# Intuition behind a particular definition regarding cycles of partitions

I'm reading this paper on cycles of partitions, and was wondering if anyone could motivate the last condition in the definition of the sets $M_n$ in terms of the partitions being examined.

In particular, for each positive integer $n$, $M_n$ consists of all bi-infinite sequences of positive integers whose terms never exceed $n$, and moreover, the $i$th term $\sigma_i$ is the number of previous terms $\sigma_j$ that are at least $i-j$. I can see that a partition of $n$ cannot have a length greater than $n$, but I'm confused about this second condition. I tried to rephrase the definition of $M_n$ as $$M_n = \{f:\mathbb{Z} \to \mathbb{N} \,\, \vert \,\, \max_i f(i) = n \text{ and } f(i) = \#\{f(j) : f(j) \ge i-j\}\}$$ and think about it on an $ij$ plane, but I'm still not grasping the intuitive reason for this condition.

Help would be appreciated!

Recall the operation of $T$: if $\lambda$ is the partition with parts $p_1,p_2,\ldots,p_\ell$, then $T(\lambda)$ is the partition with parts $p_1-1,p_2-1,\ldots,p_\ell-1,\ell$, ignoring any $0$ parts generated in the process. I’ll call $\ell$ the new part of $T(\lambda)$.

Now consider a cycle of partitions of some positive integer. Once the cycle gets going, every part of every partition in the cycle begins as a new part. Suppose that $\ell$ is the length of $\lambda_0$, where $\lambda_0$ is some partition in the cycle. Then $\ell$ is the new part of $\lambda_1=T(\lambda_0)$, and for $k=1,\ldots,\ell$ the partition $\lambda_k=T^k(\lambda_0)$ has $\ell+1-k$ as a part. In other words, the part corresponding to the length $\ell$ of $\lambda_0$ is present precisely in the $\ell$ consecutive partitions $\lambda_1,\ldots,\lambda_\ell$. Another way to say this is that $\lambda_0$ contributes a part to $\lambda_k$ iff $1\le k\le\ell$.

More generally, if $j<i$, and $|\lambda_j|$ is the length of $\lambda_j$, then $\lambda_j$ contributes a part to $\lambda_i$ iff $i-j\le|\lambda_j|$, and it follows that $\lambda_i$ must have one part for each $j<i$ such that $i-j\le|\lambda_j|$. In other words,

$$|\lambda_i|=|\{j<i:|\lambda_j|\ge i-j\}|\;.\tag{1}$$

Since the $\sigma_i$ of the paper is precisely my $|\lambda_i|$, $(1)$ says that

$$\sigma_i=|\{j<i:\sigma_j\ge i-j\}|\;,$$

which is the condition about which you were asking. As we’ve just seen, it’s a necessary condition on the sequence of partition lengths in a cycle of partitions.

When combined with a bound on the lengths, it’s also a sufficient condition, though that does not seem to me quite so obvious as the author of the paper implies.

• Thank you for such a clear explanation! I'll think more about how I'd prove the sufficiency part. – sourisse Mar 17 '15 at 10:55
• @sourisse: You’re welcome! – Brian M. Scott Mar 17 '15 at 10:57