14
$\begingroup$

Let $\mathbb T=\mathbb R/\mathbb Z$ be the circle group and $\mathbb T^\omega$ be the infinite-dimensional torus, considered as an abelian compact topological group.

Problem 1. Is it true that for any finite set $F\subset\mathbb T^\omega$ and any neighborhood $U\subseteq \mathbb T^\omega$ of zero there exists an automorphism $\alpha$ of $\mathbb T^\omega$ such that $\alpha(F)\subset U$?

This problem can be reformulated in the language of the special linear groups $SL(n,\mathbb Z)$.

Problem 2. Is it true that for any $n\in\mathbb N$, neighborhood of zero $U$ in $\mathbb R^n$ and vectors $x_1,\dots,x_n$ in $\mathbb R^\omega$ there exists $m>n$ and a matrix $A\in SL(m,\mathbb Z)$ such that $\mathrm{pr}_n\circ A\circ \mathrm{pr}_m(x_i)\in U$ for all $i\in\{1,\dots,n\}$?

Here $\mathrm{pr_k}:\mathbb R^\omega\to\mathbb R^k$, $\mathrm{pr}_k:x\mapsto x{\restriction}k$, is the projection onto the first $k$ coordinates.

Remark 1. For any field $\mathbb F$ and vectors $x_1,\dots,x_n\in\mathbb F^{2n}$ there exists a linear transformation $A\in SL(2n,\mathbb F)$ of $\mathbb F^{2n}$ such that $A(\{x_1,\dots,x_n\})\subset\{0\}^n\times\mathbb F^n$.

Remark 2. For any vector $(x,y)\in\mathbb R^2$ and any $\varepsilon>0$ there exists a matrix $A\in SL(2,\mathbb Z)$ such that $(x,y)\cdot A\in (-\varepsilon,\varepsilon)\times\mathbb R$. Such matrix $A$ can be constructed by finding relatively prime integer numbers $p,q$ such that $|xp+yq|<\varepsilon$ and then finding integer numbers $a,b$ such that $pb-qa=1$ (using the extended Euclidean algorithm). Then the matrix $A=\left(\begin{array}&p&a\\q&b\end{array}\right)\in SL(2,\mathbb Z)$ has the required property.

Taking into account Remarks 1 and 2, I would expect that the following stronger form of Problem 2 has an affirmative answer.

Problem 3. Is it true that for any $n\in\mathbb N$ and vectors $x_1,\dots,x_n\in\mathbb R^{2n}$ there exists a linear transformation $A\in SL(2n,\mathbb Z)$ such that $A(\{x_1,\dots,x_n\})\subset(-\varepsilon,\varepsilon)^n\times\mathbb R^n$?

$\endgroup$
3
  • $\begingroup$ Do you have a serious reason to restrict to matrices of determinant 1? $\endgroup$
    – YCor
    Feb 13, 2020 at 9:02
  • $\begingroup$ @YCor I far as I know, no. In order to construct a homomorphism $\alpha$ from a matrix $A$ we need that $A$ is invertible, but its determinant can be $-1$. $\endgroup$ Feb 13, 2020 at 9:22
  • $\begingroup$ @AlexRavsky Yes. I just think that restricting to $SL$ can just make some minor uninteresting complications. $\endgroup$
    – YCor
    Feb 13, 2020 at 10:17

1 Answer 1

4
$\begingroup$

This is a draft proof of an affirmative answer to Problem 3.

Proposition. For any $n\in\mathbb N$, $\varepsilon>0$ and vectors $x_1,\dots,x_n\in\mathbb R^{2n}$ there exists a linear transformation $A\in SL(2n,\mathbb Z)$ such that $A(\{x_1,\dots,x_n\})\subset[-\varepsilon,\varepsilon]^n\times\mathbb R^n$.

Proof. Let $m=2n$, $x_i=(x_{i1},\dots,x_{im})$ for each $i$ and $X=\|x_{ij}\|$. We shall call a column of a matrix small, provided all its entries have absolute value at most than $\varepsilon$, and big, otherwise. Let $k$ be the maximal number of small columns in a matrix $XB$, where $B\in SL(m,\mathbb Z)$ and $C$ be an arbitrary matrix in $SL(m,\mathbb Z)$ such that $XC$ has $k$ small columns. It suffices to show that if $k<n$ then there exists a matrix $D\in SL(m,\mathbb Z)$ such that a matrix $XCD$ has $k+1$ small columns. Without loss of generality we can suppose that the big columns $y_1,\dots, y_{l}$ of the matrix $XC$ are from the first to the $l$-th. Pick a positive integer $M$ such that the maximal absolute value of an entry of these columns is at most $M\varepsilon$. Pick an arbitrary positive integer $K>(2lM)^n$ and define a map $f$ from the subset $Q^l$ of points of the set $[-K, K]^l$ with all integer coordinates to $\mathbb R^n$ by putting $f(d)=d_1y_1+\dots + d_ly_l$ for each $d=(d_1,\dots,d_l)\in Q^l$. Since all $|d_i|\le K$ and all entries of columns $y_i$ have absolute value at most $M\varepsilon$, each coordinate of a vector $f(d)$ is at most $lKM\varepsilon$. Therefore the image $f(Q^l)$ can be covered by $(2lKM)^n$ axis-parallel cubes with side $\varepsilon$. Since $|Q^l|=(2K+1)^l>(2K)^{n+1}>(2lKM)^n$, there exist two distinct elements $d’$ and $d’$ of $Q^l$ such that each coordinate of a vector $y=f(d’)-f(d’’)$ has an absolute value at most $\varepsilon$. Put $d=d’-d’’$. Dividing entries of $d$ by their greatest common divisor, if needed, we can assume that the greatest common divisor of entries of $d$ is $1$. It is well-known (see, for instance, this MSE thread]) that there exists a matrix $D’\in SL(l,\mathbb Z)$ whose first column is $d$. Put $D=\begin{pmatrix} D’ & 0\\ 0 & I\end{pmatrix}$, where $I$ is the $(m-l)\times (m-l)$ identity matrix. Then the first column of the matrix $XCD$ is small, whereas its last $k$ columns are the same as in the matrix $XC$. $\square$

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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