Let $\mathbb{C}[S_n]$ be the regular representation of the symmetric group $S_n$, and let $\mathbb{C}^n$ be the vector representation.

Question: Does there exist a representation $V$ (of dimension $(n-1)!$) such that $V\otimes\mathbb{C}^n\cong \mathbb{C}[S_n]$? If so, does $V$ admit any particularly nice description?

For example, when $n=4$, we can take $V = V(4)\oplus V(2,1,1) \oplus V(2,2)$.

This problem has a nice solution if we only ask the isomorphism to be equivariant for the subgroup $S_{n-1}\subset S_n$. We can realize $\mathbb{C}[S_n]$ as the cohomology of the configuration space of $n$ points in $\mathbb{R}^k$ for any odd $k$ (even $k=1$ is okay). Then forgetting the $n$th point gives us a fiber bundle that behaves like a product on cohomology, and we obtain an isomorphism $\mathbb{C}[S_n]\cong \mathbb{C}[S_{n-1}]\otimes\mathbb{C}^n$. But we have broken the symmetry by choosing which point to forget, so this is only an isomorphism of $S_{n-1}$ representations. I want to do everything $S_n$-equivariantly, which is indeed possible when $n=4$, as the above example demonstrates.

Update: As Geoff Robinson observes in the comments below, an $S_n$-representation $V$ is a solution to my problem if and only if the restriction of $V$ to $S_{n-1}$ is isomorphic to the regular representation.

  • 2
    $\begingroup$ You seem to be asking for a (not necessarily irreducible) character of $S_{n}$ of degree $(n-1)!$ which vanishes on all non-identity permutations ( in the natural permutation action) which have a fixed point. This character won't be irreducible for $n > 3$. You have demonstrated the existence for $n =4$ and there is such a character (irreducible of degree $2$) when $n = 3$. It seems tricky when $n >4$, but $n=5$ should be easy to check with a computer. $\endgroup$ – Geoff Robinson Jan 2 '16 at 23:32
  • 1
    $\begingroup$ @GeoffRobinson: Just to rephrase your comment, you have observed that my question is equivalent to asking for a representation $V$ of $S_n$ with the property that the restriction of $V$ to $S_{n-1}$ is isomorphic to the regular representation of $S_{n-1}$. $\endgroup$ – Nicholas Proudfoot Jan 3 '16 at 0:23
  • 2
    $\begingroup$ @GeoffRobinson: That's equivalent? The answer is positive, and for a rather simple reason: The module $\operatorname{Lie}_n$ as defined on page 3 of www-math.mit.edu/~rstan/transparencies/whouse.pdf restricts to $S_{\left\{2,3,\ldots,n\right\}}$ as a regular representation (permuting the given basis). $\endgroup$ – darij grinberg Jan 3 '16 at 0:29
  • 2
    $\begingroup$ Actually, Corollary 8.7 of Reutenauer's Free Lie Algebras says that $\operatorname{Lie}_n \cong \operatorname{Ind}^{S_n}_{C_n} E$, where $C_n$ is the cyclic group generated by an $n$-cycle in $S_n$, and where $E$ is any faithful $1$-dimensional representation of $C_n$ (that is, the representation which sends said cycle to a primitive $n$-th root of unity). Combining this with the projection formula that is item 3 in mathoverflow.net/q/18799 , we obtain $\operatorname{Lie}_n \otimes \mathbb{C}^n \to \mathbb{C}\left[S_n\right]$ (though, annoyingly, this gives us complex coefficients). $\endgroup$ – darij grinberg Jan 3 '16 at 1:19
  • 7
    $\begingroup$ It is actually immediate (with the benefit of hindsight) that the permutation module afforded by the action of the cosets of $\langle (12 \ldots n-1 n) \rangle$ will do for $V$, which is related to the Corollary in Reutenaur's book mentioned by Darij Grinberg. It suffices by Mackey's formula to note that no non-identity power of an $n$-cycle is conjugate to an element in $S_{n-1}$. $\endgroup$ – Geoff Robinson Jan 3 '16 at 1:39

Let $H$ be a regular subgroup of $S_n$, for instance a transitive cyclic subgroup of order $n$. Then the permutation module $V$ of the action on the coset space $S_n/H$ has the requested property, as only the identity element of $S_n$ lies in a conjugate of $H$ and fixes a point in the natural action at the same time.

(Remark: Just noticed that in the very same minute Geoff Robinson gave the identical answer in a comment. So I marked my answer CW.)

  • $\begingroup$ This is soooo nice! (@both of you.) Let me just spell out the exact isomorphism. Let $\left[n\right]$ denote the left $S_n$-set $\left\{1,2,\ldots,n\right\}$. Then, the map $S_n \to \left(S_n/H\right)\times \left[n\right], \ \sigma \mapsto \left(\sigma H, \sigma\left(1\right)\right)$ is an isomorphism of left $S_n$-sets (since it is $S_n$-equivariant and injective and its domain has the same size as its target). By linearization, it becomes an isomorphism $\mathbb{C}\left[S_n/H\right] \otimes \mathbb{C}^n \to \mathbb{C}\left[S_n\right]$ of left $S_n$-modules. $\endgroup$ – darij grinberg Jan 3 '16 at 1:52
  • 4
    $\begingroup$ This all said, I still want to know my isomorphism $\operatorname{Lie}_n \otimes \mathbb{C}^n \to \mathbb{C}\left[S_n\right]$ ! $\endgroup$ – darij grinberg Jan 3 '16 at 1:54

The fact that the Lie module (as proposed by Darij Grinberg) works, as well, as an explicit isomorphism of modules, follows from the theory of cyclic operads: see Corollary 6.9 in http://sites.math.northwestern.edu/~getzler/Papers/cyclic.pdf (to be precise, restrict the statement of that Corollary to $S_n\subset S_{n+1}$).

| cite | improve this answer | |
  • $\begingroup$ Thank you! This is more than I can chew off right now, of course. Does your book, or Loday-Valette, bring me anywhere close? $\endgroup$ – darij grinberg Jan 3 '16 at 2:59
  • $\begingroup$ @darijgrinberg Neither LV nor my book with Bremner discusses enough of cyclic operads (well, we don't discuss them at all, and LV discuss them just a little bit). Section 13.1.10 of LV would help in unraveling bits on Harrison homology which may be missing in Getzler-Kapranov, but as for cyclic operads and appropriate cyclic homology complexes, I kind of feel that GK can't be avoided! $\endgroup$ – Vladimir Dotsenko Jan 3 '16 at 12:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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