ W λ, have
ℓ(λ)+δ(λ)
dimension 2 2 , thanks to ℓ(µ) = ℓ(λ).
Imitating the proof of Proposition 2.20, we obtain that
∏α∈Φ+ (1 − e−α )
(2.30)
∏
0̄
(1 + e−β )
chL(λ) = ∑ aµλ ∑ (−1)ℓ(w) ew(µ) ,
β∈Φ+
1̄
µ∈X w∈W
where X consists of Φ+
0̄
-dominant integral weights such that aµλ ̸= 0.
Since λ is typical and χλ = χµ for all µ ∈ X, it follows by Lemma 2.48 that
ℓ(λ)+δ(λ)
µ = λ. Recalling that aλλ = 2 2 , we can rewrite (2.30) as
∏β∈Φ+ (1 + e−β )
chL(λ) = 2 2
∑ (−1)ℓ(w) ew(λ+ρ) ,
1̄
w∈W
Remark 2.50. The assumption on λ in Theorem 2.49 forces λi > λi+1 for all i, and
so ℓ(λ) = n or ℓ(λ) = n − 1.
2.4.1. Extremal weights for gl(m|n). We start with some general remarks for any
basic Lie superalgebra g. Let L be a finite-dimensional g-module. Then given any
Borel subalgebra b of g associated to a positive system Φ+ , there exists a unique
weight λb for L such that λb +α is not a weight for L for any α ∈ Φ+ , and the weight
space Lλb is one-dimensional (cf. Proposition 1.35). The weight λb is called the
b-extremal weight for the g-module L. Two Borel subalgebras b and b′ of g are in
general not conjugate. Thus, the b-extremal weight and the b′ -extremal weight of L
may in general not be conjugate by the Weyl group W , in contrast to the semisimple
Lie algebra setting.
In this subsection we let g = gl(m|n). Let h be its standard Cartan subalgebra
with standard Borel subalgebra bst . For an (m|n)-hook partition λ we may regard
λ♮ as an element in h∗ as in (2.4). We shall determine the b-extremal weight of
L(g, bst , λ♮ ) for any Borel subalgebra b. The class of gl(m|n)-modules L(g, bst , λ♮ ),
for all (m|n)-hook partitions λ, are called the polyonomial modules of gl(m|n),
and they will feature significantly in Chapter 3 on Schur duality.
Recall the weights δi and ε j from Section 1.2.3. Let b be a Borel subalgebra
of g and Φ+ be its positive system. Assume that the εδ-sequence from Section 1.3
associated to b is δd1 εe1 δd2 εe2 · · · δdr εer where the exponents denote the correspond-
ing multiplicities (all di , ei are positive except possibly d1 = 0 or er = 0). There
exist a permutation s of {1, . . . , m} and a permutation t of {1, . . . , n} such that the
δ’s and ε’s appearing in the εδ-sequence from left to right are δs(1) , δs(2) , . . . , δs(m) ,
and εt(1) , εt(2) , . . . , εt(n) , respectively.
Define
u u
(2.31) du := ∑ da , and eu := ∑ ea ,
a=1 a=1
It is elementary to read off the b-Frobenius coordinates of λ from the Young dia-
gram of λ in general, as illustrated by the next example.
Example 2.51. Let b be the Borel subalgebra of gl(5|4) associated to the following
fundamental system:
2.4. Extremal weights of finite-dimensional simple modules 73
δ2 − δ3 ε3 − ε2 δ1 − δ4 ε4 − ε1
⊗ ⊗ ⊗ ⊗
⃝ ⃝ ⃝ ⃝
δ3 − ε3 ε2 − δ1 δ4 − ε4 ε1 − δ5
Proof. Set V = L(g, bst , λ♮ ). One distinguished feature for V is that all weights ν
of V are polynomial in the sense that ν(Eii ) ∈ Z+ for all i ∈ I(m|n); see Chapter 3,
Theorem 3.10.
The theorem holds for the standard Borel subalgebra bst , which corresponds to
st
the sequence of m δ’s followed by n ε’s, thanks to λb = λ♮ . By Corollary 1.27,
bst can be converted to b by a finite number of real and odd reflections. Thus, we
are led to consider two Borel subalgebras b1 and b2 , where b2 is obtained from
b1 by applying a simple reflection corresponding to a simple root α. Let µ be the
b2 -extremal weight for V . To complete the proof of the theorem, we will show that
µ = λb2 , assuming the theorem holds for b1 .
74 2. Finite-dimensional modules
If α is an even root, then rα (λb1 ) = λb2 by definition, and the validity of the
theorem for b1 implies its validity for b2 .
Now assume that α = ±(δi − ε j ) is an odd simple root. The corresponding
coroot is hα = ±(Eī,ī + E j j ).
If λb1 (hα ) = 0, then we must have λb1 (Eī,ī ) = λb1 (E j j ) = 0 (as each has to be
nonnegative), and λb1 = λb2 . Then by Lemma 1.36 we have µ = λb1 = λb2 .
Suppose that λb1 (hα ) ̸= 0 and α = δi − ε j . Diagrammatically we can represent
λb1 (Eī,ī )
= a and λb1 (E j j ) = b by Diagram (ii) below, and this forces that a > 0.
By Lemma 1.36, we have µ = λb1 − δi + ε j , and hence µ(E j j ) = b + 1 and µ(Eī,ī ) =
a − 1, which are represented by Diagram (i) below. The remaining parts of µ are
the same as those for λb1 and hence we conclude that µ = λb2 , as claimed.
Suppose that λb1 (hα ) ̸= 0 and α = ε j − δi . Diagrammatically we can represent
λb1 (Ej j ) = b + 1 and λ (Eī,ī ) = a − 1 by Diagram (i) below, and this forces that
b1
a−1 a
b+1
b
(i) (ii)
o
In this case Theorem 2.52 states the coefficients of δ and ε in λb are given
by the modified Frobenius coordinates (pi |qi )i≥1 of the partition λ (re-
spectively, λ′ ), when the first simple root is of the form δ−ε (respectively,
2.4. Extremal weights of finite-dimensional simple modules 75
7
4
2
4 2 1
2.4.2. Extremal weights for spo(2m|2n + 1). Let us denote the weights of the
natural spo(2m|2n + 1)-module C2m|2n+1 by ±δi , 0, ±ε j for 1 ≤ i ≤ m, 1 ≤ j ≤ n.
Recall that the standard Borel subalgebra bst of spo(2m|2n + 1) is associated to
the following fundamental system
δn − ε1
⊗
⃝ ⃝ ··· ⃝ ⃝ ··· ⃝=⇒⃝
δ1 − δ2 δ2 − δ3 δn−1 − δn ε1 − ε2 εm−1 − εm εm
Example 2.54. Suppose that the corresponding Dynkin diagram of a Borel subal-
gebra of spo(9|10) is as follows:
δ2 + δ3 −ε3 − ε2 δ1 − δ4 ε4 − ε1
⊗ ⊗ ⊗ ⊗
⃝ ⃝ ⃝ ⃝ =⇒ y
−δ3 + ε3 ε2 − δ1 δ4 − ε4 ε1 + δ5 −δ5
We read off a signed sequence with indices δ2 (−δ1 )(−ε1 )ε2 δ3 δ4 ε4 ε3 (−δ5 ). In par-
ticular, we obtain a sequence δδεεδδεεδ by ignoring the signs and indices. In this
case, d1 = d2 = 2, d3 = 1, and e1 = e2 = 2. Furthermore, the sequences (ξi )1≤i≤5
and (η j )1≤ j≤4 are (1, −1, 1, 1, −1) and (−1, 1, 1, 1), respectively.
Proof. The arguments here are similar to the ones given in the proof of Theorem
2.52 for gl(m|n), and so we will be sketchy.
We first note that the theorem holds for the standard Borel subalgebra bst ,
which corresponds to the sequence of m δ’s followed by n ε’s with all signs ξi
st
and η j being positive, i.e., λb = λ♮ .
All Borels are linked to the standard Borel by a sequence of even and odd
simple reflections. Let b and b′ be two Borel subalgebras be related by a simple
reflection with respect to a simple root α, and assume the theorem holds for b′ . If
′
α is even, then clearly rα (λb ) = λb , and the theorem holds for b. In the case when
α is odd, we have four cases to consider, by setting α = ±δi ± ε j . Each case is
verified by the same type of argument as in the proof of Theorem 2.52.
Example 2.56. With respect to the Borel b of spo(10|9) as in Example 2.54, the
b-extremal weight of L(g, bst , λ♮ ) for λ as given in Example 2.51 equals
λb = p1 δ2 − p2 δ3 + p3 δ1 + p4 δ4 − p5 δ5 − q1 ε3 + q2 ε2 + q3 ε4 + q4 ε1
= 6δ1 + 14δ2 − 11δ3 + 6δ4 − 3δ5 + 2ε1 + 6ε2 − 6ε3 + 3ε4 .
Proof. By a standard fact in highest weight theory, the dual module L(g, bst , λ♮ )∗
has bst -highest weight equal to the opposite of the bst -lowest weight of L(g, bst , λ♮ ).
Denote by bop the opposite Borel to the standard one bst . It follows by Theo-
rem 2.55 that the bop -extremal weight (i.e., the bst -lowest weight) of the module
L(g, bst , λ♮ ) is −λ♮ . This proves the corollary.
⊗
⃝⇐=⃝ ⃝ ··· ⃝ ⃝ ··· ⃝
−ε1 ε1 − ε2 ε2 − ε3 εn − δm δm − δm−1 δ2 − δ1
We have the following immediate corollary of Theorem 2.11 and Theorem 2.55.
2.4.3. Extremal weights for spo(2m|2n). Let n ≥ 2. Let us denote the weights of
the natural spo(2m|2n)-module C2m|2n by ±δi , ±ε j for 1 ≤ i ≤ m, 1 ≤ j ≤ n. The
standard Borel subalgebra bst of spo(2m|2n) is the one associated to the following
Dynkin diagram
⃝ εn−1 − εn
δm − ε1
⊗
⃝ ⃝ ··· ⃝ ⃝ ··· ⃝
δ1 − δ2 δ2 − δ3 δm−1 − δm ε1 − ε2 εn−2 − εn−1@@
⃝ εn−1 + εn
According to Section 1.3.4 there are two types of Dynkin diagrams and correspond-
ing Borel subalgebras for spo(2m|2n):
(i) Diagrams of |-shape, i.e., Dynkin diagrams with a long simple root ±2δi .
(ii) Diagrams of -shape, i.e., Dynkin diagrams with no long simple root.
We will follow the notation for spo(2m|2n + 1) in Section 2.4.2 for fundamen-
tal systems in terms of signed εδ-sequences, so we have permutations s,t, and signs
ξi , η j . We fix the ambiguity on the choice of the last sign ηn associated to a Borel b
of -shape, by demanding the total number of negative signs among η j (1 ≤ j ≤ n)
to be always even.
78 2. Finite-dimensional modules
The weight λb− will only be used for Borel b of -shape. Note that λb = λ♮ and
st
st
we shall denote λ♮− := λb− .
Given a Borel b of |-shape, we define s(b) to be the sign of ∏nj=1 η j .
Recall from Theorem 2.14 that a finite-dimensional irreducible spo(2m|2n)-
module of integer highest weight with respect to the standard Borel subalgebra is
of the form L(g, bst , λ♮ ) or L(λ♮− ), where λ is an (m|n)-hook partition.
Proof. Let b and b′ be Borel subalgebras. Suppose that they are related by b′ =
w(b), for some Weyl group element w. Note that the set of weights for L(g, bst , λ♮ )
′ ′
is W -invariant. By definition we observe that λb = w(λb ) and λb− = w(λb− ), and
hence the validity of the theorem for b implies its validity for b′ . Thus it suffices to
prove the theorem for a Borel subalgebra b with even roots {δi ± δ j |i ≤ j} ∪ {εk ±
εl |k < l}. We shall make this assumption for the remainder of the proof.
Suppose that the Dynkin diagram of b is of -shape. Thus the corresponding
εδ-sequence ends with an ε, i.e., it is of the form
· · · δ · · · ε · · · ε.
εn−1 − εn
⊗
⃝ ··· ⃝ ⃝ ··· ⃝⇐=⃝
ε1 − ε2 εn − δ1 δ1 − δ2 δm−1 − δm 2δm
εn−1 + εn
⊗
⃝ ··· ⃝ ⃝ ··· ⃝⇐=⃝
ε1 − ε2 −εn − δ1 δ1 − δ2 δm−1 − δm 2δm
Proof. The argument here is similar to the proof of Corollary 2.57. For n = 0, the
corollary follows by the well-known fact that the longest element in the Weyl group
for sp(2m) is −1. Now assume n ≥ 2 is even. Denote by bop the opposite Borel
to the standard one bst . It follows by Theorem 2.59 that the bop -extremal weight
of the module L(g, bst , λ♮ ) (and respectively, L(g, bst , λ♮− )) is −λ♮ (and respectively,
−λ♮− ). The corollary follows.
Remark 2.61. The remaining b-extremal weights for the modules L(g, bst , λ♮ )
when s(b) = − or for the modules L(g, bst , λ♮− ) when s(b) = + do not seem to
afford a uniform simple answer.
ε1 − ε2 ⃝
@ εn − δm
⊗
@⃝ ··· ⃝ ⃝ ··· ⃝
ε2 − ε3 εn−1 − εn δm − δm−1 δ2 − δ1
−ε1 − ε2 ⃝
80 2. Finite-dimensional modules
2.5. Notes
Section 2.1. A systematic study of the representation theory of finite-dimensional
(simple) Lie superalgebras was initiated by Kac [K2]. The classification of finite-
dimensional irreducible representations of the Lie superalgebras of types gl and
spo (Propositions 2.2 and 2.2; Theorems 2.11 and 2.14) appeared first in Kac [K1],
where the finite dimensionality condition was formulated in terms of Dynkin labels
instead of hook partitions. The proof using odd reflections in the most challenging
spo type given here follows Shu-Wang [SW] for the cases of integer weights, and
the proof using odd reflections and the Bernstein functor (see Santos [San]) for
the cases of half-integer weights here is new. Yet another closely related approach
to the classification is provided in Azam-Yamane-Yousofzadeh [AYY] by means
of Weyl groupoids. Theorem 2.16 on the classification of finite-dimensional irre-
ducible modules of q(n) was formulated by Penkov [Pe, Theorem 4], who proved
the “only if” direction. We are not aware of a written proof for the “if” direction in
the literature. Our proof uses the Bernstein functor again.
Section 2.2. The Harish-Chandra homomorphisms for Lie superalgebras was
first studied by Kac [K2], who formulated results in terms of fractional fields. The
analogue of Chevalley theorem on invariant polynomials for simple Lie superalge-
bras has been established by Sergeev [Sv3], which are presented here as a part of
Theorem 2.24 for type gl and osp. Our approach of calculating the images of the
Harish-Chandra homomorphisms in this book is different and new, and it uses the
connection to the theory of supersymmetric functions for types gl and spo and the
theory of Schur Q-functions for type q. There is another approach developed by
Gorelik [Go] following Kac [K5] on the centers of the universal enveloping alge-
bras. A variant of Theorem 2.27 on linkage principle for gl and osp was originally
stated as a conjecture in Kac [K2, §6].
The notion of typicality for basic Lie superalgebras is due to Kac [K2]. Corol-
lary 2.26 (which is due to Kac [K2, Theorem 2] by different arguments) was used
by Kac to derive the irreducible character formula for finite-dimensional typical
modules of basic classical Lie superalgebas (including Theorem 2.32).
2.5. Notes 81
Section 2.3. This section is the counterpart of Section 2.2 for the queer Lie
superalgebra q(n). Theorem 2.43 on the image of the Harish-Chandra homomor-
phism for the queer Lie superalgebra was formulated without proof in a short an-
nouncement of Sergeev [Sv4]. We fill in the details of a proof by presenting several
explicit formulas of singular vectors in Verma modules. The notion of typicality
for q(n) was formulated by Penkov [Pe]. Theorem 2.49 on the typical character
formula for q(n) is due to Penkov [Pe, Theorem 2]. Theorem 2.45 on linkage
principle for q(n) or its variant has been regarded as folklore after [Sv4].
Section 2.4. As we shall see in Chapter 3, the irreducible polynomial repre-
sentations of gl(m|n) were classified by Sergeev [Sv1] (also see Berele and Regev
[BeR]). The results on extremal weights of the irreducible polynomial representa-
tions for gl and finite-dimensional irreducible integer weight modules for spo were
due to Ngau Lam and the authors (for the spo case see [CLW] and for the gl case
see [CW5]). The special case for type gl when a fundamental system consists of
only odd simple roots goes back to [CW3].
The irreducible character problem for finite-dimensional modules over Lie su-
peralgebras have been a challenging one. For gl(m|n), there has been complete
solutions in completely different approaches due to Serganova [Sva], Brundan
[Br1], Brundan-Stroppel [BrS], and Cheng-Lam [CL2] (also see Cheng-Wang-
Zhang [CWZ]). There is also a combinatorial dimension formula by Su-Zhang
[SZ]. For q(n), there has been complete solutions due to Penkov-Serganova [PS2]
and Brundan [Br2]. There were numerous partial results in the literature over the
years, see for example [BL, CWZ2, J, JHKT, PS2, San, Sv1, Zou]. Chapter 6
of this book offers a complete solution for a wider class of (including all finite-
dimensional) modules of gl(m|n) and osp(m|2n).
Chapter 3
Schur duality
Schur duality, which was popularized in Weyl’s book The Classical Groups, is
an interplay between representations of the general linear Lie group/algebra and
representations of the symmetric group. On the combinatorial level, it explains
their mutual connections to Schur functions and Young tableaux.
In this chapter, we describe Sergeev’s superalgebra generalization of Schur du-
ality. The first generalization is a duality between the general linear Lie superalge-
bra and the symmetric group. This allows us to classify the irreducible polynomial
representations of the general linear Lie superalgebra, and obtain their character
formula in terms of supertableaux and also in terms of super Schur functions. The
second generalization is a duality between the queer Lie superalgebra and a twisted
(or spin) group algebra Hd of the hyperoctahedral group. The representation theory
of the algebra Hd is systematically developed, and its connection with symmetric
functions such as Schur Q-functions is explained in detail. The irreducible polyno-
mial representations of the queer Lie superalgebra are classified in terms of strict
partitions, and their characters are shown to be the Schur Q-functions up to some
2-powers.
83
84 3. Schur duality
We have A21̄ = A0̄ , for, otherwise A21̄ + A1̄ would be a proper ideal of A, which
contradicts with the simplicity of A. Hence, we have
(3.3) J ∩ A1̄ = 0, p1̄ (J) = A1̄ ,
by the following computations (which use (3.2) and that 1 ∈ A0̄ ):
J ∩ A1̄ = A0̄ (J ∩ A1̄ ) = A21̄ (J ∩ A1̄ ) ⊆ A1̄ (J ∩ A0̄ ) = 0,
p1̄ (J) ⊇ A1̄ p0̄ (J) = A1̄ A0̄ = A1̄ .
Now the lemma follows from (3.2) and (3.3).
Proof of Theorem 3.1. We claim that Z(|A|)0̄ = C. To see this, let z ∈ Z(|A|)0̄ .
The map ℓz : A → A defined by left multiplication by z has an eigenvalue c ∈ C, and
the kernel of ℓz − c is a nonzero 2-sided ideal of A. The simplicity of A implies
that ℓz − c = 0, and hence z = c · 1.
(1) Assume Z(|A|)1 = 0. We claim that |A| is a simple algebra (and so |A|
is a matrix algebra by a classical structure theorem of simple algebras). Indeed,
assume |A| is a non-simple algebra with a proper ideal J. Consider w = p−10̄
(1) ∈ J
and u = p1̄ (w) ∈ A1̄ , where u ̸= 0 by Lemma 3.2. Then, w = 1 + u. For any
homogeneous element x ∈ Ai for some i ∈ Z2 , we have xw = x+xu and wx = x+ux,
and so pi (xw) = x = pi (wx). By Lemma 3.2, xw = wx and so xu = ux, that is,
u ∈ Z(|A|)1 . It follows by assumption that u = 0, which is a contradiction.
Consider the automorphism σ : |A| → |A| given by σ(ai ) = (−1)i ai for ai ∈ Ai
and i ∈ Z2 . There exists u ∈ |A| such that σ(a) = uau−1 for all a ∈ |A|, since every
automorphism of a matrix algebra is inner. We have u ∈ A0̄ as σ(u) = u. It follows
from σ2 = 1 that u2 ∈ Z(|A|)0̄ = C. We can choose u such that u2 = 1, and then by
a change of basis, take u to be a diagonal matrix with m 1’s followed by n (−1)’s
in the main diagonal. Hence, we conclude that A ∼ = M(m|n).
(2) For a nonzero element w ∈ Z(|A|)1 , we have w2 ∈ Z(|A|)0 = C. We must
have w2 ̸= 0, for, otherwise the annihilator of w is a proper ideal of the super-
algebra A, contrary to the simplicity of A. Then we may assume w2 = 1. We
86 3. Schur duality
3.1.2. Wedderburn theorem and Schur’s Lemma. The basic results of finite-
dimensional semisimple (unital associative) algebras over C admit natural super
generalizations.
A module M of a superalgebra A is called semisimple, if every A-submodule
of M is a direct summand of M. The following are straightforward generalizations
of the classical structure results on semisimple modules and algebras, and they can
be proved in the same way as in the non-super setting.
Theorem 3.3. (Super Wedderburn Theorem) The following statements are equiva-
lent for a superalgebra A:
(i) Every A-module is completely reducible.
(ii) The left regular module A is a direct sum of minimal left ideals.
(iii) A is a direct sum of simple superalgebras.
where m = m(A) and q = q(A) are invariants of A. Observe that Cr|s is a simple
module (unique up to isomorphism) of the simple superalgebra M(r|s), and Cn|n is
a simple module (unique up to isomorphism) of the simple superalgebra Q(n). A
simple A-module V is annihilated by all but one such summand in (3.4). We say V
is of type M if this summand is of the form M(ri |si ) and of type Q if this summand is
of the form Q(n j ). These two types of simple modules are distinguished by a super
analogue of Schur’s Lemma.
Lemma 3.4. (Super Schur’s Lemma) If M and L are simple modules over a super-
algebra A, then
1 if M ∼= L is of type M,
dim HomA (M, L) = 2 if M ∼= L is of type Q,
0 if M ̸∼
= L.
3.1. Generalities for associative superalgebras 87
W∼
= ∑ 2−δi Ui ⊗Vi ,
i
where δi ∈ {0, 1}, {Ui } are pairwise non-isomorphic simple A-modules, {Vi } are
pairwise non-isomorphic simple B-modules. Moreover, Ui and Vi are of the same
type, and they are of type M if and only if δi = 0.
Proof. Assume that Va are all the pairwise non-isomorphic simple B-modules of
type M, and Vb are all the pairwise non-isomorphic simple B-modules of type Q.
Then the hom spaces Ua := HomB (Va ,W ) and Ub := HomB (Vb ,W ) are naturally
A-modules. By the super Schur’s Lemma, we know dimUb = kb |kb for some kb .
By the semisimplicity assumption on B, we have an isomorphism of B-modules
⊕ ⊕
W∼
= Ua ⊗Va ⊕ 2−1Ub ⊗Vb .
a b
88 3. Schur duality
By applying the super Schur’s Lemma, we obtain the following superalgebra iso-
morphisms:
⊕ ⊕
A = EndB (W ) ∼
= EndB (Ua ⊗Va ) ⊕ EndB (2−1Ub ⊗Vb )
a b
⊕ ⊕
∼
= End(Ua ) ⊗ IVa ⊕ (End(2−1Ub ) ⊗ Q(1)) ⊗ IVb
a b
⊕ ⊕
∼
= End(Ua ) ⊗ IVa ⊕ Q(Ub ) ⊗ IVb .
a b
3.1.4. Split conjugacy classes in a finite supergroup. For a finite group G, the
group algebra CG is a semisimple algebra. It follows by using the Wedderburn
Theorem and comparing two different bases for the center of CG that the number
of simple G-modules coincides with the number of conjugacy classes of G.
Now assume that a finite group G contains a subgroup G0 of index 2. We call
elements in G0 even and elements in G\G0 odd. Then the group algebra of G is
naturally a superalgebra; we shall denote it by C[G, G0 ] = CG0 ⊕ C[G\G0 ] to make
clear its Z2 -grading and its dependence on G0 . Just as for the usual group algebras,
it is standard to show that C[G, G0 ] is a semisimple superalgebra.
Since elements in a given conjugacy class of G share the same parity (i.e.,
Z2 -grading), it makes sense to talk about even and odd conjugacy classes.
Proposition 3.6. (1) The number of simple C[G, G0 ]-modules coincides with
the number of even conjugacy classes of G.
(2) The number of simple C[G, G0 ]-modules of type Q coincides with the num-
ber of odd conjugacy classes of G.
Proposition 3.7. eG
(1) The number of simple spin C[G, e0 ]-modules equals the
number of even split conjugacy classes of G.
(2) The number of simple spin C[G, eG e0 ]-modules of type Q equals the number
of odd split conjugacy classes of G.
3.2.1. Schur-Sergeev duality, I. Let g = gl(m|n). Let {ei |i ∈ I(m|n)} be the stan-
dard basis for the natural g-module V = Cm|n . Then V ⊗d is naturally a g-module
by letting
Lemma 3.8. The formula (3.7) defines a left action of the symmetric group Sd on
V ⊗d . The actions of (gl(m|n), Φd ) and (Sd , Ψd ) on V ⊗d commute with each other.
Proof. It is straightforward to check that the formula (3.7) satisfies the Coxeter
relations for Sd , and hence it defines a left action of Sd .
3.2. Schur-Sergeev duality of type A 91
The verification of the commuting action boils down to the case below when
d = 2 and i = 1 (where we denote s = (1, 2) ∈ S2 , g ∈ gl(m|n), and v1 , v2 ∈ V ):
Ψd (s)Φd (g)(v1 ⊗ v2 )
= Ψd (s)(g.v1 ⊗ v2 + (−1)|g|·|v1 | v1 ⊗ g.v2 )
= (−1)(|g|+|v1 |)|v2 | v2 ⊗ g.v1 + (−1)|g|·|v1 |+|v1 |(|g|+|v2 |) g.v2 ⊗ v1 ,
Φd (g)Ψd (s)(v1 ⊗ v2 ) = (−1)|v1 |·|v2 | (g.v2 ⊗ v1 + (−1)|g|·|v2 | v2 ⊗ g.v1 ).
Theorem 3.9 (Schur-Sergeev duality, Part I). Let g = gl(m|n). The images of Φd
and Ψd , Φd (U(g)) and Ψd (CSd ), satisfy the double centralizer property, i.e.,
Φd (U(g)) =EndSd (V ⊗d ),
EndU(g) (V ⊗d ) = Ψd (CSd ).
Assume that Yk−1 ⊆ Xk−1 . Note that ϖ(x1 , . . . , xk−1 ) · x̃k ∈ Xk . On the other
hand, we have
ϖ(x1 , . . . , xk−1 ) · x̃k
= ∑ σ.(x1 ⊗ . . . ⊗ xk−1 ⊗ 1d−k+1 ) · x̃k
σ∈Sd
d ( )
= ∑ ∑ σ. (x1 ⊗ . . . ⊗ xk−1 ⊗ 1d−k+1 ) · (1 j−1 ⊗ xk ⊗ 1d− j ) ,
j=1 σ∈Sd
and
d
A2 = ∑ ∑ σ.(x1 ⊗ . . . ⊗ xk−1 ⊗ 1 j−k ⊗ xk ⊗ 1d− j )
j=k σ∈Sd
= (d − k + 1)ϖ(x1 , . . . , xk−1 , xk ).
3.2.2. Schur-Sergeev duality, II. Recall from Definition 2.10 that an (m|n)-hook
partition µ = (µ1 , µ2 , . . .) is a partition with λm+1 ≤ n. Given an (m|n)-hook parti-
tion µ, we denote by µ+ = (µm+1 , µm+2 , . . .) and write its conjugate partition, which
is necessarily of length ≤ n, as ν = (µ+ )′ = (ν1 , . . . , νn ). We recall the weight
defined in (2.4)
µ♮ = µ1 δ1 + . . . + µm δm + ν1 ε1 + . . . + νn εn .
Denote by Pd (m|n) the set of all (m|n)-hook partitions of size d and let
P(m|n) = ∪d≥0 Pd (m|n) = {λ | λ are partitions with λm+1 ≤ n}.
Then, Pd (m) = P(d, m|0) is the set of partitions of d of length ≤ m. Accordingly,
we let
Pd = {λ | λ are partitions of d}.
We denote by L(λ♮ ) for λ ∈ P(m|n) the simple g-module of highest weight λ♮ with
respect to the standard Borel subalgebra. For a partition λ of d, we denote by
Sλ the Specht module of Sd . For example, S(d) is the trivial representation, and
d
S(1 ) = sgnd is the sign representation.
3.2. Schur-Sergeev duality of type A 93
We have the following weight space decomposition with respect to the Cartan sub-
algbra of diagonal matrices h ⊂ gl(m|n):
⊕
(3.8) W= Wν|µ ,
ν|µ∈CPd (m|n)
where Wν|µ has a linear basis ei1 ⊗ . . . ⊗ eid , with the indices satisfying the following
equality of sets:
{i1 , . . . , id } = {1, . . . , 1, . . . , m, . . . , m; 1, . . . , 1, . . . , n, . . . , n}.
| {z } | {z } | {z } | {z }
ν1 νm µ1 µn
Let
Sν|µ := Sν1 × . . . × Sνm × Sµ1 × . . . × Sµn .
⊗µ ⊗µ
The span of the vector eν|µ := e⊗ν
1
1
⊗ . . . ⊗ e⊗ν
m ⊗ e1 ⊗ . . . ⊗ en
m 1 n
can be identified
with the Sν|µ -module 1ν ⊗ sgnµ . Since the orbit Sd .eν|µ spans Wν|µ we have a
surjective Sd -homomorphism from IndS Sν|µ (1ν ⊗ sgnµ ) onto Wν|µ by the Frobenius
d
Let λ be a partition which is identified with its Young diagram. Recall I(m|n)
is totally ordered by (1.3). A supertableau T of shape λ, or a super λ-tableau
T , is an assignment of an element in I(m|n) to each box of the Young diagram λ
satisfying the following conditions:
(HT1) The numbers are weakly increasing along each row and column.
(HT2) The numbers from {1, . . . , m} are strictly increasing along each column.
(HT3) The numbers from {1, . . . , n} are strictly increasing along each row.
Such a T is said to have content ν|µ ∈ CP(m|n) if ī (1 ≤ i ≤ m) appears νi times and
j (1 ≤ j ≤ n) appears µ j times. Denote by HT(λ, ν|µ) the set of super λ-tableaux
of content ν|µ. In the case n = 0, a supertableau becomes a usual (semistandard)
tableau.
Lemma 3.11. We have Kλ,ν|µ = #HT(λ, ν|µ).
where λ is such that λ/ρ is a horizontal strip of νr boxes (that is, λ/ρ is a skew
diagram of size νr whose columns all have length at most one).
Then using the above special case (for µ = 0) / as the initial step, we complete
the proof in the general case by induction on the length of µ, in which the induction
step is exactly the conjugated Pieri’s formula.
Lemma 3.12. Let λ ∈ Pd and ν|µ ∈ CPd (m|n). Then Kλ,ν|µ = 0 unless λ ∈ Pd (m|n).
Proof. By the identity Kλ,ν|µ = #HT(λ, ν|µ), it suffices to prove that if a super
λ-tableau T of content ν|µ exists, then λm+1 ≤ n.
3.2. Schur-Sergeev duality of type A 95
3.2.3. The character formula. We shall compute the character of the gl(m|n)-
module L(λ♮ )
E
Em̄m̄ yE11 . . . yEnn and
Let us apply simultaneously the trace operator tr |L(λ♮ ) x1 1̄1̄ . . . xm 1 n
the Frobenius characteristic map chF (see Appendix A.1.4) to both sides of the
above isomorphism. Then, summing over all d ≥ 0 and using (3.9) and (A.24), we
have that
Now the theorem follows by comparing the above two equations and noting the
linear independence of sλ (z).
where the summation is taken over all super λ-tableaux T . Also, we have
(3.11) chL(λ♮ ) = ∑ #HT(λ, ν|µ) mν (x)mµ (y),
ν,µ
Proof. The formula (3.10) is simply a reformulation of Theorem 3.14 by the defi-
nition of super Schur functions in (A.35) and the combinatorial formula for (skew)
Schur functions (A.18).
3.2. Schur-Sergeev duality of type A 97
Let us collect the same monomials together in the sum (3.10). Regarding L(λ♮ )
as a module over gl(m) × gl(n), we observe that chL(λ♮ ) is symmetric with respect
to x1 , . . . , xm and symmetric with respect to y1 , . . . , yn . Hence, we must have
#HT(λ, e
ν|e
µ) = #HT(λ, ν|µ)
if e
ν and e
µ are compositions obtained by rearranging the parts from ν and µ, respec-
tively. Now (3.11) follows from this observation and (3.10).
Corollary 3.16. The following weight multiplicity formula for the module L(λ♮ )
holds:
dim L(λ♮ )ν|µ = #HT(λ, ν|µ).
Example 3.17. The standard basis vectors for S2 (Cm|n ) are ei ⊗ e j + (−1)|i|| j| e j ⊗
ei , where i, j ∈ I(m|n) satisfy i ≤ j < 0, i < 0 < j, or 0 < i < j (cf. Remark 3.13 (4)).
They are in bijection with the super tableaux of shape λ = (2):
i j
3.2.4. The classical Schur duality. We sketch below a more standard argument
for the standard Schur duality on (Cn )⊗d , which emphasizes the decomposition of
W as a gl(n)-module.
Given a composition (or a partition) µ of d of length ≤ n, we denote by Wµ the
µ-weight space of the gl(n)-module. Clearly Wµ has a basis
(3.12) ei1 ⊗ . . . ⊗ eid , where {i1 , . . . , id } = {1, . . . , 1, . . . , n, . . . , n}.
| {z } | {z }
µ1 µn
As a (gl(n), Sd )-module,
⊕
W∼
= L(λ) ⊗U λ ,
λ
On the other hand, Sd acts on the basis (3.12) of Wµ transitively, and the stabilizer
⊗µ ⊗µ
of the basis element e1 1 ⊗ . . . ⊗ en n is the Young subgroup Sµ . Therefore we
have
⊕
(3.14) Wµ ∼ S
= Ind d 1µ =
Sµ Kλµ Sλ ,
λ≥µ
where Kλµ is the Kostka number which satisfies Kλλ = 1. It is well known that Kλµ
is equal to the number of semistandard λ-tableaux of content µ.
By the double centralizer property (see Proposition 3.5 and Theorem 3.9), U λ
has to be an irreducible Sd -module for each λ. We compare the interpretations
(3.13) and (3.14) of Wµ in the special case when µ is dominant (i.e., a partition).
One by one downward along the dominance order starting with µ = (d), this pro-
vides the identification U µ = Sµ for every µ, and moreover, we obtain the well-
known equality dim L(λ)µ = Kλµ .
m − iλ
@
@
m @
@
Proof. The lemma is clearly equivalent to the claim that λm−iλ +1 ≤ n − iλ ≤ λm−iλ .
The latter is evident from the diagram in Example 3.18, as iλ is equal to the number
3.2. Schur-Sergeev duality of type A 99
We observe that the sequence {λ1 +m, λ2 +m−1, . . . , λm +1} is strictly decreasing,
while the sequence {1 − ν1 , 2 − ν2 , . . . , n − νn } is strictly increasing.
Suppose that iλ = 0. This is equivalent to saying that λm ≥ n, by Lemma 3.19.
Thus,
λi + m − i + 1 ≥ λm + 1 > j − ν j , ∀1 ≤ j ≤ n, 1 ≤ i ≤ m.
It follows by Definition 2.28 that #λ♮ = 0.
Now suppose that iλ > 0. Then, λm = n − j0 < n for some 0 < j0 ≤ n. This
implies that νn = · · · = νn− j0 +1 = 0, and thus
{n − j0 + 1 − νn− j0 +1 , . . . , n − νn } = {n − j0 + 1, . . . , n}.
It follows by Lemma 3.19 that, for 0 ≤ j ≤ iλ − 1,
n − j0 + 1 = λm + 1 ≤ λm− j + j + 1 ≤ n − iλ + j + 1 ≤ n.
Thus, in the set {n − j0 + 1 − νn− j0 +1 , . . . , n − νn } = {n − j0 + 1, . . . , n}, there is a
unique element that is equal to λm− j + j + 1, for 0 ≤ j ≤ iλ − 1. Hence, #λ♮ ≥ iλ .
Finally, for i = 1, . . . , m − iλ , Lemma 3.19 implies that
λi + m − i + 1 ≥ λm−iλ + iλ + 1 ≥ (n − iλ ) + iλ + 1 = n + 1.
Thus, any such λi + m − i + 1, for i = 1, . . . , m − iλ , cannot be equal to an element
of the form j − ν j , for j = 1, . . . , n, whence #λ♮ = iλ .
The following corollary can be read off from the proof of Proposition 3.20.
Proof. (1) Suppose that χλ♮ = χµ♮ . Then by Theorem 2.27 we have
Hλ♮ \ Iλ♮ = Hµ♮ \ Iµ♮ , Tλ♮ \ Iλ♮ = Tµ♮ \ Iµ♮ .
Thus, it suffices to show that we can reconstruct λ♮ from the sets Hλ♮ \ Iλ♮ and
Tλ♮ \ Iλ♮ . To that end, note that λ♮ + ρ and λ♮ can be recovered from the sets Hλ♮ and
Tλ♮ , which in turn are determined by the three sets Iλ♮ , Hλ♮ \ Iλ♮ and Tλ♮ \ Iλ♮ . But by
Corollary 3.21 the set Iλ♮ is determined from the set Tλ♮ \ Iλ♮ . This proves (1).
(2) Consider a short exact sequence of the form (3.16) with M = L(gl(m|n), λ♮ )
and N = L(gl(m|n), µ♮ ). First assume λ ̸= µ. It follows by (1) that there exists a
central element z such that χλ♮ (z) ̸= χµ♮ (z). Then, ker(z − χλ♮ (z)) is a nonzero
proper submodule of E, which must be isomorphic to M as it cannot be isomorphic
to N. Hence the short exact sequence splits. Now assume λ = µ. Then by weight
consideration, the two-dimensional µ♮ -weight subspace of E has to be a highest
weight space. Thus, E contains two highest weight submodules of highest weight
µ♮ that must intersect trivially. Hence the short exact sequence splits in this case as
well.
Definition 3.23. A weight µ = ∑m i=1 ai δi + ∑ j=1 b j ε j is called a polynomial weight
n
Since (Cm|n )⊗d+l is a direct sum of irreducible polynomial modules, so is the sub-
module L(gl(m|n), λ♮ ) ⊗ L(gl(m|n), µ♮ ).
Since Πn /{1, z} ≃ Zn2 , the group Ben is a double cover of the hyperoctahedral
group Bn := Zn2 o Sn , and the order |Ben | is 2n+1 n!. That is, we have a short exact
sequence of groups
θn
1 −→ {1, z} −→ Ben −→ Bn −→ 1,
Lemma 3.26. Two elements of Bn are conjugate if and only if their types are the
same.
3.3. Representation theory of the algebra Hn 103
We shall denote by Cρ+ ,ρ− the conjugacy class of type (ρ+ , ρ− ). Note that if
(bI , σ) ∈ Cρ+ ,ρ− , then Cρ+ ,ρ− is even (respectively, odd) if |I| is even (respectively,
odd).
Example 3.27. Let τ = (1, 2, 3, 4)(5, 6, 7)(8, 9), σ = (1, 3, 8, 6)(2, 7, 9)(4, 5) ∈ S10 .
It is straightforward to check that both x = ((+, +, +, −, +, +, +, −, +, −), τ) and
y = ((+, −, −, −, +, −, −, −, +, −), σ) in B10 have the same type
(ρ+ , ρ− ) = ((3), (4, 2, 1)).
Thus, x is conjugate to y in B10 .
Let
P+
n = {λ ∈ Pn | ℓ(λ) is even},
P−
n = {λ ∈ Pn | ℓ(λ) is odd}.
Given σ ∈ Sn of cycle type µ, we denote by
d(σ) = n − ℓ(µ).
Lemma 3.28. Let aI s = (aI1 s1 ) . . . (aIq sq ) be an element of Ben in its cycle decom-
position. Let J ⊆ supp(s1 ) ∩ I1c . Then
(aJ s1 )(aI s)(aJ s1 )−1 = zd(s1 )+|J||I| aI s.
104 3. Schur duality
Proof. Observe that a2I = z(|I|−1)|I|/2 for any subset I. For k > 1 we have
Proof. Assume that Cρ+ ,ρ− is an even conjugacy class such that ρ+ ̸∈ OP. Then
θ−1
n (Cρ+ ,ρ− ) contains an element aI s with a signed cycle decomposition of the form
aI s = s1 (aI2 s2 ) . . . (aIp s p ),
where s1 = (1, 2, . . . , r) for r = 2k even, I1 = 0/ and |I| is even. Consider the element
x = a12...r (1, 2, . . . , r) ∈ Ben . By Lemma 3.28 we have
x(aI s)x−1 = z(2k−1)+2k·|I| aI s = zaI s.
Therefore, if an even conjugacy class Cρ+ ,ρ− splits then ρ+ ∈ OP.
Assume that Cρ+ ,ρ− is an even conjugacy class such that ρ− ̸= 0. / Then ρ−
−1
contains at least two parts, and θn (Cρ+ ,ρ− ) contains an element of the form
aI s = (ai1 s1 )(ai2 s2 )(aI3 s3 ) . . . (aIp s p ),
where i1 ∈ supp(s1 ), i2 ∈ supp(s2 ). If both s1 and s2 are of cycle length 1, then
(i1 , i2 )(aI s)(i1 , i2 )−1 = zaI s. Assume that the order ord(s1 ) is ≥ 2, so s−1
1 (i1 ) =
′
i1 ̸= i1 . Then
(ai1 s1 )−1 aI s(ai1 s1 )
= (ai1 s1 )−1 (ai1 s1 ai2 s2 )(ai1 s1 )(aI3 s3 ) . . . (aIp s p )
= ai2 s2 (ai1 s1 )(aI3 s3 ) . . . (aIp s p )
= z(ai1 s1 )(ai2 s2 )(aI3 s3 ) . . . (aIp s p )
= zaI s.
Hence, if an even conjugacy class Cρ+ ,ρ− splits then ρ− = 0.
/ Together with the
above, we have shown that an even split conjugacy class should satisfy (1).
3.3. Representation theory of the algebra Hn 105
/ Then,
Now assume that Cρ+ ,ρ− is an odd conjugacy class such that ρ+ ̸= 0.
θ−1
n (Cρ+ ,ρ− )
contains an element aI s with the signed cycle decomposition of the
form
aI s = (s1 )(aI2 s2 ) . . . (aIq sq ),
where I1 = 0/ and |I| is odd. Let J = supp(s1 ). Then,
(aJ s1 )(aI s)(aJ s1 )−1 = z(|J|−1)+|I||J| aI s = zaI s,
/
since |I| is odd. Hence, if Cρ+ ,ρ− is an odd split conjugacy class then ρ+ = 0.
Next, assume that Cρ+ ,ρ− is an odd conjugacy class such that ρ− contains two
identical parts. Then θ−1
n (Cρ+ ,ρ− ) contains an element of the form
which must equal ai1 s1 up to a power of z. Set (aJ1 t1 )(ai1 s1 )(aJ1 t1 )−1 = z∗ ai1 s1
where ∗ is 0 or 1. We claim that ∗ is always 0. Note that aJ1 at1 (i1 ) = z∗ ai1 as1 (J1 ) ,
and so J1 differs from s1 (J1 ) by one element. Without loss of generality, we let
i1 = 1, s1 = (1, 2, . . . , k), J1 = {1, 2, . . . , r} with 0 ≤ r < k, then we have
a1 . . . ar · ar1 +1 = aJ1 at1 (i1 ) = z∗ ai1 as1 (J1 ) = z∗ a1 · a2 . . . ar+1 ,
106 3. Schur duality
which implies that r1 = r and ∗ = 0. Therefore, (aJ t)(ai1 s1 )(aJ t)−1 = z|J|−|J1 | ai1 s1 ,
and similarly we have
(aJ t)(aI s)(aJ t)−1 = (aJ t)(aI1 s1 ) . . . (aIq sq )(aJ t)−1
= zq|J|−(|J1 |+...+|Jq |) aI s = z(q−1)|J| aI s = aI s,
since q is odd. This is a contradiction.
has the advantage that one can apply the standard arguments from the theory of
finite groups directly.
Denote by R−n the Grothendieck group of Hn -mod. As in the usual (ungraded)
case, we may replace the isomorphism classes of modules by their characters, and
then regard R−n as the free abelian group with a basis consisting of the characters
of the simple spin Ben -modules. Let
∞
⊕ ∞
⊕
R− := R−
n, R−
Q := Q ⊗Z R−
n,
n=0 n=0
follows.
Denote by Bem,n the subgroup of Bem+n generated by Sm × Sn and Πm+n . Then
Bm,n can be identified with the quotient group Bem × Ben /{(1, 1), (z, z)}, where Bem ×
e
Ben denotes the product group in the super sense, i.e., elements from Bem and Ben
supercommute with each other.
Given φ ∈ R− − e
m and ψ ∈ Rn , we define a spin character φ×ψ of Bm,n by letting
ˆ
φ×ψ(x,
ˆ y) = φ(x)ψ(y),
where (x, y) is the image of (x, y) in Bem,n ∼
= (Bem × Ben )/{(1, 1), (z, z)}. We define a
−
product on R by
Be
φ · ψ = IndBem+n (φ×ψ),
ˆ
m,n
where φ ∈ R−
m, ψ∈ R−
nfor all m, n, and Ind denotes the induced character. It
follows from the properties of the induced characters that the multiplication on R−
is commutative and associative.
Remark 3.31. Equivalently, the multiplication in R− can be described as follows.
Let Hm,n be the subalgebra of Hm+n generated by Cm+n and Sm × Sn . For M ∈
Hm -mod and N ∈ Hn -mod, M ⊗ N is naturally an Hm,n -module, and we define the
product
[M] · [N] = [Hm+n ⊗Hm,n (M ⊗ N)],
and then extend it by Z-bilinearity.
For φ ∈ R−n , we shall write φα = φ(x) for x ∈ Cα , α ∈ OPn ; hence φ(y) = −φα
+
−
for y ∈ Cα . Given two partitions α, β, we let α ∪ β denote the partition obtained
by collecting the parts of α and β together and rearranging them in a descending
order.
Lemma 3.32. Let φ ∈ R− −
m , ψ ∈ Rn , and γ ∈ OPm+n . Then
zγ
(φ · ψ)γ = ∑ z z
φα ψ β .
α,β∈OP,α∪β=γ α β
108 3. Schur duality
It follows from this, Lemma 3.30, and the standard induced character formula for
finite groups that
|Bem+n |
(φ · ψ)γ =
|Bem,n | · |C+
∑ (φ×ψ)(w)
ˆ
γ | w∈C+
γ
21+ℓ(γ) zγ
= ∑ φα ψβ |C+α | · |C+β |
2m+n+1 m!n! α,β∈OP,α∪β=γ
zγ
= ∑ 2m−ℓ(α) z−1
m!n!2m+n−ℓ(γ) α,β∈OP,α∪β=γ α m!2
n−ℓ(β) −1
zβ n!φα ψβ
zγ
= ∑ z z
φα ψβ ,
α,β∈OP,α∪β=γ α β
3.3.4. The characteristic map. Recall from (A.43) in Appendix A that the ring
ΓQ := Q ⊗Z Γ has a basis given by the power-sum symmetric functions pµ for
µ ∈ OP. Moreover, ΓQ is equipped with a bilinear form ⟨·, ·⟩ given in (A.52).
We define the characteristic map
ch : R−
Q −→ ΓQ
to be the linear map given by
ch(φ) = ∑ z−1
α φα p α , φ ∈ R−
n.
α∈OPn
Proof. For φ ∈ R− −
m , ψ ∈ Rn , we have by Lemma 3.32 that
ch(φ · ψ) = ∑ z−1
γ (φ · ψ)γ pγ
γ∈OP
zγ
= ∑ ∑ z−1
γ
zα zβ
φα ψβ pγ = ch(φ)ch(ψ),
γ α,β∈OP,α∪β=γ
Denote by
⟨[M], [N]⟩ = dim HomBen (M, N)
for spin Ben -modules M, N. This defines a bilinear form ⟨·, ·⟩ on R−
n by Z-bilinearity.
The HomBen here can be either understood as in the category of Ben -modules (with
3.3. Representation theory of the algebra Hn 109
degree one morphism allowed) or in the category of (non-graded) Ben -modules, and
they give the same dimension. In light of super Schur’s Lemma 3.4, this is reduced
to a straightforward verification in the case when M ∼
= N is simple of type Q.
Then R− , and hence also R−
Q , carry a symmetric bilinear form, still denoted by
⟨·, ·⟩, which is induced from the ones on R− − −
n , for all n, such that Rn and Rm are
orthogonal whenever n ̸= m.
Lemma 3.34. For φ, ψ ∈ R−
n , we have
1
= ∑
2n+1 n! α∈OP α |φα ψα
2|C+
n
1
= ∑ 2n−ℓ(α) z−1
2n n! α∈OP α n!φα ψα
n
= ∑ 2−ℓ(α) z−1
α φα ψα .
α∈OPn
Proposition 3.35. The characteristic map ch : R− Q → ΓQ is an isometry, i.e., it
−
preserves the bilinear forms ⟨·, ·⟩ on RQ and ΓQ .
Proof. This follows from a direct computation using Lemma 3.34: for φ, ψ ∈ R−
n
we have
⟨ch(φ), ch(ψ)⟩ = ∑ z−1 −1
α zβ ⟨pα , pβ ⟩φα φβ
α,β∈OPn
= ∑ z−2
α 2
−ℓ(α)
zα φα ψα = ⟨φ, ψ⟩.
α∈OPn
3.3.5. The basic spin module. The algebra Hn acts on the Clifford algebra Cn by
the formula
ci .(ci1 ci2 . . .) = ci ci1 ci2 . . . , σ.(ci1 ci2 . . .) = cσ(i1 ) cσ(i2 ) . . . ,
for σ ∈ Sn . The Hn -module Cn is called the basic spin module. Let σ = σ1 . . . σℓ ∈
Sn be a product of disjoint cycles with cycle length of σi being µi , for i = 1, . . . , ℓ.
110 3. Schur duality
Below we will freely use the statements in Appendix A.3 on Schur Q-functions
Qλ . Recall the symmetric function qn defined by the generating function (A.40),
which is,
1 + xit
∑ qnt n = ∏ 1 − xit .
n≥0 i≥1
Proof. It follows by the definition of ch, Lemma 3.36, and (A.45) that
ch(ξn ) = ∑ 2ℓ(α) z−1
α pα = qn .
α∈OPn
n. Hence ch is an isomorphism.
We emphasize that these are precisely the same recursive relations for the Schur
Q-functions Qλ ; see (A.48).
Lemma 3.39. We have ch(ξλ ) = Qλ and ⟨ξλ , ξµ ⟩ = 2ℓ(λ) δλµ , for λ, µ ∈ SP.
Proof. By Lemma 3.37, we have ch(ξn ) = qn = Q(n) . The general case of the first
identity follows since ch is a ring homomorphism, and in addition, ξλ and Qλ are
obtained from ξn ’s and qn ’s, respectively, by the same recursive relations.
The second identity follows from the fact that ch is an isometry and the formula
⟨Qλ , Qµ ⟩ = 2ℓ(λ) δλµ from (A.54).
ℓ(λ)−δ(λ)
We define ζλ := 2− 2 ξλ , for λ ∈ SPn .
follows easily by induction using the recursive relations (3.24) and (3.25).
Proof. The first identity follows from Lemma 3.39 and the definition of ζλ .
Write pα = ∑λ∈SPn aλα Qλ for some scalars aλα . Recall from (A.52) and (A.54)
that ⟨Qλ , Qµ ⟩ = 2ℓ(λ) δλµ and ⟨pα , pβ ⟩ = 2−ℓ(α) zα δαβ . Then,
ℓ(λ)−δ(λ) ℓ(λ)+δ(λ)
aλα = 2−ℓ(λ) ⟨pα , Qλ ⟩ = 2−ℓ(λ) 2 2 z−1 λ
α ζα ⟨pα , pα ⟩ = 2
− 2 −ℓ(α)
ζλα ,
where the middle equation uses the first identity of the corollary.
Theorem 3.43. Let λ ∈ SPn and ℓ(λ) = ℓ. Then, ζλ is the character of a simple
spin Ben -module (which is to be denoted by Dλ ). Moreover, the degree of ζλ is equal
to
ℓ−δ(λ) n! λi − λ j
2n− 2 ∏
λ1 ! . . . λℓ ! i< j λi + λ j
.
Proof. Since ζλ ∈ R− by Lemma 3.40 and ⟨ζλ , ζλ ⟩ = 1 for ℓ(λ) even by Corol-
lary 3.41, ζλ or −ζλ , for ℓ(λ) even, is a simple character of type M. By Proposi-
tion 3.7 and Theorem 3.29, these are all simple characters of type M. Now since
⟨ζλ , ζλ ⟩ = 2 for ℓ(λ) odd by Corollary 3.41, we have two possibilities: (a) ζλ or
−ζλ is either simple of type Q, or (b) ζλ is of the form ±ζµ ± ζν with both ζµ and
ζν being simple of type M which means both ℓ(µ) and ℓ(ν) being even. Case (b)
cannot occur, otherwise it would contradict the linear independence of ζλ for all
λ ∈ SPn .
To show that ζλ instead of −ζλ for each λ ∈ SPn is a character of a simple
module, it suffices to know that ζλ(1n ) or ξλ(1n ) is positive. To that end, we claim that
n! λi − λ j
ξλ(1n ) = 2n ∏
λ1 ! . . . λℓ ! i< j λi + λ j
,
Laplacian type expansion of the classical Pfaffian identity (cf. Macdonald [Mac,
III.8, Ex. 5]) ( )
ti − t j ti − t j
Pf = ∏ .
ti + t j 1≤i< j≤2n ti + t j
3.4.1. A double centralizer property. Recall from Section 3.2 (by setting m = n)
that we have a representation (V ⊗d , Φd ) of gl(n|n), hence of its subalgebra q(n),
and we also have a representation (V ⊗d , Ψd ) of the symmetric group Sd . More-
over, the actions of gl(n|n) and the symmetric group Sd on V ⊗d commute with
each other.
Note in addition that the Clifford algebra Cd acts on V ⊗d , also denoted by Ψd :
Ψd (ci ).(v1 ⊗ . . . ⊗ vd ) = (−1)(|v1 |+...+|vi−1 |) v1 ⊗ . . . ⊗ vi−1 ⊗ Pvi ⊗ . . . ⊗ vd ,
where P is given in (1.10), each vi ∈ V is assumed to be Z2 -homogeneous, and
1 ≤ i ≤ n.
Lemma 3.44. Let V = Cn|n . The actions of Sd and Cd above give rise to a repre-
sentation (V ⊗d , Ψd ) of Hd . Moreover, the actions of q(n) and Hd on V ⊗d commute
with each other.
Proof. To see the relation (3.20) holds, it suffices to check for σ = ( j, j + 1), 1 ≤
j ≤ d − 1. We compute that
Ψd ((i, i + 1))Ψd (ci ).(v1 ⊗ . . . ⊗ vd )
= Ψd ((i, i + 1))(−1)(|v1 |+...+|vi−1 |) v1 ⊗ . . . ⊗ Pvi ⊗ vi+1 ⊗ . . . ⊗ vd
= (−1)(|vi |+1)|vi+1 | (−1)(|v1 |+...+|vi−1 |) v1 ⊗ . . . ⊗ vi+1 ⊗ Pvi ⊗ . . . ⊗ vd ,
Ψd (ci+1 )Ψd ((i, i + 1)).(v1 ⊗ . . . ⊗ vd )
= Ψd (ci+1 )(−1)(|vi |·|vi+1 |) v1 ⊗ . . . ⊗ vi+1 ⊗ vi ⊗ . . . ⊗ vd
= (−1)(|vi |·|vi+1 |) (−1)(|v1 |+...+|vi−1 |+|vi+1 |) v1 ⊗ . . . ⊗ vi+1 ⊗ Pvi ⊗ . . . ⊗ vd .
Hence, we have Ψd ((i, i + 1))Ψd (ci ) = Ψd (ci+1 )Ψd ((i, i + 1)). This further implies
that Ψd ((i, i + 1))Ψd (ci+1 ) = Ψd (ci )Ψd ((i, i + 1)). A similar calculation shows that
Ψd (( j, j + 1))Ψd (ci ) = Ψd (ci )Ψd (( j, j + 1)) for j ̸= i, i − 1.
114 3. Schur duality
By the definitions of q(n) and of Ψd (ci ) via P, the action of q(n) commutes
with the action of ci for 1 ≤ i ≤ d. Since gl(n|n) commutes with Sd , so does the
subalgebra q(n) of gl(n|n). Hence, the action of q(n) commutes with the action of
Hd on V ⊗d .
Theorem 3.45. The images Φd (U(q(n))) and Ψd (Hd ) satisfy the double central-
izer property, i.e.,
Φd (U(q(n))) =EndHd (V ⊗d ),
Endq(n) (V ⊗d ) = Ψd (Hd ).
3.4.2. The Sergeev duality. Recall from (1.35) that, for a partition λ of d with
length ℓ(λ),
{
0, if ℓ(λ) is even,
δ(λ) =
1, if ℓ(λ) is odd.
3.4. Schur-Sergeev duality for q(n) 115
Recall further that Dλ stands for the simple Hd -module with character ζλ (see
Theorem 3.43), and L(λ) for the simple q(n)-module with highest weight λ (see
Section 2.1.6).
Theorem 3.46. Let V = Cn|n . As a U(q(n)) ⊗ Hd -module, we have
⊕
(3.27) ∼
V ⊗d = 2−δ(λ) L(λ) ⊗ Dλ .
λ∈SPd ,ℓ(λ)≤n
Proof. Let W = V ⊗d . It follows from the double centralizer property and the
semisimplicity of the superalgebra Hd that we have a multiplicity-free decomposi-
tion of the (q(n), Hd )-module W :
⊕
W∼ = 2−δ(λ) L[λ] ⊗ Dλ ,
λ∈Q(d,n)
where L[λ] is some simple q(n)-module associated to λ, whose highest weight (with
respect to the standard Borel) is to be determined. Also to be determined is the
index set Q(d, n) = {λ ∈ SPd | L[λ] ̸= 0}.
We shall identify a weight µ = ∑ni=1 µi εi occuring in W with a composition
µ = (µ1 , . . . , µn ) ∈ CP(d, n). We have the following weight space decomposition:
⊕
(3.28) W= Wµ ,
µ∈CP(d,n)
where Wµ has a linear basis ei1 ⊗ . . . ⊗ eid , with the indices satisfying the following
equality of sets:
{i1 , . . . , id } = {1, . . . , 1, , 1, . . . , 1, . . . , n, . . . , n, n, . . . , n}.
| {z } | {z }
µ1 µn
bλλ = 1. To complete the proof of the theorem, we shall need the following.
where K
Lemma 3.47. Let µ = (µ1 , . . . , µn ) be a composition of d. We have the following
decomposition of Wµ as an Hd -module:
⊕
Wµ ∼ bλµ Dλ .
ℓ(λ)−δ(λ)
= 2 2 K
λ∈SP,λ≥µ
bλµ ∈ Z+ .
In particular, K
116 3. Schur duality
Applying the characteristic map to both sides of (3.31), and using (3.32), (3.33)
and (3.29), we obtain that
ℓ(λ)−δ(λ)
qµ = ∑ 2− 2 K̃λµ Qλ .
λ
bλµ .
ℓ(λ)−δ(λ)
It follows by a comparison of this identity with (3.30) that K̃λµ = 2 2 K
We return to the proof of the theorem. Note that the simple q(n)-module L[λ]
⊕ [λ]
has a weight space decomposition L[λ] = µ∈CP(d,n),µ≤λ Lµ , and hence, λ ∈ Pd (n)
if L[λ] ̸= 0. Among all such µ, clearly λ corresponds to a highest weight. Hence,
we conclude that L[λ] = L(λ), the simple g-module of highest weight λ, and that
Q(d, n) = {λ ∈ SPd | ℓ(λ) ≤ n}. This completes the proof of Theorem 3.46.
On the other hand, by applying ch2 to (3.27) and using (3.32), we obtain that
ℓ(λ)−δ(λ)
ch2 (V ⊗d ) = ∑ 2−δ(λ) chL(λ) · 2− 2 Qλ (z).
λ∈SPd ,ℓ(λ)≤n
Now the theorem follows by comparing the above two identities and noting the
linear independence of the Qλ (z)’s.
3.5. Notes
Section 3.1. The classification of finite-dimensional simple associative superalge-
bras was due to Wall [Wal] over a general field, and it is somewhat simplified over
C in Józefiak [Jo1]. The basics on representation theory of superalgebras and finite
supergroups, including Wedderburn Theorem, Schur’s Lemma and the role of split
conjugacy classes, have been developed in [Jo1].
Section 3.2. The superalgebra generalization of Schur duality between general
linear Lie superalgebra gl(m|n) and the symmetric group Sd was due to Sergeev
[Sv1] and Berele-Regev [BeR], and the character of irreducible polynomial repre-
sentations of gl(m|n) was given in terms of super Schur polynomials. This general-
ization is intimately related to the combinatorics of supertableaux. Proposition 3.20
on the diagrammatic interpretation of the degree of atypicality of a polynomial
dominant weight of gl(m|n) was due to Cheng-Wang [CW5], and Corollary 3.21
appeared in Moens and van der Jeugt [MJ, Proposition 2.1]. The semisimplicity
of the category of polynomial modules of gl(m|n) was expected by experts. Our
proof here, which is based on a comparison of central characters, is borrowed from
Cheng-Kwon [CK].
Section 3.3. The algebra Hd is a twisted group algebra for a double cover Bed
of the hyperoctraheral group. The classification of the split conjugacy classes for
Bed (Theorem 3.29) was due to Read [Re]. We follow Józefiak [Jo3] to develop
systematically a superalgebra approach toward the characteristic map and spin rep-
resentation theory for Bed .
Section 3.4. The Sergeev duality is a version of Schur duality between the
queer Lie superalgebra q(n) and an algebra Hd , and it was outlined in [Sv1]. This
leads to a character formula in terms of Schur Q-functions for the irreducible poly-
nomial q(n)-modules [Sv1]. Here we present complete proofs which use exten-
sively the results from Sections 3.1 and 3.3.
Various results on symmetric functions including super Schur functions and
Schur Q-functions relevant to this Chapter are collected in Appendix A.
Chapter 4
119
120 4. Classical invariant theory
and P(U) with induced G-actions are completely reducible. The basic question of
classical invariant theory is to describe
(1) the space of tensor G-invariants (U ⊗d )G ;
(2) the algebra of polynomial G-invariants P(U)G ;
(3) the algebra of G-invariants in the tensor algebra P(U) ⊗ ∧(V ).
These different versions of G-invariants turn out to be closely related to each other.
Theorem 4.1. Let G be a classical (or reductive) group, and let U,V be finite-
dimensional rational G-modules. Then, the algebra of G-invariants in P(U) ⊗
∧(V ) is finitely generated as a C-algebra.
Proof. Set R := P(U) ⊗ ∧(V ). Under the assumptions of the theorem, the G-
module R is completely reducible, and hence the subalgebra J := (P(U) ⊗ ∧(V ))G
is a direct summand of the G-module R. Let us denote by ϖ : R −→ J the natural
projection which is G-equivariant.
The algebra R is naturally Z+ -graded by the total degree, denoted by deg, on
the polynomial subalgebra and the exterior subalgebra. Denote by J+ the subspace
of J which consists of G-invariant elements in R with zero constant terms, and
denote by ⟨J+ ⟩ the (two-sided) ideal of R generated by J+ . Then ⟨J+ ⟩, when
viewed as a submodule of the finitely generated P(U)-module R = P(U) ⊗ ∧(V ),
is finitely generated over P(U), by Hilbert basis theorem. In particular, ⟨J+ ⟩ as an
ideal of R is finitely generated. We can further take a set of generators φ1 , . . . , φn
of the ideal ⟨J+ ⟩ to be homogeneous elements, say deg φi = di ≥ 1 for each i.
To complete the proof of the theorem, we shall show that φ1 , . . . , φn generate J
as a C-algebra. Indeed, let φ ∈ J+ be a homogeneous element. Write φ = ∑i fi φi
for some fi ∈ R. Then,
φ = ϖ(φ) = ϖ(∑ fi φi ) = ∑ ϖ( fi )φi ,
i i
where deg ϖ( fi ) ≤ deg fi < deg φ. By induction on the degree, we can assume that
ϖ( fi ) lies in the C-subalgebra generated by φ1 , . . . , φn .
4.1.2. Tensor and multilinear FFT for GL(V ). Now let V be a finite-dimensional
vector space, and let G = GL(V ) be the general linear Lie group on V . Let U =
V ⊗d ⊗ (V ∗ )⊗k , the space of mixed tensors of type (d, k). Denote the representation
of G on such a U by ρd,k .
4.1. FFT for the general linear Lie group 121
Recall the following commuting actions from Lemma 3.8 in Section 3.2 where
we have replaced the action of gl(V ) by G = GL(V ):
Φd Ψd
G y V ⊗d x Sd .
By Schur duality (Theorem 3.9), we have
(4.2) EndG (V ⊗d ) = Ψd (CSd ).
Theorem 4.2 (Tensor FFT for GL(V )). Let G = GL(V ). There is no non-zero G-
invariants in the tensor space V ⊗d ⊗ (V ∗ )⊗k , for k ̸= d. The space of G-invariants
in V ⊗d ⊗ (V ∗ )⊗d is spanned by the Θσ for σ ∈ Sd .
Lemma 4.3. Let V1 , . . . ,Vp be finite-dimensional vector spaces. Then, the dual
space (V1 ⊗ · · · ⊗ Vp )∗ can be naturally identified with the space of multilinear
functions on V1 ⊕ · · · ⊕Vp .
122 4. Classical invariant theory
( )∗
By Lemma 4.3, we shall regard the dual space V ⊗d ⊗ (V ∗ )⊗k as the space
of multilinear functions f : V d ⊕ V ∗k → C. There is no nonzero G-invariant mul-
tilinear functions, for k ̸= d. For, aI ∈ G sends (v, φ) = (v1 , . . . , vd , φ1 , . . . , φk ) to
(av1 , . . . , avd , a−1 φ1 , . . . , a−1 φk ); hence, if f is a G-invariant multilinear function,
then f (a(v, φ)) = ad−k f (v, φ).
Let k = d. Define the contraction ⟨ j|i∗ ⟩, for 1 ≤ i, j ≤ d, by
(4.4) ⟨ j|i∗ ⟩(v1 , . . . , vd , φ1 , . . . , φk ) := φi (v j ).
Given σ ∈ Sd , we let
fσ := ⟨1|σ(1)∗ ⟩⟨2|σ(2)∗ ⟩ · · · ⟨d|σ(d)∗ ⟩.
The tensor FFT for GL(V ) (Theorem 4.2) can now be reformulated as follows.
Theorem 4.4 (Multilinear FFT for GL(V )). Let G = GL(V ). Then there is no
nonzero G-invariant multilinear functions on V d ⊕ V ∗k , for k ̸= d. The space of
G-invariant multilinear functions on V d ⊕ V ∗d is spanned by the functions fσ for
σ ∈ Sd .
( ) ( ⊗d )∗
Proof. Thanks to V ∗∗ ∼
= V , we identify (V ∗ )⊗d ⊗V ⊗d ∼= V ⊗ (V ∗ )⊗d . Thus,
we have by (4.1) and (4.2) that
( )G ( )G
(V ∗ )⊗d ⊗V ⊗d ∼ = V ⊗d ⊗ (V ∗ )⊗d ∼ = Ψd (CSd ).
( )G
By Lemma 4.3, (V ∗ )⊗d ⊗V ⊗d can be identified with the space of G-invariant
multilinear functions on V ⊗d ⊗ (V ∗ )⊗d . Hence, the theorem follows as a dual ver-
sion of the tensor FFT for GL(V ) (see Theorem 4.2), where fσ is the counterpart
of Θσ in (4.3).
4.1.3. Formulation of the polynomial FFT for GL(V ). Again let G = GL(V ).
Then G acts naturally on the dual space V ∗ and also on the direct sums V k and
V ∗m , for k, m ≥ 0. We have natural identifications of G-modules:
∼ ∼
V k −→ Hom(Ck ,V ), V ∗m −→ Hom(V, Cm ).
This leads to a G-equivariant isomorphism between polynomial algebras
∼
P(V k ⊕V ∗m ) −→ P(Vm,k ),
where we have denoted
Vm,k := Hom(Ck ,V ) ⊕ Hom(V, Cm ).
We identify the Hom-space Hom(Ck , Cm ) with the space Mmk of m×k matrices
over C, and define the composition map
τ : Vm,k → Mmk , τ(x, y) = yx.
4.1. FFT for the general linear Lie group 123
Note that τ is G-equivariant, where we let G act on Mmk trivially. Hence, it induces
an algebra homomorphism
τ∗ : P(Mmk ) −→ P(Vm,k )G , τ∗ ( f ) = f ◦ τ.
The image of τ∗ consists of G-invariants, since for g ∈ G, x ∈ Vm,k , by the G-
equivariance of τ, we have τ∗ ( f )(g.x) = f (τ(g.x)) = f (τ(x)) = τ∗ ( f )(x).
Denote by xi j the (i, j)th matrix coefficient on Mmk , where 1 ≤ i ≤ m, 1 ≤ j ≤ k.
It follows by a direct computation that τ∗ (xi j ) coincides with the contraction ⟨ j|i∗ ⟩
(which is defined as in (4.4) with obvious modifications of indices), or equivalently,
τ∗ (xi j )(v1 , . . . , vk , v∗1 , . . . , v∗m ) = v∗i (v j ).
Theorem 4.5 (Polynomial FFT for GL(V )). The algebra of GL(V )-invariants in
P(V k ⊕V ∗m ) is generated by the contractions ⟨ j|i∗ ⟩, for 1 ≤ i ≤ m, 1 ≤ j ≤ k.
Proof. Parts (1) and (2) follow directly from the definitions.
Setting w1 = . . . = w p = w in (4.5), we obtain that
∑
r
(t1 + . . . + t p ) p f (w) = fr1 ...r p (w, . . . , w)t1r1 · · ·t pp .
r1 ,...,r p
Proof. Since G is a classical group, all the representations of G involved are com-
pletely reducible. By Proposition 4.8, R : P(W p )(1,1,...,1) → P p (W ) is surjective
and G-equivariant, and so R sends a given G-isotypic component of P(W p )(1,1,...,1)
onto the corresponding isotypic component of P p (W ). The corollary follows by
considering the isotypic component of the trivial module.
Now we are ready to derive the polynomial FFT for GL(V ) from the multilin-
ear FFT for GL(V ).
4.2.2. The symplectic and orthogonal groups. Let V be a vector space equipped
with a non-degenerate symmetric or skew-symmetric bilinear form ω : V ×V → C.
We denote by G(V, ω) the subgroup of GL(V ) which preserves the form ω, that is,
G(V, ω) = {g ∈ GL(V ) | ω(gv1 , gv2 ) = ω(v1 , v2 ), ∀v1 , v2 ∈ V }.
4.2. Polynomial FFT for classical groups 127
The choices of (4.6) and (4.7) play prominent roles in the later Sections 5.3.2
and 5.3.1, respectively.
Theorem 4.13 (Polynomial FFT for G± (V )). The homomorphism τ∗± : P(SM± k )→
± (V ) ± (V )
P(V ) k G is surjective. Equivalently, P(V )
k G is generated by the functions
(i| j), where 1 ≤ i ≤ j ≤ k in the + case and 1 ≤ i < j ≤ k in the − case.
4.2.4. From basic to general polynomial FFT. Let us now refer to the statement
in the polynomial FFT for G± (V ) as formulated in Theorem 4.13 as FFT(k), for
k ≥ 1. Set N = dimV as before.
Assuming a basic case FFT(N) holds for now, let us complete the proof of
FFT(k) for all k. Denote by Sk the subspace spanned by the functions (i| j), where
1 ≤ i, j ≤ k.
FFT(N) =⇒ FFT(k) for k < N. (Indeed, the same argument below shows that
FFT(ℓ) =⇒ FFT(k) for all ℓ ≥ k ≥ 1.) The subspace SN is GLN -stable. Moreover,
SN ∩ V k is simply Sk . Now FFT(k) follows by applying FFT(N) and Proposi-
tion 4.12 for G = G± (V ) and S = SN .
FFT(N) =⇒ FFT(k) for k > N. According to FFT(N), SN generates the al-
gebra P(V N )G . Observe that ⟨SN ⟩GLk = Sk . Hence, by Weyl’s Theorem 4.11, the
algebra P(V k )G is generated by Sk , whence FFT(k).
This completes the proof of Theorem 4.13, modulo the proof of the basic case
FFT(N). We shall prove the basic case FFT(N) as Theorem 4.16 in the next sub-
section.
4.2.5. The basic case. In this subsection, we shall prove the following basic case
of the polynomial FFT for G± (V ) (that is, Theorem 4.13 for k = dimV ).
± (V )
Theorem 4.16. Let N = dimV . Then, the algebra P(V N )G is generated by the
functions (i| j), for 1 ≤ i, j ≤ N.
The orthogonal group case. Let V = CN , and take (x, y) = ∑Ni=1 xi yi , where
x = (x1 , . . . , xN ), y = (y1 , . . . , yN ) in CN , so O(V ) = ON . Denote
V1 := CN−1 × {0} ⊆ V
and denote by (·, ·)1 the bilinear form on V1 obtained by restriction from (·, ·) on V .
For N = 1, O1 = {±1}, Theorem 4.16 is clear.
Assume now that N > 1. We have an orthogonal decomposition V = V1 ⊕ CeN
with respect to (·, ·).
Let f ∈ P(V N )ON . Since −IN ∈ ON , f (X) = f (−IN · X) = f (−X), so f cannot
contain a nonzero odd degree term. Without loss of generality, let us assume f is
homogeneous of even degree 2d. Below, we will consider the restriction of f to
the Zariski-open subset of vectors v = (v1 , . . . , vN ) of V N given by the inequality
(vN , vN ) ̸= 0.
Let κ ∈ C such that κ2 = (vN , vN ). Since (vN , vN ) = (κeN , κeN ), there exists an
orthogonal matrix, g ∈ ON , such that gvN = κeN . Set
(4.8) v′i = κ−1 gvi , (i = 1, . . . , N − 1).
Then,
f (v) = f (gv) = κ2d f (v′1 , . . . , v′N−1 , eN ) = (vN , vN )d f (v′1 , . . . , v′N−1 , eN ).
Let
(4.9) v′i = v′′i + ti eN (i = 1, . . . , N − 1),
be the orthogonal decomposition in V = V1 ⊕ CeN for v′′i ∈ V1 . Using (4.8), we
compute that
(4.10) ti = (v′i , eN ) = (κ−1 gvi , κ−1 gvN ) = (vi , vN )/(vN , vN ).
i
Writing t I = t1i1 · · ·tN−1
N−1
for multi-indices I = (i1 , . . . , iN−1 ) and regarding ti as inde-
pendent variables, we have a decomposition of the following form:
(4.11) f (v′1 , . . . , v′N−1 , eN ) = ∑ fˆI (v′′1 , . . . , v′′N−1 )t I ,
I
where fˆI are polynomial functions uniquely determined by f . Noting from (4.10)
that ti is invariant under the transformation v′i 7→ gv′i for g ∈ ON−1 , we conclude
from (4.11) that fˆI is ON−1 -invariant for each I.
By induction hypothesis, there exist polynomials φI in the functions (i| j), 1 ≤
i ≤ j ≤ N − 1, such that fˆI = φI when evaluated on (v′′1 , . . . , v′′N−1 ). Using (4.8),
(4.9) and (4.10), we compute that
(vi , v j )(vN , vN ) − (vi , vN )(v j , vN )
(4.12) (v′′i , v′′j ) = (v′i , v′j ) − (ti eN ,t j eN ) = .
(vN , vN )2
Let us recall that we have been considering the restriction of f to the open subset
defined by (vN , vN ) ̸= 0. Plugging (4.12) into fˆI = φI , we have f (v) = (vN , vN )−p F1 ,
4.2. Polynomial FFT for classical groups 131
The symplectic group case. Let V = CN be equipped with the standard basis
{e1 , . . . , eN }, where N = 2ℓ is even. For definiteness, let us take the symplectic
group SpN = Sp(V ) defined via the symplectic form ω on V :
ℓ
ω(x, y) = ∑ (x2i−1 y2i − x2i y2i−1 ),
i=1
where fˆIJ are polynomial functions uniquely determined by f . Noting from (4.15)
that si ,ti are invariant under the transformation v′i 7→ gv′i for g ∈ SpN−2 , we conclude
from (4.16) that fˆIJ is SpN−2 -invariant for every I, J.
By inductive assumption, there exist polynomials φIJ in the functions (i| j), 1 ≤
i ≤ j ≤ N − 2, such that fˆIJ = φIJ when evaluated at (v′′1 , . . . , v′′N−2 ). Using (4.13),
(4.14) and (4.15), we compute that
ω(v′′i , v′′j ) = ω(v′i , v′j ) + ti s j − sit j
(4.17) ω(vi , v j )ω(vN−1 , vN ) − ω(vi , vN−1 )ω(v j , vN ) + ω(vi , vN )ω(v j , vN−1 )
= .
ω(vN−1 , vN )2
Recall that we have been considering the restriction of f to the open subset defined
by ω(vN−1 , vN ) ̸= 0. Plugging (4.17) into fˆIJ = φIJ , we have f (v) = ω(vN−1 , vN )−p F,
for some polynomial F in ω(vi , v j )i≤ j≤N and some positive integer p, whenever
ω(vN−1 , vN ) ̸= 0. Repeating the same trick used in the orthogonal group case, we
see that the polynomial function ω(vN−1 , vN ) p divides F.
This completes the proof of the theorem.
4.3.1. Tensor FFT for classical groups. Let G = G(V, ω), where V is a vector
space equipped with a non-degenerate symmetric or skew-symmetric bilinear form
ω. Since V ∼
= V ∗ as a G-module, it suffices to consider the G-invariants on the ten-
sor space V instead of a mixed tensor space V ⊗m ⊗(V ∗ )⊗n , for m, n ≥ 0. Observe
⊗m
Then T −1 (IV ⊗k ) = θk .
Recall the following commuting actions:
Φ2k Ψ2k
GL(V ) y V ⊗2k x S2k .
Since G is a subgroup of GL(V ) which commutes with S2k , Ψ2k (σ)(θk ) is G-
invariant for any σ ∈ S2k .
Theorem 4.17 (Tensor FFT for G(V, ω)). Let G = G(V, ω), where ω is a non-
degenerate symmetric or skew-symmetric bilinear form on V . Then (V ⊗m )G = 0
for m odd. Moreover, (V ⊗2k )G has a spanning set {Ψ2k (σ)(θk ) | σ ∈ S2k }.
The tensor FFT for G(V, ω) affords an equivalent dual reformulation on the
G-invariants of (V ∗⊗m ) ∼
= (V ⊗m )∗ . We let (i| j) be the function such that
(i| j)(v1 , . . . , vm ) = ω(vi , v j ).
Denote by
θ∗k = (1|2)(3|4) . . . (2k − 1|2k)
the G-invariant function on (V ⊗m ) which corresponds to the identity under the
∼
isomorphism T ∗ : V ∗⊗2k −→ End(V ⊗k ). Denote by Ψ∗2k the natural action of S2k
on (V ∗⊗2k ), which clearly commutes with the natural action of G.
134 4. Classical invariant theory
Theorem 4.18 (Tensor FFT for G(V, ω), a dual version). Let G = G(V, ω), where
ω is non-degenerate symmetric or skew-symmetric. Then (V ∗⊗m )G = 0 for m odd.
Moreover, (V ∗⊗2k )G has a spanning set {Ψ∗2k (σ)(θ∗k ) | σ ∈ S2k }.
Theorems 4.17 and 4.18 are equivalent, and we shall only present a detailed
proof of the tensor FFT as formulated in Theorem 4.18. The strategy of the proof
can be outlined as follows. First, we introduce an auxiliary polarization variable
X ∈ End V to reformulate this tensor FFT problem as one involving the basic case
of the polynomial FFT as in Theorem 4.16 (note End V ∼
= V N for N = dimV ). Then,
by applying Theorem 4.16, we transfer the problem at hand to an FFT problem for
GL(V ), for which Theorem 4.2 applies.
4.3.2. From tensor FFT to supersymmetric FFT. Let W = W0̄ ⊕W1̄ be a finite-
dimensional vector superspace. We denote the supersymmetric algebra of W by
where the summation is taken over all nonnegative integers αa , βb , γc , and δd such
that |α| + |β| + |γ| + |δ| = r. Here we have denoted α = (α1 , . . . , αm ), |α| = ∑a αa ,
and similarly for β, |β|, and so on. We regard the space Sαβγδ as a subspace of the
following mixed tensor space
identity IV ∈ End(V ). We shall denote by ⟨a, i|b, j∗ ||⟩ the corresponding quadratic
GL(V )-invariant in the following space (which is ∼= V ⊗V ∗ ):
pairing the ith copy of V in V αa with the jth copy of V ∗ in (V ∗ )βb , where 1 ≤
a ≤ m, 1 ≤ i ≤ αa and 1 ≤ b ≤ n, 1 ≤ j ≤ βb . In similar notation, we have the
quadratic GL(V )-invariants ⟨a, i|||d, q∗ ⟩, ⟨|b, j∗ |c, p|⟩, and ⟨||c, p|d, q∗ ⟩, where 1 ≤
a ≤ m, 1 ≤ i ≤ αa , 1 ≤ b ≤ n, 1 ≤ j ≤ βb , 1 ≤ c ≤ k, 1 ≤ p ≤ γc , 1 ≤ d ≤ ℓ, 1 ≤ q ≤ δd .
By the tensor FFT for GL(V ) (Theorem 4.2), the GL(V )-invariants of T αβγδ are
spanned by Θσ for σ ∈ SN , and each Θσ is a suitable product of N invariants
among ⟨a, i|b, j∗ ||⟩, ⟨a, i|||d, q∗ ⟩, ⟨|b, j∗ |c, p|⟩, and ⟨||c, p|d, q∗ ⟩ such that each copy
of V and V ∗ in T αβγδ is used exactly in one pairing.
Note that the subspace (Sαβγδ )GL(V ) is obtained from the space (T αβγδ )GL(V )
−
by a (partial) supersymmetrization. Hence, applying the Sα × Sβ × S− γ × Sδ -
supersymmetrization operator ϖαβγδ to the Θσ gives us a spanning set {ϖαβγδ (Θσ )}
for (Sαβγδ )GL(V ) , where Sα = Sα1 × . . . × Sαm and by supersymmetrization we
mean the symmetrization over the subgroup Sα × Sβ and the anti-symmetrization
(denoted by the minus sign superscript above) over the subgroup Sγ × Sδ . Let us
denote by
The cases for G = O(V ) or Sp(V ) can be pursued in an analogous manner. Set
We have a similar decomposition of the G-module S(W ) = ⊕α,γ Sαγ where Sαγ is
modified from (4.19) with β and δ being zero. The space Sαγ is a subspace of
the tensor space T αγ , where T αγ is modified from (4.20) with β and δ being zero.
Hence, the space (Sαγ )G can be viewed as a subspace of (T αγ )G . By the tensor FFT
(Theorem 4.17), we may assume below that |α| + |γ| is even, say, equal to 2N.
4.3. Tensor and supersymmetric FFT for classical groups 137
4.4. Notes
Section 4.1. The materials here are mostly classical and standard. Weyl’s book
[We] on invariant theory was influential, where the terminology of First Funda-
mental Theorem (FFT) for invariants was introduced. See Howe [H1, H2]; also
see the notes of Kraft-Procesi [KP]. Theorem 4.1 on the finite-generation of an
invariant subalgebra of the supersymmetric algebra is somewhat novel, as it is usu-
ally formulated for a polynomial algebra only.
Section 4.2. The results here are classical. Our proof of Weyl’s theorem fol-
lows Kraft-Procesi [KP]. Weyl’s theorem allows us to reduce the proof of the
polynomial FFT for classical groups in general to a basic case. The basic case is
then proved by induction on dimensions, and our presentation here does not differ
much from Goodman-Wallach [GW]. In [GW], the general polynomial FFT was
derived using the tensor FFT instead of Weyl’s theorem, while in Weyl [We], the
Cappelli identity was used essentially in the proof of the polynomial FFT.
Section 4.3. The tensor FFT for classical groups can be found in Weyl [We].
The quick proof of the tensor FFT for the orthogonal groups here is due to Atiyah-
Bott-Patodi [ABP], and it works for the symplectic groups similarly (also see a
presentation in Goodman-Wallach [GW]). The supersymmetric FFT for classical
groups, which is a common generalization of the polynomial FFT and skew FFT,
is taken from Howe [H1]. The supersymmetric FFT will be used in next chapter
on Howe duality for Lie superalgebras.
Chapter 5
Howe duality
The goal of this chapter is to explain Howe duality for classical Lie groups, Lie
algebras and superalgebras, and applications to irreducible character formulas for
Lie superalgebras. Like Schur duality, Howe duality involves commuting actions
of a pair of Lie group G and superalgebra g′ on a supersymmetric algebra S(U), and
S(U) is naturally an irreducible module over a Weyl-Clifford algebra WC. In the
main examples, G is one of the three classical groups GL(V ), Sp(V ), or O(V ), and
g′ is a classical Lie algebra or superalgebra, and U is a direct sum of copies of the
natural G-module and its dual. According to the FFT for classical invariant theory
in Chapter 4, when applied to the G-action on the associated graded gr WC, the
basic invariants generating (gr WC)G turn out to form the associated graded space
for a Lie (super)algebra g′ . From this it follows that the algebra of G-invariants
WCG is generated by g′ .
The most important cases of S(U) are the two cases of a symmetric algebra
and of an exterior algebra, corresponding respectively to U1̄ = 0 and U0̄ = 0. In
these two cases, the Howe dualities only involve Lie groups and Lie algebras, and
they may be regarded as Lie theoretic reformulations of the classical polynomial
invariant and skew invariant theories, respectively, treated in Chapter 4.
Howe duality provides a realization of an important class of irreducible mod-
ules, called oscillator modules, of Lie (super)algebra g′ . The irreducible G-modules
and irreducible oscillator g′ -modules appearing in the decomposition of S(U) are
paired in “duality”, which is much stronger than a simple bijection. Character
formulas for the irreducible oscillator g′ -modules are then obtained, via a compar-
ison with Howe duality involving classical groups G and infinite-dimensional Lie
algebras, which we develop in detail.
139
140 5. Howe duality
We regard U0̄∗ ⊆ U ∗ by extending u∗ ∈ U0̄∗ such that ⟨u∗ , w⟩ = 0 for all w ∈ U1̄ ,
and similarly we have U1̄∗ ⊆ U ∗ . In this way, we identify U ∗ ≡ U0̄∗ ⊕U1̄∗ as a vector
superspace. For u∗ ∈ U0̄∗ , we define the even derivation Du∗ of degree (−1, 0̄), and
for w∗ ∈ U1̄∗ , we define the odd derivation Dw∗ of degree (−1, 1̄) as follows:
Du∗ : S(U) −→S(U), Dw∗ : S(U) −→ S(U),
p
∗
∑ ⟨u∗ , ui ⟩u1 . . . ûi . . . u p w1 . . . wq ,
D
u1 . . . u p w1 . . . wq −→
u
i=1
q
∗
∑ (−1)k−1 ⟨w∗ , wk ⟩u1 . . . u p w1 . . . ŵk . . . wq ,
D
u1 . . . u p w1 . . . wq −→
w
k=1
where ui ∈ U0̄ and w j ∈ U1̄ . Here and below ûi means the omittance of ui , and so
on.
The linear operators Mu and Du∗ , for u ∈ U0̄ , u∗ ∈ U0̄∗ , in the superalgebra
End(S(U)) defined above are even, while Mw , Dw∗ ∈ End(S(U)), for w ∈ U1̄ , w∗ ∈
U1̄∗ , are odd. We shall denote by [·, ·] the supercommutator among linear operators.
Lemma 5.1. For x∗ , y∗ ∈ U ∗ and x, y ∈ U, we have
[Dx∗ , Dy∗ ] = [Mx , My ] = 0, [Dx∗ , My ] = ⟨x∗ , y⟩1.
5.1. Weyl-Clifford algebra and classical Lie superalgebras 141
Proof. The two statements in the proposition are equivalent. It follows by def-
inition and (5.2) that the map m is surjective. The injectivity of m is equivalent
to the claim that the “monomials” of an ordered basis for U and U ∗ form a basis
of WC(U U ), and the claim can then be established by a straightforward inductive
argument.
Alternatively, the proposition for WC(U U ) reduces to the well-known counter-
parts for W(U U 0̄ ) and Cl(U
U 1̄ ), thanks to WC(U U) ∼= W(U U 0̄ ) ⊗ Cl(U
U 1̄ ) (Note that
W(U U 0̄ ) and Cl(U
U 1̄ ) commute).
σ(T ) = ∑ cα,β,γ,δ xα ηβ yγ ξδ ∈ Sk (U
U ),
|α|+|β|+|γ|+|δ|=k
5.1. Weyl-Clifford algebra and classical Lie superalgebras 143
Proof. The G-module WC(U U )k is semisimple and hence isomorphic to the associ-
ated graded ⊕ j=0 WC(U
k U ) j /WC(U U ) j−1 . For a fixed j, the linear span of {xα ηβ ∂γ ðδ |
|α| + |β| + |γ| + |δ| = j} as a G-module is isomorphic to WC(U U ) j /WC(UU ) j−1 ,
which in turn is isomorphic to S (Uj
U ). The proposition follows.
Hence, ι(UU 0̄ ) is a representation of the Lie algebra S2 (ι(U U 0̄ )) under the adjoint
action. With respect to the ordered basis {x1 , . . . , xm , ∂1 , . . . , ∂m } for ι(U
U 0̄ ), we
can write down the matrices of (5.3), the span of which is precisely given by the
2m × 2m matrices of the m|m-block form
( )
A B
, B = Bt ,C = Ct .
C −At
U 0̄ )) ∼
This shows that S2 (ι(U U 0̄ ) and that ι(U
= sp(U U 0̄ ) is its natural representation.
Similarly, ∧ (ι(U
2 U )) is closed under the Lie bracket, ∧2 (ι(U
1̄ U )) ∼
1̄ = so(UU ) as
1̄
Lie algebras, and ι(U U 1̄ ) under the adjoint action of ∧2 (ι(U U 1̄ )) is its natural rep-
resentation. Since the case for ∧2 (ι(U U 1̄ )) is completely parallel to the case for
S2 (ι(U
U 0̄ )) above, we shall omit the details except recording the following identity
analogous to (5.4):
[ ]
(5.5) [c, d], c1 = 2⟨d, c1 ⟩′ c − 2⟨c, c1 ⟩′ d, ∀c, d, c1 ∈ ι(U U 1̄ ).
Proof. The first statement follows from Proposition 5.5(2) and the definition of
the filtration on WC(U U ). Clearly the adjoint action of G0̄ on ι(U U ) lifts to an ac-
tion of the group Sp(U U 0̄ ) × O(U
U 1̄ ). As each finite-dimensional filtered subspace
WC(U U )k for k ≥ 1 is preserved by the adjoint action of G0̄ , the G0̄ -module WC(U U)
5.1. Weyl-Clifford algebra and classical Lie superalgebras 145
We shall need the following variant of the Jacobson density theorem, a proof
of which can be found in [GW, Corollary 4.1.6].
Proposition 5.7. Let L be a vector space of countable dimension. Assume that a
subalgebra R of End(L) acts on L irreducibly. Then, for any finite-dimensional
subspace X of L, we have R|X = Hom(X, L).
146 5. Howe duality
Now, suppose that T ∈ HomG (X, L). Regarding T ∈ Hom(X, L), we may apply
Proposition 5.7 to find an element r ∈ R such that r|X = T . Then we have
T = π(T ) = π(r|X ) = π(r)|X ,
proving the claim.
Recall M λ = HomG (L(λ), L) by definition, for λ ∈ G(ρ).b Take two nonzero
−1
elements S, T ∈ HomG (L(λ), L). Then ψ = ST : T (L(λ)) → S(L(λ)) is an iso-
morphism of G-modules. It follows by the claim above that there exists u ∈ RG
which restricts to ψ : T (L(λ)) → L. Hence, uT : L(λ) → S(L(λ)) is an isomor-
phism of irreducible G-modules. By Schur’s Lemma, we must have S = cuT for
some scalar c, and this proves the irreducibility of the RG -module HomG (L(λ), L).
b
It remains to show that, for distinct λ, µ ∈ G(ρ), an RG -module homomorphism
φ : HomG (L(λ), L) → HomG (L(µ), L) must be zero. To that end, we take an arbi-
trary element T ∈ HomG (L(λ), L). Set S := φ(T ), U := T (L(λ)) ⊕ S(L(µ)), and
let p : U → S(L(µ)) ⊆ L be the natural projection, which is clearly G-equivariant.
By the above claim, there exists r ∈ RG such that p = r|U . Then, it follows from
pT = 0 that rT = 0 and then,
0 = φ(rT ) = rφ(T ) = rS = pS = S.
Hence we conclude that φ = 0 as expected.
5.1.5. A duality for Weyl-Clifford algebras. Let U = U0̄ ⊕U1̄ be a vector super-
space, and let G be a classical Lie subgroup of GL(U0̄ ) × GL(U1̄ ). In all the exam-
ples which give rise to concrete Howe dualities in the next sections, U0̄ and U1̄ are
actually direct sums of the natural G-module. The induced G-module (ρ, S(U)) is
semisimple. Also, the G-module WC(U U ) ⊆ End(S(U)) by conjugation given by
−1
g.T = ρ(g)T ρ(g ), for g ∈ G and T ∈ WC(U U ), is semisimple.
U ), S(U) is irre-
Lemma 5.9. As a module over the Weyl-Clifford algebra WC(U
ducible.
Proof. Identify S(U) with C[x, η], and identify WC(U U ) as the superalgebra of
polynomial differential operators on U as in Section 5.1.2. Then, given any nonzero
element f in C[x, η], we can find a suitable constant-coefficient differential operator
D ∈ WC(U U ) such that D. f = 1. Now applying suitable multiplication operators in
WC(U U ) to 1, we obtain all the monomials xα ηβ which span C[x, η]. The lemma
follows.
Then letting L = S(U) and R = WC(U U ), the triple (G, L, R) satisfies Condi-
tions (i)–(iii) in 5.1.4. Hence Theorem 5.8 is applicable in light of Lemma 5.9 and
gives us the following.
U )G )-module, S(U) is
Theorem 5.10. Retain the notations above. As a (G, WC(U
strongly multiplicity-free, i.e.,
⊕
S(U) ∼
= L(λ) ⊗ M λ .
b
λ∈G(ρ)
Proof. It follows from Proposition 5.4 that the symbol map σ restricts to an iso-
k to ⊕ j=0 S (U
U )G , for each k ≥ 0.
j
morphism from WC(U U )G k
Let a ∈ WC(U U )G
k . We shall show that a is generated by T1 , . . . , Tr by in-
duction on k. For k = 0 the claim is trivial. Suppose that k ≥ 1. By assump-
tion σ(a) = f (S1 , S2 , . . . , Sr ), for some polynomial f . Now consider the element
f (T1 , T2 , . . . , Tr ) ∈ WC(U k . We note that σ(a − f (T1 , . . . , Tr )) ∈ ⊕ j=0 S (U
U )G k−1 j
U ), and
hence a − f (T1 , . . . , Tr ) ∈ WC(U U )G
k−1 . By induction hyperthesis, a − f (T1 , . . . , Tr )
is generated by T1 , . . . , Tr , and hence so is a.
148 5. Howe duality
U )G
Proposition 5.12. Suppose that the generators T1 , . . . , Tr for the algebra WC(U
′
can be chosen such that g := ⊕i=1 CTi is a Lie subalgebra of WC(U
r G
U ) under
the super commutator. Then the irreducible WC(U G λ
U ) -modules M ’s (see The-
orem 5.10) are pairwise non-isomorphic g′ -modules, and we have the following
(G, g′ )-module decomposition:
⊕
S(U) ∼
= L(λ) ⊗ M λ .
b
λ∈G(ρ)
5.2.1. Howe dual pair (GL(k), gl(m|n)). Take V = Ck , and hence identify GL(V )
with GL(k) and gl(V ) with gl(k). Given m, n ≥ 0, we consider the superspace
U = U0̄ ⊕ U1̄ with U0̄ = V m and U1̄ = V n , and identify naturally U ≡ V ⊗ Cm|n .
We let e1 , . . . , ek denote the standard basis for the natural gl(k)-module Ck , and let
ei , i ∈ I(m|n), denote the standard basis for the natural module Cm|n of the general
linear Lie superalgebra gl(m|n). We set, for 1 ≤ r ≤ k, 1 ≤ a ≤ m, 1 ≤ b ≤ n,
(5.8) xra := er ⊗ eā , ηrb := er ⊗ eb .
Then the set {xra , ηrb } is a basis for U. We will denote by C[x, η] the polynomial
superalgebra generated by the elements in (5.8), and from now on further identify
The natural commuting actions of gl(k) and gl(m|n) on V ⊗ Cm|n induce com-
muting actions on C[x, η]. The following lemma is standard and can be verified by
a direct computation.
Lemma 5.13. The commuting actions of gl(k) and gl(m|n) on S(Ck ⊗ Cm|n ) =
C[x, η] are realized by the formulas (5.10) and (5.11), respectively.
We shall denote by g′ the linear span of the elements in (5.11). Indeed, the
elements in (5.11) form a linear basis for g′ . Let U = U ⊕U ∗ as before.
Theorem 5.14. The g′ , which forms an isomorphic copy of gl(m|n), generates the
algebra of GL(V )-invariants in WC(U U ). Moreover, as a (GL(k), gl(m|n))-module,
S(Ck ⊗ Cm|n ) is strongly multiplicity-free.
Proof. Set G = GL(k). Recall from Proposition 5.4 that the associated graded of
the filtered algebra WC(UU ) is S(UU ), and that the action of gl(k) on S(U
U ) lifts to
an action of G. In light of (5.9), we identify WC(U U ) with the superalgebra of
polynomial differential operators on C[x, η], and so the elements in (5.11) may be
regarded as elements in WC(U U ).
Let {y pi } denote the basis in U0̄∗ dual to the basis {x pi } for U0̄ , and let {ξ pt }
denote the basis in U1̄∗ dual to the basis {η pt } for U1̄ . By the supersymmetric
First Fundamental Theorem for GL(V ) (see Theorem 4.19), the algebra S(U U )G =
( )G
S(U0̄ ⊕U0̄∗ ) ⊗ ∧(U1̄ ⊕U1̄∗ ) is generated by
k k
zī j¯ = ∑ x pi y p j , zts = ∑ η pt ξ ps ,
p=1 p=1
(5.12)
k k
zīt = ∑ x pi ξ pt , zt ī = ∑ η pt y pi , 1 ≤ i, j ≤ m, 1 ≤ s,t ≤ n.
p=1 p=1
Observe that the symbol map σ in 5.1.2 sends each Eab in (5.11) to zab for a, b ∈
I(m|n), and hence by Proposition 5.11, the Eab ’s form a set of generators for
WC(U U )G . Now the theorem follows from Proposition 5.12, since the Eab ’s gener-
ate the Lie superalgebra gl(m|n) by Lemma 5.13.
150 5. Howe duality
5.2.2. (GL(k), gl(m|n))-Howe duality. We will first find the multiplicity-free de-
composition of the (GL(V ), gl(m))-module S(Ck ⊗ Cm ) explicitly. To do that, we
will take advantage of the Schur duality which was established in Chapter 3. As
the formulation below will involve Lie algebra gl(k) for various k, we will add the
index k to denote by Lk (λ) the irreducible gl(k)-module of highest weight λ. For
dominant integral weight λ, the gl(k)-module Lk (λ) lifts to a GL(k)-module, which
will be denoted by the same notation.
Let us consider the natural action of gl(k) × gl(m) on Ck ⊗ Cm and its induced
action on the dth symmetric tensor Sd (Ck ⊗ Cm ), for d ≥ 0. As usual, we identify a
polynomial weight for gl(k) with a partition of length at most k. Recall Pd denotes
the set of partitions of d.
Theorem 5.16 ((GL(k), gl(m))-Howe duality). As a (GL(k), gl(m))-module, we
have ⊕
Sd (Ck ⊗ Cm ) ∼
= Lk (λ) ⊗ Lm (λ).
λ∈Pd
ℓ(λ)≤min{k,m}
Proof. We shall show that Schur duality implies Howe duality. By Schur duality
(see Theorem 3.10), we have
(5.13) ∼ ⊕ λ∈P Lk (λ) ⊗ Sλ ,
(Ck )⊗d = ∼ ⊕ µ∈P Lm (µ) ⊗ Sµ .
(Cm )⊗d =
d d
ℓ(λ)≤k ℓ(µ)≤m
In the last step, we have used the well-known fact that Sλ ∼ = (Sλ )∗ , and hence by
Schur’s lemma
(Sλ ⊗ Sµ )∆Sd ∼
= HomSd (Sλ , Sµ ) ∼
= δλ,µ C.
This completes the proof of the theorem.
5.2. Howe duality for type A and type Q 151
Now let us consider the induced action of gl(k) × gl(m) on the dth exterior
tensor space ∧d (Ck ⊗ Cm ), for 0 ≤ d ≤ km. Recall that λ′ denotes the conjugate
partition of a partition λ. We have the following skew version of Howe duality.
Proof. We shall use Schur duality to derive the skew Howe duality. For an Sd -
module M, let us denote by M Sd ,sgn the submodule of M that transforms by the
sign character. Using (5.13), we have
( )Sd ,sgn
∧d (Ck ⊗ Cm ) ∼
= (Ck ⊗ Cm )⊗d
( )∆Sd ,sgn
∼
= (C k ⊗d
) ⊗ (C m ⊗d
)
⊕ ( )∆Sd ,sgn
∼
= Lk (λ) ⊗ Lm (µ) ⊗ Sλ ⊗ Sµ
λ,µ∈Pd
ℓ(λ)≤k,ℓ(µ)≤m
⊕
∼
= Lk (λ) ⊗ Lm (λ′ ).
λ∈Pd
ℓ(λ)≤k,ℓ(λ′ )≤m
5.2.3. Formulas for highest weight vectors. In this subsection, we shall find ex-
plicit formulas for the joint highest weight vectors in the (GL(k), gl(m|n))-Howe
duality decomposition (see Theorem 5.19). These formulas will play a key role in
the subsequent section on Howe duality for symplectic and orthogonal groups.
Recalling the notation from (5.10) and (5.11) the simple root vectors of gl(k)
and gl(m|n) are respectively (1 ≤ i ≤ k − 1, 1 ≤ s ≤ m − 1, 1 ≤ t ≤ n − 1):
m
∂ n
∂
(5.14) Ei,i+1 = ∑ xi j ∂xi+1, j + ∑ ηis ∂ηi+1,s ,
j=1 s=1
(5.15)
k
∂ k
∂ k
∂
Es̄,s+1 = ∑ x ps ∂x p,s+1 , Et,t+1 = ∑ η pt ∂η p,t+1 , Em,1 = ∑ x pm ∂η p1 .
p=1 p=1 p=1
the summand Lk (λ) ⊗ Lm|n (λ♮ ) in the decomposition of C[x, η]. Such a vector is
unique up to a scalar multiple, thanks to the multiplicity-free decomposition in
Theorem 5.19.
For 1 ≤ r ≤ m, define
x11 x12 · · · x1r
x21 x22 · · · x2r
(5.16) ♢r := det . .. .. .. .
.. . . .
xr1 xr2 · · · xrr
The column determinant of an r × r matrix with possibly non-commuting
entries A = [ai j ] is defined to be
cdet A = ∑ (−1)ℓ(σ) aσ(1)1 aσ(2)2 · · · aσ(r)r .
σ∈Sr
Proof. Set µ = λ≤r and ♢ = ♢1,2,...,r,µ′ in this proof. By Remark 5.21 and a direct
computation using (5.10), (5.11) and (5.14), we can verify the following.
(1) ♢ is non-zero.
(2) ♢ has weight (µ, µ♮ ) with respect to the action of gl(k) × gl(m|n).
(3) Each factor ♢i,λ′i in ♢ is annihilated by the operators in (5.14).
154 5. Howe duality
Theorem 5.23. Let λ be a partition such that ℓ(λ) ≤ k and λm+1 ≤ n. Then, a
highest weight vector of weight (λ, λ♮ ) in the gl(k) × gl(m|n)-module C[x, η] is
given by ♢1,2,...,λ1 ,λ′ = ♢1,λ′1 ♢2,λ′2 · · · ♢λ1 ,λ′λ .
1
Proof. The vector ♢1,2,...,λ1 ,λ′ is nonzero since each factor ♢i,λ′i is nonzero by Re-
mark 5.21 and distinct ηi j are involved in different ♢i,λ′i . Each factor ♢i,λ′i in
♢1,2,...,λ1 ,λ′ is a joint highest weight vector by a special case of Lemma 5.22, and
hence so is ♢1,2,...,λ1 ,λ′ . Also it is straightforward to verify by (5.10) and (5.11) that
♢1,2,...,λ1 ,λ′ has the correct weights (λ, λ♮ ).
Remark 5.24. When n = 0, the highest weight vector in Theorem 5.23 does not in-
volve the odd generators ηi j . When m = 0, the vector becomes simply a monomial
in the anti-commuting variables ηi j .
5.2.4. (q(m), q(n))-Howe duality. In this subsection, using the (q(n), Hd )-Sergeev
duality in Theorem 3.46, we shall obtain a (q(m), q(n))-Howe duality in a similar
fashion as in 5.2.2.
Recall that by definition q(m) consists of X ∈ gl(m|m) super-commuting ( with )
A B
P given in (1.10), i.e., q(m) consists of the m|m-block matrices of the form .
B A
Recall from Remark 1.9 that we have another nonstandard realization of the Lie su-
peralgebra q(m) as e q(m) = {X ∈(gl(m|m))| XP − PX = 0}, which consists of the
A B
m|m-block matrices of the form . We let q(m) act on Cm|m as the non-
−B A
standard isomorphic copy e q(m), while we let q(n) act on Cn|n by linear maps in the
standard way, i.e. linear maps super-commuting with P. This induces an action of
q(m) × q(n) on Cm|m ⊗ Cn|n .
Define the linear map P∆ : Cm|m ⊗ Cn|n → Cm|m ⊗ Cn|n by
Lemma 5.25. The actions of P∆ and q(m) × q(n) on Cm|m ⊗ Cn|n commute with
each other.
5.2. Howe duality for type A and type Q 155
e pq =
n
∂ ∂
E ∑ (x p j ∂xq j + ξ p j ∂ξq j ),
j=1
(5.19) n
∂ ∂
∑ (x p j ∂ξq j − ξ p j ∂xq j ),
pq
E = 1 ≤ p, q ≤ m.
j=1
ei j =
m
∂ ∂
E ∑ (x pi ∂x p j + ξ pi ∂ξ p j ),
p=1
(5.20) m
∂ ∂
Ei j = ∑ (x pi ∂ξ p j + ξ pi ∂x p j ), 1 ≤ i, j ≤ n.
p=1
Proof. By the Sergeev duality (Theorem 3.46), we have as a q(m) × Bed module
⊕
(Cm|m )⊗d ∼
= 2−δ(λ) Lm (λ) ⊗ Dλ .
λ∈SPd
ℓ(λ)≤m
5.3. Howe duality for symplectic and orthogonal groups 157
where we recall that Dλ denotes an irreducible Bed -module (or equivalently, an ir-
reducible Hd -module). Therefore, combined with Lemma 5.26, this gives us
e
(5.21) Sd (Cmn|mn ) ∼
= (((Cm|m )⊗d ) ⊗ ((Cn|n )⊗d ))∆(Bd )
⊕ e
∼
= 2−δ(λ) 2−δ(µ) (Lm (λ) ⊗ Dλ ⊗ Ln (µ) ⊗ Dµ )∆(Bd )
λ,µ∈SPd
ℓ(λ)≤m,ℓ(µ)≤n
⊕ e
∼
= 2−δ(λ) 2−δ(µ) (Lm (λ) ⊗ Ln (µ)) ⊗ (Dλ ⊗ Dµ )∆(Bd ) .
λ,µ∈SPd
ℓ(λ)≤m,ℓ(µ)≤n
Recall that the character of the irreducible q(n)-module Ln (λ) is given by the
Schur Q-function Qλ up to a 2-power; see Theorem 3.48. Now by taking the char-
acters of both sides of the isomorphism in Theorem 5.28 and summing over all
d ≥ 0, we recover the following Cauchy identity (A.54) in Appendix A:
1 + xi y j
∏ 1 − xi y j = ∑ 2−ℓ(λ) Qλ (x)Qλ (y).
i, j λ∈SP
aℓ,ℓ aℓ−1,ℓ ··· a2,ℓ a1,ℓ
aℓ,ℓ−1 aℓ−1,ℓ−1 ··· a2,ℓ−1 a1,ℓ−1
aℓ,ℓ−2 aℓ−1,ℓ−2 ··· a2,ℓ−2 a1,ℓ−2
(5.22) A♭ = . .. .. .. .. .
.. . . . .
aℓ,2 aℓ−1,2 ··· a2,2 a1,2
aℓ,1 aℓ−1,1 ··· a2,1 a1,1
{( ) }
A B
(5.23) sp(k) = | D = −A , B = B,C = C .
♭ ♭ ♭
C D
The standard Borel of sp(k) consists of the k × k upper triangular matrices of the
form (5.23) and so is compatible with the standard Borel of gl(k). Furthermore,
the Cartan subalgebra of sp(k) is spanned by the basis vectors Eii − Ek+1−i,k+1−i ,
for 1 ≤ i ≤ k/2.
The action of the Lie algebra sp(k) as the subalgebra of gl(k) on C[x, η] given
in (5.10) lifts to an action of the Lie group Sp(V ) = Sp(k). On the other hand,
the action ρ of gl(m|n) on C[x, η] with ρ(Eab ) = Eab , for a, b ∈ I(m|n), is modified
from (5.11) by a shift of scalars on the diagonal matrices:
k
∂ k k
∂
Eī j¯ = ∑ x pi ∂x p j + 2 δi j , Eīs = ∑ x pi ∂η ps , 1 ≤ i, j ≤ m,
p=1 p=1
(5.24)
k
∂ k k
∂
Est = ∑ η ps − δst ,
∂η pt 2
Esī = ∑ η ps ∂x pi , 1 ≤ s,t ≤ n.
p=1 p=1
5.3. Howe duality for symplectic and orthogonal groups 159
( k
2 )
Ist = ∑ η ps ηk+1−p,t − ηk+1−p,s η pt ,
p=1
(5.25)
( ∂
k
2
∂ ∂ ∂ )
∆ī j¯ = ∑ − ,
p=1 ∂x pi ∂xk+1−p, j ∂xk+1−p,i ∂x p j
k
2 ( ∂ ∂ ∂ ∂ )
∆īs = ∑ − , 1 ≤ i < j ≤ m,
p=1 ∂x pi ∂ηk+1−p,s ∂xk+1−p,i ∂η ps
k
2 ( ∂ ∂ ∂ ∂ )
∆st = ∑ − , 1 ≤ s ≤ t ≤ n.
p=1 ∂η ps ∂ηk+1−p,t ∂ηk+1−p,s ∂η pt
Let us denote by g′ the linear span of the elements in (5.24) and (5.25). Clearly,
the elements in (5.24) and (5.25) form a linear basis for g′ .
Lemma 5.29. (1) Under the super-commutator g′ forms a Lie superalgebra,
which is isomorphic to osp(2m|2n).
(2) The actions of Sp(k) and osp(2m|2n) on S(Ck ⊗ Cm|n ) commute.
elements in U ∗ . Similarly ∆ī j¯, ∆īs , ∆st defined above are Sp(k)-invariant in WC(UU ).
On the other hand, the elements in (5.24) are GL(k)-invariant (and hence Sp(k)-
invariant) in WC(U U ).
Hence, for g ∈ G, x ∈ g′ and z ∈ C[x, η], we have gx.z = xg.z, whence (2).
Theorem 5.30 ((Sp(k), osp(2m|2n))-Howe duality, I). Sp(k) and osp(2m|2n) form
a Howe dual pair on S(Ck ⊗ Cm|n ), and the (Sp(k), osp(2m|2n))-module S(Ck ⊗
Cm|n ) is strongly multiplicity-free.
Proof. Set G = Sp(k). The action of sp(k) on S(Ck ⊗Cm|n ) clearly lifts to an action
of G. By Lemma 5.29, the actions of G and osp(2m|2n) on S(U) = S(Ck ⊗ Cm|n )
commute.
Let U = U ⊕ U ∗ . The associated graded for WC(U U ) is S(U
U ), and the action
of sp(k) on S(U U ) clearly lifts to an action of G. Let {y pi } denote the basis in U0̄∗
dual to the basis {x pi } for U0̄ , and let {ξ pt } denote the basis in U1̄∗ dual to the basis
{η pt } for U1̄ . By the supersymmetric First Fundamental Theorem ( (FFT) for G (see
Theorem 4.19), a generating set T for the algebra S(U U )G = S(U0̄ ⊕U0̄∗ ) ⊗ ∧(U1̄ ⊕
)G
U1̄∗ ) consists of zī j¯, zst , zīs , zsī in (5.12), Iī j¯, Iīs , Ist , and
(k
2 ) (
k
2 )
∑ pi k+1−p, j k+1−p,i p j ,
y y − y y ∑ pi k+1−p,s k+1−p,i ps ,
y ξ − y ξ
p=1 p=1
k
2 ( )
∑ ξ ps ξk+1−p,t − ξk+1−p,s ξ pt ,
p=1
for 1 ≤ i, j ≤ m, 1 ≤ s,t ≤ n. Observe that the symbol map σ sends the basis
elements (5.24) and (5.25) for g′ to the above elements in T for S(U
U )G . Now the
theorem follows from Proposition 5.12 and Lemma 5.29.
δ1 − δ2 ⃝
@ ⊗
@
⃝ ··· ⃝ ⃝ ··· ⃝
δ2 − δ3 δm − ε1 εn−1 − εn
−δ1 − δ2 ⃝
The root vectors associated to the simple roots −δ1 − δ2 , δi − δi+1 (1 ≤ i ≤
m − 1), δm − ε1 , εt − εt+1 (1 ≤ t ≤ n − 1), in the notations of (5.24) and (5.25), are
respectively
(5.26) ∆1̄2̄ , Eī,i+1 (1 ≤ i ≤ m − 1), Em1 , Et,t+1 (1 ≤ t ≤ n − 1).
Then gl(m|n) is a Levi subalgebra of osp(2m|2n) corresponding to the removal of
the simple root −δ1 − δ2 . We have a triangular decomposition
(5.27) osp(2m|2n) = u− ⊕ gl(m|n) ⊕ u+ .
In particular, the Lie superalgebras gl(m|n) and osp(2m|2n) share the same Cartan
subalgebra. An element f ∈ C[x, η] is called harmonic, if f is annihilated by the
+
subalgebra u+ . The space of harmonics C[x, η]u will be denoted by Sp H.
Recall that P(m|n) denotes the set of all (m|n)-hook partitions λ (i.e., λm+1 ≤
n), and also recall from (2.4) that, for an (m|n)-hook partition λ, λ♮ is a weight
for gl(m|n). Then λ♮ + ℓ1m|n can be regarded as a weight for the Lie superalgebras
gl(m|n) and osp(2m|2n), where we recall the definition of 1m|n from (1.29).
Theorem 5.31 ((Sp(k), osp(2m|2n))-Howe duality, II). Let k = 2ℓ. We have the
following decomposition of (Sp(k), osp(2m|2n))-modules:
⊕ ( )
S(Ck ⊗ Cm|n ) = L(Sp(k), λ) ⊗ L osp(2m|2n), λ♮ + ℓ1m|n .
λ∈P(m|n),ℓ(λ)≤ℓ
+
Proof. Observing that M u is a gl(m|n)-module for any osp(2m|2n)-module M
by the triangular decomposition (5.27), we may regard the space of harmonics
Sp H as an (Sp(k), gl(m|n))-module. Furthermore, since each graded subspace
in the theorem, it suffices (and is indeed equivalent) to establish the following ex-
plicit decomposition of Sp H as an (Sp(k), gl(m|n))-module:
⊕
(5.29) Sp
H∼
= L(Sp(k), λ) ⊗ Lm|n (λ♮ + ℓ1m|n ).
λ∈P(m|n),ℓ(λ)≤ℓ
We first consider the limit case n = ∞ with the space of harmonics denoted by
Sp H∞ , where the condition λ ∈ P(m|n) reduces to λ ∈ P. Set ♢ = ♢1,2,...,λ1 ,λ′ . Ob-
serve by Theorem 5.23 that the vector ♢ associated to any partition λ with ℓ(λ) ≤ ℓ
is a joint sp(k) × gl(m|n)-highest weight vector and ♢ a polynomial independent of
the variables x pi and η pt for p ≥ ℓ + 1. Each ∆-operator in (5.25) is a second order
differential operator whose summands always involve differentiation with respect
to one of these variables for p ≥ ℓ + 1. Thus, ♢ is annihilated by u+ , and hence we
have ♢ ∈ Sp H∞ . One also easily checks that the vector ♢ has weight (λ, λ♮ + ℓ1m|n )
under the transformations of sp(k) × osp(2m|2n). We note that all the summands
on the right hand side of (5.29) indeed occur in Sp H∞ (with multiplicity one), and
all irreducible representations of Sp(k) occur. We conclude that (5.29) must hold
(when n = ∞).
Now consider the finite n case. We may regard S(Cd ⊗ Cm|n ) ⊆ S(Cd ⊗ Cm|∞ )
with compatible actions of osp(2m|2n) ⊆ osp(2m|2∞). We claim the following
equation on spaces of joint highest weight vectors holds:
(Sp H)n1 ×n(n) = (Sp H∞ )n1 ×n(∞) ∩ C[x, η],
where n1 (respectively, n(n)) denotes the nilradical of the standard Borel for sp(k)
(respectively, the nilradical generated by (5.26) for osp(2m|2n)). Note by definition
that (Sp H)n1 ×n(n) ⊇ (Sp H∞ )n1 ×n(∞) ∩ C[x, η]. Since each summand of the operators
in (5.24) and the ∆-operators in (5.25) that lie in osp(2m|2∞)\osp(2m|2n) must in-
volve ∂η∂ps for s > n, we conclude that (Sp H)n1 ×n(n) ⊆ (Sp H∞ )n1 ×n(∞) ∩ C[x, η], and
hence the claim is proved. Observe from the explicit formulas of the joint highest
vectors in Sp H∞ (see Theorem 5.23) that precisely those vectors corresponding to
(m|n)-hook partitions will not involve the variables η ps for s > n, and hence they
lie in (Sp H∞ )n1 ×n(∞) ∩ C[x, η]. This together with the claim above completes the
proof of the theorem.
Corollary 5.33 ((Sp(k), sp(2n))-Howe duality). Let k = 2ℓ. We have the following
decomposition of (Sp(k), sp(2n))-modules:
⊕ ( n )
∧(Ck ⊗ Cn ) ∼
= L(Sp(k), λ) ⊗ L sp(2n), ∑ (λ′i − ℓ)εi .
λ∈P i=1
ℓ(λ)≤ℓ,ℓ(λ′ )≤n
λ′1 by k − λ′1 . This way we have obtained a complete parametrization of the simple
O(k)-modules in terms of partitions µ with ℓ(µ) ≤ ℓ and µ′1 + µ′2 ≤ k = 2ℓ + 1.
Now consider the case when k = 2ℓ is even.
We shall denote the simple O(k)-module in Proposition 5.35 (i) by L(O(k), λ).
Let τ ∈ O(k) \ SO(k) denote the element that switches the basis vector eℓ with
eℓ+1 and fixes all other standard basis vectors ei ’s of Ck . We declare L(O(k), λ) to
be the O(k)-module in Proposition 5.35(ii) on which τ acts on an SO(k)-highest
weight vector trivially. Note that τ transforms an SO(k)-highest weight vector in
the O(k)-module L(O(k), λ) ⊗ det by −1. We shall denote the simple O(k)-module
e where as before the Young diagram of the partition
L(O(k), λ) ⊗ det by L(O(k), λ),
eλ is obtained from λ by replacing the first column λ′ by k − λ′ . In this way, we
1 1
have obtained a complete parametrization of the simple O(k)-modules in terms of
partitions µ with ℓ(µ) ≤ ℓ such that µ′1 + µ′2 ≤ k = 2ℓ.
Summarizing the above discussion for k odd and even, we have obtained the
following uniform description.
5.3.4. Howe dual pair (O(k), spo(2m|2n)). We consider the superspace U = U0̄ ⊕
U1̄ with U0̄ = V m and U1̄ = V n , and identify naturally U ≡ V ⊗ Cm|n as before. We
continue the identification S(U) ≡ C[x, η] as in (5.9).
The action of the Lie algebra so(k) as the subalgebra of gl(k) on C[x, η] given
in (5.10) lifts to an action of Lie group O(V ) = O(k). On the other hand, recall
the action of gl(m|n) on C[x, η] given by (5.24). Introduce the following additional
5.3. Howe duality for symplectic and orthogonal groups 165
Let us denote by g′ the linear span of the elements in (5.24) and (5.32). Note that
the elements in (5.24) and (5.32) form a linear basis for g′ .
Theorem 5.38 ((O(k), spo(2m|2n))-Howe duality, I). O(k) and spo(2m|2n) form a
Howe dual pair on S(Ck ⊗Cm|n ), and the (O(k), spo(2m|2n))-module S(Ck ⊗Cm|n )
is strongly multiplicity-free.
Proof. Set G = O(k). The action of so(k) on S(Ck ⊗ Cm|n ) lifts to an action of
G. By Lemma 5.37, the actions of G and spo(2m|2n) on S(U) = S(Ck ⊗ Cm|n )
commute.
Let U = U ⊕U ∗ so that the associated graded for WC(U U ) is S(UU ). Thus, the
action of so(k) on WC(U U ) lifts to an action of G. Let {y pi } denote the basis in U0̄∗
dual to the basis {x pi } for U0̄ , and let {ξ pt } denote the basis in U1̄∗ dual to the basis
{η pt } for U1̄ . By the supersymmetric First Fundamental Theorem ( (FFT) for G (see
Theorem 4.19), a generating set T for the algebra S(U U )G = S(U0̄ ⊕U0̄∗ ) ⊗ ∧(U1̄ ⊕
166 5. Howe duality
)G
U1̄∗ ) consists of zī j¯, zst , zīs , zsī in (5.12), Jī j¯, Jīs , Jst , and
k k k
∑ y pi yk+1−p, j , ∑ y pi ξk+1−p,s , ∑ ξ ps ξk+1−p,t ,
p=1 p=1 p=1
for 1 ≤ i, j ≤ m, 1 ≤ s,t ≤ n. The symbol map σ sends the basis elements (5.24)
and (5.32) for g′ to the above elements in T, and hence the theorem follows from
Proposition 5.12 and Lemma 5.37.
+ +
Proof. Denote the space of harmonics C[x, η]u by O H. Then M u is a gl(m|n)-
module for any spo(2m|2n)-module M by the triangular decomposition (5.34), so
that O H is an (O(k), gl(m|n))-module. Similar as in the proof of Theorem 5.31 it
is enough to establish the following decomposition of O H as an (O(k), gl(m|n))-
module:
⊕ ( k )
(5.35) O
H= ∼ L(O(k), λ) ⊗ Lm|n λ♮ + 1m|n .
λ∈P(m|n)
2
λ′1 +λ′2 ≤k
5.3. Howe duality for symplectic and orthogonal groups 167
The validity of (5.35) for finite n follows from the case when n = ∞ in a completely
parallel way as in the proof of Theorem 5.31. Hence it remains to establish (5.35)
in the limit case n = ∞, where the condition λ ∈ P(m|n) reduces to λ ∈ P.
Assume now n = ∞. Set ♢ = ♢1,2,...,λ1 ,λ′ , for λ ∈ P with λ′1 + λ′2 ≤ k. We
claim that ♢ ∈ O H, i.e, ♢ is annihilated by u+ . By Theorem 5.23 ♢ is a joint
so(k) × gl(m|n)-highest weight vector. To establish the claim, it suffices to show
that ♢ is annihilated by the simple root vector ∇1̄1̄ , since ♢ is clearly annihilated
by the remaining simple root vectors in (5.33) which lie in gl(m|n).
To show that ∇1̄1̄ (♢) = 0, we recall that ∇1̄1̄ = ∑kp=1 ∂x∂p1 ∂xk+1−p,1
∂
and recall
from (5.18) that ♢ = ♢1,λ′1 · · · ♢λ1 ,λ′λ . We observe that each summand in the expan-
1
∂♢i,λ′ ∂♢ j,λ′
sion of the double differentiation ∇1̄1̄ (♢) is either of the form (i) i
∂x p1 ∂xk+1−p,1
j
···
∂2 ♢i,λ′
for i ̸= j, or of the form (ii) ∂x p1 ∂xk+1−p,1 · · · . Since λ′1 + λ′2 ≤ k by assumption,
i
we have λ′i + λ′j ≤ k for i ̸= j. Noting by definition (5.16) and (5.17) that the x pq
appearing in ♢i,λ′i satisfy the constraints p ≤ i, the derivative in (i) must be zero.
Expanding the determinant ♢i,λ′i along the first column, we obtain ♢i,λ′i = ∑ p x p1 A p ,
where A p is an expression not containing any xq1 . Therefore, the derivative in (ii)
must also be zero, and hence ∇1̄1̄ (♢) = 0.
When k = 2ℓ, recall the element τ ∈ O(k) \ SO(k) defined in Section 5.3.3.
When k = 2ℓ + 1, we let τ be the linear map on Ck that fixes e p , for all p ̸= ℓ + 1,
and that sends eℓ+1 to −eℓ+1 . By examining how τ transforms the so(k)-highest
weight vector ♢, for λ′1 + λ′2 ≤ k, one concludes that the O(k)-module generated
by ♢ is L(O(k), λ). Another direct computation implies that ♢ has spo(2m|2n)-
weight λ♮ + 2k 1m|n . Thus, all irreducible representations of O(k) occur, and all the
summands on the right hand side of (5.35) occur in the space of harmonics O H.
Therefore, (5.35) holds in the case when n = ∞. The theorem is proved.
uncompleted variants are preferred as we need to compare with finite rank Lie
algebras. The main results in this section make sense when formulated for these
completed Lie algebras.
We extend the Z-gradation of the Lie algebra gl∞ to a∞ by putting the degree
of K to be 0. In particular, we have a triangular decomposition
a∞ = a∞+ ⊕ a∞0 ⊕ a∞− ,
where ⊕
a∞± = gl∞,± j , a∞0 = gl∞,0 ⊕ CK.
j∈N
5.4.2. The fermionic Fock space. Recall from (A.62) that F denotes the fermionic
Fock space generated by the fermions ψ± (z), whose components ψ± n satisfy the
Clifford commutation relation (A.60).
We shall denote by Fℓ the fermionic Fock space of ℓ pairs of fermions
∑ ϕr z−r− 2
1
ϕ(z) =
r∈ 21 +Z
1
Every graded subspace of Fℓ and Fℓ+ 2 with respect to the principal gradation is
finite dimensional.
Introduce the normal ordered product (denoted by : :)
172 5. Howe duality
{ −,q
−,q −ψs ψ+,p
r , if s = −r < 0,
r ψs :
:ψ+,p = +,p −,q
ψr ψs , otherwise,
(5.40) {
−ψs ψ−,p
+,q
r , if s = −r < 0,
:ψ−,p
r ψs :
+,q
= −,p +,q
ψr ψs , otherwise.
+,q −,p −,p
+,q ±,p −,q
±,p −,q
Also let :ψ+,p +,p q q
r ψs : = ψr ψs , :ψr ψs : = ψr ψs , :ψr ϕs : = ψr ϕs ,
for all p, q, r, s.
Equivalently, we have
ℓ
(5.41) Ei j = ∑ :ψ+,p−i ψ−,p
1
2j−
:. 1
2
p=1
Such generating functions are “vertex operators” in the theory of vertex algebras,
pq
but we will avoid such terminology in this book. We will only need e∗ (0) below,
pq pq
and hence introduce a short-hand notation E∗ ≡ e∗ (0). It follows that
(5.42) E∗pq = ∑ :ψ+,p −,q
−r ψr :, p, q = 1, . . . , ℓ.
r∈ 12 +Z
(ℓ )
+,p −,p +,p −,p +,p −,p +,p −,p
= ∑ −i j− − j i−
ψ 1 ψ 1 ψ 1 ψ
2
1 − ψ 1 ψ
− j i− 1 ψ
2
1 ψ
−i j−
2
1
2 2 2 2 2
p=1
(ℓ )
+,p −,p +,p −,p
= ∑ 1 −i i− 1 1 − j j− 1
ψ ψ − ψ
2
ψ
2 2 2
p=1
(ℓ ) ( )
+,p −,p +,p −,p
= ∑ 1 −i i− 1
:ψ ψ : − :ψ 1 ψ
− j j−
2
1 : + θ{i ≤
2
0} − θ{ j ≤ 0}
2
ℓI
2
p=1
( )
= Eii − E j j + θ{i ≤ 0} − θ{ j ≤ 0} ℓI.
pv pq uq
(2) We need to check that [E∗ , Euv
∗ ] = δqu E∗ − δ pv E∗ for 1 ≤ p, q, u, v ≤ ℓ.
Indeed, it follows by (5.42) that
[ +,p −,q ]
∗ ]= ∑
−,v
[E∗pq , Euv :ψ−r ψr :, :ψ+,u
−r ψr :
r∈ 12 +Z
( +,u −,q )
= ∑ δqu :ψ+,p −,v
−r ψr : − δvp :ψ−r ψr : = δqu E∗ − δ pv E∗ .
pv uq
r∈ 12 +Z
pq
As the operators E∗ have degree zero in the principal gradation, the action of gl(ℓ)
preserves each (principally) graded subspace of Fℓ , which is finite dimensional
and of integral weight. Hence the action of gl(ℓ) lifts to an action of the Lie group
GL(ℓ) on Fℓ .
(3) It follows by (5.41) and (5.42) that
[ ℓ ]
+,u −,u
[E∗pq , Ei j ] = ∑ −r ψr :, ∑ :ψ 1 −i ψ j− 1 :
:ψ+,p −,q
2 2
r∈ 12 +Z u=1
[ ] [ ]
−,q +,q −,q +,p −,q +,p −,p
= :ψ+,p
1 ψ
−i i− 1
:, :ψ 1 ψ
−i j− 1
: + :ψ 1 ψ
− j j− 1
:, :ψ 1 ψ
−i j− 1
:
2 2 2 2 2 2 2 2
−,q −,q
= :ψ+,p
1
−i
ψ j− 1 : − :ψ+,p
1
−i
ψ j− 1 : = 0.
2 2 2 2
whose weights with respect to gl(ℓ) and a∞ are λ and Λa (λ), respectively.
Proof. The vector vaλ in (5.44) is indeed a highest weight vector for gl(ℓ) and a∞ ,
respectively, since applying any positive root vector in either gl(ℓ) or a∞ to vaλ is
either manifestly zero, or gives rise to two identical ψ∗∗
∗ in the resulting monomial,
5.4. Howe duality for infinite-dimensional Lie algebras 175
whence also zero. Another direct calculation shows that the highest weight of vaλ
for gl(ℓ) is (λ1 , . . . , λℓ ).
We recall that Emm = ∑ℓp=1 : ψ+,p ψ−,p1 :. When m ≥ 1 and n ≥ 1 one easily
2 −m m− 2
1
computes
+,q +,q −,q
[Emm , ψ−n+ 1 ] = δm,n ψ−n+ 1 , [Emm , ψ−n+ 1 ] = 0.
2 2 2
This implies that the a∞ -weight of the vector vaλ equals ℓΛa0 + ∑i λ′i εi , which is also
equal to Λaλ1 + . . . + Λaλℓ by (5.38).
r∈ 12 +Z
pq qp
Note that Ẽ pq = Ẽqp and Ẽ∗∗ = Ẽ∗∗ .
Lemma 5.46. Let k = 2ℓ.
(1) The formula (5.45) defines an action of Lie algebra c∞ on Fℓ of level ℓ.
pq pq
(2) The operators Ẽ pq and Ẽ∗∗ in (5.46) together with E∗ in (5.42), for 1 ≤
p, q ≤ ℓ, define an action of the Lie algebra sp(k) on Fℓ , which lifts to an
action of Sp(k).
176 5. Howe duality
Proof. (1) This follows from Lemma 5.43(1), (5.45), and the definition of c∞ .
(2) Let us denote a standard basis for V = Cℓ ⊕ Cℓ∗ by v±,p , for p = 1, . . . , ℓ,
and denote a standard basis for C∞ by wr with r ∈ (− 21 − Z+ ). Letting U = V ⊗C∞ ,
we naturally identify Fℓ with ∧(U) by setting ψ±,p r ≡ v±,p ⊗ wr . Note that V is a
symplectic space with a natural pairing, and so is U. Hence, we have a natural
action of sp(V ) = sp(k) on Fℓ . The formula (5.46) is simply a precise way of
writing down this action in terms of coordinates. Indeed, comparing these formulas
with (5.23) we can see easily that the matrices there with B = C = 0 correspond
pq pq
to the E∗ ’s, while the ones with A = C = 0 correspond to the Ẽ∗∗ ’s, and the ones
with A = B = 0 correspond to the Ẽ ’s. Alternatively, it can be verified by a direct
pq
computation that the operators in (5.46) generate sp(k). Note that the action of
sp(k) preserves the principal gradation of Fℓ and every graded subspace of Fℓ is
finite dimensional. Thus, the action of sp(k) lifts to that of Sp(k).
pq
(3) As part of the proof of Lemma 5.43, we see that E∗ commutes with the
action of c∞ . We will now check that [Ẽuv , c∞ ] = 0, and leave the similar verification
that [Ẽuv
∗∗ , c∞ ] = 0 to the reader. Indeed, we have that
[ uv ]
Ẽ ,Ei, j − (−1)i+ j E1− j,1−i
[ ℓ ]
+,p −,p i+ j +,p −,p
= ∑ (−1)−r− 2 :ψ−u −v
∑ −i j−
1
ψ
−r r :, (:ψ 1 ψ 1 : − (−1) :ψ j− 1 ψ 1
−i
:)
2 2 2 2
r∈ 12 +Z p=1
[
= (−1)i ψ−v ψ−u1 + (−1)i ψ−u ψ−v1 + (−1) j ψ−u ψ−v + (−1) j ψ−v ψ−u ,
2 −i i− 2 2 −i i− 2 2−j 2−j
1 1 1
j− 12 1
j− 12
]
−,u +,v −,v i+ j +,u −,u i+ j +,v −,v
ψ+,u
1 ψ
−i j− 1 +ψ1 ψ
−i j− 1 − (−1) ψ j− 1 ψ1
−i
− (−1) ψ j− 1 ψ1
−i
2 2 2 2 2 2 2 2
=(−1)i ψ−v
1 ψ−u + (−1)i ψ−u
−i j− 1 1 ψ−v − (−1)i ψ−u
−i j− 1 1 ψ−v − (−1)i ψ−v
−i j− 1 1 ψ−u = 0.
−i j− 1
2 2 2 2 2 2 2 2
where Λc (λ) := ℓΛc0 + ∑i≥1 λ′i εi = Λcλ1 + . . . + Λcλℓ . The joint highest weight vector
in Fℓ associated to λ with respect to the standard Borel for sp(k) × c∞ is vcλ :=
Ξ+,1 +,2 +,ℓ
λ1 Ξλ2 · · · Ξλℓ |0⟩.
5.4. Howe duality for infinite-dimensional Lie algebras 177
Proof. We observe that vcλ has sp(k)-weight λ, and c∞ -weight ℓΛc0 + ∑i≥1 λ′i εi , the
pq
latter of which by (5.39) equals Λcλ1 + . . . + Λcλℓ . Also Ẽ∗∗ annihilates vcλ , since in
both expressions only ψ+,p
pq
∗ ’s are involved and Ẽ∗∗ |0⟩ = 0. Since vλ is known to
c
be a joint (gl(ℓ), a∞ )-highest weight vector by Proposition 5.45, we see that vcλ is a
joint (sp(k), c∞ )-highest weight vector.
By Lemma 5.46 and Proposition 5.12, Fℓ is strongly multiplicity-free as an
(Sp(k), c∞ )-module. Since vcλ is a non-zero vector in Fℓ of sp(k)-weight λ, for every
λ ∈ P(ℓ), we also observe that all finite-dimensional Sp(k)-module appears in the
decomposition of Fℓ . So the multiplicity-free (Sp(k), c∞ )-module decomposition
of Fℓ follows.
pp
The character of the Sp(k)-module Fℓ is the trace of the operator ∏ℓp=1 zEp∗ on
E −E
Fℓ , while that of the c∞ -module Fℓ is the trace of the operator ∏i∈N yi ii 1−i,1−i on
E −E pp
Fℓ . Computing the trace of ∏i∈N yi ii 1−i,1−i ∏ℓp=1 zEp∗ on both sides of the isomor-
phism in Theorem 5.47, we obtain the following character identity:
∞ ℓ
(5.47) p )= ∑
∏ ∏ (1 + yi z p )(1 + yi z−1 ch L(Sp(k), λ) ch L(c∞ , Λc (λ)).
i=1 p=1 λ∈P(ℓ)
r∈ 12 +Z r∈ 12 +Z
pq qp
Note that Ě pq = −Ěqp and Ě∗∗ = −Ě∗∗ .
Lemma 5.48. (1) The formula (5.48) defines an action of the Lie algebra d∞
on Fℓ of level ℓ.
178 5. Howe duality
pq pq
(2) The operators Ě pq , Ě∗∗ in (5.49) together with E∗ in (5.42), for 1 ≤ p, q ≤
ℓ, define an action of so(2ℓ) on Fℓ , which lifts to an action of O(2ℓ).
(3) O(2ℓ) and d∞ form a Howe dual pair on Fℓ .
Proof. The proof is completely analogous to that of Lemma 5.46 for type C. For
example, (2) can be easily obtained by comparing the formulas in (5.49) and (5.42)
with the one in (5.30), similar to the type C case. We leave the details to the
reader.
1
Let k = 2ℓ + 1. We introduce the following operators on Fℓ+ 2 :
ℓ ( )
−,p +,p −,p
(5.50) Ei, j − E1− j,1−i = ∑ :ψ+,p
1 ψ
−i j− 1 : − :ψ
j− 1 ψ1
−i
: + :ϕ p1 −i ϕ pj− 1 :.
2 2 2 2 2 2
p=1
Equivalently, we have
∑ (Ei, j − E1− j,1−i ) zi−1 w− j
i, j∈Z
= ∑ℓp=1 (:ψ+,p (z)ψ−,p (w): − :ψ+,p (w)ψ−,p (z):) + :ϕ(z)ϕ(w):.
Introducing the short-hand notations Ě∗p = e∗p (0) and Ě p = e p (0), we have the fol-
lowing formulas:
(5.51) Ě p = ∑ :ψ−,p
−r ϕr :, Ě∗p = ∑ :ψ+,p
−r ϕr :.
r∈ 21 +Z r∈ 21 +Z
pq pq
(2) The operators Ě∗∗ , Ě pq in (5.49), Ě∗p , Ě p in (5.51), together with E∗ in
(5.42), for 1 ≤ p, q ≤ ℓ, define an action of the Lie algebra so(2ℓ + 1) on
1
Fℓ+ 2 , which lifts to an action of O(2ℓ + 1).
1
(3) O(2ℓ + 1) and d∞ form a Howe dual pair when acting on Fℓ+ 2 .
Proof. The proof is completely analogous to that of Lemma 5.46 in type C, and
will be omitted.
5.5.1. Characters for modules of Lie algebras c∞ and d∞ . For x ∈ {c, d}, let x∞0
be the Cartan subalgebra and W be the Weyl group of the Lie algebras x∞ defined
in Section 5.4. Let l be the Levi subalgebra of x∞ corresponding to the removal of
the simple root β× (as marked in the Dynkin diagrams in Section 5.4.1), and let
W0 denote the Weyl group of l. Let u+ and u− be respectively the nilradical and
opposite nilradical associated to l so that we have x∞ = u− ⊕ l ⊕ u+ . Let Wr0 be
the set of the minimal length representatives of the right cosets W0 \W of length r
for x∞ . It is well known that the Weyl group W can be written as W = W0W 0 with
⊔
W 0 = r≥0 Wr0 . For µ ∈ (x∞0 )∗ and w ∈ W we set
w ◦ µ := w(µ + ρx ) − ρx ,
where ρx ∈ (x∞0 )∗ is determined by ⟨ρx , Hix ⟩ = 1, for every simple coroot Hix .
Let λ be as in Theorem 5.47 in case of c∞ and as in Theorem 5.50 in case
of d∞ . Since ⟨Λx (λ), H xj ⟩ ∈ Z+ , for every j, it follows that ⟨w ◦ Λx (λ), Hix ⟩ ∈ Z+ ,
180 5. Howe duality
for w ∈ W 0 and Hix in the Levi subalgebra l. In our setting, noting that W0 is the
group of (finite) permutations of N, we may find explicitly a partition λw = λxw =
((λw )1 , (λw )2 , . . .) such that w ◦ Λx (λ) can be written as
{
kΛx + ∑ j>0 (λw ) j ε j , if x = d,
(5.53) w ◦ Λx (λ) = k 0x
2 Λ0 + ∑ j>0 (λw ) j ε j , if x = c.
Recall that sµ denotes the corresponding Schur function associated to a partition µ.
Proposition 5.51. Let λ be as in Theorem 5.47 in case of c∞ and as in Theorem 5.50
in case of d∞ . We have the following character formula:
1 ∞
ch L(x∞ , Λx (λ)) = ∑ (−1)r ∑ 0 sλw (y1 , y2 , . . .),
Dx r=0
for x = c, d,
w∈W r
Proof. For any such λ, set Λ = Λx (λ). The module L(x∞ , Λ) is integrable and hence
affords the Weyl-Kac character formula
eσ(Λ+ρx )−ρx
(5.54) ch L(x∞ , Λ) = ∑ (−1)ℓ(σ) ∏α∈Φ+ (1 − e−α )
,
σ∈W
where Φ+ denotes the set of positive roots. Let ρl ∈ (x∞0 )∗ be the element deter-
mined by ρl (H j ) = 1, for all j ∈ N, and ρl (H0x ) = 0. Set ρu := ρx − ρl . Then for
τ ∈ W0 we have τ(ρu ) = ρu . Let Φ+ = Φ+ l ⊔ Φu , where Φl and Φu are the subsets
+ + +
1 ∞
= x ∑ (−1)r ∑ sλw (y1 , y2 , . . .).
D r=0 w∈W 0 r
In the last identity we have used the fact that l is of type A+∞ , and so the irreducible
l-character of integrable highest weight w ◦ Λ is simply the Schur function sλw .
5.5. Character formula for Lie superalgebras 181
5.5.2. Characters for oscillator osp(2m|2n)-modules. Recall hsλ denotes the su-
per Schur function defined in Appendix (A.35). Below, by the character of a mod-
ule of Sp(k) or of osp(2m|2n), we mean the trace of the operator ∏1≤p≤ℓ zEp −E
pp k+1−p.k+1−p
E E
or ∏1≤i≤m ∏1≤ j≤n xi īī y j j j , respectively. We have the following character formula
for the irreducible oscillator osp(2m|2n)-modules appearing in the Howe duality
decomposition in Theorem 5.31.
E E
Proof. Computing the trace of the operator ∏ℓp=1 zEp −E
pp k+1−p.k+1−p
∏m n
i=1 ∏ j=1 xi y j
īī jj
Finally, on the resulting identity we set x j = 0 for j ≥ m + 1 and multiply both sides
( )ℓ
by xy11···x
···yn
m
. In this way, we have obtained the following identity which shares the
same left-hand side as (5.55):
( )
x1 · · · xm ℓ ℓ m n (1 + y j z−1 p )(1 + y j z p )
∏ ∏ ∏ = ∑ ch L(Sp(2ℓ), λ)×
y1 · · · yn p=1 i=1 j=1 (1 − xi z−1
p )(1 − xi z p ) λ∈P(ℓ)
λm+1 ≤n
182 5. Howe duality
( )ℓ
x1 ···xm
y1 ···yn ∏1≤i≤m (1 + xi ys ) ∑∞
r=0 (−1) ∑w∈Wr0 hsλw (y; x)
r
1≤s≤n
.
∏1≤i< j≤m ∏1≤s≤t≤n (1 − xi x j )(1 − ys yt )
The theorem follows by comparing this identity with (5.55) and noting the linear
independence of the characters ch L(Sp(2ℓ), λ).
both sides of the isomorphism in Theorem 5.39, we obtain the following character
identity:
( )k
x1 · · · xm 2 ℓ m n (1 + εy j z−1 p )(1 + εy j z p )(1 + εy j )
y1 · · · yn ∏ ∏ ∏ −1
p=1 i=1 j=1 (1 − εxi z p )(1 − εxi z p )(1 − εxi )
(5.56) ( k )
= ∑ ch L(O(k), λ) ch L spo(2m|2n), λ♮ + 1m|n .
λ′ +λ′ ≤k
2
1 2
λm+1 ≤n
5.5. Character formula for Lie superalgebras 183
( ) 2k
x1 · · · xm ℓ (1 + εy j z−1
p )(1 + εy j z p )(1 + εy j )
y1 · · · yn ∏ ∏ (1 − εx z−1
p )(1 − εx z
p=1 i, j i i p )(1 − εxi )
= ∑ ch L(O(k), λ)×
λ′1 +λ′2 ≤k
λm+1 ≤n
( ) 2k ∏1≤i≤m (1 + xi ys ) ∑∞
r=0 (−1) ∑w∈Wr0 hsλw (y; x)
r
x1 · · · xm 1≤s≤n
.
y1 · · · yn ∏1≤i≤ j≤m ∏1≤s<t≤n (1 − xi x j )(1 − ys yt )
The theorem follows by comparing this identity with (5.56) and noting that the
(enhanced) characters ch L(O(k), λ) are linearly independent.
We now turn to the case when k = 2ℓ is even. In this case, the trick of intro-
ducing the extra variable ε does not work, and we obtain the following.
ch L(spo(2m|2n), λ♮ + ℓ1m|n ) =
( ) ∏1≤i≤m (1 + xi ys ) · ∑∞
r=0 (−1) ∑w∈Wr0 hsλw (y; x)
r
x1 · · · xm ℓ 1≤s≤n
· .
y1 · · · yn ∏1≤i≤ j≤m ∏1≤s<t≤n (1 − xi x j )(1 − ys yt )
184 5. Howe duality
E E
Proof. Computing the trace of the operator zEp −E
pp k+1−p,k+1−p
∏i, j xi īī y j on both
jj
In the same way as in the proof of Theorem 5.53 by a formal algebraic ma-
nipulation based on (5.52), we obtain the following identity that shares the same
left-hand side as (5.57):
( )
x1 · · · xm ℓ ℓ m n (1 + y j z−1 p )(1 + y j z p )
∏ ∏ ∏ −1
= ∑ ch L(O(k), λ)×
y1 · · · yn p=1 i=1 j=1 (1 − xi z p )(1 − xi z p ) λ′ +λ ′ ≤k
1 2
λm+1 ≤n
( ∞
) ∏1≤i≤m (1 + xi ys ) · ∑r=0 (−1)r ∑w∈W 0 hsλw (y; x)
x1 · · · xm ℓ 1≤s≤n
r
· .
y1 · · · yn ∏1≤i≤ j≤m ∏1≤s<t≤n (1 − xi x j )(1 − ys yt )
The theorem follows by comparing this identity with (5.57) and noting that the
characters of L(O(k), λ) and L(O(k), eλ) coincide.
5.6. Notes
Section 5.1. The materials on Weyl-Clifford algebra and connections to classi-
cal Lie algebras/superalgebras are standard. The general duality theorem, The-
orem 5.8, is taken from Goodman-Wallach [GW, 4.2.1], and it can be regarded
as an abstract generalization of the original formulation (see Theorem 5.10 and
Proposition 5.12) of Howe duality [H1] (Howe’s paper as a preprint was dated in
1976).
Section 5.2. The (GL(k), gl(m))-Howe duality (respectively, its skew version)
in Theorems 5.16 and 5.18 (due to Howe [H1, H2]) offers a representation theoretic
interpretation of the classical Cauchy identity (respectively, its dual version). It is
equivalent to Schur duality as well as the First Fundamental Theorem of invariant
theory for general linear groups.
The Howe dual pair (GL(k), gl(m|n)) (Theorem 5.14) already appeared in
Howe’s classical paper [H1], and the precise multiplicity-free (GL(k), gl(m|n))-
Howe duality decomposition (Theorem 5.19) was obtained later independently by
Sergeev [Sv2] and Cheng-Wang [CW1]. The formula for the joint highest weight
vectors in the (GL(k), gl(m|n))-Howe duality (Theorem 5.23) was due to Cheng-
Wang [CW1], and it unifies the formulas of Howe [H2] in two (non-super) special
cases when m or n is zero; a formula of highest weight vectors for the more general
5.6. Notes 185
(gl(k|ℓ), gl(m|n))-Howe duality was also obtained in loc. cit.. The (q(m), q(n))-
Howe duality was due to Sergeev [Sv2] and Cheng-Wang [CW2] independently,
and it gives a representation theoretic interpretation of the Cauchy identity for
Schur Q-functions.
Section 5.3. The Howe dual pairs (Sp(k), osp(2m|2n)) and (O(k), spo(2m|2n))
on S(Ck ⊗ Cm|n ) (Theorems 5.30 and 5.38) were formulated in Howe [H1]; they
were based on, and in turn can be regarded as a Lie theoretic reformulation of, the
First Fundamental Theorem of invariant theory for classical groups. The orthogo-
nal group O(k) is disconnected. We follow Howe [H2] to present a parametrization
of the simple O(k)-modules in terms of partitions. The strongly multiplicity-free
decompositions in (Sp(k), osp(2m|2n))- and (O(k), spo(2m|2n))-Howe dualities
(Theorems 5.31 and 5.39) were obtained in Cheng-Zhang [CZ2], where the high-
est weight vector formula in Section 5.2 plays a key role in the proofs. The short
proofs presented in this book follow Cheng-Kwon-Wang [CKW]. We easily re-
cover the Howe duality decompositions in the non-super setting, which were due
to Howe [H2] with somewhat different arguments, by setting m or n to zero.
There are additional Howe dualities involving the spin groups (see Howe [H2])
as well as spo(2m|2n + 1) (see Cheng-Kwon-Wang [CKW, Appendix A]). There
is also a type A Howe duality involving infinite-dimensional irreducible modules of
gl(n) or more generally gl(m|n) (see Kashiwara-Vergne [KaV] and Cheng-Lam-
Zhang [CLZ]). We do not treat these cases in the book and refer the reader to the
original papers for details.
Section 5.4. Howe duality between classical groups and infinite-dimensional
Lie algebras was systematically developed by Wang [Wa], and it has been used
in Kac-Wang-Yan [KWY] for the study of representation theory of classical Lie
subalgebras of W1+∞ . The (GL(ℓ), a∞ )-Howe duality (Theorem 5.44) can also be
recovered as a limit case of a duality for affine Lie algebras of type gl given earlier
by I. Frenkel [Fr], and the proof here follows [Wa]. The (Sp(k), c∞ )-Howe duality
and (O(k), d∞ )-Howe duality (Theorems 5.47 and 5.50) are due to Wang [Wa],
where one can find several more Howe dualities in bosonic and fermionic Fock
spaces not covered in the book.
Section 5.5. Irreducible characters for the oscillator modules of Lie superalge-
bras were computed in Cheng-Zhang [CZ2] following the approach of Cheng-Lam
[CL1]. A character formula of Enright [En] played an essential role there. The
simpler and more elementary approach presented in the book bypasses Enright’s
formula, via a comparison with Howe duality involving infinite-dimensional Lie
algebras in Section 5.4, and follows Cheng-Kwon-Wang [CKW].
The reader is referred to Cheng-Kwon-Wang [CKW], where the calculation of
the characters via a comparison of two Howe dualities has been refined to compute
the corresponding u− -homology groups with coefficients in the oscillator modules.
The approach there follows the strategy of Aribaud [Ar] and Cheng-Kwon [CK].
Chapter 6
Super duality
In this chapter, we develop a super duality approach to obtain a complete and con-
ceptual solution of the irreducible character problem in certain parabolic Bernstein-
Gelfand-Gelfand categories for general linear and ortho-symplectic Lie superal-
gebras. These parabolic categories contain all the finite-dimensional irreducible
modules of these Lie superalgebras.
Super duality is an equivalence of categories between parabolic categories for
Lie superalgebras and their Lie algebra counterparts at an infinite-rank limit. A
weak version of such a category equivalence on the Grothendieck group level,
which is established first in an elementary way, already implies a solution of the
irreducible character problem of these Lie superalgebras in terms of Kazhdan-
Lusztig polynomials of classical types. It is further shown that the corresponding
Kostant u-homology groups, or equivalently the Kazhdan-Lusztig-Vogan polyno-
mials, are matched perfectly under super duality. Via the truncation functors which
we introduce, the Kazhdan-Lusztig solution for the irreducible character problem
at infinite-rank limit provides us a complete solution to the irreducible character
problem in the corresponding parabolic BGG categories for finite-dimensional ba-
sic Lie superalgebras.
187
188 6. Super duality
intermediary between them. We present explicit matrix realizations of these Lie su-
peralgebras together with their finite-dimensional counterparts, e
gxn , gxn , and gxn , for
n ∈ N. All these superalgebras implicitly depend on a fixed type x = a, b, b• , c, d.
6.1.1. Head, tail, and master diagrams. For m ∈ Z+ , consider a vector space
with basis {ε−m , . . . , ε−1 } ∪ {εr |r ∈ 12 N}, and a symmetric bilinear form (·|·) given
by
1
(εr |εs ) = (−1)2r δrs ,r, s ∈ {−m, . . . , −1} ∪ N.
2
We introduce the following notation for roots which we shall need shortly:
α× := ε−1 − ε1/2 , α j := ε j − ε j+1 , −m ≤ j ≤ −2,
(6.1) 1
β× := ε−1 − ε1 , αr := εr − εr+1/2 , βr := εr − εr+1 , r ∈ N.
2
We
define
the Tn
tail diagrams to be the following three Dynkin diagrams
,
Tn
, and
e with prescribed fundamental systems denoted by Π(Tn ), Π(Tn ),
Tn
e n ), for n ∈ N:
and Π(T
Tn ⃝ ⃝ ⃝ ··· ⃝ ⃝
β× β1 β2 βn−2 βn−1
⊗
Tn ⃝ ⃝ ··· ⃝ ⃝
α× β1/2 β3/2 βn−5/2 βn−3/2
⊗ ⊗ ⊗ ⊗ ⊗
en
T ···
α× α1/2 α1 αn−1 αn−1/2
Corresponding to these three diagrams we have the Lie superalgebras gl(1 + n),
gl(1|n), and gl(1 + n|n). For n = ∞, we have analogous diagrams that are the
Dynkin diagrams of the corresponding infinite-rank Lie superalgebras.
We shall choose another Dynkin diagram k , called a head diagram, to
connect with one of the three tail diagrams to produce the following three new
Dynkin diagrams, which will be called the master diagrams (n ∈ N ∪ {∞}):
Tn en
(6.2)
k
k
Tn
k
T
The three master diagrams will be denoted by
gn
,
gn
, and
e
gn
respectively,
and the associated Lie superalgebras gn , gn , and e
gn will be introduced subsequently.
We denote the fundamental systems corresponding to the three master diagrams by
(6.3) Πn := Π(k) ⊔ Π(Tn ), Πn := Π(k) ⊔ Π(Tn ), Π e n ).
e n := Π(k) ⊔ Π(T
e
g , Π(T), Π(T), and Π(T),g, g, and e
g, respectively.
6.1. Lie superalgebras of classical types 189
The head diagram
k used in this book is always chosen to be one of the
Dynkin diagrams
kx x we will add the superscript x to
defined below. Accordingly,
x
the general notations in 6.1.1 to write Πn ,
gn , gxn , and so on, when it is needed to
•
b,x b, c, d and m ≥ 1, introduce the Lie (super)algebras
specify the type x. For x = a,
x
k with Dynkin diagrams
k and prescribed fundamental systems denoted by
Π(k ) as follows (see (6.1) for notation of α j ):
x
ka ⃝ ⃝ ⃝ ··· ⃝ ⃝ ⃝
α−m α−m+1 α−3 α−2
kb ⃝⇐=⃝ ⃝ ··· ⃝ ⃝ ⃝
−ε−m α−m α−3 α−2
• y⇐=⃝
kb ⃝ ··· ⃝ ⃝ ⃝
−ε−m α−m α−3 α−2
kc ⃝=⇒⃝ ⃝ ··· ⃝ ⃝ ⃝
−2ε−m α−m α−3 α−2
α−m ⃝
@
@
kd ⃝ ⃝ ··· ⃝ ⃝ ⃝
α−m+1 α−3 α−2
⃝
−ε−m −ε−m+1
6.1.2. The index sets. Let us introduce some notations and conventions for in-
dex sets, which will be needed for defining the Lie superalgebras gx , gx , and e
gx
associated to the master diagrams (6.2).
For m ∈ Z+ , we introduce the following totally ordered set eIm :
3 1 1 3
··· < < 1 < < −1
| < ·{z
· · < −m} < 0 < −m
| · · · < −1} < < 1 < < · · ·
< {z
2 2 2 2
m m
For X = Im , Im , or eI+
m , define
X× := X \ {0}, X+ := X ∩ eI+
m.
vector superspaces together with the index sets for their bases in Table 1 below:
Table 1
×
Superspaces Vm V m Vem Vm× V m Vem× Vm+ V m Vem+
+
×
Index sets Im Im eIm Im eI× Im eI+
+
I×
m m I+
m m
×
Let W be one of the superspaces Vem , Vem× , Vem+ ,Vm ,Vm× ,Vm+ ,V m ,V m or V m , regarded
+
realizations of the Lie superalgebras e ga , ga , and ga , respectively. We summarize
this in the Table 2 below.
6.1. Lie superalgebras of classical types 191
• • •
Superalgebras e
gb gb gb e
gc gc gc
×
Matrix forms spo(Vem ) spo(Vm ) spo(V m ) spo(Vem× ) spo(Vm× ) spo(V m )
192 6. Super duality
c e+
Φ = {±εr − εs |r < s (r, s ∈ eI+
m )} ∪ {−2εi (i ∈ Im )},
+
b•
Φ+ = {±εi − ε j |i < j (i, j ∈ I+
m )} ∪ {−εi , −2εi (i ∈ Im )},
+
c
Φ+ = {±εi − ε j |i < j (i, j ∈ I+
m )} ∪ {−2εi (i ∈ Im )},
+
b• + + +
Φ = {±εr − εs |r < s (r, s ∈ Im )} ∪ {−2εi (−m ≤ i ≤ −1)} ∪ {−εr (r ∈ Im )},
+ +
c
Φ = {±εr − εs |r < s (r, s ∈ Im )} ∪ {−2εi (−m ≤ i ≤ −1)}.
Superalgebras egb gb gb e
gd gd gd
×
Matrix forms osp(Vem ) osp(Vm ) osp(V m ) osp(Vem× ) osp(Vm× ) osp(V m )
d e+
Φ = {±εr − εs |r < s (r, s ∈ eI+
+
m )} ∪ {−2εs (s ∈ I0 )},
b
Φ+ = {±εi − ε j |i < j (i, j ∈ I+
m )} ∪ {−εi (i ∈ Im )},
+
d
Φ+ = {±εi − ε j |i < j (i, j ∈ I+
m )},
+ + + +
b
Φ = {±εr − εs |r < s (r, s ∈ Im )} ∪ {−2εs (s ∈ I0 )} ∪ {−εr (r ∈ Im )},
+ + +
d
Φ = {±εr − εs |r < s (r, s ∈ Im )} ∪ {−2εs (s ∈ I0 )}.
e + , x Φ+ , and x Φ+ for
6.1.4. The case of m = 0. Let x = a, b, b• , c, d. The sets x Φ
m = 0 still make sense, and they are the positive systems for the Lie superalge-
bras e
gx , gx and gx whose Dynkin diagrams and fundamental systems are as follows.
Some of the Dynkin diagrams in the case of m = 0 differ somewhat from the coun-
terparts in the case when m ≥ 1, and this is why we treat the two cases separately.
For m = 0, the corresponding Lie superalgebras are realized similarly as in Sec-
tion 6.1.3, and we shall keep the same notations as before.
⊗ ⊗ ⊗ ⊗ ⊗ ⊗
e
ga ··· ···
α1/2 α1 α3/2 αr αr+1/2 αr+1
⊗ ⊗ ⊗ ⊗ ⊗
y
⇐= ··· ···
e
gb
−ε1/2 α1/2 α1 αr−1/2 αr αr+1/2
• ⊗ ⊗ ⊗ ⊗ ⊗
e
gb ⃝⇐= ··· ···
−ε1/2 α1/2 α1 αr−1/2 αr αr+1/2
⊗
α1/2
@ ⊗ ⊗ ⊗ ⊗ ⊗
@
e
gc ··· ···
⊗ ⊗ ⊗ ⊗ ⊗
e
gd ⃝=⇒ ··· ···
−2ε1/2 α1/2 α1 αr−1/2 αr αr+1/2
ga ⃝ ⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
β1 β2 β3 βn βn+1 βn+2
gb ⃝⇐=⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−ε1 β1 β2 βn−1 βn βn+1
• y
gb ⇐=⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−ε1 β1 β2 βn−1 βn βn+1
194 6. Super duality
gc ⃝=⇒⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−2ε1 β1 β2 βn−1 βn βn+1
β1 ⃝
@
@
gd ⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
β2 β3 βn−1 βn βn+1
−ε1 − ε2 ⃝
ga ⃝ ⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
β1/2 β3/2 β5/2 βr βr+1 βr+2
y
gb ⇐=⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−ε1/2 β1/2 β3/2 βr−1 βr βr+1
•
gb ⃝⇐=⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−ε1/2 β1/2 β3/2 βr−1 βr βr+1
β1/2 ⃝
@
@
gc ⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
β3/2 β5/2 βr−1 βr βr+1
−ε1/2 − ε3/2 ⃝
gd ⃝=⇒⃝ ⃝ ··· ⃝ ⃝ ⃝ ···
−2ε1/2 β1/2 β3/2 βr−1 βr βr+1
Λ0 (K) = 1, Λ0 (Ebr ) = 0,
196 6. Super duality
for all admissible r in each case. The C-span of Λ0 and the εr ’s in the dual of hx
(respectively, h and e hx ) will be denoted by hx∗ (respectively, h and e
x x∗
hx∗ ), where r
lies in I+
+ e+
m (respectively, Im and Im ).
In case x = a it will also be advantageous to set Ebr ≡ Err , for r > 0. For nota-
tional convenience later on, we shall declare Λa0 to be 0.
From now on, we shall adopt a convention of dropping the superscript x.
We
write g, g,
and e
g for gx , gx , and e
gx , with associated Dynkin diagrams
g ,
g
e
g
, and
•
, respectively, where x denotes a fixed type among a, b, b , c, d.
6.2.1. Category of polynomial modules revisited. Recall from 6.1.5 that Ve0,k +
is
the (k|k)-dimensional superspace spanned by vr , for r = 1/2, 1, 3/2, . . . , k − 1/2, k,
for k ∈ N, and Ve0+ = Ve0,∞
+
. We introduce the short-hand notations
el+ = gl(Ve + ), el+ = gl(Ve + ), k ∈ N.
0 k 0,k
where ν satisfies (ν, εr ) = 0, for r > n. (trkn as a functor will be defined and studied
systematically in a general setting in 6.2.5).
Let λ be a (k|k)-hook partition. The character of the irreducible el+ k -module
e + θ
L(lk , λ ) is given by hsλ (x1 , . . . , xk ; y1/2 , y3/2 , . . . , yk−1/2 ) according to Theorem 3.14,
where x j = eε j and y j−1/2 = eε j−1/2 for j ∈ N. Hence the character of the el+ n -module
( + θ )
e
trn L(lk , λ ) is equal to hsλ (x1 , . . . , xn ; y1/2 , y3/2 , . . . , yn−1/2 ) or zero, depending
k
The L(el+ , λθ ), for λ ∈ P, has character given by the super Schur function hsλ .
( )
Lemma 6.1. Let λ, µ ∈ P. Then Exte1l+ L(el+ , λθ ), L(el+ , µθ ) = 0.
By Theorem 3.25, (6.12) splits. Hence, we can find a singular vector w in the el+ n-
∞ e+ e + θ
module trn E so that U(ln )w = L(ln , µ ). Since there are no weight vectors in E
of weight µθ + εi − εi+1/2 , for i ≥ n, w is a singular vector in E with respect to the
standard Borel subalgebra of el+ too.
Now we consider the el+ -submodule L generated by w in E. We claim that
L∼ = L(el+ , µθ ), and so (6.11) splits. For if L is not isomorphic to L(el+ , µθ ), then L
contains L(el+ , λθ ), and in particular contains a highest weight vector v of weight
λθ . This implies that µθ − λθ is a positive integral combination of positive roots of
el+ . By the choice of n, µθ − λθ has to be a positive integral combination of posi-
tive roots of el+ e+ e+ e+ θ
n , and hence v ∈ U(ln )w and U(ln )w ⊇ L(ln , λ ). This contradicts
U(el+ e+ θ
n )w = L(ln , µ ). So (6.11) splits, and the lemma is proved.
The notions of polynomial weights and polynomial modules for el+ can be de-
fined just as in Definition 3.23. As in Proposition 3.24, the irreducible polynomial
el+ -modules are precisely L(el+ , λθ ), for λ ∈ P. The following theorem is an infinite-
rank analogue of Theorem 3.25, and it follows easily from Lemmas 6.1 and 6.2.
6.2.2. Parabolic subalgebras and dominant weights. Recall the Lie superalge-
bras g, g, and e
g from Section 6.1.6, which implicitly depends on a fixed m ∈ Z+ .
We shall in addition fix an arbitrary subset Y0 of Π(k), with the convention that
Y0 = 0/ when m = 0. Let Y , Y , and Ye be the following subsets of Π, Π, and Π,e
respectively:
Y = Y0 ⊔ Π(T) \ {β× }, Y = Y0 ⊔ Π(T) \ {α× }, e \ {α× }.
Ye = Y0 ⊔ Π(T)
As fixing Y0 also fixes the sets Y , Y , and Ye , we will make a convention of suppress-
ing them from notations below.
Let l, l, and el be the standard Levi subalgebras of g, g, and e g corresponding
to the subsets Y , Y , and Ye , respectively. The standard Borel subalgebras of g, g, and
e
g, spanned by the central element K and upper triangular matrices, are denoted by
b, b, and eb, respectively. Let p = l + b, p = l + b, and ep = el + e
b be the corresponding
parabolic subalgebras with nilradicals u, u, and e u and opposite nilradicals u− ,
ū− , and e
u− , respectively.
Let − λ = (λ−m , . . . , λ−1 ) ∈ Cm and let + λ be a partition. Recall also that hα
denotes the coroot for a root α. Associated to a given d ∈ C and a tuple (− λ; + λ)
such that ⟨∑−1
i=−m λi εi , hα ⟩ ∈ Z+ for all α ∈ Y0 ⊆ Π(k), we define the weights
−1
(6.13) λ := dΛ0 + ∑ λi εi + ∑ + λ j ε j ∈ h∗ ,
i=−m j∈N
−1
+ ′ ∗
(6.14) λ♮ := dΛ0 + ∑ λi εi + ∑ λs+ 1 εs
2
∈h ,
i=−m s∈ 21 +Z+
−1
(6.15) λθ := dΛ0 + ∑ λi εi + ∑ θ(+ λ)r εr ∈ e
h∗ .
i=−m r∈ 12 N
For λ ∈ P+ , let L(l, λ) denote the highest weight irreducible l-module of highest
weight λ. We extend L(l, λ) to a p-module by letting u act trivially. Define the
parabolic Verma g-module ∆(λ) as
∆(λ) := Indgp L(l, λ) = U(g) ⊗U(l) L(l, λ),
and denote its unique irreducible quotient g-module by L(λ).
200 6. Super duality
where the summation is over ν ∈ ∑−1 i=−m Cεi + ∑0< j≤n Z+ ε j +CΛ0 . When it is clear
from the context we shall also write trn instead of trkn . Analogously, truncation
ek → O
functors trkn : Ok → On and trkn : O e n are defined. These functors are obviously
exact. (The notation trkn used earlier in 6.2.1 corresponds to an extreme case here
when k = 0.)
Recall from (6.6) and Section 6.1.6 the notation E j and Ebj for basis elements
of the Cartan subalgebras.
Proposition 6.7. Let n < k ≤ ∞ and X = L, ∆.
{
( ) Xn (µ), if ⟨µ, Ebj ⟩ = 0, ∀ j > n,
(i) For µ ∈ Pk we have trn Xk (µ) =
+
0, otherwise.
{
+ ( ) X n (µ), if ⟨µ, Ebj ⟩ = 0, ∀ j > n,
(ii) For µ ∈ Pk we have trn X k (µ) =
0, otherwise.
{
( ) Xen (µ), if ⟨µ, Ebj ⟩ = 0, ∀ j > n,
(iii) For µ ∈ Pek+ we have trn Xek (µ) =
0, otherwise.
Proof. We shall only show (i), as similar arguments prove (ii) and (iii). Since
trkn ◦ trlk = trln , it suffices to show (i) for k = ∞.
First suppose that ⟨µ, Ebn+1 ⟩ > 0. Then every weight ν of L(l, µ) also satisfies
⟨ν, Ebn+1 ⟩ > 0. Recall from the proof of Proposition 6.5 that as an l-module u−
decomposes into a direct sum of irreducibles with highest weights in P+ . Thus,
every weight υ in ∆(µ) = S(u− )⊗L(l, µ) must also satisfy ⟨υ, Ebn+1 ⟩ > 0. Therefore,
trn (∆(µ)) = 0, and hence trn (L(µ)) = 0.
Now suppose that ⟨µ, Ebn+1 ⟩ = 0. Since µ ∈ Pk+ , this implies that ⟨µ, Ebj ⟩ = 0
for all j > n. Let l′ and p′ denote the standard Levi and parabolic subalgebras
of g corresponding to the removal of the vertex βn of the Dynkin diagram of g.
Then l′ ∼ = gn ⊕ gl(W ), where W is the subspace of V0+ spanned by the vectors vi ,
i > n. Consider the parabolic Verma module Indgp′ Ln (µ), where Ln (µ) extends to an
l′ -module by a trivial action of gl(W )-module and is then extended to a p′ -module
in a trivial way. Clearly, L(µ) is the unique irreducible quotient of Indgp′ Ln (µ).
Since trn (Indgp′ Ln (µ)) = Ln (µ) and trn (L(µ)) is a nonzero gn -module (as it contains
a non-zero vector of weight µ), we conclude that trn (L(µ)) = Ln (µ).
To complete the proof, we observe that ∆(µ) ∼ = S(u− ) ⊗ L(l, µ), and that the
gn -module trn (∆(µ)) has highest weight µ with character equals ch ∆n (µ). From
these we conclude that trn (∆(µ)) = ∆n (µ).
parabolic Verma modules, respectively. The main ingredients for studying these
functors are two sequences of odd reflections on the standard Borel subalgebra of
e
g, leading to new Borel subalgebras which are “approximately compatible” with
the standard Borel subalgebras of g and g. As a consequence, the solution of the
irreducible character problem in O via the classical Kazhdan-Lusztig theory also
e
provides a complete solution to the irreducible character problem for O and O.
6.3.1. Two sequences of Borel subalgebras of e g. Let us recall briefly the basics
on odd reflections from Chapter 1, Sections 1.3 and 1.4, which are applied now to
the Lie superalgebra e g. Under the odd reflection with respect to an isotropic odd
simple root α in a fundamental system Π e ′ , the resulting new fundamental system
e α is given by (1.30) as follows:
Π
e α = {β ∈ Π
Π e ′ | (β, α) = 0, β ̸= α} ∪ {β + α | β ∈ Π e ′ , (β, α) ̸= 0} ∪ {−α}.
⊗ ⊗ ⊗ ⊗ ⊗
k ⃝ ⃝
α5/2
···
β× ε1 − ε1/2 ε1/2 − ε2 −α3/2 ε3/2 − ε5/2 α3
⊗ ⊗ ⊗
k ⃝ ⃝ ⃝ ⃝
α5/2
···
β× β1 ε2 − ε1/2 β1/2 β3/2 α3
Then, to the last diagram we apply consecutively the three odd reflections corre-
sponding to α5/2 = ε5/2 − ε3 , α3/2 + α2 + α5/2 = ε3/2 − ε3 , and α1/2 + α1 + α3/2 +
α2 + α5/2 = ε1/2 − ε3 and obtain the following three diagrams:
⊗ ⊗ ⊗ ⊗
k ⃝ ⃝ ⃝ ⃝
α7/2
···
β× β1 ε2 − ε1/2 β1/2 ε3/2 − ε3 −α5/2 β5/2
⊗ ⊗ ⊗ ⊗
k ⃝ ⃝
ε2 − ε1/2 ε1/2 − ε3 ε3 − ε3/2
⃝ ⃝
α7/2
···
β× β1 β3/2 β5/2
⊗ ⊗
k ⃝ ⃝ ⃝
ε3 − ε1/2
⃝ ⃝ ⃝
α7/2
···
β× β1 β2 β1/2 β3/2 β5/2
Continue this way until we finally apply the n odd reflections corresponding to
αn−1/2 , αn−3/2 + αn−1 + αn−1/2 , . . ., ∑2n−1 i=1 αi/2 , which equal εn−1/2 − εn , εn−3/2 −
εn , . . . , ε1/2 − εn , respectively. The resulting new Borel subalgebra for e g in the end
will be denoted by e c
b (n), and the corresponding positive system will be denoted
e
by Φ (n). The corresponding fundamental system, denoted by Π
c e c (n), is listed in
the following Dynkin diagram:
⊗ ⊗ ⊗
(6.18)
k
Tn ⃝ ··· ⃝ ···
εn − ε1/2 β1/2 βn−1/2 αn+1/2 αn+1
⊗
The crucial point here is that the subdiagram to the left of the first in (6.18) is
the Dynkin diagram of gn and the precise detail of the remaining part of (6.18) is
not needed.
On the other hand, starting with the standard Dynkin diagram (6.16) of e g , we
may apply a different sequence of n(n+1)
2 odd reflections associated to the following
206 6. Super duality
n(n+1)
ordered sequence of 2 odd roots:
ε1 − ε3/2 ,
ε2 − ε5/2 , ε1 − ε5/2 ,
(6.19) ε3 − ε7/2 , ε2 − ε7/2 , ε1 − ε7/2 ,
··· ,··· ,··· ,··· ,
εn − εn+1/2 , εn−1 − εn+1/2 , ..., ε2 − εn+1/2 , ε1 − εn+1/2 .
Since the procedure here is completely parallel to the previous one for the sequence
(6.19), we skip the details. The resulting new Borel subalgebra for e g in the end
es
will be denoted by b (n), and the corresponding positive system will be denoted by
e s (n). The corresponding fundamental system, denoted by Π
Φ e s (n), is listed in the
following Dynkin diagram:
⊗ ⊗ ⊗
(6.20) Tn+1
k
⃝ ··· ⃝ ···
εn+1/2 − ε1 β1 βn αn+1 αn+3/2
Again the crucial fact here is that the subdiagram to the left of the odd simple root
εn+1/2 − ε1 in (6.20) is the Dynkin diagram of gn+1 .
+
Recall that Φ+ and Φ denote the standard positive systems of g and g, re-
spectively. Also, b and b denote the corresponding standard Borel subalgebras of
g and g, respectively. By definition, g and g are naturally subalgebras of e
g.
e c (n), and b = e
Proposition 6.8. We have Φ+ ⊂ Φ
+ e s (n), and
bc (n) ∩ g. Also, Φ ⊂ Φ
b=ebs (n) ∩ g.
e c (n) and Π ⊂ Φ
Proof. It suffices to check that Π ⊂ Φ e s (n), where Π and Π are the
standard fundamental systems of g and g, respectively. These inclusions can then
be observed directly from (6.18) and (6.20).
6.3.2. Odd reflections and highest weight modules. Recall the standard Levi
subalgebra el of e
g with nilradical e u− from Section 6.2.2.
u and opposite nilradical e
Lemma 6.9. The sequences of odd reflections (6.17) and (6.19) leave the set of
roots of e u− invariant.
u and the set of roots of e
Proof. The isotropic odd roots used in the sequences of odd reflections (6.17) and
(6.19) are of the form αi + αi+1/2 + . . . + α j for 0 < i < j, hence they are all roots
of el. Since e u− are both invariant under the action of el, the lemma follows by
u and e
applying Lemma 1.26.
Proof. The proof of (2) using the sequence (6.19) is completely parallel to the
proof of (1) which uses (6.17). We shall only prove (1).
We observe that theodd e
reflections (6.17) only affect the tail diagram
T and
leave the head diagram
k unchanged, and the summand
−1
(6.22) λ|k := dΛ0 + ∑ λi εi
i=−m
where ⟨q⟩ := max{q, 0} for q ∈ Z. Part (1) follows as a special case of (6.23) for
k = n. Note that λ[0] = λθ = λ|k + ∑ j≥1 ⟨λ′j − j + 1⟩ε j−1/2 + ∑ j≥1 ⟨λ j − j⟩ε j .
Suppose that k = 1. If ℓ(+ λ) < 1, then + λ = 0 and λθ = λ|k . So in particular
⟨λθ , Eb1/2 + Eb1 ⟩ = 0, and thus by Lemma 1.36, the e bec (1)-highest weight is λ[1] = λθ .
l
If ℓ(+ λ) ≥ 1, then ⟨λθ , Eb1/2 + Eb1 ⟩ > 0, as we recall that
′
λθ = λ|k + + λ1 ε1/2 + (+ λ1 − 1)ε1 + · · · .
Hence, by Lemma 1.36, the e
bec (1)-highest weight after the odd reflection with re-
l
spect to ε1/2 − ε1 is
′
λ[1] = λ|k + + λ1 ε1 + (+ λ1 − 1)ε1/2 + · · · ,
208 6. Super duality
Now we apply the next (k + 1) odd reflection in the sequence (6.17) consecutively.
As ⟨λ[k] , Ebk+1/2 + Ebk+1 ⟩ > 0, by Lemma 1.36 we calculate the new weight after the
odd reflection with respect to εk+1/2 − εk+1 to be
k k
′ ′
λ[k,1] = λ|k + ∑ + λi εi + ∑ (+ λi − k)εi−1/2 + (+ λk+1 − k − 1)εk+1/2
i=1 i=1
′
+ ( λk+1 − k)εk+1 +
+
∑ ⟨+ λ j − j + 1⟩ε j−1/2 + ∑ ⟨+ λ j − j⟩ε j .
j≥k+2 j≥k+2
Now ⟨λ[k,1] , Ebk−1/2 + Ebk+1 ⟩ > 0, so by Lemma 1.36 we calculate the new weight
after the odd reflection with respect to εk−1/2 − εk+1 to be
k k−1
′ ′
λ[k,2] =λ|k + ∑ + λi εi + ∑ (+ λi − k)εi−1/2 + (+ λk − k − 1)εk−1/2
i=1 i=1
+ ′
+ ( λk+1 − k − 1)εk+1/2 + (+ λk+1 − k + 1)εk+1
′
+ ∑ ⟨+ λ j − j + 1⟩ε j−1/2 + ⟨+ λ j − j⟩ε j . ∑
j≥k+2 j≥k+2
Continuing this way, after a total of (k + 1) odd reflections we end up with the
weight
k+1 k+1
′
λ[k,k+1] =λ|k + ∑ + λi εi + ∑ (+ λi − k − 1)εi−1/2
i=1 i=1
′
+ ∑ ⟨+ λ j − j + 1⟩ε j−1/2 + ∑ ⟨+ λ j − j⟩ε j ,
j≥k+2 j≥k+2
′
(ii) Now suppose that + λk+1 = s < k + 1. Since + λ is a partition, the weight
(6.23) becomes
k s
′
λ[k] = λ|k + ∑ + λi εi + ∑ (+ λi − k)εi−1/2 ,
i=1 i=1
where + λ′ − k
> 0, for i ≤ s. It follows by Lemma 1.36 that odd reflections with
i
respect to εk+1/2 − εk+1 , · · · , εs+1/2 − εk+1 leave λ[k] unchanged, while odd refec-
tions with respect to εs−1/2 − εk+1 , · · · , ε1/2 − εk+1 do affect λ[k] . In a similar way,
the new weight after these (k + 1) odd reflections is calculated to be
k+1 s
′
λ[k,k+1] = λ|k + ∑ + λi εi + ∑ (+ λi − k − 1)εi−1/2 ,
i=1 i=1
e is an h-submodule of M,
Note that T (M) e and T (M)
e is an h-submodule of M.
e One
⊕ e
e e e e
checks that if M = γ∈eh∗ Mγ is an l-module, then T (M) is an l-submodule of M
210 6. Super duality
e is an l-submodule of M.
and T (M) e Furthermore, if M e is a e e is
g-module, then T (M)
a g-submodule of Me and T (M)
e is a g-submodule of M. e
The direct sum decomposition in Me gives rise to the natural projections
e −−−−→ T (M)
TMe : M e and e −−−−→ T (M)
T Me : M e
Proof. We prove the first formula using a comparison of the characters. The proof
of the second formula is similar and will be omitted.
Let λ = (dΛ0 , − λ, + λ) ∈ P+ . To be consistent with the notation of λ|k in (6.22),
it is convenient to make a convention that k contains the central element K, and let
L(el ∩ k, λ|k ) denote the irreducible el ∩ k-module of highest weight λ|k . It follows by
the discussion in Section 6.2.1 that
(6.26) ch L(el, λθ ) = ch L(el ∩ k, λ|k ) hs+ λ′ (x1/2 , x3/2 , . . . ; x1 , x2 , . . .),
where xr := eεr for r ∈ 12 N.
Note that el ∩ k = l ∩ k. Now L(el, λθ ) is completely reducible as an l-module,
since a polynomial module of gl(Ve0+ ) = gl(∞|∞) is completely reducible when
restricted to gl(∞). On the character level, applying T to L(el, λθ ) corresponds to
setting x1/2 , x3/2 , x5/2 , . . . in the character formula (6.26) to zero. Thus, by (A.34),
( )
T L(el, λθ ) is an l-module with character being ch L(el∩k, λ|k ) s+ λ (x1 , x2 , . . .), which
is precisely the character of L(l, λ). This proves the first formula.
6.3. The irreducible character formulas 211
Proof. We will only prove the statement for T (Ve (λθ )), as the case of T (Ve (λθ )) is
analogous.
By Proposition 6.11, for n > ℓ(+ λ), Ve (λθ ) is a e
bc (n)-highest weight module of
highest weight λ, and thus a nonzero vector v in Ve (λθ ) of weight λ (unique up to
a scalar multiple and independent of n) is a ebc (n)-highest weight vector of Ve (λθ ).
Evidently we have v ∈ T (Ve (λθ )), and by Proposition 6.8, v is a b-singular vector
in T (Ve (λθ )).
212 6. Super duality
−1 n−1
(6.27) λ−µ = ∑ ai εi + ∑ b j ε j , ai , b j ∈ Z.
i=−m j=1
(We emphasize that only ε j with positive integer indices j appear in (6.27).) But
e c (n), i.e.,
λ − µ is also a Z+ -linear combination of simple roots from Π
λ−µ = ∑ aα α,
e c (n)
α∈Π
6.3.4. Character formulas. The following theorem is the first main result of this
chapter.
Theorem 6.15. Let λ ∈ P+ . We have
( ) ( θ )
e θ ) = ∆(λ), T L(λ
T ∆(λ e ) = L(λ);
( ) ( θ )
e θ ) = ∆(λ♮ ), T L(λ
T ∆(λ e ) = L(λ♮ ).
Proof. We will prove only the statements involving T , and the statements involving
T can be proved in the same way.
Let us write ∆(λ) = U(u− ) ⊗C L(l, λ) and ∆(λ e θ ) = U(e u− ) ⊗C L(el, λθ ). We
−
observe that all the weights in U(u ), L(l, λ), U(e u ), and L(el, λθ ) are of the form
−
e = L(λ
Set M e θ ). Suppose that M := T (M) e is not irreducible. Since by Propo-
sition 6.14 the M is a highest weight g-module, it must have a b-singular vector,
say w of weight µ ∈ P+ , inside M that is not a highest weight vector. We choose
n > max{ℓ(+ λ), ℓ(+ µ)}. By Proposition 6.11, the e g-module M e is a e
bc (n)-highest
weight module of highest weight λ, say with a highest weight vector vλ . By Propo-
sition 6.14, vλ is a b-highest weight vector of M, and hence w ∈ U(a)vλ , where a
is the subalgebra of n− generated by the root vectors corresponding to the roots in
−Π = −Π(k) ⊔ {−β−1 , −β1 , . . . , −βk }, for some k.
Now choose q > max{n, k + 1}. Note that vλ is also a e bc (q)-highest weight
vector of the eg-module M e of weight λ. Since w is b-singular, it is annihilated by
the root vectors corresponding to the roots in Π(k) ⊔ {β−1 , β1 , β2 , . . .}. But w is
also annihilated by the root vectors corresponding to the roots in Π e c (q) comple-
mentary to the subset Π = Π(k) ⊔ {β−1 , β1 , β2 , . . . , βq−1 }, since these root vectors
commute with a and w ∈ U(a)vλ . It follows that w is a e bc (q)-singular vector in M,e
e
contradicting the irreducibility of M.
It is standard to write
(6.28) ch L(λ) = ∑ aµλ ch ∆(µ),
µ∈P+
for λ ∈ P+ and aµλ ∈ Z. The triangular transition matrix (aµλ ) in Theorem 6.16
or its inverse is known in the Kazhdan-Lusztig theory, since the Kazhdan-Lusztig
polynomials determine the composition multiplicities of parabolic Verma modules
in the parabolic BGG category O of g-modules; see, e.g., Soergel [So, p. 455]. The
infinite-rank of g here does not cause any difficulty in light of Proposition 6.7. We
have the following character formulas for g and e g.
Theorem 6.16. Let λ ∈ P+ . Then
(1) ch L(λ♮ ) = ∑µ∈P+ aµλ ch ∆(µ♮ ),
e θ ).
e θ ) = ∑µ∈P+ aµλ ch ∆(µ
(2) ch L(λ
perfectly the corresponding Kostant homology groups (and also Kostant cohomol-
ogy groups). Such matchings are then interpreted as equalities of the Kazhdan-
e
Lusztig-Vogan polynomials for the categories O, O, and O.
Some basic facts on Lie algebra homology and cohomology needed in this
section can be found in the book of Kumar [Ku].
∑ (−1)s+t+|xs | ∑i=1 |xi |+|xt | ∑ j=1 |x j |+|xs ||xt | [xs , xt ]x1 · · · xbs · · · xbt · · · xn ⊗ v
s−1 t−1
1≤s<t≤n
n
+ ∑ (−1)s+|xs | ∑i=s+1 |xi | x1 · · · xbs · · · xn ⊗ xs v,
n
s=1
Now suppose that L is a Lie superalgebra (or simply a vector superspace) which
contains a sequence of finite-dimensional subalgebras (or subspaces) Lk , for k ∈ N,
such that Lk ⊆ Lk+1 for each k and that L = ∪k Lk . Further suppose that L contains
a natural basis B such that B ∩ Lk is a basis for Lk , for all k. Then with respect to
the basis B, we may extend by zero to regard Lk∗ ⊆ Lk+1 ∗ for each k. In this way, the
restricted dual of L, denoted by L , is given by L := ∪k Lk∗ . All the examples of
∗ ∗
Lie superalgebras in this book satisfy these assumptions, and there is such a basis
for L consisting of root or weight vectors.
Fix n ∈ Z+ , and let V be a vector superspace of countable dimension. Then,
∧n Lk ⊆ ∧n Lk+1 , and the basis B of L induces a natural basis, denoted formally by
∧n B, of ∧n L which is compatible with ∧n Lk for all k. Regard Hom(∧n Lk ,V ) ⊆
Hom(∧n Lk+1 ,V ) by an extension by zero with respect to the basis ∧n B, for each k.
In this way, we define the (restricted) Hom-space
∪
(6.30) Hom(∧n L,V ) := Hom(∧n Lk ,V ).
k
+ ∑(−1)s+t+|xs | ∑i=1 |xi |+|xt | ∑ j=1 |x j |+|xs ||xt | f ([xs , xt ], x1 , . . . , xbs , . . . , xbt , . . . , xn+1 ).
s−1 t−1
s<t
6.4.2. Kostant u− -homology and u-cohomology. Recall the Lie algebra g with
parabolic subalgebra p, Levi subalgebra l, nilradical u, and opposite radical u−
from Section 6.2.2. Let M ∈ O. Then we have M = ⊕µ∈h∗ Mµ , with dim Mµ < ∞. In
this case, the restricted dual M ∗ of M is given by M ∗ = ⊕µ∈h∗ (Mµ )∗ . Let τ : g → g
be the Chevalley automorphism of g from Section 1.1.3 in Chapter 1. We denote
M ∗ with a τ-twisted g-action by M ∨ , i.e.,
M ∨ := ⊕µ∈h∗ (Mµ )∗ ,
where the action of g is now given by (x · f )(v) = (−1)|x|·| f |+1 f (τ(x) · v). Then
M ∨ lies in O, for M ∈ O. We shall refer M ∨ the dual of M in O, and sending
each M ∈ O to M ∨ defines an exact functor, called the duality functor, on O. The
following is a summary of some standard properties of the duality functor on O, cf.
Humphreys [Hum, Theorem 3.2].
Proposition 6.19. The duality functor is a an exact and contravariant functor,
inducing a self-equivalence on the category O. For each M ∈ O, M and M ∨ have
the same character, and hence they have the same composition factor multiplicities.
In particular, we have L(λ)∨ ∼
= L(λ), for λ ∈ P+ .
Let M ∈ O. We note that u and u− are l-modules so that the chain and cochain
complexes of the Lie superalgebra u or u− with coefficients in M are semisimple
l-modules. It is a standard fact that the boundary and coboundary operators are
l-invariant, so the homology and cohomology groups of u and u− are naturally
semisimple l-modules.
Lemma 6.20. Let M ∈ O and n ∈ Z+ . As semisimple l-modules, we have
∼ Hn (u, M ∗ )∗ .
H n (u, M) =
the complex C• (u, M) and the coboundary operator Hom(d, C) of the complex
Hom (C• (u, M ∗ ) , C). Since the operators are also l-module homomorphisms, we
obtain the corresponding isomorphisms of the cohomology groups as l-modules.
Hn (u, M ∗ )∗ ∼
= Hn (u− , M ∨ ).
Proof. We use a superscript τ to indicate the l-action obtained by twisting the usual
l-action by the Chevalley automorphism τ. Since uτ ∼ τ
= u− and (M ∗ ) = M ∨ , we see
that
( i )τ
∧ u ⊗ M∗ ∼ = ∧i u− ⊗ M ∨ .
Hn (u, M ∗ )τ ∼
= Hn (u− , M ∨ ).
Now since the l-modules Hn (u, M ∗ )τ and Hn (u, M ∗ )∗ are direct sums of simple l-
modules with highest weights in P+ and they have the same character, we have an
l-module isomorphism Hn (u, M ∗ )τ ∼
= Hn (u, M ∗ )∗ . The lemma follows.
e are analogous.
Proof. We only prove the statements for O, while those for O and O
Combining Lemma 6.20 and Lemma 6.21, we obtain an l-module isomorphism
∼ H n (u, M) . This implies that Hn (u− , L(λ)) ∼
Hn (u− , M ∨ ) = = H n (u, L(λ)), thanks to
L(λ)∨ ∼
= L(λ) by Proposition 6.19.
218 6. Super duality
∧(u− ) ⊗ M , the l-module ∧(ū− ) ⊗ M, and the el-module ∧(e u− ) ⊗ M e are completely
reducible by Theorem 6.3.
Proof. We will only prove (1), as the argument for (2) is parallel.
( )
It follows from eu− ∩ g = u− and the definition of T that T ∧ (e u− ) = ∧(u− ).
Now, since all modules involved have weights of the ( form ∑i<0) ai εi + ∑r>0 br εr
(possibly modulo dΛ0 ) with br ∈ Z+ , we see that T ∧ (e u− ) ⊗ Me and ∧(u− ) ⊗ M
( )
have the same character. Hence, it follows from T ∧ (e u− ) ⊗ Me ⊇ ∧(u− ) ⊗ M that
( ) ( )
T ∧ (e e = ∧(u− ) ⊗ M. By Theorem 6.15, T L(λ
u− ) ⊗ M e θ ) = L(λ), and so we
( )
have T ∧ (e e θ ) = ∧(u− ) ⊗ L(λ).
u− ) ⊗ L(λ
¿From the definitions of d,e d, and d in (6.29), we have d(v) e = d(v) for all
[ ]
v ∈ ∧(u− ) ⊗ M and d(w)e = d(w) for all v ∈ ∧(ū− ) ⊗ M. Thus, T de = d and
[ ]
T de = d. This completes the proof of (1).
Proof. We only prove (1) as (2) is similar. By Lemma 6.23, we have a surjec-
tive l-module homomorphism T : ∧(e e → ∧(u− ) ⊗ M. By Lemma 6.12,
u− ) ⊗ M
( )
T L(el, µθ ) = L(l, µ), for all µ ∈ P+ . Now (1) follows from the exactness of T .
Proof. We shall only prove (1), as the argument for (2) is similar.
We regard ∧(u− ) ⊗ M ⊆ ∧(e e By Lemma 6.23 and (6.34), we have
u− ) ⊗ M.
( )
e = Ker de∩ ∧(u− ) ⊗ M = Ker d,
T (Ker d)
and ( )
e = Im de∩ ∧(u− ) ⊗ M = Im d.
T (Im d)
Since T is an exact functor, we have
( )
⊕ ⊕
T u− , M)
Hn (e e (Im d)
e = T (Ker d)/T e = Ker d/Im d = Hn (u− , M).
n≥0 n≥0
This completes the proof.
Corollary 6.26. Let M e λ ∈ P+ , and n ∈ Z+ . Furthermore, set T (M)
e ∈ O, e = M and
e
T (M) = M. We have
(1) T (H n (e e ∼
u, M)) = H n (u, M), and thus T (H n (e e θ ))) ∼
u, L(λ = H n (u, L(λ)), as
l-modules;
(2) T (H n (e e ∼
u, M)) = H n (u, M), and thus T (H n (e e θ ))) ∼
u, L(λ = H n (u, L(λ♮ )), as
l-modules.
(3) Moreover, we have
( ( ))
Homl (L(l, µ), H n (u, M)) ∼
= Homel L(el, µθ ), H n e e
u, M
( ( ))
∼
= Homl L(l, µ♮ ), H n ū, M ,
220 6. Super duality
and so in particular
( ( ))
Homl (L(l, µ), H n (u, L(λ))) ∼
= Homel L(el, µθ ), H n e e θ)
u, L(λ
( ( ))
∼
= Homl L(l, µ♮ ), H n ū, L(λ♮ ) .
( ) ( )
∼ e ∨ ∼ n
= Hn u− , T (M) = H u, T (M) . e
The first and the last isomorphisms follow from Theorem 6.22. The second iso-
morphism is due to Theorem 6.25(1), while the third isomorphism is due the com-
patibility of the functors T and ·∨ . This proves the first statement of (1). Setting
Me = L(λ
e θ ) and using Theorem 6.15 we get the second statement of (1). The first
( )
“∼
=” in (3) follows from (1) and that T L(el, µθ ) = L(l, µ) (see Lemma 6.12).
The argument for (2) is similar to (1), and the second “∼ =” in (3) follows from
(2) and Lemma 6.12.
∞ ( )
ℓµλ (q) := ∑ (−q)−n dim Homl L(l, µ), Hn (u− , L(λ)) ,
n=0
∞ ( )
ℓµ♮ λ♮ (q) := ∑ (−q)−n dim Homl L(l, µ♮ ), Hn (ū− , L(λ♮ )) ,
n=0
∞ ( )
ℓeµθ λθ (q) := ∑ (−q)−n dim Homel L(el, µθ ), Hn (e
u− , L(λθ )) .
n=0
Recall from (6.28) that ch L(λ) = ∑µ∈P+ aµλ ch ∆(µ), for λ ∈ P+ . By definition,
ℓµλ (q), ℓµ♮ λ♮ (q), ℓeµθ λθ (q) are power series in q−1 .
Proposition 6.28. The ℓµλ (q), ℓµ♮ λ♮ (q), ℓeµθ λθ (q) are polynomials in q−1 . Moreover,
we have ℓµλ (1) = aµλ .
6.4. Kostant homology and KLV polynomials 221
Proof. We will only show that ℓµλ (q) is a polynomial in q−1 , and similar arguments
can be applied to ℓµ♮ λ♮ (q), ℓeµθ λθ (q).
Take λ ∈ P+ , and recall that Cn (u− , L(λ)) = ∧(ū− ) ⊗ L(λ). Since Cn (u− , L(λ))
has finite-dimensional weight spaces, the module L(l, µ) appears with finite multi-
plicity in Cn (u− , L(λ)), for each n. Also observe that, for a given weight µ ∈ P+ ,
µ cannot be a weight of Cn (u− , L(λ)), for n ≫ 0. This proves that ℓµλ (q) is a
polynomial in q−1 .
We can now evaluate the polynomial ℓµλ (q) at q = 1. Applying the Euler-
Poincaré principle to the chain complex C• (u− , L(λ)) of h-modules, we have
ch L(λ) ∑ (−1)n ch ∧n (u− )
n≥0
Theorem 6.29. For µ, λ ∈ P+ we have ℓµλ (q) = ℓeµθ λθ (q) = ℓµ♮ λ♮ (q).
We now employ the same method as in Section 6.4.3 to study the effect of the
truncation functors trkn on the corresponding u− k -homology groups with coefficients
in the module Lk (λ), for λ ∈ Pk . Let d : ∧(u−
+ (n) −
n ) ⊗ Ln (λ) → ∧(un ) ⊗ Ln (λ) denote
the corresponding boundary operator of the chains of the finite-dimensional Lie
algebra u−n with coefficients in Ln (λ). It is evident that d |∧(u−
(k)
n )⊗Ln (λ)
= d (n) , and
furthermore we have d (n) trkn = trkn d (k) . The proof of Theorem 6.25 can be adapted
to prove the following.
Theorem 6.31. Let λ ∈ Pk+ and i ∈ Z+ . We have
{
k
( −
) Hi (u−
n , Ln (λ)) , if λ ∈ Pn+ ,
trn Hi (uk , Lk (λ)) =
0, otherwise.
i=0
equals the parabolic Kazhdan-Lusztig polynomial of the Lie algebra gn via Vogan’s
homological interpretation. Theorem 6.31 and (6.36) imply that the Kazhdan-
Lusztig polynomials for finite-rank semisimple or reductive Lie algebras admit a
remarkable stability, i.e., we have
(n)
ℓµλ (q) = ℓµλ (q), λ, µ ∈ Pn+ .
C Ext1C (·, A) is a contravariant functor. With this notation ·γA = Ext1C (γ, A) we shall
write ·γA = ·γ, when A is clear from the context.
Dually, suppose we are given an object A′ ∈ C, a morphism α : A → A′ , and an
element E ∈ Ext1C (C, A), so that we have the following diagram:
E: 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
α
y
A′
We obtain uniquely an element E ′ ∈ Ext1C (C, A) such that the following diagram is
commutative:
E : 0 −−−−→ A −−−−→ B −−−−→ C −−−−→ 0
α β
y y
E ′ : 0 −−−−→ A′ −−−−→ B′ −−−−→ C −−−−→ 0
The resulting element E ′ is denoted by αE. Fixing E we thus obtain a map
·E
·E : HomC (A, A′ ) → Ext1C (C, A′ ) given by α → αE. On the other hand, letting
E vary, we can show that a fixed morphism α : A → A′ induces a well-defined map
′
C α· : ExtC (C, A) → ExtC (C, A ) so that ExtC (C, ·) is a covariant functor. With this
1 1 1
notation we have C α· = ExtC (C, α). Thus, Ext1C (·, ·) gives a bi-functor on C. We
1
shall write C α· = α·, when C is clear from the context. For an extension E, with
morphisms α and γ as above, we have (αE)γ = α(Eγ).
We define an abelian group structure on Ext1C (C, A). For A ∈ C, we define a
morphism ∆ = ∆A : A → A ⊕ A, ∆(a) := (a, a), and a morphism ∇ = ∇A : A ⊕ A → A
by ∇(a, a′ ) := a + a′ . Given two elements E, E ′ ∈ Ext1C (C, A) we let E ⊕ E ′ be the
obvious element in Ext1C (C ⊕C, A ⊕ A). To define the group operation we consider
(6.38) ∇A (E ⊕ E ′ )∆C ,
which is an exact sequence of the form (6.37). Define E + E ′ to be the ele-
ment in Ext1C (C, A) corresponding to (6.38). It can be shown that the operation
+ : Ext1C (C, A) × Ext1C (C, A) → Ext1C (C, A) is well-defined, abelian, and associa-
tive. Furthermore, it can be shown that the equivalence class of the split exact
sequence E0 is an additive identity and that (−1A )E is the additive inverse of E.
For A,C ∈ C, it is convenient to set Ext0C (C, A) := HomC (C, A).
Let X be an object in C. Applying the functor HomC (X, ·) to the exact sequence
(6.37) gives us an exact sequence
(6.39) 0 → HomC (X, A) → HomC (X, B) → HomC (X,C).
Now for each γ ∈ HomC (X,C), we obtain Eγ ∈ Ext1C (X, A). It can be shown that
the map E· extends (6.39) to the following exact sequence:
(6.40) 0 −→HomC (X, A) −→ HomC (X, B) −→ HomC (X,C)
6.5. Super duality as an equivalence of categories 225
E· i· j·
−→Ext1C (X, A) −→ Ext1C (X, B) −→ Ext1C (X,C).
∂k (a ⊗ x1 x2 · · · xk )
( )
Homl L(l, λ♮ ), H i (u,V ) ∼ i
= ExtO (∆(λ♮ ),V ).
Proof. We shall only prove the first isomorphism for l for i = 0, 1, and the remain-
ing two isomorphisms can be proved in an analogous fashion.
Let V ∈ O. Recall the (unrestricted) cohomology group Hi (u,V ) from Re-
mark 6.18 in contrast to the (restricted) cohomology group H i (u,V ) used mostly in
this chapter. Let Hi (u,V )ss be the maximal h-semisimple submodule of H i (u,V ).
We have the following
∼ H i (u,V ).
Claim. Hi (u,V )ss =
To see this, observe by definition that we have a natural injective map from
ι
H i (u,V ) → Hi (u,V )ss . Since the coboundary operator ∂i is an h-homomorphism
and Hom(∧ (u),V ) is a direct product of finite-dimensional h-weight spaces, Ker ∂i
i
and Im ∂i are also direct products of finite-dimensional h-weight spaces. Now given
x ∈ Ker ∂i representing a cohomology class in Hi (u,V )ss , there exists an element
y ∈ Im ∂i−1 such that x − y ∈ Ci (u,V ) ∩ Ker ∂i . We have then ι(x − y) ∈ x + Im ∂i−1 ,
and hence ι is also a surjection. This proves the claim.
Now since L(l, λ) is h-semisimple and Hi (u,V ) is a direct product of finite-
dimensional h-weight spaces, it follows by the claim that
( ) ( )
(6.44) Homl L(l, λ), Hi (u,V ) ∼ = Homl L(l, λ), H i (u,V ) .
where the last “∼ =” is due to Frobenius reciprocity. This proves the isomorphism
for l with i = 0 in the theorem.
( )
It remains to show that Homl L(l, λ), H 1 (u,V ) ∼ = Ext1O (∆(λ),V ). We shall
construct the isomorphism explicitly.
( )
First, take f¯ ∈ Homl L(l, λ), H 1 (u,V ) . Recall H 1 (u,V ) = Ker d1 /Im d0 , with
Ker d1 ⊆ Hom(u,V ). Let f ∈ Homl (L(l, λ), Hom(u,V )) ∼ = Homl (u ⊗ L(l, λ),V )
be a representative,
( which, by
) Frobenius reciprocity, is regarded as an element in
Homp Indpl (u ⊗ L(l, λ)) ,V .
Applying the exact functor U(g) ⊗U(p) - to (6.43), we obtain an exact sequence
( ) ∂′2 ∂′1
U(g) ⊗U(l) ∧2 (p/l) ⊗ L(l, λ) −→U(g) ⊗U(l) (p/l ⊗ L(l, λ)) −→
∂′
U(g) ⊗U(l) L(l, λ) −→
0
∆(λ) −→ 0,
6.5. Super duality as an equivalence of categories 227
where ∂′i denotes the differentials induced from ∂i . This gives rise to the following
diagram
∂′
U(g) ⊗U(l) (p/l ⊗ L(l, λ)) −−−1−→ U(g) ⊗U(l) L(l, λ) −−−−→ ∆(λ) −−−−→ 0
f
y
V
Taking the pushout E f of f and ∂′1 , we obtain the following commutative diagram
U(g) U(g)
IndU(l) (p/l ⊗ L(l, λ)) −−−−→ IndU(l) L(l, λ) −−−−→ ∆(λ) −−−−→ 0
f
y y
α
V −−−−→ Ef −−−−→ ∆(λ) −−−−→ 0
Since f represents a cocycle, we have f ◦∂′2 = 0. This implies that α is an injection.
Now since V and ∆(λ) lie in O, so does E f , and hence we obtain an element in
Ext1O (∆(λ),V ). One verifies in a straightforward manner that if f is a coboundary,
then E f ∼
= V ⊕ ∆(λ), and hence the map f¯ → E f¯ is well-defined.
Conversely, suppose that we have an extension of the form
0 −→ V −→ E −→ ∆(λ) −→ 0.
Note that E ∈ O, and hence there exists an l-submodule of E that is mapped iso-
morphically to L(l, λ) ⊆ ∆(λ). We denote this l-module also by L(l, λ) and note
that uL(l, λ) ⊆ V . This allows us to define an element fE ∈ Homl (u ⊗ L(l, λ),V ) ∼
=
Homl (L(l, λ), Hom(u,V )) by fE (u ⊗ w) = uw. Now fE corresponds to a cocycle in
Homl (L(l, λ), Hom(u,V )), because V is a submodule of the u-module E. Also, one
sees that if E is the trivial extension, then fE must be a coboundary. Thus, sending
E to f¯E gives us a well-defined map Ext1O (∆(λ),V ) → Homl (L(l, λ), H1 (u,V )) ∼ =
Homl (L(l, λ), H 1 (u,V )); see (6.44).
It can be
( checked1 that these
) two1 maps are inverses to each other. Hence, we
have Homl L(l, λ), H (u,V ) ∼ = ExtO (∆(λ),V ).
e
It makes sense to apply T and T to the short exact sequences (i.e., Ext1e ) in O.
O
Proof. We shall only prove the identity for T , as the proof for T is analogous.
By Theorem 6.32, for i = 0, 1, we have
( ) ( )
ExtiO ∆(λ), L(µ) ∼ = Homl L(l, λ), H i (u, L(µ)) ,
228 6. Super duality
( ) ( )
(6.45) ExtiOe ∆(λ e θ) ∼
e θ ), L(µ = Homel L(el, λθ ), H i (e e θ )) .
u− , L(µ
( )
Recall that T L(el, λθ ) = L(l, λ) by Lemma 6.12, and that T (H n (e e θ ))) ∼
u, L(λ =
n
H (u, L(λ)) by Corollary 6.26. Now the identity for T in the corollary follows
by applying the functor T to (6.45) and the naturality of T .
f e f . The following proposition is standard and well-
6.5.3. Categories O f , O and O
known. For a proof the reader may consult Kumar [Ku, Lemma 2.1.10]. Recall Π
denotes the standard fundamental system of g.
Proposition 6.34. Let M ∈ O. Then there exists a (possibly infinite) increasing
filtration of g-modules
(6.46) 0 = M0 ⊆ M1 ⊆ M2 ⊆ · · · ⊆ Mi−1 ⊆ Mi ⊆ · · ·
such that
∪
(1) M = i≥0 Mi ;
(2) Mi /Mi−1 is a highest weight module of highest weight νi with νi ∈ P+ ,
for i ≥ 1;
(3) the condition νi − ν j ∈ ∑α∈Π Z+ α implies that i < j;
(4) for any weight µ of M, there exists an r ∈ N such that (M/Mr )µ = 0.
e
e ∈ O.
Similar statements hold for M ∈ O and M
Corollary 6.35. Let M ∈ O. Then M ∈ O f if and only if there exists a finite increas-
ing filtration 0 = M0 ⊆ M1 ⊆ M2 ⊆ · · · ⊆ Mk = M of g-modules such that Mi /Mi−1
is a highest weight module of highest weight νi with νi ∈ P+ , for 1 ≤ i ≤ k. Similar
statements hold for M ∈ O and M e
e ∈ O.
Proof. If M is a g-module with a finite filtration of the form (6.46), then clearly
M is finitely generated. Conversely, suppose that u1 , u2 , . . . , un are vectors in M of
weights ν1 , ν2 , . . . , νn generating M over U(g). By Proposition 6.34(4), M has a
(possibly infinite) filtration of the form (6.46) such that there exists r ∈ N with Mr
containing all vectors in M of weights ν1 , ν2 , . . . , νn . Thus ui ∈ Mr , for 1 ≤ i ≤ n,
and hence Mr = M. The proofs for e g and g are analogous and omitted.
6.5.4. Lifting highest weight modules. The following proposition is the converse
of Proposition 6.14.
Proposition 6.36. (1) Suppose that V (λ) is a highest weight g-module of
g-module Ve (λθ ) of
highest weight λ ∈ P+ . Then there is a highest weight e
highest weight λ such that T (Ve (λ )) = V (λ).
θ θ
6.5. Super duality as an equivalence of categories 229
6.5.5. Super duality and strategy of proof. Recall the functors T and T from
Section 6.3.3. The following is a main result of this chapter.
230 6. Super duality
e is a semisimple e ⊕
Recall that Ve ∈ O h-module with Ve = γ∈eh∗ Veγ . Then Ve gives rise
g-module Ve ′ :
to a Z2 -graded e
⊕ ⊕
(6.47) Ve ′ := Ve0̄ Ve1̄ , Veε := Veµ (ε = 0̄, 1̄),
µ∈e
h∗ε
now on drop the superscript 0̄ and use O, O, e O f and O e f to denote the respective
0̄ e 0̄ e f ,0̄ . Henceforth, when we write ∆(λ ef,
categories O , O
f ,0̄
, O and O e θ ), L(λ
e θ) ∈ O
λ ∈ P+ , we will mean the corresponding modules equipped with the Z2 -gradation
(6.47). Similar convention applies to ∆(λ♮ ) and L(λ♮ ).
In the remainder of this subsection, we outline the strategy of the proof of
Theorem 6.37. The detailed proof will be carried out in the next subsection. We
shall restrict our proof for T entirely, since the arguments for the functors T and
T are completely parallel. First, we study the relation between the morphisms
6.5. Super duality as an equivalence of categories 231
6.5.6. The proof of super duality. We shall freely use the following short-hand
notations without further notices. Set M = T (M) e and M = T (M), e for M
e ∈ O.e Sim-
e ′ e ′′ e e
ilar convention applies to the images of T and T of M , M , N ∈ O. In addition, set
V (λ) = T (Ve (λθ )) and V (λ♮ ) = T (Ve (λθ )), for a highest weight eg-module Ve (λθ ) of
highest weight λθ with λ ∈ P+ ; see Proposition 6.14 for a justification of notations.
e and let
e ∈ O,
Lemma 6.38. Let N
e
Ee : e ′ −→ e e ′′ −→ 0
i
(6.48) 0 −→ M M−→M
be an exact sequence of e e
g-modules in O.
(1) The exact sequence (6.48) induces the following commutative diagram
with exact rows. (We will use subscripts to distinguish various maps in-
duced by T .)
( ′′ ) ( ) ( ′ )
0 −−→ HomOe M e ,N e −−→ Hom e M, e Ne −−→ Hom e M e ,N e
O O
T e ′′ e T e e T e ′ e
y M ,N y M,N y M ,N
( ′′ ) ( ) ( )
0 −−→ HomO M , N −−→ HomO M, N −−→ HomO M ′ , N
·Ee ( ′′ ) ( ) ( ′ )
−−→ Ext1Oe Me ,N e −−→ Ext1 M,e N e −−→ Ext1 M e ,N e
e
O e
O
T 1 T 1 T 1
y Me ′′ ,Ne y M,e Ne y Me ′ ,Ne
e
·T (E) ( ) ( ) ( )
−−−→ Ext1O M ′′ , N −−→ Ext1O M, N −−→ Ext1O M ′ , N .
Proof. By (6.40), the rows are exact, and it remains to show that the following
diagram is commutative:
( ′ ) ·Ee ( ′′ )
HomOe Me ,N e −−− −→ Ext1Oe Me ,N e
T e ′ e T 1
(6.49) y M ,N y Me ′′ ,Ne
( ) ·T (E)
e ( )
HomO M ′ , N −−−−→ Ext1O M ′′ , N
232 6. Super duality
( ′ ) ( ′′ )
Let fe ∈ HomOe M e . Then feEe ∈ Ext1 M
e ,N
e
O
e ,N
e is the bottom exact row of
the following commutative diagram:
f′ −−−ei−→
0 −−−−→ M e −−−−→
M e ′′ −−−−→ 0
M
(6.50) e
yf y
e −−−−→ Fe −−−−→ M
0 −−−−→ N e ′′ −−−−→ 0
This gives rise to the following commutative diagram with exact rows:
( ) ( ) ( )
0 −−−→ HomOe Ve (λθ ), N e −−−→ Hom e ∆(λe θ ), Ne −−−→ Hom e M, e Ne
O O
Te θ e Te θ e T e e
y V (λ ),N y ∆(λ ),N y M,N
( ) ( ) ( )
0 −−−→ HomO V (λ), N −−−→ HomO ∆(λ), N −−−→ HomO M, N .
Corollary 6.26(3) and Theorem 6.32 imply that T∆(λ e is an isomorphism. By
e θ ),N
Lemma 6.39, TM, e is an injection. Now a standard diagram chasing implies that
eN
TVe (λθ ),Ne is an isomorphism.
Proof. Let
e
(6.51) e −→ Ee −→
0 −→ N
f
Ve (λθ ) −→ 0
be an exact sequence of e
g-modules. Suppose that (6.51) gives rise to a split ex-
T [ fe]
act sequence of g-modules 0 → N → E → V (λ) → 0. Thus there exists ψ ∈
HomO (V (λ), E) such that T [ fe] ◦ ψ = 1V (λ) . By Lemma 6.40, there exists ψ e∈
e θ e e
e ] = ψ. Thus T [ f ◦ ψ
HomOe (V (λ ), E) such that T [ψ e
e ] = T [ f ] ◦ T [ψ
e ] = 1V (λ) . By
e e
Lemma 6.40 again, we have f ◦ ψ = 1Ve (λθ ) , and hence (6.51) splits.
e N
Proposition 6.42. Let M, e We have
e ∈ O.
e N)
(1) T : HomOe (M, e −→ HomO (M, N) is an isomorphism.
e N)
(2) T : Hom e (M, e −→ Hom (M, N) is an isomorphism.
O O
The sequence (6.52) induces the following commutative diagram with exact rows.
( ) ( )
0 −−−−→ HomOe Zei , N e −−−−→ Hom e M
O
ei , N
e
Te e T e e
y Zi ,N y M,N
( ) ( )
0 −−−−→ HomO Zi , N −−−−→ HomO Mi , N
·eιi ( ) ( )
−−−− → HomOe Mei−1 , N
e −−−−→ Ext1 Zei , Ne
e
O
T e e T 1
y Mi−1 ,N y Zei ,Ne
·ιi ( ) ( )
−−−− → HomO Mi−1 , N −−−−→ Ext1O Zi , N .
The map TMei−1 ,Ne is an isomorphism by induction. The map TZe1 ,Ne is an injection
i
by Lemma 6.41. Also TZei ,Ne is an isomorphism by Lemma 6.40. Now a standard
diagram chasing implies that TMei ,Ne is an isomorphism.
Now we consider the general case (i.e. M e By Proposition 6.34, we may
e ∈ O).
choose an increasing filtration of e g-modules 0 = M e0 ⊂ M e1 ⊂ M
e2 ⊂ · · · such that
∪
e e e e
i≥0 Mi = M and Mi /Mi−1 is a highest weight module of highest weight νi with
θ
e N
Lemma 6.43. Let M, e We have
e ∈ O.
e N)
(1) T : Ext1Oe (M, e → Ext1 (M, N) is an injection,
O
e N)
(2) T : Ext1Oe (M, e → Ext1 (M, N) is an injection.
O
Proof. The proof is virtually identical to the proof of Lemma 6.41, where we use
Proposition 6.42 in place of Lemma 6.40.
Proposition 6.44. Let Ve (λθ ) be a highest weight e
g-module of highest weight λθ
with λ ∈ P , and let N
+ e
e ∈ O. Then we have
(1) T : Ext1Oe (Ve (λθ ), N)
e → Ext1 (V (λ), N) is an isomorphism.
O
·e
π ( ) ( )
e θ ), N
−−−−→ Ext1Oe ∆(λ e −−−−→ Ext1 M,e N e
e
O
T 1 T 1
y ∆(λ
e θ e
),N y M,e Ne
·π ( ) ( )
−−−−→ Ext1O ∆(λ), N −−−−→ Ext1O M, N .
The map TM,1 is an injection by Lemma 6.43. The map T 1
eNe e is an isomorphism
e θ ),N
∆(λ
by Lemma 6.33. Also T∆(λ e and TM,
e θ ),N eN e are isomorphisms by Lemma 6.42. Now a
standard diagram chasing implies that TVe1(λθ ),Ne is an isomorphism.
6.6. Notes
This chapter is an exposition on the formulation and proof of super duality, which
is an equivalence of categories O and O, in Cheng-Lam-Wang [CLW], which in
turn was built on Cheng-Lam [CL2] for type A. The super duality conjecture for
type A was first formulated in Cheng-Wang-Zhang [CWZ] and more generally in
Cheng-Wang [CW4], motivated and based in part by Brundan [Br1]. There is a
different approach of Brundan-Stroppel [BrS] to the special case of super duality
conjecture in [CWZ].
Section 6.1. This preliminary section follows closely Cheng-Lam-Wang [CLW].
Section 6.2. The truncation functors and their basic properties regarding Verma
and irreducible modules (Proposition 6.7) have counterparts in the algebraic group
setting, which was developed by Donkin [Don]. In the super duality context, the
truncation functors first appeared in Cheng-Wang-Zhang [CWZ], and then gener-
alized in [CW4, CL2, CLW].
Section 6.3. The odd reflection approach leading to Theorem 6.15 and the
character formulas in Theorem 6.16 was developed in [CL2, CLW]. It is remark-
able that, when combined with Proposition 6.7, this solves completely the irre-
ducible character problem for modules in the category On , which include all finite-
dimensional simple modules of types gl and osp. This approach can be generalized
in a straightforward manner to provide irreducible characters of Kac-Moody super-
algebras.
Totally different approaches and solutions for the finite-dimensional irreducible
character problem have been developed in Serganova [Sva], Brundan [Br1], and
Brundan-Stroppel [BrS] for gl(m|n), and in Gruson-Serganova [GS] for osp.
Section 6.4. We follow Tanaka [Ta] for a formula of the (co)boundary operator
for Lie superalgebra (see also Iohara-Koga [IK]). The results on the identification
6.6. Notes 237
Symmetric functions
Consider the Z-module of formal power series Z[[x]] in the set of infinite inde-
terminates x := {x1 , x2 , x3 , . . .}. For α = (α1 , α2 , . . . , αk ) ∈ Zk+ we write
where by definition σ(i) = i, for all i > n. Since the action of Sn+1 is compatible
with that of Sn , we obtain an action of the direct limit, denoted by S∞ , on Z[[x]].
239
240 A. Symmetric functions
that are called the nth power, elementary, and complete symmetric function,
respectively. Furthermore, we set e0 = h0 = 1, and er = hr = 0, for r < 0.
E(t) := ∑ ent n
= ∏(1 + xit),
n≥0 i≥1
1
H(t) := ∑ hnt n = ∏ 1 − xit .
n≥0 i≥1
It follows that
E(−t)H(t) = 1,
( )
pr t r
E(t) = exp ∑ (−1)r−1 ,
(A.1) r≥1 r
( )
pr t r
H(t) = exp ∑ .
r≥1 r
A.1. The ring Λ and Schur functions 241
(A.2) The set {pλ | λ ⊢ n} forms a linear basis for ΛnQ , for n ∈ N. Moreover,
ΛQ = Q[p1 , p2 , . . .].
The sets {eλ | λ ⊢ n} and {hλ | λ ⊢ n} are Z-bases for Λn , for n ∈ N. Moreover,
Λ = Z[e1 , e2 , . . .] = Z[h1 , h2 , . . .].
with cλµ ∈ Z+ and cλλ > 0. This implies that the matrix (cλµ ), with sub-indices or-
dered compatibly with the dominance order, is an invertible upper triangular matrix
with coefficients in Z. Since {mλ | λ ⊢ n} is a basis for ΛnQ , so is {pλ | λ ⊢ n}.
For a partition λ, every summand in eλ′ is of the form
(A.4) xi1 · · · xiλ′ x j1 · · · x jλ′ · · · xkℓ · · · xkλ′ ,
1 2 ℓ
where i1 < i2 < · · · < iλ′1 , j1 < j2 < · · · < jλ′2 , et cetera. Let us write the expression
in (A.4) as xα = xαj11 xαj22 · · · xαjss , such that α1 ≥ α2 ≥ · · · ≥ αs and all the ji ’s are
distinct. Note that λ ≥ α. Hence we conclude that
eλ′ = ∑ aλµ mµ ,
µ≤λ
with aλµ ∈ Z+ and aλλ = 1. This implies that the set {eλ | λ ⊢ n} is a Z-basis for
Λn .
It follows by (A.1) that, for all n ∈ N,
n
(A.5) ∑ (−1)r er hn−r = 0.
r=0
Thus we can express en as a polynomial of hr ’s and er−1 ’s, for r ≤ n. By induction
on n it follows that every en is a polynomial of {h1 , . . . , hn } with integer coefficients.
Therefore, the Z-module Z[h1 , . . . , hn ]∩Λn = ∑λ⊢n Zhλ contains the Z-span of {eλ |
λ ⊢ n}. Hence ∑λ⊢n Zhλ = Λn , and so {hλ | λ ⊢ n} is a Z-basis for Λn .
(A.6) The sets {p1 , p2 , . . .}, {e1 , e2 , . . .}, and {h1 , h2 , . . .} are algebraically inde-
pendent in Λ.
By (A.6), we can define a ring homomorphism ω : Λ → Λ by letting
ω(en ) = hn , ∀n ∈ N.
(A.7) We have ω(hn ) = en and ω(pn ) = (−1)n−1 pn . Thus, ω is an involution of the
ring Λ.
A.1.2. Schur functions. We first suppose that x = {x1 , x2 , . . . , xn }, and denote the
ring of symmetric polynomials in n variables by Λn . There is a natural ring homo-
morphism πm,n : Λm → Λn , for m > n, given by sending each x j for j > n to 0, and
Λ is the inverse limit of Λn .
For a partition λ of length no greater than n, we let
λn
aλ := ∑ λ1
sgn(σ)xσ(1) λ2
xσ(2) · · · xσ(n) ,
σ∈Sn
and let
n−1
∑ er t = ∏(1 + xit).
(k) r
E (k) (t) =
r=0 i̸=k
Then
1
H(t)E (k) (−t) = .
1 − xk t
This implies that
n
∑ hα −n+ j (−1)n− j en− j = xkα .
(k) i
i
j=1
1
∏ 1 − xi y j = ∑ sλ (x)sλ (y),
i, j λ∈P
Proof. We compute
( )
1
∏ 1 − xi y j = ∏ H(y j ) = ∏ ∑ hr (x)yrj = ∑ hα (x)yα1 1 yα2 2 · · · ,
i, j j j r≥0 α
where the last sum is over all compositions α = (α1 , α2 , . . .). Hence it equals
∑λ hλ (x)mλ (y), proving the first identity in (A.11). The second identity follows
by applying ω in x variables to the first one and using (A.10).
Assume that the number of variables of x and y both equal to n. The gen-
eral
( case is−1proved
) by letting n go to infinity, as usual. Consider the n × n matrix
(1 − xi y j ) . If we multiply the ith row of this matrix by ∏nj=1 (1 − xi y j ), then
we obtain a matrix whose (i, k)th entry is equal to
n
∏(1 − xi yr ) = ∑ (−1)n− j en− j (y)xi
(k) n− j
.
r̸=k j=1
That is, it is the (i, k)th entry of the matrix Aρ (x)M(y), where Aρ and M are as in
the proof of (A.10). This implies that
( ) 1
(A.12) det (1 − xi y j )−1 = aρ (x)aρ (y) ∏ .
i, j 1 − xi y j
Proof. We write
uλ = ∑ aλν hν , vµ = ∑ bµη mη , aλν , bµ,η ∈ Q.
ν η
Then (1) is equivalent to ∑ν aλν bµν = δλµ , which is equivalent to ∑λ aλν bλη = δνη .
Now (2) is equivalent to ∑λ ∑ν,η aλν bλ,η hν (x)mη (y) = ∑λ hλ (x)mλ (y) by using
(A.11). This is equivalent to ∑λ aλν bλη = δνη .
A.1.3. Skew Schur functions. For partitions µ and ν, we write the product sµ sν
as
sµ sν = ∑ cλµν sλ , cλµν ∈ Z.
λ
The cλµν associated with a triple of partitions are the Littlewood-Richardson coeffi-
cients, and they allow us to define the skew Schur function associated to partitions
λ, µ by the formula
sλ/µ := ∑ cλµν sν .
ν
This definition is equivalent to the following identity:
(A.17) (sµ f , sλ ) = ( f , sλ/µ ), ∀ f ∈ Λ.
This can be seen by checking for f = sν . It follows from the definitions that sλ/0/ =
sλ , and by (A.10) that
ω(sλ ) = sλ′ .
Since ω is a ring isomorphism, we obtain by the definition of the Littlewood-
′
Richardson coefficients above that cλµ′ ν′ = cλµν . It now follows from the definition
of the skew Schur functions that
ω(sλ/µ ) = sλ′ /µ′ .
Let µ ⊆ λ be partitions which we may regard as Young diagrams. The skew
diagram λ/µ is obtained from λ by removing the sub-diagram µ. A tableau of
shape λ/µ is a filling of the boxes of λ/µ with elements from the set N. A tableau
T of λ/µ is called semistandard, if the entries in T are strictly increasing from
A.1. The ring Λ and Schur functions 247
top to bottom along each column and (weakly) increasing from left to right along
each row. Given such a semistandard tableau T with entries t ∈ T , we define xT :=
∏t∈T xt .
The skew Schur function sλ/µ admits the following combinatorial description,
a proof of which can be found in [Mac, (5.12)].
(A.18) For partitions λ ⊇ µ, we have
sλ/µ = ∑ xT ,
T
where the summation is over all semistandard tableaux of shape λ/µ.
The following is an easy consequence of (A.18).
(A.19) Let x = {x1 , x2 , . . .} and y = {y1 , y2 , . . .} be two independent sets of vari-
ables. We have
sλ (x, y) = ∑ sµ (x)sλ/µ (y).
µ⊆λ
∑ sν′ = ∏ (1 − xi x j )−1 .
ν even 1≤i< j
Since the involution ω interchanges sλ with sλ′ we conclude from these identities
that
( )
(A.22) ω ∏ (1 − xi x j )−1 = ∏ (1 − xi x j )−1 .
1≤i≤ j 1≤i< j
248 A. Symmetric functions
A.1.4. The Frobenius characteristic map. Let Rn denote the Grothendieck group
of the category of Sn -modules, which has [Sλ ] for λ ∈ Pn as a Z-basis. Let
∞
⊕
R= Rn .
n=0
The Frobenius characteristic map chF : R → Λ is the Z-linear map such that, for
n ≥ 0,
(A.23) chF (χ) = ∑ z−1
µ χµ pµ , χ ∈ Rn ,
µ∈Pn
Proof. One can use the induced character formula to prove that chF is a ring homo-
morphism. By (A.8) and (A.9) we have respectively chF (1n ) = hn and chF (sgnn ) =
en . This immediately implies the first two identities since chF is a homomorphism.
It is known (e.g., as a special case of Lemma 3.11) that
⊕
(A.25) CSn
IndCSµ
1n ∼
= Kλµ Sλ ,
λ≥µ
where the Kostka number Kλµ is equal to the number of semistandard λ-tableaux
of content µ. Note that Kµµ = 1.
Also from (A.18) we conclude that
sλ = ∑ Kλµ mµ .
µ≤λ
Using the relations of dual bases (A.14) and (A.16), this identity admits the fol-
lowing dual version:
(A.26) hµ = ∑ Kλµ sλ .
λ≥µ
A.2. Supersymmetric Polynomials 249
It follows by comparing (A.25) and (A.26) and using the first identity in (A.24) that
chF ([Sλ ]) = sλ . Since [Sλ ] and sλ for λ ∈ P form Z-bases for R and Λ respectively,
we conclude that chF is an isomorphism.
Proof. We proceed by induction on n (for arbitrary m), with the case n = 0 being
clear.
Let n ≥ 1. We may assume, without loss of generality, that q is homogeneous
of degree k ≥ 0, and we proceed by induction on k. Consider
qm,n |yn =0 = q|yn =0 x1 · · · xm m,n−1 .
We can define the ring of supersymmetric functions Λm|∞ and prove the fol-
lowing n = ∞ version of (A.31).
(A.32) The algebra Λ(m|∞),Q of supersymmetric functions is generated by σrm,∞ for
all r ≥ 1.
A.2.2. Super Schur functions. Define Λ e (m|n) to be the subring of the ring of dou-
bly symmetric polynomials over Z in the two sets of variables {x1 , . . . , xm } and
{y1 , . . . , yn } consisting of polynomials f such that f |xm =t=−yn is independent of t.
Evidently f (x, y) ∈ Λ e m|n if and only if f (x, −y) is supersymmetric, hence we have
an isomorphism of rings
∼
= e
(A.33) ρy : Λ(m|n) −→ Λ (m|n) ,
e m|∞ .
of infinite variables {y1 , y2 , . . .} we obtain a ring homomorphism ωy : Λ → Λ
Indeed, by (A.7), we have
(A.34) ωy (pr ) = x1r + . . . + xm
r e m|∞ .
+ (−1)r−1 (yr1 + yr2 + . . .) ∈ Λ
Since the pr ’s, for r ≥ 1, are algebraically independent generators of Λ by (A.2) and
(A.6), we conclude that ωy (ΛQ ) ⊆ Λ e m|∞,Q , and hence ωy (Λ) ⊆ Λe m|∞,Q ∩ Z[x, y] =
e m|∞ . Now (A.32) and the fact that ρ−1
Λ −1
y (ωy (pr )) = σm,∞ , r ≥ 1 , imply that ρy ◦
r
µ⊆λ
The definition of hsλ (x; y) makes sense for x and y being finite and infinite. By
(A.19), we have for y infinite
( )
(A.36) ωy sλ (x, y) = hsλ (x; y).
(A.37) The map ωy : Λ → Λ e m|∞ is an isomorphism of rings. Also, the set {hsλ | λ ∈
e m|∞ .
P} is a Z-basis for Λ
The map Λe (m|∞) → Λ
e (m|n) obtained by setting the variables yi = 0, for i > n, is
a ring epimorphism. Furthermore, it sends the super Schur function corresponding
to an (m|n)-hook partition to the respective super Schur polynomial corresponding
to the same (m|n)-hook partition. The other super Schur functions are sent to zero.
Evidently, the super Schur polynomials corresponding to (m|n)-hook partitions are
linearly independent. Thus we have the following.
(A.38) The set {hsλ }, as λ runs over all (m|n)-hook partitions, forms a Z-basis for
e m|n .
Λ
The the Cauchy identities in (A.11) admit the following super generalization.
(A.39) Let x = {x1 , x2 , . . .}, y = {y1 , y2 , . . .}, z = {z1 , z2 , . . .} be three sets of (possi-
bly infinite) indeterminates. The following identities hold:
∏ j,k (1 + y j zk )
= ∑ hsλ (x; y)sλ (z).
∏i,k (1 − xi zk ) λ∈P
Proof. If suffices to prove the claim in the case when all three sets of indetermi-
nates are infinite. A variant of the Cauchy identity in (A.11) can be written as
1 1
∏ (1 − xi zk ) ∏ (1 − y j zk ) = ∑ sλ (x, y)sλ (z).
i,k j,k λ∈P
A.3. The ring Γ and Schur Q-functions 253
Now the claim follows by applying the involution ωy on the y variables to both
sides of the above equation and using (A.36).
∑ (−1)r qr qs = 0, n ≥ 1.
r+s=n
We shall denote by OP and SP the sets of odd and strict partitions, respectively.
(A.43) The ring Γ and ΓQ enjoy the following remarkable properties:
(1) ΓQ is a polynomial algebra with polynomial generators p2r−1 for r ≥ 1.
(2) ΓQ is a polynomial algebra with polynomial generators q2r−1 for r ≥ 1.
(1′ ) {pµ | µ ∈ OP} forms a linear basis for ΓQ .
(2′ ) {qµ | µ ∈ OP} forms a linear basis for ΓQ .
Proof. Recall the generating function for the power-sums: P(t) = ∑r≥1 pr t r−1 . We
have
Q′ (t)
= (ln Q(t))′ = P(t) + P(−t) = 2 ∑ p2r+1t 2r .
Q(t) r≥0
254 A. Symmetric functions
Proof. We claim that qλ lies in the Z-span of {qµ | µ ∈ SPn , µ ≥ λ}, for each par-
tition λ of a given n. This can be seen by induction downward on the dominance
order on λ as follows. The initial step for λ = (n) is clear, as (n) is strict. For non-
strict λ, we have λi = λ j = m for some i < j and some m > 0. Applying (A.42) to
rewrite qλ easily provides the inductive step.
According to a formula of Euler, |SPn | = |OPn |. Now (A.44) follows from
Part (2) of (A.43) and the above claim. We record here the short proof of Euler’s
formula:
∏ (1 + qr )(1 − qr )
∑ |SPn |qn = ∏(1 + qr ) = r≥1∏r≥1 (1 − qr )
n≥0 r≥1
∏r≥1 (1 − q2r ) 1
= = = ∑ |OPn |qn .
∏r≥1 (1 − qr ) ∏r≥1,odd (1 − qr ) n≥0
(A.45) We have qn = ∑α∈OPn 2ℓ(α) z−1
α pα .
A.3.2. Schur Q-functions. We shall define the Schur Q-functions Qλ , for λ ∈ SP.
Let
Q(n) = qn , n ≥ 1.
Consider the generating function
t1 − t2
Q(t1 ,t2 ) := (Q(t1 )Q(t2 ) − 1) .
t1 + t2
By (A.41), Q(t1 ,t2 ) is a power series in t1 and t2 , and we write
Q(t1 ,t2 ) = ∑ Q(r,s)t1rt2s .
r,s≥0
The following can be checked easily from the definition by noting Q(t1 ,t2 ) =
−Q(t2 ,t1 ).
(A.46) We have Q(r,s) = −Q(s,r) , Q(r,0) = qr . In addition,
s
Q(r,s) = qr qs + 2 ∑ (−1)i qr+i qs−i , r > s.
i=1
λ2n
Remark A.6. Equivalently, Qλ is the coefficient of t1λ1 . . .t2n in
Q(t1 , . . . ,t2n ) := Pf (Q(ti ,t j )).
It can be shown that Q(t1 , . . . ,t2n ) is alternating in the sense that
Q(tσ(1) , . . . ,tσ(2n) ) = sgn(σ)Q(t1 , . . . ,t2n ), ∀σ ∈ S2n .
It follows that Qλ = 0 unless all parts of λ are distinct.
256 A. Symmetric functions
It follows from the recursive relation (A.48) that Qλ is equal to qλ plus a linear
combination of qν for ν > λ (not necessarily strict). Now (A.49) follows since each
qν is expressible as a linear combination of qµ with strict partitions µ ≥ ν by (A.42)
(this fact has been used in the induction step in the proof of (A.44)).
As an immediately corollary of (A.44) and (A.49), we have
(A.50) The Qλ for all strict partitions λ form a Z-basis for Γ. Moreover, for any
partition µ, we have
qµ = ∑ Kbλµ Qλ ,
λ∈SPn ,λ≥µ
where K bλµ ∈ Z and Kbλλ = 1. (Clearly, one can further assume that µ is a composi-
tion instead of a partition in the statement).
bλµ ≥ 0
Indeed from representation-theoretic consideration it can be seen that K
(cf. Chapter 3, Lemma 3.47) .
= ∑ mµ (x)qµ (y).
µ∈P
A.3. The ring Γ and Schur Q-functions 257
Now the first identity follows since pα (xy) = pα (x)pα (y). We further compute that
∞
1 + xi y j
∏ 1 − xi y j = ∏ ∑ qr (y)xir i
i
i, j i ri =0
= ∑ mµ (x)qµ (y).
µ∈P
A rather nontrivial direct proof of (A.54) can be found in Jozefiak [Jo2]. Alter-
natively, one first works on the generality of Hall-Littlewood symmetric functions
Qλ (x;t) for all partitions λ, and then (A.54) can be obtained as a specialization at
t = −1. For details we refer to [Mac].
¿From the Hall-Littlewood approach, the Schur Q-function, for a strict parti-
tion λ with ℓ(λ) = ℓ and m ≥ ℓ, is equal to
( zi + z j )
λℓ
(A.55) Qλ (z1 , . . . , zm ) = 2ℓ ∑ w z λ1
· · · zℓ ∏ ∏ ,
1≤i≤ℓ i< j≤m zi − z j
1
w∈S /S m m−ℓ
where the symmetric group Sm acts by permuting the variables z1 , . . . , zm and Sm−ℓ
is the subgroup acting on zℓ+1 , . . . , zm .
258 A. Symmetric functions
Since the Schur polynomials sλ (z1 , . . . , zm ) form a basis for the space Λm , this
proves the lemma.
Proof. Clearly, the power sums pr for r odd satisfy the above cancelation property,
and so does any element in Γ.
To prove the converse, it suffices to do so in the setting Γm of m variables. We
shall show that if g ∈ Λm is a symmetric polynomial in z1 , . . . , zm such that g|zi =−z j =t
is independent of t, then g can be written as a polynomial in the odd power sums
{p2k+1 | k ∈ Z+ }. Observe that g|zi =−z j =t is independent of t for some particular
choice of i, j with i ̸= j if and only it is independent of t for all (nonidentical) pairs
i, j.
We first note by (A.56) that q ∈ Γm for any symmetric polynomial q. We
proceed by induction on m. If m = 0, 1, then the cancelation property is vacuous.
Assume that m ≥ 2. Let p′2k+1 denote the (2k + 1)st power sum in the variables
z1 , . . . , zm−2 . Then the polynomial g|zm−1 =−zm =0 is symmetric in z1 , . . . , zm−2 , and
it satisfies the cancelation property. By induction hypothesis on m, there exists a
polynomial f such that
g|zm−1 =−zm =0 = f (p′1 , . . . , p′2r+1 ).
A.4. Boson-Fermion Correspondence 259
− − + + + + + + + +
mb : ··· p p p p p p p p p p ···
−3 −2 −1 0 1 2 3 4
− + − + + – + – + +
mc : ··· p p p p p p p p p p ···
−3 −2 −1 0 1 2 3 4
Recall that P denotes the set of all partitions. The following can be viewed as the
Boson-Fermion correspondence at combinatorial level.
(A.59) The map m 7→ (ℓm , λm ) is a bijection between the set of Maya diagrams and
the set Z × P of “charged partitions”.
Proof. We will use primarily the example of mc in Example A.7 to illustrate how
to visualize the Young diagram associated to λmc . Let us draw a new diagram
following the signs on a Maya diagram from far left to far right as follows. We
move to the southeast direction by one unit for each unit interval with a minus
sign on the Maya diagram of mc , and move to northeast by one unit for each unit
interval with a plus sign on. As a result, we obtain the zigzag path in the first
diagram below. The path is labeled by the unit intervals on the Maya diagram.
Extending the two line segments coming from minus infinity and going to plus
infinity, respectively, a finite bounded region emerges. This region can be further
partitioned by solid lines as in the second diagram, and from which we read off the
partition λmc = (3, 2, 2, 1) starting from the southwest row. It is easy to identify the
partition obtained this way for a Maya diagram m with the partition λm defined in
(A.58). The vertical dashed line on the first diagram indicates that the intersection
of the two infinite line segments corresponds to the coordinate 0 separating the two
unit intervals − 21 and 12 . From this we read off the charge ℓmc = 0.
@
@ ·
@ 1 3 5 7 ··
2 @
2 2 @2
..
@ @ @
9
. − 25 − 32 − 12 2
@
−2
7 @
@ @
@
@
@
@
@
0
@
@ ·
@ ··
@
@ @
..
@ @
9
. 2
@ @ @ @ @
−2
7
@ @ @ 2
7
@ @
− 25@ @ @ 2
5
@ @
− 32 @ @ 2
3
@
− 12 @ 1
2
Observe that uniformly shifting every sign on a Maya diagram m one unit in-
terval to the left gives rise to a Maya diagram whose charge is ℓm − 1. For example,
the mb can be obtained from ma by such a uniform shifting to the left twice in a
A.4. Boson-Fermion Correspondence 261
row, and so ℓmb = ℓma − 2. Such a shifting clearly corresponds to a uniform shift-
ing of the labels on the corresponding zigzag path in the first diagram. Hence, it
suffices to show that the set of Maya diagrams with charge 0 is in bijection with the
set of partitions via m 7→ λm . This bijection follows since we can easily reconstruct
the labeled zigzag path (and hence the Maya diagram) from a given partition λ by
first using a dashed diagonal line on the upside-down Young diagram for λ (whose
first southwest row has λ1 boxes) to locate the coordinate 0 point.
A.4.3. Femions and fermionic Fock space. Let F be the vector space with a
linear basis given by the symbols |m⟩ parameterized by the Maya diagrams m.
Note that different Maya diagrams are related by changing signs at finitely
many intervals, and let us consider changing signs to a Maya diagram at one in-
terval at a time. Accordingly, we define the linear operators ψ+ −
n , ψn on F, for
n ∈ 2 + Z, as follows:
1
{
b i , mi+1 , . . .⟩, if mi = −n for some i,
(−1)i−1 | . . . , mi−1 , m
ψn |m⟩ =
+
0, otherwise,
{
(−1) | . . . , mi , n, mi+1 , . . .⟩, if mi < n < mi+1 for some i,
i
ψ−n |m⟩ =
0, otherwise,
where the hat m b i denotes the removal of the term. That is, ψ+
n corresponds to the
−
sign change at −n if the initial sign at −n is +, and ψn corresponds to the sign
change at n if the initial sign at n is −.
Remark A.8. Let V be a vector space with a standard basis vi , i ∈ 21 + Z, and let
{v∗i } denote its dual basis. It is instructive to regard each vector |m⟩ as a semi-
infinite wedge vm := vm1 ∧ vm2 ∧ vm3 ∧ · · · . The signs in the above definition of
ψ±n can be explained naturally by identifying ψn as the contraction operator v−n
+ ∗
and ψ− n as the exterior product with vn . In this way, the Maya diagram model is
equivalent to a semi-infinite wedge model of the fermionic Fock space.
In particular, (ψ+ 2 − 2
n ) = (ψn ) = 0.
262 A. Symmetric functions
for p1 > p2 > . . . > pr > 0 and q1 > q2 > . . . > qs > 0 with r, s ≥ 0. The basis
element (A.63) is equal to m (up to a sign), where m is obtained from 12 + Z+ by
deleting q1 , . . . , qs and adding −p1 , . . . , −pr .
The Fock space F is an irreducible module of the Clifford algebra C generated
by the vacuum vector |0⟩.
Proof. The identification of the vector (A.63) with m up to a sign follows from the
definitions of ψ±
n , which correspond to the additions and removals of the + signs.
The irreducibility follows from the fact that a Maya diagram can be transformed
into another by changing the signs at one interval at a time.
Example A.9. We have |mc ⟩ = −ψ− −
−5/2 ψ−1/2 ψ−7/2 ψ−3/2 |0⟩, for m from Exam-
+ + c
ple A.7.
[Mac, p.3]). For example the modified Frobenius coordinates for the partition
(3, 2, 2, 1) is ( 52 , 12 | 72 , 23 ).
It turns out that the indices pi , q j in (A.63) (at least when r = s) have natural
interpretations in terms of partitions.
(A.64) Let m be a Maya diagram of charge 0. Write (p1 , p2 , . . . , pr |q1 , q2 , . . . , qr )
for the modified Frobenius coordinates for the partition λm . Then, up to a sign,
|m⟩ is equal to ψ− − −
−p1 ψ−p2 . . . ψ−pr ψ−q1 ψ−q2 . . . ψ−qr |0⟩.
+ + +
r2
The sign in (A.64) can be computed to be (−1)q1 +...+qr − 2 .
A.4. Boson-Fermion Correspondence 263
Proof. The proof follows from two observations which work for any general Maya
diagram of charge 0, and we will illustrate below by the running example mc from
Example A.7: (i) In the diagrams appearing in the proof of (A.59), the 2 labels
corresponding to the + sign to the left of the dashed vertical line are − 25 , − 21 , and
the opposites of the 2 labels corresponding to the − sign to the right of the dashed
vertical line are − 32 , − 72 ; they are exactly the indices of the ψ± s in the formula
|mc ⟩ = −ψ− −
−5/2 ψ−1/2 ψ−7/2 ψ−3/2 |0⟩ by (A.62). (ii) The modified Frobenius coor-
+ +
dinates for the partition (3, 2, 2, 1) are ( 52 , 21 | 72 , 32 ), and they precisely correspond to
the 2 + 2 labels specified in (i).
∑ −n− 2
ψ− (z) = ∑ ψ− −n− 2
1 1
(A.65) ψ+ (z) = ψ+
nz , nz .
n∈ 21 +Z n∈ 12 +Z
A.4.4. Charge and energy. We define the energy operator L0 as a linear opera-
tor on F which diagonalizes the basis elements (A.63) with eigenvalue p1 + . . . +
pr + q1 + . . . + qs . In addition, we define a charge operator α0 as a linear operator
on F which diagonalizes the basis elements (A.63) with eigenvalue s − r. The ele-
ment (A.63) will be said of having energy p1 + . . . + pr + q1 + . . . + qs and charge
s − r. We have the following charge decomposition:
⊕
F= F(ℓ) ,
ℓ∈Z
where F(ℓ)
is spanned by the basis vectors (A.63) of charge ℓ. We shall see that the
notion of charge here matches with the notion of charge ℓm for a Maya diagram m.
(A.66) For each ℓ ∈ Z, F(ℓ) has a basis |m⟩ parameterized by the Maya diagrams
m = {m j } j≥1 that satisfy ℓm = ℓ.
For ℓ ∈ Z, we let
+
ψ 12 −ℓ . . . ψ− 32 ψ− 12 |0⟩,
+ +
if ℓ > 0
(A.67) |ℓ⟩ = |0⟩, if ℓ = 0
ψ−1 . . . ψ− 3 ψ− 1 |0⟩,
+ℓ − −
if ℓ < 0.
2 2 2
264 A. Symmetric functions
In particular, |1⟩ = ψ+
−1
|0⟩ and |-1⟩ = ψ−
−1
|0⟩. The following is easily verified.
2 2
(A.68) For each ℓ ∈ Z, the element |ℓ⟩ ∈ F(ℓ) has the minimal energy, which is
ℓ2 /2, among all vectors (A.63) of charge ℓ.
A.4.5. From Boson to Fermion. The Heisenberg Lie algebra Heis is the Lie alge-
bra generated by hk (k ∈ Z) and a cental element c with the commutation relation:
(A.69) [hm , hn ] = mδm,−n c, m, n ∈ Z.
In particular, h0 is central. The Heisenberg algebra Heis acts irreducibly on a
polynomial algebra B(ℓ) := C[p1 , p2 , . . .] in infinitely many variables p1 , p2 , . . ., by
letting h−k (k > 0) act as the multiplication operator by kpk , hk (k > 0) act as the
differentiation operator ddpk , c acts as the identity I, and h0 act as ℓI, for ℓ ∈ C. The
Heis-module B(ℓ) is irreducible, since any nonzero polynomial in B(ℓ) can be trans-
formed to a nonzero constant polynomial by a suitable sequence of differentiation
operators and then produce arbitrary monomials by the multiplication operators.
The Heis-module B(ℓ) is a highest weight module in the sense that hn .1 = 0 for
n > 0, and any highest weight Heis-module generated by a highest weight vector v
such that c.v = v and h0 .v = ℓv is isomorphic to B(ℓ) .
Similarly as in (5.40) we introduce the normal ordered product
{ −
− −ψ+ n ψm , if n = −m < 0
:ψ+ ψ : = −
m n
ψm ψn , otherwise.
+
1
[αk , ψ± ±
n ] = ±ψk+n , k ∈ Z, n ∈ + Z.
2
The identity (A.73) follows by another computation using the commutator identity.
By definition the operators αk have charge 0. Hence, αk (F(ℓ) ) ⊆ F(ℓ) , that is,
F(ℓ) is a Heis-module.
For p, q > 0 and m of charge ℓ, ψ− ′
−p ψ−q |m⟩ = |m ⟩ if it is nonzero, where m
+ ′
On the other hand, note that the energy operator L0 satisfies (and is indeed
characterized by) the following properties: (i) L0 |0⟩ = 0; (ii) [L0 , ψ± ±
−n ] = nψ−n , for
all n ∈ 2 + Z. It follows from (A.71) that
1
Note that αk |ℓ⟩ = 0, for k > 0, and α0 |ℓ⟩ = ℓ|ℓ⟩. Hence, the irreducible Heis-
submodule of F(ℓ) generated by |ℓ⟩ is isomorphic to B(ℓ) and has q-dimension
ℓ2
q2
equal to
∏∞ k , the same as the graded dimension of F(ℓ) . It follows that the
k=1 (1−q )
Heis-module F(ℓ) is irreducible.
together with α0 |0⟩ = 0 implies that α0 here can be identified with the charge
operator defined earlier.
Proof. We compute the trace of the operator qL0 yα0 on F in two ways.
First, by the linear basis (A.62) of F and knowing each fermionic operator ψ±
−ℓ
contributes ℓ to the energy and ±1 to the charge, we have
tr|F qL0 yα0 = ∏ (1 + qn y)(1 + qn y−1 ).
n∈ 12 +Z+
On the other hand, F(ℓ) for each ℓ ∈ Z is identified with the bosonic Fock space
B(ℓ) by (A.72). Using (A.68), we compute that
ℓ2
q 2 yℓ
L0 α0
tr|F q y = ∑ tr|B (ℓ)
L0 α0
q y =∑ ∞ .
ℓ∈Z ∏k=1 (1 − q )
k
ℓ∈Z
The Jacobi triple product identity follows now by equating the two formulas for
the trace and clearing the denominator.
A.5. Notes
Section A.1. The materials on symmetric functions and the Frobenius characteristic
map are fairly standard, and they can be found in Macdonald [Mac] in possibly
different order.
Section A.2. The characterization of supersymmetric polynomials in A.2.1 ap-
peared in Stembridge [St]. The results on super Schur functions and super Cauchy
identity in A.2.2 can be found in [BeR, Sv1].
Section A.3. The materials can be found in Macdonald [Mac] and in Józefiak
[Jo2]. The characterization of the ring Γ in A.3.4 appeared in Pragacz [Pr, Theo-
rem 2.11], and is used in Section 2.3.2.
Section A.4. The materials on Boson-Fermion correspondence are standard.
The use of Maya diagrams follows Miwa-Jimbo-Date [MJD]. Additional applica-
tions of Boson-Fermion correspondence, most notably to soliton equations, can be
found in Kac-Raina [KR] and Miwa-Jimbo-Date [MJD].
Bibliography
267
268 Bibliography
[Car] R. Carter, Lie Algebras of Finite and Affine Type. Cambridge Studies in Advanced Mathe-
matics 96. Cambridge University Press, Cambridge, 2005. xviii+632 pp.
[CK] S.-J. Cheng and J.-H. Kwon, Howe duality and Kostant homology formula for infinite-
dimensional Lie superalgebras, Int. Math. Res. Not. 2008, Art. ID rnn 085, 52 pp.
[CKL] S.-J. Cheng, J.-H. Kwon, and N. Lam, A BGG-type resolution for tensor modules over
general linear superalgebra, Lett. Math. Phys. 84 (2008), 75–87.
[CKW] S.-J. Cheng, J.-H. Kwon, and W. Wang, Kostant homology formulas for oscillator modules
of Lie superalgebras, Adv. Math. 224 (2010), 1548–1588.
[CL1] S.-J. Cheng and N. Lam, Infinite-dimensional Lie superalgebras and hook Schur functions,
Commun. Math. Phys. 238 (2003), 95–118.
[CL2] S.-J. Cheng and N. Lam, Irreducible characters of general linear superalgebra and super
duality, Commun. Math. Phys. 280 (2010), 645–672.
[CLW] S.-J. Cheng, N. Lam and W. Wang, Super duality and irreducible characters of ortho-
symplectic Lie superalgebras, Invent. Math. 183 (2011), 189–224.
[CLW2] S.-J. Cheng, N. Lam and W. Wang, Super duality for general linear Lie superalgebras and
applications, preprint (2011).
[CLZ] S.-J. Cheng, N. Lam, and R.B. Zhang, Character formula for infinite dimensional unitariz-
able modules of the general linear superalgebra, J. Algebra 273 (2004), 780–805.
[CW2] S.-J. Cheng and W. Wang, Remarks on Schur-Howe-Sergeev duality, Lett. Math. Phys. 52
(2000), 143–153.
[CW1] S.-J. Cheng and W. Wang, Howe duality for Lie superalgebras, Compositio Math. 128
(2001), 55–94.
[CW3] S.-J. Cheng and W. Wang, Lie subalgebras of differential operators on the super circle,
Publ. Res. Inst. Math. Sci. 39 (2003), 545–600.
[CW4] S.-J. Cheng and W. Wang, Brundan-Kazhdan-Lusztig and Super Duality Conjectures, Publ.
Res. Inst. Math. Sci. 44 (2008), 1219–1272.
[CW5] S.-J. Cheng and W. Wang, Dualities for Lie superalgebras, Lecture notes for Shanghai
summer school 2009, Surveys of Modern Mathematics 2, International Press and Higher
Education Press, pp. 1–46, arXiv:1001.0074.
[CWZ2] S.-J. Cheng, W. Wang and R.B. Zhang, A Fock space approach to representation theory of
osp(2|2n), Transform. Groups 12 (2007), 209–225.
[CWZ] S.-J. Cheng, W. Wang, and R.B. Zhang, Super duality and Kazhdan-Lusztig polynomials,
Trans. Amer. Math. Soc. 360 (2008) 5883–5924.
[CZ1] S.-J. Cheng and R.B. Zhang, Analogue of Kostant’s u-homology formula for general linear
Superalgebras, Int. Math. Res. Not. 2004, 31–53.
[CZ2] S.-J. Cheng and R.B. Zhang, Howe duality and combinatorial character formula for or-
thosymplectic Lie superalgebras, Adv. Math. 182 (2004), 124–172.
[DJKM] E. Date, M. Jimbo, M. Kashiwara and T. Miwa: Transformation groups for soliton equa-
tions. III. Operator approach to the Kadomtsev-Petviashvili equation, J. Phys. Soc. Japan
50 (1981), 3806–3812. Transformation groups for soliton equations. IV. A new hierarchy of
soliton equations of KP-type, Phys. D 4 (1981/82), 343–365.
[DES] M. Davidson, E. Enright, R. Stanke, Differential Operators and Highest Weight Represen-
tations, Mem. Amer. Math. Soc. 94 (1991) no. 455.
[Don] S. Donkin, On tilting modules for algebraic groups, Math. Z. 212 (1993), 39–60.
[En] T. Enright, Analogues of Kostant’s u-cohomology formulas for unitary highest weight mod-
ules, J. Reine Angew. Math. 392 (1988), 27–36.
Bibliography 269
[FSS] L. Frappat, A. Sciarrino and P. Sorba, Dictionary on Lie algebras and superalgebras. Aca-
demic Press, Inc., San Diego, CA, 2000.
[Fr] I. Frenkel, Representations of affine Lie algebras, Hecke modular forms and Kortweg-de
Vries type equations, Lect. Notes Math. 933 (1982), 71–110.
[Ger] J. Germoni, Indecomposable representations of osp(3, 2), D(2, 1; α) and G(3). Colloquium
on Homology and Representation Theory (Spanish) (Vaquerias, 1998). Bol. Acad. Nac.
Cienc. (Cordoba) 65 (2000), 147–163.
[GW] R. Goodman and N. Wallach, Representations and invariants of the classical groups. Ency-
clopedia of Mathematics and its Applications, 68. Cambridge University Press, Cambridge,
1998.
[Go] M. Gorelik, The Kac construction of the centre of U(g) for Lie superalgebras, J. Nonlinear
Math. Phys. 11 (2004), 325–349.
[GS] C. Gruson and V. Serganova, Cohomology of generalized supergrassmannians and char-
acter formulae for basic classical Lie superalgebras, Proc. Lond. Math. Soc. 101 (2010),
852–892.
[H1] R. Howe, Remarks on classical invariant theory, Trans. Amer. Math. Soc. 313 (1989), 539–
570.
[H2] R. Howe, Perspectives on invariant theory: Schur duality, multiplicity-free actions and
beyond, The Schur Lectures, Israel Math. Conf. Proc. 8, Tel Aviv (1992), 1–182.
[HLT] P.-Y. Huang, N. Lam, and T.-M. Tu, Super duality and homology of unitarizable modules
of Lie algebras, Pub. RIMS (to appear), arXiv:1012.1087.
[Hum] J. E. Humphreys, Representations of Semisimple Lie Algebras in the BGG Category O,
Graduate Studies in Mathematics, 94. American Mathematical Society, Providence, RI,
2008.
[IK] K. Iohara and Y. Koga, Second homology of Lie superalgebras, Math. Nachr. 278 (2005),
1041–1053.
[Jo1] T. Józefiak, Semisimple superalgebras, In: Algebra–Some Current Trends (Varna, 1986),
pp. 96–113, Lect. Notes in Math. 1352, Springer-Berlag, Berlin-New York, 1988.
[Jo2] T. Józefiak, Characters of projective representations of symmetric groups, Expo. Math. 7
(1989), 193–247.
[Jo3] T. Józefiak, A class of projective representations of hyperoctahedral groups and Schur Q-
functions, Topics in Algebra, Banach Center Publ., 26, Part 2, PWN-Polish Scientific Pub-
lishers, Warsaw (1990), 317–326.
[K4] V. Kac, Classification of simple Lie superalgebras (in Russian), Funkcional. Anal. i Priložen
9 (1975), 91–92.
[K1] V. Kac, Lie superalgebras, Adv. Math. 26 (1977), 8–96.
[K2] V. Kac, Representations of classical Lie superalgebras. Differential geometrical methods
in mathematical physics, II (Proc. Conf., Univ. Bonn, Bonn, 1977), pp. 597–626, Lecture
Notes in Math. 676, Springer, Berlin, 1978.
[K5] V. Kac, Laplace operators of infinite-dimensional Lie algebras and theta functions, Proc.
Nat. Acad. Sci. USA 81 (1984), 645–647.
[K3] V. Kac, Infinite dimensional Lie algebras. Third edition. Cambridge University Press, Cam-
bridge, 1990.
[KR] V. Kac and A. Raina, Bombay Lectures on Highest Weight Representations of Infinite Di-
mensional Lie Algebras, Advanced Series in Mathematical Physics 2. World Scientific Pub-
lishing Co., Inc., Teaneck, NJ, 1987.
270 Bibliography
[KW] V. Kac and M. Wakimoto, Integrable highest weight modules over affine superalgebras and
Appell’s function, Commun. Math. Phys. 215 (2001), 631–682.
[KWY] V. Kac, W. Wang, and C. Yan, Quasifinite representations of classical Lie subalgebras of
W1+∞ , Adv. Math. 139 (1998) 56–140.
[KK] S.-J. Kang and J.-H. Kwon, Graded Lie superalgebras, supertrace formula, and orbit Lie
superalgebras, Proc. London Math. Soc. 81 (2000), 675–724.
[KaV] M. Kashiwara and M. Vergne, On the Segal-Shale-Weil representations and harmonic poly-
nomials, Invent. Math. 44 (1978), 1–47.
[KL] D. Kazhdan and G. Lusztig, Representations of Coxeter groups and Hecke algebras, Invent.
Math. 53 (1979), 165–184.
[KV] A. Knapp and D. Vogan, Cohomological Induction and Unitary Representations, Princeton
Mathematical Series, 45. Princeton University Press, Princeton, NJ, 1995.
[Kos] B. Kostant, Lie Algebra Cohomology and the generalized Borel-Weil Theorem,
Ann. Math. 74 (1961), 329–387.
[KP] H. Kraft and C. Procesi, Classical Invariant Theory, A Primer. 1996, 128 pp. Available at
http://www.math.unibas.ch/ kraft/Papers/KP-Primer.pdf.
[Ku] S. Kumar, Kac-Moody groups, their flag varieties and representation theory. Progress in
Mathematics, 204. Birkhauser Boston, Inc., Boston, MA, 2002.
[LZ] N. Lam and R.B. Zhang, Quasi-finite modules for Lie superalgebras of infinite rank,
Trans. Amer. Math. Soc. 358 (2006), 403–439.
[LSS] D. Leites, M. Saveliev, and V. Serganova, Embedding of osp(N/2) and the associated non-
linear supersymmetric equations. Group theoretical methods in physics, Vol. I (Yurmala,
1985), 255–297, VNU Sci. Press, Utrecht, 1986.
[Liu] L. Liu, Kostant’s Formula for Kac-Moody Lie Algebras, J. Algebra 149 (1992), 155–178.
[Mac] I. G. Macdonald, Symmetric functions and Hall polynomials, Second Edition, Oxford Math-
ematical Monographs. Oxford Science Publications. The Clarendon Press, Oxford Univer-
sity Press, New York, 1995.
[Ma] Yu. Manin, Gauge field theory and complex geometry, Grundlehren der mathematischen
Wissenschaften 289, Second Edition, Springer-Verlag, Berlin, 1997.
[MM] J. Milnor and J. Moore, On the Structure of Hopf Algebras, Ann. Math. 81 (1965), 211–264.
[Mi] B. Mitchell, Theory of categories, Pure and Applied Mathematics XVII, Academic Press,
New York-London 1965.
[MJD] T. Miwa, M. Jimbo, and E. Date, Solitons. Differential equations, symmetries and infinite-
dimensional algebras, Cambridge Tracts in Mathematics 135. Cambridge University Press,
Cambridge, 2000.
[MJ] E. Moens and J. van der Jeugt, A determinantal formula for supersymmetric Schur polyno-
mials, J. Algebraic Combin. 17 (2003), 283–307.
[Mu] I. Musson, Lie superalgebras, Clifford algebras, induced modules and nilpotent orbits,
Adv. Math. 207 (2006), 39–72.
[Naz] M. Nazarov, Capelli identities for Lie superalgebras, Ann. Sci. École Norm. Sup. 30 (1997),
847–872.
[Pe] I. Penkov, Characters of typical irreducible finite-dimensional q(n)-modules, Funct. Anal.
App. 20 (1986), 30–37.
[PS] I. Penkov and V. Serganova, Cohomology of G/P for classical complex Lie supergroups G
and characters of some atypical G-modules, Ann. Inst. Fourier 39 (1989), 845–873.
[PS2] I. Penkov and V. Serganova, Characters of finite-dimensional irreducible q(n)-modules,
Lett. Math. Phys. 40 (1997), 147–158.
Bibliography 271
[Po] N. Popescu, Abelian categories with applications to rings and modules. London Mathemat-
ical Society Monographs 3, Academic Press, London-New York, 1973.
[Pr] P. Pragacz, Algebro-geometric applications of Schur S- and Q-polynomials. Topics in in-
variant theory (Paris, 1989/1990), 130–191, Lecture Notes in Math. 1478, Springer, Berlin,
1991.
[Re] E. W. Read, The α-regular classes of the generalized symmetric groups, Glasgow Math. J.
17 (1976), 144–150.
[RW] A. Rocha-Caridi and N. Wallach: Projective modules over graded Lie algebras I, Math.
Z. 180 (1982), 151–177.
[Ro] L. Ross, Representations of graded Lie algebras, Trans. Amer. Math. Soc. 120 (1965) 17–
23.
[Sa1] B. Sagan, Shifted tableaux, Schur Q-functions, and a conjecture of R. Stanley, J. Com-
bin. Theory Ser. A 45 (1981), 62–103.
[San] J. Santos, Foncteurs de Zuckermann pour les superalgébres de Lie, J. Lie Theory 9 (1999),
69–112.
[Sch] M. Scheunert, The Theory of Lie superalgebras, Lect. Notes in Math. 716. Springer, Berlin,
1979.
[SNR] M. Scheunert, W. Nahm, and V. Rittenberg, Classification of all simple graded Lie algebras
whose Lie algebra is reductive I, II. Construction of the exceptional algebras, J. Math. Phys.
17 (1976), 1626–1639, 1640–1644.
[Sva] V. Serganova, Kazhdan-Lusztig polynomials and character formula for the Lie superalge-
bra gl(m|n), Selecta Math. (N.S.) 2 (1996), 607–651.
[Sva2] V. Serganova, Kac-Moody superalgebras and integrability, In: Developments and trends in
infinite-dimensional Lie theory, 169–218, Progr. Math. 288, Birkhäuser, 2011.
[Sv4] A. Sergeev, The centre of enveloping algebra for Lie superalgebra Q(n, C).
Lett. Math. Phys. 7 (1983), no. 3, 177–179.
[Sv1] A. Sergeev, The tensor algebra of the identity representation as a module over the Lie
superalgebras gl(n, m) and Q(n), Math. USSR Sbornik 51 (1985), 419–427.
[Sv3] A. Sergeev, The invariant Polynomials of simple Lie superalgebras, Represent. Theory 3
(1999), 250–280 (electronic).
[Sv2] A. Sergeev, An Analog of the Classical Invariant Theory, I, II, Michigan J. Math. 49 (2001),
113–146, 147–168.
[SVe] A. Sergeev and A. Veselov, Grothendieck rings of basic classical Lie superalgebras, Annals
of Math. 173 (2011), 663-703.
[SW] B. Shu and W. Wang, Modular representations of the ortho-symplectic supergroups, Proc.
London Math. Soc. 96 (2008), 251–271.
[So] W. Soergel, Character formulas for tilting modules over Kac-Moody algebras, Represent.
Theory (electronic) 2 (1998), 432–448.
[St] J. Stembridge, A characterization of supersymmetric polynomials, J. Algebra 95 (1985),
439–444.
[Su] Y. Su, Composition factors of Kac modules for the general linear Lie superalgebras, Math.
Z. 252 (2006), 731–754.
[SZ] Y. Su and R.B. Zhang, Character and dimension formulae for general linear superalgebra,
Adv. Math. 211 (2007), 1–33.
[Ta] J. Tanaka, On homology and cohomology of Lie superalgebras with coefficients in their
finite-dimensional representations, Proc. Japan Acad. Ser. A Math. Sci. 71 (1995), 51–53.
272 Bibliography
[J] J. Van der Jeugt, Character formulae for Lie superalgebra C(n), Comm. Algebra 19,
(1991), 199–222.
[JHKT] J. Van der Jeugt, J.W.B. Hughes, R. C. King and J. Thierry-Mieg, Character formulas for
irreducible modules of the Lie superalgebras sl(m/n), J. Math. Phys. 31 (1990), 2278–
2304.
[Vo] D. Vogan, Irreducible characters of semisimple Lie Groups II: The Kazhdan-Lusztig Con-
jectures, Duke Math. J. 46 (1979), 805–859.
[Wal] C.T.C. Wall, Graded Brauer groups, J. Reine Angew. Math. 213 (1964), 187–199.
[Wa] W. Wang, Duality in infinite dimensional Fock representations, Commun. Contem. Math. 1
(1999), 155–199.
[WZ] W. Wang and L. Zhao, Representations of Lie superalgebras in prime characteristic I, Proc.
London Math. Soc. 99 (2009), 145–167.
[We] H. Weyl, The classical groups. Their invariants and representations. Fifteenth printing.
Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ, 1997.
[Zou] Y. Zou, Categories of finite-dimensional weight modules over type I classical Lie superal-
gebras, J. Algebra 180 (1996), 459–482.
Index
εδ-sequence, 22 degree
atypicality, 62, 70
even, 2
adjoint action, 3
odd, 2
adjoint map, 3
derivation, 3
algebra
diagram
Clifford, 141
head, 188
Hecke-Clifford, 106
master, 188
univeral enveloping, 30
Maya, 259
Weyl, 141
charge of, 259
Weyl-Clifford, 141
skew, 246
tail, 188
bilinear form dimension, 2
even, 6 duality
invariant, 3 super, 230
odd, 6
skew-supersymmetric, 6 energy, 263
supersymmetric, 6 extension, 223
boundary, 214 exterior algebra, 10
273
274 Index
T , 209 module
T , 209 Kac, 40
duality, 216 oscillator
truncation, 202 osp(2m|2n), 162
fundamental system, 19 spo(2m|2n), 167
parabolic Verma, 199
group polynomial, 100, 198
orthogonal, 127 spin, 89
symplectic, 127 basic, 109
type M, 86
harmonic, 161 type Q, 86
highest weight, 32 multiplicity-free
highest weight vector, 33 strongly, 146
homology, 214
Kostant, 216 nilradical, 199
homomorphism opposite, 199
Harish-Chandra, 52 normal ordered product, 171
Lie superalgebra, 3
module, 2 odd reflection, 26
Howe dual pair, 148 odd trace, 18
operator
ideal, 2 boundary, 214
charge, 263
Jacobi triple product identity, 266 coboundary, 215
energy, 263
Kac module, 40
Kazhdan-Lusztig-Vogan polynomial, 220 parity, 2
partition, 239
level, 169 size of, 239
Lie superalgebra, 3 dominance order, 239
e
g, g, g, 196 even, 247
basic, 13 generalized, 174
Cartan type hook, 45
H(n), 12 length of, 239
S(n), 11 odd, 103
W (n), 10 part of, 239
e
S(n), 11 strict, 103
classical, 13 Pfaffian, 255
exceptional Pieri’s formula, 247
D(2, 1, α), 9 Poincaré-Birkhoff-Witt Theorem, 30
F(4), 10 polarization, 123
G(3), 10 polynomial
general linear, 4 doubly symmetric, 249
ortho-symplectic, 6 Kazhdan-Lusztig-Vogan, 220
periplectic, 9 Schur, 243
queer, 8 supersymmetric, 249
solvable, 31 positive system, 19
special linear, 5
type a, 190 reflection
type b• , 191 odd, 26
type b, 192 restitution, 123
type c, 191 restricted dual, 215
type d, 192 ring of symmetric functions, 240
linkage principle root
gl,osp, 62 even, 14
q, 70 isotropic, 15
Littlewood-Richardson coefficient, 246 odd, 14
Index 275
simple, 19 half-integer, 44
root system, 13 highest, 32
integer, 44
Schur function, 243 linked
skew, 246 gl,osp, 60
super, 252 q, 69
Schur polynomial, 243 polynomial, 100
strongly multiplicity-free, 146 Weyl group, 14
subalgebra, 2 Weyl symbol, 142
Borel, 20 Weyl vector, 20
Cartan, 13
Levi, 199
parabolic, 199
subspace, 2
super duality, 230
superalgebra, 2
semisimple, 86
simple, 2
supercharacter, 51
superdimension, 2
supergroup, 88
superspace, 2
supertableau, 94
content of, 94
supertrace, 5
supertranspose, 5
symbol map, 143
symmetric function
complete, 240
elementary, 240
involution ω, 242
monomial, 240
power, 240
ring of, 240
system
fundamental, 19
positive, 19
root, 13
tableau, 246
semistandard, 246
trace
odd, 18
typical, 62, 70
vector
highest weight, 33
singular, 65
vacuum, 261
virtual character, 42
weight
dominant, 199, 202
dominant integral, 41
extremal, 71