1: \documentclass[a4paper,11pt,reqno]{amsart}
2: \usepackage{graphicx,latexsym,amssymb,xypic}
3:
4:
5: \newcommand{\imply}{\,\Rightarrow\,}
6: \renewcommand{\iff}{\,\Leftrightarrow\,}
7: \newcommand{\sym}{\mathcal{S}}
8: \newcommand{\N}{{\mathbb N}}
9: \newcommand{\Z}{{\mathbb Z}}
10: \newcommand{\Q}{{\mathbb Q}}
11: \newcommand{\R}{{\mathbb R}}
12: \newcommand{\C}{{\mathbb C}}
13: \newcommand{\ns}{\hspace{-0.2ex}} %Very thin negative space.
14:
15: \renewcommand{\theequation}{\arabic{equation}}
16:
17: %\swapnumbers
18: \theoremstyle{plain}
19: \newtheorem{theorem}{Theorem}
20: \newtheorem*{theorem*}{Theorem}
21: \newtheorem{corollary}[theorem]{Corollary}
22: \newtheorem*{corollary*}{Corollary}
23: \newtheorem{porism}[theorem]{Porism}
24: \newtheorem*{porism*}{Porism}
25: \newtheorem{lemma}[theorem]{Lemma}
26: \newtheorem*{lemma*}{Lemma}
27: \newtheorem{proposition}[theorem]{Proposition}
28: \newtheorem*{proposition*}{Proposition}
29:
30: \theoremstyle{definition}
31: \newtheorem{definition}[theorem]{Definition}
32: \newtheorem*{definition*}{Definition}
33: \newtheorem{example}[theorem]{Example}
34: \newtheorem*{example*}{Example}
35:
36: \theoremstyle{remark}
37: \newtheorem{remark}{Remark}
38: \newtheorem*{remark*}{Remark}
39:
40:
41: \renewcommand{\theenumi}{\Roman{enumi}}
42:
43: \newcommand{\wpi}{\widehat\pi}
44: \newcommand{\emptyword}{{\epsilon}}
45:
46: \DeclareMathOperator{\proj}{\mathrm{proj}}
47: \DeclareMathOperator{\des}{des}
48:
49: \newcommand{\nop}{\mathrm{NOP}}
50: \newcommand{\mono}{\mathcal{M}}
51:
52:
53: % ### BEGIN PATTERNS
54: \def\dd{\hspace{1pt}\mbox{\textup{\text{-}\makebox[0pt]{\text{-}}}}\hspace{2pt}}
55:
56: \def\axb{\ensuremath{a{\dd}b}}
57: \def\bxa{\ensuremath{b{\dd}a}}
58: \def\ab{\ensuremath{ab}}
59: \def\ba{\ensuremath{ba}}
60:
61: \def\axbc{\ensuremath{a{\dd}bc}}
62: \def\axcb{\ensuremath{a{\dd}cb}}
63: \def\bxac{\ensuremath{b{\dd}ac}}
64: \def\bxca{\ensuremath{b{\dd}ca}}
65: \def\cxab{\ensuremath{c{\dd}ab}}
66: \def\cxba{\ensuremath{c{\dd}ba}}
67:
68: \def\abxc{\ensuremath{ab{\dd}c}}
69: \def\acxb{\ensuremath{ac{\dd}b}}
70: \def\baxc{\ensuremath{ba{\dd}c}}
71: \def\bcxa{\ensuremath{bc{\dd}a}}
72: \def\caxb{\ensuremath{ca{\dd}b}}
73: \def\cbxa{\ensuremath{cb{\dd}a}}
74:
75: \def\axbxc{\ensuremath{a{\dd}b{\dd}c}}
76: \def\axcxb{\ensuremath{a{\dd}c{\dd}b}}
77: \def\bxaxc{\ensuremath{b{\dd}a{\dd}c}}
78: \def\bxcxa{\ensuremath{b{\dd}c{\dd}a}}
79: \def\cxaxb{\ensuremath{c{\dd}a{\dd}b}}
80: \def\cxbxa{\ensuremath{c{\dd}b{\dd}a}}
81:
82: \def\cba{\ensuremath{cba}}
83:
84: \def\caxdb{\ensuremath{ca{\dd}db}}
85: \def\cbxda{\ensuremath{cb{\dd}da}}
86: \def\abxdc{\ensuremath{ab{\dd}dc}}
87: \def\baxdc{\ensuremath{ba{\dd}dc}}
88: \def\dcxab{\ensuremath{dc{\dd}ab}}
89: \def\dcxba{\ensuremath{dc{\dd}ba}}
90:
91: \def\axbxcxd{\ensuremath{a{\dd}b{\dd}c{\dd}d}}
92: % ### END PATTERNS
93:
94:
95: % ### BEGIN FIGS
96: \newcommand{\mA}{\ensuremath{\bar{0}}}
97: \newcommand{\mB}{\ensuremath{\bar{1}}}
98: \newcommand{\mC}{\ensuremath{\bar{2}}}
99: \newcommand{\mD}{\ensuremath{\bar{3}}}
100: \newcommand{\mE}{\ensuremath{\bar{4}}}
101: \newcommand{\mF}{\ensuremath{\bar{5}}}
102: \newcommand{\mG}{\ensuremath{\bar{6}}}
103: \newcommand{\mH}{\ensuremath{\bar{7}}}
104: \newcommand{\mI}{\ensuremath{\bar{8}}}
105: \newcommand{\mJ}{\ensuremath{\bar{9}}}
106:
107: \newcommand{\size}{1.1ex} \newcommand{\x}[1]{\makebox[\size]{$#1$}}
108: \newcommand{\xxx}{\makebox[\size]{ }}
109: \newcommand{\SLA}{\makebox[\size]{/}}
110: \newcommand{\BAC}{\makebox[\size]{\ensuremath{\backslash}}}
111: \newcommand{\CIR}{\makebox[\size]{\ensuremath{\circ}}}
112: \newcommand{\DOT}{\makebox[\size]{\ensuremath{\bullet}}}
113: \newcommand{\DAS}{\makebox[\size]{\ensuremath{-}}}
114: \newcommand{\uxx}{\makebox[\size]{\ensuremath{u}}}
115: \newcommand{\dxx}{\makebox[\size]{\ensuremath{d}}}
116: \newcommand{\ONE}{\makebox[\size]{1}}
117: \newcommand{\TWO}{\makebox[\size]{2}}
118: \newcommand{\THR}{\makebox[\size]{3}}
119: \newcommand{\FOU}{\makebox[\size]{4}}
120: \newcommand{\FIV}{\makebox[\size]{5}}
121: \newcommand{\SIX}{\makebox[\size]{6}}
122: \newcommand{\SEV}{\makebox[\size]{7}}
123: \newcommand{\EIG}{\makebox[\size]{8}}
124: \newcommand{\NIN}{\makebox[\size]{9}}
125: \newcommand{\Axx}{\makebox[\size]{10}}
126: \newcommand{\Bxx}{\makebox[\size]{11}}
127: \newcommand{\Cxx}{\makebox[\size]{12}}
128: \newcommand{\Dxx}{\makebox[\size]{13}}
129: \newcommand{\Exx}{\makebox[\size]{14}}
130: \newcommand{\Fxx}{\makebox[\size]{15}}
131: \newcommand{\Gxx}{\makebox[\size]{16}}
132: \newcommand{\Hxx}{\makebox[\size]{17}}
133: \newcommand{\Ixx}{\makebox[\size]{18}}
134: \newcommand{\Jxx}{\makebox[\size]{19}}
135:
136: \newcommand{\FIGmono}{
137: \begin{smallmatrix}
138: \xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\CIR\xxx\xxx\CIR\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\CIR\\
139: \xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\CIR\DAS\DAS\DAS\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\xxx\xxx\xxx\\
140: \xxx\xxx\xxx\xxx\xxx\xxx\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\\
141: \CIR\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\CIR\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\\
142: \ONE\xxx\xxx\TWO\xxx\xxx\THR\xxx\xxx\FOU\xxx\xxx\FIV\xxx\xxx\SIX\xxx\xxx\SEV\xxx\xxx\EIG\xxx\xxx\NIN\xxx\xxx\Axx\xxx\xxx\Bxx\xxx\xxx\Cxx\xxx\xxx\Dxx\\
143: \end{smallmatrix}
144: }
145:
146: \newcommand{\FIGnop}{
147: \begin{smallmatrix}
148: \xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\CIR\DAS\DAS\DAS\DAS\DAS\CIR\DAS\DAS\CIR\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\xxx\\
149: \xxx\xxx\xxx\xxx\xxx\xxx\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\xxx\xxx\CIR\xxx\xxx\CIR\DAS\DAS\CIR\DAS\DAS\CIR\xxx\xxx\xxx\\
150: \CIR\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\DAS\CIR\\
151: \ONE\xxx\xxx\TWO\xxx\xxx\THR\xxx\xxx\FOU\xxx\xxx\FIV\xxx\xxx\SIX\xxx\xxx\SEV\xxx\xxx\EIG\xxx\xxx\NIN\xxx\xxx\Axx\xxx\xxx\Bxx\xxx\xxx\Cxx\xxx\xxx\Dxx\\
152: \end{smallmatrix}
153: }
154:
155: % ### END FIGS
156:
157:
158:
159: \begin{document}
160:
161: \title{Generalised pattern avoidance}
162:
163: \author{Anders Claesson}
164: \address{Matematik\\
165: Chalmers tekniska h\"ogskola och G\"oteborgs universitet\\
166: S-412 96 G\"oteborg, Sweden} \email{claesson@math.chalmers.se}
167: %\thanks{thanks}
168: %\keywords{keywords}
169: %\subjclass{Primary: subject; Secondary: subject}
170: \date{\today}
171:
172: \begin{abstract}
173: Recently, Babson and Steingr\'{\i}msson have introduced generalised
174: permutation patterns that allow the requirement that two adjacent
175: letters in a pattern must be adjacent in the permutation. We
176: consider pattern avoidance for such patterns, and give a complete
177: solution for the number of permutations avoiding any single pattern
178: of length three with exactly one adjacent pair of letters. We also
179: give some results for the number of permutations avoiding two
180: different patterns. Relations are exhibited to several well studied
181: combinatorial structures, such as set partitions, Dyck paths,
182: Motzkin paths, and involutions. Furthermore, a new class of set
183: partitions, called monotone partitions, is defined and shown to be
184: in one-to-one correspondence with non-overlapping partitions.
185: \end{abstract}
186:
187:
188: \maketitle\thispagestyle{empty}
189:
190:
191: \section{Introduction}
192:
193: In the last decade a wealth of articles has been written on the
194: subject of pattern avoidance, also known as the study of ``restricted
195: permutations'' and ``permutations with forbidden subsequences''.
196: Classically, a pattern is a permutation $\sigma\in\sym_k$, and a
197: permutation $\pi\in\sym_n$ avoids $\sigma$ if there is no subsequence
198: in $\pi$ whose letters are in the same relative order as the letters
199: of $\sigma$. For example, $\pi\in\sym_n$ avoids $132$ if there is no
200: $1\leq i\leq j\leq k\leq n$ such that $\pi(i)\leq\pi(k)\leq\pi(j)$.
201: In \cite{Kn73v1} Knuth established that for all $\sigma\in\sym_3$,
202: the number of permutations in $\sym_n$ avoiding $\sigma$ equals the
203: $n$th Catalan number, $C_n = \frac{1}{1+n}\binom{2n}{n}$. One may also
204: consider permutations that are required to avoid several patterns. In
205: \cite{SiSc85} Simion and Schmidt gave a complete solution for
206: permutations avoiding any set of patterns of length three. Even
207: patterns of length greater than three have been considered. For
208: instance, West showed in \cite{We95} that permutations avoiding
209: both $3142$ and $2413$ are enumerated by the Shr\"oder numbers, $S_n =
210: \sum_{i=0}^n\binom {2n-i} i C_{n-i}$.
211:
212: In \cite{BaSt00} Babson and Steingr\'{\i}msson introduced generalised
213: permutation patterns that allow the requirement that two adjacent
214: letters in a pattern must be adjacent in the permutation. The
215: motivation for Babson and Steingr\'{\i}msson in introducing these
216: patterns was the study of Mahonian statistics, and they showed that
217: essentially all Mahonian permutation statistics in the literature can
218: be written as linear combinations of such patterns. An example of a
219: generalised pattern is $(\axcb)$. An $(\axcb)$-subword of a
220: permutation $\pi = a_1 a_2 \cdots a_n$ is a subword $a_i a_j a_{j+1}$,
221: $(i<j)$, such that $a_i<a_{j+1}<a_j$. More generally, a pattern $p$ is
222: a word over the alphabet $a<b<c<d\cdots$ where two adjacent letters
223: may or may not be separated by a dash. The absence of a dash between
224: two adjacent letters in a $p$ indicates that the corresponding letters
225: in a $p$-subword of a permutation must be adjacent. Also, the ordering
226: of the letters in the $p$-subword must match the ordering of the
227: letters in the pattern. This definition, as well as any other
228: definition in the introduction, will be stated rigorously in
229: Section~\ref{prel}. All classical patterns are generalised patterns
230: where each pair of adjacent letters is separated by a dash. For
231: example, the generalised pattern equivalent to $132$ is $(\axcxb)$.
232:
233: We extend the notion of pattern avoidance by defining that a
234: permutation avoids a (generalised) pattern $p$ if it does not contain
235: any $p$-subwords. We show that this is a fruitful extension, by
236: establishing connections to other well known combinatorial structures,
237: not previously shown to be related to pattern avoidance. The main
238: results are given below.
239: \vspace{.3ex}
240: $$
241: \vspace{.3ex}
242: \begin{array}{l|l|l}
243: P & |\sym_n(P)| & Description\\ \hline
244: \axbc & B_n & \text{ Partitions of } [n] \\
245: \axcb & B_n & \text{ Partitions of } [n] \\
246: \bxac & C_n & \text{ Dyck paths of length } 2n \\
247: \axbc,\,\abxc & B^*_n & \text{ Non-overlapping partitions of } [n] \\
248: \axbc,\,\axcb & I_n & \text{ Involutions in } \sym_n \\
249: \axbc,\,\acxb & M_n & \text{ Motzkin paths of length } n \\
250: \end{array}
251: $$
252: Here $\sym_n(P) = \{ \pi\in\sym_n : \pi \text{ avoids } p \text{ for
253: all } p\in P \}$, and $[n]=\{1,2,\ldots,n\}$. When proving that
254: $|\sym_n(\axbc,\,\abxc)| = B^*_n$ (the $n$th Bessel number), we first
255: prove that there is a one-to-one correspondence between
256: $\{\axbc,\abxc\}$-avoiding permutations and \emph{monotone
257: partitions}. A partition is monotone if its non-singleton blocks can
258: be written in increasing order of their least element and increasing
259: order of their greatest element, simultaneously. This new class of
260: partitions is then shown to be in one-to-one correspondence with
261: non-overlapping partitions.
