1
$\begingroup$

Let $r$ and $h$ be a real numbers and $n>0$. Write the fractional parts $\{k*r+h\}$, for $k = 1,2, . . . n$, in increasing order as $$ a_1 < a_2 < \cdots < a_n.$$ Let $D_n$ be the set of all the differences $a_{i+1} - a_i$ for $k = 1,2, \ldots, n-1.$ Can someone cite a reference or give a proof that $D_n$ contains at most $3$ elements for every $n$?

For the case $h = 0$, a certain paper calls this a "somewhat surprising and apparently little-known fact", without proof or reference. The proposition appears to be factual even when $h$ isn't $0$.

$\endgroup$

1 Answer 1

3
$\begingroup$

The case $h=0$ is known as the "Three-Distance Theorem"; just google for numerous references or look here for discussion and nice pictures, or here for an interesting historical comment.

A standard reformulation of the theorem is as follows: if, for an irrational $\alpha$, the unit-length circle is partitioned "in the natural way" into $n$ arcs by the points $\alpha k$ with $k\in[1,n]$, then the lengths of these $n$ arcs take just two or three distinct values. This easily implies the case $h\ne 0$ where you basically select one of the arcs (that containing the point corresponding to $h$) and confine to the lengths of the remaining $n-1$ arcs.

$\endgroup$
3
  • $\begingroup$ Thanks - however, it appears that the three-distance theorem, alias "three-gap theorem", is only for $h=0$. I still need a reference or proof that answers the question (where $h$ is an arbitrary real number). $\endgroup$ Jun 30, 2015 at 20:22
  • $\begingroup$ Isn't the extension to general $h$ trivial? Doesn't it fall out of the usual proofs? $\endgroup$ Jul 1, 2015 at 0:05
  • $\begingroup$ @Clark Kimberling: I expanded my answer a little to address the general case of $h\ne 0$. Does this answer your question? $\endgroup$
    – Seva
    Jul 3, 2015 at 6:36

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.