This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Roots of identity in finite groups of Lie type

Saikat Panja [email protected] Harish-Chandra Research Institute- Main Building, Chhatnag Rd, Jhusi, Uttar Pradesh 211019, India
Abstract.

Given an integer M2M\geq 2, we deploy the generating function techniques to compute the number of MM-th roots of identity in some of the well-known finite groups of Lie type, more precisely for finite general linear groups, symplectic groups, orthogonal groups of all types and unitary groups over finite fields of odd characteristics.

Key words and phrases:
word maps, finite groups of Lie type, MM-th root
2020 Mathematics Subject Classification:
20G40, 20D06
The author is supported by a PDF-Math fellowship from the Harish-Chandra Research Institute

1. Introduction

A significant amount of modern mathematics is centred around finding solutions to equations in objects belonging to different categories, for example, the category of groups, the category of associative algebras, the category of Lie algebras etc.. In the world of groups, these are known as word problems. It has come to the consideration of many great mathematicians of the current century, especially after the settlement of the long-standing Ore’s conjecture (which states that every element of a finite non-abelian simple group is a commutator). Let us recall the word problem in the context of group theory in brief. Fix a free group n\mathscr{F}_{n} on nn generators. Then for any word ωn\omega\in\mathscr{F}_{n} and a group GG, we get a map ω~:GnG\widetilde{\omega}:G^{n}\longrightarrow G, by means of evaluations. Several astonishing results have been established regarding these maps. A typical word problem asks if the map induced by the given word is surjective on the group or not. The other related problem i,e, Waring-type problem asks for the existence of N(ω)N(\omega)\in\mathbb{N} such that ω~(G)=ω~(G)N(ω)\langle\widetilde{\omega}(G)\rangle=\widetilde{\omega}(G)^{N(\omega)}. The most surprising result in this direction states that, for a given word ω\omega, there exists NωN_{\omega}\in\mathbb{N} such that for all finite non-abelian simple groups of size greater or equal to NwN_{w}, we have N(ω)=3N(\omega)=3, see the great work [31] by Shalev. Later he together with Larsen and Tiep improved the result by proving N(ω)=2N(\omega)=2, in [18] and [20]. Unfortunately the number N(ω)N(\omega) can not be reduced further. An easy example arises out of the power map, which is induced by the word omega=xM1\ omega=x^{M}\in\mathscr{F}_{1}, for some integer M2M\geq 2. The Waring-type problems have also been studied in case of Lie groups and Chevalley groups in [13], unipotent algebraic groups in [17], residually finite groups in [19], pp-adic and Adelic groups in [1] et cetera. For a survey of these results and further problems in the context of group theory, we refer the reader to the excellent survey due to Shalev [31] and the references therein.

There have been studies to find the number of elements which are images of the map xMx^{M} in classical groups. This was done for the finite general linear groups in [16], for the finite symplectic and orthogonal groups in [29] and for finite unitary groups in [30]. The exceptional cases remain open. After knowing that an equation has a solution in the concerned object, it is desirable to ask how many solutions exist in this case. Some authors also call them as the problem of finding fibres of a word map. The enumeration of the solutions of xp=1x^{p}=1 has been carried out for symmetric and alternating groups in many works like [4], [28], [15], [14] et cetera. The problem of finding fibres of xM=1x^{M}=1 in some of the finite groups of Lie type has been studied by several authors, for example, proportions of elements in finite groups of Lie type have been estimated in [27], algorithms to identify abundant pp-singular elements in finite classical groups have been studied in [26], Proportions of pp-regular elements in finite classical groups have been studied in [2], abundant pp-singular elements in finite classical groups have been studied in [25] et ceetra. In this paper we derive the generating functions for the probability of an element satisfying xM=1x^{M}=1, where MM is an integer greater than 22. We call these elements to be roots of identity. Note that if an element is a root of identity, then so are all its conjugates. Thus we first find the conjugacy class representatives which are MM-th root of identity. In this regard, we brush through a brief description of the conjugacy classes in Section 2. Hereafter in Section 3, we first compute the roots of identity which are either semisimple or unipotent. Finally using the Jordan decomposition of an element, in Section 4, we derive the generating functions for the number of MM-th roots of identity in finite classical groups of Lie type; more precisely in the general linear group, the symplectic group, the orthogonal group and the unitary group. Finally in Section 5, we derive an exact formula for the probability of a matrix to be MM-th root of identity in the case of the general linear group and the symplectic group, where MM is an odd non-defining prime and the defining prime satisfies q1(modM)q\equiv-1\pmod{M}.

Running assumption

Throughout the article it will be assumed that the finite fields under consideration are of odd characteristics.

Notation

The set of units in /n\mathbb{Z}/n\mathbb{Z} will be denoted by /n×\mathbb{Z}/n\mathbb{Z}^{\times}. By the notation e(n)e(n), we will always denote the order of qq in the multiplicative group /n×\mathbb{Z}/n\mathbb{Z}^{\times}. The Euler’s totient function will be denoted by ϕ(n)\phi(n). We denote the set of irreducible monic polynomials over 𝔽q\mathbb{F}_{q} by Φ\Phi. Let us also fix the notation for the set of all partitions of nn to be Λn\Lambda^{n}. By the symbol (uq)i\left(\dfrac{u}{q}\right)_{i}, we denote the quantity (1uq)(1uq2)(1uqi)\left(1-\dfrac{u}{q}\right)\left(1-\dfrac{u}{q^{2}}\right)\ldots\left(1-\dfrac{u}{q^{i}}\right). For a partition λn\lambda\vdash n, the number of occurrences of ii will be denoted as mi(λ)m_{i}(\lambda). The transpose of a partition λ\lambda will be denoted as λ\lambda^{\prime}. The number of partitions of a positive integer nn will be denoted as 𝐏(n)\mathbf{P}(n) and those with lesser than or equal to equal to mm many (not necessarily distinct) parts will be denoted as 𝐏m(n)\mathbf{P}_{m}(n).

2. The groups under consideration

In this section, we briefly recall the groups. Note that if an element is an MM-th root of identity, then so are its conjugates. Hence we embark to find first the conjugacy classes which are MM-th root of identity. This makes it necessary for us to know about the conjugacy classes and the centralizer of an element. We will follow the treatment of [11], [23], [36] and, [5]. For the case of the general linear group, it is easy to find the conjugacy representative for the conjugacy classes, but for other concerned groups, they are a little tricky. One may look into [34], [33] and, [35], to handle the other cases. Without further delay let us start with the subsection where we talk about the general linear group.

2.1. General linear group

Given a nn-dimensional vector space VV over 𝔽q\mathbb{F}_{q}, the set of all automorphisms of VV is denoted by GL(V)\mathrm{GL}(V). After fixing an appropriate basis, GL(V)\mathrm{GL}(V) can be identified with GLn(q)\mathrm{GL}_{n}(q). The conjugacy classes of GLn(q)\mathrm{GL}_{n}(q) is determined by functions λ:ΦΛ\lambda:\Phi\longrightarrow\Lambda by fλff\mapsto\lambda_{f} such that fdeg(f)|λf|=n\sum\limits_{f}\deg(f)\cdot|\lambda_{f}|=n. The combinatorial data attached via this correspondence, for an element αGLn(q)\alpha\in\mathrm{GL}_{n}(q) will be denoted by Δ(α)\Delta(\alpha). The centralizer size is given by

φΔ(α)(qdegφi(λi,φ)2i1(1qdegφ)mi(λφ)).\displaystyle\prod\limits_{\varphi\in\Delta(\alpha)}\left(q^{\deg\varphi\cdot\sum\limits_{i}(\lambda_{i,\varphi}^{\prime})^{2}}\prod_{i\geq 1}\left(\dfrac{1}{q^{\deg\varphi}}\right)_{m_{i}(\lambda_{\varphi})}\right).

This quantity sometimes can be shortened using a general expression, which we recall from [6, Theorem 2] below;

cGL,φ,q(λ)=(qdegφi(λi)2i1(1qdegφ)mi(λφ)).\displaystyle c_{\mathrm{GL},\varphi,q}(\lambda)=\left(q^{\deg\varphi\cdot\sum\limits_{i}(\lambda_{i}^{\prime})^{2}}\prod_{i\geq 1}\left(\dfrac{1}{q^{\deg\varphi}}\right)_{m_{i}(\lambda_{\varphi})}\right).

2.2. Symplectic and orthogonal groups

2.2.1. Symplectic group

Let VV be a vector space of dimension 2n2n over 𝔽q\mathbb{F}_{q}. There is a unique non-degenerate alternating bilinear form on VV. We consider the form given by

(xi)i=12n,(yj)j=12n=j=1nxjy2n+1ji=0n1x2niyi+1.\left<(x_{i})_{i=1}^{2n},(y_{j})_{j=1}^{2n}\right>=\sum_{j=1}^{n}x_{j}y_{2n+1-j}-\sum_{i=0}^{n-1}x_{2n-i}y_{i+1}.

The symplectic group is the subgroup of GL(V)\mathrm{GL}(V) consisting of those elements which preserve this alternating form on VV. By fixing an appropriate basis, the matrix of the form is J=(0ΠnΠn0)J=\begin{pmatrix}0&\Pi_{n}\\ -\Pi_{n}&0\end{pmatrix} where Πn=(000100101000)\Pi_{n}=\begin{pmatrix}0&0&\cdots&0&1\\ 0&0&\cdots&1&0\\ \vdots&\vdots&\reflectbox{$\ddots$}&\vdots&\vdots\\ 1&0&\cdots&0&0\end{pmatrix} and

Sp2n(q)={AGL2n(q)AtJA=J}.\mathrm{Sp}_{2n}(q)=\{A\in\mathrm{GL}_{2n}(q)\mid{}^{t}\!AJA=J\}.

Since all alternating forms are equivalent over 𝔽q\mathbb{F}_{q}, the symplectic groups obtained with respect to different forms are conjugate within GL2n(q)\mathrm{GL}_{2n}(q).

2.2.2. Orthgonal Groups

Let VV be an mm-dimensional vector space over a finite field 𝔽q\mathbb{F}_{q}. Then there are at most two non-equivalent non-degenerate quadratic forms on VV. The orthogonal group consists of elements of GL(V)\mathrm{GL}(V) which preserve a non-degenerate quadratic form QQ.

When m=2nm=2n for some n1n\geq 1, up to equivalence there are two such forms denoted as Q+Q^{+} and QQ^{-}. These are as follows. Fix a𝔽qa\in\mathbb{F}_{q} such that t2+t+a𝔽q[t]t^{2}+t+a\in\mathbb{F}_{q}[t] is irreducible. Then the two non-equivalent forms are given by

  1. (1)

    Q+(x1,,xm)=x1x2+x3x4++x2n1x2nQ^{+}(x_{1},\ldots,x_{m})=x_{1}x_{2}+x_{3}x_{4}+\cdots+x_{2n-1}x_{2n}, and

  2. (2)

    Q(x1,,xm)=x12+x1x2+ax22+x3x4++x2n1x2nQ^{-}(x_{1},\ldots,x_{m})=x_{1}^{2}+x_{1}x_{2}+ax_{2}^{2}+x_{3}x_{4}+\cdots+x_{2n-1}x_{2n}.