262:
263:
264: \section{Preliminaries}\label{prel}
265:
266: By an \emph{alphabet} $X$ we mean a non-empty set. An element of $X$
267: is called a \emph{letter}. A \emph{word} over $X$ is a finite sequence
268: of letters from $X$. We consider also the \emph{empty word}, that is,
269: the word with no letters; it is denoted by $\emptyword$. Let
270: $x=x_1x_2\cdots x_n$ be a word over $X$. We call $|x|:=n$ the
271: \emph{length} of $x$. A \emph{subword} of $x$ is a word
272: $v=x_{i_1}x_{i_2}\cdots x_{i_k}$, where $1 \leq i_1<i_2<\cdots<i_k
273: \leq n$. A \emph{segment} of $x$ is a word $v=x_{i}x_{i+1}\cdots
274: x_{i+k}$. If $X$ and $Y$ are two linearly ordered alphabets, then two
275: words $x=x_1x_2 \cdots x_n$ and $y = y_1 y_2 \cdots y_n$ over $X$ and
276: $Y$, respectively, are said to be \emph{order equivalent} if $x_i<x_j$
277: precisely when $y_i < y_j$.
278:
279: Let $X = A \cup \{\dd\}$ where $A$ is a linearly ordered alphabet. For
280: each word $x$ let $\bar x$ be the word obtained from $x$ by deleting
281: all dashes in $x$. A word $p$ over $X$ is called a \emph{pattern} if
282: it contains no two consecutive dashes and $\bar{p}$ has no repeated
283: letters. By slight abuse of terminology we refer to the \emph{length
284: of a pattern} $p$ as the length of $\bar{p}$. Two patterns $p$ and
285: $q$ of equal length are said to be \emph{dash equivalent} if the $i$th
286: letter in $p$ is a dash precisely when the $i$th letter in $q$ is a
287: dash. If $p$ and $q$ are dash and order equivalent, then $p$ and
288: $q$ are \emph{equivalent}. In what follows a pattern will usually be
289: taken to be over the alphabet $\{a,b,c,d,\ldots\}\cup\{\dd\}$ where
290: $\{a,b,c,d,\ldots\}$ is ordered so that $a<b<c<d<\cdots$.
291:
292: Let $[n]:=\{1,2,\ldots,n\}$ (so $[0]=\emptyset$). A \emph{permutation}
293: of $[n]$ is bijection from $[n]$ to $[n]$. Let $\sym_n$ be the set of
294: permutations of $[n]$. We shall usually think of a permutation $\pi$
295: as the word $\pi(1)\pi(2)\cdots\pi(n)$ over the alphabet $[n]$. In
296: particular, $\sym_0 = \{\emptyword\}$, since there is only one
297: bijection from $\emptyset$ to $\emptyset$, the empty map. We say that
298: a subword $\sigma$ of $\pi$ is a $p$-\emph{subword} if by replacing
299: (possibly empty) segments of $\pi$ with dashes we can obtain a pattern
300: $q$ equivalent to $p$ such that $\bar q = \sigma$. However, all
301: patterns that we will consider will have a dash at the beginning and
302: one at the end. For convenience, we therefore leave them out. For
303: example, $(\axbc)$ is a pattern, and the permutation $491273865$
304: contains three $(\axbc)$-subwords, namely $127$, $138$, and $238$. A
305: permutation is said to be \emph{$p$-avoiding} if it does not contain
306: any $p$-subwords. Define $\sym_n(p)$ to be the set of $p$-avoiding
307: permutations in $\sym_n$ and, more generally, $\sym_n(A) =
308: \bigcap_{p\in A} \sym_n(p)$.
309:
310: We may think of a pattern $p$ as a permutation statistic, that is,
311: define $p\,\pi$ as the number of $p$-subwords in $\pi$, thus regarding
312: $p$ as a function from $\sym_n$ to $\N$. For example,
313: $(\axbc)\,491273865 = 3$. In particular, $\pi$ is $p$-avoiding if and
314: only if $p\,\pi = 0$. We say that two permutation statistics
315: $\mathrm{stat}$ and $\mathrm{stat}'$ are \emph{equidistributed} over
316: $A\subseteq\sym_n$, if
317: $$
318: \sum_{\pi\in A}x^{\mathrm{stat}\, \pi} = \sum_{\pi\in
319: A}x^{\mathrm{stat}'\, \pi}.
320: $$
321: In particular, this definition applies to patterns.
322:
323: Let $\pi=a_1 a_2 \cdots a_n \in \sym_n$. An $i$ such that
324: $a_i>a_{i+1}$ is called a \emph{descent} in $\pi$. We denote by $\des
325: \pi$ the number of descents in $\pi$. Observe that $\des$ can be
326: defined as the pattern $(ba)$, that is, $\des\pi = (ba)\pi$. A
327: \emph{left-to-right minimum} of $\pi$ is an element $a_i$ such that
328: $a_i < a_j$ for every $j<i$. The number of left-to-right minima is a
329: permutation statistic. Analogously we also define \emph{left-to-right
330: maximum}, \emph{right-to-left minimum}, and \emph{right-to-left
331: maximum}.
332:
333: In this paper we will relate permutations avoiding a given set of
334: patterns to other better known combinatorial structures. Here follows
335: a brief description of these structures.
336:
337: \subsection*{Set partitions}
338:
339: A \emph{partition} of a set $S$ is a family, $\pi = \{A_1 , A_2,
340: \ldots, A_k \}$, of pairwise disjoint non-empty subsets of $S$ such
341: that $S=\cup_i A_i$. We call $A_i$ a \emph{block} of $\pi$. The total
342: number of partitions of $[n]$ is called a \emph{Bell number} and is
343: denoted $B_n$. For reference, the first few Bell numbers are
344: $$
345: 1,1,2,5,15,52,203,877,4140,21147,115975,678570,4213597. $$
346: Let
347: $S(n,k)$ be the number of partitions of $[n]$ into $k$ blocks; these
348: numbers are called the \emph{Stirling numbers of the second kind}.
349:
350: \subsection*{Non-overlapping partitions}
351:
352: Two blocks $A$ and $B$ of a partition $\pi$ \emph{overlap} if
353: $$\min A < \min B < \max A < \max B.$$
354: A partition is
355: \emph{non-overlapping} if no pairs of blocks overlap.
356: Thus
357: $$\pi = \{\{ 1,2,5,13\},\{3,8\},\{4,6,7\},\{9\},\{10,11,12\}\}$$
358: is non-overlapping. A pictorial representation of
359: $\pi$ is
360: $$\pi \,=\, \FIGnop .$$
361: Let $B^*_n$ be
362: the number of non-overlapping partitions of $[n]$; this number is
363: called the $n$th \emph{Bessel number} \cite[p. 423]{FlSc90}. The
364: first few Bessel numbers are
365: $$
366: 1,1,2,5,14,43,143,509,1922,7651,31965,139685,636712. $$
367: We denote
368: by $S^*(n,k)$ the number of non-overlapping partitions of $[n]$ into
369: $k$ blocks.
370:
371: \subsection*{Involutions}
372:
373: An \emph{involution} is a permutation which is its own inverse. We
374: denote by $I_n$ the number of involutions in $\sym_n$. The sequence
375: $\{I_n\}_0^{\infty}$ starts with
376: $$
377: 1,1,2,4,10,26,76,232,764,2620,9496,35696,140152. $$
378:
379: \subsection*{Dyck paths}
380:
381: A \emph{Dyck path} of length $2n$ is a lattice path from $(0,0)$ to
382: $(2n,0)$ with steps $(1,1)$ and $(1,-1)$ that never goes below the
383: $x$-axis. Letting $u$ and $d$ represent the steps $(1,1)$ and $(1,-1)$
384: respectively, we code such a path with a word over $\{u,d\}$. For
385: example, the path
386: \begin{center}
387: \includegraphics[width=25ex]{dyckpath.eps}
388: \end{center}
389: is coded by $uuduuddd$. The $n$th \emph{Catalan number} $C_n=\frac 1
390: {n+1} \binom {2n} n$ counts the number of Dyck paths of length $2n$.
391: The sequence of Catalan numbers starts with
392: $$1,1,2,5,14,42,132,429,1430,4862,16796,58786,208012.$$
393:
394: \subsection*{Motzkin paths}
395:
396: A \emph{Motzkin path} of length $n$ is a lattice path from $(0,0)$ to
397: $(n,0)$ with steps $(1,0)$, $(1,1)$, and $(1,-1)$ that never goes
398: below the $x$-axis. Letting $\ell$, $u$, and $d$ represent the steps
399: $(1,0)$, $(1,1)$, and $(1,-1)$ respectively, we code such a path with
400: a word over $\{\ell, u,d\}$. For example, the path
401: \begin{center}
402: \includegraphics[width=25ex]{motzkinpath.eps}
403: \end{center}
404: is coded by $u\ell \ell ud\ell d\ell $. The $n$th \emph{Motzkin
405: number} $M_n$ is the number of Motzkin paths of length $n$. The
406: first few of the Motzkin numbers are
407: $$
408: 1,1,2,4,9,21,51,127,323,835,2188,5798,15511. $$
409:
410:
411: \section{Three classes of patterns}
412:
413: Let $\pi=a_1 a_2 \cdots a_n \in \sym_n$. Define the \emph{reverse} of
414: $\pi$ as $\pi^r:=a_n \cdots a_2 a_1$, and define the \emph{complement}
415: of $\pi$ by $\pi^c(i) = n+1-\pi(i)$, where $i\in[n]$.
416:
417: \begin{proposition}
418: With respect to being equidistributed, the twelve pattern statistics
419: of length three with one dash fall into the following three classes.
420: \begin{itemize}
421: \item[(i)] \hspace{.8ex}\axbc,\; \cxba,\; \abxc,\; \cbxa.
422: \item[(ii)] \hspace{.8ex}\axcb,\; \cxab,\; \baxc,\; \bcxa.
423: \item[(iii)] \hspace{.8ex}\bxac,\; \bxca,\; \acxb,\; \caxb.
424: \end{itemize}
425: \end{proposition}
426:
427: \begin{proof}
428: The bijections $\pi\mapsto\pi^r$, $\pi\mapsto\pi^c$, and
429: $\pi\mapsto(\pi^r)^c$ give the equidistribution part of the result.
430: Calculations show that these three distributions differ pairwise on
431: $\sym_4$.
432: \end{proof}
433:
434: % ------------------------------------------------------------------------
435: \section{Permutations avoiding a pattern of class one or two}
436:
437:
438: \begin{proposition}\label{bell1}
439: Partitions of $[n]$ are in one-to-one correspondence with
440: $(\axbc)$-avoiding permutations in $\sym_n$. Hence
441: $|\sym_n(\axbc)|=B_n$.
442: \end{proposition}
443:
444: \begin{proof}[First proof]
445: Recall that the Bell numbers satisfy $B_0=1$, and
446: $$B_{n+1}= \sum_{k=0}^n \binom n k B_k.$$
447: We show that
448: $|\sym_n(\axbc)|$ satisfy the same recursion. Clearly,
449: $\sym_0(\axbc) = \{\emptyword\}$. For $n > 0$, let
450: $M=\{2,3,\ldots,n+1\}$, and let $S$ be a $k$ element subset of $M$.
451: For each $(\axbc)$-avoiding permutation $\sigma$ of $S$ we construct
452: a unique $(\axbc)$-avoiding permutation $\pi$ of $[n+1]$. Let $\tau$
453: be the word obtained by writing the elements of $M\setminus S$ in
454: decreasing order. Define $\pi := \sigma 1 \tau$.
455:
456: Conversely, if $\pi = \sigma 1 \tau$ is a given $(\axbc)$-avoiding
457: permutation of $[n+1]$, where $|\sigma|=k$, then the letters of
458: $\tau$ are in decreasing order, and $\sigma$ is an $(\axbc)$-avoiding
459: permutation of the $k$ element set $\{2,3,\ldots,n+1\}\setminus \{i
460: : i \text{ is a letter in }\tau\}$.
461: \end{proof}
462:
463: \begin{proof}[Second proof]
464: Given a partition $\pi$ of $[n]$, we introduce a standard
465: representation of $\pi$ by requiring that:
466: \begin{itemize}
467: \item[(a)] Each block is written with its least element first, and
468: the rest of the elements of that block are written in decreasing
469: order.
470: \item[(b)] The blocks are written in decreasing order of their least
471: element, and with dashes separating the blocks.
472: \end{itemize}
473:
474: Define $\wpi$ to be the permutation we obtain from $\pi$ by writing
475: it in standard form and erasing the dashes. We now argue that
476: $\wpi:=a_1a_2\cdots a_n$ avoids $(\axbc)$. If $a_i<a_{i+1}$, then
477: $a_i$ and $a_{i+1}$ are the first and the second element of some
478: block. By the construction of $\wpi$, $a_i$ is a left-to-right
479: minimum, hence there is no $j\in[i-1]$ such that $a_j<a_i$.
480:
481: Conversely, $\pi$ can be recovered uniquely from $\wpi$ by inserting
482: a dash in $\wpi$ preceding each left-to-right minimum, apart from
483: the first letter in $\wpi$. Thus $\pi\mapsto\wpi$ gives the desired
484: bijection.
485: \end{proof}
486:
487: \begin{example}
488: As an illustration of the map defined in the above proof, let
489: $$\pi=\{\{1,3,5\},\{2,6,9\},\{4,7\},\{8\}\}.$$
490: Its standard form is
491: $8 \dd 47 \dd 296 \dd 153$. Thus $\wpi= 847296153$.
492: \end{example}
493:
494: \begin{porism}
495: Let $L(\pi)$ be the number of left-to-right minima of $\pi$. Then
496: $$
497: \sum_{\pi\in\sym_n(\axbc)} \!\!\!\!\!\!\!\! x^{L(\pi)} =
498: \sum_{k\geq 0} S(n,k)x^k.
499: $$
500: \end{porism}
501:
502: \begin{proof}
503: This result follows readily from the second proof of
504: Proposition~\ref{bell1}. We here give a different proof, which is
505: based on the fact that the Stirling numbers of the second kind
506: satisfy
507: \begin{equation*}
508: S(n,k) = S(n-1,k-1) + kS(n-1,k).
509: \end{equation*}
510:
511: Let $T(n,k)$ be the number of permutations in $\sym_n(\axbc)$ with
512: $k$ left-to-right minima. We show that the $T(n,k)$ satisfy the same
513: recursion as the $S(n,k)$.
514:
515: Let $\pi$ be an $(\axbc)$-avoiding permutation of $[n-1]$. To insert
516: $n$ in $\pi$, preserving $(\axbc)$-avoidance, we can put $n$ in
517: front of $\pi$ or we can insert $n$ immediately after each
518: left-to-right minimum. Putting $n$ in front of $\pi$ creates a new
519: left-to-right minimum, while inserting $n$ immediately after a
520: left-to-right minimum does not.
521: \end{proof}
522:
523: % ------------------------------------------------------------------------
524:
525: \begin{proposition}\label{bell2}
526: Partitions of $[n]$ are in one-to-one correspondence with
527: $(\axcb)$-avoiding permutations in $\sym_n$. Hence
528: $|\sym_n(\axcb)|=B_n$.
529: \end{proposition}
530:
531: \begin{proof}
532: Let $\pi$ be a partition of $[n]$. We introduce a standard
533: representation of $\pi$ by requiring that:
534: \begin{itemize}
535: \item[(a)] The elements of a block are written in increasing order.
536: \item[(b)] The blocks are written in decreasing order of their least
537: element, and with dashes separating the blocks.
538: \end{itemize}
539: Notice that this standard representation is different from the one
540: given in the second proof of Proposition~\ref{bell1}. Define $\wpi$
541: to be the permutation we obtain from $\pi$ by writing it in standard
542: form and erasing the dashes. It easy to see that $\wpi$ avoids
543: $(\axcb)$. Conversely, $\pi$ can be recovered uniquely from $\wpi$
544: by inserting a dash in between each descent in $\wpi$.
545: \end{proof}
546:
547: \begin{example}
548: As an illustration of the map defined in the above proof, let
549: $$\pi=\{\{1,3,5\},\{2,6,9\},\{4,7\},\{8\}\}.$$
550: Its standard form is
551: $8 \dd 47 \dd 269 \dd 135$. Thus $\wpi= 847269135$.
552: \end{example}
553:
554: \begin{porism}
555: $$
556: \sum_{\pi\in\sym_n(\axcb)} \!\!\!\!\!\!\!\! x^{1+\des \pi } =
557: \sum_{k\geq 0} S(n,k)x^k.
558: $$
559: \end{porism}
560:
561: \begin{proof}
562: From the proof of Proposition~\ref{bell2} we see that $\pi$ has
563: $k+1$ blocks precisely when $\wpi$ has $k$ descents.
564: \end{proof}
565:
566: % ------------------------------------------------------------------------
567:
568: \begin{proposition}\label{involutions}
569: Involutions in $\sym_n$ are in one-to-one correspondence with
570: permutations in $\sym_n$ that avoid $(\axbc)$ and $(\axcb)$. Hence
571: $$ |\sym_n(\axbc, \axcb)|=I_n.$$
572: \end{proposition}
573:
574: \begin{proof}
575: We give a combinatorial proof using a bijection that is essentially
576: identical to the one given in the second proof of
577: Proposition~\ref{bell1}.
578:
579: Let $\pi\in\sym_n$ be an involution. Recall that $\pi$ is an
580: involution if and only if each cycle of $\pi$ is of length one or
581: two. We now introduce a standard form for writing $\pi$ in cycle
582: notation by requiring that:
583: \begin{itemize}
584: \item[(a)] Each cycle is written with its least element first.
585: \item[(b)] The cycles are written in decreasing order of their least
586: element.
587: \end{itemize}
588: Define $\wpi$ to be the permutation obtained from $\pi$ by writing
589: it in standard form and erasing the parentheses separating the
590: cycles.
591:
592:
593: Observe that $\wpi$ avoids $(\axbc)$: Assume that $a_i<a_{i+1}$,
594: that is $(a_i\,a_{i+1})$ is a cycle in $\pi$, then $a_i$ is a
595: left-to-right minimum in $\pi$. This is guaranteed by the
596: construction of $\wpi$. Thus there is no $j<i$ such that $a_j<a_i$.
597:
598: The permutation $\wpi$ also avoids $(\axcb)$: Assume that
599: $a_i>a_{i+1}$, then $a_{i+1}$ must be the smallest element of some
600: cycle. Then $a_i$ is a left-to-right minimum in $\pi$.
601:
602: Conversely, if $\wpi:=a_1\ldots a_n$ is an $\{\axbc, \axcb\}$-avoiding
603: permutation then the involution $\pi$ is given by: $(a_i\,a_{i+1})$ is a
604: cycle in $\pi$ if and only if $a_i<a_{i+1}$.
605: \end{proof}
606:
607: \begin{example}
608: The involution $\pi=826543719$ written in standard form is
609: $$(9)(7)(4\,5)(3\,6)(2)(1\,8),$$
610: and hence $\wpi = 974536218$.
611: \end{example}
612:
613:
614: \begin{porism}
615: $$\sym_n(\axbc, \axcb) = \sym_n(\axbc, acb)
616: = \sym_n(abc, \axcb)
617: = \sym_n(abc, acb).$$
618: \end{porism}
619:
620: \begin{proof}
621: Kitaev \cite{Ki00} observed that the dashes in the patterns $(\axbc)$
622: and $(\axcb)$ are immaterial for the proof of
623: Proposition~\ref{involutions}. The result may, however, also be proved
624: directly. For an example of such a proof see the proof of
625: Lemma~\ref{lemma2}.
626: \end{proof}
627:
628: \begin{porism}\label{des_fix}
629: The number of permutations in $\sym_{n+k}(\axbc,\axcb)$ with $n-1$
630: descents equals the number of involutions in $\sym_{n+k}$ with $n-k$
631: fixed points.
632: \end{porism}
633:
634: \begin{proof}
635: Under the bijection $\pi\mapsto\wpi$ in the proof of
636: Proposition~\ref{involutions}, a cycle of length two in $\pi$
637: corresponds to an occurrence of $(\ab)$ in $\wpi$. Hence, if $\pi$
638: has $n-2k$ fixed points, then $\wpi$ has $n-k-1$ descents.
639: Substituting $n+k$ for $n$ we get the desired result.
640: \end{proof}
641:
642: To take the analysis of descents in $\{\axbc,\axcb\}$-avoiding
643: permutations further, we introduce the polynomial
644: $$
645: A_n(x) = \!\!\! \!\!\! \sum_{\pi\in\sym_n(\axbc,\axcb)}
646: \!\!\!\!\!\! \!\!\!\!\!\! \! x^{1+\des\,\pi},
647: $$
648: and call it the $n$th \emph{Eulerian polynomial for
649: $\{\axbc,\axcb\}$-avoiding permutations}. Direct enumeration shows
650: that the sequence $\{A_n(x)\}$ starts with
651: $$
652: \begin{array}{rclclclclclclclcl}
653: A_0(x) &=& 1 \\
654: A_1(x) &=& &\!\!+\!\!& x\\
655: A_2(x) &=& & & x &\!\!+\!\!& x^2\\
656: A_3(x) &=& & & & & 3x^2 &\!\!+\!\!& x^3\\
657: A_4(x) &=& & & & & 3x^2 &\!\!+\!\!& 6x^3 &\!\!+\!\!& x^4\\
658: A_5(x) &=& & & & & & & 15x^3 &\!\!+\!\!& 10x^4 &\!\!+\!\!& x^5\\
659: A_6(x) &=& & & & & & & 15x^3 &\!\!+\!\!& 45x^4 &\!\!+\!\!& 15x^5 &\!\!+\!\!& x^6\\
660: A_7(x) &=& & & & & & & & & 105x^4 &\!\!+\!\!& 105x^5 &\!\!+\!\!& 21x^6 &\!\!+\!\!& x^7.\\
661: \end{array}
662: $$
663: We will relate these polynomials to the so called Bessel
664: polynomials. The $n$th \emph{Bessel polynomial} $y_n(x)$ is defined
665: by
666: \begin{equation}\label{bessel_expl}
667: y_n(x) = \sum_{k=0}^n \binom {n+k} k \binom n k \frac {k!} {2^k}
668: x^k.
669: \end{equation}
670: The first six of the Bessel polynomials are
671: $$
672: \begin{array}{rclclclclclclcl}
673: y_0(x) &=& 1 \\
674: y_1(x) &=& 1 &\!\!+\!\!& x\\
675: y_2(x) &=& 1 &\!\!+\!\!& 3x &\!\!+\!\!& 3x^2\\
676: y_3(x) &=& 1 &\!\!+\!\!& 6x &\!\!+\!\!& 15x^2 &\!\!+\!\!& 15x^3\\
677: y_4(x) &=& 1 &\!\!+\!\!& 10x &\!\!+\!\!& 45x^2 &\!\!+\!\!& 105x^3 &\!\!+\!\!& 105x^4\\
678: y_5(x) &=& 1 &\!\!+\!\!& 15x &\!\!+\!\!& 105x^2 &\!\!+\!\!& 420x^3 &\!\!+\!\!& 945x^4 &\!\!+\!\!& 945x^5.\\
679: \end{array}
680: $$
681: These polynomials satisfy the second order differential
682: equation
683: $$x^2\frac {d^2y} {dx^2} + 2(x+1) \frac {dy} {dx} = n(n+1)y.$$
684: Moreover, the Bessel polynomials satisfy the recurrence relation
685: \begin{equation}\label{bessel_rec}
686: y_{n+1}(x) = (2n+1)xy_n(x) + y_{n-1}(x).
687: \end{equation}
688:
689: \begin{proposition}\label{euler}
690: Let $y_n(x)$ be the $n$th Bessel polynomial, and let $A_n(x)$ be the
691: $n$th Eulerian polynomial for $\{\axbc,\axcb\}$-avoiding
692: permutations. Then
693: \begin{itemize}
694: \item[(i)] $\sum_n y_n(x) (xt)^n$ generates $\{A_n(t)\}$, that is
695: $$\sum_{n\geq 0} A_n(t) x^n = \sum_{n\geq 0} y_n(x) (xt)^n.$$
696: \item[(ii)] $A_0(x) = 1$, $A_1(x) = x$, and for $n\geq2$, we have
697: $$A_{n+2}(x) = x(1+x+2x\frac d {dx}) A_n(x).$$
698: \item[(iii)] $A_n(x)$ is explicitly given by
699: $$A_n(x) = \sum_{k=0}^n \binom n k \binom {n-k} k \frac {k!} {2^k}
700: x^{n-k}.$$
701: \end{itemize}
702: \end{proposition}
703:
704: \begin{proof}
705: Let $I_n^k$ denote the number of involutions in $\sym_n$ with $k$
706: fixed points. Then Porism~\ref{des_fix} is equivalently stated as
707: \begin{equation}\label{eulerian_poly}
708: A_n(x) = \sum_{k\geq 0} I_n^{2k-n} x^k.
709: \end{equation}
710: In \cite{DuFa91} Dulucq and Favreau showed that the Bessel
711: polynomials are given by
712: \begin{equation}\label{bessel_poly}
713: y_n(x) = \sum_{k\geq 0} I_{n+k}^{n-k} x^k.
714: \end{equation}
715: To prove (i), multiply Equation~(\ref{bessel_poly}) by $(xt)^n$ and
716: sum over $n$.
717: \begin{align*}
718: \sum_{n\geq 0} y_n(x)(xt)^n
719: &\,=\, \sum_{n\geq 0} \sum_{k\geq 0} I_{n+k}^{n-k} t^n x^{n+k} \\
720: &\,=\, \sum_{k\geq 0} \sum_{n\geq 0} I_{k}^{2n-k} t^n x^k
721: && \text{By substituting $n-k$ for $k$.}\\
722: &\,=\, \sum_{k\geq 0} A_k(t) x^k
723: && \text{By Equation~(\ref{eulerian_poly}).}
724: \end{align*}
725: We now multiply Equation~(\ref{bessel_rec}) by $(xt)^n$ and sum over
726: $n$. Tedious but straightforward calculations then yield (ii) from
727: (i). Finally, we obtain (iii) from Equation~(\ref{bessel_expl}) by
728: identifying coefficients in (i).
729: \end{proof}
730:
731: % ------------------------------------------------------------------------
732:
733: \begin{definition}
734: Let $\pi$ be an arbitrary partition whose non-singleton blocks
735: $\{A_1,\ldots,A_k\}$ are ordered so that for all $i\in[k-1]$, $\min
736: A_i>\min A_{i+1}$. If $\max A_i>\max A_{i+1}$ for all $i\in [k-1]$,
737: then we call $\pi$ a \emph{monotone partition}. The set of monotone
738: partitions of $[n]$ is denoted by $\mono_n$.
739: \end{definition}
740:
741: \begin{example}
742: The partition
743: $$\pi \,=\, \FIGmono $$
744: is monotone.
745: \end{example}
746:
747: \begin{proposition}\label{mono}
748: Monotone partitions of $[n]$ are in one-to-one correspondence with
749: permutations in $\sym_n$ that avoid $(\axbc)$ and $(\abxc)$. Hence
750: $$|\sym_n(\axbc,\abxc)| = |\mono_n|.$$
751: \end{proposition}
752:
753: \begin{proof}
754: Given $\pi$ in $\mono_n$, let $A_1\dd A_2\dd\cdots\dd A_k$ be the
755: result of writing $\pi$ in the standard form given in the second
756: proof of Proposition~\ref{bell1}, and let $\wpi=A_1 A_2\cdots A_k$.
757: By the construction of $\wpi$ the fist letter in each $A_i$ is a
758: left-to-right minimum. Furthermore, since $\pi$ is monotone the
759: second letter in each non-singleton $A_i$ is a right-to-left
760: maximum. Therefore, if $xy$ is an $(ab)$-subword of $\wpi$, then $x$
761: is left-to-right minimum and $y$ is a right-to-left maximum. Thus
762: $\wpi$ avoids both $(\axbc)$ and $(\abxc)$.
763:
764: Conversely, given $\wpi$ in $\sym_n(\axbc,\abxc)$, let $A_1\dd
765: A_2\dd\cdots\dd A_k$ be the result of inserting a dash in $\wpi$
766: preceding each left-to-right minimum, apart from the first letter in
767: $\wpi$. Since $\wpi$ is $(\abxc)$-avoiding, the second letter in
768: each non-singleton $A_i$ is a right-to-left maximum. The second
769: letter in $A_i$ is the maximal element of $A_i$ when $A_i$ is viewed
770: as a set. Thus $\pi = \{A_1,A_2,\ldots,A_k \}$ is monotone.
771: \end{proof}
772:
773: What is left for us to show is that there is a one-to-one
774: correspondence between monotone partitions and non-overlapping
775: partitions. The proof we give is strongly influenced by the work of
776: Flajolet \cite{FlSc90}.
777:
778: \begin{proposition}\label{nop-mono}
779: Monotone partitions of $[n]$ are in one-to-one correspondence with
780: non-overlapping partitions of $[n]$. Hence $|\mono_n| = B_n^*$.
781: \end{proposition}
782:
783: \begin{proof}
784: If $k$ is the minimal element of a non-singleton block, then call
785: $k$ the \emph{first} element of that block. Similarly, If $k$ is the
786: maximal element of a non-singleton block, then call $k$ the
787: \emph{last} element of that block. An element of a non-singleton
788: that is not a first or last element is called an \emph{intermediate}
789: element. Let us introduce an ordering of the blocks of a partition. A
790: block $A$ is smaller than a block $B$ if $\min A < \min B$.
791:
792: We define a map $\Phi$ that to each non-overlapping partition $\pi$
793: of $[n]$ gives a unique monotone partition $\Phi(\pi)$ of
794: $[n]$. Let the integer $k$ range from $1$ to $n$.
795: \begin{itemize}
796: \item[(a)] If $k$ is the first element of a block of $\pi$, then
797: open a new block in $\Phi(\pi)$ by letting $k$ be its first
798: element. (A block $B$ is open if $\max B<k$.)
799: \item[(b)] If $k$ is the last element of a block of $\pi$, then
800: close the smallest open block of $\Phi(\pi)$ by letting $k$ be its
801: last element.
802: \item[(c)] If $k$ is an intermediate element of some block $B$ of
803: $\pi$, and $B$ is the $i$:th largest open block of $\pi$, then let
804: $k$ belong to the $i$th largest open block of $\Phi(\pi)$.
805: \item[(d)] If $\{k\}$ is a singleton block of $\pi$, then let
806: $\{k\}$ be a singleton block of $\Phi(\pi)$.
807: \end{itemize}
808:
809: Observe that $\Phi(\pi)$ is monotone. Indeed, it is only in (b) that
810: we close a block of $\Phi(\pi)$, and we always close the smallest
811: open block of $\Phi(\pi)$.
812:
813: Conversely, we give a map $\Psi$ that to each monotone partition
814: $\pi$ be a of $[n]$ gives a unique non-overlapping partition
815: $\Psi(\pi)$ of $[n]$. Define $\Psi$ the same way as $\Phi$ is
816: defined, except for case (c), where we instead of closing the
817: smallest open block close the largest open block.
818:
819: It is easy to see that $\Phi$ and $\Psi$ are each others inverses
820: and hence they are bijections.
821: \end{proof}
822:
823: \begin{corollary}\label{nop}
824: The NOPs (non-overlapping partitions) of $[n]$ are in one-to-one
825: correspondence with permutations in $\sym_n$ that avoid $(\axbc)$
826: and $(\abxc)$. Hence
827: $$|\sym_n(\axbc, \abxc)|=B^*_n.$$
828: \end{corollary}
829:
830: \begin{proof}
831: Follows immediately from Proposition~\ref{mono} together with
832: Proposition~\ref{nop-mono}.
833: \end{proof}
834:
835: \begin{example}
836: By the proof of Proposition~\ref{nop-mono}, the non-overlapping partition
837: \begin{align*}
838: \pi &\,=\, \FIGnop \\
839: \intertext{corresponds to the monotone partition}
840: \Phi(\pi) &\,=\, \FIGmono \\
841: \intertext{that according to the proof of Proposition~\ref{mono}
842: corresponds to the $\{\axbc,\abxc\}$-avoiding permutation}
843: \widehat{\Phi(\pi)} &\,=\, 1\ns 0\;1\ns 3\;1\ns 1\;9\;4\;1\ns 2\;6\;3\;8\;1\;7\;5\;2.
844: \end{align*}
845: \end{example}
846:
847:
848: \begin{porism}
849: Let $L(\pi)$ be the number of left-to-right minima of $\pi$. Then
850: $$
851: \sum_{\pi\in\sym_n(\axbc,\abxc)} \!\!\!\!\!\!\!\!\!\!\!\!\!\!
852: x^{L(\pi)} = \sum_{k\geq 0} S^*(n,k)x^k.
853: $$
854: \end{porism}
855:
856: \begin{proof}
857: Under the bijection $\pi\mapsto\wpi$ in the proof of
858: Proposition~\ref{mono}, the number of blocks in $\pi$ determines the
859: number of left-to-right minima of $\wpi$, and vice versa. The
860: number of blocks is not changed by the bijection $\Phi_1\circ\Psi_2$
861: in the proof of Proposition~\ref{nop-mono}.
862: \end{proof}
863:
864:
865: % ------------------------------------------------------------------------
866: \section{Permutations avoiding a pattern of class three}
867:
868: In \cite{Kn73v1} Knuth observed that there is a
869: one-to-one correspondence between $(\bxaxc)$-avoiding permutations and
870: Dyck paths. For completeness and future reference we give this result
871: as a lemma, and prove it using one of the least known bijections.
872: First we need a definition. For each word $x=x_1x_2\cdots x_n$
873: without repeated letters, we define the \emph{projection} of $x$ onto
874: $S_n$, which we denote $\proj(x)$, by
875: $$
876: \proj(x) = a_1 a_2 \cdots a_n \,,\;\text{ where }\; a_i = |\{j\in
877: [n] : x_i\geq x_j \}|.
878: $$
879: Equivalently, $\proj(x)$ is the permutation in $\sym_n$ which is
880: order equivalent to $x$. For example, $\proj(265) = 132$.
881:
882: \begin{lemma}\label{lemma1}
883: $|\sym_n(\bxaxc)|=C_n.$
884: \end{lemma}
885:
886: \begin{proof}
887: Let $\pi=a_1 a_2 \cdots a_n$ be a permutation of $[n]$ such that
888: $a_k=1$. Then $\pi$ is $(\bxaxc)$-avoiding if and only if
889: $\pi=\sigma 1 \tau$, where $\sigma:=a_1\cdots a_{k-1}$ is a
890: $(\bxaxc)$-avoiding permutation of $\{n,n-1,\ldots,n-k+1\}$, and
891: $\tau:=a_{k+1}\cdots a_n$ is a $(\bxaxc)$-avoiding permutation of
892: $\{2,3,\ldots,k\}$.
893:
894: We define recursively a mapping $\Phi$ from $\sym_n(\bxaxc)$ onto
895: the set of Dyck paths of length $2n$. If $\pi$ is the empty word,
896: then so is the Dyck path determined by $\pi$, that is,
897: $\Phi(\emptyword) = \emptyword$. If $\pi \neq \emptyword$, then we
898: can use the factorisation $\pi=\sigma 1 \tau$ from above, and define
899: $\Phi(\pi) = u\,(\Phi\circ\proj)(\sigma)\,d\,(\Phi\circ\proj)(\tau)$. It
900: is easy to see that $\Phi$ may be inverted, and hence is a
901: bijection.
902: \end{proof}
903:
904:
905: \begin{lemma}\label{lemma2}
906: A permutation avoids $(\bxac)$ if and only if it avoids $(\bxaxc)$.
907: \end{lemma}
908:
909: \begin{proof}
910: The sufficiency part of the proposition is trivial. The necessity
911: part is not difficult either. Assume that $\pi$ contains a
912: $(\bxaxc)$-subword. Then there exist
913: $$
914: A,B,C, n_1,n_2,\ldots,n_r \in [n], \text{ where } A < B < C,
915: $$
916: such that $B\hspace{-1pt}AC$ is a subword of $\pi$, and $A
917: n_1\cdots n_r C$ is a segment of $\pi$. If $n_1>B$, then
918: $B\hspace{-1pt}A n_1$ form a $(\bxac)$-subword in $\pi$. Assume that
919: $n_1<B$. Indeed, to avoid forming a $(\bxac)$-subword we will have
920: to assume that $n_i<B$ for all $i\in[r]$, but then $B n_r C$ is a
921: $(\bxac)$-subword. Accordingly we conclude that there exists at
922: least one $(\bxac)$-subword in $\pi$.
923: \end{proof}
924:
925: \begin{proposition}\label{catalan}
926: Dyck paths of length $2n$ are in one-to-one correspondence with
927: $(\bxac)$-avoiding permutations in $\sym_n$. Hence
928: $$|\sym_n(\bxac)|=\frac 1 {n+1} \binom {2n} n.$$
929: \end{proposition}
930:
931: \begin{proof}
932: Follows immediately from Lemma~\ref{lemma1} and Lemma~\ref{lemma2}.
933: \end{proof}
934:
935: \begin{proposition}
936: Let $L(\pi)$ be the number of left-to-right minima of $\pi$. Then
937: $$
938: \sum_{\pi\in\sym_n(\bxac)} \!\!\!\!\!\!\!\! x^{L(\pi)} =
939: \sum_{k\geq 0} \frac k {2n-k} \binom{2n-k} n x^k.
940: $$
941: \end{proposition}
942:
943: \begin{proof}
944: A \emph{return step} in a Dyck path $\delta$ is a $d$ such that
945: $\delta = \alpha u \beta d \gamma$, for some Dyck paths $\alpha$,
946: $\beta$, and $\gamma$. A useful observation is that every non-empty
947: Dyck path $\delta$ can be uniquely decomposed as $\delta = u \alpha
948: d \beta$, where $\alpha$ and $\beta$ are Dyck paths. This is the
949: so-called \emph{first return decomposition} of $\delta$. Let
950: $R(\delta)$ denote the number of return steps in $\delta$.
951:
952: In \cite{De99} Deutsch showed that the distribution of $R$ over all
953: Dyck paths of length $2n$ is the distribution we claim that $L$ has
954: over $\sym_n(\bxac)$.
955:
956: Let $\gamma$ be a Dyck path of length $2n$, and let $\gamma = u
957: \alpha d \beta$ be its first return decomposition. Then $R(\gamma) =
958: 1 + R(\beta)$. Let $\pi\in\sym_n(\bxac)$, and let $\pi=\sigma 1
959: \tau$ be the decomposition given in the proof of Lemma~\ref{lemma1}.
960: Then $L(\pi) = 1 + L(\sigma)$. The result now follows by induction.
961: \end{proof}
962:
963: In addition, it is easy to deduce that left-to-right minima,
964: left-to-right maxima, right-to-left minima, and right-to-left maxima
965: all share the same distribution over $\sym_n(\bxac)$.
966:
967: % ------------------------------------------------------------------------
968:
969: \begin{proposition}
970: Motzkin paths of length $n$ are in one-to-one correspondence with
971: permutations in $\sym_n$ that avoid $(\axbc)$ and $(\acxb)$. Hence
972: $$|\sym_n(\axbc, \acxb)|=M_n.$$
973: \end{proposition}
974:
975: \begin{proof}
976: We mimic the proof of Lemma~\ref{lemma1}. Let $\pi\in\sym_n(\axbc,
977: \acxb)$. Since $\pi$ avoids $(\acxb)$ it also avoids $(\axcxb)$ by
978: Lemma~\ref{lemma2} via $\pi\mapsto (\pi^c)^r$. Thus we may write
979: $\pi = \sigma n \tau$, where $\pi(k)=n$, $\tau$ is an $\{\axbc,
980: \acxb\}$-avoiding permutation of $\{n-1,n-2,\ldots,n-k+1\}$, and
981: $\tau$ is an $\{\axbc, \acxb\}$-avoiding permutation of $[n-k]$. If
982: $\sigma\neq\emptyword$ then $\sigma=\sigma'r$ where $r=n-k+1$, or
983: else an $(\axbc)$-subword would be formed with $n$ as the '$c$' in
984: $(\axbc)$. Define a map $\Phi$ from $\sym_n(\axbc, \acxb)$ to the
985: set of Motzkin paths by $\Phi(\emptyword) = \emptyword$ and
986: $$\Phi(\pi) =
987: \begin{cases}
988: \ell\,(\Phi\circ\proj)(\sigma) & \text{ if } \pi = n\sigma, \\
989: u\,(\Phi\circ\proj)(\sigma)\,d\,\Phi(\tau) & \text{ if }
990: \pi = \sigma r n \tau \text{ and } r=n-k+1.\\
991: \end{cases}
992: $$
993: Its routine to find the inverse of $\Phi$.
994: \end{proof}
995:
996: \begin{example}
997: Let us find the Motzkin path associated to the
998: $\{\axbc,\acxb\}$-avoiding permutation $76453281$.
999: \begin{eqnarray*}
1000: \Phi(76453{\bf 28}1) &=& u \Phi({\bf 5}4231) d \Phi({\bf 1}) \\
1001: &=& u\ell \Phi({\bf 4}231) d\ell \\
1002: &=& u\ell\ell \Phi({\bf 23}1) d\ell \\
1003: &=& u\ell\ell u d \Phi({\bf 1}) d\ell \\
1004: &=& u\ell\ell u d\ell d\ell \\
1005: \end{eqnarray*}
1006: \end{example}
1007:
1008: \section*{Acknowledgement}
1009:
1010: I am greatly indebted to my advisor Einar Steingr\'{\i}msson, who put
1011: his trust in me and gave me the opportunity to study mathematics on a
1012: postgraduate level. This work has benefited from his knowledge,
1013: enthusiasm and generosity.
1014:
1015:
1016: \bibliographystyle{plain}
1017: \nocite{SlPl95}
1018: \nocite{St97}
1019: \bibliography{gpa}
1020:
1021: \end{document}
1022:
1023: