2
$\begingroup$

Observe $\mathbb{Z}_q^n = \mathbb{Z}_q \times \cdots \times\mathbb{Z}_q$ as a module over $\mathbb{Z}_q\equiv\mathbb{Z}/q\mathbb{Z}$, for general $q$. I am interested in the following questions:

How many submodules of size $q^k$, $k\leq n$, does it have? How many of them are free? Can something be said about the ratio of these two numbers when $n\to\infty$ and $k=\lambda n$, $\lambda\in(0,1)$?

Can someone give me a reference where such or similar problems have been studied? Or else provide pointers how to tackle them?

$\endgroup$
5
  • $\begingroup$ What exactly did you try to do so far? It certainly is a problem where one can do a lot before getting stuck as such. Did you solve the (standard) case of $q$ a prime $q$? Of a $q$ a prime power? $\endgroup$ Commented Nov 24, 2015 at 10:47
  • $\begingroup$ Yes, the case of prime $q$ is very easy. This looked to me like a problem that has been studied before, so I was hoping for a reference... Should the prime power case also be obvious? $\endgroup$
    – aleph
    Commented Nov 24, 2015 at 11:43
  • $\begingroup$ Towards that end, I added the "reference-request" tag. Also, if you're specifically interested in $q$ a prime power (as the letter suggests), you should probably indicate that. $\endgroup$ Commented Nov 24, 2015 at 15:54
  • $\begingroup$ Actually, I had in mind the general case, not necessarily prime powers. But of course, if something can be said in certain special cases, that would also be interesting (apart from the case of prime $q$). $\endgroup$
    – aleph
    Commented Nov 24, 2015 at 16:30
  • $\begingroup$ You might find the accepted question to this answer relevant (not as a reference, but as a way to easily figure out the answer by yourself - it really does not seem to be a hard question at all, hence my first comment): mathoverflow.net/questions/81937/… . [I think that the prime power case is just a bit better because the answer may be less messy.] $\endgroup$ Commented Nov 24, 2015 at 18:22

3 Answers 3

4
$\begingroup$

First, if $q=rs$ with $(r,s)=1$ then $\mathbb{Z}_q^n=\mathbb{Z}_r^n\times\mathbb{Z}_s^n$, and every submodule of $\mathbb{Z}_q^n$ splits uniquely as a direct sum of a submodule of $\mathbb{Z}_r^n$ and a submodule of $\mathbb{Z}_s^n$. Using this, we reduce easily to the case where $q$ is a prime power, say $q=p^m$.

Now let $F$ denote the set of free submodules of rank $k$ in $\mathbb{Z}_q^n$, and let $F_0$ be the corresponding set for $\mathbb{Z}_p^n$. By a standard argument which you say you have seen, we have $|F_0|=\pi(n)/(\pi(k)\pi(n-k))$, where $\pi(k)=(p-1)(p^2-1)\dotsb(p^k-1)$. The reduction map $\rho\colon F\to F_0$ is easily seen to be surjective, so you just need to understand $|\rho^{-1}\{A_0\}|$ for $A_0\in F_0$. We can choose $A\in\rho^{-1}\{A_0\}$, and then choose a complement $B$ such that $\mathbb{Z}_q^n=A\times B$. Now $\rho^{-1}\{A_0\}$ is the set of submodules $C\leq A\times B$ that are free of rank $r$ and agree with $A$ mod $p$. For any such $C$, we have projections $A\xleftarrow{f}C\xrightarrow{g}B$, and we see that $f(C)+pA=A$ and $g(C)\leq pB$. As $f(C)+pA=A$ we see that $f$ is surjective, but $|C|=|A|=q^k$ so $f$ is an isomorphism. It follows that $C=\{(a,h(a)):a\in A\}$, where $h=gf^{-1}\in\text{Hom}(A,pB)$. This construction gives a bijection from $\rho^{-1}\{A_0\}$ to $\text{Hom}(A,pB)$, so $|\rho^{-1}\{A_0\}|=p^{(m-1)k(n-k)}$. This is independent of $A_0$, so $$ |F| = p^{(m-1)k(n-k)} \pi(n)/(\pi(k)\pi(n-k)). $$

One can say interesting things about non-free subgroups as well, but I do not have time to write that now.

$\endgroup$
0
6
$\begingroup$

Aren't you just asking for the number of subgroups of $\mathbb{Z}_q^n$ of size $q^k$? There isn't going to be a simple answer. The classification and enumeration of subgroups of finite abelian groups goes back to Garrett Birkhoff. See http://plms.oxfordjournals.org/content/s2-38/1/385.citation. Some more recent work is mentioned at http://www.northeastern.edu/iloseu/bhmn/Ringel04.html. One can give an answer in terms of symmetric functions. If $G$ is a finite abelian group of order $p_1^{a_1}p_2^{a_2}\cdots$ (where the $p_i$'s are distinct primes) then the face lattice $L(G)$ is the product of the lattices for each Sylow subgroup. This reduces the problem to abelian $p$-groups. Let $P_\lambda(x;t)$ be a Hall-Littlewood symmetric function. Let $$ P_\mu(x;t)P_\nu(x;t) =\sum_\lambda f^\lambda_{\mu\nu}(t) P_\lambda(x;t). $$ Let $n(\lambda) =\sum (i-1)\lambda_i$. Then $p^{n(\lambda)}f^\lambda_{\mu\nu}(1/p)$ is the number of subgroups $H$ of type $\mu$ of an abelian $p$-group of type $\lambda$ such that $G/H$ has type $\nu$ (or something close to this). In particular, one can deduce that if $h_i$ is a complete symmetric function and $$ h_mh_n = \sum_\lambda c_{m,n}^\lambda(t)P_\lambda(x;t), $$ then $p^{n(\lambda)}c_{m,n}^\lambda(1/p)$ is the number of subgroups of order $p^m$ in an abelian $p$-group of type $\lambda\vdash m+n$. All this is explained in Chapter II of Macdonald, Symmetric functions and Hall Polynomials.

$\endgroup$
0
1
$\begingroup$

Here's a somewhat different approach that doesn't directly yield a formula for what you want, but can be used to glean information. Fix $n$ and let $$ \nu(m) = \text{\# of subgroups of $(\mathbb Q/\mathbb Z)^n$ of order $m$.} $$ Then the Dirichlet series associated to the sequence $\nu(m)$ is given by the beautiful formula $$ \sum_{m=1}^\infty \frac{\nu(m)}{m^s} = \zeta(s)\zeta(s-1)\cdots\zeta(s-n+1). $$ Here $\zeta(s)$ is the Riemann zeta function. From this formula, properties of $\zeta(s)$, and a standard Tauberian theorem, one deduces, for example, that for all $k\ge-(n-1)$, we have $$ \sum_{m\le X} m^k\nu(m) \sim \frac{X^{n+k}}{n+k} \quad\text{as $X\to\infty$.} $$ I believe that this is all standard. It follows from the formulas for Dirichelet series with Hecke operator coefficients [1, Theorem 3.21] and some algebra. (I thank Sol Friedberg and Dan Bump for giving me this reference.)

[1] G. Shimura, Introduction to the Arithmetic Theory of Automorphic Forms, Princeton University Press, 1971.

$\endgroup$
0

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