-1

Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\alpha_1,\cdots,\alpha_m$ be a finite sequence of elements of $G$. Suppose $\alpha_i = \sum_j a_{ij} \omega_j$ for $i = 1,\cdots, m$. Let $H$ be the subgroup of $G$ generated by $\alpha_1,\cdots,\alpha_m$. Can we compute $|G/H|$ by the data $a_{ij}, 1 \le i \le m, 1 \le n\le n$? If yes, how?

Makoto Kato
  • 42,602
  • The function which sends $(a_{ij})$ to $|G/H|$ vanishes if two of the $\alpha_i$ coincide, and $|G/G| = 1$, so perhaps $|G/H|$ is related to the determinant of $(a_{ij})$. – Adam Saltz Dec 08 '13 at 04:26
  • @AdamSaltz The determinant does not make sense unless $n = m$. Regards, – Makoto Kato Dec 08 '13 at 04:30

2 Answers2

3

Yes:

1) if $\;m<n\;$ then the index is "obviously" (why?) infinite ;

2) If $\;m=n\;$ then form the square $\;n\times n\;$ matrix $\;A=(a_{ij})\;$ , then

$$|G/H|=[G:H]=\left|\det A\right|$$

The proof of (b) is a rather beautiful subject within "finitely generated abelian groups", which include the canonical Smith form for (integer) matrices and stuff. You can google this to read more about it.

DonAntonio
  • 211,718
  • 17
  • 136
  • 287
  • I have two questions. One: How do you prove that $|G/H|$ is infinite when the determinant is zero? Two: What about the case $m \gt n$? – Makoto Kato Dec 08 '13 at 07:18
  • 1
    @Makoto Kato This is standard theory of finitely generated abelian groups, and you would do better to learn the basics of the topic before asking specific questions. The general algorithm works for all $m$ and $n$ and is known as the Smith Normal Form algorithm. – Derek Holt Dec 08 '13 at 08:12
  • Derek's comment answers you, @MakotoKato, but if you want an intuitive insight: if $;\det A=0;$ then at least one of the rows/columns in $;A;$ depend $;\Bbb Z-$linearly on the others ones, and this means the rank of $;H;$ has to be less than $;n;$ ... – DonAntonio Dec 08 '13 at 12:17
  • @DonAntonio I still have two questions.
    1. Could you explain why the case $m \lt n$ obviously implies $|G/H|$ is infinite?
    2. Could you explain why the Smith normal form also solves the problem in the case $m \gt n$?
    – Makoto Kato Dec 08 '13 at 19:19
  • The case $;m<n;$ is just the same as with $;\det A=0;$ : the rank of the subgroup is less than $;n;$ , and the case $;m>n;$ is impossible: the rank of a subgroup of a group with rank $;n;$ is at most $;n;$ . What Derek probably meant is that working with the Smith Normal Form solves the problem for any $;m\le n;$, without splitting in two cases. – DonAntonio Dec 08 '13 at 19:57
  • @DonAntonio [The case m < n is just the same as with det A = 0 : the rank of the subgroup is less than n] Could you explain why the fact that the rank of $H$ is less than $n$ implies $|G/H|$ is infinite?

    [the case m > n is impossible: the rank of a subgroup of a group with rank n is at most n.] I don't understand why $m \gt n$ is impossible. $\alpha_1, \cdots,\alpha_m$ need not be linearly independent over $\mathbb{Z}$.

