2

For a group $G$ with operation $+$, I'm interested in the set $\mathscr H$ of its subsets that can be constructed using only the following two rules:

  • $\{g\}$ (so a singleton set) is in $\mathscr H$ for $g \in G$;
  • for any $H \in \mathscr H$ and any $g \in G$, the set "$H$ mod $g$" $= \{ h + n \times g \mid h \in H \land n \in \mathbb Z \}$ is also in $\mathscr H$.

As an example (actually the one I'm interested in), take $G = {\mathbb Z}^2$ with component-wise addition, then $H\text{ mod }(7,5)\text{ mod }(-1,1) = H\text{ mod }(-1,1)\text{ mod }(7,5)$. Here $\{g\}$ represents a musical note (like B#3); $\{g\}\text{ mod }(7,5)$ a note name in an arbitrary octave (like B#); $\{g\}\text{ mod }(-1,1)$ a piano or MIDI key (like B#3 = middle C); and $\{g\}\text{ mod }(-1,1)\text{ mod }(7,5)$ one of the 12 piano keys in an arbitrary octave (like C).

I am looking for some background on this structure. And specifically, I would like to find a way to represent elements of $\mathscr H$ in software, so that it is possible to efficiently compare them to see if two given elements are equal.

Lee Mosher
  • 135,265
  • I found it hard to pose this question in a clear way, and to find a clear title; feel free to ask for clarifications, and suggestions are very welcome. – MarnixKlooster ReinstateMonica May 23 '23 at 19:29
  • There is no need for the \; either side of your $\rm\LaTeX$ code. The spacing is just odd. – Shaun May 23 '23 at 19:43
  • @Shaun I find the added spacing reads better, and use it in many questions and answers. But I removed it since I don't want my formatting to detract from the actual question. :-) – MarnixKlooster ReinstateMonica May 23 '23 at 19:47
  • I edited in an attempt to clarify the logic of the second clause in the definition of $\mathscr H$. Feel free to revert the edit if you feel it is not appropriate. – Lee Mosher May 23 '23 at 20:01
  • @LeeMosher I don't see the improvement, but if it helps you / someone it's fine, leaving it as is for now. Thanks for the attention! – MarnixKlooster ReinstateMonica May 23 '23 at 20:03
  • 2
    In $\mathbb Z^2$ or any other finitely generated abelian group $G$, $\mathscr H$ is the set of all cosets of all subgroups. Standard methods of linear algebra (over $\mathbb Z$) can be used to give a normal form for any coset; comparing normal forms tells you when two cosets are the same. – Lee Mosher May 24 '23 at 01:07
  • I don't think much, if anything, can be said for nonabelian groups. – Lee Mosher May 24 '23 at 01:07
  • @LeeMosher Could you fill out that outline a bit more, in an answer perhaps? I'm an just an amateur and beginner in most topics you mention... The subgroups of ${\mathbb Z}^2$ I can read about in https://math.stackexchange.com/q/193863/11994, cosets I'll try to learn from https://en.wikipedia.org/wiki/Coset. But what would that "normal form for [a] coset" look like? – MarnixKlooster ReinstateMonica May 24 '23 at 05:40

1 Answers1

2

Building up on Lee Mosher's comments...

Theorem. Let $G$ be an abelian group, written additively. Then the set $\mathscr{H}$ consists precisely of all cosets of finitely generated subgroups of $G$.

Proof. Note that if $\mathscr{A}$ and $\mathscr{B}$ are two collections of subsets of $G$ that contain all singletons such that if they include $H$ then they include $H\bmod g$ for all $g\in G$, then so does $\mathscr{A}\cap\mathscr{B}$. Therefore, the smallest such collection will be the intersection of all collections satisfying these properties.

First, we show that $\mathscr{H}$ has as elements all cyclic subgroups of $G$. We have $\{0\}\in\mathscr{H}$, by assumption. If $g\in G$, then $\mathscr{H}$ contains $$\{0\}\bmod g = \{ ng\mid n\in\mathbb{Z}\} = \langle g\rangle.$$ So $\mathscr{H}$ contains all cyclic subgroups.

Next we show inductively that $\mathscr{H}$ contains all finitely generated subgroups. Assume that $\mathscr{H}$ contains all subgroups generated by at most $k\geq 1$ elements, and let $K=\langle g_1,\ldots,g_k,g_{k+1}\rangle$ be a subgroup generated by $k+1$ elements. Then $H=\langle g_1,\ldots,g_k\rangle\in\mathscr{H}$, and $K=H+\langle g_{k+1}\rangle$; since $H\bmod g\in\mathscr{H}$, and $$H\bmod g_{k+1} = \{h+ng\mid n\in\mathbb{Z},h\in H\} = H+\langle g_{k+1}\rangle = K,$$ we conclude that $K\in\mathscr{H}$. Thus, $\mathscr{H}$ contains all finitely generated subgroups.

Next we show that $\mathscr{H}$ contains any coset of a finitely generated subgroup of $G$. Let $K=\langle g_1,\ldots,g_n\rangle$ be a finitely generated subgroup of $G$, and let $x\in G$. Then $\mathscr{H}$ contains $\{h\}\bmod g_1 = h+\langle g_1\rangle$. Therefore, it also contains $$\begin{align*} (h+\langle g_1\rangle) \bmod g_2 &= (h+\langle g_1\rangle)+\langle g_2\rangle\\ &= \{h + ng_1 + mg_2\mid n,m\in\mathbb{Z}\} = h+ \langle g_1,g_2\rangle\\ (h+\langle g_1,g_2\rangle) \bmod g_3 &= h+\langle g_1,g_2,g_3\rangle\\ &\vdots\\ (h+\langle g_1,\ldots,g_{n-1}\rangle\bmod g_n &= h+\langle g_1,\ldots,g_n\rangle = h+K. \end{align*}$$ Thus, $\mathscr{H}$ contains all cosets of finitely generated subgroups of $G$.

Finally, we show that the collection $\mathscr{S}$ of all cosets of finitely generated subgroups of $G$ is a collection that includes all singletons and is closed under the "mod" operation you have defined. Indeed, the singleton $\{g\}$ is the coset of $g$ relative to the subgroup $\{e\}$, so $\mathscr{S}$ contains all singletons. Now let $h+K$ be a coset of a finitely generated subgroup $K$, and let $g\in G$. Then $$(h+K)\bmod g = \{ h+k+ng\mid k\in K,n\in\mathbb{Z}\} = h+\langle K,g\rangle,$$ and since $K$ is finitely generated so is $\langle K,g\rangle$; therefore, $(h+K)\bmod g\in\mathscr{S}$. Thus, $\mathscr{S}\subseteq \mathscr{H}\subseteq\mathscr{S}$, proving the desired equality. $\Box$

The situation is more complicated for nonabelian groups $G$. It is clear that $\mathscr{H}$ will contain all left cosets of cyclic subgroups of $G$, since (switching to multiplicative notation) it contains $\{x\}$ for all $x\in G$, and $\{x\}\bmod g = \{xg^n\mid n\in\mathbb{Z}\}=x\langle g\rangle$. But it will contains sets that are not cosets in general, as well. For example, consider the Heisenberg group of order $27$, $$\langle x,y,z\mid x^3=y^3=z^3=1, yx=xyz, xz=zx, yz=zy\rangle$$ which can be realized as the multiplicative group of all unitriangular matrices over $\mathbb{Z}/3\mathbb{Z}$: $$\left(\begin{array}{ccc} 1 & a & c\\ 0 & 1 & b\\ 0 & 0 & 1 \end{array}\right).$$ Then $\mathscr{H}$ contains the singletons (27 of them); also the subgroup $\langle x\rangle = \{1,x,x^2\}$. Then it also contains $$\langle x\rangle\bmod y = \{1, y, y^2, x, xy, xy^2, x^2, x^2y, x^2y^2\} = \langle y\rangle \cup x\langle y \rangle \cup x^2\langle y\rangle,$$ the union of three cosets of $\langle y\rangle$. However, this is not a coset of a subgroup of $G$, since it is not a subgroup: it contains $y$ and $x$, but does not contain $yx = xyz$; and if it were a coset, it would have to be a subgroup since it contains $1$. It will then get more complicated to describe further iterations, so I agree that there might not be much that can be said for arbitrary nonabelian $G$.

Arturo Magidin
  • 417,286