Let $p$ be a prime and $n\geq 1$, it is known that there are exactly two extraspecial groups of order $p^{1+2n}$, denoted by $p^{1+2n}_+$ and $p^{1+2n}_-$. They fit into central extensions $$0\to(\mathbb{Z}/p)\to E\to(\mathbb{Z}/p)^{2n}\to 0.$$ When $p$ is odd, $E=p^{1+2n}_+$ has exponent $p$ and an explicit $2$-cocycle $\beta:V\times V\to\mathbb{Z}/p$ for $E$ (where $V=(\mathbb{Z}/p)^{2n}$, seen as an $\mathbb{F}_p$-vector space) is given by $\beta(v,w)=\frac{1}{2}[v,w]$, with $$[v,w]:=\displaystyle\sum_{i=0}^n{v_iw_{n+i}-w_iv_{n+i}}.$$ Is there any explicit description for $\beta$ when $p=2$?
-
$\begingroup$ For $p = 2$, the explicit description of the cocycle $\beta$ is: $$ \beta(v, w) = \sum_{i=1}^n v_i w_{n+i} + v_{n+i} w_i, $$ where $v$ and $w$ are elements of $(\mathbb{Z}/2)^{2n}$. $\endgroup$– zeraoulia rafikCommented Jul 16 at 13:23
-
1$\begingroup$ @zeraouliarafik I do not think it is that easy. $\endgroup$– AntoineCommented Jul 16 at 14:33
-
$\begingroup$ @zeraouliarafik where did you get that formula from? Looks to me like a copy of the formula in the original post but using that +1=-1 in characteristic 2. $\endgroup$– David Roberts ♦Commented Jul 17 at 4:00
2 Answers
When $p=2$, the bilinear form given by commutators no longer specifies the extraspecial group. You need to use a quadratic form $q\colon V \to \mathbb{Z}/2$ giving the square map. A commutator in a group is always a product of three squares (via $x^{-1}y^{-1}xy = (x^{-1})^2(xy^{-1})^2y^2$ ) and so $q$ determines the commutator map via $[x,y]=q(x+y)+q(x)+q(y)$.
Given the quadratic form $q$ determining the extraspecial group, there are many cocycles representing the cohomology class. They satisfy $\beta(x,x)=q(x)$, and $\beta(x,y)+\beta(y,x)=q(x+y)+q(x)+q(y)$. To choose such a $\beta$, choose a basis of $V$, say $v_1,\dots,v_{2n}$. Then set $\beta(v_i,v_i)=q(v_i)$, choose $\beta(v_i,v_j)$ arbitrarily for $i<j$, and then set $\beta(v_j,v_i)=q(v_i+v_j)+q(v_i)+q(v_j)+\beta(v_i,v_j)$. Now extend bilinearly.
There are two types of nonsingular quadratic forms, distinguished by their Arf invariants, and the recipe above works fine for both cases.
-
$\begingroup$ It might have made it clearer what was going on if I had written some of the plusses as minuses. But then people would complain that it makes no difference, so why do it? You can't win, you know. $\endgroup$ Commented Jul 16 at 15:30
-
$\begingroup$ Thanks a lot! I know where the condition $\beta(x,x)=q(x)$ comes from, but the other condition is not evident to me, does it follow from the cocycle equation? $\endgroup$– AntoineCommented Jul 16 at 16:28
-
1$\begingroup$ I guess the point is that if you lift elements of $V$ to $E$ then $\beta(x,y)$ tells you how the lift of the product differs from the product of the lifts. So $\beta(x,y)-\beta(y,x)$ tells you the commutator of the lifts because $V$ is commutative. The formula for the commutator in terms of the square tells you this is $q(x-y)-q(x)+q(y)$. And as mentioned before, minus is plus. I hope that helps. $\endgroup$ Commented Jul 16 at 16:32
-
-
1
The extraspecial group $P = 2_{+}^{1+2n}$ has generators $z$, $v_1$, $\ldots$, $v_n$, $w_1$, $\ldots$, $w_n$, with $Z(P) = \langle z \rangle$ cyclic of order $2$ and
\begin{align*} v_i^2 = w_i^2 = 1 & & \text{ for all } i\\ [v_i,v_j] = [w_i,w_j] = 1 && \text{ for all } i,j \\ [v_i,w_j] = 1 && \text{ for } i \neq j \\ [v_i,w_i] = z && \text{ for all } i \\ \end{align*}
So for $V = P/Z(P)$ you get the $2$-cocycle $f: V \times V \rightarrow Z(P)$ defined by $$f(v,v') = z^{\sum_{k = 1}^n i_k' j_k}$$
for $v = \prod_k v_k^{i_k} \prod_k w_k^{j_k}$ and $v' = \prod_k v_k^{i_k'} \prod_k w_k^{j_k'}$.
You can of course interpret this in terms of quadratic forms as in Dave Benson's answer. In this case $q: V \rightarrow \mathbb{F}_2$ with $q(v_i) = q(w_j) = 0$, with polarization $\beta$ defined by $\beta(v_i,v_j) = \beta(w_i,w_j) = 0$ and $\beta(v_i,w_j) = \delta_{i,j}$.
With some abuse of notation, the quadratic form on $V$ is defined by the formula $x^2 = z^{q(x)}$ for all $x \in P$.
-
$\begingroup$ thank you, I guess you mean $f(v,v')=z^{\sum_{k=1}^n{i_k j'_k}}$, right? $\endgroup$– AntoineCommented Jul 17 at 5:50
-
$\begingroup$ With what I wrote $vv' = f(v,v') \prod_k v_k^{i_k+i_k'} \prod_k w_k^{j_k+j_k'}$ in $P$, isn't that correct? $\endgroup$ Commented Jul 17 at 14:55
-
$\begingroup$ Well, the thing is that your formula for $f(v,v')$ implies that $f(v_i,w_j)=0$ for all $i,j$. $\endgroup$– AntoineCommented Jul 18 at 4:31
-
$\begingroup$ @Antoine I think he must mean $f(w_i,v_j)=\delta_{i,j}$. $\endgroup$ Commented Jul 18 at 6:49
-