8
$\begingroup$

Given the group $\mathrm{SL}(2,\mathbb{Q}_p)$ one can describe precisely the number of vertices in the Bruhat-Tits building of distance $k$ from a fixed vertex $v$, where the distance between two vertices $v$ and $v'$ is the minimal number of edges in a path connecting the two. Denoting this quantity by $N_k$, we see in particular that for $\mathrm{SL}(2,\mathbb{Q}_p)$ $$N_k=(q+1)q^{k-1},$$ where $q$ is the size of the residue field.

I am curious if one can come up with an analogous closed formula for $G=\mathrm{SL}(n,\mathbb{Q}_p)$. In the case that $k=1$, this is not difficult and well known by using the lattice interpretation of the building for $\mathrm{SL}$ and counting the number of non-trivial $k$-dimensional subspaces of $\mathbb{F}_p^n$ (one can also use spherical buildings for this and more general groups). For instance, in the case of $n=3$, we have $$N_1=2(q^2+q+1).$$ Continuing the example where $n=3$, to compute $N_2$ is equivalent to finding the number of lattices $\Lambda$ such that if $$\Lambda_0=\mathbb{Z}_p\oplus\mathbb{Z}_p\oplus\mathbb{Z}_p,$$ then $$p^2\Lambda_0\subsetneq \Lambda\subsetneq\Lambda_0,$$ with the additional requirements that $\Lambda$ does not contain $p\Lambda_0$ and that $\Lambda$ is not contained in $p\Lambda_0$ (this ensures that $\Lambda$ is truly distance $2$). However, I have not been able to make the combinatorics of this count work on paper. In general when $n=3$, to compute $N_k$ one would need to count the analogous $\Lambda$s in a chain with $p^2\Lambda_0$ replaced by $p^k\Lambda_0$ subject to the condition that $\Lambda$ does not contain any $p^i\Lambda_0$ and is not contained in any $p^j\Lambda_0$ where $1\le i,j\le k-1$ (to ensure genuine distance $k$).

I am looking to see if a closed formula for $N_k$ is known in the literature or is easy to compute and I am missing the computation. I would be extremely pleased with just the case of $G=\mathrm{SL}(3,\mathbb{Q}_p)$, even if just the case of $n=3$ and $p=2$.

$\endgroup$
  • $\begingroup$ You will have $n-1$ types of vertices. The $i$th type will correspond to the number of submodules of $(\mathbb{Z}/p^k\mathbb{Z})^n$ isomorphic to $(\mathbb{Z}/p^k\mathbb{Z})^i$. It is not hard to make an explicit computation (which I currently don't have the time to do), but you can surely find it also in Macdonald "Symmetric functions and Hall polynomials". $\endgroup$ – Uri Bader Mar 1 '17 at 18:39
  • $\begingroup$ @UriBader Thank you for the response! If I understand it correctly, I think this gives you some but not all of the vertices, since vertices of the same type as the original vertex may show up. For instance if $k=2$, $n=3$ then the class of the lattice $\mathbb{Z}_p\oplus p\mathbb{Z}_p\oplus p^2\mathbb{Z}_p$ is such a vertex. In the case of $k=2$ these are fairly easy to count (by counting vertices of distance 2 lying in a chamber which intersects another chamber containing the original vertex on a codimension 1 simplex), however I don't quite see the count for these types in general. $\endgroup$ – Guest Mar 2 '17 at 4:25
  • $\begingroup$ Sorry, I guess I was not focused when commenting. Fix $n$ integers $k= \lambda_1\geq \lambda_2 \ldots \lambda_n=0$ and let $\lambda=(\lambda_1,\ldots,\lambda_n)$. A module of type $\lambda$ is one which is isomorphic to $\oplus\mathbb{Z}/p^{\lambda_i}\mathbb{Z}$. You want to count all the submodules of type $\lambda$ in a module of type $(k,k,\ldots,k)$. Now you want to sum up over all such $\lambda$s. In claim 4.5(1) in arxiv.org/pdf/math/0404408.pdf you will find such a computation (with notation introduced before prop 3.2). Probably you can find it already in Macdonald book. $\endgroup$ – Uri Bader Mar 2 '17 at 7:32
  • $\begingroup$ @UriBader Thanks, this is exactly what I was looking! I was able to turn it into a closed formula for $n=3$ (surely for larger $n$ it can't be much harder, just more tedious). I'm happy to accept your comment as an answer if you want to turn it into one. $\endgroup$ – Guest Mar 2 '17 at 18:37
  • $\begingroup$ In my post below we have a summation over various partitions. I didn't make an effort to give a simpler expression. If you happen to get one, please share it with me. $\endgroup$ – Uri Bader Mar 4 '17 at 13:56
1
$\begingroup$

We consider the Bruhat-Tits building of $\text{SL}_n(\mathbb{Q}_p)$. As usual, we identify the vertices with lattices, that is subgroups of the of $\mathbb{Q}^n$ commensurated with the standard lattice $\Lambda_0=\mathbb{Z}_p^n$, defined up to homothety, namely multiplication by $p^k$ for some $k\in \mathbb{Z}$. Getting rid of the homothety relation, we can represent each lattice uniquely as a subgroup $\Lambda$ of $\Lambda_0$ which is not contained in $p\Lambda_0$. Such a $\Lambda$ will contain $p^k\Lambda_0$ for some natural $k$, and the minimal such $k$ is said to be the distance from $\Lambda$ to $\Lambda_0$.

From now on by vertex I mean a representation of homothety class of lattices given as above. Thus the set all vertices of distance $k$ from $\Lambda_0$ is given by $$ \{\Lambda\mid p^k\Lambda_0<\Lambda<\Lambda_0,~\Lambda\notin p\Lambda_0 \mbox{ and the rank of } p^k\Lambda_0/\Lambda\mbox{ is less than }n\}.$$ Dividing by $p^k\Lambda_0$, the set above is in bijection with the set of all $p$-abelian subgroups $$ \{A \mid A < (\mathbb{Z}/p^k\mathbb{Z})^n,\mbox{ the ranks of } A\mbox{ and } (\mathbb{Z}/p^k\mathbb{Z})^n/A \mbox{ is less than }n\}.$$

Recall that every for every $p$-abelian group $A$ there exists a unique vector of natural numbers $\lambda=(\lambda_1,\lambda_2,\ldots,\lambda_m)$ with $\lambda_1\geq \lambda_2\geq \ldots\geq \lambda_m> 0$ such that $$ A\simeq \bigoplus_{i=1}^m \mathbb{Z}/p^{\lambda_i}\mathbb{Z}.$$ In that case we say that $A$ is of (isomorphism) type $\lambda$. If for all $i$ $\lambda_i=k$ we say $\lambda=k^m$.

Given types $\nu$ and $\lambda$ we denote by ${\nu\choose \lambda}_p$ the number of submodules of type $\lambda$ of a given module of type $\nu$. We conclude that the number of vertices of distance $k$ is given by $\sum {k^n\choose \lambda}_p$ where the sum runs over all $\lambda=(\lambda_1,\lambda_2,\ldots,\lambda_m)$ with $\lambda_1=k$ and $m<n$.

We are left with a concrete computation. Fixing $\nu$ and $\lambda$, ${\nu\choose \lambda}_p$ is a polynomial in the variable $p$ which is computed in claim 4.5(1) in https://arxiv.org/pdf/math/0404408.pdf: $$ {\nu\choose \lambda}_p=p^{\langle \nu'-\lambda',\lambda'\rangle}\prod_i\begin{bmatrix} \nu'_i-\lambda'_{i+1} \\ \nu'_i-\lambda'_i\end{bmatrix}_p$$ where $\nu',\lambda'$ are the transpose diagrams of $\nu,\lambda$, $\langle \nu,\lambda\rangle=\sum \nu_i\lambda_i$ and we use the notations: $$ [i]_1=1-p^{-i},~[m]_p!=\prod_{i=1}^m [i]_p,~\begin{bmatrix} n \\ m\end{bmatrix}_p =\frac{[n]_p}{[m]_p[n-m]_p}.$$

$\endgroup$

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.