1: %%% Check spelling
2:
3: % sections
4: % 1. {Introduction} \label{sec-intro}
5: % 2. {Triangulations and frieze patterns} \label{sec-frieze}
6: % 3. {The sideways recurrence and its periodicity} \label{sec-sideways}
7: % 4. {Snake graphs} \label{sec-snake}
8: % 5. {A tropical analogue} \label{sec-tropical}
9: % 6. {A variant} \label{sec-variant}
10: % 7. {Markoff numbers} \label{sec-markoff}
11: % 8. {Other directions for exploration} \label{sec-other}
12:
13: \documentclass[12pt]{article}
14: \usepackage{mathdots}
15:
16: %\usepackage[dvips]{graphics}
17: \usepackage{times,mathptmx,amsthm,amsmath}
18: \usepackage{pstricks,pst-node,graphics}
19: \usepackage{epsf}
20: \usepackage{graphicx}
21:
22: % Theorem definitions
23: \newtheorem{theorem}{Theorem}[section]
24: \newtheorem{proposition}{Proposition}[section]
25: \newtheorem{lemma}[theorem]{Lemma}
26: \theoremstyle{definition}
27: \newtheorem{definition}[theorem]{Definition}
28: \newtheorem{example}[theorem]{Example}
29: \newtheorem{xca}[theorem]{Exercise}
30: \theoremstyle{remark}
31: \newtheorem{remark}[theorem]{Remark}
32: \numberwithin{equation}{section}
33:
34: \newcommand{\Z}{{\bf Z}}
35: \newcommand{\R}{{\bf R}}
36: \newcommand{\vu}{{\bf u}}
37: \newcommand{\vv}{{\bf v}}
38: \newcommand{\vw}{{\bf w}}
39: \newcommand{\vzero}{{\bf 0}}
40: \newcommand{\ve}{{\bf e}}
41: \newcommand{\tT}{\frac{2}{3}}
42: \newcommand{\fT}{\frac{5}{3}}
43: \newcommand{\sT}{\frac{7}{3}}
44: \newcommand{\Un}{\ 1 \ }
45: \newcommand{\jd}{\ddots}
46: \newcommand{\ji}{\iddots}
47:
48: % figures
49: % 1. A triangulated 6-gon.
50: % 2. The graph $G_{1,4}$.
51: % 3. A triangulated 6-gon with edge-weights.
52: % 4. The graph $G_{1,4}$ with edge-weights.
53: % 6. A snake of order 6.
54: % 6. The snake-graph recurrence.
55: % 7. Another snake of order 6.
56: % 8. A paths model for snakes.
57: % 9. A planar paths model for snakes.
58: % 10. Another planar paths model for snakes.
59: % 11. Conway and Coxeter's marking scheme.
60: % 12. A hexagon snake.
61: % 13. From path systems to perfect matchings.
62: % 14. A hexagon snake with regular hexagons.
63: % 15. A (short) straight snake with multiplicities.
64: % 16. A dual snake.
65: % 17. A dual snake in a frieze pattern.
66: % 18. An integral lamination.
67: % 19. A fundamental parallelogram.
68: % 20. A Markoff snake.
69: % 21. Kuo condensation for snakes.
70: % 22. Herriot's theorem.
71:
72: \begin{document}
73:
74: \begin{center}
75: \bf The combinatorics of frieze patterns and Markoff numbers \rm \\
76: \end{center}
77: \bigskip
78: \begin{center}
79: James Propp \\
80: Department of Mathematics, \\
81: University of Massachusetts Lowell \\
82: James\_Propp@uml.edu.ignorethis
83: \end{center}
84:
85: \bigskip
86:
87: \small
88:
89: \noindent
90: {\sc Abstract}: This article, based on joint work with
91: Gabriel Carroll, Andy Itsara, Ian Le, Gregg Musiker,
92: Gregory Price, Dylan Thurston, and Rui Viana,
93: presents a combinatorial model based on perfect matchings
94: that explains the symmetries of the numerical arrays
95: that Conway and Coxeter dubbed frieze patterns.
96: This matchings model is a combinatorial interpretation
97: of Fomin and Zelevinsky's cluster algebras of type $A$.
98: One can derive from the matchings model
99: an enumerative meaning for the Markoff numbers,
100: and prove that the associated Laurent polynomials
101: have positive coefficients
102: as was conjectured (much more generally) by Fomin and Zelevinsky.
103: Most of this research was conducted under the auspices of REACH
104: (Research Experiences in Algebraic Combinatorics at Harvard).
105:
106: \normalsize
107:
108: \bigskip
109:
110: \section{Introduction} \label{sec-intro}
111:
112: This article is part of a recent burst of activity
113: relating to what Sergey Fomin and Andrei Zelevinsky
114: have dubbed the ``Laurent phenomenon''
115: (described in greater detail below),
116: This phenomenon has algebraic, topological, and combinatorial aspects,
117: and it the third of these aspects that is developed here.
118: In particular, we show how two examples of rational recurrences ---
119: the two-dimensional frieze patterns of Conway and Coxeter,
120: and the tree of Markoff numbers ---
121: relate to one another and to the Laurent phenomenon.
122:
123: A {\it Laurent polynomial\/} in the variables $x,y,\dots$
124: is a rational function in $x,y,\dots$ that can be expressed as
125: a polynomial in the variables $x,x^{-1},y,y^{-1},\dots$;
126: for example,
127: the function $f(x) = (x^2+1)/x = x + x^{-1}$ is a Laurent polynomial,
128: but the composition $f(f(x)) = (x^4+3x^2+1)/x(x^2+1)$ is not.
129: The preceding example shows
130: that the set of Laurent polynomials in a single variable
131: is not closed under composition.
132: This failure of closure also holds in the multivariate setting;
133: for instance, if $f(x,y)$, $g(x,y)$ and $h(x,y)$
134: are Laurent polynomials in $x$ and $y$,
135: then we would not expect to find
136: that $f(g(x,y),h(x,y))$ is a Laurent polynomial as well.
137: Nonetheless, it has been discovered that,
138: in broad class of instances (embraced as yet by no general rule),
139: ``fortuitous'' cancellations occur that cause Laurentness to be preserved.
140: This is the ``Laurent phenomenon''
141: discussed by Fomin and Zelevinsky~\cite{FZL}.
142:
143: Furthermore, in many situations where the Laurent phenomenon holds,
144: there is a positivity phenomenon at work as well,
145: and all the coefficients of the Laurent polynomials turn out to be positive.
146: In these cases, the functions being composed
147: are Laurent polynomials with positive coefficients;
148: that is, they are expressions
149: involving only addition, multiplication, and division.
150: It should be noted that subtraction-free expressions
151: do not have all the closure properties one might hope for,
152: as the example $(x^3+y^3)/(x+y)$ illustrates:
153: although the expression is subtraction-free,
154: its reduced form $x^2 - xy + y^2$ is not.
155:
156: Fomin and Zelevinsky have shown that a large part
157: of the Laurentness phenomenon
158: fits in with their general theory of cluster algebras.
159: This article will treat one important special case
160: of the Laurentness-plus-positivity phenomenon,
161: namely the case associated with cluster algebras of type $A$,
162: discussed in detail in~\cite{FZY}.
163: The purely combinatorial approach taken
164: in sections~\ref{sec-frieze} and~\ref{sec-sideways} of this article
165: obscures the links with deeper issues
166: (notably the representation-theoretic questions
167: that motivated the invention of cluster algebras),
168: but it provides the quickest and most self-contained way
169: to prove the Laurentness-plus-positivity assertion in this case
170: (Theorem~\ref{thm-frieze}).
171: The frieze patterns of Conway and Coxeter,
172: and their link with triangulations of polygons,
173: will play a fundamental role,
174: as will perfect matchings of graphs derived from these triangulations.
175: (For a different, more algebraic way
176: of thinking about frieze patterns, see~\cite{CaCh}.
177: For an extension of the result of this article into a broader setting,
178: see~\cite{BM}.)
179:
180: In sections~\ref{sec-tropical} and~\ref{sec-variant} of this article,
181: two variations on the theme of frieze patterns are considered.
182: One is the tropical analogue,
183: which has bearing on graph-metrics in trees.
184: The other variant is based on a recurrence
185: that looks very similar to the frieze relation;
186: the variant recurrence
187: appears to give rise to tables of positive integers
188: possessing the same glide-reflection symmetry as frieze patterns,
189: but positivity and integrality are still unproved.
190: %% Really? Didn't Hickerson prove it? -- 5/19/08
191:
192: In section~\ref{sec-snake},
193: the graphs constructed in section~\ref{sec-frieze}
194: are viewed from a number of different perspectives
195: that relate them to existing literature.
196:
197: In section~\ref{sec-markoff},
198: the constructions of sections~\ref{sec-frieze} and~\ref{sec-sideways}
199: are specialized to a case
200: in which the triangulated polygons come from pairs of mutually visible points
201: in a dissection of the plane into equilateral triangles.
202: In this case, counting the matchings of the derived graphs
203: gives us an enumerative interpretation of Markoff numbers
204: (numbers satisfying the ternary cubic $x^2+y^2+z^2=3xyz$).
205: This yields a combinatorial proof of a Laurentness assertion
206: proved by Fomin and Zelevinsky in~\cite{FZL}
207: (namely a special case of their Theorem 1.10)
208: that falls outside of the framework of cluster algebras in the strict sense.
209: Fomin and Zelevinsky proved Theorem 1.10
210: by use of their versatile ``Caterpillar Lemma'',
211: but this proof did not settle the issue of positivity.
212: The combinatorial approach adopted here shows
213: that all of the Laurent polynomials that occur in
214: the three-variable rational-function analogue of the Markoff numbers ---
215: the ``Markoff polynomials'' ---
216: are in fact positive (Theorem~\ref{thm-triples}).
217:
218: Section~\ref{sec-other} concludes with some problems suggested
219: by the main result of section~\ref{sec-markoff}.
220: One can try to generalize the combinatorial picture
221: by taking other dissections of the plane into triangles,
222: or one can try to generalize by considering other Diophantine equations.
223: There are tantalizing hints of a link between
224: the two proposed avenues of generalization,
225: but its nature is still obscure.
226:
227: This work was supported by funds
228: from the National Science Foundation
229: and the National Security Agency.
230: It would have been impossible without
231: the deep insight and hard work
232: of Gabriel Carroll, Andy Itsara, Ian Le,
233: Gregg Musiker, Gregory Price, and Rui Viana
234: (all of whom were undergraduates
235: at the time the work was done)
236: and conversations with Dylan Thurston,
237: as well as earlier conversations with
238: Rick Kenyon and David Wilson.
239:
240: \section{Triangulations and frieze patterns} \label{sec-frieze}
241:
242: A {\it frieze pattern}~\cite{CoCo}
243: is an infinite array such as
244: $$
245: \begin{array}{cccccccccccccccccccc}
246: ...&\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &... \\[1.2ex]
247: ...& &\fT& & 2 & &\Un& & 5 & &\tT& & 3 & &\fT& & 2 &... \\[1.2ex]
248: ...& 4 & &\sT& &\Un& & 4 & &\sT& &\Un& & 4 & &\sT& &... \\[1.2ex]
249: ...& & 5 & &\tT& & 3 & &\fT& & 2 & &\Un& & 5 & &\tT&... \\[1.2ex]
250: ...&\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &\Un& &...
251: \end{array}
252: $$
253: consisting of $n-1$ rows, each periodic with period $n$,
254: such that all entries in the top and bottom rows are equal to 1
255: and all entries in the intervening rows satisfy the relation
256: $$
257: \begin{array}{ccc}
258: & A & \\[1.6ex]
259: B & & C \\[1.6ex]
260: & D &
261: \end{array} \ \ : \ \ AD = BC-1 \ \ .
262: $$
263: The rationale for the term ``frieze pattern'' is that
264: such an array automatically possesses glide-reflection symmetry
265: (as found in some decorative friezes):
266: for $1 \leq m \leq n-1$, the $(n-m)$th row
267: is the same as the $m$th row, shifted rightward
268: (or, equivalently, leftward) by $n/2$ positions.
269: Hence the relation $AD=BC-1$
270: will be referred to below as the ``frieze relation''
271: even though its relevance to friezes and their symmetries
272: is not apparent from the algebraic definition.
273:
274: Frieze patterns arose from Coxeter's study
275: of metric properties of polytopes,
276: and served as useful scaffolding for various sorts of metric data;
277: see e.g.~\cite{C1} (page 160),~\cite{C2}, and~\cite{C3}.
278: Typically some of the entries in a metric frieze pattern are irrational.
279: Conway and Coxeter completely classify
280: the frieze patterns whose entries are positive integers,
281: and show that these frieze patterns constitute
282: a manifestation of the Catalan numbers.
283: Specifically, there is a natural association
284: between positive integer frieze patterns
285: and triangulations of regular polygons with labeled vertices.
286: (In addition to~\cite{CoCo}, see the shorter discussion
287: on pp.\ 74--76 and 96--97 of~\cite{CG}.)
288: Note that for each fixed $n$, any convex $n$-gon would serve here
289: just as well as the regular $n$-gon,
290: since we are only viewing triangulations combinatorially.
291:
292: $$
293: \pspicture(0,-.75)(9,5.796)
294: %\pspicture(0,-.5)(6,3.864)
295: %\pspicture(0,-1.0)(12,7.728)
296: %\psset{xunit=2,yunit=2}
297: \psset{xunit=1.5,yunit=1.5}
298: \pspolygon(2,0)(4,0)(5,1.732)(4,3.464)(2,3.464)(1,1.732)
299: \psline(2,0)(2,3.464)
300: \psline(2,3.464)(4,0)
301: \psline(4,0)(4,3.464)
302: \rput(1.8,-.4){6}
303: \rput(4.2,-.4){5}
304: \rput(5.4,1.732){4}
305: \rput(4.2,3.864){3}
306: \rput(1.8,3.864){2}
307: \rput(0.6,1.732){1}
308: \endpspicture
309: $$
310: \begin{center}
311: Figure 1. A triangulated 6-gon.
312: \end{center}
313:
314: From every triangulation $T$ of a regular $n$-gon
315: with vertices cyclically labeled 1 through $n$,
316: Conway and Coxeter build an $(n-1)$-rowed frieze pattern
317: determined by the numbers $a_1,a_2,\dots,a_n$,
318: where $a_k$ is the number of triangles in $T$ incident with vertex $k$.
319: Specifically:
320: (1) the top row of the array is $\dots,1,1,1,\dots$;
321: (2) the second row (offset from the first)
322: is $\dots,a_1,a_2,\dots,a_n,a_1,\dots$ (with period $n$); and
323: (3) each succeeding row (offset from the one before)
324: is determined by the frieze recurrence
325: $$
326: \begin{array}{ccc}
327: & A & \\[1.6ex]
328: B & & C \\[1.6ex]
329: & D &
330: \end{array} \ \ : \ \ D = (BC-1)/A \ \ .
331: $$
332: E.g., the triangulation shown in Figure 1
333: determines the data $(a_1,\dots,a_6)=(1,3,2,1,3,2)$
334: and the 5-row frieze pattern
335: $$
336: \begin{array}{ccccccccccccccc}
337: ...&1& &1& &1& &1& &1& &1& &1&... \\[1.1ex]
338: ...& &1& &3& &2& &1& &3& &2& &... \\[1.1ex]
339: ...&1& &2& &5& &1& &2& &5& &1&... \\[1.1ex]
340: ...& &1& &3& &2& &1& &3& &2& &... \\[1.1ex]
341: ...&1& &1& &1& &1& &1& &1& &1&...
342: \end{array}
343: $$
344:
345: Conway and Coxeter show that the frieze relation,
346: applied to the initial rows $\dots,1,1,1,\dots$
347: and $\dots,a_1,a_2,\dots,a_n,\dots$,
348: yields a frieze pattern.
349: Note that implicit in this assertion
350: is the proposition that every entry in rows 1 through $n-3$ is non-zero
351: (so that the recurrence $D = (BC-1)/A$ never involves division by 0).
352: It is not a priori obvious that
353: each of the entries in the array is positive
354: (since the recurrence involves subtraction)
355: or that each of the entries is an integer
356: (since the recurrence involves division).
357: Nor is it immediately clear why
358: for $1 \leq m \leq n-1$, the $(n-m)$th row of the table
359: given by repeated application of the recurrence
360: should be the same as the $m$th row, shifted,
361: so that in particular the $n-1$st row, like the first row,
362: consists entirely of 1's.
