1 Introduction
Given a closed nonorientable surface, a result of Frobenius and Schur [Reference Frobenius and Schur4] counts the number of homomorphisms from the fundamental group of the surface to a finite group G in terms of dimensions of the irreducible complex representations of G, the Frobenius-Schur indicators of these representations, the order of G, and the nonorientable genus of the surface. Given a closed orientable surface, a result of Mednykh [Reference Mednykh7] counts the number of homomorphisms from the fundamental group of the surface to a finite group G in terms of the dimensions of the irreducible complex representations of G, the order of G, and the genus of the surface. Given a compact oriented surface with labeled boundary components, a finite group G, and a choice of conjugacy class in G for each boundary component of the surface, Dijkgraaf and Witten [Reference Dijkgraaf and Witten1] as well as Freed and Quinn [Reference Freed and Quinn2] gave formulas (in the context of mathematical physics) of a formula for the number of homomorphisms from the fundamental group of the surface to G, sending the conjugacy classes of the boundaries to the chosen conjugacy classes in terms of the dimensions of the irreducible complex representations of G, the values of the characters of these representations on the chosen conjugacy classes, the order of G, the sizes of the conjugacy classes, and the genus of the surface. An elementary character theoretic proof of this was supplied by Zagier [Reference Zagier14]. In the case where the genus of the surface is zero, the result was also known to Frobenius [Reference Frobenius3].
Theorem 3.1 is the missing case where the surface is compact and nonorientable with labeled boundary components. In this paper, we supply simple character-theoretic proofs of all of these results following the treatments given by Zagier [Reference Zagier14] and Mulase and Yu (see section 4 of [Reference Mulase and Yu8]). We then draw some topological and combinatorial conclusions from these results. For a more extensive overview of the history of these equations in the case of closed surfaces, see [Reference Snyder12].
In Section 2, we discuss a framework for understanding all of these equations as consequence of character-theoretic ideas. In Section 3, we make the connection to 2-dimensional topology, and in Section 4, we show how these equations can be used to deduce some symmetric function identities, following Stanley [Reference Stanley11].
2 Counting solutions to equations in groups
All of our representations are over $\mathbb {C}$ . Let r be positive integer, let $F_r$ be the free group on r letters, and $\gamma \in F_r$ . Then for any finite group G, we have the class function $f_{\gamma } = f_{\gamma , G}$ with
where $\gamma (g_1,...,g_r)$ is the image in G of the homomorphism from $F_r \to G$ , given by sending the ith generator of $F_r$ to $g_i$ . Since $f_\gamma $ is a class function, we have
for some coefficients $a_\chi ^\gamma \in \mathbb {C}$ where the sum is over irreducible characters $\chi $ of G. Two basic questions are as follows: When is $f_\gamma $ a character for all G (i.e., all of the $a_\chi ^\gamma $ are nonnegative integers), and when is $f_\gamma $ a virtual character for all G (i.e., all of the $a_\chi ^\gamma $ are integers)?
In general, we have
since with respect to the inner product of class functions given by
the characters $\chi $ form an orthonormal basis. Therefore,
This has been observed before, for example, in Proposition 3.1 in [Reference Parzanchevski and Schul9]. However, in special cases, such as $x_1^2x_2^2 \cdots x_k^2 \in F_k$ and $[x_1,y_1]\cdots [x_g,y_g] \in F_{2g}$ , the coefficients $a_\chi ^\gamma $ have a much more explicit form that we now discuss (see [Reference Parzanchevski and Schul9] for additional results on computing $a^\gamma _\chi $ for specific types of elements $\gamma $ ). As a word of warning, note that $f_\gamma $ depends on “where $\gamma $ is”; for example, $f_{x_1}$ depends on if $x_1 \in F_1$ or in $F_2$ .
Given a irreducible character $\chi $ of a finite group G, the Frobenius-Schur indicator of $\chi $ is
For all $\chi $ , $\nu (\chi ) \in \{ -1, 0, 1\}$ with $\nu (\chi ) \neq 0$ if and only if there is a nonzero G-invariant bilinear form on the representation with character $\chi $ , $\nu (\chi ) = 1$ if and only if there exists a symmetric such form, and $\nu (\chi ) = -1$ if and only if there exists a skew-symmetric such form [Reference Frobenius and Schur4]. At the root of our discussion is the following theorem of Frobenius [Reference Frobenius3] (for the first equation) and Frobenius and Schur [Reference Frobenius and Schur4] (for the second equation):
Theorem 2.1 (Frobenius, Schur) Let G be a finite group and $w \in G$ an element. Then
and
where the sums are over the irreducible characters of G.
Equivalently, Theorem 2.1 says that for every irreducible character $\chi $ , we have
for $xyx^{-1}y^{-1}$ in a free group generated by x and y and
for $x^2$ in the free group generated by x.
The following result (see [Reference Stanley11] Exercise 7.69 (d)) shows how to obtain an expression for $a_\chi ^{\gamma _1\cdots \gamma _m}$ in terms of $a_\chi ^{\gamma _1},..., a_\chi ^{\gamma _m}$ when the words $\gamma _1, ..., \gamma _m$ contain disjoint letters.
Proposition Let G be a finite group and let $f_1,...,f_m$ be class functions on G. Define the class function $F = F_{f_1,...,f_m}$ by
Let $\chi $ be an irreducible character of G. Then
Using this, we note that
and
Thus, from Theorem 2.1 together with Proposition 1, we obtain the following formula of Mednykh [Reference Mednykh7] for the first equation and Frobenius and Schur [Reference Frobenius and Schur4] for the second equation – see also [Reference Mulase and Yu8] where this phrasing is used:
Corollary (Mednykh, Frobenius, Schur) Let G be a finite group and $w \in G$ an element. For all integers $g, k \geq 0$ , we have
for $[x_1,y_1]\cdots [x_g,y_g]$ in the free group of rank $2g$ with generators $x_1,y_1,...,x_g,y_g$ and
for $x_1^2x_2^2 \cdots x_k^2$ in the free group of rank k generated by $x_1,...,x_k$ .
Remark 2.2 Note, for example, that setting $w = 1$ and $g = 1$ in Corollary 1, we see that $|G|$ divides
This is in fact a special case of a theorem of Solomon [Reference Solomon10] that implies that for any $\gamma \in F_r$ with $r> 1$ , $|G|$ divides $f_{\gamma ,G}(1)$ .
We now mention a few more character-theoretic results that we use in the sequel (for a proof, see, for example, chapter 3 of [Reference Isaacs5]).
Proposition Let $\chi $ be an irreducible character of a finite group G. Let $Z(\mathbb {C}G)$ be the center of the group algebra. Then the map
is an algebra homomorphism.
Proposition Let $C_1,...,C_n$ be not necessarily distinct conjugacy classes in a finite group G and let $\chi $ be an irreducible character of G. Then
where $\chi (C_i)$ denotes the value of $\chi $ on an element of $C_i$ .
Proof Let $C_i^+$ be
and note that $C_i^+$ is in the center $Z(\mathbb {C}G)$ . Then, by applying the algebra homomorphism $\omega _\chi $ from Proposition 2 to the product $C_1^+ \cdots C_n^+$ , we obtain
as desired.
We are now prepared to prove the following result that we need in the next section. The statement of the result is much more natural from a topological perspective, which we discuss in the next section.
Theorem 2.3 Let $\gamma \in F_r$ , G be a finite group, and $C_1,...,C_n$ be not necessarily distinct conjugacy classes of G. Then
Proof
where the removal of the inverse in the third equation is justified by resumming over the complex conjugates of the characters and recalling that $\overline {\chi }(g^{-1}) = \chi (g)$ .
Thus, if we have a more explicit formula for $a_\chi ^\gamma $ , we obtain a more explicit formula for the expression in the left-hand side of Theorem 2.3. We now do exactly that.
3 Relationship with 2-dimensional topology
In this section, we demonstrate how the following result is proven and how it relates to topology:
Theorem 3.1 Let $C_1, ..., C_n$ be a collection of not necessarily distinct conjugacy classes in a finite group G. Then
where the sum is over the characters of the irreducible complex representations of G and where $\nu (\chi )$ denotes the Frobenius-Schur indicator of $\chi $ .
Let $S_g$ be a closed orientable surface of genus g and let $N_k$ be a closed nonorientable surface of nonorientable genus k. Let G be a finite group. Noting that
and
we have
and
Thus, we can rewrite the formula of Mednykh in Corollary 1 with $w=1$ as
and similarly, we can rewrite the formula of Frobenius and Schur in Corollary 1 with $w=1$ as
Let $S_{g,n}$ denote the compact surface of genus g with n boundary components labeled from 1 to n. Further, fix an orientation on $S_{g,n}$ which thus induces an orientation on all of the boundary components of $S_{g,n}$ . Let $C_1,...,C_n$ be a choice of n not necessarily distinct conjugacy classes in G. Let
denote the set of homomorphisms from $\pi _1(S_{g,n})$ to G such that the conjugacy class given by the ith boundary component using the given orientation is sent to the conjugacy class $C_i$ for $1 \leq i \leq n$ . Noting that
we have that
Frobenius [Reference Frobenius3] proved for $g = 0$ , and Dijkgraaf and Witten [Reference Dijkgraaf and Witten1] as well as Freed and Quinn [Reference Freed and Quinn2] proved in general that for $n \geq 1$ that
where here $\chi (C_i)$ denotes the value of $\chi $ on any element in $C_i$ . Note that by setting $n = 0$ , this gives exactly Mednykh’s formula in Corollary 1.
