3
$\begingroup$

I apologize if this notion is well-known, but I couldn't find anything useful and I am not sure what key words to look for.

Suppose we have a lattice $\Lambda \subset \mathbb{Z}^n$, given by in the form

$$\displaystyle \Lambda = \left\{M \mathbf{u} : \mathbf{u} \in \mathbb{Z}^n \right \}$$

for some matrix $M$ with integer entries and non-zero determinant. By 'transpose' lattice I mean the corresponding lattice given by

$$\displaystyle \Lambda^T = \left\{M^T \mathbf{u} : \mathbf{u} \in \mathbb{Z}^n \right \}$$.

Is there a name for $\Lambda^T$? What properties can be deduced about $\Lambda^T$ given $\Lambda$?

For example, it is clear that $\det \Lambda = \det M = \det M^T = \det \Lambda^T$.

$\endgroup$
5
  • $\begingroup$ @SamHopkins I am thinking of $\mathbf{u}$ as column vectors, if that helps $\endgroup$ Commented Jun 6, 2020 at 0:01
  • $\begingroup$ I think @SamHopkins was pointing out that, in the notation $\{-:-\}$, the part after the colon should be a statement, not a vector. $\endgroup$ Commented Jun 6, 2020 at 0:07
  • $\begingroup$ @AndreasBlass ah yes I see what you mean now. I have edited the question $\endgroup$ Commented Jun 6, 2020 at 0:09
  • 9
    $\begingroup$ I'm pretty sure that the "transpose" lattice can't generally be determined by the "original" lattice alone; the lattice defined by $M$ is also defined by $MA$ for any $A \in SL_n(\mathbb{Z})$, but the lattice defined by $(MA)^T = A^TM^T$ is not necessarily the lattice defined by $M^T$. As such, most of the properties are probably more about $M$ than the lattice itself. Equivalently, lattices correspond to $SL_n(\mathbb{Z})$-orbits of the subset of $\mathbb{M}_n(\mathbb{Z})$ with nonzero determinants, but transposing doesn't respect these orbits. $\endgroup$
    – user44191
    Commented Jun 6, 2020 at 0:13
  • 1
    $\begingroup$ As the comment by @user44191 shows, the answer very much depends on what the data exactly are. Do you insist that $\mathbb{Z}^n$ is the standard lattice, with a basis? Is $u$ really part of the data, and if so, is it a matrix, or an injective map of free $\mathbb{Z}$-modules of rank $n$, or an injective endomorphism of one free $\mathbb{Z}$-module of rank $n$? Do you insist that $\Lambda^T$ should live in the same $\mathbb{Z}^n$ as $\Lambda$? $\endgroup$ Commented Jun 6, 2020 at 9:43

2 Answers 2

3
$\begingroup$

Let $V = \mathbb{Z}^n/(M \mathbb{Z}^n), V' = \mathbb{Z}^n/(M^T \mathbb{Z}^n)$. I claim that $V \simeq V'$ as abelian groups.

By the classification of finite abelian groups, $V \simeq \oplus \mathbb{Z}/d_i \mathbb{Z}$ for some $d_1 | d_2 | \dots | d_n$, and with this condition, the set $\{d_i\}$ is unique. Let $D$ be the diagonal matrix with $\{d_i\}$ on the diagonals; then it's not hard (but it is slightly tedious) to show that $M = ADB$ for some $A, B \in SL_n(\mathbb{Z})$ (in fact, this is one proof of the above classification). Then $M^T = B^TDA^T$, so by reversing the above reasoning, $V' \simeq \oplus \mathbb{Z}/d_i \mathbb{Z} \simeq V$.

In fact, we can use the above reasoning for the converse: if $\Lambda, \Lambda'$ are sublattices of $\mathbb{Z}^n$ such that $\mathbb{Z}^n/\Lambda \simeq \mathbb{Z}^n/\Lambda'$, then there are some $M_0, M'_0$ such that $\Lambda = M_0 \mathbb{Z}^n, \Lambda' = M'_0 \mathbb{Z}^n$. If $D$ is the common Smith Normal Form, then $M_0 = ADB, M'_0 = A'DB'$; let $M = ADA'^T$. Then $\Lambda = M_0 \mathbb{Z}^n = ADB\mathbb{Z}^n = AD\mathbb{Z}^n = ADA'^T\mathbb{Z}^n = M\mathbb{Z}^n$, and similarly $\Lambda' = M^T\mathbb{Z}^n$. So two lattices come from transpose matrices iff their quotients are isomorphic.

As Sam Hopkins said in a comment, the key word here is Smith Normal Form.

$\endgroup$
3
  • 1
    $\begingroup$ Sure, this is just the Smith Normal Form. $\endgroup$ Commented Jun 6, 2020 at 3:43
  • $\begingroup$ @SamHopkins I couldn't remember the name, but yeah, that was the idea; I've edited the post to include the key word. $\endgroup$
    – user44191
    Commented Jun 6, 2020 at 3:58
  • $\begingroup$ In fact, we have canonically $V'\cong\mathrm{Ext}^1_\mathbb{Z}(V,\mathbb{Z})\cong\mathrm{Hom}_\mathbb{Z}(V,\mathbb{Q}/\mathbb{Z})$; the latter is non-canonically isomorphic to $V$ by the classification. $\endgroup$ Commented Jun 7, 2020 at 10:36
1
$\begingroup$

To follow up on user44191's comment, here is a concrete example: consider the matrices $$ M = \begin{pmatrix} 1 & 0 \\ 0 & 2 \end{pmatrix}, \qquad M' = \begin{pmatrix} 1 & 1 \\ 0 & 2 \end{pmatrix}.$$ The columns of $M$ and $M'$ span the same lattice $\Lambda$. However, for the transpose matrices $$ M^T = M,\qquad (M')^T = \begin{pmatrix} 1 & 0 \\ 1 & 2 \end{pmatrix},$$ the columns do not span the same lattice.

This shows that the definition of $\Lambda^T$ is not well-defined, as a subset of $\mathbb Z^n$. It depends on the choice of basis $M$ for $\Lambda$.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .