\subsection{Description}

\subsection{Canonicity}

\subsection{Experimental Results}

We compare sizes of canonical mod-$p$ decision diagrams with
the results for MTDDs, shared MBTDDs, and MDDs for characteristic
functions found in \cite{sasao96}. 

\section{Framework for Edge Valuation}

BITS FROM THESIS:

All of these proposed sets of edge valuations have the common
property of being groups on the range of the
diagram. In Figure~\ref{fig:edgexf}, we show why: the node shown
in the figure is being created using {\sc MakeUnique}; in
order to maintain uniqueness, it needs to be normalized, with
the transformer on the lowest edge being the identity. On the
left, the transformer on the lowest edge is $f$; so $f^{-1}$
is calculated, and applied (on the left) to all of the
transformers on the outgoing edges. This normalized node
is either already in the unique table, or it is created;
{\sc MakeUnique} then returns that table entry, together
with the edge transformer $f$ leading into it.

MORE: GIVE THE SUFFICIENT CONDITIONS FOR THE CANONICITY PROOF.