2
$\begingroup$

What conditions show that a sequence can be permuted so that no two successive members are equal?

  • 0
    An obvious necessary condition is that no collection of more than half the members of the sequence can have the same value. This might be a sufficient condition as well.2011-11-02
  • 0
    @Didier Piau: I'm wondering, what does "more than half the members of the sequence" mean exactly?2011-11-02
  • 0
    @zoo: With permutation, do you mean an arbitrary self-bijection of the natural numbers or one which fixed almost all elements?2011-11-02
  • 0
    @Rasmus, This means $\geqslant n+1$ if the sequence has length $2n$ or $2n-1$. But you knew it, didn't you?2011-11-02
  • 1
    Finite sequence, or infinite?2011-11-02
  • 0
    @Didier Piau: Well, for a finite sequence this is clear. But what if it is infinite?2011-11-02
  • 1
    @Rasmus, Unsurprisingly, the expression *half the members of the sequence* refers to finite sequences. And, unless I am missing something, the infinite sequence version is trivial: you lose if there is exactly one infinite class and you win otherwise.2011-11-02
  • 0
    @Brian: Eh, I'll figure out how to reword in when it's not 4am here anymore.2011-11-02
  • 0
    @Rasmus: Correction: for infinite sequences, you lose if there is exactly one infinite class which is also co-finite, and you win otherwise (that is, if there are no infinite class, or exactly one infinite class, also co-infinite, or at least two infinite classes). The additional co-finiteness condition is not needed to ensure that you lose if there are finitely many classes.2011-11-02

2 Answers 2

2

Here’s a proof that Christian’s greedy algorithm works.

Suppose that there are $n$ balls and $k$ colors. Suppose further that color $i$ occurs $n_i$ times, so that $n_1+\dots+n_k=n$, and that $n_i\le\lceil n/2 \rceil$ for each $i$. Finally, assume that $n_1\ge n_2\ge\dots\ge n_k$.

Even if we’ve already placed some balls, at most one color is unavailable, so we may take the first ball to be of color $1$ or color $2$. A problem would arise only if color $1$ is unavailable and $n_2=0$, but in that case $n_1=n$, which is impossible. To show that the induction goes through, it suffices to show that after the new ball is placed, there is no color with more than $\lceil(n-1)/2\rceil$ balls left.

Suppose first that the ball just placed is of color $1$. Clearly $n_1-1\le\lceil(n-1)/2\rceil$. Suppose that $n_i>\lceil(n-1)/2\rceil$ for some $i>1$. Then $\lceil(n-1)/2\rceil. If $n=2m$, this is $m, which is absurd, so $n=2m+1$. But then we have $m, so $n_i=m+1$, and $n\ge n_1+n_i\ge 2n_i=2m+2>n$, which is also absurd. Thus, the induction goes through just fine if the ball just placed is of color $1$.

Now suppose that it’s of color $2$. If $n_1>\lceil(n-1)/2\rceil$, then $\lceil(n-1)/2\rceil. As in the last paragraph $n$ must be odd, say $n=2m+1$, and $n_1=m+1$, which is a problem. However, this case arises only if color $1$ is unavailable, which means that we just placed a ball of color $1$ at the previous step. At the beginning of that step, therefore, there must have been $n+1=2m+2$ balls, of which $m+2$ were of color $1$, and that’s impossible. Thus, we must have $n_1\le\lceil(n-1)/2\rceil$, as desired. The colors $i$ for $i>2$ can then be handled as in the last paragraph.

1

Assume everything is finite. So basically we have finitely many balls of different colors assorted into heaps according to colors. I think that Didier Piau's conjecture is true and that the following greedy algorithm works: Take as next ball a ball from an allowed heap of maximal size.

  • 0
    It does work; a proof by induction isn’t very difficult.2011-11-02