Let $G$ be a finite group, and let $k$ be a field whose characteristic divides $\left|G\right|$. Let $\rho:G\to \mathrm{End} V$ be a (finite-dimensional) representation of $G$ over $k$. Prove or disprove that

$\sum\limits_{g\in G} \frac{1}{\det\left(\mathrm{id}-T\rho\left(g\right)\right)} = 0$ as an equality between power series in $k\left[T\right]$.


Let $G$ be a finite group, and $k$ be any field. Let $\rho:G\to \mathrm{End} V$ be a (finite-dimensional) representation of $G$ over $k$. For every $d\geq 0$, let $n_d$ denote the dimension of the space $\left(\mathrm{Sym}^d V^{\ast}\right)^G$ of the $G$-invariant symmetric $d$-ary forms on $V$. Molien's formula states that

$\left|G\right|\cdot \sum\limits_{n=0}^{\infty} n_dT^d = \sum\limits_{g\in G} \frac{1}{\det\left(\mathrm{id}-T\rho\left(g\right)\right)}$ as an equality between power series in $k\left[T\right]$

if $\mathrm{char} k$ does not divide $\left|G\right|$. I am wondering whether this formula still holds if $\mathrm{char} k\mid \left|G\right|$. The standard proof, using the Maschke projection, does not make much sense in this case...


A simple-minded argument:

Pick any element $g\in G$ of order $p$, the characteristic of $k$, and let $C(g)$ be the centralizer of $g$.

Now $G\setminus C(g)$ is the disjoint union of orbits under the action of the inner automorphism $\iota_g:h\in G\mapsto ghg^{-1}\in G$, and those orbits are of size $p$. The sum of the terms in your sum corresponding to the elements of $g$ in one of those orbits is then zero, for those terms are all equal.

If $h\in C(g)$, then the terms in your sum corresponding to the elements $h$, $gh$, $g^2h$, $\dots$, $g^{p-1}h$ are all equal---because $g$ and $h$ commute, you can take them simutaneously to Jordan canonical form and $g$ has only $1$ as an eigenvalue---so that their sum is also zero.

We have thus partitioned $G$ into, on one hand, the orbits of $\iota_g$ in $G\setminus C(g)$, and, on the other, the cosets of $\langle g\rangle$ in $C(g)$, and checked that the sum of the terms in each part of this partition is zero. Therefore your sum is zero.

NB: Notice that the specific form of the terms in your sum does not really matter, as long as it only depend on the eigenvalues of the $g\in G$. Thus, for example, exactly the same reasoning shows that the "other" Molien formula $$\sum_{g\in G}\det(I-t\rho(g))$$ also vanishes.

  • $\begingroup$ Wow, this is beautiful! $\endgroup$ – darij grinberg Feb 23 '11 at 16:23

By a well known lemma in group theory, we can write any $g\in G$ uniquely as $h_1h_2$ where $h_1$ has order a power of $p$, $h_2$ has order prime to $p$, and $h_1h_2=h_2h_1$. Thus we may write $G$ as a disjoint union

$G=\cup_g P_g\cdot g$

where $g$ ranges over the elements of order prime to $p$ and $P_g$ denotes the set of elements of $p$-power order which commute with $g$. Fix a char. p representation $\rho$. Note that every element of $\rho(P_g\cdot g)$ has the same characteristic polynomial as $\rho(g)$. Also, $|P_g|$ is either divisible by $p$ or equal to one (copy the proof of Cauchy's Theorem). In the former case, we clearly have

$\sum\limits_{\tilde{g}\in P_g\cdot g} \det\left(\mathrm{id}-T\rho\left(\tilde{g}\right)\right)^{-1} = 0$.

The remaining $g$'s all have centralizer of order prime to $p$, so each of their conjugacy classes has order divisible by $p$. It follows from this observation that the sum over the remaining elements must vanish as well.

  • $\begingroup$ Thanks. I didn't even read your post to the end; after the "well known lemma" it was immediately clear to me. I am going to accept your solution as soon as you add a reference to this lemma. $\endgroup$ – darij grinberg Feb 23 '11 at 12:55
  • $\begingroup$ @darij: The reference is "folklore". Write $ord(g)=ab$ with relative prime $a,b$ and choose $s,t$ such that $sa+bt=1$. Then $g=g^1=(g^a)^s \cdot (g^b)^t$. Obviously the two factors commute and $ord((g^a)^s) | b$ and $ord((g^b)^t) | a$. In fact the orders equal those terms because $ab=ord(g)=ord((g^a)^s)ord((g^b)^t) | ba$. One the other hand: If $g=xy$ with $ord(x)=b, ord(y)=a$ and $xy=yx$ then $g^{sa} = (x^{sa})(y^{sa})=x^{sa}=x^{sa \mod b}=x$ and $g^{tb}=(x^{tb})(y^{tb})=y^{tb}=y^{tb \mod a}=y$. $\endgroup$ – Johannes Hahn Feb 23 '11 at 15:16
  • $\begingroup$ Ah, thanks! I was stupid. I confused the "well known lemma" with a lemma I wanted to use in my proof and couldn't show. They only see what they want to see. $\endgroup$ – darij grinberg Feb 23 '11 at 15:58
  • $\begingroup$ Oh. Wait. In $G=\bigcup_g P_g\cdot g$, why is the union disjoint? $\endgroup$ – darij grinberg Feb 23 '11 at 16:08
  • 1
    $\begingroup$ The union is disjoint by uniqueness of the decomposition $g=h_1h_2$. Anyway, Mariano's argument is cleaner. $\endgroup$ – Kevin Ventullo Feb 23 '11 at 18:25

I am not specialist of the subject but seems that the equality does not work in the special case:

$k=\mathbb{F}_2,$ $G=\{0,1\}$ the additive group of the same field $\mathbb{F}_2,$ the vector space still the same field $V=k,$ and the representation $$ \rho(0)=id,\;\;\rho(1)=0 $$

  • 1
    $\begingroup$ This is not a representation: $\rho(1)\cdot\rho(1)\neq \rho(1+1)$. $\endgroup$ – Kevin Ventullo Feb 23 '11 at 6:10
  • $\begingroup$ @Kevin: You are right. I misunderstood the definition... $\endgroup$ – Luis H Gallardo Feb 23 '11 at 8:11
  • $\begingroup$ @Kevin, more importantly, $\rho(1)$ is not an automorphism :) $\endgroup$ – Mariano Suárez-Álvarez Feb 23 '11 at 16:25

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.