Remark: the decomposition $p = p_L * p_R$ above is not unique in general. In order to have a useful unique decomposition we need to organise partitions in a slightly non-canonical way, for example as follows. The `low-shell' of $p \in {\pbase} (n,m)$ is the partition obtained by replacing each propagating part by a part containing just the two lowest numbered elements (one primed, one unprimed), with all other elements being singletons. We say two propagating parts are low-non-crossing if their image in the low-shell is non-crossing in the obvious sense. We say $p$ is low-non-crossing if each pair of propagating parts is low-non-crossing. For example $1_n$ is (equal to its low-shell and) low-non-crossing. Low-shell example: \begin{figure} \caption{\label{fig0x} A partition drawn low-neat, and its low-shell} \includegraphics[width=3in]{xfig/partit-low1.gif} \end{figure} \begin{figure} \caption{\label{fig1x} A partition in $\pbase^{3}(7,3)$ and its low-shell} \includegraphics[width=2in]{xfig/partit-low2.gif} \end{figure} Note that the middle (dashed) horizontal line drawn in Fig.\ref{fig1x} separates the partition into a low-non-crossing partition and a permutation (in $S_3$, or in $\pbase^{3}(3,3)$, as it were). Note that $p$ can be low-non-crossing without being a plane partition.
%Let
${\pbase}^{=l}(n,m)$ denotes the subset of ${\pbase}^l (n,m)$
consisting of low-non-crossing elements.
\begin{proposition}
Every $p \in {\pbase}^{l} (n,l)$
can be expressed in the form $p_L * p_R$ where
$p_L \in {\pbase}^{=l} (n,l)$ and $p_R \in {\pbase}^l (l,l)$.
\end{proposition}
Proof: Elements $p_R$ of the given form can be used to `uncross any crossing'.
And they are invertible.
\begin{proposition}
The category composition defines bijections
\begin{eqnarray}
* : {\pbase}^{l} (n,l) \times {\pbase}^{=l} (l,m)
& \rightarrow & {\pbase}^{l} (n,m)
\label{eq01}
\\
(p_L , p_R ) & \mapsto & p_L * p_R
\end{eqnarray}
\begin{eqnarray}
** : {\pbase}^{=l} (n,l) \times {\pbase}^l (l,l) \times {\pbase}^{=l} (l,m)
& \rightarrow & {\pbase}^{l} (n,m)
\\
(p_L , p_C , p_R) & \mapsto & p_L * p_C * p_R
\end{eqnarray}
\end{proposition}
\subsection{Useful elements of $P_n$}
Fix a natural inclusion of algebras
\[
P_{n-1} \subset P_{n}
\]
by
$
p \mapsto p \cup \{ \{ n , n' \}\} \; = p \circc 1_1
$.
The identity element in $P_0$ is the empty partition. Note that this
and the (iterated) inclusion map above determines the identity in $P_n$.
Define an element
\[
%a_1 = \{ \{ 1 \}, \{ 1' \}, ... \}
a_1 = \{ \{ 1 \}, \{ 1' \} \}
\in P_1
\]
Define $a_1$ in $P_n$ by inclusion
(so it is otherwise the same as the identity element).
That is $a_1 = \{ \{ 1 \} , \{ 1' \} \} \otimes 1_{n-1}$ in $P_n$.
Define $a_i$ similarly.
Assuming (for now) that $\delta$ is a unit in $k$, define
\[
e_i = \frac{1}{\delta} a_i
\]
Note that this is idempotent.
Define
$E_n = e_1 e_2 ... e_n$
and
$E_n^l = e_1 e_2 ... e_{n-l}$.
(The point of this notation is that, in $P_n$,
$E_n^l$ is a scalar multiple of a partition with $l$ propagating components.)
Note the useful isomorphism
\begin{equation}
E_n^l P_n E_n^l \cong P_l .
\end{equation}
In particular
we have
\[
%e_1 e_2 ... e_n
E_n P_n E_n = k E_n
\]
Now for any ring $R$ and idempotent $e \in R$, $eRe$ a local ring
implies that $e$ is primitive.
Thus, $E_n$ is a primitive idempotent.
\subsection{Ideals}
Consider the chain of ideals
\[
P_n \supset P_n e_1 P_n \supset P_n e_1 e_2 P_n \supset ...
\supset P_n e_1 e_2 ... e_n P_n
\]
Define $P_n^l = P_n e_1 e_2 ... e_{n-l} P_n$.
(Explicitly, $P_n^n = P_n$, $P_n^{n-1} = P_n e_1 P_n$,
... , $P_n^{0} = P_n e_1 e_2 ... e_n P_n$,
and we take also $P_n^{-1} = 0$.)
Proposition.
The ideal
$P_n^l$ is spanned by partitions
with at most $l$ propagating components.
Proof: Exercise.
Define quotient algebra
$$
P_n^{/l} = P_n / P_n^l .
$$
This is like $P_n$, except that any partition basis element with $l$ or fewer
propagating components is congruent to 0.
In particular the idempotent $E_n^m = 0$ unless $m > l$ in $P_n^{/l}$.
Let $S_n$ denote the symmetric group.
\begin{proposition}
We have isomorphisms of $k$-algebras
\[
E_n^{l+1} P_n^{/l} E_n^{l+1} \cong P_{l+1}^{/l}
\cong k S_{l+1}
\]
\end{proposition}
Proof:
We can choose a subset of the partition basis of $P_n$ to form a basis for the algebra
on the left.
No such element on the left can have more than $l+1$ propagating components
by virtue of the factor $E_n^{l+1}$;
nor less than $l+1$ by virtue of the quotient.
Consider a
complete orthogonal idempotent decomposition of $1 \in kS_{l+1}$:
$$
1 = \sum_\lambda f_\lambda
$$
% in $kS_{l+1}$
Recall that such a decomposition exists for any artinian ring;
and that each $f_\lambda kS_{l+1} f_\lambda$ is a local ring
- see e.g. [AndersonFuller].
In our case, if $k = \C$ say, the idempotents may be indexed by
$\lambda \in \Lambda_{l+1}^{st}$,
the set of standard Young tableau associated to all the integer partitions of $l+1$.
This will now
give us a certain collection of primitive idempotents in $P_n^{/l}$:
for any $\lambda \in \Lambda_{l+1}^{st}$ setting
$$
F_\lambda = E_{n-({l+1})} \circc f_{\lambda}
= E_n ^{l+1} (1_{n-(l+1)} \circc f_\lambda )
$$
(in the side-by-side notation)
we have
\[
F_\lambda P_n^{/l} F_\lambda =
( E_{n-({l+1})} \circc f_\lambda )
P_n^{/l}
( E_{n-({l+1})} \circc f_\lambda )
= E_{n -( {l+1})} \circc ( f_\lambda P_{l+1}^{/l} f_\lambda )
\cong f_\lambda (k S_{l+1} ) f_\lambda
%= kf_\lambda
\]
a local ring.
Thus the left ideal
\[
\Pi_\lambda =
P_n^{/l} F_\lambda
\]
is indecomposable projective in the quotient $P_n^{/l}$.
Hence for each $l\in \{ -1,0,...,n-1 \}$ and each $\lambda \in \Lambda_{l+1}$
the module $\Pi_\lambda$ is also an
indecomposable $P_n$-module with simple head.
\subsection{Modules}
Define $P_n,P_n$-bimodule
\[
\bar{P}_n^l = P_n^l / P_n^{l-2}
\]
Note that this has a basis $\pbase^l(n,n)$ (of
the $(n,n)$-partitions with $l$ propagating parts).
%`lines' (components).
We have the following useful decomposition as a left-module
\[
\bar{P}_n^l = \oplus_{p_R \in \pbase^{=l}(l,n)} k \pbase^{l}(n,l) * p_R
\]
Cf. Equation(\ref{eq01}), noting that each $k \pbase^{l}(n,l) * p_R$
is a $P_n$-module by virtue of the quotient
(action of $P_n$ from the left on some $p_L * p_R$ does not
change $p_R$ unless it reduces the number of propagating lines,
in which case the action gives 0).
It will be clear that each of these summands is isomorphic, so we can
study $\bar{P}_n^l$ as a left-module by studying $k \pbase^{l}(n,l) * p_R$,
or indeed $k \pbase^{l}(n,l) $.
Note that the latter is also a right $kS_l$-module in an obvious way,
indeed it is a free module. Thus we have a functor
\[
S : kS_l - mod \rightarrow P_n-mod
\]
given by $M \mapsto k \pbase^{l}(n,l) \otimes_{kS_l} M$.
The idempotent decomposition of 1 in $kS_l$ as above induces
a decomposition
\[
k \pbase^{l}(n,l) \cong \oplus_\lambda \Pi_\lambda
\]
Altogether then, we see that the left regular module may be filtered by
the indecomposable modules of form $\Pi_\lambda$ (all possible $\lambda$).
In the complex case (for example) it will be clear that
$kS_l f_\lambda \cong kS_l f_\mu$ if and only if
the underlying Young diagrams of $\lambda$ and $\mu$ are the same.
Thus a complete set of $\Pi_\lambda$ modules is obtained by taking
one of each shape.
It will be clear that $e_1 P_n e_1 \cong P_{n-1}$.
Indeed $e_i P_n e_i \cong P_{n-1}$ for any suitable $i$.
It follows that
we have a functor
\[
G : P_{n-1}-mod \rightarrow P_n-mod
\]
given by $M \mapsto P_n e_n \otimes M$.
Next we will show that $G(\Pi_\lambda ) = \Pi_\lambda$
(the reader should take care to consider the meaning of this identity).
We will be able to use this to determine the structure of $P_n$ by
induction on $n$ (using the fact that $P_0$ and $P_1$ are simple base
cases which may be analysed by brute force).
...