We now demonstrate the use of Theorem 2.3 by giving a short proof of this formula. For another elementary proof, see [Reference Zagier14].
Proof (of Equation 3.1) Let $\gamma = [x_1,y_1] \cdots [x_g,y_g]$ in the free group $F_{2g}$ generated by $x_1,y_1,...,x_g,y_g$ . By Corollary 1, we have
for all irreducible representations $\chi $ of G. Therefore, by Theorem 2.3 together with the observation that
for all $\chi $ , the result follows.
This formula shows that the ordering of the conjugacy classes as $(C_1,...,C_n)$ does not affect $|\operatorname {\mathrm {Hom}}^{(C_1,...,C_n)}(\pi _1(S_{g,n}), G)|$ . This also follows directly without the use of the formula, as noted in [Reference Zagier14], since by using the identity
we have a bijection between the set
and the respective set
given by interchanging the order of the conjugacy classes $C_{i}$ and $C_{i+1}$ . Thus, as far as the number $|\operatorname {\mathrm {Hom}}^{(C_1,...,C_n)}(\pi _1(S_{g,n}), G)|$ is concerned, we only need to know $(C_1, ..., C_n)$ as a multiset.
Let $N_{k,n}$ be the compact nonorientable surface of nonorientable genus k with n boundary components labeled with the numbers 1 through n. Let
denote the set of homomorphisms from $\pi _1(N_{n,k})$ to G such that the conjugacy class given by the ith boundary component using the given orientation is sent to the conjugacy class $C_i$ for $1 \leq i \leq n$ . Noting that
we have that
Just as above in the orientable case, we see (with or without the help of Theorem 3.1) that this does not depend on the number of these homomorphisms and does not depend on the given ordering of the conjugacy classes $C_1,...,C_n$ .
Thus, Theorem 3.1 can be restated as
The proof of Theorem 3.1 is analogous the above proof in the orientable case.
Proof (of Theorem 3.1) Let $\gamma = x_1^2 x_2^2 \cdots x_k^2$ in the free group $F_{2g}$ generated by $x_1,x_2,...,x_k$ . By Corollary 1, we have
for all irreducible representations $\chi $ of G. Therefore, by Theorem 2.3 together with the observation that $\nu (\overline {\chi }) = \nu (\chi )$ and thus
for all $\chi $ , the result follows.
Note that, by setting $n = 0$ , this again gives Frobenius and Schur’s result in Corollary 1 (though this is not another proof of that result, just an observation).
Note that $\pi _1(S_{g,n})$ and $\pi _1(N_{k,n})$ are both free groups with
and
Note that for any finite group G, we have
Using this fact and summing over the possible tuples of conjugacy classes in Equation (3.1) and Theorem 3.1, we obtain the following:
Corollary Let G be a finite group and let $g, n, k$ be positive integers. Then
and
where the first sums are over all ordered n-tuples of conjugacy classes in G and the second sums are over all irreducible complex characters of G.
Then, taking the limit as $g \to \infty $ and $k \to \infty $ , and noting that any bilinear form on a 1-dimensional space is automotically symmetric and therefore 1-dimensional representations never have Frobenius-Schur indicator equal to -1, we have the following:
Corollary Let G be a finite group and let n be a positive integer.
where the first sums are over all ordered n-tuples of conjugacy classes $(C_1,...,C_n)$ in G and the second sums are over all irreducible complex characters of G that satisfy the specified conditions.
As remarked in [Reference Zagier14] in the orientable case, Equation (3.1) and Theorem 3.1 have a topological interpretation. In the orientable case, fill in each of the n boundary components of $S_{g,n}$ with disks and let $p_i$ denote the points at the center of these disks and denote the resulting closed surface by $\overline {S_{g,n}}$ . Let G have a faithful action on some set F. Then each element of $\operatorname {\mathrm {Hom}}^{(C_1,...,C_n)}( \pi _1(S_{g,n}), G)$ gives rise to a (not necessarily connected) branched covering of $\overline {S_{g,n}}$ with Galois group G and ramification points $p_i$ . Conversely, each such branched covering comes from $\operatorname {\mathrm {Hom}}^{(C_1,...,C_n)}( \pi _1(S_{g,n}), G)$ . So Equation (3.1) gives a method for computing the number of such branched coverings. The construction extends analogously to branched covers of nonorientable surfaces, and Theorem 3.1 then supplies the relevant counting formula.
As a closing remark for this section, we mention another family of words $\gamma $ where there is a nice formula for $a_\chi ^\gamma $ for all G and $\chi $ . Let
(these are sometimes referred to as generalized commutators). Then a formula of Leitz [Reference Leitz6] (see also [Reference Tambour13]) says that
where
Using this together with Proposition 1, a formula for the $f_\gamma $ follows where $\gamma $ is a product of generalized commutators just as in Theorem 1 (and this generalizes the formula for products of commutators in Theorem 1). Similarly, applying Theorem 2.3 with these words, we obtain a result generalizing Equation (3.1) (although with no apparent topological interpretation).
In a similar vein but generalizing the case of the elements $ x_1^2 x_2^2 \cdots x_k^2$ , we can consider words of the form $\gamma = x^n$ in a free group of rank one generated by x. In this case, the class function $f_\gamma $ counts the number of nth roots that each element in G has. Therefore, in this case, we have
where $\nu _n$ is the gereralized Frobenius-Schur indicator given by
(see, for example, Chapter 4 of [Reference Isaacs5]). In general, $\nu _n \in \mathbb {Z}$ so $f_\gamma $ is a virtual character. From Proposition 1, taking $\gamma = x_1^{n_1} x_2^{n_2} \cdots x_m^{n_m}$ in a free group of rank m generated by $x_1,...,x_m$ with $n_1,...,n_m \in \mathbb {Z}$ (note that, in general, $f_\gamma = f_{\gamma ^{-1}}$ ), we have
We could also involve conjugacy classes by applying Theorem 2.3. Similarly, we could take products of generalized commutators and powers of elements and do this again, obtaining a common generalization of all of the results.
4 Some symmetric function identities
In this section, we apply some of the identities from the previous sections to the symmetric group and use an isomorphism between the set of class functions on the symmetric group with a certain space of symmetric functions in order to obtain a few identities among symmetric functions. Let $S_n$ be the symmetric group on an n element set. The relevant definitions and background come from [Reference Stanley11]. Let $\Lambda $ be the ring of symmetric functions over the complex numbers and let $\Lambda ^n$ be the subspace of $\Lambda $ spanned by symmetric functions of degree n so that $\Lambda $ has a grading
Given a partition $\lambda $ of n, denoted $\lambda \vdash n$ , let $p_\lambda $ and $s_\lambda $ be the associated power and Schur symmetric functions. Let $\chi ^\lambda $ be the irreducible character of $S_n$ coming from the Specht module $S^\lambda $ . Let $H_\lambda $ denote the product of the hook lengths of the Young tableau associated to $\lambda $ . Given an element $w \in S_n$ , let $\rho (w)$ denote the partition of n given by cycle type of w. Let $R^n$ denote the set of class functions on $S_n$ . Then we have a vector space isomorphism
and this has the property that $\operatorname {\mathrm {ch}}(\chi ^\lambda ) = s_\lambda $ .
Given $\gamma \in F_r$ , then we have the class function $f_\gamma $ and
and using the expansion
we also have
Therefore, we have shown the following:
Theorem 4.1 Given $\gamma \in F_r$ ,
This is a generalization of exercise 7.68 (c) in [Reference Stanley11]. Taking the examples of $x_1^2 x_2^2 \cdots x_k^2 \in F_k$ and $[x_1,y_1]\cdots [x_g,y_g] \in F_{2g}$ and applying the hook length formula, which says that
together with the fact that all of the Frobenius-Schur indicators for the symmetric group are $1$ , we have the following:
Corollary For integers $n,k,g \geq 1$ , we have
and
Let $1^q$ denote the vector that is q ones followed by zeroes. Recalling that
where $\kappa (w)$ is the number of cycles in w, and that
where $c(t)$ denotes the content of $\lambda $ at t, then by specializing to $1^q$ , we obtain the following:
Corollary For integers $n,k,g \geq 1$ , we have
and
The first of these is a generalization of exercise 7.68 (e) of [Reference Stanley11]. More such identities can be obtained by applying the same ideas to the identities mentioned at the end of the last section.
As a final remark, fix a word $\gamma = x_1^2x_2^2 \cdots x_k^2 \in F_k$ or $[x_1,y_1]\cdots [x_g,y_g] \in F_{2g}$ and consider the sequence $f_{\gamma , S_n}$ (here, we have made the group explicitly a part of the notation) as n varies. The exponential generating function for this sequence in the case where $\gamma = x_1^2x_2^2 \cdots x_k^2 \in F_k$ is, by Corollary 1 together with the hook length formula,
which in the specific case of $k=2$ has the particularly nice form
(see [Reference Stanley11] exercise 5.12). Similarly for $\gamma = [x_1,y_1]\cdots [x_g,y_g] \in F_{2g}$ , we have the exponential generating function
which for $g = 1$ again gives the product in Equation (4.1).