1

I am reading this [Six Lonely Runners]https://www.researchgate.net/publication/220343204_Six_Lonely_Runners, specifically Chapter 4. I do not understand their use of Kronecker's Theorem here and the paper they cited their generalisation from is in German so I cannot understand that. I believe their generalisation can be summarised by this:

Given an n-tuple $(u_1,...,u_n)\in\mathbb{R}^n$ consider all solutions $(x_1,...,x_n)\in\mathbb{Q}^n$ to the equation \begin{equation}\label{independents} u_1x_1+...+u_nx_n=0 \end{equation} and let $J$ be a maximally independent set (over $\mathbb{R}$) of such solutions (for example if $(u_1,...,u_n)$ are linearly independent over $\mathbb{Q}$ then $J=\{(0,...,0)\})$. Write $J=\{(a_{11},a_{12},...,{a_1n}),...,(a_{m1},a_{m2},...,a_{mn}\}$ and write \begin{equation*} A = \begin{pmatrix} a_{11} & a_{12} & \dots &a_{1n} &\\ a_{21} & a_{22} & \dots & a_{2n} &\\ \vdots & \vdots & \dots & \vdots &\\ a_{m1} & a_{m2} & \dots & a_{mn} \end{pmatrix} \end{equation*} Then $\mathbb{R}\mathbf{u}+\mathbb{Z}^n$ is dense in $\ker_{\mathbb{R}}A+\mathbb{Z}^n$.

But I am struggling to even know where to start to try and prove this. Any help in proving this would be greatly appreciated.

  • With tools like google translate it is not impossible to get an English translation of Oskar Perron's Satz 65 in his book Irrationalzahlen: It says "Is $n$ a positive integer and $p$ prime then $\sqrt[n]{p}$ is an algebraic number of degree $n$." I used the Göschen edition from 1921 and hope it is the right one. This book has had many newer editions all the way into the 1960ies. – Kurt G. Mar 06 '23 at 17:11
  • @KurtG.: The German construction that literally translates as "Is $n$ a positive integer" means "If $n$ is a positive integer". – joriki Mar 06 '23 at 17:19
  • @joriki Sure. A typo. Thanks. The theorem that research gate paper quotes is however #65 from the 1951 edition of which I found not even a screenshot in the internet (google books skips those pages). So we are not sure if we have the right one. – Kurt G. Mar 06 '23 at 17:29
  • @KurtG.: This looks more like the right theorem, and this snippet search in the 1951 edition confirms that it had that number already in that edition. I'll be providing an English translation. – joriki Mar 06 '23 at 18:00

1 Answers1

1

Here’s an English translation of this version of Theorem 65 in Irrationalzahlen by Oskar Perron; this search seems to confirm that this theorem had that number already in the 1951 edition cited by Bohman, Holzman and Kleitman. Let me know if anything is unclear or you need translations of the passages referred to (p. $158$ and Theorem $64$).


German original


Satz $\mathbf{65}$. Wenn die Zahlen $\xi_1,\xi_2,\ldots,\xi_{n+1}$ voneinander rational unabhängig sind, so läßt sich das System der $n+1$ Ungleichungen

$$ \quad\quad\quad|\xi_\nu y-x_\nu-\eta_\nu|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n+1), $$

wie klein auch die positive Zahl $\epsilon$ sei, stets durch ganze Zahlen $x_\nu$ und eine reelle (im allgemeinen nicht ganze) Zahl $y$ befriedigen.

Sind aber die Zahlen voneinander rational abhängig und bestehen zwischen ihnen genau $n+1-s$ linear unabhängige Relationen

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\xi_\nu=0\quad\quad\quad(\kappa=1,2,\ldots,n+1-s) $$

mit ganzzahligen Koeffizienten $b_{\kappa\nu}$, so besteht die notwendige und hinreichende Bedingung für die Erfüllbarkeit der obigen Ungleichungen darin, daß die Zahlen $\eta_\nu$ den Bedingungen

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu=\sum_{\nu=1}^{n+1}b_{\kappa\nu}g_\nu\quad\quad\quad(\kappa=1,2,\ldots,n+1-s) $$

genügen, wo die $g_\nu$ ganze Zahlen sind.

Beweis. Daß die Bedingungen notwendig sind, ist wieder leicht zu sehen. Denn aus den zu erfüllenden Ungleichungen folgt sogleich

$$ \left|\sum_{\nu=1}^{n+1}b_{\kappa\nu}x_\nu+\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu\right|=\left|\sum_{\nu=1}^{n+1}b_{\kappa\nu}(-\xi_\nu y+x_\nu+\eta_\nu)\right|\lt\epsilon\sum_{\nu=1}^{n+1}\left|b_{\kappa\nu}\right|\;. $$

Läßt man hier $\epsilon$ eine nach Null abnehmende Folge durchlaufen, so kann man schließen wie Seite $158$ Mitte.

Daß die Bedingungen auch hinreichend sind, ist, falls alle $\xi_\nu$ verschwinden, d.h. im Fall $s=0$, trivial, da sie dann augenscheinlich besagen, daß alle $\eta_\nu$ ganze Zahlen sind. Im andern Fall dürfen wir speziell etwa $\xi_{n+1}\ne0$ voraussetzen. Alsdann sind, wenn wir zu jedem ganzzahligen $x_{n+1}$ ein reelles $y$ aus der Gleichung

$$ \xi_{n+1}y-x_{n+1}-\eta_{n+1}=0\tag{14} $$

bestimmen, nur noch die $n$ Ungleichungen

$$ \quad\quad\quad\left|\xi_\nu y-x_\nu-\eta_\nu\right|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n) $$

zu erfüllen. Diese gehen aber, wenn man für $y$ den Wert aus $(14)$ einsetzt, über in

$$ \quad\quad\quad\left|\frac{\xi_\nu}{\xi_{n+1}}x_{n+1}-x_\nu-\left(\eta_\nu-\frac{\xi_\nu\eta_{n+1}}{\xi_{n+1}}\right)\right|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n)\;. $$

Für die Erfüllbarkeit dieser Ungleichungen durch ganze Zahlen $x_1,\ldots,x_{n+1}$ ist aber, da zwischen den $\frac{\xi_\nu}{\xi_{n+1}}$ die Relationen

$$ \quad\quad\quad b_{\kappa,n+1}+\sum_{\nu=1}^nb_{\kappa\nu}\frac{\xi_\nu}{\xi_{n+1}}=0\quad\quad\quad(\kappa=1,2,\ldots,n+1-s) $$

bestehen, nach Satz $64$ notwendig und hinreichend

$$ \sum_{\nu=1}^nb_{\kappa\nu}\left(\eta_\nu-\frac{\xi_\nu\eta_{n+1}}{\xi_{n+1}}\right)=b_{\kappa,n+1}g_{n+1}+\sum_{\nu=1}^nb_{\kappa\nu}g_\nu\;, $$

was mit Rücksicht auf die vorausgehende Gleichung soviel besagt wie

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu=\sum_{\nu=1}^{n+1}b_{\kappa\nu}g_\nu\quad\quad\quad(\kappa=1,2,\ldots,n+1-s)\;. $$

Damit ist Satz $65$ bewiesen.


English translation


Theorem $\mathbf{65}$. If the numbers $\xi_1,\xi_2,\ldots,\xi_{n+1}$ are rationally independent, the system of $n+1$ inequalities

$$ \quad\quad\quad|\xi_\nu y-x_\nu-\eta_\nu|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n+1), $$

can always be satisfied by integers $x_\nu$ and a real number $y$ (generally not an integer), no matter how small $\epsilon$ is.

However, if the numbers are rationally dependent and exactly $n+1-s$ linearly independent relations

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\xi_\nu=0\quad\quad\quad(\kappa=1,2,\ldots,n+1-s) $$

obtain between them, with integer coefficients $b_{\kappa\nu}$, then the necessary and sufficient condition for the satisfiability of the above inequalities consists in the numbers $\eta_\nu$ fulfilling the conditions

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu=\sum_{\nu=1}^{n+1}b_{\kappa\nu}g_\nu\quad\quad\quad(\kappa=1,2,\ldots,n+1-s)\;, $$

where the $g_\nu$ are integers.

Proof. That the conditions are necessary is again easy to see. For the inequalities to be fulfilled immediately imply

$$ \left|\sum_{\nu=1}^{n+1}b_{\kappa\nu}x_\nu+\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu\right|=\left|\sum_{\nu=1}^{n+1}b_{\kappa\nu}(-\xi_\nu y+x_\nu+\eta_\nu)\right|\lt\epsilon\sum_{\nu=1}^{n+1}\left|b_{\kappa\nu}\right|\;. $$

If $\epsilon$ is taken to traverse a sequence decreasing towards zero, an analogous conclusion as in the middle of page $158$ can be drawn.

That the conditions are also sufficient is trivial if all the $\xi_\nu$ vanish, i.e. in the case $s=0$, since in this case they evidently express that all the $\eta_\nu$ are integers. Otherwise, we may assume, say, $\xi_{n+1}\ne0$. If we then determine a real $y$ for each integer $x_{n+1}$ from the equation

$$ \xi_{n+1}y-x_{n+1}-\eta_{n+1}=0\;,\tag{14} $$

only the $n$ inequalities

$$ \quad\quad\quad\left|\xi_\nu y-x_\nu-\eta_\nu\right|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n) $$

remain to be fulfilled. But if the value from $(14)$ is substituted for $y$, these become

$$ \quad\quad\quad\left|\frac{\xi_\nu}{\xi_{n+1}}x_{n+1}-x_\nu-\left(\eta_\nu-\frac{\xi_\nu\eta_{n+1}}{\xi_{n+1}}\right)\right|\lt\epsilon\quad\quad\quad(\nu=1,2,\ldots,n)\;. $$

But since the relations

$$ \quad\quad\quad b_{\kappa,n+1}+\sum_{\nu=1}^nb_{\kappa\nu}\frac{\xi_\nu}{\xi_{n+1}}=0\quad\quad\quad(\kappa=1,2,\ldots,n+1-s) $$

obtain among the $\frac{\xi_\nu}{\xi_{n+1}}$, by Theorem $64$ it is necessary and sufficient for the satisfiability of these inequalities by integers $x_1,\ldots,x_{n+1}$ that

$$ \sum_{\nu=1}^nb_{\kappa\nu}\left(\eta_\nu-\frac{\xi_\nu\eta_{n+1}}{\xi_{n+1}}\right)=b_{\kappa,n+1}g_{n+1}+\sum_{\nu=1}^nb_{\kappa\nu}g_\nu\;, $$

which in view of the preceding equation is equivalent to

$$ \quad\quad\quad\sum_{\nu=1}^{n+1}b_{\kappa\nu}\eta_\nu=\sum_{\nu=1}^{n+1}b_{\kappa\nu}g_\nu\quad\quad\quad(\kappa=1,2,\ldots,n+1-s)\;. $$

This concludes the proof of Theorem $65$.

joriki
  • 238,052
  • Thank you so much for this! Any ideas on what $\eta_v$ is in the third line? Can it be anything (I'd assume so because you could just change $x_v$)? – crewmate Mar 08 '23 at 19:06
  • @crewmate: As I understand it, the $\eta_\nu$ are given real numbers, as are the $\xi_\nu$, and the resulting inequalities are to be satisfied by finding $x_\nu$ and $y$. You can't just change $x_\nu$ to cancel a change in $\eta_\nu$ because the $x_\nu$ must be integers. – joriki Mar 09 '23 at 07:00
  • Are we able to choose the $g_\upsilon$? So the inequalities are satisfied if and ony if we can find integers $g_1,...,g_{n+1}$ such that $\sum_{\upsilon =1}^{n+1}b_{\kappa \upsilon}\eta_{\upsilon} = \sum_{\upsilon =1}^{n+1}b_{\kappa \upsilon}g_{\upsilon}$? – crewmate Mar 09 '23 at 17:21
  • @crewmate: That's how I'd interpret it, yes. – joriki Mar 09 '23 at 17:26