The orthogonal group preserving Q+Q^{+} will be denoted as Om+(q)\mathrm{O}^{+}_{m}(q), and the orthogonal group preserving QQ^{-} will be denoted as Om(q)\mathrm{O}^{-}_{m}(q).

When m=2n+1m=2n+1, for qq even there is only one (up to equivalence) quadratic form, namely Q(x1,,xm)=x12+i=1nx2ix2i+1Q(x_{1},\ldots,x_{m})=x_{1}^{2}+\sum\limits_{i=1}^{n}x_{2i}x_{2i+1} and hence there is only one (up to conjugacy) orthogonal group. If qq is odd, then up to equivalence, there are two non-degenerate quadratic forms. But, these two forms give isomorphic orthogonal groups. We take Q(x1,,xm)=x12++xm2Q(x_{1},\ldots,x_{m})=x_{1}^{2}+\cdots+x_{m}^{2}. Thus, in case m=2n+1m=2n+1, up to conjugacy, we have only one orthogonal group. This will be denoted as O(m,q)\mathrm{O}(m,q).

As it is common in literature, we will use the notation Omϵ(q)\mathrm{O}^{\epsilon}_{m}(q) to denote any of the orthogonal group above where ϵ{,+,}\epsilon\in\{~{},+,-\}. With respect to an appropriate basis, we will fix the matrices of the symmetric bilinear forms (associated to the quadratic forms QϵQ^{\epsilon}) as follows:

J0=(00Πn0α0Πn00),J+=(0ΠnΠn0),J=(000Πn1010000δ0Πn1000)J_{0}=\begin{pmatrix}0&0&\Pi_{n}\\ 0&\alpha&0\\ \Pi_{n}&0&0\end{pmatrix},J_{+}=\begin{pmatrix}0&\Pi_{n}\\ \Pi_{n}&0\end{pmatrix},J_{-}=\begin{pmatrix}0&0&0&\Pi_{n-1}\\ 0&1&0&0\\ 0&0&-\delta&0\\ \Pi_{n-1}&0&0&0\end{pmatrix}

where α𝔽q×,δ𝔽q𝔽q2\alpha\in\mathbb{F}_{q}^{\times},\delta\in\mathbb{F}_{q}\setminus\mathbb{F}_{q}^{2}, and Πl=(000100101000)\Pi_{l}=\begin{pmatrix}0&0&\cdots&0&1\\ 0&0&\cdots&1&0\\ \vdots&\vdots&\reflectbox{$\ddots$}&\vdots&\vdots\\ 1&0&\cdots&0&0\end{pmatrix}, an l×ll\times l matrix. Then, the orthogonal group in matrix form is

Omϵ(q)={AGLm(q)AtJϵA=Jϵ}.\mathrm{O}^{\epsilon}_{m}(q)=\{A\in\mathrm{GL}_{m}(q)\mid{}^{t}\!{A}J_{\epsilon}A=J_{\epsilon}\}.

Adapting the notations of [8], we define the type of an orthogonal space as follows.

The type of an orthogonal space (V,Q)(V,Q) of dimension mm is

τ(V)={1if m is odd, q1(mod  4),Qxi2,1if m is odd, q1(mod  4),Qbxi2,ιmif m is odd, q3(mod  4),Qxi2,(ι)mif m is odd, q3(mod  4),Qbxi2,\displaystyle\tau(V)=\begin{cases}1&\text{if }m\text{ is odd, }q\equiv 1\allowbreak\mkern 10.0mu({\operator@font mod}\,\,4),Q\sim\sum x_{i}^{2},\\ -1&\text{if }m\text{ is odd, }q\equiv 1\allowbreak\mkern 10.0mu({\operator@font mod}\,\,4),Q\sim b\sum x_{i}^{2},\\ \iota^{m}&\text{if }m\text{ is odd, }q\equiv 3\allowbreak\mkern 10.0mu({\operator@font mod}\,\,4),Q\sim\sum x_{i}^{2},\\ (-\iota)^{m}&\text{if }m\text{ is odd, }q\equiv 3\allowbreak\mkern 10.0mu({\operator@font mod}\,\,4),Q\sim b\sum x_{i}^{2},\end{cases}

where ι\iota\in\mathbb{C} satisfies ι2=1\iota^{2}=-1, and b𝔽q𝔽q2b\in\mathbb{F}_{q}\setminus\mathbb{F}_{q}^{2}. More generally, when VV is orthogonal direct sum V1V2VlV_{1}\oplus V_{2}\oplus\cdots\oplus V_{l}, the type is defined by τ(V)=i=1lτ(Vi)\tau(V)=\prod\limits_{i=1}^{l}\tau(V_{i}).

Now we describe the conjugacy classes and the centralizer of an element in finite symplectic and orthogonal groups. This is the classic work of [36].We recall briefly the results therein, which will be used further. A symplectic signed partition is a partition of a number kk, such that the odd parts have even multiplicity and even parts have a sign associated with it. The set of all symplectic signed partitions will be denoted as ΛSp\Lambda_{\mathrm{Sp}}. An orthogonal signed partition is a partition of a number kk, such that all even parts have even multiplicity, and all odd parts have a sign associated with it. The set of all orthogonal signed partitions will be denoted as ΛO\Lambda_{\mathrm{O}}. The dual of a monic degree rr polynomial f(x)k[x]f(x)\in k[x] satisfying f(0)0f(0)\neq 0, is the polynomial given by f(x)=f(0)1xrf(x1)f^{*}(x)=f(0)^{-1}x^{r}f(x^{-1}).

According to [36], [32], the conjugacy classes of Sp2n(q)\mathrm{Sp}_{2n}(q) are parameterized by the functions λ:ΦΛ2nΛSp2n\lambda:\Phi\rightarrow\Lambda^{2n}\cup\Lambda_{\mathrm{Sp}}^{2n}, where Φ\Phi denotes the set of all monic, non-constant, irreducible polynomials, Λ2n\Lambda^{2n} is the set of all partitions of 1k2n1\leq k\leq 2n and ΛSp2n\Lambda_{\mathrm{Sp}}^{2n} is the set of all symplectic signed partitions of 1k2n1\leq k\leq 2n. Such a λ\lambda represent a conjugacy class of Sp\mathrm{Sp} if and only if (a) λ(u)=0\lambda(u)=0, (b) λφ=λφ\lambda_{\varphi^{*}}=\lambda_{\varphi}, (c) λφΛSpn\lambda_{\varphi}\in\Lambda^{n}_{\mathrm{Sp}} iff φ=u±1\varphi=u\pm 1 (we distinguish this λ\lambda, by denoting it λ±\lambda^{\pm}) and, (d) φ|λφ|deg(φ)=2n\displaystyle\sum_{\varphi}|\lambda_{\varphi}|\textup{deg}(\varphi)=2n. Also from [36], [32], we find that a similar kind of statement is true for the groups Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q). The conjugacy classes of Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q) are parameterized by the functions λ:ΦΛnΛOn\lambda:\Phi\rightarrow\Lambda^{n}\cup\Lambda_{\mathrm{O}{}}^{n}, where Φ\Phi denotes the set of all monic, non-constant, irreducible polynomials, Λn\Lambda^{n} is the set of all partitions of 1kn1\leq k\leq n and ΛOn\Lambda_{\mathrm{O}}^{n} is the set of all symplectic signed partitions of 1kn1\leq k\leq n. Such a λ\lambda represent a conjugacy class of Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q) for ϵ=±\epsilon=\pm, if and only if (a) λ(x)=0\lambda(x)=0, (b) λφ=λφ\lambda_{\varphi^{*}}=\lambda_{\varphi}, (c) λφΛOn\lambda_{\varphi}\in\Lambda^{n}_{\mathrm{O}} iff φ=u±1\varphi=u\pm 1 (we distinguish this λ\lambda, by denoting it λ±\lambda^{\pm}) and, (d) φ|λφ|deg(φ)=n\displaystyle\sum_{\varphi}|\lambda_{\varphi}|\textup{deg}(\varphi)=n. Class representative corresponding to given data can be found in [34], [33], [10] and we will mention them whenever needed. We mention the following results about the conjugacy class size (and hence the size of the centraliser) of elements corresponding to given data, which can be found in [36].

Lemma 2.1.

[36, pp. 36] Let XSp2n(q)X\in\mathrm{Sp}_{2n}(q) be a matrix corresponding to the data Δ(X)={(ϕ,μϕ):ϕΦXΦ}\Delta(X)=\{(\phi,\mu_{\phi}):\phi\in\Phi_{X}\subset\Phi\}. Then the conjugacy class of XX in Sp2n(q)\mathrm{Sp}_{2n}(q) is of size |Sp2n(q)|ϕB(ϕ)\dfrac{|\mathrm{Sp}_{2n}(q)|}{\prod\limits_{\phi}B(\phi)} where B(ϕ)B(\phi) and A(ϕμ)A(\phi^{\mu}) are defined as follows

A(ϕμ)={|Umμ(Q)|if ϕ(x)=ϕ(x)x±1|GLmμ(Q)|12if ϕϕ|Spmμ(q)|if ϕ(x)=x±1,μ odd|q12mμOmμϵ(q)|if ϕ(x)=x±1,μ even,\displaystyle A(\phi^{\mu})=\begin{cases}|\mathrm{U}_{m_{\mu}}(Q)|&\text{if }\phi(x)=\phi^{*}(x)\neq x\pm 1\\ |\mathrm{GL}_{m_{\mu}}(Q)|^{\frac{1}{2}}&\text{if }\phi\neq\phi^{*}\\ |\mathrm{Sp}_{m_{\mu}}(q)|&\text{if }\phi(x)=x\pm 1,~{}\mu\text{ odd}\\ |q^{\frac{1}{2}m_{\mu}}\mathrm{O}^{\epsilon}_{m_{\mu}}(q)|&\text{if }\phi(x)=x\pm 1,~{}\mu\text{ even}\end{cases},

where ϵ\epsilon gets determined by the sign of the corresponding partition, Q=q|ϕ|Q=q^{|\phi|}, mμ=m(ϕμ)m_{\mu}=m(\phi^{\mu}) and

B(ϕ)=Qμ<νμmμmν+12μ(μ1)mμ2μA(ϕμ).\displaystyle B(\phi)=Q^{\sum\limits_{\mu<\nu}\mu m_{\mu}m_{\nu}+\frac{1}{2}\sum\limits_{\mu}(\mu-1)m_{\mu}^{2}}\prod\limits_{\mu}A(\phi^{\mu}).
Lemma 2.2.

[36, pp. 39] Let XOnϵ(q)X\in\mathrm{O}^{\epsilon}_{n}(q) be a matrix corresponding to the data Δ(X)={(ϕ,μϕ):ϕΦXΦ}\Delta(X)=\{(\phi,\mu_{\phi}):\phi\in\Phi_{X}\subset\Phi\}. Then the conjugacy class of XX in Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q) is of size |Onϵ(q)|ϕB(ϕ)\dfrac{|\mathrm{O}^{\epsilon}_{n}(q)|}{\prod\limits_{\phi}B(\phi)} where B(ϕ)B(\phi) and A(ϕμ)A(\phi^{\mu}) are defined as before, except when ϕ(x)=x±1\phi(x)=x\pm 1,

A(ϕμ)={|Omμϵ(q)|if μ oddq12mμ|Spmμ(q)|if μ even,\displaystyle A(\phi^{\mu})=\begin{cases}|\mathrm{O}^{\epsilon^{\prime}}_{m_{\mu}}(q)|&\text{if }\mu\text{ odd}\\ q^{-\frac{1}{2}m_{\mu}}|\mathrm{Sp}_{m_{\mu}}(q)|&\text{if }\mu\text{ even}\end{cases},

where ϵ\epsilon^{\prime} in Omμϵ(q)\mathrm{O}^{\epsilon^{\prime}}_{m_{\mu}}(q) gets determined by the corresponding sign of the part, of the partition.

The quantity B(ϕ)B(\phi) will also be denoted as cSp,ϕ,q(λ±)c_{\mathrm{Sp},\phi,q}(\lambda^{\pm}) in case of symplectic groups and as cOϵ,ϕ,q(λ±)c_{\mathrm{O}^{\epsilon},\phi,q}(\lambda^{\pm}) in case of orthogonal groups when ϕ=x±1\phi=x\pm 1.

2.3. Unitary groups

Consider the field k=𝔽q2k=\mathbb{F}_{q^{2}}, and the involution σ:𝔽q2𝔽q2\sigma\colon\mathbb{F}_{q^{2}}\longrightarrow\mathbb{F}_{q^{2}} defined as σ(a):=a¯=aq\sigma(a):=\overline{a}=a^{q} of the field. We further get an automorphism of 𝔽q2[t]\mathbb{F}_{q^{2}}[t], the polynomial ring over 𝔽q2\mathbb{F}_{q^{2}}, by action on the coefficients of the polynomials. The image of f𝔽q2[t]f\in\mathbb{F}_{q^{2}}[t] will be denoted as f¯\overline{f}. Consider the Hermitian form given by the matrix

Πn=(0001001001001000).\Pi_{n}=\begin{pmatrix}0&0&\ldots&0&1\\ 0&0&\ldots&1&0\\ \vdots&\vdots&\ddots&\vdots&\vdots\\ 0&1&\ldots&0&0\\ 1&0&\ldots&0&0\end{pmatrix}.

Then we identify the unitary group (corresponding to the above Hermitian form) with the set

Un(𝔽q2)={AGLn(𝔽q2)AΠnA¯t=Λn}.\mathrm{U}_{n}(\mathbb{F}_{q^{2}})=\left\{A\in\mathrm{GL}_{n}(\mathbb{F}_{q^{2}})\mid A\Pi_{n}\overline{A}^{t}=\Lambda_{n}\right\}.

Since all non-degenerate Hermitian forms on an nn-dimensional vector space are unitarily congruent to the Hermitian form given by the above matrix, it is evident that all the unitary groups arising are conjugate with each other inside GLn(q2)\mathrm{GL}_{n}(q^{2}) (see theorem 10.310.3 of [12]). From the work of Wall, it is known that two elements g1,g2Un(q2)g_{1},g_{2}\in\mathrm{U}_{n}(q^{2}) are conjugate to each other if and only if they are conjugate in GLn(q2)\mathrm{GL}_{n}(q^{2}). These conjugacy classes are parameterized by special polynomials and partitions. We call a polynomial ff of degree dd to be \sim-symmetric if f~=f\widetilde{f}=f where f~(t)=f(0)¯1tdf¯(t1)\widetilde{f}(t)=\overline{f(0)}^{-1}t^{d}\bar{f}(t^{-1}). A polynomial without a proper \sim-symmetric factor will be called a \sim-irrdeucible polynomial. Note that a \sim-irreducible polynomial can be reducible in the usual sense. Denote by Φ~\widetilde{\Phi} to be the set of all \sim-irreducible polynomials. Let 𝒫n\mathcal{P}_{n} denote the set of all partitions of numbers n\leq n. Then a conjugacy class in U(n,𝔽q2)\mathrm{U}(n,\mathbb{F}_{q^{2}}) is determined by a function λ:Φ~𝒫n\lambda\colon\widetilde{\Phi}\longrightarrow\mathcal{P}_{n} which satisfies (a) λ(f)=λ(f~)\lambda(f)=\lambda(\widetilde{f}) and, (b) ϕ|λ(ϕ)|degϕ=n\sum\limits_{\phi}|\lambda(\phi)|\deg\phi=n. The centralizer six=ze can be found in [36] and we are not quoting it here.

Using these information, we determine when a matrix AA is an MM-th root of identity. This information is further used in subsequent sections to determine the desired generating functions, with the help of the concept of central join of two matrices, following [34], [33], [35].

3. Semisimple & unipotent roots of identity

3.1. Semisimple classes

Note that an element in G(q)\mathrm{G}(q) is semisimple if and only if it has order coprime to qq. In this subsection, we classify all the semisimple elements, which give Id\mathrm{Id}, when raised to the power MM. In this subsection, we will assume that (M,q)=1(M,q)=1. Let ff denote a product of distinct irreducible polynomials, and AA be the semisimple matrix with ff as its characteristic polynomial. Then if AM=IdA^{M}=\mathrm{Id}, then ff must have irreducible factors coming from irreducible factors of xM1𝔽q[x]x^{M}-1\in\mathbb{F}_{q}[x]. We recall some results and definitions from [21]. For a positive integer nn, the splitting field of xn1x^{n}-1 over a field KK is called the nn-th cyclotomic field over KK and denoted by K(n)K^{(n)}. The roots of xn1x^{n}-1 in K(n)K^{(n)} are called the nn-th roots of unity over KK and the set of all these roots is denoted by E(n)E^{(n)}.

Lemma 3.1.

[21, Theorem 2.42] Let nn be a positive integer and KK a field of characteristic pp. If (p,n)=1(p,n)=1, then E(n)E^{(n)} is a cyclic group of order nn with respect to multiplication in K(n)K^{(n)}.

A generator of the cyclic group is called primitive nn-th root of unity over KK. If (n,p)=1(n,p)=1 and ζ\zeta is a primitive nn-th root of unity over KK, the polynomial

Qn(x)=s=1(s,n)=1n(xζs),\displaystyle Q_{n}(x)=\prod\limits_{\begin{subarray}{c}s=1\\ (s,n)=1\end{subarray}}^{n}\left(x-\zeta^{s}\right),

is called the nn-th cyclotomic polynomial over KK. When (n,p)=1(n,p)=1, we have that xn1=d|nQd(x)x^{n}-1=\prod_{d|n}Q_{d}(x), see [21, Theorem 2.45]. Note that if AM=1A^{M}=1, for each irreducible factor gg of ff, there exists an m(g)m(g) such that f|Qm(g)(x)f|Q_{m(g)}(x). Thus we need to know how Qn(x)Q_{n}(x) factors over 𝔽q[x]\mathbb{F}_{q}[x]. We recall the following result.

Lemma 3.2.

[21, Theorem 2.47(ii)] When K=𝔽qK=\mathbb{F}_{q} and (n,q)=1(n,q)=1, then the cyclotomic polynomial QnQ_{n} factors into ϕ(n)/e(n)\phi(n)/e(n) distinct monic irreducible polynomials in K[x]K[x] of the same degree e(n)e(n), where e(n)e(n) is order of qq in /n×\mathbb{Z}/n\mathbb{Z}^{\times}.

Since (M,p)=1(M,p)=1, the polynomial xM1x^{M}-1 is separable and hence for ddd\neq d^{\prime} dividing MM, the cyclotomic polynomials Qd(x)Q_{d}(x) and Qd(x)Q_{d^{\prime}}(x) are coprime. Hence we have the following lemma.

Proposition 3.3.

Let (M,q)=1(M,q)=1 and Gn=GLn(q)\mathrm{G}_{n}=\mathrm{GL}_{n}(q). Let ana_{n} denote the number of semisimple conjugacy classes xn,iGx_{n,i}^{G}, such that xn,iM=Idnx_{n,i}^{M}=\mathrm{Id}_{n} for some indices ii. Then we have that

(3.1) 1+n=1anzn=d|M(11ze(d))ϕ(d)e(d),1+\sum\limits_{n=1}^{\infty}a_{n}z^{n}=\prod\limits_{\begin{subarray}{c}d|M\end{subarray}}\left(\dfrac{1}{1-z^{e(d)}}\right)^{\frac{\phi(d)}{e(d)}},

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

Let AA be a semisimple matrix satisfying AM=1A^{M}=1. Then the characteristic polynomial of AA, divides xM1x^{M}-1. Since (M,q)=1(M,q)=1, using factorization of xM1x^{M}-1 and Lemma 3.2, we get that the irreducible factors of xM1x^{M}-1 are of degree ϕ(d)/e(d)\phi(d)/e(d) for d|Md|M, where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}. Furthermore, (M,q)=1(M,q)=1 implies that the polynomial xM1x^{M}-1 is separable. Hence for ddd\neq d^{\prime} dividing MM, we have that (Qd(x),Qd(x))=1(Q_{d}(x),Q_{d^{\prime}}(x))=1. Hence the result follows. ∎

To deduce the result on the probability of a semisimple element to be an MM-th root of identity, we need to know the order of the centralizer of each element. This has been discussed in Section 2.

Corollary 3.4.

Let (M,q)=1(M,q)=1 and Gn=GLn(q)\mathrm{G}_{n}=\mathrm{GL}_{n}(q). Let bnb_{n} denote the proportion of semisimple MM-th roots of identity in Gn\mathrm{G}_{n}. Then we have that

1+n=1bnzn=d|M(1+m=1zme(d)|GLm(qe(d))|)ϕ(d)e(d),1+\sum\limits_{n=1}^{\infty}b_{n}z^{n}=\prod\limits_{\begin{subarray}{c}d|M\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{me(d)}}{|\mathrm{GL}_{m}(q^{e(d)})|}\right)^{\frac{\phi(d)}{e(d)}},

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

To calculate the number of semisimple conjugacy classes in Sp2n(q)\mathrm{Sp}_{2n}(q), which are MM-th root of Id\mathrm{Id}, we need to know the factorization of xM1x^{M}-1. Before proceeding further, for an integer mm, let us define

Dm={x>0:x|qm+1,xqt+1, for all 0t<m}.\displaystyle D_{m}=\{x\in\mathbb{Z}_{>0}:x|q^{m}+1,x\nmid q^{t}+1,\text{ for all }0\leq t<m\}.

The set DmD_{m} plays a crucial role in counting the semisimple elements in finite symplectic and orthogonal groups, which gives Id\mathrm{Id} on raising the power to MM. We collect the following results from [37].

Lemma 3.5.

[37, Theorem 8, and Theorem 11] We have the following results.

  1. (1)

    An irreducible polynomial of degree 2m2m over 𝔽q\mathbb{F}_{q} is SRIM iff ord(f)Dm\mathrm{ord}(f)\in D_{m}

  2. (2)

    For dDmd\in D_{m} the dd-th cyclotomic polynomial QdQ_{d} factors into product of all distinct SRIM polynomials of degree 2m2m and order dd.

Corollary 3.6.

The polynomial Qd(x)Q_{d}(x) has a (and hence all) *-symmetric irreducible factor if and only if dDmd\in D_{m} for some mm. Furthermore, if e(d)e(d) is odd, QdQ_{d} has no SRIM factor.

Proof.

By definition, QnQ_{n} is a *-symmetric polynomial. Hence its factors are either irreducible *-symmetric or product of the form gggg^{*}, where ggg\neq g^{*} and gg are irreducible. If e(n)e(n) is odd, then there is no irreducible *-symmetric factor of QnQ_{n}, since all *-symmetric irreducible polynomials are of even degree. So we assume that e(n)e(n) is even. So assume e(n)e(n) is even. The backward direction is the content of Lemma 3.5. In the other direction, assume that Qn(x)Q_{n}(x) has a *-symmetric polynomial, say ff. Then ord(f)Dm\mathrm{ord}(f)\in D_{m} for some mm. But, by definition of QnQ_{n}, we have that ord(f)=n\mathrm{ord}(f)=n and hence nDmn\in D_{m}. ∎

Proposition 3.7.

Let (M,q)=1(M,q)=1 and G=Sp2n(q)G=\mathrm{Sp}_{2n}(q). Let ana_{n} denote the number of semisimple conjugacy classes xn,iGx_{n,i}^{G}, such that xn,iM=Idnx_{n,i}^{M}=\mathrm{Id}_{n} for some indices ii. Then we have that

(3.2) 1+n=1anzn=m=1(d|MdDm(11ze(d))ϕ(d)e(d)d|MdDm(11z2e(d))ϕ(d)2e(d))1+\sum\limits_{n=1}^{\infty}a_{n}z^{n}=\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in D_{m}\end{subarray}}\left(\dfrac{1}{1-z^{e(d)}}\right)^{\frac{\phi(d)}{e(d)}}\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in D_{m}\end{subarray}}\left(\dfrac{1}{1-z^{2e(d)}}\right)^{\frac{\phi(d)}{2e(d)}}\right)

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

Let ASp2n(q)A\in\mathrm{Sp}_{2n}(q) be a matrix satisfying AM=IdA^{M}=\mathrm{Id}. Then we have that, the minimal polynomial of AA over 𝔽q\mathbb{F}_{q} should divide xM1x^{M}-1. Since xM1=d|MQd(x)x^{M}-1=\prod\limits_{d|M}Q_{d}(x) and Qd(x)Q_{d}(x) further decomposes into irreducibles we need to do a case-by-case study. Because of Corollary 3.6 we have to use DmD_{m}’s. So first assume that dDmd\in D_{m}, then each of the factors of QdQ_{d} determine a Symplectic matrix of Spe(d)\mathrm{Sp}_{e(d)} which gives Id\mathrm{Id} when raised to the power MM. Since QdQ_{d} has in total ϕ(d)/e(d)\phi(d)/e(d) many irreducible factors, this justifies the first factor in the product.

Next, suppose dDmd\not\in D_{m}. Then for any ggg\neq g^{*}, a factor of QdQ_{d}, we need to pair up gg and gg^{*} to get a conjugacy class of Sp2e(d)\mathrm{Sp}_{2e(d)}, which contributes to the counting. Since QdQ_{d} is *-symmetric, gg and gg^{*} occur together in the factorization of QdQ_{d}. Furthermore, QdQ_{d} is separable and hence all the irreducible factors are coprime to each other. Since gg and gg^{*} are combined together, the exponent of zz is taken to be 2e(d)2e(d). But in the process, we are grouping gg and gg^{*} together, resulting in the power of the factor being ϕ(d)/2e(d)\phi(d)/2e(d). This finishes the proof. ∎

Corollary 3.8.

Let (M,q)=1(M,q)=1 and Gn=Sp2n(q)\mathrm{G}_{n}=\mathrm{Sp}_{2n}(q). Let bnb_{n} denote the proportion of semisimple MM-th roots of identity in Gn\mathrm{G}_{n}. Then we have that

1+n=1bnzn=(1+m=1um|Sp2m(q)|)o(M)m=1(d|MdDm(1+m=1zme(d)|Um(q2e(d))|)ϕ(d)e(d)×d|MdDmd1,2(1+m=1zme(d)|GLm(qe(d))|)ϕ(d)2e(d))\displaystyle\begin{split}1+\sum\limits_{n=1}^{\infty}b_{n}z^{n}&=\left(1+\sum\limits_{m=1}^{\infty}\dfrac{u^{m}}{|\mathrm{Sp}_{2m}(q)|}\right)^{o(M)}\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in D_{m}\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{me(d)}}{|\mathrm{U}_{m}(q^{2e(d)})|}\right)^{\frac{\phi(d)}{e(d)}}\right.\\ &\times\left.\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in D_{m}\\ d\neq 1,2\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{me(d)}}{|\mathrm{GL}_{m}(q^{e(d)})|}\right)^{\frac{\phi(d)}{2e(d)}}\right)\end{split}

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}, and o(M)=2o(M)=2 if 2|M2|M and 11 otherwise.

Proof.

Start with the case when MM is even. In this case, both Id\mathrm{Id} and Id\mathrm{Id} are MM-th roots of identity. Now let us assume that AA is an MM-th root of identity, such that ±1\pm 1 is not an eigenvalue of AA. Then the characteristic polynomial of AA is a product of factors of QdQ_{d}, where d|Md|M and the polynomials are irreducible of degree more than 11. The rest of the proof follows from the description of the centralizers when the irreducible factor is a SRIM polynomial or not. In the latter case, we have that ggg\neq g^{*} occurs together in the factorization of the characteristic polynomial of AA. For MM being even Id-\mathrm{Id} is not an MM-th root of identity, which proves the existence of the function o(M)o(M) in the formula. This finishes the proof. ∎

Corollary 3.9.

Let (M,q)=1(M,q)=1 and G2n=O2nϵ(q)\mathrm{G}_{2n}=\mathrm{O}^{\epsilon}_{2n}(q) and G2n+1=O2n+1(q)\mathrm{G}_{2n+1}=\mathrm{O}_{2n+1}(q), where ϵ=±\epsilon=\pm. Let bnϵb_{n}^{\epsilon} and bnb_{n} denote the proportion of semisimple MM-th roots of identity in G2n\mathrm{G}_{2n} and G2n+1\mathrm{G}_{2n+1} respectively. Let us define

B+(z)=1+n=1bn+zn,B(z)=n=1bnzn,B(z)=1+n=1bnzn.\displaystyle B_{+}(z)=1+\sum\limits_{n=1}^{\infty}b_{n}^{+}z^{n},B_{-}(z)=\sum\limits_{n=1}^{\infty}b_{n}^{-}z^{n},B(z)=1+\sum\limits_{n=1}^{\infty}b_{n}z^{n}.

Then we have that

B+(z2)+B(z2)+2zB(z2)\displaystyle B_{+}(z^{2})+B_{-}(z^{2})+2zB(z^{2})
=\displaystyle= (1+m=1(1|O2m+(q)|+1|O2m(q)|)z2m+z(1+m=1z2m|Sp2m(q)|))o(M)\displaystyle\left(1+\sum\limits_{m=1}^{\infty}\left(\dfrac{1}{|\mathrm{O}^{+}_{2m}(q)|}+\dfrac{1}{|\mathrm{O}^{-}_{2m}(q)|}\right)z^{2m}+z\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2m}}{|\mathrm{Sp}_{2m}(q)|}\right)\right)^{o(M)}
×\displaystyle\times m=1(d|MdDm(1+m=1z2me(d)|Um(q2e(d))|)ϕ(d)e(d)d|MdDmd1(1+m=1z2me(d)|GLm(qe(d))|)ϕ(d)2e(d))\displaystyle\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in D_{m}\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2me(d)}}{|\mathrm{U}_{m}(q^{2e(d)})|}\right)^{\frac{\phi(d)}{e(d)}}\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in D_{m}\\ d\neq 1\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2me(d)}}{|\mathrm{GL}_{m}(q^{e(d)})|}\right)^{\frac{\phi(d)}{2e(d)}}\right)

and

B+(z2)B(z2)\displaystyle B_{+}(z^{2})-B_{-}(z^{2})
=\displaystyle= (1+m=1(1|O2m+(q)|1|O2m(q)|)z2m)o(M)\displaystyle\left(1+\sum\limits_{m=1}^{\infty}\left(\dfrac{1}{|\mathrm{O}^{+}_{2m}(q)|}-\dfrac{1}{|\mathrm{O}^{-}_{2m}(q)|}\right)z^{2m}\right)^{o(M)}
×\displaystyle\times m=1(d|MdDm(1+m=1(1)mz2me(d)|Um(q2e(d))|)ϕ(d)e(d)d|MdDmd1(1+m=1z2me(d)|GLm(qe(d))|)ϕ(d)2e(d))\displaystyle\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in D_{m}\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{(-1)^{m}z^{2me(d)}}{|\mathrm{U}_{m}(q^{2e(d)})|}\right)^{\frac{\phi(d)}{e(d)}}\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in D_{m}\\ d\neq 1\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2me(d)}}{|\mathrm{GL}_{m}(q^{e(d)})|}\right)^{\frac{\phi(d)}{2e(d)}}\right)

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}, and o(M)=2o(M)=2 if 2|M2|M and 11 otherwise.

Proof.

The proof is almost similar to Corollary 3.8, but care needs to be taken as we are giving it in a combinatorial formula. This formula occurs because of the existence of different quadratic forms. Let XX be an orthogonal MM-th root of identity, preserving a non-degenerate quadratic form on an NN dimensional vector space VV, where NN is even or odd. Consider the characteristic polynomial of XX, which decomposes into irreducibles with factors being φ=φ\varphi=\varphi^{*} and irreducible or ψψ\psi\psi^{*}, with ψψ\psi\neq\psi^{*} irreducible. We use the centralizer information from Section 2.

Let us start with the first equation. The characteristic polynomial of XX might have root to be 11 or 1-1. But, if MM is odd, it will not have 1-1 as a root. This justifies the presence of o(M)o(M). Note that for NN being even and positive the coefficient of zNz^{N} is sum of the probabilities in case of ON+(q)\mathrm{O}^{+}_{N}(q) and ON(q)\mathrm{O}^{-}_{N}(q). Furthermore for odd NN this is two times the probability in the case of ON(q)\mathrm{O}_{N}(q), which occurs because of two quadratic forms. In case of odd NN, note that |ON(q)|=2|SpN1(q)||\mathrm{O}_{N}(q)|=2|\mathrm{Sp}_{N-1}(q)|, which we are using in the formula. This proves the equality in the first equation.

We prove the next equation by modifying the first equation. We need the value of type of an orthogonal space, as was discussed in Section 2. This consideration doesn’t alter the first factor, as they correspond to the case whether Id\mathrm{Id} and Id-\mathrm{Id} are MM-th roots of identity or not. In the right side of the first equation, for the self conjugate irreducible factors, replace zz by ±iz\pm iz, according to their types. This proves the second equality. ∎

Let us now deduce the SCIM factors of xM1x^{M}-1 in 𝔽q2\mathbb{F}_{q^{2}}. According to [5, Lemma 2], if ff is a SCIM polynomial of odd degree dd, then for any root ζ\zeta of ff, it satisfies ζqd+1=1\zeta^{q^{d}+1}=1. This means that ord(f)|qd+1\mathrm{ord}(f)|q^{d}+1. For a positive integer mm define

D~m={y>0:y|q2m+1+1,yqt+1 for odd 1t<2m+1}.\displaystyle\widetilde{D}_{m}=\{y\in\mathbb{Z}_{>0}:y|q^{2m+1}+1,y\nmid q^{t}+1\text{ for odd }1\leq t<2m+1\}.

It is evident that if ff is a SCIM polynomial of degree dd, then ord(f)D~(d1)/2\mathrm{ord}(f)\in\widetilde{D}_{(d-1)/2}. If dDmd\in D_{m}, and β\beta is a primitive dd-th root of unity, then the set {β,βq2,βq4,,βq2(d1)}\{\beta,\beta^{q^{2}},\beta^{q^{4}},\cdots,\beta^{q^{2(d-1)}}\} is of cardinality dd. Indeed, if βq2i=βq2j\beta^{q^{2i}}=\beta^{q^{2j}} for some 0i<jd10\leq i<j\leq d-1, then d|q2(ji)1d|q^{2(j-i)}-1. Hence 2(ji)2(j-i) is an even multiple of dd, which forces 2(ji)=02(j-i)=0, see [37, Prposition 1]. For eD~me\in\widetilde{D}_{m} and a primitive ee-th root of unity, let us define

fβ(x)=i=0d1(xβq2i).\displaystyle f_{\beta}(x)=\prod\limits_{i=0}^{d-1}\left(x-\beta^{q^{2i}}\right).

Then by [22, Theorem 3.4.8] this is an irreducible polynomial. We claim that fβf_{\beta} is a SCIM polynomial. Note that (βq2j)q=βq2(m+i+1)\left(\beta^{q^{2j}}\right)^{-q}=\beta^{q^{2(m+i+1)}}, since βq2m+1+1=1\beta^{q^{2m+1}+1}=1. This implies that (βq2j)q\left(\beta^{q^{2j}}\right)^{-q} is also a root, and hence fβf_{\beta} is a SCIM polynomial, see [9, pp. 23]. Using [21, Theorem 2.47], we conclude that for an irreducible polynomial ff of degree 2d+12d+1 over 𝔽q2\mathbb{F}_{q^{2}}, the following statements are equivalent;

  1. (1)

    ff is SCIM,

  2. (2)

    ord(f)D~d\mathrm{ord}(f)\in\widetilde{D}_{d},

  3. (3)

    f=f~βf=\widetilde{f}_{\beta} for some β\beta being a tt-th root of unity for some tD~dt\in\widetilde{D}_{d}.

For tD~dt\in\widetilde{D}_{d}, we have that Qd(x)=(xγ)Q_{d}(x)=\prod(x-\gamma), where γ\gamma runs over all tt-th primitive root of unity. It follows that in this case QdQ_{d} factors into a product of all SCIM polynomials. Using a similar argument as in Corollary 3.6 we conclude the following.

Lemma 3.10.

The polynomial Qd(x)Q_{d}(x) has a (and hence all) \sim-symmetric irreducible factor if and only if dD~md\in\widetilde{D}_{m} for some mm. Furthermore, if e(d)e(d) is even, QdQ_{d} has no SCIM factor.

Proposition 3.11.

Let (M,q)=1(M,q)=1 and G=Un(q2)G=\mathrm{U}_{n}(q^{2}). Let ana_{n} denote the number of semisimple conjugacy classes xn,iGx_{n,i}^{G}, such that xn,iM=Idnx_{n,i}^{M}=\mathrm{Id}_{n} for some indices ii. Then we have that

(3.3) 1+n=1anzn=m=1(d|MdD~m(11ze(d))ϕ(d)e(d)d|MdD~m(11z2e(d))ϕ(d)2e(d))1+\sum\limits_{n=1}^{\infty}a_{n}z^{n}=\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in\widetilde{D}_{m}\end{subarray}}\left(\dfrac{1}{1-z^{e(d)}}\right)^{\frac{\phi(d)}{e(d)}}\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in\widetilde{D}_{m}\end{subarray}}\left(\dfrac{1}{1-z^{2e(d)}}\right)^{\frac{\phi(d)}{2e(d)}}\right)

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

Let AUn(q2)A\in\mathrm{U}_{n}(q^{2}) be a matrix satisfying AM=IdA^{M}=\mathrm{Id}. Then we have that, the minimal polynomial of AA over 𝔽q2\mathbb{F}_{q^{2}} should divide xM1x^{M}-1. Since xM1=d|MQd(x)x^{M}-1=\prod\limits_{d|M}Q_{d}(x) and Qd(x)Q_{d}(x) further decomposes into irreducibles we examine this each dd case-by-case depending on whether dd is inside or outside D~m\widetilde{D}_{m} for some mm, and use Lemma 3.10. So first assume that dD~md\in\widetilde{D}_{m}, then each of the factors of QdQ_{d} determine a Unitary matrix of Ue(d)(q2)\mathrm{U}_{e(d)}(q^{2}) which gives Id\mathrm{Id} when raised to the power MM. Since QdQ_{d} has in total ϕ(d)/e(d)\phi(d)/e(d) many irreducible factors, this justifies the first factor in the product.

Next, suppose dDmd\not\in D_{m}. Then for any gg~g\neq\widetilde{g}, a factor of QdQ_{d}, we need to pair up gg and gg^{*} to get a conjugacy class of Ue(d)(q2)\mathrm{U}_{e(d)}(q^{2}), which contributes to the counting. Since QdQ_{d} is \sim-symmetric (because α\alpha and αq\alpha^{-q} are roots of QdQ_{d}, as (d,M)=1(d,M)=1), gg and g~\widetilde{g} occur together in the factorization of QdQ_{d}. Furthermore, QdQ_{d} is separable and hence all the irreducible factors are coprime to each other. Hence the exponent of zz is taken to be 2e(d)2e(d). But in the process, we are grouping gg and gg^{*}, resulting in the power of the factor being ϕ(d)/2e(d)\phi(d)/2e(d). This finishes the proof. ∎

Corollary 3.12.

Let (M,q)=1(M,q)=1 and Gn=Un(q2)\mathrm{G}_{n}=\mathrm{U}_{n}(q^{2}). Let bnb_{n} denote the proportion of semisimple MM-th roots of identity in Gn\mathrm{G}_{n}. Then we have 1+n=1bnzn1+\sum\limits_{n=1}^{\infty}b_{n}z^{n} to be equal to

(3.4) m=1(d|MdD~m(1+m=1zme(d)|Um(q2e(d))|)ϕ(d)e(d)d|MdD~m(1+m=1z2me(d)|GLm(q2e(d))|)ϕ(d)2e(d))\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|M\\ d\in\widetilde{D}_{m}\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{me(d)}}{|\mathrm{U}_{m}(q^{2e(d)})|}\right)^{\frac{\phi(d)}{e(d)}}\prod\limits_{\begin{subarray}{c}d|M\\ d\not\in\widetilde{D}_{m}\end{subarray}}\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2me(d)}}{|\mathrm{GL}_{m}(q^{2e(d)})|}\right)^{\frac{\phi(d)}{2e(d)}}\right)

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

This follows easily from the information on centralizers and size of the conjugacy classes. The flow of reasoning is as same as Corollary 3.4, Corollary 3.8 and Corollary 3.9. Hence it is left to the reader. ∎

3.2. Unipotent classes

Recall that an element xG(q)x\in G(q) is a unipotent element if and only if ord(x)\mathrm{ord}(x) is a power of the defining prime. Hence to find all unipotent elements of order MM, it is enough to find elements of order prp^{r} where M=tprM=t\cdot p^{r} and ptp\nmid t. In the following result, we find the conjugacy classes of unipotent elements of order prp^{r} when G=GLnG=\mathrm{GL}_{n}. In the other finite groups of Lie type two unipotent elements u1u_{1} and u2u_{2} need not be GG-conjugate, even if they are GLn\mathrm{GL}_{n} conjugate. But it so happens that the unipotent conjugacy class in other Finite groups of Lie type descend from those in GLn\mathrm{GL}_{n}. Thus it is necessary that we first resolve the case of GLn(q)\mathrm{GL}_{n}(q). The treatments for the other groups will follow easily.

Proposition 3.13.

Let λ=λ1i1λ2i2λkik\lambda=\lambda_{1}^{i_{1}}\lambda_{2}^{i_{2}}\ldots\lambda_{k}^{i_{k}} be a partition of |λ||\lambda|, where λ1>λ2>>λk\lambda_{1}>\lambda_{2}>\ldots>\lambda_{k}. Consider JλJ_{\lambda} to be the matrix corresponding to the Jordan canonical form of the unipotent matrix attached with λ\lambda. Then JλJ_{\lambda} is of order prp^{r} if and only if pr1<λ1prp^{r-1}<\lambda_{1}\leq p^{r}.

Proof.

Note that Nλ=JλId=(aij)N_{\lambda}=J_{\lambda}-\mathrm{Id}=(a_{ij}) is a nilpotent matrix with the (i,i+1)(i,i+1)-th entry aii+1{0,1}a_{ii+1}\in\{0,1\} for all 1i|λ|11\leq i\leq|\lambda|-1 and aij=0a_{ij}=0 for all (i,j)(i,i+1)(i,j)\neq(i,i+1). Also JλIdJ_{\lambda}-\mathrm{Id} is a nilpotent matrix satisfies (JλId)|λ|=0(J_{\lambda}-\mathrm{Id})^{|\lambda|}=0. Given that JλGL|λ|(𝔽p)J_{\lambda}\in\mathrm{GL}_{|\lambda|}(\mathbb{F}{p}) is a nontrivial unipotent element, it must have order ptp^{t} for some t>0t>0. Using the binomial theorem,

Jλpt=Id+i=1pt(pti)Nλi.\displaystyle J_{\lambda}^{p^{t}}=\mathrm{Id}+\displaystyle\sum\limits_{i=1}^{p^{t}}{p^{t}\choose i}N_{\lambda}^{i}.

Since λ=1i12i2kik\lambda=\ell_{1}^{i_{1}}\ell_{2}^{i_{2}}\ldots\ell_{k}^{i_{k}}, the tt should satisfy pt1p^{t}\geq\ell_{1}. This proves both sides of the statement. ∎

Lemma 3.14.

Let Q(r,[s])Q(r,[s]) denote the number of partitions of rr, with parts not exceeding ss. Then the generating function for QQ in indeterminate zz is given by

1+r=1Q(r,[s])zr=t=1s11zt.\displaystyle 1+\sum\limits_{r=1}^{\infty}Q(r,[s])z^{r}=\prod\limits_{t=1}^{s}\dfrac{1}{1-z^{t}}.

4. Generating functions for all roots of identity

To detect an element xG(q)x\in\mathrm{G}(q), such that xM=Idx^{M}=\mathrm{Id}, we use the Jordan decomposition. First, let us write M=tprM=t\cdot p^{r}, where ptp\nmid t. Using Jordan decomposition (see [24]), we write x=xsxux=x_{s}x_{u}, where xsx_{s} is semisimple and xux_{u} is unipotent. It can be easily seen that xM=Idx^{M}=\mathrm{Id} if and only if xst=Idx_{s}^{t}=\mathrm{Id} and xupr=Idx_{u}^{p^{r}}=\mathrm{Id}. Since we use the combinatorial data to conclude our results, it is important to know the unipotent part of xx, when xx is determined by a combinatorial data Δ(x)\Delta(x). For any element xG(q)x\in\mathrm{G}(q), we consider it inside ambient GLn(q)\mathrm{GL}_{n}(q), and since the order of an element is unaltered under this consideration, we discuss the case for an element in GLn\mathrm{GL}_{n} only. The other cases will follow easily. Recall from Section 2, for an element xGLn(q)x\in\mathrm{GL}_{n}(q), if the combinatorial data is given by {(f,λf(x)):fΦ{z}}\{(f,\lambda_{f}(x)):f\in\Phi\setminus\{z\}\}, then the companion matrix corresponding to the combinatorial data is given by the sum of the blocks of the form

(CfIdCfCf),\displaystyle\begin{pmatrix}C_{f}&\mathrm{Id}&&\\ &C_{f}&&\\ &&\ddots&\\ &&&C_{f}\end{pmatrix},

where CfC_{f} is the companion matrix corresponding to ff, and Id\mathrm{Id} is a degf×degf\deg f\times\deg f matrix. Also, the size of this blocks are λideg(f)×λideg(f)\lambda_{i}\deg(f)\times\lambda_{i}\deg(f), where λ=(λ1,λ2,)\lambda=(\lambda_{1},\lambda_{2},\ldots). It is well known that two matrices x,xGLnx,x^{\prime}\in\mathrm{GL}_{n} are conjugate to each other if and only if they are conjugate in GLn(𝔽q¯)\mathrm{GL}_{n}(\overline{\mathbb{F}_{q}}), see [3]. Hence we consider the matrix xx in GLn(𝔽q¯)\mathrm{GL}_{n}(\overline{\mathbb{F}_{q}}), up to conjugacy. Then the combinatorial data (f,λf(x))(f,\lambda_{f}(x)) becomes {(xμ1,λf(x)),,(xμs,λf(x))}\{(x-\mu_{1},\lambda_{f}(x)),\ldots,(x-\mu_{s},\lambda_{f}(x))\}, where f(x)=(xμ1)(xμs)f(x)=(x-\mu_{1})\ldots(x-\mu_{s}). Hence to summarize we have the following result, using Proposition 3.13.

Lemma 4.1.

Let Δ(x)={(f,λf(x)):fΦ}\Delta(x)=\{(f,\lambda_{f}(x)):f\in\Phi\} be the combinatorial data corresponding to xx and M=tprM=t\cdot p^{r}, ptp\nmid t. If x=xsxux=x_{s}x_{u} is the Jordan decomposition of xx, then xx is an MM-th root of identity in GLn(q)\mathrm{GL}_{n}(q) if and only if xsx_{s} is a tt-th root of identity in GLn(q)\mathrm{GL}_{n}(q) and each part of λ(f)\lambda(f) has size less than or equal to prp^{r}.

In this section, we use cycle indices for classical groups to conclude the final set of results. We will recall the cycle indices from [6] and [7] whenever needed and then deduce our results. We assume the polynomials to be in the indeterminate uu. Let xφ,λx_{\varphi,\lambda} be variables corresponding to pairs of polynomials and partitions. The cycle index for a group G(q)\mathrm{G}(q) is defined to be

1+n=1zn|Gn(q)|(αGn(q)φuxφ,λφ(α)).\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{z^{n}}{|\mathrm{G}_{n}(q)|}\left(\sum\limits_{\alpha\in\mathrm{G}_{n}(q)}\prod\limits_{\begin{subarray}{c}\varphi\neq u\end{subarray}}x_{\varphi,\lambda_{\varphi}(\alpha)}\right).

In the next couple of results, we are first going to recall the factorization of the cycle indices, due to the works of Jason Fulman. After that, we are going to substitute zero or one, on the basis of whether such an element contributes to the quantity or not. We present the first lemma now.

Lemma 4.2.

[7, pp. 55] For G(q)=GLn(q)G(q)=\mathrm{GL}_{n}(q), we have

1+n=1zn|GLn(q)|(αGLn(q)φuxφ,λφ(α))\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{z^{n}}{|\mathrm{GL}_{n}(q)|}\left(\sum\limits_{\alpha\in\mathrm{GL}_{n}(q)}\prod\limits_{\begin{subarray}{c}\varphi\neq u\end{subarray}}x_{\varphi,\lambda_{\varphi}(\alpha)}\right)
=\displaystyle= φu(1+n1λnxφ,λzndegφqdegφ(i(λi)2)i1(1qdegφ)mi(λφ)).\displaystyle\prod\limits_{\varphi\neq u}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}\dfrac{z^{n\cdot\deg\varphi}}{q^{\deg\varphi\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{q^{\deg\varphi}}\right)_{m_{i}(\lambda_{\varphi})}}\right).
Theorem 4.3.

Let ana_{n} denote the number of elements in GLn(q)\mathrm{GL}_{n}(q) which are MM-th root of identity. Let M=tprM=t\cdot p^{r}, where ptp\nmid t. Then the generating function of the probability an/|GLn(q)|a_{n}/|\mathrm{GL}_{n}(q)| is given by

1+n=1an|GLn(q)|zn=d|t(1+m1λmλ1przme(d)qe(d)(i(λi)2)i1(1qe(d))mi(λφ))ϕ(d)e(d),\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{a_{n}}{|\mathrm{GL}_{n}(q)|}z^{n}=\prod\limits_{d|t}\left(1+\sum\limits_{m\geq 1}\sum\limits_{\begin{subarray}{c}\lambda\vdash m\\ \lambda_{1}\leq p^{r}\end{subarray}}\dfrac{z^{me(d)}}{q^{e(d)\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{q^{e(d)}}\right)_{m_{i}(\lambda_{\varphi})}}\right)^{\frac{\phi(d)}{e(d)}},

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

It follows from the proof of Proposition 3.3 and Proposition 3.13 that, for an element xGLn(q)x\in\mathrm{GL}_{n}(q),

  1. (1)

    the semisimple part xsx_{s} has order tt if and only if the irreducible factors of the characteristic polynomial of xsx_{s} divides QdQ_{d} for some d|td|t,

  2. (2)

    the unipotent part xux_{u} has order prp^{r} if and only if the partition corresponding to xux_{u} has all parts lesser than or equal to prp^{r}.

Hence in the formula of Lemma 4.2, we substitute xφ,λφ(α)x_{\varphi,\lambda_{\varphi}(\alpha)} to be 11 when

  1. (1)

    all the polynomials occurring in Δ(α)\Delta(\alpha) are divisors of QdQ_{d}, for some d|td|t,

  2. (2)

    all the λφ(α)\lambda_{\varphi}(\alpha) occurring in Δ(α)\Delta(\alpha) has highest part to be lesser than or equal to prp^{r}.

and put all other xφ,λφ(α)x_{\varphi,\lambda_{\varphi}(\alpha)} to be zero. The occurrence of the degrees follows easily. This proves the equality among both sides. ∎

The theorems for the case of finite symplectic, orthogonal and unitary groups will be stated without detailed proof, since the arguments will be as same as Theorem 4.3, so we won’t be repeating them. However, we will indicate the results from which the proofs follow.

Lemma 4.4.

[6] For G(q)=Sp2n(q)G(q)=\mathrm{Sp}_{2n}(q), we have

1+n=1z2n|Sp2n(q)|(αSp2n(q)φ=u±1xφ,λφ±(α)φu±1xφ,λφ(α))\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{z^{2n}}{|\mathrm{Sp}_{2n}(q)|}\left(\sum\limits_{\alpha\in\mathrm{Sp}_{2n}(q)}\prod\limits_{\varphi=u\pm 1}x_{\varphi,\lambda^{\pm}_{\varphi}(\alpha)}\prod\limits_{\varphi\neq u\pm 1}x_{\varphi,\lambda_{\varphi}(\alpha)}\right)
=\displaystyle= φ=u±1(1+n1λ±nxφ,λ±zncSp,u±1,q(λ±))φ=φφu±1(1+n1λnxφ,λ(zdegφ)ncGL,u1,(qdegφ)1/2(λ))\displaystyle\prod\limits_{\varphi=u\pm 1}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda^{\pm}\vdash n}x_{\varphi,\lambda^{\pm}}\dfrac{z^{n}}{c_{\mathrm{Sp},u\pm 1,q}(\lambda^{\pm})}\right)\prod\limits_{\begin{subarray}{c}\varphi=\varphi^{*}\\ \varphi\neq u\pm 1\end{subarray}}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}\dfrac{(-z^{\deg\varphi})^{n}}{c_{\mathrm{GL},u-1,-(q^{\deg\varphi})^{1/2}}(\lambda)}\right)
×\displaystyle\times {φ,φ}φφ(1+n1λnxφ,λxφ,λz2ndegφcGL,u1,qdegφ(λ))\displaystyle\prod\limits_{\begin{subarray}{c}\{\varphi,\varphi^{*}\}\\ \varphi\neq\varphi^{*}\end{subarray}}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}x_{\varphi^{*},\lambda}\dfrac{z^{2n\deg\varphi}}{c_{\mathrm{GL},u-1,q^{\deg\varphi}}(\lambda)}\right)
Theorem 4.5.

Let ana_{n} denote the number of elements in Sp2n(q)\mathrm{Sp}_{2n}(q) which are MM-th root of identity. Let M=tprM=t\cdot p^{r}, where ptp\nmid t. Then the generating function of the probability an/|Sp2n(q)|a_{n}/|\mathrm{Sp}_{2n}(q)| is given by

1+n=1an|Sp2n(q)|zn\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{a_{n}}{|\mathrm{Sp}_{2n}(q)|}z^{n}
=\displaystyle= (1+n1λ±nλ1±przncSp,u±1,q(λ±))o(t)m=1(d|tdDm[1+n1λ(z)ne(d)cGL,u1,(qe(d))1/2(λ)]ϕ(d)e(d)\displaystyle\left(1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda^{\pm}\vdash n\\ \lambda^{\pm}_{1}\leq p^{r}\end{subarray}}\dfrac{z^{n}}{c_{\mathrm{Sp},u\pm 1,q}(\lambda^{\pm})}\right)^{o(t)}\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|t\\ d\in D_{m}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda\end{subarray}}\dfrac{(-z)^{n}e(d)}{c_{\mathrm{GL},u-1,-(q^{e(d)})^{1/2}}(\lambda)}\right]^{\frac{\phi(d)}{e(d)}}\right.
d|tdDm,d1,2[1+n1λnz2ne(d)cGL,u1,qe(d)(λ)]ϕ(d)2e(d))\displaystyle\left.\prod\limits_{\begin{subarray}{c}d|t\\ d\not\in D_{m},d\neq 1,2\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}\dfrac{z^{2ne(d)}}{c_{\mathrm{GL},u-1,{q^{e(d)}}}(\lambda)}\right]^{\frac{\phi(d)}{2e(d)}}\right)
Proof.

This follows from the proof of Corollary 3.8 and Proposition 3.13. The occurrence of o(t)o(t) can be justified as was done in Corollary 3.8. Finally, we need to use Lemma 4.4. ∎

Lemma 4.6.

[6] Define the cycle index for sum of the both type of orthogonal groups to be

1+n=1\displaystyle 1+\sum_{n=1}^{\infty} (zn|On+(q)|αO+(q)φ=u±1xφ,λφ±(α)φu,u±1xφ,λφ(α)\displaystyle\left(\dfrac{z^{n}}{|\mathrm{O}^{+}_{n}(q)|}\sum\limits_{\alpha\in\mathrm{O}^{+}(q)}\prod\limits_{\varphi=u\pm 1}x_{\varphi,\lambda_{\varphi}^{\pm}(\alpha)}\prod\limits_{\varphi\neq u,u\pm 1}x_{\varphi,\lambda_{\varphi}(\alpha)}\right.
+\displaystyle+ zn|On(q)|αO(q)φ=u±1xφ,λφ±(α)φu,u±1xφ,λφ(α)).\displaystyle\left.\dfrac{z^{n}}{|\mathrm{O}^{-}_{n}(q)|}\sum\limits_{\alpha\in\mathrm{O}^{-}(q)}\prod\limits_{\varphi=u\pm 1}x_{\varphi,\lambda_{\varphi}^{\pm}(\alpha)}\prod\limits_{\varphi\neq u,u\pm 1}x_{\varphi,\lambda_{\varphi}(\alpha)}\right).

This quantity factorises as

φ=u±1(1+n1λ±nxφ,λ±zncO,u±1,q(λ±))φ=φφu±1(1+n1λnxφ,λ(zdegφ)ncGL,u1,(qdegφ)1/2(λ))\displaystyle\prod\limits_{\varphi=u\pm 1}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda^{\pm}\vdash n}x_{\varphi,\lambda^{\pm}}\dfrac{z^{n}}{c_{\mathrm{O},u\pm 1,q}(\lambda^{\pm})}\right)\prod\limits_{\begin{subarray}{c}\varphi=\varphi^{*}\\ \varphi\neq u\pm 1\end{subarray}}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}\dfrac{(-z^{\deg\varphi})^{n}}{c_{\mathrm{GL},u-1,-(q^{\deg\varphi})^{1/2}}(\lambda)}\right)
×\displaystyle\times {φ,φ}φφ(1+n1λnxφ,λxφ,λz2ndegφcGL,u1,qdegφ(λ))\displaystyle\prod\limits_{\begin{subarray}{c}\{\varphi,\varphi^{*}\}\\ \varphi\neq\varphi^{*}\end{subarray}}\left(1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}x_{\varphi^{*},\lambda}\dfrac{z^{2n\deg\varphi}}{c_{\mathrm{GL},u-1,q^{\deg\varphi}}(\lambda)}\right)
Theorem 4.7.

Let anϵa^{\epsilon}_{n} denote the number of elements in Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q) which are MM-th root of identity, where ϵ{±}\epsilon\in\{\pm\}. Let M=tprM=t\cdot p^{r}, where ptp\nmid t. Then the generating function of the sum of the probabilities anϵ/|Onϵ(q)|a^{\epsilon}_{n}/|\mathrm{O}^{\epsilon}_{n}(q)| is given by

1+n=1(an+|On+(q)|+an|On(q)|)zn\displaystyle 1+\sum\limits_{n=1}^{\infty}\left(\dfrac{a_{n}^{+}}{|\mathrm{O}^{+}_{n}(q)|}+\dfrac{a_{n}^{-}}{|\mathrm{O}^{-}_{n}(q)|}\right)z^{n}
=\displaystyle= (1+n1λ±nλ1±przncO,u±1,q(λ±))o(t)m=1(d|tdDm[1+n1λ(z)ne(d)cGL,u1,(qe(d))1/2(λ)]ϕ(d)e(d)\displaystyle\left(1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda^{\pm}\vdash n\\ \lambda^{\pm}_{1}\leq p^{r}\end{subarray}}\dfrac{z^{n}}{c_{\mathrm{O},u\pm 1,q}(\lambda^{\pm})}\right)^{o(t)}\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|t\\ d\in D_{m}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda\end{subarray}}\dfrac{(-z)^{n}e(d)}{c_{\mathrm{GL},u-1,-(q^{e(d)})^{1/2}}(\lambda)}\right]^{\frac{\phi(d)}{e(d)}}\right.
d|tdDm,d1,2[1+n1λnz2ne(d)cGL,u1,qe(d)(λ)]ϕ(d)2e(d))\displaystyle\left.\prod\limits_{\begin{subarray}{c}d|t\\ d\not\in D_{m},d\neq 1,2\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}\dfrac{z^{2ne(d)}}{c_{\mathrm{GL},u-1,{q^{e(d)}}}(\lambda)}\right]^{\frac{\phi(d)}{2e(d)}}\right)
Proof.

The reason for clubbing these two probabilities is related to how the cycle indices for the different types of orthogonal groups are treated. This follows from the proof of Corollary 3.9 and Proposition 3.13. The occurrence of o(M)o(M) can be justified as was done in Corollary 3.9. Finally, we need to use Lemma 4.6. ∎

A formula for the generating function for the difference of the probabilities an+|On+(q)|an|O(q)|\dfrac{a^{+}_{n}}{|\mathrm{O}^{+}_{n}(q)|}-\dfrac{a^{-}_{n}}{|\mathrm{O}^{-}(q)|} can be formulated easily, and using the same techniques we can obtain a formula for the generating function. This is omitted here. But, a treatment for a special case can be found in [9].

Lemma 4.8.

[6, 7, Theorem 10, pp 63.] For G(q)=Un(q2)G(q)=\mathrm{U}_{n}(q^{2}), we have

1+n=1zn|Un(q)|(αUn(q2)φuxφ,λφ(α))\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{z^{n}}{|\mathrm{U}_{n}(q)|}\left(\sum\limits_{\alpha\in\mathrm{U}_{n}(q^{2})}\prod\limits_{\begin{subarray}{c}\varphi\neq u\end{subarray}}x_{\varphi,\lambda_{\varphi}(\alpha)}\right)
=\displaystyle= φuφ=φ~[1+n1λnxφ,λ(z)ndegφ(q)degφ(i(λi)2)i1(1(q)degφ)mi(λφ)]\displaystyle\prod\limits_{\begin{subarray}{c}\varphi\neq u\\ \varphi=\widetilde{\varphi}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}\dfrac{(-z)^{n\cdot\deg\varphi}}{(-q)^{\deg\varphi\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{(-q)^{\deg\varphi}}\right)_{m_{i}(\lambda_{\varphi})}}\right]
×\displaystyle\times φφ~{φ,φ~}[1+n1λnxφ,λxφ~,λz2ndegφq2degφ(i(λi)2)i1(1q2degφ)mi(λφ)]\displaystyle\prod\limits_{\begin{subarray}{c}\varphi\neq\widetilde{\varphi}\\ \{\varphi,\widetilde{\varphi}\}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\lambda\vdash n}x_{\varphi,\lambda}x_{\widetilde{\varphi},\lambda}\dfrac{z^{2n\cdot\deg\varphi}}{q^{2\deg\varphi\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{q^{2\deg\varphi}}\right)_{m_{i}(\lambda_{\varphi})}}\right]
Theorem 4.9.

Let ana_{n} denote the number of elements in Un(q2)\mathrm{U}_{n}(q^{2}) which are MM-th root of identity. Let M=tprM=t\cdot p^{r}, where ptp\nmid t. Then the generating function for the probability an/|Un(q2)|a_{n}/|\mathrm{U}_{n}(q^{2})| is given by

1+n=1an|Un(q2)|zn=\displaystyle 1+\sum\limits_{n=1}^{\infty}\dfrac{a_{n}}{|\mathrm{U}_{n}(q^{2})|}z^{n}= m=1(d|tdD~m[1+n1λnλ1pr(z)ne(d)(q)e(d)(i(λi)2)i1(1(q)e(d))mi(λφ)]ϕ(d)e(d)\displaystyle\prod\limits_{m=1}^{\infty}\left(\prod\limits_{\begin{subarray}{c}d|t\\ d\in\widetilde{D}_{m}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda\vdash n\\ \lambda_{1}\leq p^{r}\end{subarray}}\dfrac{(-z)^{n\cdot e(d)}}{(-q)^{e(d)\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{(-q)^{e(d)}}\right)_{m_{i}(\lambda_{\varphi})}}\right]^{\frac{\phi(d)}{e(d)}}\right.
×\displaystyle\times d|tdD~m[1+n1λnλ1prz2ne(d)q2e(d)(i(λi)2)i1(1q2e(d))mi(λφ)]ϕ(d)2e(d)),\displaystyle\left.\prod\limits_{\begin{subarray}{c}d|t\\ d\not\in\widetilde{D}_{m}\end{subarray}}\left[1+\sum\limits_{n\geq 1}\sum\limits_{\begin{subarray}{c}\lambda\vdash n\\ \lambda_{1}\leq p^{r}\end{subarray}}\dfrac{z^{2n\cdot e(d)}}{q^{2e(d)\cdot(\sum_{i}(\lambda_{i}^{\prime})^{2})}\prod\limits_{i\geq 1}\left(\dfrac{1}{q^{2e(d)}}\right)_{m_{i}(\lambda_{\varphi})}}\right]^{\frac{\phi(d)}{2e(d)}}\right),

where e(d)e(d) denotes the multiplicative order of qq in /d×\mathbb{Z}/d\mathbb{Z}^{\times}.

Proof.

This follows from the proof of Corollary 3.12 and Proposition 3.13. Finally, we need to use Lemma 4.8. ∎

5. An example

In this section, we compute the exact probility when q1(modM)q\equiv-1\pmod{M}. All the groups are defined over 𝔽q\mathbb{F}_{q}. We concentrate mainly on the case when M2M\not=2 is a prime and q1(modM)q\equiv-1\pmod{M}, for example, (q,M)=(41,7)(q,M)=(41,7). Then we get

xM1=(x1)QM(x).\displaystyle x^{M}-1=(x-1)\cdot Q_{M}(x).

Using Lemma 3.2, it will factor into (M1)/e(M)(M-1)/e(M) many distinct monic polynomials of degree e(M)=2e(M)=2. In this case, the elements contributing to MM-th root of identity will all be semisimple (see the discussion at the beginning of Section 3).

5.1. The case of GLn(q)\mathrm{GL}_{n}(q)

Let bnb_{n} denote the proportion of MM-th roots of identity in GLn(q)\mathrm{GL}_{n}(q). Then using Corollary 3.4, we get that

1+n=1bnzn\displaystyle 1+\sum\limits_{n=1}^{\infty}b_{n}z^{n} =(1+m=1zm|GLm(q)|)(1+m=1z2m|GLm(q2)|)(M1)/2.\displaystyle=\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{m}}{|\mathrm{GL}_{m}(q)|}\right)\left(1+\sum\limits_{m=1}^{\infty}\dfrac{z^{2m}}{|\mathrm{GL}_{m}(q^{2})|}\right)^{{(M-1)}/{2}}.

We now divide the computation into two cases. The first is nn being odd. In this case, we should have an odd power of zz, coming from the first term of the product. Other contributing powers of zz will have all even power. Hence probability of being an MM-th root is

1jMj=odd1|GLj(q)|(λMj21|GLλ(q2)|),\displaystyle\sum\limits_{\begin{subarray}{c}1\leq j\leq M\\ j=\text{odd}\end{subarray}}\dfrac{1}{|\mathrm{GL}_{j}(q)|}\cdot\left(\sum\limits_{\lambda\vdash\frac{M-j}{2}}\prod\limits_{\ell}\dfrac{1}{|\mathrm{GL}_{\lambda_{\ell}}(q^{2})|}\right),

where \ell runs over the subscripts of the parts of λ=(λ1,λ2,)\lambda=(\lambda_{1},\lambda_{2},\ldots). When nn is even, using the same argument as before, we get the resulting probability to be

0jMj=even1|GLj(q)|(λMj21|GLλ(q2)|),\displaystyle\sum\limits_{\begin{subarray}{c}0\leq j\leq M\\ j=\text{even}\end{subarray}}\dfrac{1}{|\mathrm{GL}_{j}(q)|}\cdot\left(\sum\limits_{\lambda\vdash\frac{M-j}{2}}\prod\limits_{\ell}\dfrac{1}{|\mathrm{GL}_{\lambda_{\ell}}(q^{2})|}\right),

where \ell runs over the subscripts of the parts of λ=(λ1,λ2,)\lambda=(\lambda_{1},\lambda_{2},\ldots) and |GL0(q)||\mathrm{GL}_{0}(q)| is 11 by convention.

5.2. The case of Sp2n(q)\mathrm{Sp}_{2n}(q)

Since MM is odd, we have that o(M)=1o(M)=1. We can have two possibilities, either MDmM\in D_{m} for some mm, or MDmM\not\in D_{m} for all m1m\geq 1. If MDmM\in D_{m} for some mm, then we have that the probability of an element of Sp2n(q)\mathrm{Sp}_{2n}(q) to be MM-th root, using Corollary 3.8, is

j=1n1|Sp2j(q)|(λ𝐏M12(Mj)1|Uλ(q4)|),\displaystyle\sum\limits_{j=1}^{n}\dfrac{1}{|\mathrm{Sp}_{2j}(q)|}\cdot\left(\sum\limits_{\lambda\in\mathbf{P}_{\frac{M-1}{2}}({M-j})}\prod\limits_{\ell}\dfrac{1}{|\mathrm{U}_{\lambda_{\ell}}(q^{4})|}\right),

and when MDmM\not\in D_{m}, then the resulting probability will be

j=1n1|Sp2j(q)|(λ𝐏M14(Mj)1|GLλ(q2)|).\displaystyle\sum\limits_{j=1}^{n}\dfrac{1}{|\mathrm{Sp}_{2j}(q)|}\cdot\left(\sum\limits_{\lambda\in\mathbf{P}_{\frac{M-1}{4}}({M-j})}\prod\limits_{\ell}\dfrac{1}{|\mathrm{GL}_{\lambda_{\ell}}(q^{2})|}\right).

The cases for the orthogonal groups Onϵ(q)\mathrm{O}^{\epsilon}_{n}(q) and the unitary group Un(q2)\mathrm{U}_{n}(q^{2}) are similar and we omit them from the display.

References

  • [1] Nir Avni, Tsachik Gelander, Martin Kassabov and Aner Shalev “Word values in pp-adic and adelic groups” In Bull. Lond. Math. Soc. 45.6, 2013, pp. 1323–1330
  • [2] László Babai, Simon Guest, Cheryl E. Praeger and Robert A. Wilson “Proportions of rr-regular elements in finite classical groups” In J. Lond. Math. Soc. (2) 88.1, 2013, pp. 202–226
  • [3] Timothy C. Burness and Michael Giudici “Classical groups, derangements and primes” 25, Australian Mathematical Society Lecture Series Cambridge University Press, Cambridge, 2016, pp. xviii+346
  • [4] S. Chowla, I.. Herstein and W.. Scott “The solutions of xd=1x^{d}=1 in symmetric groups” In Norske Vid. Selsk. Forh., Trondheim 25, 1952, pp. 29–31
  • [5] Veikko Ennola “On the conjugacy classes of the finite unitary groups” In Ann. Acad. Sci. Fenn. Ser. A I 313, 1962, pp. 13
  • [6] Jason Fulman “Cycle indices for the finite classical groups” In J. Group Theory 2.3, 1999, pp. 251–289
  • [7] Jason Fulman “Random matrix theory over finite fields” In Bull. Amer. Math. Soc. (N.S.) 39.1, 2002, pp. 51–85
  • [8] Jason Fulman, Peter M. Neumann and Cheryl E. Praeger “A generating function approach to the enumeration of matrices in classical groups over finite fields” In Mem. Amer. Math. Soc. 176.830, 2005, pp. vi+90
  • [9] Jason Fulman, Peter M. Neumann and Cheryl E. Praeger “A generating function approach to the enumeration of matrices in classical groups over finite fields” In Mem. Amer. Math. Soc. 176.830, 2005, pp. vi+90
  • [10] Samuel Gonshaw, Martin W. Liebeck and E.. O’Brien “Unipotent class representatives for finite classical groups” In J. Group Theory 20.3, 2017, pp. 505–525
  • [11] J.. Green “The characters of the finite general linear groups” In Trans. Amer. Math. Soc. 80, 1955, pp. 402–447
  • [12] Larry C. Grove “Classical groups and geometric algebra” 39, Graduate Studies in Mathematics American Mathematical Society, Providence, RI, 2002, pp. x+169
  • [13] Chun Yin Hui, Michael Larsen and Aner Shalev “The Waring problem for Lie groups and Chevalley groups” In Israel J. Math. 210.1, 2015, pp. 81–100
  • [14] Hideki Ishihara, Hiroyuki Ochiai, Yugen Takegahara and Tomoyuki Yoshida pp-divisibility of the number of solutions of xp=1x^{p}=1 in a symmetric group” In Ann. Comb. 5.2, 2001, pp. 197–210
  • [15] Tatsuhiko Koda, Masaki Sato and Yugen Takegahara “2-adic properties for the numbers of involutions in the alternating groups” In J. Algebra Appl. 14.4, 2015, pp. 1550052\bibrangessep21
  • [16] Rijubrata Kundu and Anupam Singh “Generating functions for the powers in GL(n, q)” In Israel J. Math. 259.2, 2024, pp. 887–936
  • [17] Michael Larsen and Dong Quan Ngoc Nguyen “Waring’s problem for unipotent algebraic groups” In Ann. Inst. Fourier (Grenoble) 69.4, 2019, pp. 1857–1877
  • [18] Michael Larsen and Aner Shalev “Word maps and Waring type problems” In J. Amer. Math. Soc. 22.2, 2009, pp. 437–466
  • [19] Michael Larsen and Aner Shalev “Words, Hausdorff dimension and randomly free groups” In Math. Ann. 371.3-4, 2018, pp. 1409–1427
  • [20] Michael Larsen, Aner Shalev and Pham Huu Tiep “The Waring problem for finite simple groups” In Ann. of Math. (2) 174.3, 2011, pp. 1885–1950
  • [21] Rudolf Lidl and Harald Niederreiter “Finite fields” With a foreword by P. M. Cohn 20, Encyclopedia of Mathematics and its Applications Cambridge University Press, Cambridge, 1997, pp. xiv+755
  • [22] San Ling and Chaoping Xing “Coding theory” A first course Cambridge University Press, Cambridge, 2004, pp. xii+222
  • [23] I.. Macdonald “Numbers of conjugacy classes in some finite classical groups” In Bull. Austral. Math. Soc. 23.1, 1981, pp. 23–48
  • [24] Gunter Malle and Donna Testerman “Linear algebraic groups and finite groups of Lie type” 133, Cambridge Studies in Advanced Mathematics Cambridge University Press, Cambridge, 2011, pp. xiv+309
  • [25] Alice C. Niemeyer, Tomasz Popiel and Cheryl E. Praeger “Abundant pp-singular elements in finite classical groups” In J. Algebra 408, 2014, pp. 189–204
  • [26] Alice C. Niemeyer, Tomasz Popiel and Cheryl E. Praeger “Algorithms to identify abundant pp-singular elements in finite classical groups” In Bull. Aust. Math. Soc. 86.1, 2012, pp. 50–63
  • [27] Alice C. Niemeyer and Cheryl E. Praeger “Estimating proportions of elements in finite groups of Lie type” In J. Algebra 324.1, 2010, pp. 122–145
  • [28] Alice C. Niemeyer and Cheryl E. Praeger “On permutations of order dividing a given integer” In J. Algebraic Combin. 26.1, 2007, pp. 125–142
  • [29] Saikat Panja and Anupam Singh “Powers in finite orthogonal and symplectic groups: A generating function approach” In Israel Journal of Mathematics, 2023, pp. To appear
  • [30] Saikat Panja and Anupam Singh “Powers in finite unitary groups”, 2023 arXiv:2304.13735 [math.GR]
  • [31] Aner Shalev “Word maps, conjugacy classes, and a noncommutative Waring-type theorem” In Ann. of Math. (2) 170.3, 2009, pp. 1383–1416
  • [32] Ken-ichi Shinoda “The characters of Weil representations associated to finite fields” In J. Algebra 66.1, 1980, pp. 251–280
  • [33] D.. Taylor “Conjugacy Classes in Finite Orthogonal Groups” online notes, 2020 URL: https://www.maths.usyd.edu.au/u/don/code/Magma/GOConjugacy
  • [34] D.. Taylor “Conjugacy Classes in Finite Symplectic Groups” online notes, 2020 URL: https://www.maths.usyd.edu.au/u/don/code/Magma/SpConjugacy
  • [35] D.. Taylor “Conjugacy Classes in Finite Unitary Groups” online notes, 2020 URL: https://www.maths.usyd.edu.au/u/don/code/Magma/GUConjugacy.pdf
  • [36] G.. Wall “On the conjugacy classes in the unitary, symplectic and orthogonal groups” In J. Austral. Math. Soc. 3, 1963, pp. 1–62
  • [37] Joseph L. Yucas and Gary L. Mullen “Self-reciprocal irreducible polynomials over finite fields” In Des. Codes Cryptogr. 33.3, 2004, pp. 275–281