On the Unicity and the Ambiguity of Lusztig Parametrizations for Finite Classical Groups

Let G be a classical group defined over a finite field ${\mathbf{F}}_{q}$ of odd characteristic, and let $F$ be the corresponding Frobenius endomorphism. Let $G=\text{\bf G}^{F}$ denote the finite subgroup of rational points, and let ${\mathcal{E}}(\text{\bf G})$ denote the set of irreducible characters (i.e., the characters of irreducible representations) of $G$.

Let $R^{\text{\bf G}}_{\text{\bf T}^{*},s}$ denote the Deligne-Lusztig virtual characters (cf. [Car85], [GM20]) indexed by conjugacy class of pair $(\text{\bf T}^{*},s)$ where $\text{\bf T}^{*}$ is a rational maximal torus in the dual group $\text{\bf G}^{*}$ and $s$ is a rational semisimple element contained in $\text{\bf T}^{*}$. Let ${\mathcal{V}}(\text{\bf G})$ be the space of (complex valued) class function on $G$, and let ${\mathcal{V}}(\text{\bf G})^{\sharp}$ denote the subspace spanned by Deligne-Lusztig virtual characters. Note that ${\mathcal{V}}(\text{\bf G})$ is an inner product space with the inner product $\langle,\rangle_{\text{\bf G}}$ given by

for $f_{1},f_{2}\in{\mathcal{V}}(\text{\bf G})$, and $\overline{f_{2}(g)}$ denotes the complex conjugate of the value $f_{2}(g)$. For $f\in{\mathcal{V}}(\text{\bf G})$ the orthogonal projection of $f$ onto ${\mathcal{V}}(\text{\bf G})^{\sharp}$ is denoted by $f^{\sharp}$ and called the uniform projection. A class function $f\in{\mathcal{V}}(\text{\bf G})$ is called uniform if $f^{\sharp}=f$.

A natural question is how much an irreducible character $\rho$ of a classical group $G$ can be determined by its uniform projection $\rho^{\sharp}$?

If G is a general linear group ${\rm GL}_{n}$ or a unitary group ${\rm U}_{n}$, then every irreducible character is uniform (i.e., $\rho=\rho^{\sharp}$) and the above question is trivial. However, if G is a symplectic group or an orthogonal group, the question is more subtle. Recall that $\rho\in{\mathcal{E}}(\text{\bf G})$ is called unipotent if $\langle\rho,R^{\text{\bf G}}_{\text{\bf T}^{*},1}\rangle_{\text{\bf G}}\neq 0$ for some $\text{\bf T}^{*}$. If $\rho$ is unipotent and G is connected, it is known that $\rho$ is uniquely determined by its uniform projection, i.e., $\rho^{\prime\sharp}=\rho^{\sharp}$ if and only if $\rho^{\prime}=\rho$ (cf. [DM90] proposition 6.3 and [GM20] theorem 4.4.23). If $\rho$ is unipotent and $\text{\bf G}={\rm O}^{\epsilon}_{2n}$, it is also known that $\rho^{\prime\sharp}=\rho^{\sharp}$ if and only if $\rho^{\prime}=\rho$ or $\rho^{\prime}=\rho\cdot{\rm sgn}$ (cf. [Pan19b] proposition 3.6).

In this paper, the above question will be answered completely (cf. Corollary 7.5, Corollary 9.4 and Corollary 8.4) for classical groups:

Here “${\rm sgn}$” denotes the sign character of an orthogonal group, and “$\rho^{c}$” denotes the character obtained from $\rho$ by conjugating an element in the corresponding similitude group (cf. [Wal04] §4.3, §4.10). Note that if $\rho$ is unipotent then $\rho^{c}=\rho$, and then (1.2) is reduced to the above known result.

In fact, the above theorem is a consequence of a more precise result on the ambiguity of the Lusztig parametrization of irreducible characters of finite classical groups. From now on, we assume that G is a symplectic group or an orthogonal group. It is known by Lusztig that there is a partition

where the union $\bigcup_{(s)}$ runs over conjugacy classes of semisimple elements in the connected component of the dual group $G^{*}$ and

Each ${\mathcal{E}}(\text{\bf G},s)$ is called a Lusztig series. In particular, the subset ${\mathcal{E}}(\text{\bf G},1)$ consists of unipotent characters.

Now we first focus on the set of unipotent characters. Let ${\mathcal{V}}(\text{\bf G},1)$ denote the subspace spanned by elements in ${\mathcal{E}}(\text{\bf G},1)$, and let ${\mathcal{V}}(\text{\bf G},1)^{\sharp}$ denote the uniform projection of ${\mathcal{V}}(\text{\bf G},1)$. Following from Lusztig (cf. [Lus81], [Lus82]) we can define a set ${\mathcal{S}}_{\text{\bf G}}^{\sharp}$ (cf. (3.7)) and class functions $R_{\Sigma}^{\text{\bf G}}$ (cf. (3.5)) for $\Sigma\in{\mathcal{S}}_{\text{\bf G}}^{\sharp}$ such that the set $\{\,R_{\Sigma}^{\text{\bf G}}\mid\Sigma\in{\mathcal{S}}_{\text{\bf G}}^{\sharp}\,\}$ forms an orthonormal basis for ${\mathcal{V}}(\text{\bf G},1)^{\sharp}$. Lusztig constructs a set ${\mathcal{S}}_{\text{\bf G}}$ (cf. (2.7)) of similar classes of symbols and a bijective mapping ${\mathcal{L}}_{1}^{\text{\bf G}}\colon{\mathcal{S}}_{\text{\bf G}}\rightarrow{\mathcal{E}}(\text{\bf G},1)$ denoted by $\Lambda\mapsto\rho_{\Lambda}$ such that the value $\langle\rho_{\Lambda},R^{\text{\bf G}}_{\Sigma}\rangle_{\text{\bf G}}$ is specified (cf. Proposition 3.12). Such a mapping ${\mathcal{L}}_{1}^{\text{\bf G}}$ is called a Lusztig parametrization of unipotent characters of G. Because the uniform projection $\rho_{\Lambda}^{\sharp}$ can be obtained by the values $\langle\rho_{\Lambda},R^{\text{\bf G}}_{\Sigma}\rangle_{\text{\bf G}}$, the problem of the uniqueness of ${\mathcal{L}}_{1}^{\text{\bf G}}$ is equivalent to the problem whether a unipotent character $\rho_{\Lambda}$ can be uniquely determined by its uniform projection $\rho_{\Lambda}^{\sharp}$.

As described in Subsection 1.1, ${\mathcal{L}}_{1}^{\text{\bf G}}$ is known to be unique if $\text{\bf G}={\rm SO}_{2n+1}$ or ${\rm Sp}_{2n}$. However, due to the disconnectedness of ${\rm O}^{\epsilon}_{2n}$, the mapping ${\mathcal{L}}_{1}^{{\rm O}^{\epsilon}_{2n}}$ is not uniquely determined. By using the result of “cells” by Lusztig, we determine how ambiguous a Lusztig parametrization ${\mathcal{L}}_{1}^{{\rm O}^{\epsilon}_{2n}}$ could be and we show in Proposition 5.9 that ${\mathcal{L}}_{1}^{{\rm O}^{\epsilon}_{2n}}$ can be chosen to be unique if we require it to be

• •

  compatible with parabolic induction and

• •

  compatible with theta correspondence on cuspidal characters or ${\bf 1}_{{\rm O}^{+}_{2}},{\rm sgn}_{{\rm O}^{+}_{2}}$.

In particular, ${\mathcal{L}}_{1}^{{\rm O}^{\epsilon}_{2n}}$ (and ${\mathcal{L}}_{1}^{{\rm Sp}_{2n^{\prime}}}$) can be given so that $(\rho_{\Lambda},\rho^{\prime}_{\Lambda^{\prime}})\in\Theta_{\text{\bf G},\text{\bf G}^{\prime}}$ if and only if $(\Lambda,\Lambda^{\prime})\in{\mathcal{B}}_{\text{\bf G},\text{\bf G}^{\prime}}$ for $(\text{\bf G},\text{\bf G}^{\prime})=({\rm O}^{\epsilon}_{2n},{\rm Sp}_{2n^{\prime}})$ where ${\mathcal{B}}_{\text{\bf G},\text{\bf G}^{\prime}}$ is a relation between ${\mathcal{S}}_{\text{\bf G}}$ and ${\mathcal{S}}_{\text{\bf G}^{\prime}}$ defined in Subsection 5.2.

For a general Lusztig series ${\mathcal{E}}(\text{\bf G},s)$, Lusztig shows (cf. [Lus84]) that there exists a bijection

satisfying

where $\epsilon_{\text{\bf G}}=(-1)^{\kappa(\text{\bf G})}$, $\kappa(\text{\bf G})$ denotes the rational rank of G, and $C_{\text{\bf G}^{*}}(s)$ denotes the centralizer of $s$ in $\text{\bf G}^{*}$. Such a bijection ${\mathfrak{L}}_{s}$ will be called a Lusztig correspondence in this article (it is called a Jordan decomposition in [GM20]).

Now the question is to understand whether the Lusztig correspondence ${\mathfrak{L}}_{s}$ is uniquely determined by (1.3). If the answer is negative, then we want to know what kind of conditions need to be enforced to make ${\mathfrak{L}}_{s}$ unique. Some discussion on this problem can be founded in [GM20] appendix A.5. If G is a connected group with connected center, it is shown in [DM90] theorem 7.1 that ${\mathfrak{L}}_{s}$ can be uniquely determined by (1.3) and some extra conditions.

For $s\in G^{*}$, we have a decomposition $s=s^{(0)}\times s^{(1)}\times s^{(2)}$ where $s^{(1)}$ (resp. $s^{(2)}$) is the part whose eigenvalues are all equal to $-1$ (resp. $1$), and $s^{(0)}$ is the part whose eigenvalues do not contain $1$ or $-1$. Then we can define groups $\text{\bf G}^{(0)}(s)$, $\text{\bf G}^{(-)}(s)$ and $\text{\bf G}^{(+)}(s)$ (cf. (6.12)) so that there is a bijective mapping

Combining Lusztig parametrization ${\mathcal{L}}_{1}$ of unipotent characters for $\text{\bf G}^{(0)}$, $\text{\bf G}^{(-)}(s)$, $\text{\bf G}^{(+)}(s)$, and above bijection and the inverse ${\mathfrak{L}}_{s}^{-1}$ of a Lusztig correspondence, we obtain a bijective mapping

denoted by $(x,\Lambda_{1},\Lambda_{2})\mapsto\rho_{x,\Lambda_{1},\Lambda_{2}}$ which is called a modified Lusztig correspondence. Then we prove the following results on the unicity and ambiguity of ${\mathcal{L}}_{s}$ (or equivalently, the unicity and ambiguity of ${\mathfrak{L}}_{s}$) for classical groups:

1. (1)

   Suppose that $\text{\bf G}={\rm SO}_{2n+1}$. There is a unique modified Lusztig correspondence ${\mathcal{L}}_{s}$ (cf. Theorem 7.1). Note that ${\rm SO}_{2n+1}$ is a connected group with connected center, so this case is covered by [DM90] theorem 7.1. However, for this group, no extra condition other than (1.3) is needed to ensure the uniqueness of ${\mathcal{L}}_{s}$.

2. (2)

   Suppose that $\text{\bf G}={\rm O}^{\epsilon}_{2n}$ where $\epsilon=+$ or $-$. There exists a unique modified Lusztig correspondence ${\mathcal{L}}_{s}$ which is compatible with the parabolic induction and some other conditions (on basic characters). (cf. Theorem 8.12).

3. (3)

   Suppose that $\text{\bf G}={\rm Sp}_{2n}$. Here we provide two choices of the modified Lusztig correspondence ${\mathcal{L}}_{s}$:

   1. (a)

      There exists a unique modified Lusztig correspondence ${\mathcal{L}}_{s}$ which is compatible with the parabolic induction and compatible with the theta correspondence for the dual pair $(\text{\bf G},\text{\bf G}^{\prime})=({\rm Sp}_{2n},{\rm SO}_{2n^{\prime}+1})$, i.e., we show that $(\rho_{x,\Lambda_{1},\Lambda_{2}},\rho_{x^{\prime},\Lambda_{1}^{\prime},\Lambda_{2}^{\prime}})\in\Theta_{\text{\bf G},\text{\bf G}^{\prime}}^{\psi}$ if and only if

      • •

        $s^{(0)}=-s^{\prime(0)}$ and $x=x^{\prime}$,

      • •

        $\Lambda_{2}=\Lambda^{\prime}_{1}$, and

      • •

        $(\Lambda_{1},\Lambda^{\prime}_{2})\in{\mathcal{B}}_{\text{\bf G}^{(-)}(s),\text{\bf G}^{(+)}(s^{\prime})}$

      where $\rho_{x^{\prime},\Lambda_{1}^{\prime},\Lambda_{2}^{\prime}}$ is given by the unique modified Lusztig correspondence in (1) (cf. Theorem 9.7 and Theorem 9.9).

   2. (b)

      There exists a unique modified Lusztig correspondence ${\mathcal{L}}_{s}$ which is compatible with the parabolic induction and compatible with the theta correspondence for the dual pair $(\text{\bf G},\text{\bf G}^{\prime})=({\rm Sp}_{2n},{\rm O}^{\epsilon}_{2n^{\prime}})$, i.e., $(\rho_{x,\Lambda_{1},\Lambda_{2}},\rho_{x^{\prime},\Lambda_{1}^{\prime},\Lambda_{2}^{\prime}})\in\Theta_{\text{\bf G},\text{\bf G}^{\prime}}^{\psi}$ if and only if

      • •

        $s^{(0)}=s^{\prime(0)}$ and $x=x^{\prime}$,

      • •

        $\Lambda_{1}=\Lambda^{\prime}_{1}$, and

      • •

        $(\Lambda_{2},\Lambda^{\prime}_{2})\in{\mathcal{B}}_{\text{\bf G}^{(+)}(s),\text{\bf G}^{(+)}(s^{\prime})}$

      where $\rho_{x^{\prime},\Lambda_{1}^{\prime},\Lambda_{2}^{\prime}}$ is given by the unique modified Lusztig correspondence in (2) (cf. Theorem 9.11 and Theorem 9.12).

   It seems that two choices in (a) and (b) of the modified Lusztig correspondence ${\mathcal{L}}_{s}$ for ${\rm Sp}_{2n}$ should be the same. However, we do not know how to obtain the conclusion yet.

The contents of this article are as follows. In Section 2, we recall the notion and some basic result of “symbols” and “special symbols” by Lusztig from [Lus77]. In Section 3 we first recall the notion of “almost characters” by Lusztig from [Lus81] and [Lus82]. Then we record some results of cells from [Pan19a]. In Section 4 we show the uniqueness of ${\mathcal{L}}_{1}^{\text{\bf G}}$ for $\text{\bf G}={\rm Sp}_{2n}$ and ${\rm SO}_{2n+1}$ by using the results in the previous section. Moreover, we also discuss the ambiguity of ${\mathcal{L}}_{1}^{\text{\bf G}}$ for $\text{\bf G}={\rm O}^{\epsilon}_{2n}$. In Section 5 we discuss the relation between the theta correspondence $\Theta^{\psi}_{\text{\bf G},\text{\bf G}^{\prime}}$ on unipotent characters for $(\text{\bf G},\text{\bf G}^{\prime})=({\rm Sp}_{2n},{\rm O}^{\epsilon}_{2n^{\prime}})$ and Lusztig parametrizations ${\mathcal{L}}_{1}^{\text{\bf G}}\colon{\mathcal{S}}_{\text{\bf G}}\rightarrow{\mathcal{E}}(\text{\bf G},1)$ for $\text{\bf G}={\rm Sp}_{2n},{\rm O}^{\epsilon}_{2n}$. In Section 6 we discuss the relation between the theta correspondence $\Theta^{\psi}_{\text{\bf G},\text{\bf G}^{\prime}}$ on certain Lusztig series for $(\text{\bf G},\text{\bf G}^{\prime})=({\rm Sp}_{2n},{\rm SO}_{2n^{\prime}+1})$ or $({\rm Sp}_{2n},{\rm O}^{\epsilon}_{2n^{\prime}})$ and the Lusztig correspondence ${\mathfrak{L}}_{s}\colon{\mathcal{E}}(\text{\bf G},s)\rightarrow{\mathcal{E}}(C_{\text{\bf G}^{*}}(s),1)$. In Section 7 we show that the Lusztig correspondence ${\mathfrak{L}}_{s}\colon{\mathcal{E}}(\text{\bf G},s)\rightarrow{\mathcal{E}}(C_{\text{\bf G}^{*}}(s),1)$ is unique for $\text{\bf G}={\rm SO}_{2n+1}$. In Section 8 we show that ${\mathfrak{L}}_{s}$ can be chosen to be unique for $\text{\bf G}={\rm O}^{\epsilon}_{2n}$ if we require ${\mathfrak{L}}_{s}$ to be compatible with the parabolic induction and some other conditions on “basic characters”. In the final section we discuss the uniqueness of the Lusztig correspondence ${\mathfrak{L}}_{s}$ for $\text{\bf G}={\rm Sp}_{2n}$. In particular, we show that a unique ${\mathfrak{L}}_{s}$ can be chosen to be compatible with the theta correspondence for the dual pair $({\rm Sp}_{2n},{\rm SO}_{2n+1})$ or for the dual pair $({\rm Sp}_{2n},{\rm O}^{\epsilon}_{2n})$.

Let ${\mathcal{P}}(n)$ denote the set partitions of $n$. It is known that the set of irreducible characters ${\mathcal{E}}(S_{n})$ of the symmetric group $S_{n}$ is parametrized by ${\mathcal{P}}(n)$. For $\lambda\in{\mathcal{P}}(n)$, we write $|\lambda|=n$, and the corresponding irreducible character of $S_{n}$ is denoted by $\varphi_{\lambda}$.

Let $W_{n}$ denote the Coxeter group of type $B_{n}$, i.e., $W_{n}$ consists of all permutations on $\{1,2,\ldots,n,n^{*},(n-1)^{*},\ldots,1^{*}\}$ which commutes with the involution

where $(i,j)$ denote the transposition of $i,j$. For $i=1,\ldots,n-1$, let

It is known that $W_{n}$ is generated by $\{s_{1},\ldots,s_{n-1},\sigma_{n}\}$. Each element of $W_{n}$ induces a permutation of $\{1,2,\ldots,n\}$. So we have a surjective homomorphism $W_{n}\rightarrow S_{n}$ with kernel isomorphic to $({\mathbb{Z}}/2{\mathbb{Z}})^{n}$. Therefore, $\varphi_{\lambda}$ can be regarded as an irreducible character of $W_{n}$.

An ordered pair $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$ of two partitions is called a bi-partitions. Let ${\mathcal{P}}_{2}(n)$ denote the set of bipartitions of $n$, i.e.,

For a bi-partition $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$, we define its transpose by $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}^{\rm t}=\genfrac{[}{]}{0.0pt}{}{\nu}{\mu}$. A bi-partition $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$ is called degenerate if $\mu=\nu$, and it is called non-degenerate otherwise. For $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}\in{\mathcal{P}}_{2}(n)$ such that $\mu\in{\mathcal{P}}(k)$ and $\nu\in{\mathcal{P}}(l)$ where $k+l=n$, we define

where $\varepsilon_{l}\colon W_{l}\rightarrow\{\pm 1\}$ is given by $s_{i}\mapsto 1$ and $\sigma_{l}\mapsto-1$. It is known that $\varphi_{\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}}$ is an irreducible character of $W_{n}$, and the mapping ${\mathcal{P}}_{2}(n)\rightarrow{\mathcal{E}}(W_{n})$ by $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}\mapsto\varphi_{\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}}$ gives a parametrization of ${\mathcal{E}}(W_{n})$ such that

• •

  $\varphi_{\genfrac{[}{]}{0.0pt}{}{n}{-}}={\bf 1}_{W_{n}}$

• •

  $\varphi_{\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}}\cdot\varepsilon_{n}=\varphi_{\genfrac{[}{]}{0.0pt}{}{\nu}{\mu}}$, in particular $\varphi_{\genfrac{[}{]}{0.0pt}{}{-}{n}}=\varepsilon_{n}$

(cf. [GP00] theorem 5.5.6).

The kernel $W_{n}^{+}$ of $\varepsilon_{n}$ is a subgroup of index two generated by $\{s_{1},\ldots,s_{n-1},\sigma_{n}s_{n-1}\sigma_{n}\}$. Let $W_{n}^{-}=W_{n}\smallsetminus W_{n}^{+}$. It is well known that if $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$ is non-degenerate, then $\varphi_{\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}}|_{W_{n}^{+}}=\varphi_{\genfrac{[}{]}{0.0pt}{}{\nu}{\mu}}|_{W_{n}^{+}}$ which is an irreducible character of $W_{n}^{+}$; if $\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$ is degenerate, then $\varphi_{\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}}|_{W_{n}^{+}}$ is a sum of two irreducible characters of $W_{n}^{+}$.

In this subsection, we recall some basic notations of “symbols” from [Lus77]. A symbol $\Lambda$ is an ordered pair

of two finite sets $A,B$ (possibly empty) of non-negative integers. The sets $A,B$ are also denoted by $\Lambda^{*},\Lambda_{*}$ and called the first row, the second row of $\Lambda$ respectively. A symbol $\Lambda$ is called reduced if $0\not\in A\cap B$. If $\Lambda=\binom{A}{B}$, then we define its transpose by $\Lambda^{\rm t}=\binom{B}{A}$. We denote $\Lambda_{1}\subset\Lambda_{2}$ and call $\Lambda_{1}$ a subsymbol of $\Lambda_{2}$ if both $\Lambda_{1}^{*}\subset\Lambda_{2}^{*}$ and $(\Lambda_{1})_{*}\subset(\Lambda_{2})_{*}$. If $\Lambda_{1}\subset\Lambda_{2}$, their difference is defined by $\Lambda_{2}\smallsetminus\Lambda_{1}=\binom{\Lambda_{2}^{*}\smallsetminus\Lambda_{1}^{*}}{(\Lambda_{2})_{*}\smallsetminus(\Lambda_{1})_{*}}$. If both $\Lambda_{1}^{*}\cap\Lambda_{2}^{*}=\emptyset$ and $(\Lambda_{1})_{*}\cap(\Lambda_{2})_{*}=\emptyset$, we define $\Lambda_{1}\cup\Lambda_{2}=\binom{\Lambda_{1}^{*}\cup\Lambda_{2}^{*}}{(\Lambda_{1})_{*}\cup(\Lambda_{2})_{*}}$.

For a symbol $\Lambda$ given in (2.2), its rank and defect are defined by

From the definition, it is not difficult to check that

A symbol $\Lambda$ is called degenerate if $\Lambda^{\rm t}=\Lambda$. If $\Lambda$ is degenerate, then ${\rm rk}(\Lambda)$ is even and ${\rm def}(\Lambda)=0$.

We define an equivalence relation “$\sim$” on symbols generated by

If $\Lambda_{1}\sim\Lambda_{2}$, two symbols $\Lambda_{1},\Lambda_{2}$ are called similar. It is not difficult to see that two symbols in the same similar class have the same rank and the same defect, and each similar class contains a unique reduced symbol. Let ${\mathcal{S}}$ denote the set of similar classes of symbols, and let ${\mathcal{S}}_{n,\delta}\subset{\mathcal{S}}$ denote the subset of similar classes of symbols of rank $n$ and defect $\delta$.

A symbol $\Lambda$ is called cuspidal if (2.4) is an equality. It is not difficult to see that a symbol is cuspidal if and only if it is similar to a symbol of the forms $\binom{k,k-1,\ldots,0}{-}$ or $\binom{-}{k,k-1,\ldots,0}$ for some non-negative integer $k$. Note that $\binom{A}{-}$ means that the second row of the symbol is the empty set.

A mapping $\Upsilon$ from symbols to bi-partitions is given by

If $\Upsilon(\Lambda)=\genfrac{[}{]}{0.0pt}{}{\mu}{\nu}$, we write $\Upsilon(\Lambda)^{*}=\mu$ and $\Upsilon(\Lambda)_{*}=\nu$ to denote the first row and the second row of the bi-partition $\Upsilon(\Lambda)$. We can check that $\Upsilon(\Lambda_{1})=\Upsilon(\Lambda_{2})$ if and only if $\Lambda_{1}\sim\Lambda_{2}$, and then $\Upsilon$ gives a bijection

Modified from Lusztig, we define

Note that $\Lambda\in{\mathcal{S}}_{{\rm O}^{\epsilon}_{2n}}$ if and only if $\Lambda^{\rm t}\in{\mathcal{S}}_{{\rm O}^{\epsilon}_{2n}}$ where $\epsilon=+$ or $-$. Then we define

i.e., in ${\mathcal{S}}_{{\rm SO}^{\epsilon}_{2n}}$ a non-degenerate symbol is identified with its transpose, and a degenerated symbol $\Lambda$ occurs with multiplicity $2$ and the two copies are denoted by $\Lambda^{\rm I},\Lambda^{\rm II}$ respectively. Note that ${\mathcal{S}}_{{\rm O}^{-}_{2n}}$ does not contain any degenerate symbols and so ${\mathcal{S}}_{{\rm SO}^{-}_{2n}}={\mathcal{S}}_{{\rm O}^{-}_{2n}}/\{\Lambda,\Lambda^{\rm t}\}$.

Let $\text{\bf G}={\rm Sp}_{2n}$, ${\rm SO}_{2n+1}$, ${\rm O}^{\epsilon}_{2n}$ or ${\rm SO}^{\epsilon}_{2n}$ where $\epsilon=+$ or $-$. A symbol $Z=\binom{a_{1},a_{2},\ldots,a_{m+1}}{b_{1},b_{2},\ldots,b_{m}}$ of defect $1$ is called special if $a_{1}\geq b_{1}\geq a_{2}\geq b_{2}\geq\cdots\geq a_{m}\geq b_{m}\geq a_{m+1}$; similarly, a symbol $Z=\binom{a_{1},a_{2},\ldots,a_{m}}{b_{1},b_{2},\ldots,b_{m}}$ of defect $0$ is called special if $a_{1}\geq b_{1}\geq a_{2}\geq b_{2}\geq\cdots\geq a_{m}\geq b_{m}$. Define

For a special symbol $Z$ of defect $\delta_{0}$, we define

i.e., ${\mathcal{S}}_{Z}$ is the subset of ${\mathcal{S}}$ consisting of the symbols of the exactly the same entries of $Z$. Note that in the above definition of ${\mathcal{S}}_{Z}^{\text{\bf G}}$, the special symbol $Z$ is not required to be in ${\mathcal{S}}_{\text{\bf G}}$. It is clear that

where $\bigcup_{Z}$ runs over

• •

  special symbols of rank $n$ and defect $1$ if $\text{\bf G}={\rm Sp}_{2n}$ or ${\rm SO}_{2n+1}$;

• •

  special symbols of rank $n$ and defect $0$ if $\text{\bf G}={\rm SO}^{+}_{2n}$ or ${\rm O}^{+}_{2n}$;

• •

  non-degenerate special symbols of rank $n$ and defect $0$ if $\text{\bf G}={\rm SO}^{-}_{2n}$ or ${\rm O}^{-}_{2n}$.

For a special symbol $Z$, let $Z_{\rm I}$ denote the subsymbol consisting of “singles”, i.e.,

and we define the degree of $Z$ by

where $|Z_{\rm I}|$ denotes the number of entries in $Z_{\rm I}$, i.e., $|Z_{\rm I}|=|(Z_{\rm I})^{*}|+|(Z_{\rm I})_{*}|$. Note that $\deg(Z)$ is always a non-negative integer. For a subsymbol $M\subset Z_{\rm I}$, we define a symbol $\Lambda_{M}\in{\mathcal{S}}_{Z}$ by

It is not difficult to see that

Note that

where $Z^{\rm I},Z^{\rm II}$ are both equal to $Z$ but are regarded as two elements. Then we have

Note that the family ${\mathcal{S}}_{Z}^{\text{\bf G}}$ is empty if $\text{\bf G}={\rm O}^{-}_{2n},{\rm SO}^{-}_{2n}$ and $\deg(Z)=0$.

Finally we define a pairing $\langle,\rangle\colon{\mathcal{S}}_{Z}^{\text{\bf G}}\times{\mathcal{S}}_{Z,\delta_{0}}\rightarrow{\mathbf{F}}_{2}$ by

In this subsection, let G be a general linear group ${\rm GL}_{n}$ or a unitary group ${\rm U}_{n}$. It is well known that the Weyl group $W_{\text{\bf G}}=S_{n}$, and ${\mathcal{E}}(S_{n})$ is parametrized by ${\mathcal{P}}(n)$. For $\lambda\in{\mathcal{P}}(n)$, we define

where $\varphi_{\lambda}$ denotes the irreducible character of $S_{n}$ corresponding to $\lambda$. It is known that $R_{\lambda}^{\text{\bf G}}$ is an irreducible unipotent character of $G$, i.e., $R_{\lambda}^{\text{\bf G}}\in{\mathcal{E}}(\text{\bf G},1)$. Let ${\mathcal{S}}_{{\rm GL}_{n}}={\mathcal{S}}_{{\rm U}_{n}}={\mathcal{P}}(n)$. Then the mapping

is a bijection. To distinguish ${\mathcal{S}}_{{\rm U}_{n}}$ from ${\mathcal{S}}_{{\rm GL}_{n}}$, an element in ${\mathcal{S}}_{{\rm U}_{n}}$ will be denoted by $[\bar{\lambda}_{1},\ldots,\bar{\lambda}_{m}]$ for $[\lambda_{1},\ldots,\lambda_{m}]\in{\mathcal{P}}(n)$. An element of the form $[\bar{k},\bar{k-1},\ldots,\bar{1}]\in{\mathcal{S}}_{{\rm U}_{k(k+1)/2}}$ for some $k$ is called cuspidal. For a unitary group, it is well known that $\lambda$ is cuspidal if and only if $\rho_{\lambda}$ is cuspidal.

It is known that the parametrization ${\mathcal{L}}_{1}$ above is compatible with the parabolic induction on unipotent characters. More precisely, let $\text{\bf G}_{n}={\rm GL}_{n}$, ${\rm U}_{2n}$ or ${\rm U}_{2n+1}$. For $\rho\in{\mathcal{E}}(\text{\bf G}_{n},1)$, define

where $R^{\text{\bf G}_{n+1}}_{\text{\bf G}_{n}\times{\rm GL}^{\dagger}_{1}}$ is the standard parabolic induction, ${\rm GL}^{\dagger}_{1}={\rm GL}_{1}$ defined over ${\mathbf{F}}_{q}$ if $\text{\bf G}_{n}={\rm GL}_{n}$; and ${\rm GL}^{\dagger}_{1}$ is the restriction to ${\mathbf{F}}_{q}$ of ${\rm GL}_{1}$ defined over a quadratic extension of ${\mathbf{F}}_{q}$ if $\text{\bf G}_{n}={\rm U}_{2n}$ or ${\rm U}_{2n+1}$. For $\lambda\in{\mathcal{S}}_{\text{\bf G}_{n}}$, we define $\Omega(\lambda)$ to be a subset of ${\mathcal{S}}_{\text{\bf G}_{n+1}}$ consisting of partitions of the following types: