1: \documentclass[12pt]{article}
2:
3: %\usepackage{showlabels}
4: %\usepackage{fullpage}
5: \usepackage{amsmath}
6: \usepackage{amsthm}
7: \usepackage{amssymb}
8: \usepackage{latexsym}
9: %\usepackage{subfigure}
10: %\usepackage{hyperref}
11: %\usepackage{units}
12: \usepackage{graphicx, pst-plot, pst-node, pst-text, pst-tree}
13: %\usepackage{mathabx}
14:
15: \newtheorem{theorem}{Theorem}[section]
16: \newtheorem{proposition}[theorem]{Proposition}
17: \newtheorem{lemma}[theorem]{Lemma}
18: \newtheorem{definition}[theorem]{Definition}
19: \newtheorem{corollary}[theorem]{Corollary}
20: \newtheorem{example}[theorem]{Example}
21: \newtheorem{conjecture}[theorem]{Conjecture}
22: \newtheorem{question}[theorem]{Question}
23:
24: \newcounter{todocounter}
25: \newcommand{\todo}[1]{
26: \addtocounter{todocounter}{1}
27: \bigskip
28: \noindent{\bf $\ll$ To-do \#\arabic{todocounter}:\rule{10pt}{0pt}#1 $\gg$}\bigskip
29: }
30:
31: \newcommand{\note}[1]{
32: \bigskip\noindent{\it Note to Sophie: #1\/}\bigskip
33: }
34:
35: \setlength{\textwidth}{6.4in} \setlength{\textheight}{8.9in}
36: \setlength{\topmargin}{0pt} \setlength{\headsep}{0pt}
37: \setlength{\headheight}{0pt} \setlength{\oddsidemargin}{0pt}
38: \setlength{\evensidemargin}{0pt} \makeatletter
39: \newfont{\footsc}{cmcsc10 at 8truept}
40: \newfont{\footbf}{cmbx10 at 8truept}
41: \newfont{\footrm}{cmr10 at 10truept}
42: \pagestyle{plain}
43:
44: \title{Grid classes and the Fibonacci dichotomy for restricted permutations}
45: \author{Sophie Huczynska\thanks{Supported by a Royal Society Dorothy Hodgkin Research Fellowship.}\ \ and Vincent Vatter\thanks{Supported by EPSRC grant GR/S53503/01.}\\[-5pt]
46: \small School of Mathematics and Statistics\\[-5pt]
47: \small University of St.\ Andrews\\[-5pt]
48: \small St.\ Andrews, Fife, Scotland\\[-5pt]
49: \small \texttt{\{sophieh, vince\}@mcs.st-and.ac.uk}\\[-5pt]
50: \small \texttt{http://turnbull.mcs.st-and.ac.uk/\~{}\{sophieh,
51: vince\}}}
52:
53: \date{\today \\[6pt]
54: \begin{flushleft}
55: \small Key Words: Fibonacci number, grid class, permutation class, restricted permutation\\[6pt]
56: \small AMS classification:
57: \small Primary 05A05;
58: \small Secondary 05A15, 05A16.
59: \end{flushleft}
60: }
61:
62: \begin{document}
63: \maketitle
64:
65: \newcommand{\st}{\operatorname{st}}
66: \newcommand{\Av}{\operatorname{Av}}
67: \newcommand{\C}{\mathcal{C}}
68: \newcommand{\vecv}{{\bf v}}
69: \newcommand{\x}{{\bf x}}
70: \newcommand{\y}{{\bf y}}
71: \newcommand{\zpm}{0/\mathord{\pm} 1}
72: \newcommand{\Grid}{\operatorname{Grid}}
73: \renewcommand{\sb}{\mathcal{SB}}
74:
75: \begin{abstract}
76: We introduce and characterise grid classes, which are natural generalisations of other well-studied permutation classes. This characterisation allows us to give a new, short proof of the Fibonacci dichotomy: the number of permutations of length $n$ in a permutation class is either at least as large as the $n$th Fibonacci number or is eventually polynomial.
77: \end{abstract}
78:
79: \section{Introduction}
80:
81: A permutation $\pi$ of $[n]$
82: \footnote{Here
83: $[n]=\{1,2,\ldots,n\}$ and, more generally, for $a,b \in
84: \mathbb{N}$ ($a<b$), the interval $\{a,a+1,\ldots,b\}$ is
85: denoted by $[a,b]$, the interval $\{a+1,a+2,\ldots,b\}$ is denoted by
86: $(a,b]$, and so on.}
87: contains the permutation $\sigma$ of $[k]$ ($\sigma\le\pi$) if $\pi$ has a subsequence of length $k$
88: in the same relative order as $\sigma $.
89: For example, $\pi=391867452$ (written in list, or one-line notation)
90: contains $\sigma=51342$, as can be seen by considering the subsequence
91: $91672$ ($=\pi(2),\pi(3),\pi(5),\pi(6),\pi(9)$).
92: A {\it permutation class\/} is a downset of permutations under this order, or in other words, if $\C$ is a permutation class, $\pi\in\C$, and $\sigma\le\pi$, then $\sigma\in\C$.
93: We shall denote by $\C_n$ ($n \in \mathbb{N}$)
94: the set $\C \cap S_n$, i.e.\ those permutations in $C$ of length
95: $n$.
96: Recall that
97: an {\it antichain} is a set of pairwise incomparable elements.
98: For any permutation class $\C$, there is a unique (and possibly infinite) antichain $B$ such that $\C=\Av(B)=\{\pi: \beta \not \leq\pi\mbox{ for all } \beta \in B\}$. This antichain $B$ is called the {\it basis} of $\C$.
99: Permutation classes arise naturally in a variety of disparate fields, ranging from the analysis of sorting machines (dating back to Knuth~\cite{knuth1}, who proved that a permutation is stack-sortable if and only if it lies in the class $\Av(231)$) to the study of Schubert varieties (see, e.g., Lakshmibai and Sandhya~\cite{ls:smooth}).
100:
101: The Stanley-Wilf Conjecture, recently proved by Markus and Tardos~\cite{mt:swc}, states that all permutation classes except the set of all permutations have at most exponential growth, i.e., for every class $\C$ with a nonempty basis, there is a constant $K$ so that $|\C_n|<K^n$ for all $n$. Less is known regarding the exact enumeration of permutation classes. Natural enumerative questions include:
102: \begin{enumerate}
103: \item[(i)] Which permutation classes are finite?
104: \item[(ii)] Which permutation classes are enumerated by a polynomial?
105: \item[(iii)] Which permutation classes have rational generating functions? (We refer to $\sum |\C_n|x^n$ as the generating function of $\C$.)
106: \item[(iv)] Which permutation classes have algebraic generating functions?
107: \item[(v)] Which permutation classes have $P$-recursive enumeration?
108: \end{enumerate}
109: The answer to the first question on this list follows easily from the Erd\H os-Szekeres Theorem\footnote{{\bf The Erd\H os-Szekeres Theorem~\cite{es:acpig}.}
110: Every permutation of length $n$ contains a monotone subsequence of length at least $\sqrt{n}$.}: the class $\Av(B)$ is finite if and only if $B$ contains both an increasing permutation and a decreasing permutation.
111: The answer to the second question is provided in this paper. Questions (iii)--(v) remain unanswered.
112:
113: \bigskip\noindent{\bf Downsets of vectors.} Perhaps the simplest interesting context in which to study downsets is finite vectors of nonnegative integers, and in this context there is also a polynomial enumeration result which we shall employ in our proofs. Let
114: $\x=(x_1,x_2,\dots,x_m),\y=(y_1,y_2,\dots,y_m)\in\mathbb{N}^m$ for
115: some $m$. We write $\x\le\y$ if $x_i\le y_i$ for all $i\in[m]$.
116: This order is often called the {\it product order\/}. The weight
117: of the vector $\x$, denoted $\|\x\|$, is the sum of the entries of
118: $\x$.
119:
120: \begin{figure}
121: \begin{center}
122: \psset{xunit=0.01in, yunit=0.01in} \psset{linewidth=1\psxunit}
123: \begin{pspicture}(0,0)(150,150)
124: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
125: showorigin=false, labels=none](0,0)(140,140)
126: \multips(0,0)(10,0){15}{\pscircle*{4.0\psxunit}}
127: \multips(0,10)(0,10){14}{\pscircle*{4.0\psxunit}}
128: \multips(10,10)(10,0){14}{\pscircle*{4.0\psxunit}}
129: \multips(10,20)(0,10){13}{\pscircle*{4.0\psxunit}}
130: \multips(20,20)(0,10){13}{\pscircle*{4.0\psxunit}}
131: \multips(30,20)(0,10){13}{\pscircle*{4.0\psxunit}}
132: \pscircle*(40,20){4.0\psxunit} \pscircle*(50,20){4.0\psxunit}
133: \pscircle*(60,20){4.0\psxunit}
134: \multips(40,30)(0,10){5}{\pscircle*{4.0\psxunit}}
135: \pscircle(70,20){4.0\psxunit} \pscircle(50,30){4.0\psxunit}
136: \pscircle(40,80){4.0\psxunit}
137: \end{pspicture}
138: \end{center}
139: \caption{The plot of downset in $\mathbb{N}^2$; the elements
140: of the class are drawn with solid circles, while the elements of
141: the basis are drawn with hollow circles.}
142: \end{figure}
143:
144: \begin{theorem}\label{vector-poly}
145: Let $\C$ denote a downset in $\mathbb{N}^m$. For
146: sufficiently large $n$, the number of vectors in $\C$ of weight
147: $n$ is given by a polynomial.
148: \end{theorem}
149:
150: Stanley~\cite{s:E2546} posed Theorem~\ref{vector-poly} as a
151: {\it Monthly\/} problem in 1976 and offered two solutions. One of these solutions is elementary while the other follows from viewing the number of vectors in question as a Hilbert function.
152:
153: \bigskip\noindent{\bf Downsets of other objects.} Downsets of other combinatorial objects have been extensively studied, and other polynomial enumeration results are known. These have often been established by ideas analogous to the grid classes of matchings we use.
154:
155: For example, downsets of graphs w.r.t.\ the induced subgraph ordering that are closed under isomorphism are called {\it hereditary properties\/}. Let $\mathcal{P}$ denote a hereditary property, and let $\mathcal{P}_n$ denote the set of graphs in $\mathcal{P}$ with vertex set $[n]$. Scheinerman and Zito~\cite{sz:hered} proved that $|\mathcal{P}_n|$ either has polynomial growth (meaning that $|\mathcal{P}_n|=\Theta(n^k)$ for some $k$) or $|\mathcal{P}_n|$ has at least exponential growth. Balogh, Bollob{\'a}s, and Weinreich~\cite{bbw:growth2000} later showed that polynomial growth hereditary properties are enumerated exactly by a polynomial for large $n$. Their proof of this result uses ``canonical properties,'' which are quite like our grid classes of matchings.
156:
157: Moving to a more general context, Pouzet and Thi\'ery~\cite{pt:poly} study polynomial growth (although not exact polynomial enumeration) for certain downsets of relational structures. While summarising their work would take us too far afield, we remark first that permutations can be viewed as relational structures\footnote{E.g., $\pi\in S_n$ can be taken to correspond to the relational structure on $[n]$ with two linear orders, $<$ and $\prec$, where $<$ is the normal ordering of $[n]$ and $i\prec j\iff \pi(i)<\pi(j)$.} and second that the grid classes of matchings we use essentially correspond to their concept of ``monomorphic decompositions into finitely many parts.''
158:
159: %
160: %
161: %
162: %
163: %
164: %
165: %
166: %
167: %
168: %
169: %
170: %
171: %
172: %
173: \section{Grid classes}\label{grid-classes}
174:
175: %
176: %
177: %
178: %
179: %
180: %
181: %
182: %
183: %
184: %
185: %
186: %
187: %
188: \subsection{The skew-merged permutations}\label{skew-merged}
189:
190: \begin{figure}[t]
191: \begin{center}
192: \psset{xunit=0.01in, yunit=0.01in} \psset{linewidth=1\psxunit}
193: \begin{pspicture}(0,0)(100,100)
194: %\psline[linecolor=gray](55,0)(55,100)
195: %\psline[linecolor=gray](0,55)(100,55)
196: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
197: showorigin=false, labels=none](0,0)(90,90)
198: \pscircle*(10,90){4.0\psxunit} \pscircle*(20,10){4.0\psxunit}
199: \pscircle*(30,70){4.0\psxunit} \pscircle*(40,40){4.0\psxunit}
200: \pscircle*(50,50){4.0\psxunit} \pscircle*(60,60){4.0\psxunit}
201: \pscircle*(70,30){4.0\psxunit} \pscircle*(80,20){4.0\psxunit}
202: \pscircle*(90,80){4.0\psxunit}
203: \end{pspicture}
204: \end{center}
205: \caption{The plot of the skew-merged permutation
206: $917456328$.}\label{fig-skew-merged}
207: \end{figure}
208:
209: We begin with an example of a grid class. A permutation is said to be {\it skew-merged\/} if it is the union of an increasing subsequence and a decreasing subsequence. For example, the permutation shown in Figure~\ref{fig-skew-merged} is skew-merged. Stankova~\cite{stankova:fs} was the first to find the basis of this class. Later, K\'ezdy, Snevily, and Wang~\cite{ksw:incdec} observed that the basis follows easily from F\"oldes and Hammer's characterisation of split graphs%
210: \footnote{A graph $G$ is {\it split\/} if its vertices can be
211: partitioned into a disjoint union $V(G)=V_1\uplus V_2$ s.t.\ $G[V_1]$
212: is complete and $G[V_2]$ is edgeless. F\"oldes and
213: Hammer proved that a graph is split if and only if it does not
214: contain $K_2\uplus K_2$, $C_4$, or $C_5$ as induced subgraphs.} in
215: \cite{fh:sg}.
216:
217: \begin{theorem}[Stankova~\cite{stankova:fs}; K\'ezdy, Snevily, and Wang~\cite{ksw:incdec}; and Atkinson~\cite{a:skewmerged}]\label{skewmerged}
218: The skew-merged permutations are $\Av(2143, 3412)$.
219: \end{theorem}
220:
221: Atkinson~\cite{a:skewmerged} gave another proof of
222: Theorem~\ref{skew-merged} and solved the enumeration problem for
223: the skew-merged permutations; he showed that their generating
224: function is given by
225: $$
226: \frac{1-3x}{(1-2x)\sqrt{1-4x}}.
227: $$
228:
229: K\'ezdy, Snevily, and Wang~\cite{ksw:incdec} studied one
230: generalization of skew-merged permutations, the class of
231: permutations which can be partitioned into $r$ increasing
232: subsequences and $s$ decreasing subsequences. Grid classes
233: provide a different generalization.
234:
235: %
236: %
237: %
238: %
239: %
240: %
241: %
242: %
243: %
244: %
245: %
246: %
247: %
248: \subsection{Definitions}
249:
250: First an important warning: when discussing grid classes,
251: we index matrices beginning from the lower left-hand corner, and we
252: reverse the rows and columns; for example $M_{3,2}$ denotes for us
253: the entry of $M$ in the $3$rd column from the left and $2$nd row
254: from the bottom. Below we include a matrix with its entries
255: labeled:
256: $$
257: \left(
258: \begin{footnotesize}
259: \begin{array}{rrr}
260: (1,2)&(2,2)&(3,2)\\
261: (1,1)&(2,1)&(3,1)
262: \end{array}
263: \end{footnotesize}
264: \right).
265: $$
266:
267: \bigskip
268:
269: Roughly, the {\it grid class\/} of a matrix $M$ is the set of all
270: permutations that can be divided in a prescribed manner (dictated
271: by $M$) into a finite number of blocks, each containing a monotone subsequence. We have already
272: introduced the best-studied grid class, the skew-merged
273: permutations. We previously defined them as the permutations that
274: can be written as the union of an increasing subsequence and a
275: decreasing subsequence. As a grid class, the skew-merged
276: permutations can be defined as the permutations that can be
277: divided into four monotonic blocks, two increasing and two
278: decreasing, as indicated in Figure~\ref{fig-skew-merged-gridded},
279: and our notation for this class is
280: $$
281: \Grid\left(\begin{footnotesize}\begin{array}{rr}-1&1\\1&-1\end{array}\end{footnotesize}\right),
282: $$
283: but before reaching that point we need to introduce some notation.
284:
285: \begin{figure}[t]
286: \begin{center}
287: \psset{xunit=0.01in, yunit=0.01in} \psset{linewidth=1\psxunit}
288: \begin{pspicture}(0,0)(100,100)
289: \psline[linecolor=gray](55,0)(55,100)
290: \psline[linecolor=gray](0,55)(100,55) \psaxes[dy=10, Dy=1, dx=10,
291: Dx=1, tickstyle=bottom, showorigin=false, labels=none](0,0)(90,90)
292: \pscircle*(10,90){4.0\psxunit} \pscircle*(20,10){4.0\psxunit}
293: \pscircle*(30,70){4.0\psxunit} \pscircle*(40,40){4.0\psxunit}
294: \pscircle*(50,50){4.0\psxunit} \pscircle*(60,60){4.0\psxunit}
295: \pscircle*(70,30){4.0\psxunit} \pscircle*(80,20){4.0\psxunit}
296: \pscircle*(90,80){4.0\psxunit}
297: \end{pspicture}
298: \end{center}
299: \caption{A gridding of the skew-merged permutation
300: $917456328$.}\label{fig-skew-merged-gridded}
301: \end{figure}
302:
303: Given a permutation $\pi\in S_n$ and sets $A,B\subseteq[n]$,
304: we write
305: $\pi(A\times B)$ for the subsequence of $\pi$ with indices
306: from $A$ which has values in $B$. For example, applying this
307: operation to the permutation shown in
308: Figure~\ref{fig-skew-merged-gridded}, we get
309: $$
310: 917456328([5]\times[5])=1,4,5,
311: $$
312: and this (increasing) subsequence gives the points in the
313: lower left-hand box of Figure~\ref{fig-skew-merged-gridded}. The
314: increasing subsequence in the upper right-hand box is
315: $$
316: 917456328([6,9]\times[6,9])=6,8,
317: $$
318: while the decreasing subsequence in the lower right-hand box is
319: $$
320: 917456328([6,9]\times[5])=3,2.
321: $$
322:
323: Now suppose that $M$ is a $t\times u$ matrix (meaning, in the
324: notation of this paper, that it has $t$ columns and $u$ rows).
325: An {\it $M$-gridding\/} of the permutation $\pi\in S_n$ is a pair
326: of sequences $1=c_1\le c_2\le\cdots\le c_{t+1}=n+1$ (the column
327: divisions) and $1=r_1\le r_2\le\cdots\le r_{u+1}=n+1$ (the row
328: divisions) such that for all $k\in[t]$ and $\ell\in[u]$,
329: $\pi([c_k,c_{k+1})\times[r_\ell,r_{\ell+1}))$ is:
330: \begin{itemize}
331: \item increasing if $M_{k,\ell}=1$, \item decreasing if
332: $M_{k,\ell}=-1$, \item empty if $M_{k,\ell}=0$.
333: \end{itemize}
334: We define the {\it grid class of $M$\/}, written $\Grid(M)$, to be
335: the set of all permutations that possess an $M$-gridding. We say
336: that $\pi$ is {\it $t\times u$-griddable\/} if it is $M$-griddable
337: for some $t\times u$ matrix $M$.
338:
339: A class $\C$ is said to be {\it $t\times u$-griddable\/} if every
340: permutation in $\C$ is $t\times u$-griddable, and it is said to be
341: {\it griddable\/} if it is $t\times u$-griddable for some
342: $t,u\in\mathbb{N}$. Note that all griddable classes lie in some
343: particular grid class (suppose that $\C$ is $t\times u$ griddable
344: and take a larger matrix $M$ containing every $t\times u$ matrix,
345: then $\C$ lies in $\Grid(M)$).
346:
347: Two special types of grid classes have been extensively studied.
348: One type is the profile classes of Atkinson~\cite{a:rp}, which in
349: our language are grid classes of permutation matrices. Another
350: example of grid classes are the $W$-classes introduced by
351: Atkinson, Murphy, and Ru\v{s}kuc~\cite{amr:pwocsop}, which are the
352: grid classes of $\zpm$ row vectors.
353:
354: Atkinson, Murphy, and Ru\v{s}kuc~\cite{amr:pwocsop} introduced $W$-classes in their study of partially well-ordered (pwo)\footnote{Recall that a partially ordered set is said to be {\it partially well-ordered (pwo)\/} if it contains neither an infinite properly decreasing sequence
355: nor an infinite antichain.} permutation classes, and proved that
356: grid classes of $\zpm$ row vectors are pwo. This result
357: does not extend to arbitrary grid classes, i.e., some grid classes
358: contain infinite antichains, e.g., there is an infinite antichain of
359: skew-merged permutations.
360: In order to characterise the pwo grid
361: classes, we associate a graph to each grid class. For any
362: $t\times u$ matrix $M$ we construct the bipartite graph $G(M)$
363: with vertices $x_1$, $x_2$, $\dots$, $x_t$ and $y_1$, $y_2$,
364: $\dots$, $y_u$ and edges $x_ky_\ell$ precisely when
365: $M_{k,\ell}\neq 0$. For example, the bipartite graph of a vector
366: is a star together with isolated vertices, while the bipartite
367: graph of
368: $\left(\begin{footnotesize}\begin{array}{rr}-1&1\\1&-1\end{array}\end{footnotesize}\right)$
369: is a cycle with $4$ vertices. The pwo properties of a grid class depend only on its graph.
370:
371: \begin{theorem}[Murphy and Vatter~\cite{profile}]\label{prof-pwo}
372: The grid class of $M$ is pwo if and only if $G(M)$ is a forest.
373: \end{theorem}
374:
375:
376: %
377: %
378: %
379: %
380: %
381: %
382: %
383: %
384: %
385: %
386: %
387: %
388: %
389: \subsection{The characterisation of griddable classes}
390:
391: It appears surprisingly difficult to compute the basis of
392: $\Grid(M)$ when $M$ is neither a vector nor a permutation matrix.
393: Waton [private communication] has computed the bases of
394: $\Grid(M)$ for all $2\times 2$ matrices $M$, but we know of
395: no such results for larger matrices.
396: In particular, the following remains a conjecture.
397:
398: \begin{conjecture}
399: All grid classes are finitely based.
400: \end{conjecture}
401:
402: We instead take a coarser approach here and ask only for a
403: characterisation of the griddable classes, that is, the
404: permutation classes that lie in some grid class.
405:
406: It will prove useful to have an alternative interpretation of
407: griddability. We say that the permutation $\pi\in S_n$ can be
408: {\it covered by $s$ monotonic rectangles\/} if there are
409: $[w_1,x_1]\times[y_1,z_1]$,$\dots$,$[w_s,x_s]\times[y_s,z_s]\subseteq[n]\times[n]$
410: such that
411: \begin{itemize}
412: \item for each $i\in[s]$, $\pi([w_i,x_i]\times[y_i,z_i])$ is
413: monotone, and \item $\displaystyle\bigcup_{i\in[s]}
414: [w_i,x_i]\times[y_i,z_i]=[n]\times[n]$.
415: \end{itemize}
416: Note that we allow these rectangles to intersect. By definition
417: every $t\times u$-griddable permutation can be covered by $tu$
418: monotonic rectangles. The following proposition gives the
419: other direction.
420:
421: \begin{proposition}\label{grid-coverings}
422: Every permutation that may be covered by $s$ monotonic rectangles
423: is $(2s-1)\times (2s-1)$-griddable.
424: \end{proposition}
425: \begin{proof}
426: Suppose that $\pi\in S_n$ is covered by the $s$ monotonic
427: rectangles $[w_1,x_1]\times[y_1,z_1]$, $\dots$,
428: $[w_s,x_s]\times[y_s,z_s]\subseteq[n]\times[n]$. Define the
429: indices $c_1,c_2,\dots,c_{2s}$ and $r_1,r_2,\dots,r_{2s}$ by
430: \begin{eqnarray*}
431: \{c_1\le c_2\le\cdots\le c_{2s}\}&=&\{w_1,x_1,w_2,x_2,\dots,w_s,x_s\},\\
432: \{r_1\le r_2\le\cdots\le
433: r_{2s}\}&=&\{y_1,z_1,y_2,z_2,\dots,y_s,z_s\}.
434: \end{eqnarray*}
435: Since these rectangles cover $\pi$, we must have $c_1=r_1=1$ and
436: $c_{2s}=r_{2s}=n$. Now we claim that these sets of indices form
437: an $M$-gridding of $\pi$ for some $2s-1\times 2s-1$ matrix $M$.
438:
439: To prove this claim it suffices to show that
440: $\pi([c_k,c_{k+1}]\times[r_\ell,r_{\ell+1}])$ is monotone for
441: every $k,\ell\in[2s-1]$, since we can then construct the matrix
442: $M$ based on whether this subsequence is increasing or decreasing.
443: Because the rectangles given cover $\pi$, the point $(c_k,r_\ell)$
444: lies in at least one rectangle, say $[w_m,x_m]\times[y_m,z_m]$.
445: Thus $c_k\ge w_m$ and $r_\ell\ge y_m$ and, because of the ordering
446: of the $c$'s and $r$'s, we have $c_{k+1}\le x_m$ and
447: $r_{\ell+1}\le z_m$. Therefore
448: $[c_k,c_{k+1}]\times[r_\ell,r_{\ell+1}]$ is contained in
449: $[w_m,x_m]\times[y_m,z_m]$ and so
450: $\pi([c_k,c_{k+1}]\times[r_\ell,r_{\ell+1}])$ is monotone.
451: \end{proof}
452:
453: With this new interpretation of griddability established, we need only two more definitions before characterising the griddable classes. Given two permutations $\pi\in S_m$ and $\sigma\in S_n$, we define their {\it direct sum\/}, written $\pi\oplus\sigma$ by
454: $$
455: (\pi\oplus\sigma)(i)
456: =
457: \left\{
458: \begin{array}{ll}
459: \pi(i)&\mbox{if $i\in [m]$,}\\
460: \sigma(i-m)+m&\mbox{if $i\in[m+n]\setminus[m]$,}
461: \end{array}
462: \right.
463: $$
464: and similarly define their {\it skew sum\/} by
465: $$
466: (\pi\ominus\sigma)(i)
467: =
468: \left\{
469: \begin{array}{ll}
470: \pi(i)+n&\mbox{if $i\in [m]$,}\\
471: \sigma(i-m)&\mbox{if $i\in[m+n]\setminus[m]$.}
472: \end{array}
473: \right.
474: $$
475:
476: \begin{theorem}\label{griddable-characterisation}
477: A permutation class is griddable if and only if it does not
478: contain arbitrarily long direct sums of $21$ or skew sums of $12$.
479: \end{theorem}
480: \begin{proof}
481: If a permutation class does contain arbitrarily long direct sums
482: of $21$ or skew sums of $12$, then it is clearly not griddable.
483:
484: For the other direction, let $\C$ be a permutation class that does
485: not contain $\ominus^{a+1} 12$ or $\oplus^{b+1} 21$. We show
486: by induction on $a+b$ that there is a function $f(a,b)$ so that
487: every permutation in $\C$ can be covered by $f(a,b)$ monotonic
488: rectangles, and thus we will be done by
489: Proposition~\ref{grid-coverings}.
490:
491: First note that if either $a$ or $b$ is $0$ then $\C$ can only
492: contain monotone permutations, so we can set $f(a,0)=f(0,b)=1$.
493: The next case is $a+b=2$, and since we may assume that $a,b\neq
494: 0$, we have $a=b=1$. Thus $\C$ contains neither $\ominus^2
495: 12=3412$ nor $\oplus^2 21=2143$, so $\C$ is a subclass of the
496: skew-merged permutations and thus every permutation in $\C$ may be
497: covered by $4$ monotonic rectangles and we may take $f(1,1)=4$.
498:
499: %\definecolor{gray}{rgb}{.75, .75, .75}
500:
501: \begin{figure}
502: \begin{center}
503: \begin{tabular}{ccccccc}
504:
505: \psset{xunit=0.02in, yunit=0.02in} \psset{linewidth=0.5\psxunit}
506: \begin{pspicture}(0,0)(50,50)
507: \psframe[linecolor=white,fillstyle=vlines,hatchangle=45,hatchcolor=gray](0,0)(20,20)
508: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](20,40)(20,0)
509: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](0,20)(40,20)
510: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
511: showorigin=false, labels=none](0,0)(50,50)
512: \pscircle*(10,30){2\psxunit} \pscircle*(20,40){2\psxunit}
513: \pscircle*(30,10){2\psxunit} \pscircle*(40,20){2\psxunit}
514: \end{pspicture}
515:
516: &\rule{10pt}{0pt}&
517:
518: \psset{xunit=0.02in, yunit=0.02in} \psset{linewidth=0.5\psxunit}
519: \begin{pspicture}(0,0)(50,50)
520: \psframe[linecolor=white,fillstyle=vlines,hatchangle=45,hatchcolor=gray](20,0)(50,30)
521: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](10,30)(50,30)
522: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](20,40)(20,0)
523: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
524: showorigin=false, labels=none](0,0)(50,50)
525: \pscircle*(10,30){2\psxunit} \pscircle*(20,40){2\psxunit}
526: \pscircle*(30,10){2\psxunit} \pscircle*(40,20){2\psxunit}
527: \end{pspicture}
528:
529: &\rule{10pt}{0pt}&
530:
531: \psset{xunit=0.02in, yunit=0.02in} \psset{linewidth=0.5\psxunit}
532: \begin{pspicture}(0,0)(50,50)
533: \psframe[linecolor=white,fillstyle=vlines,hatchangle=45,hatchcolor=gray](0,20)(30,50)
534: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](30,10)(30,50)
535: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](0,20)(40,20)
536: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
537: showorigin=false, labels=none](0,0)(50,50)
538: \pscircle*(10,30){2\psxunit} \pscircle*(20,40){2\psxunit}
539: \pscircle*(30,10){2\psxunit} \pscircle*(40,20){2\psxunit}
540: \end{pspicture}
541:
542: &\rule{10pt}{0pt}&
543:
544: \psset{xunit=0.02in, yunit=0.02in} \psset{linewidth=0.5\psxunit}
545: \begin{pspicture}(0,0)(50,50)
546: \psframe[linecolor=white,fillstyle=vlines,hatchangle=45,hatchcolor=gray](30,30)(50,50)
547: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](10,30)(50,30)
548: \psline[linecolor=gray,linestyle=dashed,linewidth=1\psxunit](30,10)(30,50)
549: \psaxes[dy=10, Dy=1, dx=10, Dx=1, tickstyle=bottom,
550: showorigin=false, labels=none](0,0)(50,50)
551: \pscircle*(10,30){2\psxunit} \pscircle*(20,40){2\psxunit}
552: \pscircle*(30,10){2\psxunit} \pscircle*(40,20){2\psxunit}
553: \end{pspicture}
554:
555: \\[4pt]
556:
557: (i)&&(ii)&&(iii)&&(iv)
558:
559: \end{tabular}
560: \end{center}
561: \caption{The regions of $\pi$ referred to in the proof of
562: Theorem~\ref{griddable-characterisation}.}\label{characterisation-fig}
563: \end{figure}
564:
565: By symmetry and the cases we have already handled, we may assume
566: that $a\ge 2$ and $b\ge 1$. Let $\pi\in\C_n$ be a
567: $3412$-containing permutation (if there are no such permutations,
568: then we are done by induction) and suppose that
569: $\pi(i_1)\pi(i_2)\pi(i_3)\pi(i_4)$ is in the same relative order
570: as $3412$ where $1\le i_1<i_2<i_3<i_4\le n$. By induction we have
571: the following (see Figure~\ref{characterisation-fig} for an
572: illustration of these regions):
573: \begin{enumerate}
574: \item[(i)] $\pi([i_2]\times[\pi(i_4)])$ avoids $\ominus^{a+1} 12$
575: and $\oplus^{b} 21$ so it can be covered by $f(a,b-1)$ monotonic
576: rectangles, \item[(ii)] $\pi([i_2,n]\times[\pi(i_1)])$ avoids
577: $\ominus^{a} 12$ and $\oplus^{b+1} 21$ so it can be covered by
578: $f(a-1,b)$ monotonic rectangles, \item[(iii)]
579: $\pi([i_3]\times[\pi(i_4),n])$ avoids $\ominus^{a} 12$ and
580: $\oplus^{b+1} 21$ so it can be covered by $f(a-1,b)$ monotonic
581: rectangles, and \item[(iv)] $\pi([i_3,n]\times[\pi(i_1),n])$
582: avoids $\ominus^{a+1} 12$ and $\oplus^{b} 21$ so it can be covered
583: by $f(a,b-1)$ monotonic rectangles.
584: \end{enumerate}
585: Because the four regions in (i)--(iv) cover $\pi$, it may be
586: covered by $2f(a-1,b)+2f(a,b-1)$ monotonic rectangles.
587: Furthermore, the $3412$-avoiding permutations in $\C$ may be
588: covered by $f(1,b)\le f(a-1,b)$ monotonic rectangles by induction,
589: so we may take $f(a,b)=2f(a-1,b)+2f(a,b-1)$, completing the proof.
590: \end{proof}
591:
592: %
593: %
594: %
595: %
596: %
597: %
598: %
599: %
600: %
601: %
602: %
603: %
604: %
605: \subsection{The enumeration of grid classes}
606:
607: To date only scattered results are known about the enumeration of
608: grid classes and their subclasses. The only general results are the following two.
609:
610: \begin{theorem}[Atkinson~\cite{a:rp}]\label{perm-mat-poly}
611: If $M$ is a permutation matrix, then $\Grid(M)$ and all its
612: subclasses have eventually polynomial enumeration.
613: \end{theorem}
614:
615: \begin{theorem}[Albert, Atkinson, and Ru\v{s}kuc~\cite{aar:regular}]\label{star-rational}
616: If $G(M)$ is a star, then $\Grid(M)$ and all its subclasses have
617: rational (and readily computable) generating functions.
618: \end{theorem}
619:
620: It is very tempting to speculate that the enumerative properties
621: of a grid class depend only on its graph\footnote{For example:
622: \begin{conjecture}\label{forest-rational}
623: If $G(M)$ is a forest then $\Grid(M)$ and all its subclasses have
624: rational generating functions.
625: \end{conjecture}}. Our contribution to
626: this suspicion is to show (in Theorem~\ref{grid-poly}) that when
627: $G(M)$ is a matching\footnote{We take a {\it matching\/} to be a
628: graph without incident edges, i.e., a graph with maximum
629: degree $1$.} then $\Grid(M)$ and all its subclasses have
630: eventually polynomial enumeration, thus generalising
631: Theorem~\ref{perm-mat-poly}. For brevity, we refer to such
632: classes as the {\it grid classes of matchings\/}.
633:
634: \begin{theorem}\label{grid-poly}
635: If the permutation class $\C$ lies in the grid class of a matching
636: then there is a polynomial $p(n)$ so that $|\C_n|=p(n)$ for all
637: sufficiently large $n$.
638: \end{theorem}
639: \begin{proof}
640: Let $M$ be a $t\times u$ matrix whose graph is a matching, let
641: $\C$ be a subclass of $\Grid(M)$, and let $\pi\in\C$. We define
642: the {\it greedy $M$-gridding\/} of $\pi$ to be the gridding given
643: by $1=c_1\le c_2\le\cdots\le c_{t+1}=n+1$ (the column divisions)
644: and $1=r_1\le r_2\le\cdots\le r_{u+1}=n+1$ (the row divisions)
645: where for each $k$, $c_k$ is chosen so as to maximise
646: $c_1+c_2+\cdots+c_k$. Because $G(M)$ is a matching, this uniquely
647: defines the $r$'s.
648:
649: \begin{figure}
650: \begin{center}
651: \psset{xunit=0.01in, yunit=0.01in} \psset{linewidth=1\psxunit}
652: \begin{pspicture}(0,0)(150,150)
653: \psline[linecolor=gray](15,0)(15,150)
654: \psline[linecolor=gray](85,0)(85,150)
655: \psline[linecolor=gray](105,0)(105,150)
656: \psline[linecolor=gray](0,25)(150,25)
657: \psline[linecolor=gray](0,95)(150,95)
658: \psline[linecolor=gray](0,105)(150,105) \psaxes[dy=10, Dy=1,
659: dx=10, Dx=1, tickstyle=bottom, showorigin=false,
660: labels=none](0,0)(140,140) \pscircle(10,100){4.0\psxunit}
661: \pscircle(20,90){4.0\psxunit} \pscircle*(30,80){4.0\psxunit}
662: \pscircle*(40,70){4.0\psxunit} \pscircle*(50,60){4.0\psxunit}
663: \pscircle*(60,50){4.0\psxunit} \pscircle*(70,40){4.0\psxunit}
664: \pscircle(80,30){4.0\psxunit} \pscircle(90,10){4.0\psxunit}
665: \pscircle(100,20){4.0\psxunit} \pscircle(110,140){4.0\psxunit}
666: \pscircle*(120,130){4.0\psxunit} \pscircle*(130,120){4.0\psxunit}
667: \pscircle(140,110){4.0\psxunit}
668: \end{pspicture}
669: \end{center}
670: \caption{A greedy gridding of a permutation, showing its peg
671: points as hollow circles; the peg permutation for this permutation
672: is $5431276$ while its non-peg vector is $(0,5,0,2)$. Note that
673: since this is a greedy gridding, the $(1,3)$ entry of the
674: corresponding matrix must be $1$.} \label{fig-pegs-example}
675: \end{figure}
676:
677: We define a {\it peg point\/} of $\pi$ to be a point which is
678: either first or last (either horizontally or vertically; since the
679: blocks are monotone, it doesn't matter) in its block in the greedy
680: $M$-gridding of $\pi$. An example is shown in
681: Figure~\ref{fig-pegs-example}. The {\it peg permutation\/},
682: $\rho^\pi$, of $\pi$ is then the permutation formed by its peg
683: points. We also associate to each permutation $\pi\in\C$ its {\it
684: non-peg vector\/} $\y^\pi=(y_1,y_2,\dots,y_t)$, where $y_i$
685: denotes the number of non-peg points in
686: $\pi([c_i,c_{i+1})\times[n])$. Because the $M$-gridding was
687: chosen greedily, the pair $(\rho^\pi, \y^\pi)$ uniquely determines
688: $\pi$.
689:
690: We now partition the class $\C$ based upon peg permutations. Since
691: there can be at most $3^t$ different peg permutations of members
692: of $\C$ (for every column of $M$ a peg permutation can have $0$,
693: $1$, or $2$ elements), this is a partition into a finite number of
694: subsets. Let $\C^\rho$ denote the subset of $\C$ with peg
695: permutation $\rho$. This is not a permutation class (the peg
696: permutation of $\sigma\le\pi$ need not be the peg permutation of
697: $\pi$), but the set of non-peg vectors of permutations in this
698: class, $\{\y^\pi : \pi\in\C^\rho\}$, is a downset of vectors
699: in $\mathbb{N}^t$. Therefore Theorem~\ref{vector-poly} shows that
700: $\C^\rho$ has eventually polynomial enumeration, and so $\C$ does
701: as well.
702: \end{proof}
703:
704:
705:
706:
707:
708: %
709: %
710: %
711: %
712: %
713: %
714: %
715: %
716: %
717: %
718: %
719: %
720: %
721: \section{The Fibonacci dichotomy}\label{polynomials}
722:
723: The Fibonacci dichotomy for permutation classes, first proved by
724: Kaiser and Klazar~\cite{kk:growth}, states that all sub-Fibonacci
725: permutation classes\footnote{We call a class $\C$ {\it sub-Fibonacci\/}
726: if $|\C_n|$ is strictly less than the $n$th Fibonacci number for some $n$.
727: The definition of sub-$2^{n-1}$ is analogous.}
728: have eventually
729: polynomial enumeration.
730: Here we give a new proof using the
731: characterisation of grid classes.
732: We have already shown, in
733: Theorem~\ref{grid-poly}, that grid classes of matchings and their
734: subclasses have eventually polynomial enumeration. It remains
735: only to show that all sub-Fibonacci classes lie in grid classes of
736: matchings. We do this in two parts. First we observe in
737: Proposition~\ref{sub-fib-griddable} that all sub-Fibonacci classes
738: are griddable, and then we show in
739: Proposition~\ref{interleavings-matchings} that all sub-$2^{n-1}$
740: griddable classes (which includes sub-Fibonacci classes) lie in
741: grid classes of matchings.
742:
743:
744: \begin{proposition}\label{sub-fib-griddable}
745: All sub-Fibonacci classes are griddable.
746: \end{proposition}
747: \begin{proof}
748: Let $\C$ denote a non-griddable class, so, by
749: Theorem~\ref{griddable-characterisation} and symmetry we may
750: assume that $\C$ contains arbitrarily long direct sums of $21$.
751: Since $\C$ is a permutation class, it must also contain every
752: permutation that embeds into an arbitarily long direct sum of
753: $21$. These permutations have the form
754: $\sigma_1\oplus\sigma_2\oplus\cdots\oplus\sigma_k$ where each
755: $\sigma_i$ is either $1$ or $21$. Thus there are precisely as
756: many permutations of this form of length $n$ as there are ways of
757: writing $n$ as an ordered sum of $1$'s and $2$'s, of which there
758: are $F_n$.
759: \end{proof}
760:
761: \begin{figure}
762: \begin{center}
763: \begin{tabular}{ccc}
764: \psset{xunit=0.01in, yunit=0.01in} \psset{linewidth=1\psxunit}
765: \begin{pspicture}(0,0)(150,150)
766: \psline[linecolor=gray](0,75)(150,75) \psaxes[dy=10, Dy=1, dx=10,
767: Dx=1, tickstyle=bottom, showorigin=false,
768: labels=none](0,0)(140,140) \pscircle*(10,70){4.0\psxunit}
769: \pscircle*(20,100){4.0\psxunit} \pscircle*(30,30){4.0\psxunit}
770: \pscircle*(40,120){4.0\psxunit} \pscircle*(50,50){4.0\psxunit}
771: \pscircle*(60,80){4.0\psxunit} \pscircle*(70,60){4.0\psxunit}
772: \pscircle*(80,140){4.0\psxunit} \pscircle*(90,10){4.0\psxunit}
773: \pscircle*(100,110){4.0\psxunit} \pscircle*(110,40){4.0\psxunit}
774: \pscircle*(120,130){4.0\psxunit} \pscircle*(130,20){4.0\psxunit}
775: \pscircle*(140,90){4.0\psxunit}
776: \end{pspicture}
777: & \rule{20pt}{0pt} & \psset{xunit=0.01in, yunit=0.01in}
778: \psset{linewidth=1\psxunit}
779: \begin{pspicture}(0,0)(150,150)
780: \psline[linecolor=gray](75,0)(75,150) \psaxes[dy=10, Dy=1, dx=10,
781: Dx=1, tickstyle=bottom, showorigin=false,
782: labels=none](0,0)(140,140) \pscircle*(10,90){4.0\psxunit}
783: \pscircle*(20,130){4.0\psxunit} \pscircle*(30,30){4.0\psxunit}
784: \pscircle*(40,110){4.0\psxunit} \pscircle*(50,50){4.0\psxunit}
785: \pscircle*(60,70){4.0\psxunit} \pscircle*(70,10){4.0\psxunit}
786: \pscircle*(80,60){4.0\psxunit} \pscircle*(90,140){4.0\psxunit}
787: \pscircle*(100,20){4.0\psxunit} \pscircle*(110,100){4.0\psxunit}
788: \pscircle*(120,40){4.0\psxunit} \pscircle*(130,120){4.0\psxunit}
789: \pscircle*(140,80){4.0\psxunit}
790: \end{pspicture}
791: \end{tabular}
792: \end{center}
793: \caption{A vertical alternation (left) and its inverse, a
794: horizontal alternation (right).}\label{fig-alternations}
795: \end{figure}
796:
797: We define a {\it vertical alternation\/} to be a permutation
798: $\pi\in S_{2n}$ so that either
799: $$
800: \pi(1), \pi(3), \dots, \pi(2n-1)<\pi(2), \pi(4),\dots, \pi(2n)
801: $$
802: or
803: $$
804: \pi(1), \pi(3), \dots, \pi(2n-1)>\pi(2), \pi(4),\dots, \pi(2n).
805: $$
806: A {\it horizontal alternation\/} is then the group-theoretic
807: inverse of a vertical alternation. Examples are shown in
808: Figure~\ref{fig-alternations}. We begin by observing that classes
809: with arbitrarily long alternations are not small.
810:
811: \begin{proposition}\label{alternations-not-small}
812: If the permutation class $\C$ contains arbitrarily long
813: alternations, then $|\C_n|\ge 2^{n-1}$ for all $n$.
814: \end{proposition}
815: \begin{proof}
816: By symmetry, let us suppose that $\C$ contains arbitrarily long
817: horizontal alternations. By the Erd\H{o}s-Szekeres Theorem, $\C$
818: contains arbitrarily long horizontal alternations in which both
819: sides are monotone. Therefore $\C$ contains either $\Grid(1\ 1)$,
820: $\Grid(1\ -1)$, $\Grid(-1\ 1)$, or $\Grid(-1\ -1)$. It is easy
821: to compute that the first and last of these classes contain
822: $2^n-n$ permutations of length $n$ for $n\ge 1$ while the second
823: and third contain $2^{n-1}$ permutations of length $n\ge 1$,
824: establishing the proposition.
825: \end{proof}
826:
827:
828: Therefore a sub-Fibonacci class cannot contain arbitrarily long
829: alternations. We now prove that this implies that these
830: classes lie in grid classes of matchings.
831:
832: We say that a set of indices $\{i_1,i_2,\dots,i_s\}$ in $\pi$ is
833: an {\it uninterrupted monotone interval\/} if $|i_{j+1}-i_j|=1$
834: and $|\pi(i_{j+1})-\pi(i_j)|=1$ for all $j\in[s-1]$. Note that if
835: $G(M)$ is a matching, then an $M$-gridding of $\pi$ is a division
836: of the elements of $\pi$ into uninterrupted monotone intervals.
837: Conversely, every division of $\pi$ into uninterrupted monotone
838: intervals gives an $M$-gridding of $\pi$ for some $M$ where $G(M)$
839: is a matching.
840:
841: \begin{proposition}\label{interleavings-matchings}
842: A griddable class lies in the grid class of a matching if and only
843: if it does not contain arbitrarily long alternations.
844: \end{proposition}
845: \begin{proof}
846: One direction is obvious: if a permutation class contains
847: arbitrarily long alternations then it cannot lie in the grid class
848: of a matching. The other direction is almost as clear, but a
849: formal proof takes a modest amount of effort.
850:
851: Let $\C\subseteq\Grid(N)$ for some $t\times u$ matrix $N$, and
852: suppose that $\C$ does not contain any alternations (either
853: horizontal or vertical) with more than $d$ elements. It suffices
854: to show that there is a constant $m$ such that every permutation
855: $\pi\in\C$ lies in $\Grid(M)$ where $G(M)$ is a matching and $M$
856: (which we allow to depend on $\pi$) has at most $m$ nonzero
857: entries. This is because we can ignore the all-$0$ rows and columns,
858: so the size of $M$ can be bounded, and then there are only
859: finitely many such matrices, so $\C$ will lie in the grid class of
860: their direct sum (which also has a matching for its graph).
861: Equivalently, by our remarks above, it suffices to show that every
862: permutation in $\C$ can be divided into a bounded number of
863: uninterrupted monotone intervals.
864:
865: To this end, take some permutation $\pi\in\C$ of length $n$ with
866: $N$-gridding given by $1=c_1\le c_2\le\cdots\le c_{t+1}=n+1$ and
867: $1=r_1\le r_2\le\cdots\le r_{u+1}=n+1$ and consider a particular
868: block in this gridding, say
869: $$
870: \pi^{(k,\ell)}:= \pi([c_k,c_{k+1})\times[r_\ell,r_{\ell+1})).
871: $$
872: We consider four types of alternations that elements of this
873: block can participate in: vertical alternations either with blocks
874: of the form $\pi^{(k,\ell^+)}$ for $\ell^+>\ell$ or of the form
875: $\pi^{(k,\ell^-)}$ for $\ell^-<\ell$, and horizontal alternations
876: with blocks of the form $\pi^{(k^+,\ell)}$ for $k^+>k$ or of the
877: form $\pi^{(k^-,\ell)}$ for $k^-<k$.
878:
879: Every time that two consecutive elements in a block are separated
880: either horizontally or vertically (that is, every time that two
881: consecutive elements in a block fail to lie in an uninterrupted
882: monotone interval together) then they contribute to the length of
883: at least one of these four alternations. Therefore, at most $4d$
884: such separations can occur, so $\pi^{(k,\ell)}$ can be divided
885: into at most $4d+1$ uninterrupted monotone intervals. Hence $\pi$
886: itself can be divided into at most $(4d+1)tu$ uninterrupted
887: monotone intervals, proving the proposition.
888: \end{proof}
889:
890: Having established that sub-$2^{n-1}$ griddable classes (and in
891: particular, sub-Fibonacci classes) lie in grid classes of
892: matchings, we now have another proof of the Fibonacci dichotomy:
893:
894: \begin{corollary}\label{dichotomy}
895: For every permutation class $\C$, one of the following occurs:
896: \begin{itemize}
897: \item $|\C_n|\ge F_n$ for all $n$, or \item $\C$ lies in the grid
898: class of a matching and there is a polynomial $p(n)$ so that
899: $|\C_n|=p(n)$ for all sufficiently large $n$.
900: \end{itemize}
901: \end{corollary}
902:
903: %
904: %
905: %
906: %
907: %
908: %
909: %
910: %
911:
912: \section{Concluding remarks}
913:
914: \noindent{\bf Decidability.} It is not hard to see that the hypotheses of our characterisation theorems are decidable from the basis of a finitely based class. For example, in order to determine if $\Av(B)$ contains arbitrarily long direct sums of $12$ one needs only check if any element of $B$ lies in $\Av(231,312,321)$, which is the set of permutations that are contained in arbitrarily long direct sums of $12$. Thus we have the following result.
915:
916: \begin{corollary}
917: Given a finite set of permutations $B$, it is decidable whether or not $\Av(B)$ is griddable.
918: \end{corollary}
919:
920: Similar arguments show that polynomial enumeration is decidable for finitely based classes. One first needs to check whether the class is griddable and then decide whether the class contains arbitrarily long alternations.
921:
922: \begin{corollary}
923: Given a finite set of permutations $B$, it is decidable whether or not $|\Av_n(B)|$ agrees with a polynomial for all sufficiently large $n$.
924: \end{corollary}
925:
926: \bigskip\noindent{\bf Finite bases.} The decidability results above only apply to finitely based classes, however, it happens that permutation classes with polynomial enumeration must be finitely based. Because these classes lie in grid classes of matchings, they also lie in grid classes of $\zpm$ row vectors. Now one needs only to apply the result of Atkinson, Murphy, and Ru\v{s}kuc~\cite{amr:pwocsop} that every subclass of the grid class of a $\zpm$ row vector is finitely based.
927:
928: \bigskip\noindent{\bf Enumeration.} While Corollary~\ref{dichotomy} characterises the sub-Fibonacci
929: classes and shows that they have eventually polynomial
930: enumeration, it does not address the issue of how one might find
931: these formulas. This could presumably be settled
932: by strengthening the results given here to obtain bounds
933: (computable from the basis of $\C$) on the degree of the
934: polynomial and the values of $n$ for which $|\C_n|$ agrees with
935: this polynomial, but there are already three general methods which can be used to count these classes:
936: \begin{enumerate}
937: \item[(i)] Since permutation classes with polynomial growth lie in grid classes of matchings, they also lie in grid classes of $\zpm$ row vectors. One can therefore use Theorem~\ref{star-rational} to enumerate them.
938: \item[(ii)] Permutation classes with polynomial growth contain only finitely many simple permutations, and so the results of Albert and Atkinson~\cite{aa:simple:alg} apply to them.
939: \item[(iii)] Permutation classes with polynomial growth correspond to regular languages via the insertion encoding of Albert, Linton, and Ru\v{s}kuc~\cite{insertion}, so this method can also be used.
940: \end{enumerate}
941:
942: \bigskip\noindent{\bf Classes with two basis elements.} Albert, Atkinson, and Brignall~\cite{aab:poly} have recently studied doubleton-based classes with polynomial enumeration. They give a characterisation of the pairs of permutations $\beta_1$ and $\beta_2$ for which $\Av(\beta_1,\beta_2)$ has polynomial enumeration and give bounds on the degree of this polynomial.
943:
944: \bigskip
945:
946: \noindent{\it Acknowledgment.} We thank Nik Ru\v{s}kuc and Bruce Sagan for their helpful comments.
947:
948: \bibliographystyle{acm}
949: \begin{small}
950: \bibliography{refs}
951: \end{small}
952:
953: \end{document}
954: