TITLE: Perfect matchings between levels in products of posets
QUESTION [8 upvotes]: Let $P$ be a finite poset. Assign the following diagram to it - put the maximal element of $P$ on the first level, the maximal of the rest to the second level, etc. Assume that the diagram is connected, that is, that between each pair of elements in the diagram we have a chain connecting them.
Let $L_n=P\times P\times\cdots\times P$ ($n$-times) to be the product poset, that is, one element in $L_n$ is larger than another if we have all inequalities coordinate-wise. Is it true that there exists and $N=N(P)$ such that for $n>N$ the following is true:
In the diagram of L_n there is a perfect matching between all consecutive levels (from the one with smaller size to the one with larger size).
Note: due to the useful feedback of Andres Caicedo and David Speyer the question was altered.

REPLY [6 votes]: The statement is still false when $P$ is connected and graded. It is even false if we add in addition that $P$ has a unique minimal and maximal element.
Theorem For any graded finite poset $P$, let $\rho(P)$ be the average of the ranks of the elements of $P$. Suppose that $Q$ is an upper ideal of $P$ and $\rho(Q) < \rho(P)$. (And, equivalently, $\rho(P) < \rho(P \setminus Q)$.) For all sufficiently large $n$, there is no perfect matching between levels of $P^n$ and $Q^n$.
Proof Let $P$ be graded with $p_i$ elements in grade $i$. Let $p(t) = \sum p_i t^i$. So $P^n$ is graded with corresponding polynomial $P(t)^n$. I'll write $P(t)^n = \sum p^n_i t^i$. Define $Q(t)$ similarly and $q^n_i$. Suppose that $p^n_i \leq p^n_{i+1}$ but $q^n_i > q^n_{i+1}$. Then there could be no matching between levels $i$ and $i+1$ of $P^n$. The first inequality would force the matching to be upwards (level $i$ injecting into level $i+1$), but the second inequality would imply that there were not enough elements to match level $i$ of $Q^n$ to.
Now, by a result of Odlyzko and Richmond, for sufficiently large $n$, the coefficients of $P(t)^n$ will be unimodal. Also, by the central limit theorem, the coefficient of $t^k$ in $P(t)^n$ will be much larger when $k$ is near $\rho(P) n$ then anywhere else. Combining these results, the coefficients of $P(t)$ increase up to $ (\rho(P) + o(1))n$ and then decrease down the other side.
Similarly, the coefficients of $Q(t)^n$ increase up to $(\rho(Q)+o(1)) n$ and
then decrease. Since $\rho(Q) < \rho(P)$, this shows that the largest coefficient of $Q(t)^n$ occurs before the largest coefficient of $P(t)^n$ for $n$ large. Thus, for $n$ large, there is no matching at the levels between $(\rho(Q)+o(1)) n$ and $(\rho(P)+o(1)) n$. $\square$
Here is are explicit counterexamples. The poset $P$ has $6$ elements, denoted by the symbols $Q$ and $p$; the poset $Q$ is the $3$ elements labeled $Q$:

Then $P(t) = 3+3t$ and $P(t)^n$ has its largest coefficient at $t^{n/2}$, while $Q(t) = 2+t$ and has its largest coefficient at $t^{n/3}$. So, choosing $i$ between $n/3$ and $n/2$, there is no matching between level $i$ and level $i+1$.
Here is a similar example where $P$ has a maximal and minimal element. 

We have $\rho(P) = 3/2=1.5$ and $\rho(Q) = (5/7) 1 + (1/7) 2 + (1/7) 3 = 10/7 \approx 1.4$ as desired.