    – Makoto Kato Dec 09 '13 at 02:16
  • Just try to do some basic examples to get a grip of the general case, @MakotoKato : for example, if $;G=\Bbb Z\times \Bbb Z;$ and $;H={0}\times \Bbb Z\le G;$, then the order of the element $;(1,0)+H\in G/H;$ is infinite...$;m>n;$ is impossible if we assume the $;\alpha_i$'s are $;\Bbb Z$-independente, otherwise we're back in the above case. – DonAntonio Dec 09 '13 at 04:52
  • @DonAntonio It is easy to see $G/H$ is infinite if $H$ has a basis of the form $a_1\omega_1,\cdots, a_m\omega_m$, where $m \lt n$. But $H$ does not have such a basis in general. [m>n is impossible if we assume the αi's are Z-independente, otherwise we're back in the above case.] What do you mean by "we're back in the above case"? We need to compute explicitly $|G/H|$ using the data $a_{ij}$. – Makoto Kato Dec 09 '13 at 05:38
  • The elementary divisors theorem gives you the way to do that, @MakotoKato, and again: I'm assuming the elements $;\alpha_i;$ are free abelian generators of $;H;$, otherwise the answer to your question is, apparently, no . – DonAntonio Dec 09 '13 at 13:34
  • @DonAntonio Why do you think it's no? – Makoto Kato Dec 09 '13 at 19:23
  • Because we can add any number of bogus $;\alpha_i;$ and this won't change $;H;$ , so I think the "wisest" think to do here is first to find out what relations are unnecessary and then take free generators for $;H;$. – DonAntonio Dec 09 '13 at 19:30
  • @DonAntonio [so I think the "wisest" think to do here is first to find out what relations are unnecessary and then take free generators for H.] My question asks how we do it. And you think it's impossible? – Makoto Kato Dec 09 '13 at 20:15
  • No, your question didn't ask how to do it, and no: I don't think it is impossible to do it. And what did you add as an answer? It looks like a rather nice piece of work, which prompts the question: why did you ask what you asked if you knew, or could find, all that, @MakotoKato ? – DonAntonio Dec 09 '13 at 20:18
  • @DonAntonio By the way, please ping me by writing my name at the top your comment when you reply to me. Otherwise I may not notice it. – Makoto Kato Dec 09 '13 at 20:22
  • @DonAntonio It's almost the same thing. My question asks how we compute $|G/H|$. If we can construct a free basis by the given generators, we can compute it at once. – Makoto Kato Dec 09 '13 at 20:28
  • @MakotoKato Nop, your question is NOT how to compute $;|G/H|;$ but whether you can do it with the elements $;\alpha_i;$ written there. This already has been addressed, now: if you knew all what you wrote in your answer, why did you ask the question in the first place?! – DonAntonio Dec 09 '13 at 21:13
  • @DonAntonio [Nop, your question is NOT how to compute |G/H|] My question asks [If yes, how?].

    [if you knew all what you wrote in your answer, why did you ask the question in the first place?!]There are several different solutions to a problem. I would like to know as many different such solutions as possible. Another reason is that I would like to use the result(s), if possible, to answer other questions in this site. Yet another reason is that I would like to know errors in my proof if any.

    – Makoto Kato Dec 09 '13 at 21:46
-1

Let $x_1,\cdots, x_m$ be a sequence of elements of an abelian group $G$. We denote by $[x_1,\cdots,x_m]$ the subgroup of $G$ generated by $x_1,\cdots, x_m$.

Lemma 1 Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\alpha_1,\cdots,\alpha_m$ be a finite sequence of elements of $G$. Then there exists an algorithm to find $\beta_1,\cdots,\beta_n \in G$ such that $[\alpha_1,\cdots,\alpha_m] = [\beta_1,\cdots,\beta_n]$ and $\beta_i \in [\omega_1,\cdots,\omega_i]$ for all $i = 1,\cdots,n$. Moreover we may choose each $\beta_i$ such that its coefficient with respect to $\omega_i$ is non-negative.

Proof: We use induction on the rank $n$ of $G$. Let $\alpha_i = \sum_j a_{ij}\omega_j$ for $i = 1,\cdots,m$. Suppose $n = 1$. Then $\alpha_1 = a_{11}\omega_1,\cdots, \alpha_m = a_{m1}\omega_1$. If all of $a_{11},\cdots,a_{m1}$ are zero, let $\beta_1 = 0$ and we are done. Otherwise let $d =$ gcd$(a_{11},\cdots,a_{m1})$. Let $\beta_1 = d\omega_1$ and we are done.

Suppose $n \gt 1$. If all of $a_{1n},\cdots,a_{mn}$ are zero, let $\beta_n = 0$. Since $[\alpha_1, \cdots\alpha_m] \subset [\omega_1,\cdots,\omega_{n-1}]$, by the induction hypothesis, there exists an algorithm to find $\beta_1,\cdots,\beta_{n-1} \in [\omega_1,\cdots,\omega_{n-1}]$ such that $[\alpha_1,\cdots,\alpha_m] = [\beta_1,\cdots,\beta_{n-1}]$ and $\beta_i \in [\omega_1,\cdots,\omega_i]$ for all $i = 1,\cdots,{n-1}$. Hence $\beta_1,\cdots,\beta_n$ satisfies the condition of the lemma. Suppose not all of $a_{1n},\cdots,a_{mn}$ are zero. Let $d =$ gcd$(a_{1n},\cdots,a_{mn})$. Let $a_{in} = b_id$ for $i = 1,\cdots,m$. There exist integers $c_1,\cdots,c_m$ such that $d = c_1a_{1n} + \cdots + c_ma_{mn}$. Let $\beta_n = c_1\alpha_1 + \cdots + c_m\alpha_m$. Let $\gamma_i = \alpha_i- b_i\beta_n$ for $i = 1,\cdots, m$. Then that $[\gamma_1,\cdots,\gamma_m, \beta_n] = [\alpha_1,\cdots,\alpha_m]$. Since $\gamma_i \in [\omega_1,\cdots,\omega_{n-1}]$ for $i = 1,\cdots, m$, by the induction hypothesis, there exists an algorithm to find $\beta_1,\cdots,\beta_{n-1} \in [\omega_1,\cdots,\omega_{n-1}]$ such that $[\gamma_1,\cdots,\gamma_m] = [\beta_1,\cdots,\beta_{n-1}]$ and $\beta_i \in [\omega_1,\cdots,\omega_i]$ for all $i = 1,\cdots,n-1$. Hence $\beta_1,\cdots,\beta_n$ satisfies the condition of the lemma. QED

Lemma 2 Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\alpha_1,\cdots,\alpha_n \in G$ be such that

$\alpha_1 = a_{11}\omega_1$

$\alpha_2 = a_{21}\omega_1 + a_{22}\omega_2$

. . .

$\alpha_i = a_{i1}\omega_1 + \cdots + a_{ii}\omega_i$

. . .

$\alpha_n = a_{n1}\omega_1 + \cdots + a_{nn}\omega_n$

Suppose $a_{ii} \ne 0$ for all $i = 1,\cdots, n$. Let $H = [\alpha_1,\cdots,\alpha_n]$. Let $x = x_1\omega_1 + \cdots + x_i\omega_i \in H$ for some $i$ such that $1 \le i \le n$, where $x_1,\cdots,x_i$ are integers. Then $x_i$ is divisible by $a_{ii}$.

Proof: There exist integers $b_1,\cdots,b_n$ such that $x = b_1\alpha_1 + \cdots + b_n\alpha_n$. Then $n$-th coefficient of $x$ is $b_na_{nn}$ which must be $0$. Hence $b_n = 0$. Similarly we get $b_{i+1} = \cdots = b_n = 0$. Hence $x_i = b_ia_{ii}$. QED

Lemma 3 Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\alpha_1,\cdots,\alpha_n \in G$ be such that

$\alpha_1 = a_{11}\omega_1$

$\alpha_2 = a_{21}\omega_1 + a_{22}\omega_2$

. . .

$\alpha_i = a_{i1}\omega_1 + \cdots + a_{ii}\omega_i$

. . .

$\alpha_n = a_{n1}\omega_1 + \cdots + a_{nn}\omega_n$

Suppose $a_{ii} \gt 0$ for all $i = 1,\cdots, n$. Let $H = [\alpha_1,\cdots,\alpha_n]$. Then $|G/H| = a_{11}\cdots a_{nn}$.

Proof: It suffices to prove that $\{\sum_i r_i\omega_i\ |\ 0 \le r_i \lt a_{ii}, i = 1\cdots,n\}$ is a complete set of representatives for $G/H$.

Let $x = x_1\omega_1 + \cdots + x_n\omega_n \in G$. there exist integers $q_n, r_n$ such that $x_n = a_{nn}q_n + r_n, 0 \le r_n \lt a_{nn}$. Then $x - q_n\alpha_n = y_1\omega_1 +\cdots + y_{n-1}\omega_{n-1} + r_n\omega_n$, where $y_1,\cdots, y_{n-1}$ are integers. Applying this process to $x - q_n\alpha_n$, etc. we get $x - \sum_i q_i\alpha_i = \sum_i r_i\omega_i$, where $0 \le r_i \lt a_{ii}$ for all $i$. Hence $x \equiv \sum_i r_i\omega_i$ (mod $H$).

Let $\sum_i r_i\omega_i \equiv \sum_i s_i\omega_i$ (mod $H$), where $0 \le r_i, s_i \lt a_{ii}$ for all $i$. We need to prove that $r_i = s_i$ for all $i$. By Lemma 2, $r_n = s_n$. Hence $r_1\omega_1 + \cdots + r_{n-1}\omega_{n-1} \equiv s_1\omega_1 + \cdots + s_{n-1}\omega_{n-1}$ (mod $H$). Hence agian by Lemma 2, $r_{n-1} = s_{n-1}$. Repeating this process, we get $r_i = s_i$ for all $i$. QED

Lemma 4 Let $G$ be a finitely generated abelian group. Then $G\otimes_{\mathbb{Z}} \mathbb{Q}= 0$ if and only if $G$ is a finite group.

Proof: Suppose $G$ is a finite group. Let $n$ be the order of $G$. Consider an element $x \otimes r$ of $G\otimes_{\mathbb{Z}} \mathbb{Q}$. Then $x \otimes r = x \otimes n(r/n) = nx\otimes r/n = 0$. Hence $G\otimes_{\mathbb{Z}} \mathbb{Q}= 0$.

Conversely suppose $G\otimes_{\mathbb{Z}} \mathbb{Q}= 0$. Since $G$ is finitely generated, $G$ is a finite product of cyclic groups. Let $G = G_1 \oplus \cdots \oplus G_m$, where each $G_i$ is a cyclic group. Since $G\otimes_{\mathbb{Z}} \mathbb{Q}$ is isomorphic to $G_1\otimes_{\mathbb{Z}} \mathbb{Q} \oplus \cdots \oplus G_m\otimes_{\mathbb{Z}} \mathbb{Q}$, $G_i\otimes_{\mathbb{Z}} \mathbb{Q} = 0$ for all $i$. If $G_i$ is an infinite cyclic group, then $G_i\otimes_{\mathbb{Z}} \mathbb{Q}$ is isomorphic to $\mathbb{Q}$. This is a contradiction. Hence each $G_i$ is a finite cyclic group. Therefore $G$ is a finite group. QED

Lemma 5 Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\alpha_1,\cdots,\alpha_n \in G$ be such that $\alpha_i = \sum_i a_{ij}\omega_j$ for $i = 1,\cdots,n$. Let $H = [\alpha_1,\cdots,\alpha_n]$. Then the following assertions are equivalent.

  1. det $(a_{ij}) \ne 0$.

  2. $\alpha_1,\cdots,\alpha_n$ is a free basis of $H$.

  3. $|G/H|$ is finite.

Proof: $1. \Rightarrow 2.$ Let $A$ be the matrix $(a_{ij})$. Suppose $\alpha_1,\cdots,\alpha_n$ are linearly dependent over $\mathbb{Z}$. Then the row vectors of $A$ are linearly dependent over $\mathbb{Z}$. Hence they are linearly dependent over $\mathbb{Q}$. Hence det $(a_{ij}) = 0$.

$2. \Rightarrow 3.$ Consider the following exact sequence.

$0 \rightarrow H \rightarrow G \rightarrow G/H \rightarrow 0$.

This yields the following exact sequence.

$H\otimes_{\mathbb{Z}} \mathbb{Q} \rightarrow G \otimes_{\mathbb{Z}} \mathbb{Q}\rightarrow (G/H) \otimes_{\mathbb{Z}} \mathbb{Q} \rightarrow 0$.

Since the rank of $H$ is $n$, the map $H\otimes_{\mathbb{Z}} \mathbb{Q} \rightarrow G \otimes_{\mathbb{Z}} \mathbb{Q}$ is surjective. Hence $(G/H) \otimes_{\mathbb{Z}} \mathbb{Q} = 0$. Hence $G/H$ is a finite group by Lemma 4.

$3. \Rightarrow 1.$ Suppose det $(a_{ij}) = 0$. Then the map $H\otimes_{\mathbb{Z}} \mathbb{Q} \rightarrow G \otimes_{\mathbb{Z}}\mathbb{Q}$ in the above sequence is not surjective. Hence $(G/H) \otimes_{\mathbb{Z}} \mathbb{Q} \ne 0$. Hence $G/H$ is an infinite group by Lemma 4. QED

Combining Lemma 1, 2, 5, we get an answer to the question. The following proposition is also useful to compute $G/H$.

Proposition Let $G$ be a finitely generated free abelian group. Let $\omega_1,\cdots,\omega_n$ be its basis. Let $\beta_1,\cdots,\beta_n \in G$ be such that $\beta_i = \sum_i b_{ij}\omega_j$ for $i = 1,\cdots,n$. Let $H = [\beta_1,\cdots,\beta_n]$. If det $(b_{ij}) = 0$, then $|G/H|$ is infinite. If det $(b_{ij}) \ne 0$, then $|G/H| = |$det $(b_{ij})|$.

Proof: If det $(b_{ij}) = 0$, then, by Lemma 4, $|G/H|$ is infinite. Suppose det $(b_{ij}) \ne 0$. By Lemma 5, $\{\beta_1,\cdots,\beta_n\}$ is a free basis of $H$ and $G/H$ is a finite group.

Let $B = (b_{ij})$. Let $\beta$ be the column vector $(\beta_1,\cdots,\beta_n)^t$. Let $\omega$ be the column vector $(\omega_1,\cdots,\omega_n)^t$. Then $\beta = B\omega$.

By Lemma 1, there exist $\alpha_1,\cdots,\alpha_n \in H$ such that $\alpha_i \in [\omega_1,\cdots,\omega_i]$ for all $i$ and $H = [\alpha_1,\cdots,\alpha_n]$. Let $\alpha_i = \sum_j a_{ij}\omega_j$ for all $i$. We may suppose $a_{ii} \ge 0$ for all $i$. Since $G/H$ is a finite group, det $(a_{ij}) \ne 0$ by Lemma 5. Hence $a_{ii} \gt 0$ for all $i = 1,\cdots,n$. By Lemma 2, $|G/H| = a_{11}\cdots a_{nn}$. Let $A = (a_{ij})$. Let $\alpha$ be the column vector $(\alpha_1,\cdots,\alpha_n)^t$. Then $\alpha = A\omega$. Since $\{\alpha_1,\cdots,\alpha_n\}$ and $\{\beta_1,\cdots,\beta_n\}$ are free bases of $H$, there exists a matrx $P \in GL_n(\mathbb{Z})$ such that $\beta = P\alpha$. Then $P\alpha = B\omega$. Hence $PA\omega = B\omega$. Hence $PA = B$. Hence det $B =$ det $P$ det $A$. Since det $P = \pm 1$, |det $B| = |$det $A| = a_{11}\cdots a_{nn} = |G/H|$. QED

Makoto Kato
  • 42,602
  • 3
    Please use a personal website for this sort of thing, rather and asking and answering your own questions in a seemingly solipsistic manner. The reasons for this have been explained on several meta threads in the past. – Carl Mummert Dec 10 '13 at 02:55
  • @CarlMummert [The reasons for this have been explained on several meta threads in the past.] Could you show me the link to one of the threads? Regards, – Makoto Kato Dec 10 '13 at 03:49
  • 4
    You have participated in them yourself - in several cases, you opened them yourself. I have no desire to rehash the arguments there, but I may downvote when you misuse the site in this way. You do make useful posts in which you answer others' questions, and I hope you will continue to participate productively in that way, while discontinuing this sort of question and answer. – Carl Mummert Dec 10 '13 at 03:55
  • @CarlMummert As far as I remember you have never told me the reason why I should not answer my own questions. I guess you just don't like it. Anyway, I wonder why you don't leave me alone rather than minding your own business. – Makoto Kato Dec 10 '13 at 04:06
  • 2
    I was able to find several explanations by reading through answers that you opened on meta (these were by several people, not all by me, perhaps none by me). – Carl Mummert Dec 10 '13 at 04:08
  • @CarlMummert Please tell me one, for example. – Makoto Kato Dec 10 '13 at 04:12
  • The "only if" part of Lemma 4 can be proved without using the fundamental theorem of finitely generated abelian groups. See the answer to this question. – Makoto Kato Dec 10 '13 at 10:11