7
$\begingroup$

Let $G$ be a finite group, $g \geq 0, k\geq 1 $ integers, and $(C_1,...,C_k)$ a tuple of conjugacy classes of $G$. I am interested in proofs of the following identity

$$ \sum_{(c_1,...,c_k) \in C_1 \times \cdots \times C_k} \sum_{V_\lambda} d(V_\lambda)^{1-2g} \chi_\lambda(c_1^{-1}\cdots c_k^{-1}) = |C_1|\cdots|C_k|\sum_{V_\lambda} d(V_\lambda)^{2-2g-k}\chi_\lambda(c_1)\cdots \chi_\lambda(c_k) $$

where the sums with $V_\lambda$ are over all irreducible characters of $G$ and $d(V_\lambda) = \chi_\lambda(1)$ is the dimension of the representation. I obtained this from messing around with a theorem due to Frobenius (for $g=0$ and later generalized originally using a quantum topology approach - hence the $g$ for genus).

Is there some nice (ideally purely character-theoretic) proof of this identity?

$\endgroup$
5
  • $\begingroup$ This seems closely related to the functional identity $\lvert C_2\rvert\chi_\lambda(c_1)\chi_\lambda(c_2) = \chi_\lambda(1)\sum_{c_2' \in C_2} \chi_\lambda(c_1 c_2')$. I don't see the proof immediately, but I'll think about it. (Note that the inverses don't change anything, as re-indexing by $\lambda \mapsto \overline\lambda$ in the first sum shows.) $\endgroup$
    – LSpice
    Mar 30, 2021 at 16:13
  • 1
    $\begingroup$ is not it true that for fixed $\lambda$ we have (the contribution of $\lambda$ to LHS)=(the contribution of $\bar{\lambda}$ to RHS)? $\endgroup$ Mar 30, 2021 at 16:26
  • $\begingroup$ @FedorPetrov, yup, I'm writing the proof now. $\endgroup$
    – LSpice
    Mar 30, 2021 at 16:26
  • $\begingroup$ What is the $d$ in $d(V_\lambda)^{1 - 2g}$ ? $\endgroup$ Mar 30, 2021 at 18:38
  • $\begingroup$ @InesInstitoris It is the dimension of $V_\lambda$ - thanks I edited the question. $\endgroup$
    – user101010
    Mar 31, 2021 at 17:45

2 Answers 2

8
$\begingroup$

$\newcommand\card[1]{\lvert#1\rvert}$I use the functional identity $\card{C_1}\chi(c_1)\chi(c_2) = \chi(1)\sum_{c_1' \in C_1} \chi(c_1' c_2)$, which identifies multiples of irreducible characters $\chi$, and which I'll re-write in the form $$ \sum_{c_1 \in C_1} \chi(c_1)\chi(c_2) = \chi(1)\sum_{c_1 \in C_1} \chi(c_1 c_2). $$ This was my advisor (Paul Sally)'s all-time favourite result, and I am delighted to have a chance to use it.

I replace the cardinality on the RHS by a sum over $C_1 \times \dotsb \times C_k$. I also re-index the LHS by $\overline\lambda$ instead of $\lambda$, so that there's no need for inverses. Finally, I work with the $\chi$-by-$\chi$ version, as suggested by @FedorPetrov in the comments, in the sense that $$ \chi(1)^{1 - 2g}\sum_{(c_1, \dotsc, c_k) \in C_1 \times \dotsb \times C_k} \chi(c_1\dotsm c_k) = \chi(1)^{2 - 2g - k}\sum_{(c_1, \dotsc, c_k) \in C_1 \times \dotsb \times C_k} \chi(c_1)\dotsm\chi(c_k), $$ or rather that $$ \chi(1)^{k - 1}\sum_{(c_1, \dotsc, c_k) \in C_1 \times \dotsb \times C_k} \chi(c_1\dotsm c_k) = \sum_{(c_1, \dotsc, c_k) \in C_1 \times \dotsb \times C_k} \chi(c_1)\dotsm\chi(c_k). $$

We prove this modified result by induction. It's obvious for $k = 1$. (Most inductions that I know that start at $k = 1$ have an interesting degenerate case at $k = 0$, but I don't know what the proper degenerate case is here! Plugging in $k = 0$ naïvely doesn't work.) In general, we have \begin{align*} & \sum_{c_1 \in C_1} \chi(c_1)\Bigl( \sum_{(c_2, \dotsc, c_{k + 1}) \in C_2 \times \dotsb \times C_{k + 1}} \chi(c_2\dotsm c_{k + 1}) \Bigr) \\ ={} & \sum_{c_1 \in C_1} \chi(c_1)\Bigl( \chi(1)^{k - 1}\sum_{(c_2, \dotsc, c_{k + 1}) \in C_2 \times \dotsb \times C_{k + 1}} \chi(c_2\dotsm c_{k + 1})\Bigr) \\ ={} & \chi(1)^k\sum_{(c_1, \dotsc, c_{k + 1}) \in C_1 \times \dotsb C_{k + 1}} \chi(c_1 c_2\dotsm c_{k + 1}), \end{align*} as desired.

$\endgroup$
2
  • 1
    $\begingroup$ Thank you for the lovely answer! That is very nice and I was unaware of the identity that Paul Sally liked so much :-) $\endgroup$
    – user101010
    Mar 31, 2021 at 17:47
  • $\begingroup$ This works equally well, with sums over conjugacy classes replaced by normalised integrals, for compact groups. $\endgroup$
    – LSpice
    Apr 1, 2021 at 13:02
8
$\begingroup$

Here is another way to view things: it uses the fact that for each complex irreducible character $\chi$ of $G$, there is an algebra homomorphism $\omega_{\chi} : Z(\mathbb{C}G) \to \mathbb{C}$ defined by $\omega_{\chi}(X) = \frac{\chi(X)}{\chi(1)} .$ This is the underlying root (from a modern viewpoint) of the formula of Frobenius quoted in the question, and the formula beloved of Paul Sally quoted in L. Spice's answer.

For $S$ a subset of $G$ invariant under conjugation, let $S^{+} = \sum_{s \in S} s$ in $\mathbb{C}G.$

Hence for each $\lambda$ we find that $$\frac{\chi_{\lambda}(C_{1}^{+}C_{2}^{+} \ldots C_{k}^{+})}{\chi_{\lambda}(1)} = \prod_{i=1}^{k} |C_{i}|\frac{\chi_{\lambda} (c_{i})}{\chi_{\lambda}(1)}.$$

Expressing the element $C_{1}^{+} \ldots C_{k}^{+}$ as a sum of products of group elements explains the formula of the question.

$\endgroup$
3
  • $\begingroup$ Thank you very much for the answer - that is a very enlightening perspective :-). Is there a name for the map $\omega_lambda$? $\endgroup$
    – user101010
    Mar 31, 2021 at 17:49
  • $\begingroup$ It's sort of the infinitesimal character, isn't it? $\endgroup$
    – LSpice
    Mar 31, 2021 at 17:52
  • 1
    $\begingroup$ As a finite group theorist, I only know it as the "central character associated to $\chi$. Its existence comes via Schur's Lemma ( which tells us that all central elements of $\mathbb{C}G$ are represented by scalar matrices), and consideration of what $\omega_{\chi}$ does to the central idempotents associated to each irreducible character. Clearly it takes value $1$ at $e_{\chi}$ and value $0$ at $e_{\mu}$ for every irreducible character $\mu$ different from $\chi$.. $\endgroup$ Mar 31, 2021 at 17:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.