363:
364: These and many other properties of frieze patterns
365: are explained by a combinatorial model of frieze patterns
366: discovered by Carroll and Price~\cite{CP}
367: (based on earlier work of Itsara, Le, Musiker, Price, and Viana;
368: see~\cite{I} and~\cite{M},
369: as well as~\cite{CILP}).
370: Given a triangulation $T$ as above, define a bipartite graph $G = G(T)$
371: whose $n$ black vertices $v$ correspond to the vertices of $T$,
372: whose $n-2$ white vertices $w$ correspond to the triangular faces of $T$,
373: and whose edges correspond to all incidences between vertices and faces in $T$
374: (that is, $v$ and $w$ are joined by an edge precisely if
375: $v$ is one of the three vertices of the triangle in $T$ associated with $w$).
376: For $i \neq j$ in the range $1,...,n$,
377: let $G_{i,j}$ be the graph obtained from $G$
378: by removing black vertices $i$ and $j$ and all edges incident with them,
379: and let $m_{i,j}$ be the number of perfect matchings of $G_{i,j}$
380: (that is, the number of ways to pair all $n-2$ of the black vertices
381: with the $n-2$ white vertices,
382: so that every vertex is paired to a vertex of the opposite color
383: adjacent to it).
384: For instance, for the triangulation $T$ of the 6-gon
385: defined in Figure 1, the graph $G_{1,4}$ is as shown in Figure 2,
386: and we have $m_{1,4} = 5$ since this graph has 5 perfect matchings.
387: \bigskip
388: $$
389: %\pspicture(3,0)(15,10.392)
390: \pspicture(1.5,0)(7.5,5.196)
391: \psset{xunit=.5,yunit=.5}
392: \psline(5,5.196)(6,10.392)
393: \psline(5,5.196)(6,0)
394: \psline(8,3.464)(6,10.392)
395: \psline(8,3.464)(12,0)
396: \psline(8,3.464)(6,0)
397: \psline(10,6.928)(6,10.392)
398: \psline(10,6.928)(12,10.392)
399: \psline(10,6.928)(12,0)
400: \psline(13,5.196)(12,10.392)
401: \psline(13,5.196)(12,0)
402: %\psset{linestyle=dashed}
403: %\psline(6,0)(12,0)
404: %\psline(12,0)(15,5.196)
405: %\psline(15,5.196)(12,10.392)
406: %\psline(12,10.392)(6,10.392)
407: %\psline(6,10.392)(3,5.196)
408: %\psline(3,5.196)(6,0)
409: %\psline(6,0)(6,10.392)
410: %\psline(6,10.392)(12,0)
411: %\psline(12,0)(12,10.392)
412: \pscircle[fillstyle=solid,fillcolor=black](6,0){.2}
413: \pscircle[fillstyle=solid,fillcolor=white](8,3.464){.2}
414: \pscircle[fillstyle=solid,fillcolor=black](12,0){.2}
415: \pscircle[fillstyle=solid,fillcolor=white](13,5.196){.2}
416: \pscircle[fillstyle=solid,fillcolor=black](12,10.392){.2}
417: \pscircle[fillstyle=solid,fillcolor=white](10,6.928){.2}
418: \pscircle[fillstyle=solid,fillcolor=black](6,10.392){.2}
419: \pscircle[fillstyle=solid,fillcolor=white](5,5.196){.2}
420: \endpspicture
421: $$
422: \begin{center}
423: Figure 2. The graph $G_{1,4}$.
424: \end{center}
425:
426: \bigskip
427:
428: \begin{theorem}[Carroll and Price~\cite{CP}] \label{thm-cp}
429: The Conway-Coxeter frieze pattern of a triangulation $T$ is just the array
430: \bigskip
431: $$
432: \begin{array}{ccccccccc}
433: \dots & m_{1,2} & & m_{2,3} & & m_{3,4} & & m_{4,5} & \dots \\[2.0ex]
434: \dots & & m_{1,3} & & m_{2,4} & & m_{3,5} & & \dots \\[2.0ex]
435: \dots & m_{n,3} & & m_{1,4} & & m_{2,5} & & m_{3,6} & \dots \\[2.0ex]
436: \dots & & m_{n,4} & & m_{1,5} & & m_{2,6} & & \dots \\[2.0ex]
437: & \vdots & & \vdots & & \vdots & & \vdots &
438: \end{array}
439: $$
440: \bigskip
441: where here as hereafter we interpret all subscripts mod $n$.
442: \end{theorem}
443:
444: \noindent
445: Note that this claim makes the glide-reflection symmetry of frieze patterns
446: a trivial consequence of the fact that $G_{i,j} = G_{j,i}$.
447:
448: \begin{proof}
449: Here is a sketch of the main steps of the proof:
450:
451: (1) $m_{i,i+1} = 1$: This holds because there is a tree structure on
452: the set of triangles in $T$ that induces a tree structure on the set
453: of white vertices of $G$. If we examine the white vertices of $G$,
454: proceeding from outermost to innermost, we find that we have no
455: freedom in how to match them with black vertices, when we keep in
456: mind that every black vertex must be matched with a white vertex.
457: (In fact, the same reasoning shows that $m_{i,j} = 1$ whenever
458: the triangulation $T$ contains a diagonal connecting vertices
459: $i$ and $j$.)
460:
461: (2) $m_{i-1,i+1} = a_i$: The argument is similar, except now we
462: have some freedom in how the $i$th black vertex is matched: it
463: can be matched with any of the $a_i$ adjacent white vertices.
464:
465: (3) $m_{i,j} \, m_{i-1,j+1} = m_{i-1,j} \, m_{i,j+1} - 1$:
466: If we move the 1 to the left-hand side, we can use (1) to write
467: the equation in the form
468: $$m_{i,j} \, m_{i-1,j+1} + m_{i-1,i} \, m_{j,j+1} = m_{i-1,j} \, m_{i,j+1}.$$
469: This relation is a direct consequence of a lemma due to Eric Kuo
470: (Theorem 2.5 in~\cite{Kuo}),
471: which is stated here for the reader's convenience:
472:
473: {\it Condensation lemma:}
474: If a bipartite planar graph $G$ has 2 more black vertices than white vertices,
475: and the black vertices $a,b,c,d$ lie in cyclic order on some face of $G$, then
476: $$m(a,c) m(b,d) = m(a,b) m(c,d) + m(a,d) m(b,c),$$
477: where $m(x,y)$ denotes the number of perfect matchings
478: of the graph obtained from $G$
479: by deleting vertices $x$ and $y$ and all incident edges.
480:
481: (1) and (2) tell us that Carroll and Price's theorem
482: applies to the first two rows of the frieze pattern,
483: and (3) tells us (by induction) that the theorem
484: applies to all subsequent rows.
485: \end{proof}
486:
487: It should be mentioned that Conway and Coxeter
488: give an alternative way of describing the entries in frieze patterns,
489: as determinants of tridiagonal matrices.
490: Note that $m_{i-1,i+1} = a_i$
491: which equals the determinant of the 1-by-1 matrix
492: whose sole entry is $a_i$,
493: while $m_{i-1,i+2} = a_{i} a_{i+1} - 1$
494: which equals the determinant of the 2-by-2 matrix
495: $$\left( \begin{array}{cc}
496: a_i & 1 \\
497: 1 & a_{i+1} \end{array} \right).$$
498: One can show by induction using Dodgson's determinant identity
499: (for a statement and a pretty proof of this identity see~\cite{Z2})
500: that $m_{i-1,i+k}$ equals the Euler continuant $[a_i,\dots,a_{i+k-1}]$,
501: that is, the determinant of the $k$-by-$k$ matrix
502: with entries $a_i,\dots,a_{i+k-1}$ down the diagonal,
503: 1's in the two flanking diagonals, and 0's everywhere else.
504: This is true for any array satisfying the frieze relation
505: whose initial row consists of 1's, whether or not it is a frieze pattern.
506: Thus, any numerical array constructed via the frieze relation
507: from initial data consisting of a first row of 1's
508: and a second row of integers
509: will be an array of integers,
510: since entries in subsequent rows are
511: equal to determinants of integer matrices.
512: (One caveat is in order here:
513: although the table of tridiagonal determinants
514: always satisfies the frieze relation,
515: it may not be possible to compute the table
516: using just the frieze relation,
517: since some of the expressions that arise
518: might be indeterminate fractions of the form $0/0$.)
519: %% This passage could be clearer, given how it passes back and forth
520: %% between the frieze-pattern context and a broader context. -- 5/19/08
521: However, for most choices of positive integers $a_1,\dots,a_n$,
522: the resulting table of integers
523: will not be an $(n-1)$-rowed frieze pattern,
524: because some entries lower down in the table will be negative (or vanish).
525: Indeed, Conway and Coxeter show that
526: every $(n-1)$-rowed frieze pattern whose entries are positive integers
527: arises from a triangulated $n$-gon in the fashion described above.
528:
529: \section{The sideways recurrence and its periodicity} \label{sec-sideways}
530:
531: Recall that any $(n-1)$-rowed array of real numbers
532: that begins and ends with rows of 1's
533: and satisfies the frieze relation in between,
534: with all rows having period $n$,
535: qualifies as a frieze pattern.
536:
537: Note that if the vertices $1,\dots,n$ of an $n$-gon lie on a circle
538: and we let $d_{i,j}$ be the distance between points $i$ and $j$,
539: Ptolemy's theorem on the lengths of the sides and diagonals
540: of an inscriptible quadrilateral gives us the three-term quadratic relation
541: $$d_{i,j} \, d_{i-1,j+1} + d_{i-1,i} \, d_{j-1,j} = d_{i-1,j} \, d_{i,j+1}$$
542: (with all subscripts interpreted mod $n$).
543: Hence the numbers $d_{i,j}$ with $i \neq j$,
544: arranged just as the numbers $m_{i,j}$ were,
545: form an $(n-1)$-rowed array that almost qualifies as a frieze pattern
546: (the array satisfies the frieze relation
547: and has glide-reflection symmetry
548: because $d_{i,j}=d_{j,i}$ for all $i,j$,
549: but the top and bottom rows do not in general consist of 1's).
550: The nicest case occurs when the $n$-gon is a regular $n$-gon of side-length 1;
551: then we get a genuine frieze pattern
552: and each row of the frieze pattern is constant.
553:
554: Another source of frieze patterns is an old result from spherical geometry:
555: the {\it pentagramma mirificum\/} of Napier and Gauss
556: embodies the assertion that the arc-lengths of the sides
557: in a right-angled spherical pentagram
558: can be arranged to form the middle two rows of a four-rowed frieze pattern.
559:
560: Conway and Coxeter show that frieze patterns
561: are easy to construct if one proceeds not from top to bottom
562: (since one is unlikely to choose numbers $a_1,\dots,a_n$ in the second row
563: that will yield all 1's in the $(n-1)$st row) but from left to right,
564: starting with a zig-zag of entries connecting the top and bottom rows
565: (where the zig-zag path need not alternate
566: between leftward steps and rightward steps
567: but may consist of any pattern of leftward steps and rightward steps),
568: using the sideways frieze recurrence
569: $$
570: \begin{array}{ccc}
571: & A & \\[1.5ex]
572: B & & C \\[1.5ex]
573: & D &
574: \end{array} \ \ : \ \ C = (AD+1)/B
575: $$
576: Although a priori one might imagine that repeating this recurrence
577: would lead one to non-integer rational numbers
578: whose numerators and denominators would get increasingly large
579: as one goes from left to right,
580: it turns out that the resulting pattern
581: necessarily repeats with period $n$,
582: and that all the numbers that appear are whole numbers
583: (provided that all the entries in the initial zig-zag
584: are equal to 1).
585:
586: E.g., consider the partial frieze pattern
587: $$
588: \begin{array}{ccccccccccc}
589: ... & 1 & & 1 & & 1 & & 1 & & 1 & ... \\[1.5ex]
590: & & x & & x' & & & & & & \\[1.5ex]
591: & y & & y' & & & & & & & \\[1.5ex]
592: & & z & & z' & & & & & & \\[1.5ex]
593: ... & 1 & & 1 & & 1 & & 1 & & 1 & ... \end{array}
594: $$
595: Given non-zero values of $x$, $y$, and $z$, one can successively compute
596: $y' = (xz+1)/y$, % xy^{-1}z + y^{-1}
597: $x' = (y'+1)/x$, % y^{-1}z + x^{-1}y^{-1} + x^{-1}
598: and
599: $z' = (y'+1)/z$, % xy^{-1} + y^{-1}z^{-1} + z^{-1}
600: obtaining a new zig-zag of entries $x',y',z'$
601: connecting the top and bottom rows.
602: It is clear that for generic choices of non-zero $x,y,z$,
603: one has $x',y',z'$ non-zero as well,
604: so the procedure can be repeated, yielding further zig-zags of entries.
605: After six iterations of the procedure
606: one recovers the original numbers $x,y,z$
607: six places to the right of their original position
608: (unless one has unluckily chosen $x,y,z$
609: so as to cause one to encounter an indeterminate expression
610: of the form $0/0$ from the recurrence),
611: and if we specialize to $x=y=z=1$,
612: we get the 5-row frieze pattern associated with Figure 1.
613:
614: To dodge the issue of indeterminate expressions of the form $0/0$,
615: we embrace indeterminacy of another sort by regarding $x,y,z$
616: as formal quantities, not specific numbers,
617: so that $x',y',z'$, etc.\ become rational functions of $x$, $y$, and $z$.
618: Then our recurrence ceases to be problematic.
619: Indeed, one finds that the rational functions that arise
620: are of a special kind,
621: namely, Laurent polynomials with positive coefficients.
622:
623: We can see why this is so by incorporating weighted edges
624: into our matchings model.
625: Returning to the triangulated hexagon from section~\ref{sec-frieze},
626: associate the values $x$, $y$, and $z$
627: with the diagonals joining vertices 2 and 6,
628: vertices 2 and 5, and vertices 3 and 5, respectively.
629: Call these the formal weights of the diagonals.
630: Also assign weight 1 to each of the 6 sides of the hexagon;
631: see Figure 3.
632:
633: $$
634: \pspicture(0,-.75)(9,5.796)
635: %\pspicture(0,-.5)(6,3.864)
636: %\pspicture(0,-1.0)(12,7.728)
637: %\psset{xunit=2,yunit=2}
638: \psset{xunit=1.5,yunit=1.5}
639: \pspolygon(2,0)(4,0)(5,1.732)(4,3.464)(2,3.464)(1,1.732)
640: \psline(2,0)(2,3.464)
641: \psline(2,3.464)(4,0)
642: \psline(4,0)(4,3.464)
643: \rput(1.4,.666){1}
644: \rput(1.4,2.798){1}
645: \rput(3.0,3.664){1}
646: \rput(4.6,2.798){1}
647: \rput(4.6,.666){1}
648: \rput(3.0,-.2){1}
649: \rput(1.8,1.732){$x$}
650: \rput(2.8,1.732){$y$}
651: \rput(4.2,1.732){$z$}
652: % \rput(1.8,-.4){6}
653: % \rput(4.2,-.4){5}
654: % \rput(5.4,1.732){4}
655: % \rput(4.2,3.864){3}
656: % \rput(1.8,3.864){2}
657: % \rput(0.6,1.732){1}
658: \endpspicture
659: $$
660: \begin{center}
661: Figure 3. A triangulated 6-gon with edge-weights.
662: \end{center}
663:
664: \bigskip
665:
666: \noindent
667: Now construct the graph $G$ from the triangulation as before,
668: this time assigning weights to all the edges.
669: Specifically, if $v$ is a black vertex of $G$
670: that corresponds to a vertex of the $n$-gon
671: and $w$ is a white vertex of $G$
672: that corresponds to a triangle in the triangulation $T$
673: that has $v$ as one of its three vertices
674: (and has $v'$ and $v''$ as the other two vertices),
675: then the edge in $G$ that joins $v$ and $w$ should be assigned the weight
676: of the side or diagonal in $T$ that joins $v'$ and $v''$;
677: see Figure 4.
678:
679: $$
680: %\pspicture(3,0)(15,10.392)
681: \pspicture(1.5,0)(7.5,5.196)
682: \psset{xunit=.5,yunit=.5}
683: \psline(5,5.196)(6,10.392)
684: \rput(5.1,7.796){$1$} % A
685: \psline(5,5.196)(6,0)
686: \rput(5.1,2.598){$1$} % B
687: \psline(8,3.464)(6,10.392)
688: \rput(7.4,6.928){$1$} % 1
689: \psline(8,3.464)(12,0)
690: \rput(9.8,1.532){$x$} % D
691: \psline(8,3.464)(6,0)
692: \rput(7.4,1.732){$y$} % E
693: \psline(10,6.928)(6,10.392)
694: \rput(8,9.080){$z$} % F
695: \psline(10,6.928)(12,10.392)
696: \rput(10.6,8.660){$y$} % G
697: \psline(10,6.928)(12,0)
698: \rput(10.6,3.464){$1$} % h
699: \psline(13,5.196)(12,10.392)
700: \rput(12.9,7.794){$1$} % I
701: \psline(13,5.196)(12,0)
702: \rput(12.9,2.598){$1$} % J
703: \psline(3,5.196)(5,5.196)
704: \rput(4,5.596){$x$} % K
705: \psline(15,5.196)(13,5.196)
706: \rput(14,5.596){$z$} % L
707: %\psset{linestyle=dashed}
708: %\psline(6,0)(12,0)
709: %\psline(12,0)(15,5.196)
710: %\psline(15,5.196)(12,10.392)
711: %\psline(12,10.392)(6,10.392)
712: %\psline(6,10.392)(3,5.196)
713: %\psline(3,5.196)(6,0)
714: %\psline(6,0)(6,10.392)
715: %\psline(6,10.392)(12,0)
716: %\psline(12,0)(12,10.392)
717: \pscircle[fillstyle=solid,fillcolor=black](6,0){.2}
718: \pscircle[fillstyle=solid,fillcolor=white](8,3.464){.2}
719: \pscircle[fillstyle=solid,fillcolor=black](12,0){.2}
720: \pscircle[fillstyle=solid,fillcolor=white](13,5.196){.2}
721: \pscircle[fillstyle=solid,fillcolor=black](12,10.392){.2}
722: \pscircle[fillstyle=solid,fillcolor=white](10,6.928){.2}
723: \pscircle[fillstyle=solid,fillcolor=black](6,10.392){.2}
724: \pscircle[fillstyle=solid,fillcolor=white](5,5.196){.2}
725: \pscircle[fillstyle=solid,fillcolor=black](3,5.196){.2}
726: \pscircle[fillstyle=solid,fillcolor=black](15,5.196){.2}
727: \endpspicture
728: $$
729: \begin{center}
730: Figure 4. The graph $G_{1,4}$ with edge-weights.
731: \end{center}
732:
733: \bigskip
734:
735: \noindent
736: We now define $W_{i,j}$ as the sum of the weights of
737: all the perfect matchings of the graph $G_{i,j}$
738: obtained from $G$ by deleting vertices $i$ and $j$
739: (and all their incident edges),
740: where the weight of a perfect matching
741: is the product of the weights of its constituent edges,
742: and we define $M_{i,j}$ as $W_{i,j}$ divided by
743: the product of the weights of all the diagonals
744: (this product is $xyz$ in our running example);
745: e.g., $W_{1,4} = 1 + 2y + y^2 + xz$
746: and $M_{1,4} = x^{-1} y^{-1} z^{-1} + 2 x^{-1} z^{-1} + x^{-1} y z^{-1} + y^{-1}$.
747: These $M_{i,j}$'s, which are rational functions of $x$, $y$, and $z$,
748: generalize the numbers denoted by $m_{i,j}$ earlier,
749: since we recover the $m_{i,j}$'s from the $M_{i,j}$'s
750: by setting $x=y=z=1$.
751: It is clear that each $W_{i,j}$ is a polynomial with positive coefficients,
752: so each $M_{i,j}$ is a Laurent polynomial with positive coefficients.
753: And, because of the normalization (division by $xyz$),
754: we have gotten each $M_{i,i+1}$ to equal 1.
755: So the table of rational functions $M_{i,j}$
756: is exactly what we get by running our recurrence from left to right.
757: When we pass from $x,y,z$ to $x',y',z'$,
758: we are effectively rotating our triangulation by one-sixth of a full turn;
759: six iterations bring us back to where we started.
760:
761: It is not hard to see that
762: the same approach works for any triangulation of an $n$-gon for any $n$,
763: and in this way we can prove:
764:
765: \begin{theorem} \label{thm-frieze}
766: Given any assignment of formal weights
767: to $n-3$ entries in an $(n-1)$-rowed table
768: that form a zig-zag joining the top row (consisting of all 1's)
769: to the bottom row (consisting of all 1's),
770: there is a unique assignment of rational functions
771: to all the entries in the table so that the frieze relation is satisfied.
772: These rational functions of the original $n-3$ variables
773: have glide-reflection symmetry that gives each row period $n$.
774: Furthermore, each of the rational functions in the table
775: is a Laurent polynomial with positive coefficients.
776: \end{theorem}
777:
778: Note that a zig-zag joining the top row to the bottom row
779: corresponds to a triangulation $T$ whose dual tree is just a path.
780: Not every triangulation is of this kind
781: (for instance, consider the hexagon shown in Figure 1
782: triangulated by diagonals joining vertices 2 and 4,
783: vertices 4 and 6, and vertices 2 and 6).
784: In general, the entries in a frieze pattern
785: that correspond to the diagonals of a triangulation $T$
786: do not form a zig-zag path,
787: so it is not clear from the frieze pattern
788: how to extend the known entries to the unknown entries
789: (e.g., for the triangulation described in the parenthetical remark
790: in the previous sentence,
791: if one assigns respective weights $x$, $y$, and $z$
792: to the specified diagonals,
793: one obtains the partial frieze pattern
794: $$
795: \begin{array}{cccccccccccccccccccc}
796: ...& 1 & & 1 & & 1 & & 1 & & 1 & & 1 & &... \\[1.2ex]
797: ...& & ? & & x & & ? & & y & & ? & & z &... \\[1.2ex]
798: ...& ? & & ? & & ? & & ? & & ? & & ? & &... \\[1.2ex]
799: ...& & y & & ? & & z & & ? & & x & & ? &... \\[1.2ex]
800: ...& 1 & & 1 & & 1 & & 1 & & 1 & & 1 & &...
801: \end{array}
802: $$
803: where the question marks refer to entries
804: whose values do not follow immediately
805: from the frieze recurrence).
806: In such a case, it is best to refer directly to the triangulation itself,
807: and to use a generalization of the frieze relation,
808: namely the (formal) Ptolemy relation~\cite{CP}
809: $$M_{i,j} \, M_{k,l} + M_{j,k} \, M_{i,l} = M_{i,k} \, M_{j,l}$$
810: where $i,j,k,l$ are four vertices of the $n$-gon listed in cyclic order.
811: (Conway and Coxeter~\cite{CoCo}
812: give spatially extended versions of the frieze relation
813: that are equivalent to special cases of the Ptolemy relation.)
814: Since every triangulation of a convex $n$-gon
815: can be obtained from every other by means of flips
816: that replace one diagonal of a quadrilateral by the other diagonal
817: (an observation that goes back at least as far as 1936~\cite{W}),
818: we can iterate the Ptolemy relation so as to solve for all of the $M_{i,j}$'s
819: in terms of the ones whose values were given.
820:
821: Up until now we have associated indeterminates
822: with the $n-3$ diagonals, but not the $n$ sides,
823: of our triangulated $n$-gon.
824: If we assign formal indeterminates to the sides as well as the diagonals
825: and carry out the construction of the edge-weighted graph $G_{i,j}$
826: (incorporating the $n$ new variables)
827: and the polynomial $W_{i,j}$
828: (the sum of the weights of all the perfect matchings of $G_{i,j}$),
829: and we define the Laurent polynomial $M_{i,j}$
830: as $W_{i,j}$ divided by the product of all $n-3$ diagonal-weights,
831: the proof of Theorem~\ref{thm-frieze} still goes through,
832: and one sees that the $M_{i,j}$'s form an array
833: in which the top and bottom rows
834: contain the indeterminates associated with the sides of the $n$-gon
835: and the intervening rows satisfy the modified frieze relation
836: $$
837: \begin{array}{ccccccccl}
838: X & & & & & & Y & & \\
839: & \jd & & A & & \ji & & & \\[1.2ex]
840: & & B & & C & & & : & \ \ BC = AD + XY \\
841: & \ji & & D & & \jd & & & \\
842: Y & & & & & & X & & \\
843: \end{array}
844: $$
845: where $X$ is the top entry in the diagonal containing $B$ and $D$
846: as well as the bottom entry in the diagonal containing $A$ and $C$
847: and $Y$ is the top entry in the diagonal containing $C$ and $D$
848: as well as the bottom entry in the diagonal containing $A$ and $B$.
849: Each entry in this generalized frieze pattern
850: is a Laurent monomial in the $2n-3$ variables
851: in which the $n$ variables associated with sides of the $n$-gon
852: occur in only with non-negative exponents.
853:
854: Our combinatorial construction
855: of Laurent polynomials associated with the sides and diagonals of an $n$-gon
856: is essentially nothing more than
857: the type $A$ case (more precisely, the $A_{n-3}$ case)
858: of the cluster algebra construction
859: of Fomin and Zelevinsky~\cite{FZY}.
860: The result that our matchings model yields,
861: stated in a self-contained way, is as follows:
862:
863: \begin{theorem}
864: Given any assignment of formal weights
865: $x_{i,j}$ to the $2n-3$ edges of a triangulated convex $n$-gon,
866: (where $x_{i,j}$ is associated with the edge joining vertices $i$ and $j$),
867: there is a unique assignment of rational functions
868: to all $n(n-3)/2$ diagonals of the $n$-gon
869: such that the rational functions assigned
870: to the four sides and two diagonals of any quadrilateral
871: determined by four of the $n$ vertices satisfy the Ptolemy relation.
872: These rational functions of the original $2n-3$ variables
873: are Laurent polynomials with positive coefficients.
874: \end{theorem}
875:
876: The formal weights are precisely the cluster variables
877: in the cluster algebra of type $A_{n-3}$,
878: and the triangulations are the clusters.
879: The periodicity phenomenon
880: is a special case of a more general periodicity conjectured by Zamolodchikov
881: and proved in the type $A$ case independently by Frenkel and Szenes
882: and by Gliozzi and Tateo; see~\cite{FZY} for details.
883: %% Also mention Dylan Thurston's work!
884: %% It gives a nice interpretation of these cluster variables
885: %% in terms of Penner coordinates (?) for hyperbolic structures
886: %% of the circle with punctures on its boundary. -- 5/19/08
887:
888: %% Talk about the denominators and their interpretation -- 5/28/08
889:
890: \section{Snake graphs} \label{sec-snake}
891:
892: The bipartite graphs $G_{i,j}$ obtained in section~\ref{sec-frieze},
893: when shorn of their forced edges (edges that belong to every perfect matching)
894: and their forbidden edges (edges that belong to no perfect matching),
895: have a direct combinatorial construction as ``snakes'' of 4-cycles,
896: obtained by repeating the process of adding a new 4-cycle
897: at the end of a snake.
898: More precisely: a snake of order 0 is just two vertices joined by an edge;
899: a snake of order 1 is a 4-cycle;
900: a snake of order 2 is a pair of 4-cycles sharing a single edge,
901: obtained by adjoining one 4-cycle to another along an edge;
902: and if one has a snake of order $k-1$
903: whose most recently added 4-cycle $C$ was adjoined along edge $e$,
904: one obtains a snake of order $k$
905: by adjoining a new 4-cycle
906: that shares some edge of $C$ other than $e$.
907: For example, Figure 5 shows a snake of order 6
908: obtained from a triangulated 9-gon
909: whose vertices are shown in black
910: (the two vertices represented by smaller black dots
911: are not part of the snake,
912: but they are included for clarity).
913: Given a triangulation $T$,
914: the only edges of $G_{i,j}$ that are neither forced nor forbidden
915: are those whose white endpoint
916: corresponds to a triangle in $T$
917: on the path of triangles joining vertices $i$ and $j$.
918: These edges form a snake-graph
919: whose twists and turns mimic those of the path of triangles.
920:
921: Up until now, we have used $n$ to denote the number of sides
922: of the polygon being triangulated,
923: but in this section it will be more convenient
924: to let $n+3$ denote the number of sides of the polygon,
925: and to make the additional assumption
926: that every triangle in the triangulation of the $(n+3)$-gon
927: occurs on the path of triangles from $i$ to $j$,
928: so that the graph $G_{i,j}$ has no forced or forbidden edges
929: and the snake-graph has order $n$.
930:
931: $$
932: \pspicture(-2.5,-2.5)(2.5,2.5)
933: \psset{xunit=2.5,yunit=2.5}
934: \psline(-0.830,0.146)(-.866,-.500)
935: \psline(-0.830,0.146)(-.643,.766)
936: \psline(-.643,.766)(-0.502,0.421)
937: \psline(-.866,-.500)(-0.502,0.421)
938: \psline(.000,1.000)(-0.502,0.421)
939: \psline(-.866,-.500)(-0.074,0.421)
940: \psline(.000,1.000)(-0.074,0.421)
941: \psline(.643,.766)(-0.074,0.421)
942: \psline(-.866,-.500)(0.254,0.146)
943: \psline(.643,.766)(0.254,0.146)
944: \psline(.985,.174)(0.254,0.146)
945: \psline(-.866,-.500)(-0.074,-0.422)
946: \psline(.985,.174)(-0.074,-0.422)
947: \psline(-.342,-.940)(-0.074,-0.422)
948: \psline(.985,.174)(0.328,-0.568)
949: \psline(-.342,-.940)(0.328,-0.568)
950: \psline(.342,-.940)(0.328,-0.568)
951: \psline(.985,.174)(0.730,-0.422)
952: \psline(.342,-.940)(0.730,-0.422)
953: \pscircle[fillstyle=solid,fillcolor=black](-.985,.174){.1}
954: \pscircle[fillstyle=solid,fillcolor=black](-.866,-.500){.2}
955: \pscircle[fillstyle=solid,fillcolor=black](-.643,.766){.2}
956: \pscircle[fillstyle=solid,fillcolor=black](.000,1.000){.2}
957: \pscircle[fillstyle=solid,fillcolor=black](.643,.766){.2}
958: \pscircle[fillstyle=solid,fillcolor=black](.985,.174){.2}
959: \pscircle[fillstyle=solid,fillcolor=black](-.342,-.940){.2}
960: \pscircle[fillstyle=solid,fillcolor=black](.342,-.940){.2}
961: \pscircle[fillstyle=solid,fillcolor=black](.866,-.500){.1}
962: \pscircle[fillstyle=solid,fillcolor=white](-0.830,0.146){.2}
963: \pscircle[fillstyle=solid,fillcolor=white](-0.502,0.421){.2}
964: \pscircle[fillstyle=solid,fillcolor=white](-0.074,0.421){.2}
965: \pscircle[fillstyle=solid,fillcolor=white](0.254,0.146){.2}
966: \pscircle[fillstyle=solid,fillcolor=white](-0.074,-0.422){.2}
967: \pscircle[fillstyle=solid,fillcolor=white](0.328,-0.568){.2}
968: \pscircle[fillstyle=solid,fillcolor=white](0.730,-0.422){.2}
969: \endpspicture
970: $$
971: \begin{center}
972: Figure 5. A snake of order 6.
973: \end{center}
974:
975: Although the operation of adding on a new 4-cycle, or ``box'',
976: can be done in three ways at each step
977: in the iterative construction of a snake,
978: for purposes of counting perfect matchings
979: there are really only two choices at each stage:
980: for $k \geq 3$,
981: the edge that joins the $k-1$st box to the $k$th
982: can either (1) be disjoint from
983: the edge that joins the $k-2$nd box to the $k-1$st
984: or (2) have an endpoint in common with it.
985: If we let $m_{k}$, $m_{k-1}$, and $m_{k-2}$
986: denote the number of perfect matchings of
987: the $k$th, $k-1$st, and $k-2$nd snakes
988: in the iterative process,
989: then (as we will now show)
990: in case 1, $m_{k} = m_{k-1} + m_{k-2}$,
991: while in case 2, $m_{k} = m_{k-1} + (m_{k-1} - m_{k-2})$;
992: that is, the three numbers
993: are either in ``Fibonacci progression'' or in arithmetic progression.
994: Refer to Figure 6,
995: where $u$ and $v$ are the vertices of the $k-2$nd snake
996: that are not part of the $k-3$rd snake,
997: $w$ and $x$ are the vertices of the $k-1$st snake
998: that are not part of the $k-2$nd snake,
999: and $y$ and $z$ are the vertices of the $k$th snake
1000: that are not part of the $k-1$st snake,
1001: in the fashion shown.
1002: In both cases, the difference $m_{k} - m_{k-1}$
1003: counts those perfect matchings of the $k$th snake
1004: that do not contain the edge $yz$
1005: and that are therefore forced to contain
1006: all the edges shown in bold in the figure.
1007: In case 1, these perfect matchings
1008: correspond to perfect matchings of the $k-2$nd snake:
1009: simply delete edges $wy$ and $xz$ (and the vertices they contain).
1010: In case 2, these perfect matchings
1011: correspond to those perfect matchings of the $k-1$st snake
1012: that do not arise from a perfect matching of the $k-2$nd snake
1013: by adjoining the edge $wx$:
1014: simply delete edges $vy$ and $xz$ and adjoin edge $vx$.
1015: % Original version
1016: $$
1017: \pspicture(0.5,0.5)(7.5,4.5)
1018: \rput(2,4){Case (1)}
1019: \rput(6.5,4){Case (2)}
1020: \psline(0.5,3)(0.8,3)
1021: \psline(1.2,3)(1.8,3)
1022: \psline[linewidth=3pt](2.2,3)(2.8,3)
1023: \psline(0.5,2)(0.8,2)
1024: \psline(1.2,2)(1.8,2)
1025: \psline[linewidth=3pt](2.2,2)(2.8,2)
1026: \psline(1,2.2)(1,2.8)
1027: \psline(2,2.2)(2,2.8)
1028: \psline(3,2.2)(3,2.8)
1029: \psline(5.5,3)(5.8,3)
1030: \psline[linewidth=3pt](6.2,3)(6.8,3)
1031: \psline(5.5,2)(5.8,2)
1032: \psline(6.2,2)(6.8,2)
1033: \psline(6.2,1)(6.8,1)
1034: \psline(6,2.2)(6,2.8)
1035: \psline(7,2.2)(7,2.8)
1036: \psline[linewidth=3pt](6,1.2)(6,1.8)
1037: \psline[linewidth=3pt](7,1.2)(7,1.8)
1038: \rput(1,3){$u$}
1039: \rput(1,2){$v$}
1040: \rput(2,3){$w$}
1041: \rput(2,2){$x$}
1042: \rput(3,3){$y$}
1043: \rput(3,2){$z$}
1044: \rput(6,3){$u$}
1045: \rput(6,2){$v$}
1046: \rput(7,3){$w$}
1047: \rput(7,2){$x$}
1048: \rput(6,1){$y$}
1049: \rput(7,1){$z$}
1050: \endpspicture
1051: $$
1052: \begin{center}
1053: Figure 6. The snake-graph recurrence.
1054: \end{center}
1055: % Modified version
1056: % $$
1057: % \pspicture(0.5,0.5)(7.5,4.5)
1058: % \rput(1.5,4){Case (1)}
1059: % \psline(0.5,3)(0.8,3)
1060: % \psline[linewidth=3pt](1.2,3)(1.8,3)
1061: % \psline(0.5,2)(0.8,2)
1062: % \psline(1.2,2)(1.8,2)
1063: % \psline(1.2,1)(1.8,1)
1064: % \psline(1,2.2)(1,2.8)
1065: % \psline(2,2.2)(2,2.8)
1066: % \psline[linewidth=3pt](1,1.2)(1,1.8)
1067: % \psline[linewidth=3pt](2,1.2)(2,1.8)
1068: % \rput(1,3){$u$}
1069: % \rput(1,2){$v$}
1070: % \rput(2,3){$w$}
1071: % \rput(2,2){$x$}
1072: % \rput(1,1){$y$}
1073: % \rput(2,1){$z$}
1074: % \rput(7,4){Case (2)}
1075: % \psline(5.5,3)(5.8,3)
1076: % \psline(6.2,3)(6.8,3)
1077: % \psline[linewidth=3pt](7.2,3)(7.8,3)
1078: % \psline(5.5,2)(5.8,2)
1079: % \psline(6.2,2)(6.8,2)
1080: % \psline[linewidth=3pt](7.2,2)(7.8,2)
1081: % \psline(6,2.2)(6,2.8)
1082: % \psline(7,2.2)(7,2.8)
1083: % \psline(8,2.2)(8,2.8)
1084: % \rput(6,3){$u$}
1085: % \rput(6,2){$v$}
1086: % \rput(7,3){$w$}
1087: % \rput(7,2){$x$}
1088: % \rput(8,3){$y$}
1089: % \rput(8,2){$z$}
1090: % \endpspicture
1091: % $$
1092: % \begin{center}
1093: % Figure 6. The snake-graph recurrence.
1094: % \end{center}
1095:
1096: In terms of the triangulation picture,
1097: a snake-graph of order $k$ corresponds to a chain of $k+1$ triangles,
1098: in which the $i$th triangle (for $2 \leq i \leq k+1$)
1099: consists of one edge $uv$ of the $i-1$st triangle
1100: (not the edge joining the $i-1$st and $i-2$nd triangles)
1101: together with a new vertex $x$
1102: and two edges $ux$, $vx$.
1103: Any two consecutive triangles in this chain share two vertices,
1104: and any three consecutive triangles in this chain share one vertex.
1105: If the last four consecutive triangles in the chain
1106: have no vertex in common, then we are in case 1;
1107: if they do have a vertex in common, we are in case 2.
1108: We can make a {\it code\/} of length $n-2$
1109: that contains this information.
1110: Thus, the snake-graph shown in Figure 5
1111: and the triangulation it arises from
1112: can be described (from left to right) by the code
1113: {\bf 2212}, indicating that as we move through the snake
1114: from left to right,
1115: we encounter case 2, case 2, case 1, and case 2.
1116: Two extreme cases are the snake with code {\bf 11}...{\bf 1}
1117: (the ``straight snake'')
1118: and the snake with code {\bf 22}...{\bf 2}
1119: (the ``fan snake'').
1120:
1121: For purposes of enumeration of matchings,
1122: every snake graph can be built as a chain of boxes
1123: where each new box is either added at the right of the preceding box
1124: or at the bottom of the preceding box.
1125: This is because the two geometrically distinct subcases of case 2
1126: are the same from the point of view of enumeration of matchings,
1127: even though they are not isomorphic as graphs.
1128: For instance, consider a triangulated $(n+3)$-gon
1129: in which all the diagonals emanate from a single vertex.
1130: Strictly speaking, the associated snake
1131: (with code {\bf 22}...{\bf 2})
1132: has all $n$ of its boxes sharing a single vertex.
1133: However, we can replace this by a snake of
1134: squares in the square grid,
1135: where new squares are alternately added
1136: at the right or at the bottom of the growing snake.
1137: Both snake-graphs have exactly $n+1$ perfect matchings.
1138: Similarly,
1139: the snake-graph shown in Figure 5
1140: and the snake-graph shown in Figure 7
1141: both have the code {\bf 2212} and both have 13 perfect matchings.
1142:
1143: $$
1144: \pspicture(0.9,0.9)(5.1,4.1)
1145: \psline(4,1)(5,1)
1146: \psline(2,2)(5,2)
1147: \psline(1,3)(5,3)
1148: \psline(1,4)(3,4)
1149: \psline(1,3)(1,4)
1150: \psline(2,2)(2,4)
1151: \psline(3,2)(3,4)
1152: \psline(4,1)(4,3)
1153: \psline(5,1)(5,3)
1154: \pscircle[fillstyle=solid,fillcolor=black](1,4){.1}
1155: \pscircle[fillstyle=solid,fillcolor=black](2,4){.1}
1156: \pscircle[fillstyle=solid,fillcolor=black](3,4){.1}
1157: \pscircle[fillstyle=solid,fillcolor=black](1,3){.1}
1158: \pscircle[fillstyle=solid,fillcolor=black](2,3){.1}
1159: \pscircle[fillstyle=solid,fillcolor=black](3,3){.1}
1160: \pscircle[fillstyle=solid,fillcolor=black](4,3){.1}
1161: \pscircle[fillstyle=solid,fillcolor=black](5,3){.1}
1162: \pscircle[fillstyle=solid,fillcolor=black](2,2){.1}
1163: \pscircle[fillstyle=solid,fillcolor=black](3,2){.1}
1164: \pscircle[fillstyle=solid,fillcolor=black](4,2){.1}
1165: \pscircle[fillstyle=solid,fillcolor=black](5,2){.1}
1166: \pscircle[fillstyle=solid,fillcolor=black](4,1){.1}
1167: \pscircle[fillstyle=solid,fillcolor=black](5,1){.1}
1168: \endpspicture
1169: $$
1170: \begin{center}
1171: Figure 7. Another snake of order 6.
1172: \end{center}
1173:
1174: Graphs made of chains of hexagons have been considered before,
1175: starting in the chemical literature
1176: on account of their relevance to the study of benzenoid hydrocarbons;
1177: an analogous theory applies there.
1178: %% Give a (correct) reference.
1179: %% Gordon and Davison? Or John and Sachs?
1180: %% Who gave the recursive counting scheme?
1181: To add the $k$th hexagon to the chain,
1182: we choose one of the three edges of the $k-1$st hexagon
1183: that has no endpoints in common with the
1184: edge that joins the $k-1$st and $k-2$nd hexagons in the chain.
1185: If these two edges are diametrically opposite one another
1186: in the hexagon that they both belong to,
1187: we are in case 2, and the relation
1188: $m_{k} - m_{k-1} = m_{k-1} - m_{k-2}$ applies;
1189: otherwise, we are in case 1, and the relation
1190: $m_{k} - m_{k-1} = m_{k-2}$ applies.
1191:
1192: A good way to understand what is going on here
1193: comes from consideration of products of the matrices
1194: $A = \left( \begin{array}{cc} 0 & 1 \\ 1 & 1 \end{array} \right)$
1195: and
1196: $B = \left( \begin{array}{cc} 1 & 1 \\ 1 & 0 \end{array} \right)$.
1197: A product of $n-1$ such matrices
1198: corresponds to a snake with $n$ boxes,
1199: where the presence of an $A$ (resp.\ $B$)
1200: as the $i$th factor in the matrix product
1201: (with $1 \leq i \leq n-1$)
1202: corresponds to a horizontal (resp.\ vertical)
1203: segment of the snake,
1204: with the $i+1$st box in the snake
1205: lying to the right of (resp.\ below)
1206: the $i$th box in the snake.
1207: For instance, the matrix product
1208: $ABAAB = \left( \begin{array}{cc} 2 & 1 \\ 7 & 3 \end{array} \right)$
1209: corresponds to the snake-graph shown in Figure 7,
1210: with code ${\bf 2212}$.
1211: More generally, a product of $n-1$ matrices,
1212: each of which is either $A$ or $B$,
1213: corresponds to a snake of order $n$
1214: whose code can be read off from the product by the following rule:
1215: If the $i$th and $i+1$st matrices in the product are the same,
1216: the $i$th element of the code is 1;
1217: otherwise, the $i$th element of the code is 2.
1218:
1219: The number of perfect matchings of a snake
1220: is equal to the sum of the entries of the associated matrix
1221: (so that in the specific example shown
1222: the number of perfect matchings is $2+1+7+3$).
1223: More specifically:
1224: the matrix entry in the upper left counts the perfect matchings
1225: in which both the upper-left vertex of the snake
1226: and the lower-right vertex of the snake
1227: are matched horizontally;
1228: the matrix entry in the upper right counts the perfect matchings
1229: in which the upper-left vertex is matched horizontally
1230: and the lower-right vertex is matched vertically;
1231: the matrix entry in the lower left counts the perfect matchings
1232: in which the upper-left vertex is matched vertically
1233: and the lower-right vertex is matched horizontally; and
1234: the matrix entry in the lower right counts the perfect matchings
1235: in which both the upper-left vertex
1236: and the lower-right vertex are matched vertically.
1237: This interpretation of the entries of the product matrix
1238: is easily verified by induction.
1239:
1240: $$
1241: \pspicture(0.5,0.5)(6.5,2.5)
1242: \psline(1,1)(2,2)
1243: \psline(2,1)(3,2)
1244: \psline(3,1)(4,2)
1245: \psline(4,1)(5,2)
1246: \psline(5,1)(6,2)
1247: \psline(1,2)(2,1)
1248: \psline(2,2)(3,1)
1249: \psline(3,2)(4,1)
1250: \psline(4,2)(5,1)
1251: \psline(5,2)(6,1)
1252: \psline(1,1)(2,1)
1253: \psline(2,2)(3,2)
1254: \psline(3,1)(5,1)
1255: \psline(5,2)(6,2)
1256: \pscircle[fillstyle=solid,fillcolor=black](1,1){.1}
1257: \pscircle[fillstyle=solid,fillcolor=black](2,1){.1}
1258: \pscircle[fillstyle=solid,fillcolor=black](3,1){.1}
1259: \pscircle[fillstyle=solid,fillcolor=black](4,1){.1}
1260: \pscircle[fillstyle=solid,fillcolor=black](5,1){.1}
1261: \pscircle[fillstyle=solid,fillcolor=black](6,1){.1}
1262: \pscircle[fillstyle=solid,fillcolor=black](1,2){.1}
1263: \pscircle[fillstyle=solid,fillcolor=black](2,2){.1}
1264: \pscircle[fillstyle=solid,fillcolor=black](3,2){.1}
1265: \pscircle[fillstyle=solid,fillcolor=black](4,2){.1}
1266: \pscircle[fillstyle=solid,fillcolor=black](5,2){.1}
1267: \pscircle[fillstyle=solid,fillcolor=black](6,2){.1}
1268: \endpspicture
1269: $$
1270: \begin{center}
1271: Figure 8. A paths model for snakes.
1272: \end{center}
1273:
1274: A different combinatorial model
1275: that gives the same numbers as the perfect matchings model
1276: arises from the fact that these numbers
1277: can be expressed as the sum of the entries in a matrix
1278: that is written as the product of matrices
1279: whose entries are all 0's and 1's
1280: (namely the matrices $A$ and $B$).
1281: More specifically, we can create a graph
1282: in which the number of paths from either of two source vertices
1283: to either of two target vertices
1284: is the same as the number of perfect matchings of a snake-graph.
1285: Figure 8 shows what the paths-graph looks like
1286: for the snake associated with the matrix-product $ABAAB$.
1287: Each factor of $A$ corresponds to a 4-vertex bipartite graph
1288: containing all edges from the left to the right
1289: except the edge connecting the top left to the top right,
1290: and each factor of $B$ corresponds to a 4-vertex bipartite graph
1291: containing all edges from the left to the right
1292: except the edge connecting the bottom left to the bottom right.
1293: Multiplication of matrices corresponds to adjunction of graphs,
1294: and the definition of matrix multiplication
1295: ensures that $i,j$th entry of the product
1296: equals the number of left-right paths
1297: joining the $j$th of the two leftmost vertices
1298: to the $i$th of the two rightmost vertices ($1 \leq i,j \leq 2$).
1299: (For more on this combinatorial aspect of matrix multiplication,
1300: see~\cite{Z1}.)
1301: Note that changing all $A$'s into $B$'s and vice versa
1302: simply flips the picture upside down.
1303:
1304: We can improve on this model
1305: by making a slight twist in our matrix-product,
1306: working instead with the matrices
1307: $L = \left( \begin{array}{cc} 1 & 1 \\ 0 & 1 \end{array} \right)$
1308: and
1309: $R = \left( \begin{array}{cc} 1 & 0 \\ 1 & 1 \end{array} \right)$.
1310: To turn an $AB$-product into an essentially equivalent $LR$-product,
1311: work from left to right,
1312: with the proviso that two factors in the $LR$ product should be equal
1313: if and only if two factors in the $AB$ product are {\it not\/}.
1314: Thus, the product $ABAAB$
1315: corresponds to either the product $LLLRR$ or the product $RRRLL$.
1316: Either way, we get a product-matrix
1317: whose four entries are, up to permutation,
1318: the same as the four entries of the $AB$ product,
1319: with the virtue that the picture no longer involves crossings.
1320: Figure 9, for instance, is the picture for $RRRLL$.
1321:
1322: $$
1323: \pspicture(0.5,0.5)(6.5,2.5)
1324: \psline(1,2)(6,2)
1325: \psline(1,1)(6,1)
1326: \psline(1,2)(2,1)
1327: \psline(2,2)(3,1)
1328: \psline(3,2)(4,1)
1329: \psline(4,1)(5,2)
1330: \psline(5,1)(6,2)
1331: \pscircle[fillstyle=solid,fillcolor=black](1,1){.1}
1332: \pscircle[fillstyle=solid,fillcolor=black](2,1){.1}
1333: \pscircle[fillstyle=solid,fillcolor=black](3,1){.1}
1334: \pscircle[fillstyle=solid,fillcolor=black](4,1){.1}
1335: \pscircle[fillstyle=solid,fillcolor=black](5,1){.1}
1336: \pscircle[fillstyle=solid,fillcolor=black](6,1){.1}
1337: \pscircle[fillstyle=solid,fillcolor=black](1,2){.1}
1338: \pscircle[fillstyle=solid,fillcolor=black](2,2){.1}
1339: \pscircle[fillstyle=solid,fillcolor=black](3,2){.1}
1340: \pscircle[fillstyle=solid,fillcolor=black](4,2){.1}
1341: \pscircle[fillstyle=solid,fillcolor=black](5,2){.1}
1342: \pscircle[fillstyle=solid,fillcolor=black](6,2){.1}
1343: %\rput(1,0.75){a}
1344: %\rput(2,0.8){b}
1345: %\rput(3,0.75){c}
1346: %\rput(4,0.8){d}
1347: %\rput(5,0.75){e}
1348: %\rput(6,0.8){f}
1349: %\rput(1,2.2){g}
1350: %\rput(2,2.3){h}
1351: %\rput(3,2.3){i}
1352: %\rput(4,2.3){j}
1353: %\rput(5,2.3){k}
1354: %\rput(6,2.3){l}
1355: \endpspicture
1356: $$
1357: \begin{center}
1358: Figure 9. A planar paths model for snakes.
1359: \end{center}
1360:
1361: A variant of this picture is shown in Figure 10.
1362: This is just like the Figure 9,
1363: except that we have added a vertex at the upper left
1364: that connects to the two previously leftmost vertices,
1365: and we have added a vertex at the lower right
1366: that connects to the two previously rightmost vertices,
1367: so that, where before we counted paths
1368: from either of the two leftmost vertices
1369: to either of the two rightmost vertices
1370: (obtaining four numbers that get added together),
1371: we now count paths
1372: from the unique leftmost vertex
1373: to the unique rightmost vertex.
1374: We have marked each vertex $v$ by a number
1375: that indicates the number of paths
1376: from the leftmost vertex to $v$.
1377: The leftmost vertex gets marked with a 1,
1378: and every other vertex gets marked with the sum
1379: of the numbers marking its (one or two) leftward neighbors.
1380: In terms of the triangulation,
1381: this means that we put 1's
1382: at the vertices of the initial triangle in the snake,
1383: and we proceed marking vertices along the snake all the way to its tail,
1384: where each new vertex is marked with the sum
1385: of the markings of the other two vertices
1386: of the triangle being added to the snake.
1387:
1388: $$
1389: \pspicture(-0.5,0.5)(7.5,2.5)
1390: \psline(0,2)(6,2)
1391: \psline(1,1)(7,1)
1392: \psline(0,2)(1,1)
1393: \psline(1,2)(2,1)
1394: \psline(2,2)(3,1)
1395: \psline(3,2)(4,1)
1396: \psline(4,1)(5,2)
1397: \psline(5,1)(6,2)
1398: \psline(6,2)(7,1)
1399: \pscircle[fillstyle=solid,fillcolor=black](1,1){.1}
1400: \pscircle[fillstyle=solid,fillcolor=black](2,1){.1}
1401: \pscircle[fillstyle=solid,fillcolor=black](3,1){.1}
1402: \pscircle[fillstyle=solid,fillcolor=black](4,1){.1}
1403: \pscircle[fillstyle=solid,fillcolor=black](5,1){.1}
1404: \pscircle[fillstyle=solid,fillcolor=black](6,1){.1}
1405: \pscircle[fillstyle=solid,fillcolor=black](7,1){.1}
1406: \pscircle[fillstyle=solid,fillcolor=black](0,2){.1}
1407: \pscircle[fillstyle=solid,fillcolor=black](1,2){.1}
1408: \pscircle[fillstyle=solid,fillcolor=black](2,2){.1}
1409: \pscircle[fillstyle=solid,fillcolor=black](3,2){.1}
1410: \pscircle[fillstyle=solid,fillcolor=black](4,2){.1}
1411: \pscircle[fillstyle=solid,fillcolor=black](5,2){.1}
1412: \pscircle[fillstyle=solid,fillcolor=black](6,2){.1}
1413: \rput(1,0.6){1}
1414: \rput(2,0.6){2}
1415: \rput(3,0.6){3}
1416: \rput(4,0.6){4}
1417: \rput(5,0.6){4}
1418: \rput(6,0.6){4}
1419: \rput(7,0.6){13}
1420: \rput(0,2.4){1}
1421: \rput(1,2.4){1}
1422: \rput(2,2.4){1}
1423: \rput(3,2.4){1}
1424: \rput(4,2.4){1}
1425: \rput(5,2.4){5}
1426: \rput(6,2.4){9}
1427: \endpspicture
1428: $$
1429: \begin{center}
1430: Figure 10. Another planar paths model for snakes.
1431: \end{center}
1432:
1433: The marking scheme of Figure 10 is in fact nothing more than
1434: a slight variation on
1435: Conway and Coxeter's method of computing entries in frieze patterns
1436: by successively marking the vertices in a triangulation.
1437: Conway and Coxeter mark a single vertex with a 0, all its neighbors with 1's,
1438: and proceed from there;
1439: since we are pruning away all the side-branches
1440: of the graph $G_{i,j}$ until all that remains is a snake,
1441: we are effectively limiting ourselves to the case
1442: where the vertex to be marked with a 0 has only two neighbors.
1443: In this case, the only difference
1444: between our marking scheme and Conway and Coxeter's
1445: is that they mark the initial vertex with a 0
1446: where we mark it with a 1.
1447: Figure 11 shows what Conway and Coxeter's scheme looks like for the snake
1448: whose different representations are shown in Figures 5 through 10.
1449: The 9-gon being triangulated is not convex,
1450: but that is not a problem since we are
1451: dealing with triangulations purely combinatorially.
1452:
1453: $$
1454: \psset{xunit=0.7,yunit=0.7}
1455: % \pspicture(0,.866)(6,7.794)
1456: \pspicture(0,1.366)(6,7.294)
1457: \psline(3,1.732)(5,1.732)
1458: \psline(2,3.464)(4,3.464)
1459: \psline(1,5.196)(5,5.196)
1460: \psline(2,6.928)(4,6.928)
1461: \psline(3,1.732)(5,5.196)
1462: \psline(2,3.464)(4,6.928)
1463: \psline(1,5.196)(2,6.928)
1464: \psline(2,3.464)(3,1.732)
1465: \psline(2,6.928)(5,1.732)
1466: \psline(4,6.928)(5,5.196)
1467: \rput(0.8,5.196){0}
1468: \rput(1.8,6.928){1}
1469: \rput(2.7,4.996){1}
1470: \rput(4.2,6.928){2}
1471: \rput(5.2,5.196){3}
1472: \rput(4.2,3.464){4}
1473: \rput(1.8,3.464){5}
1474: \rput(2.8,1.732){9}
1475: \rput(5.2,1.732){13}
1476: \endpspicture
1477: $$
1478: \begin{center}
1479: Figure 11. Conway and Coxeter's marking scheme.
1480: \end{center}
1481:
1482: At this point it should be mentioned that there is a link
1483: between the directed path model of Figure 10,
1484: the hexagon snake model,
1485: and the square snake model,
1486: by way of a multigraph matching model
1487: that is in turn related
1488: to the strip-tiling model of Benjamin and Quinn~\cite{BQ}.
1489: We start by making use of a correspondence
1490: that has been rediscovered a number of times,
1491: starting as far back as 1952 \cite{GD} \cite{S} \cite{Kup}:
1492:
1493: Proposition: Let $D$ be a directed acyclic graph with vertex set $V$,
1494: where $m$ vertices $s_1,\dots,s_m$ have been designated as sources
1495: and $m$ other vertices $t_1,\dots,t_m$ have been designated as targets.
1496: (Since $D$ is acyclic, self-loops are forbidden,
1497: but $D$ is permitted to have multiple edges.)
1498: Create an undirected graph $G$ with vertex set $V \times \{1,2\}$
1499: and two kinds of edges:
1500: for each vertex $v$ of $D$,
1501: $D'$ contains an edge joining $(v,1)$ and $(v,2)$,
1502: and for each directed edge $e: v \rightarrow w$ of $D$,
1503: $D'$ contains an edge joining $(v,2)$ and $(w,1)$.
1504: (If $D$ has more than one directed edge from $v$ to $w$,
1505: $G$ has just as many edges joining $(v,2)$ and $(w,1)$.)
1506: Let $H$ be the induced subgraph of $G$
1507: obtained by deleting all the vertices $(s_i,1)$ and $(t_i,2)$
1508: ($1 \leq i \leq m$)
1509: and all incident edges.
1510: Then the perfect matchings of $H$
1511: are equinumerous with the ways to join
1512: the $m$ sources to the $m$ targets by $m$ edge-disjoint paths in $D$
1513: (which need not connect $s_1$ to $t_1$ etc.).
1514: Specifically, given such a collection of $m$ paths,
1515: take each arc $e: v \rightarrow w$ that belongs to one of the paths
1516: and replace it by the corresponding edge joining $(v,2)$ and $(w,1)$ in $H$,
1517: and replace each vertex $v$ in $D$ that does not lie on any of the paths
1518: by the edge joining $(v,1)$ and $(v,2)$ in $H$.
1519: It is easy to check that this yields a perfect matching of $H$,
1520: and it is also easy to show that every perfect matching
1521: arises in a unique fashion in this way.
1522:
1523: If we carry out this operation with the directed graph $D$
1524: shown in Figure 10
1525: (where all edges are oriented from left to right
1526: with the leftmost vertex the sole source
1527: and the rightmost vertex the sole target),
1528: we obtain a graph $G$ composed of 6-cycles (hexagons),
1529: as shown in Figure 12.
1530: $$
1531: \psset{xunit=1.0,yunit=1.0}
1532: \pspicture(-0.5,0.5)(7.5,2.5)
1533: \psline(0,2)(6.2,2)
1534: \psline(0.8,1)(6.8,1)
1535: \psline(0,2)(0.8,1)
1536: \psline(1.2,2)(1.8,1)
1537: \psline(2.2,2)(2.8,1)
1538: \psline(3.2,2)(3.8,1)
1539: \psline(4.2,1)(4.8,2)
1540: \psline(5.2,1)(5.8,2)
1541: \psline(6.2,2)(6.8,1)
1542: \pscircle[fillstyle=solid,fillcolor=black](0.8,1){.1}
1543: \pscircle[fillstyle=solid,fillcolor=black](1.2,1){.1}
1544: \pscircle[fillstyle=solid,fillcolor=black](1.8,1){.1}
1545: \pscircle[fillstyle=solid,fillcolor=black](2.2,1){.1}
1546: \pscircle[fillstyle=solid,fillcolor=black](2.8,1){.1}
1547: \pscircle[fillstyle=solid,fillcolor=black](3.2,1){.1}
1548: \pscircle[fillstyle=solid,fillcolor=black](3.8,1){.1}
1549: \pscircle[fillstyle=solid,fillcolor=black](4.2,1){.1}
1550: \pscircle[fillstyle=solid,fillcolor=black](4.8,1){.1}
1551: \pscircle[fillstyle=solid,fillcolor=black](5.2,1){.1}
1552: \pscircle[fillstyle=solid,fillcolor=black](5.8,1){.1}
1553: \pscircle[fillstyle=solid,fillcolor=black](6.2,1){.1}
1554: \pscircle[fillstyle=solid,fillcolor=black](6.8,1){.1}
1555: \pscircle[fillstyle=solid,fillcolor=black](0,2){.1}
1556: \pscircle[fillstyle=solid,fillcolor=black](0.8,2){.1}
1557: \pscircle[fillstyle=solid,fillcolor=black](1.2,2){.1}
1558: \pscircle[fillstyle=solid,fillcolor=black](1.8,2){.1}
1559: \pscircle[fillstyle=solid,fillcolor=black](2.2,2){.1}
1560: \pscircle[fillstyle=solid,fillcolor=black](2.8,2){.1}
1561: \pscircle[fillstyle=solid,fillcolor=black](3.2,2){.1}
1562: \pscircle[fillstyle=solid,fillcolor=black](3.8,2){.1}
1563: \pscircle[fillstyle=solid,fillcolor=black](4.2,2){.1}
1564: \pscircle[fillstyle=solid,fillcolor=black](4.8,2){.1}
1565: \pscircle[fillstyle=solid,fillcolor=black](5.2,2){.1}
1566: \pscircle[fillstyle=solid,fillcolor=black](5.8,2){.1}
1567: \pscircle[fillstyle=solid,fillcolor=black](6.2,2){.1}
1568: \endpspicture
1569: $$
1570: \begin{center}
1571: Figure 12. A hexagon snake.
1572: \end{center}
1573:
1574: \noindent
1575: Figure 13 shows how a particular path in $D$
1576: corresponds to a particular perfect matching in $G$
1577: (as described by the above proof).
1578:
1579: $$
1580: \pspicture(-0.5,0.5)(7.5,5.5)
1581: \psline(0,5)(6,5)
1582: \psline(1,4)(7,4)
1583: \psline(0,5)(1,4)
1584: \psline(1,5)(2,4)
1585: \psline(2,5)(3,4)
1586: \psline(3,5)(4,4)
1587: \psline(4,4)(5,5)
1588: \psline(5,4)(6,5)
1589: \psline(6,5)(7,4)
1590: \psline[linewidth=3pt](0,5)(2,5)
1591: \psline[linewidth=3pt](2,5)(3,4)
1592: \psline[linewidth=3pt](3,4)(5,4)
1593: \psline[linewidth=3pt](5,4)(6,5)
1594: \psline[linewidth=3pt](6,5)(7,4)
1595: \pscircle[fillstyle=solid,fillcolor=black](1,4){.1}
1596: \pscircle[fillstyle=solid,fillcolor=black](2,4){.1}
1597: \pscircle[fillstyle=solid,fillcolor=black](3,4){.1}
1598: \pscircle[fillstyle=solid,fillcolor=black](4,4){.1}
1599: \pscircle[fillstyle=solid,fillcolor=black](5,4){.1}
1600: \pscircle[fillstyle=solid,fillcolor=black](6,4){.1}
1601: \pscircle[fillstyle=solid,fillcolor=black](7,4){.1}
1602: \pscircle[fillstyle=solid,fillcolor=black](0,5){.1}
1603: \pscircle[fillstyle=solid,fillcolor=black](1,5){.1}
1604: \pscircle[fillstyle=solid,fillcolor=black](2,5){.1}
1605: \pscircle[fillstyle=solid,fillcolor=black](3,5){.1}
1606: \pscircle[fillstyle=solid,fillcolor=black](4,5){.1}
1607: \pscircle[fillstyle=solid,fillcolor=black](5,5){.1}
1608: \pscircle[fillstyle=solid,fillcolor=black](6,5){.1}
1609: \rput(3.5,3){\Large{$\downarrow$}}
1610: \psline(0.2,2)(6.2,2)
1611: \psline(0.8,1)(6.8,1)
1612: \psline(0.2,2)(0.8,1)
1613: \psline(1.2,2)(1.8,1)
1614: \psline(2.2,2)(2.8,1)
1615: \psline(3.2,2)(3.8,1)
1616: \psline(4.2,1)(4.8,2)
1617: \psline(5.2,1)(5.8,2)
1618: \psline(6.2,2)(6.8,1)
1619: \psline[linewidth=3pt](0.8,1)(1.2,1)
1620: \psline[linewidth=3pt](1.8,1)(2.2,1)
1621: \psline[linewidth=3pt](3.2,1)(3.8,1)
1622: \psline[linewidth=3pt](4.2,1)(4.8,1)
1623: \psline[linewidth=3pt](5.8,1)(6.2,1)
1624: \psline[linewidth=3pt](0.2,2)(0.8,2)
1625: \psline[linewidth=3pt](1.2,2)(1.8,2)
1626: \psline[linewidth=3pt](2.8,2)(3.2,2)
1627: \psline[linewidth=3pt](3.8,2)(4.2,2)
1628: \psline[linewidth=3pt](4.8,2)(5.2,2)
1629: \psline[linewidth=3pt](2.2,2)(2.8,1)
1630: \psline[linewidth=3pt](5.2,1)(5.8,2)
1631: \psline[linewidth=3pt](6.2,2)(6.8,1)
1632: \pscircle[fillstyle=solid,fillcolor=black](0.8,1){.1}
1633: \pscircle[fillstyle=solid,fillcolor=black](1.2,1){.1}
1634: \pscircle[fillstyle=solid,fillcolor=black](1.8,1){.1}
1635: \pscircle[fillstyle=solid,fillcolor=black](2.2,1){.1}
1636: \pscircle[fillstyle=solid,fillcolor=black](2.8,1){.1}
1637: \pscircle[fillstyle=solid,fillcolor=black](3.2,1){.1}
1638: \pscircle[fillstyle=solid,fillcolor=black](3.8,1){.1}
1639: \pscircle[fillstyle=solid,fillcolor=black](4.2,1){.1}
1640: \pscircle[fillstyle=solid,fillcolor=black](4.8,1){.1}
1641: \pscircle[fillstyle=solid,fillcolor=black](5.2,1){.1}
1642: \pscircle[fillstyle=solid,fillcolor=black](5.8,1){.1}
1643: \pscircle[fillstyle=solid,fillcolor=black](6.2,1){.1}
1644: \pscircle[fillstyle=solid,fillcolor=black](6.8,1){.1}
1645: \pscircle[fillstyle=solid,fillcolor=black](0.2,2){.1}
1646: \pscircle[fillstyle=solid,fillcolor=black](0.8,2){.1}
1647: \pscircle[fillstyle=solid,fillcolor=black](1.2,2){.1}
1648: \pscircle[fillstyle=solid,fillcolor=black](1.8,2){.1}
1649: \pscircle[fillstyle=solid,fillcolor=black](2.2,2){.1}
1650: \pscircle[fillstyle=solid,fillcolor=black](2.8,2){.1}
1651: \pscircle[fillstyle=solid,fillcolor=black](3.2,2){.1}
1652: \pscircle[fillstyle=solid,fillcolor=black](3.8,2){.1}
1653: \pscircle[fillstyle=solid,fillcolor=black](4.2,2){.1}
1654: \pscircle[fillstyle=solid,fillcolor=black](4.8,2){.1}
1655: \pscircle[fillstyle=solid,fillcolor=black](5.2,2){.1}
1656: \pscircle[fillstyle=solid,fillcolor=black](5.8,2){.1}
1657: \pscircle[fillstyle=solid,fillcolor=black](6.2,2){.1}
1658: \endpspicture
1659: $$
1660: \begin{center}
1661: Figure 13. From path systems to perfect matchings.
1662: \end{center}
1663:
1664: The hexagon snake of Figure 12
1665: can be drawn as a snake of regular hexagons,
1666: as shown in Figure 14.
1667: If we turn the figure on its side,
1668: so that the leftmost square is at the top,
1669: we can see how the turns of the snake
1670: correspond to the symbols in its $L,R$-string $RRRLL$.
1671:
1672: $$
1673: \psset{xunit=0.25,yunit=0.433}
1674: \pspicture(-1.5,1.5)(21,6.5)
1675: %\psset{xunit=0.5,yunit=0.866}
1676: \psline(1,3)(2,2)
1677: \psline(1,3)(2,4)
1678: \psline(2,2)(4,2)
1679: \psline(2,4)(4,4)
1680: \psline(4,2)(5,3)
1681: \psline(4,4)(5,3)
1682: \psline(4,4)(5,5)
1683: \psline(5,3)(7,3)
1684: \psline(5,5)(7,5)
1685: \psline(7,3)(8,4)
1686: \psline(7,5)(8,4)
1687: \psline(7,5)(8,6)
1688: \psline(8,4)(10,4)
1689: \psline(8,6)(10,6)
1690: \psline(10,4)(11,5)
1691: \psline(10,6)(11,5)
1692: \psline(10,6)(11,7)
1693: \psline(11,5)(13,5)
1694: \psline(11,7)(13,7)
1695: \psline(13,5)(14,4)
1696: \psline(13,5)(14,6)
1697: \psline(13,7)(14,6)
1698: \psline(14,4)(16,4)
1699: \psline(14,6)(16,6)
1700: \psline(16,4)(17,3)
1701: \psline(16,4)(17,5)
1702: \psline(16,6)(17,5)
1703: \psline(17,3)(19,3)
1704: \psline(17,5)(19,5)
1705: \psline(19,3)(20,4)
1706: \psline(19,5)(20,4)
1707: \pscircle[fillstyle=solid,fillcolor=black](1,3){.1}
1708: \pscircle[fillstyle=solid,fillcolor=black](2,2){.1}
1709: \pscircle[fillstyle=solid,fillcolor=black](2,4){.1}
1710: \pscircle[fillstyle=solid,fillcolor=black](4,2){.1}
1711: \pscircle[fillstyle=solid,fillcolor=black](4,4){.1}
1712: \pscircle[fillstyle=solid,fillcolor=black](5,3){.1}
1713: \pscircle[fillstyle=solid,fillcolor=black](5,5){.1}
1714: \pscircle[fillstyle=solid,fillcolor=black](7,3){.1}
1715: \pscircle[fillstyle=solid,fillcolor=black](7,5){.1}
1716: \pscircle[fillstyle=solid,fillcolor=black](8,4){.1}
1717: \pscircle[fillstyle=solid,fillcolor=black](8,6){.1}
1718: \pscircle[fillstyle=solid,fillcolor=black](10,4){.1}
1719: \pscircle[fillstyle=solid,fillcolor=black](10,6){.1}
1720: \pscircle[fillstyle=solid,fillcolor=black](11,5){.1}
1721: \pscircle[fillstyle=solid,fillcolor=black](11,7){.1}
1722: \pscircle[fillstyle=solid,fillcolor=black](13,5){.1}
1723: \pscircle[fillstyle=solid,fillcolor=black](13,7){.1}
1724: \pscircle[fillstyle=solid,fillcolor=black](14,4){.1}
1725: \pscircle[fillstyle=solid,fillcolor=black](14,6){.1}
1726: \pscircle[fillstyle=solid,fillcolor=black](16,4){.1}
1727: \pscircle[fillstyle=solid,fillcolor=black](16,6){.1}
1728: \pscircle[fillstyle=solid,fillcolor=black](17,3){.1}
1729: \pscircle[fillstyle=solid,fillcolor=black](17,5){.1}
1730: \pscircle[fillstyle=solid,fillcolor=black](19,3){.1}
1731: \pscircle[fillstyle=solid,fillcolor=black](19,5){.1}
1732: \pscircle[fillstyle=solid,fillcolor=black](20,4){.1}
1733: \endpspicture
1734: $$
1735: \begin{center}
1736: Figure 14. A hexagon snake with regular hexagons.
1737: \end{center}
1738:
1739: We also make use of an even simpler proposition
1740: that is part of the folklore of perfect matchings:
1741: Suppose $v$ is a vertex of degree 2 in a graph $G$,
1742: with neighbors $w_1$ and $w_2$.
1743: Let $G'$ be the graph obtained from $G$
1744: by deleting $v$ and its two edges
1745: and identifying vertices $w_1$ and $w_2$,
1746: so that the new amalgamated vertex (call it $w$)
1747: inherits the neighbors of $w_1$ and $w_2$.
1748: (Specifically, if $x$ is some vertex
1749: that in $G$ is connected to $w_1$ by $m_1$ edges
1750: and connected to $w_2$ by $m_2$ edges,
1751: then in $G'$, $x$ is connected to $w$ by $m_1+m_2$ edges.)
1752: Then the perfect matchings of $G$
1753: are equinumerous with the perfect matchings of $G'$.
1754: Specifically, given a perfect matching of $G$
1755: in which $v$ is connected to one of $w_1,w_2$
1756: and the other is connected a vertex $x$,
1757: construct a perfect matching of $G'$
1758: in which $w$ is connected to $x$
1759: and all other edges are unaffected
1760: (in the case where there are multiple edges from $w$ to $x$,
1761: one uses the edge that is associated with
1762: the specific edge of the matching of $G$ that contains $x$).
1763:
1764: Using this path-contraction operation,
1765: one can show that enumeration of perfect matchings
1766: of an arbitrary snake formed from $n$ cycles of even order
1767: (i.e., any combination of 4-cycles, 6-cycles, etc., arranged in a chain
1768: consisting of $n$ cycles)
1769: reduces to enumeration of perfect matchings
1770: of an ordinary straight snakes (made of 4-cycles)
1771: in which the edges shared by one 4-cycle and the next
1772: are allowed to have multiplicity,
1773: with multiplicities adding up to $n+1$.
1774: These multiplicities can be easily read off from
1775: the $L,R$-string associated with the snake:
1776: simply duplicate the first and last symbols of the string,
1777: and then write down the run-lengths.
1778: For instance, the $L,R$-string $RRRLL$
1779: becomes $RRRRLLL$ when the first and last symbols are duplicated,
1780: which gives the sequence of run-lengths $4,3$,
1781: so that the graphs shown in Figures 7 and 13,
1782: when contracted,
1783: both become the multigraph shown in Figure 15,
1784: where the 4 represents 4 parallel edges
1785: and the 3 represents 3 parallel edges.
1786:
1787: $$
1788: \psset{xunit=0.7,yunit=0.7}
1789: \pspicture(0.5,0.5)(3.5,3.5)
1790: \psline(1,1)(3,1)
1791: \psline(1,3)(3,3)
1792: \psline(1,1)(1,3)
1793: \psline(3,1)(3,3)
1794: \rput(0.8,2){4}
1795: \rput(3.2,2){3}
1796: \pscircle[fillstyle=solid,fillcolor=black](1,1){.1}
1797: \pscircle[fillstyle=solid,fillcolor=black](1,3){.1}
1798: \pscircle[fillstyle=solid,fillcolor=black](3,1){.1}
1799: \pscircle[fillstyle=solid,fillcolor=black](3,3){.1}
1800: \endpspicture
1801: $$
1802: \begin{center}
1803: Figure 15. A (short) straight snake with multiplicities.
1804: \end{center}
1805:
1806: The perfect matchings of such a weighted graph
1807: can in turn be associated with strip-tilings
1808: of the sort considered by Benjamin and Quinn~\cite{BQ}.
1809: Specifically, suppose we have a straight snake of order $n$
1810: whose $n+1$ vertical edges have multiplicities $r_0,r_1,\dots,r_N$.
1811: Then we associate this with a 1-by-$(n+1)$ rectangular strip
1812: that is to be covered by stackable 1-by-1 square tiles
1813: and non-stackable 1-by-2 rectangular tiles (``dominos''),
1814: where each square in the strip must be covered by at least one tile,
1815: and where square tiles may be stacked to height $r_i$
1816: at the $i$th square of the strip.
1817: E.g., for the graph shown in Figure 15,
1818: the associated strip-tiling problem
1819: would involve a strip consisting of two squares,
1820: which can either be tiled by a single domino
1821: or by two non-empty stacks of squares
1822: (up to 4 squares in the left stack
1823: and up to 3 squares in the right stack).
1824:
1825: By this point in the article, many readers will have recognized
1826: that our combinatorial model is not too far removed
1827: from the theory of continued fractions.
1828: Work of Benjamin and Quinn, in the context of the strip-tiling model,
1829: shows how combinatorial models
1830: can illuminate facts about continued fractions
1831: (especially those like~\cite{BQS} and~\cite{BZ} that involve
1832: reversing the order of the convergents:
1833: this operation seems unnatural from the point of view
1834: of the definition of continued fractions,
1835: inasmuch as it switches the high-order and low-order parts
1836: of the continued fraction representation,
1837: but the operation is extremely natural
1838: for tilings of a strip).
1839:
1840: %Here we pause only to mention how
1841: %the Stern-Brocot tree is related to
1842: %our snake-graph construction.
1843: %Specifically, consider a snake of order $n$,
1844: %and the snake of order $n-1$ obtained from it
1845: %by deleting the last box.
1846: %The former has $a$ perfect matchings
1847: %while the latter has $b$ perfect matchings,
1848: %where the fraction $b/a$ is the rational number
1849: %associated with the $L,R$-representation of the snake.
1850: %Note that $b/a$ is also equal to the probability
1851: %that a perfect matching of the snake of order $n$
1852: %chosen uniformly at random
1853: %contains the ``head'' of the snake
1854: %(the edge whose two endpoints do not belong
1855: %to the snake of order $n-1$).
1856: %For example, the snake in Figure 7 has 13 perfect matchings,
1857: %and the snake obtained by deleting the last box has 9 perfect matchings.
1858: %Hence the probability that a uniformly random perfect matching
1859: %of the larger snake
1860: %contains the bottom-most horizontal edge
1861: %is 9/13,
1862: %which can also be obtained from the Stern-Brocot process:
1863:
1864: There is a different way to relate frieze patterns to snake-graphs,
1865: where we count paths in the snake-graphs themselves.
1866: For instance, the number 13,
1867: whose various enumerative interpretations
1868: we have followed throughout this section,
1869: also occurs as the number of paths
1870: from the leftmost vertex to the rightmost vertex
1871: in the hexagon snake shown in Figure 14.
1872: To see why, note that for purposes of enumerating such paths,
1873: we can shrink each horizontal edge in Figure 14 to a point
1874: (identifying the two endpoints),
1875: obtaining a square snake (see Figure 16)
1876: that is combinatorially the same as
1877: the square snake shown in Figure 10.
1878: It should be stressed that this square snake is not the square snake
1879: we started with (shown in Figure 7).
1880: It is ``dual'' to the original square snake,
1881: making a bend where the original snake goes straight
1882: and going straight where the original snake makes a bend.
1883: (Equivalently, the code of the first snake has a {\bf 1}
1884: where the code of the second snake has a {\bf 2},
1885: and vice versa.)
1886: Enumerating perfect matchings of each snake
1887: is equivalent to counting paths its dual (from head to tail).
1888: For instance, the snake in Figure 7 has 13 perfect matchings
1889: and 19 paths from head to tail,
1890: while the snake in Figure 16 has 19 perfect matchings
1891: and 13 paths from head to tail.
1892:
1893: $$
1894: \psset{xunit=0.7,yunit=0.7}
1895: \pspicture(1,1)(8,6)
1896: \psline(1,2)(2,1)
1897: \psline(1,2)(2,3)
1898: \psline(2,1)(3,2)
1899: \psline(2,3)(3,2)
1900: \psline(2,3)(3,4)
1901: \psline(3,2)(4,3)
1902: \psline(3,4)(4,3)
1903: \psline(3,4)(4,5)
1904: \psline(4,3)(5,4)
1905: \psline(4,5)(5,4)
1906: \psline(4,5)(5,6)
1907: \psline(5,4)(6,3)
1908: \psline(5,4)(6,5)
1909: \psline(5,6)(6,5)
1910: \psline(6,3)(7,2)
1911: \psline(6,3)(7,4)
1912: \psline(6,5)(7,4)
1913: \psline(7,2)(8,3)
1914: \psline(7,4)(8,3)
1915: \endpspicture
1916: $$
1917: \begin{center}
1918: Figure 16. A dual snake.
1919: \end{center}
1920:
1921: $$
1922: \psset{xunit=0.7,yunit=0.7}
1923: \pspicture(0,0)(9,9)
1924: \rput(4,1){$1_d$}
1925: \rput(3,2){$1_c$}
1926: \rput(5,2){4}
1927: \rput(2,3){1}
1928: \rput(4,3){3}
1929: \rput(6,3){7}
1930: \rput(1,4){1}
1931: \rput(3,4){2}
1932: \rput(5,4){5}
1933: \rput(7,4){10}
1934: \rput(2,5){1}
1935: \rput(4,5){3}
1936: \rput(6,5){7}
1937: \rput(8,5){13}
1938: \rput(3,6){1}
1939: \rput(5,6){4}
1940: \rput(7,6){9}
1941: \rput(4,7){$1_b$}
1942: \rput(6,7){5}
1943: \rput(5,8){$1_a$}
1944: \psline(3.8,1.2)(3.2,1.8)
1945: \psline(2.8,2.2)(2.2,2.8)
1946: \psline(1.8,3.2)(1.2,3.8)
1947: \psline(4.8,2.2)(4.2,2.8)
1948: \psline(3.8,3.2)(3.2,3.8)
1949: \psline(2.8,4.2)(2.2,4.8)
1950: \psline(3.8,5.2)(3.2,5.8)
1951: \psline(4.8,6.2)(4.2,6.8)
1952: \psline(5.8,7.2)(5.2,7.8)
1953: \psline(4.2,1.2)(4.8,1.8)
1954: \psline(3.2,2.2)(3.8,2.8)
1955: \psline(2.2,3.2)(2.8,3.8)
1956: \psline(3.2,4.2)(3.8,4.8)
1957: \psline(4.2,5.2)(4.8,5.8)
1958: \psline(5.2,6.2)(5.8,6.8)
1959: \psline(1.2,4.2)(1.8,4.8)
1960: \psline(2.2,5.2)(2.8,5.8)
1961: \psline(3.2,6.2)(3.8,6.8)
1962: \psline(4.2,7.2)(4.8,7.8)
1963: \endpspicture
1964: $$
1965: \begin{center}
1966: Figure 17. A dual snake in a frieze pattern.
1967: \end{center}
1968:
1969: There is a nice way to see a dual square snake of order $n$
1970: as residing within a frieze pattern of order $n+3$:
1971: rotate the snake by 90 degrees,
1972: so that its first cell is at the top
1973: and its last cell is at the bottom,
1974: and put its top vertex (call it $u$) in the initial row of 1's
1975: of an initially blank frieze pattern with $n+2$ rows,
1976: so that its bottom vertex (call it $v$) lands in the final row of 1's,
1977: and the $R$'s and $L$'s indicate
1978: whether each successive box in the snake
1979: lies to the right or left (respectively)
1980: of the previous box.
1981: If we put 1's along the left border of the snake,
1982: we get a zig-zag of the kind discussed earlier,
1983: so we obtain a frieze-pattern.
1984: Moreover, within the part of the frieze-pattern
1985: that is bounded by the line of slope $-1$ through $u$,
1986: the line of slope $+1$ through $v$,
1987: and the snake itself,
1988: each entry admits an enumerative interpretation
1989: relating to paths in the snake graph.
1990: Specifically, given any location $w$ in the table
1991: in the aforementioned region,
1992: let $u'$ be the leftmost place
1993: where the line through $w$ of slope $-1$ meets the snake,
1994: and let $v'$ be the leftmost place
1995: where the line through $w$ of slope $+1$ meets the snake;
1996: then the entry at $w$ is equal to
1997: the number of paths in the snake
1998: from $u'$ to $v'$.
1999: For instance, in Figure 17
2000: which shows what happens for the $LLLRR$ snake
2001: (with subscripts attached to some of the 1's for purposes of labeling),
2002: we find that
2003: 9 is the number of paths from $1_a$ to $1_c$,
2004: 7 is the number of paths from $1_b$ to $1_c$,
2005: 13 is the number of paths from $1_a$ to $1_d$, and
2006: 10 is the number of paths from $1_b$ to $1_d$.
2007: (The reader may find it instructive to compare this picture
2008: with the corresponding picture for the $RRRLL$ snake;
2009: the arithmetic calculations are different,
2010: but the number 13 still emerges as the rightmost entry.)
2011:
2012: To see why this connection between dual snakes and frieze patterns holds,
2013: we can use Lindstr\"om's lemma~\cite{L}
2014: (rediscovered later by John and Sachs~\cite{JS}
2015: and by Gessel and Viennot~\cite{GV}).
2016: The $m=2$ case of this lemma states that
2017: if we have a directed graph
2018: with sources $s_1,s_2$ and targets $t_1,t_2$,
2019: and there is no way to create a pair of vertex-disjoint paths
2020: that join $s_1$ to $t_2$ and $s_2$ to $t_1$ respectively,
2021: then the number of ways to create a pair of vertex-disjoint paths
2022: that join $s_1$ to $t_1$ and $s_2$ to $t_2$ respectively
2023: is equal to the 2-by-2 determinant $p_{11} p_{22} - p_{12} p_{22}$,
2024: where $p_{ij}$ denotes the number of paths
2025: from $s_i$ to $t_j$.
2026: In our example, putting
2027: $s_1 = 1_a$, $s_2 = 1_b$, $t_1 = 1_d$, and $t_2 = 1_c$ [sic],
2028: we see that there is no way to create
2029: a path from $1_a$ to $1_c$
2030: and a path from $1_b$ to $1_d$
2031: that do not cross,
2032: so the hypotheses are satisfied.
2033: Furthermore, there is exactly 1 way
2034: to create a path from $1_a$ to $1_d$
2035: and a path from $1_b$ to $1_c$
2036: that do not cross,
2037: so we may conclude that
2038: $p_{11} p_{22} - p_{12} p_{22}$ equals 1,
2039: which (given that the $p_{ij}$'s are entries in our table)
2040: is exactly the frieze relation.
2041:
2042: We mention two other combinatorial models of frieze patterns,
2043: for the sake of completeness:
2044: Gregory Price's paths model~\cite{CP}
2045: and the model of Broline, Crowe and Isaacs~\cite{BCI}.
2046: The former (which has been significantly generalized
2047: by Schiffler and Thomas; see~\cite{ST})
2048: is related to the perfect matching model
2049: by the bijection of Carroll and Price,
2050: and the latter is closely related
2051: to the Conway-Coxeter marking scheme.
2052:
2053: \section{A tropical analogue} \label{sec-tropical}
2054:
2055: Since the sideways frieze relation
2056: involves only subtraction-free expressions in the cluster variables,
2057: our whole picture admits a tropical analogue
2058: (for background on tropical mathematics, see~\cite{SS})
2059: in which multiplication is replaced by addition,
2060: division by subtraction, addition by max, and 1 by 0.
2061: (One could use min instead of max, but max will be more useful for us.)
2062: In this new picture, the Ptolemy relation
2063: $$d_{i,j} \, d_{k,l} + d_{j,k} \, d_{i,l} = d_{i,k} \, d_{j,l}$$
2064: becomes the ultrametric relation
2065: $$\max(d_{i,j}+d_{k,l}, d_{j,k}+d_{i,l}) = d_{i,k}+d_{j,l}.$$
2066: %% Should this really be called an "ultrametric relation"?
2067: %% If not, change the usage here and elsewhere in the article!
2068: %% Ask David Speyer. -- 5/22/08
2069: Metrics satisfying this relation
2070: arise from finite collections of non-intersecting arcs
2071: that join points on the sides of the $n$-gon in pairs,
2072: where the endpoints of such an arc are not permitted
2073: to be vertices of the $n$-gon.
2074: We will call such a collection of arcs
2075: an {\it integral lamination\/}.
2076: %% I don't think this is right. Ask Dylan! -- 5/22/08
2077: Figure 18 shows an integral lamination
2078: of a hexagon.
2079:
2080: $$
2081: \pspicture(0,-.75)(9,5.796)
2082: %\pspicture(0,-.5)(6,3.864)
2083: %\pspicture(0,-1.0)(12,7.728)
2084: %\psset{xunit=2,yunit=2}
2085: \psset{xunit=1.5,yunit=1.5}
2086: \pspolygon(2,0)(4,0)(5,1.732)(4,3.464)(2,3.464)(1,1.732)
2087: \psarc(1,1.732){.8}{300}{60}
2088: \psarc(2,3.464){.8}{240}{360}
2089: \psarc(4.5,.866){.7}{60}{240}
2090: %\psline(1.5,2.598)(2.8,0)
2091: %\psline(4.5,2.598)(3.2,0)
2092: \pscurve(1.5,2.598)(2.4,1.299)(2.8,0)
2093: \pscurve(4.5,2.598)(3.6,1.299)(3.2,0)
2094: \endpspicture
2095: $$
2096: \begin{center}
2097: Figure 18. An integral lamination.
2098: \end{center}
2099:
2100: \noindent
2101: For any pair of vertices $i,j$,
2102: we define $d_{i,j}$ as the smallest possible number of intersections
2103: between a path in the $n$-gon from $i$ to $j$
2104: and the arcs in the integral lamination
2105: (we choose the path so as to avoid crossing any arc
2106: in the integral lamination more than once).
2107: Then these quantities $d_{i,j}$ satisfy the ultrametric relation,
2108: and thus can be arranged to form a tropical frieze pattern
2109: satisfying the relation
2110: $$
2111: \begin{array}{ccccccccl}
2112: X & & & & & & Y & & \\
2113: & \jd & & A & & \ji & & & \\[1.2ex]
2114: & & B & & C & & & : & \ \ B + C = \max(A+D,X+Y) \\
2115: & \ji & & D & & \jd & & & \\
2116: Y & & & & & & X & & \\
2117: \end{array}
2118: $$
2119: For instance, the integral lamination of Figure 18
2120: gives rise to the tropical frieze pattern
2121: $$
2122: \begin{array}{ccccccccccccccc}
2123: ...&3& &1& &1& &0& &2& &1& &3&... \\[1.1ex]
2124: ...& &2& &2& &1& &2& &3& &2& &... \\[1.1ex]
2125: ...&1& &3& &2& &1& &3& &2& &1&... \\[1.1ex]
2126: ...& &2& &3& &2& &2& &2& &1& &... \\[1.1ex]
2127: ...&0& &2& &1& &3& &1& &1& &0&...
2128: \end{array}
2129: $$
2130: As in the non-tropical case, we can find all the quantities $d_{i,j}$
2131: once we know the values for all $(i,j)$ associated with
2132: with the sides and diagonals
2133: belonging to some triangulation of the $n$-gon.
2134:
2135: For an alternative picture,
2136: one can divide the laminated $n$-gon into a finite number of sub-regions,
2137: each of which is bounded by pieces of the boundary of the $n$-gon
2138: and/or arcs of the integral lamination;
2139: the vertices of the $n$-gon correspond to $n$ special sub-regions
2140: (some of which may coincide with one another,
2141: if there is no arc in the integral lamination
2142: separating the associated vertices of the $n$-gon).
2143: Then the dual of this dissection of the $n$-gon
2144: is a tree with $n$ specified leaf vertices (some of which may coincide),
2145: and $d_{i,j}$ is the graph-theoretic distance
2146: between leaf $i$ and leaf $j$ (which could be zero).
2147: We see that if we know $2n-3$ of these leaf-to-leaf distances,
2148: and the $2n-3$ pairs of leaves correspond
2149: to the sides and diagonals of a triangulated $n$-gon,
2150: then all of the other leaf-to-leaf distances
2151: can be expressed as piecewise-linear functions
2152: (involving just plus, minus, and max)
2153: of the $2n-3$ specified distances.
2154: %% Is it max, and not min? This seems wrong to me. -- 5/22/08
2155: (For more on the graph metric on trees, see~\cite{Bu}.)
2156:
2157: Going back to our lamination picture,
2158: we can associate to each arc
2159: a non-negative real numbers, called its weight.
2160: Such a weighted collection is a {\it measured lamination\/}.
2161: Then, for any pair of vertices $i,j$,
2162: we define $d_{i,j}$ as the sum of the weights of all the arcs
2163: that separate $i$ from $j$.
2164: This again gives a metric that satisfies
2165: the ultrametric relation.
2166: In the dual (tree-metric) picture,
2167: this corresponds to assigning weights to edges,
2168: and measuring distance between leaves
2169: by summing weights along the path between them
2170: rather than merely counting the edges.
2171:
2172: For an extensive generalization of the foregoing picture,
2173: see~\cite{FST}.
2174:
2175: \section{A variant} \label{sec-variant}
2176:
2177: An open problem concerns a variant
2178: of Conway and Coxeter's definition,
2179: in which the frieze recurrence is replaced by the recurrence
2180: $$
2181: \begin{array}{ccc}
2182: & A & \\[1.8ex]
2183: B & C & D \\[1.8ex]
2184: & E &
2185: \end{array} \ \ : \ \ E = (BD-C)/A
2186: $$
2187: and its sideways version
2188: $$
2189: \begin{array}{ccc}
2190: & A & \\[1.8ex]
2191: B & C & D \\[1.8ex]
2192: & E &
2193: \end{array} \ \ : \ \ D = (AE+C)/B \ \ .
2194: $$
2195: We can construct arrays
2196: that have the same sort of symmetries as frieze patterns
2197: by starting with a suitable zig-zag of entries
2198: (where successive downwards steps can go left, right, or straight)
2199: and proceeding from left to right.
2200: E.g., consider the partial table
2201: $$
2202: \begin{array}{ccccccc}
2203: ...& 1 & 1 & 1 & 1 & 1 & ...\\
2204: & & A & D & x & & \\
2205: & B & E & y & & & \\
2206: & & C & F & z & & \\
2207: ...& 1 & 1 & 1 & 1 & 1 & ...
2208: \end{array}
2209: $$
2210: where $A,...,F$ are pre-specified,
2211: and where we compute
2212: $y = (AC+E)/B$,
2213: $x = (y+D)/A$,
2214: $z = (y+F)/C$,
2215: etc.
2216: Then after exactly fourteen iterations of the procedure,
2217: one gets back the original numbers (in their original order).
2218: Moreover, along the way one sees
2219: Laurent polynomials with positive coefficients.
2220:
2221: Define a ``double zig-zag''
2222: to be a subset of the entries of an $(n-2)$-rowed table
2223: consisting of a pair of adjacent entries in each of the middle $n-4$ rows,
2224: such that the pair in each row is displaced
2225: with respect to the pair in the preceding and succeeding rows
2226: by at most one position.
2227: (Thus the entries $A, B, C, D, E, F$ in the previous table
2228: form a double zig-zag,
2229: as do the entries $D, E, F, x, y, z$.)
2230:
2231: {\sc Conjecture}:
2232: Given any assignment of formal weights
2233: to the $2(n-4)$ entries in a double zig-zag
2234: in an $(n-2)$-rowed table,
2235: there is a unique assignment of rational functions
2236: to all the entries in the table
2237: so that the variant frieze relation is satisfied.
2238: These rational functions of the original $2(n-4)$ variables
2239: have glide-reflection symmetry that gives each row period $2n$.
2240: Furthermore, each of the rational functions in the table
2241: is a Laurent polynomial with positive coefficients.
2242:
2243: There ought to be a way to prove this
2244: by constructing the numerators of these Laurent polynomials
2245: as sums of weights of perfect matchings of some suitable graph
2246: (or perhaps sums of weights of combinatorial objects
2247: more general than perfect matchings),
2248: and the numerators undoubtedly contain abundant clues
2249: as to how this can be done.
2250:
2251: For $n=5,6,7,8$,
2252: it appears that the number of positive integer arrays
2253: satisfying the variant frieze relation
2254: is 1, 5, 51, 868 (respectively).
2255: This variant of the Catalan sequence
2256: does not appear to have been studied before.
2257: However, it should be said that these numbers
2258: were not computed in a rigorous fashion.
2259: Indeed, it is conceivable that beyond some point,
2260: the numbers becomes infinite
2261: (i.e., for some $n$ there could be
2262: infinitely many $(n-2)$-rowed positive integer arrays
2263: satisfying the variant frieze relation).
2264:
2265: Dean Hickerson~\cite{Hi}~has shown
2266: that any $(n-2)$-rowed array that begins and ends with a row of 1's
2267: and satisfies the variant frieze relation in between
2268: has glide-reflection symmetry and period $2n$.
2269: This implies that if one generates such a variant frieze pattern
2270: starting with a double zig-zag of 1's,
2271: one gets a periodic array of positive rational numbers.
2272: However, it is not apparent that one can modify Hickerson's
2273: (purely algebraic) proof to show that these rational numbers are integers.
2274: Furthermore, if one uses formal weights instead of 1's,
2275: Hickerson's argument does not seem to show
2276: that the resulting rational functions are Laurent polynomials
2277: (let alone that the Laurent polynomials have positive coefficients).
2278:
2279: %% Mention Reid Barton's number fences?
2280:
2281: \section{Markoff numbers} \label{sec-markoff}
2282:
2283: A {\it Markoff triple} is a triple $(x,y,z)$ of positive integers
2284: satisfying $x^2+y^2+z^2=3xyz$, such as the triple (2,5,29).
2285: A {\it Markoff number} is a positive integer
2286: that occurs in at least one such triple.
2287:
2288: Writing the Markoff equation as $z^2 - (3xy) z + (x^2+y^2) = 0$,
2289: a quadratic equation in $z$, we see that if $(x,y,z)$ is a Markoff triple,
2290: then so is $(x,y,z')$, where $z' = 3xy-z = (x^2+y^2)/z$,
2291: the other root of the quadratic in $z$.
2292: ($z'$ is positive because $z'=(x^2+y^2)/z$,
2293: and is an integer because $z'=3xy-z$.)
2294: Likewise for $x$ and $y$.
2295:
2296: The following claim is well-known
2297: (for an elegant proof, see~\cite{Ba}):
2298: Every Markoff triple $(x,y,z)$
2299: can be obtained from the Markoff triple $(1,1,1)$
2300: by a sequence of such exchange operations,
2301: in fact, by a sequence of exchange operations
2302: that leaves two numbers alone and increases the third.
2303: E.g., $(1,1,1) \rightarrow (2,1,1) \rightarrow (2,5,1)$ $\rightarrow (2,5,29)$.
2304:
2305: Create a graph whose vertices are the Markoff triples
2306: and whose edges correspond to the exchange operations
2307: $(x,y,z) \leftrightarrow (x',y,z)$,
2308: $(x,y,z) \leftrightarrow (x,y',z)$,
2309: $(x,y,z) \leftrightarrow (x,y,z')$
2310: where
2311: $x' = \frac{y^2+z^2}{x}$,
2312: $y' = \frac{x^2+z^2}{y}$,
2313: $z' = \frac{x^2+y^2}{z}$.
2314: This 3-regular graph is connected (see the claim in the preceding paragraph),
2315: and it is not hard to show that it is acyclic.
2316: Hence the graph is the 3-regular infinite tree.
2317:
2318: This tree can be understood as the dual of the
2319: triangulation of the upper half plane
2320: by images of the modular domain under the action of the modular group.
2321: Concretely, we can describe this picture by using Conway's terminology of
2322: ``lax vectors'', ``lax bases'', and ``lax superbases'' (\cite{C}).
2323:
2324: A {\it primitive} vector $\vu$ in a lattice $L$
2325: is one that cannot be written as $k\vv$
2326: for some vector $\vv$ in $L$, with $k>1$.
2327: A {\it lax vector} is a primitive vector defined only up to sign;
2328: if $\vu$ is a primitive vector,
2329: the associated lax vector is written $\pm \vu$.
2330: A {\it lax base} for $L$ is a set of two lax vectors $\{\pm \vu, \pm \vv\}$
2331: such that $\vu$ and $\vv$ form a basis for $L$.
2332: A {\it lax superbase} for $L$ is a set of three lax vectors
2333: $\{\pm \vu, \pm \vv, \pm \vw\}$ such that $\pm\vu\pm\vv\pm\vw=\vzero$
2334: (with appropriate choice of signs)
2335: and any two of $\vu,\vv,\vw$ form a basis for $L$.
2336:
2337: Each lax superbase
2338: $\{\pm \vu, \pm \vv, \pm \vw\}$
2339: contains the three lax bases
2340: $\{\pm \vu, \pm \vv\}$, $\{\pm \vu, \pm \vw\}$, $\{\pm \vv, \pm \vw\}$
2341: and no others.
2342: In the other direction, each lax base
2343: $\{\pm \vu, \pm \vv\}$
2344: is in the two lax superbases
2345: $\{\pm \vu, \pm \vv, \pm (\vu+\vv)\}, \ \{\pm \vu, \pm \vv, \pm (\vu-\vv)\}$
2346: and no others.
2347:
2348: The {\it topograph} is the graph whose vertices are lax superbases
2349: and whose edges are lax bases,
2350: where each lax superbase is incident with the three lax bases in it.
2351: This gives a 3-valent tree
2352: whose vertices correspond to the lax superbases of $L$,
2353: whose edges correspond to the lax bases of $L$,
2354: and whose ``faces'' correspond to the lax vectors in $L$.
2355:
2356: The lattice $L$ that we will want to use is the triangular lattice
2357: $L = \{(x,y,z) \in \Z^3 \ : \ x+y+z=0\}$
2358: (or $\Z^3 / \Z\vv$ where $\vv=(1,1,1)$, if you prefer).
2359:
2360: Using this terminology, it is now possible to state
2361: the main idea of this section
2362: (with details and proof to follow):
2363: Unordered Markoff triples are associated with
2364: lax superbases of the triangular lattice,
2365: and Markoff numbers are associated with
2366: lax vectors of the triangular lattice.
2367: For example, the unordered Markoff triple $2,5,29$
2368: corresponds to the lax superbase
2369: $\{\pm \vu,\pm \vv,\pm \vw\}$
2370: where $\vu = \vec{OA}$, $\vv = \vec{OB}$, and $\vw = \vec{OC}$,
2371: with $O$, $A$, $B$, and $C$ forming a fundamental parallelogram
2372: for the triangular lattice, as shown in Figure 19.
2373: The Markoff numbers 1, 2, 5, and 29 correspond to the primitive vectors
2374: $\vec{AB}$, $\vec{OA} = \vec{BC}$,
2375: $\vec{OB} = \vec{AC}$, and $\vec{OC}$.
2376:
2377: \begin{center}
2378: \begin{pspicture}*(0.5,0.4)(9.5,6.528) % clip what lies outside the frame
2379: %\begin{pspicture}*(0.5,0)(9.5,6.928) % clip what lies outside the frame
2380: \psline(0.5,1.732)(9.5,1.732)
2381: \psline(0.5,3.464)(9.5,3.464)
2382: \psline(0.5,5.196)(9.5,5.196)
2383: \psline(-1.5,0.866)(1.5,6.062) % clipped
2384: \psline(0.5,0.866)(3.5,6.062)
2385: \psline(2.5,0.866)(5.5,6.062)
2386: \psline(4.5,0.866)(7.5,6.062)
2387: \psline(6.5,0.866)(9.5,6.062)
2388: \psline(8.5,0.866)(11.5,6.062) % clipped
2389: \psline(-1.5,6.062)(1.5,0.866) % clipped
2390: \psline(0.5,6.062)(3.5,0.866)
2391: \psline(2.5,6.062)(5.5,0.866)
2392: \psline(4.5,6.062)(7.5,0.866)
2393: \psline(6.5,6.062)(9.5,0.866)
2394: \psline(8.5,6.062)(11.5,0.866) % clipped
2395: \rput(1,1.732){{\huge $O$}}
2396: \rput(4,3.464){{\huge $A$}}
2397: \rput(6,3.464){{\huge $B$}}
2398: \rput(9,5.196){{\huge $C$}}
2399: \end{pspicture}
2400: \end{center}
2401: \noindent
2402:
2403: \begin{center}
2404: Figure 19. A fundamental parallelogram.
2405: \end{center}
2406:
2407: To find the Markoff number associated with a primitive vector $\vec{OX}$,
2408: take the union $R$ of all the triangles that segment $OX$ passes through.
2409: The underlying lattice provides a triangulation of $R$.
2410: E.g., for the vector $\vu = \vec{OC}$ from Figure 19,
2411: the triangulation is as shown in Figure 20.
2412: \begin{center}
2413: \begin{pspicture}(0,0)(10,3.464) % don't clip what lies outside the frame
2414: \psline(0,0.000)(4,0.000)
2415: \psline(1,1.732)(7,1.732)
2416: \psline(4,3.464)(8,3.464)
2417: \psline(0,0.000)(1,1.732)
2418: \psline(2,0.000)(4,3.464)
2419: \psline(4,0.000)(6,3.464)
2420: \psline(7,1.732)(8,3.464)
2421: \psline(1,1.732)(2,0.000)
2422: \psline(3,1.732)(4,0.000)
2423: \psline(4,3.464)(5,1.732)
2424: \psline(6,3.464)(7,1.732)
2425: \rput(0,0.000){{\huge $O$}}
2426: \rput(3,1.732){{\huge $A$}}
2427: \rput(5,1.732){{\huge $B$}}
2428: \rput(8,3.464){{\huge $C$}}
2429: \psset{linestyle=dashed}
2430: \psline(0,0)(8,3.464)
2431: \end{pspicture}
2432: \end{center}
2433: \begin{center}
2434: Figure 20. A Markoff snake.
2435: \end{center}
2436:
2437: Turn the triangulation into a planar bipartite graph as in section 2,
2438: let $G(\vu)$ be the graph that results
2439: from deleting vertices $O$ and $C$,
2440: and let $M(\vu)$ be the number of perfect matchings of $G(\vu)$.
2441: (If $\vu$ is a shortest vector in the lattice, put $M(\vu)=1$.)
2442:
2443: \begin{theorem}[Carroll, Itsara, Le, Musiker, Price, and Viana
2444: \cite{CILP} \cite{CP} \cite{I} \cite{M}]
2445: If $\{\vu,\vv,\vw\}$ is a lax superbase of the triangular lattice,
2446: then $(M(\vu),M(\vv),M(\vw))$ is a Markoff triple.
2447: Every Markoff triple arises in this fashion.
2448: In particular,
2449: if $\vu$ is a primitive vector, then $M(\vu)$ is a Markoff number,
2450: and every Markoff number arises in this fashion.
2451: \end{theorem}
2452:
2453: \noindent
2454: (The association of Markoff numbers with the topograph is not new;
2455: what is new is the combinatorial interpretation
2456: of the association, by way of perfect matchings.)
2457:
2458: \begin{proof}
2459: The base case, with $$(M(\ve_1),M(\ve_2),M(\ve_3)) = (1,1,1),$$ is clear.
2460: The only non-trivial part of the proof is the verification that
2461: $$M(\vu+\vv) = (M(\vu)^2 + M(\vv)^2) / M(\vu-\vv).$$
2462: E.g., in Figure 21, we need to verify that
2463: $$M(\vec{OC}) M(\vec{AB}) = M(\vec{OA})^2 + M(\vec{OB})^2.$$
2464: But if we rewrite the desired equation as
2465: $$M(\vec{OC}) M(\vec{AB}) = M(\vec{OA}) M(\vec{BC}) + M(\vec{OB}) M(\vec{AC})$$
2466: we see that this is just Kuo's lemma
2467: (see the proof of Theorem \ref{thm-cp}).
2468: \begin{center}
2469: \begin{pspicture}(0,0)(10,3.464) % don't clip what lies outside the frame
2470: \psline(0,0.000)(4,0.000)
2471: \psline(1,1.732)(7,1.732)
2472: \psline(4,3.464)(8,3.464)
2473: \psline(0,0.000)(1,1.732)
2474: \psline(2,0.000)(4,3.464)
2475: \psline(4,0.000)(6,3.464)
2476: \psline(7,1.732)(8,3.464)
2477: \psline(1,1.732)(2,0.000)
2478: \psline(3,1.732)(4,0.000)
2479: \psline(4,3.464)(5,1.732)
2480: \psline(6,3.464)(7,1.732)
2481: \rput(0,0.000){{\huge $O$}}
2482: \rput(3,1.732){{\huge $A$}}
2483: \rput(5,1.732){{\huge $B$}}
2484: \rput(8,3.464){{\huge $C$}}
2485: %\psset{linestyle=dashed}
2486: %\psline(0,0)(8,3.464)
2487: \end{pspicture}
2488: \end{center}
2489: \begin{center}
2490: Figure 21. Kuo condensation for snakes.
2491: \end{center}
2492: \end{proof}
2493:
2494: %% These two historical remarks should be combined
2495: %% and redistributed somehow. -- 5/27/08
2496:
2497: Remark: Some of the work done by Carroll et al.\ during the years of the
2498: Research Experiences in Algebraic Combinatorics at Harvard (2001 to 2003)
2499: used the square lattice picture of section~\ref{sec-snake};
2500: this way of interpreting the Markoff numbers combinatorially
2501: was actually conjectured first, in 2001--2002, by Musiker,
2502: and subsequently proved in 2002--2003 by Itsara, Le, Musiker, and Viana
2503: (see~\cite{M},~\cite{I}, and~\cite{CILP},
2504: and section~\ref{sec-snake} of this article).
2505: Subsequently, the group's first combinatorial model
2506: for frieze patterns, found by Price,
2507: involved paths rather than perfect matchings.
2508: It is reminiscent of, but apparently distinct from,
2509: the paths model considered in section~\ref{sec-snake}.
2510: Carroll turned Price's paths model into a perfect matchings model,
2511: which made it possible to arrive at the snake-graph model
2512: via a different route.
2513:
2514: More generally, one can put $M(\ve_1)=x$, $M(\ve_2)=y$, and $M(\ve_3)=z$
2515: (with $x,y,z>0$)
2516: and recursively define $$M(\vu+\vv) = (M(\vu)^2 + M(\vv)^2) / M(\vu-\vv).$$
2517: Then for all primitive vectors $\vu$,
2518: $M(\vu)$ is a Laurent polynomial in $x,y,z$;
2519: that is, it can be written in the form $P(x,y,z)/$ $x^a y^b z^c$,
2520: where $P(x,y,z)$ is an ordinary polynomial
2521: in $x,y,z$ (with non-zero constant term).
2522: The numerator $P(x,y,z)$ of each Markoff polynomial is the sum of the weights
2523: of all the perfect matchings of the graph $G(\vu)$,
2524: where edges have weight $x$, $y$, or $z$ according to orientation.
2525: The triples $X=M(\vu)$, $Y=M(\vv)$, $Z=M(\vw)$
2526: of rational functions associated with lax superbases
2527: are solutions of the equation
2528: $$X^2+Y^2+Z^2 = \frac{x^2+y^2+z^2}{xyz} XYZ.$$
2529: Theorem~\ref{thm-frieze} implies that these numerators $P(x,y,z)$
2530: are polynomials with positive coefficients.
2531: This proves the following theorem:
2532:
2533: \begin{theorem} \label{thm-triples}
2534: Consider the initial triple $(x,y,z)$,
2535: along with every triple of rational functions in $x$, $y$, and $z$
2536: that can be obtained from the initial triple
2537: by a sequence of operations of the form
2538: $(X,Y,Z) \mapsto (X',Y,Z)$,
2539: $(X,Y,Z) \mapsto (X,Y',Z)$, or
2540: $(X,Y,Z) \mapsto (X,Y,Z')$,
2541: where $X' = (Y^2+Z^2)/X$, $Y' = (X^2+Z^2)/Y$, and $Z' = (X^2+Y^2)/Z$.
2542: Every rational function of $x$, $y$, and $z$ that occurs in such a triple
2543: is a Laurent polynomial with positive coefficients.
2544: \end{theorem}
2545:
2546: Fomin and Zelevinsky proved in~\cite{FZL}
2547: (Theorem 1.10)
2548: that the rational functions $X(x,y,z),Y(x,y,z),Z(x,y,z)$
2549: are Laurent polynomials,
2550: but their methods did not prove positivity.
2551: An alternative proof of positivity, based on topological ideas,
2552: was given by Dylan Thurston~\cite{T}.
2553: %% Ask Dylan for an update -- 5/27/08
2554: %% There is probably a proof based on quivers; cite it! -- 5/19/08
2555:
2556: It can be shown that
2557: if $\vu$ is inside the cone generated by $+\ve_1$ and $-\ve_3$,
2558: then $a < b > c$ and $(c+1) \ve_1 - (a+1) \ve_3 = \vu$.
2559: (Likewise for the other sectors of $L$.)
2560: This implies that all the ``Markoff polynomials'' $M(\vu)$ are distinct
2561: (aside from the fact that $M(\vu)$ $=M(-\vu)$),
2562: and thus $M(\vu)(x,y,z)$ $\neq M(\vv)(x,y,z)$
2563: for all primitive vectors $\vu \neq \pm \vv$
2564: as long as $(x,y,z)$ lies in a dense $G_\delta$ set of real triples.
2565: This fact can be used to show~\cite{T}
2566: that, for a generic choice of hyperbolic structure on the once-punctured torus,
2567: no two simple geodesics have the same length.
2568: (It should be mentioned that for the specific choice $x=y=z=1$,
2569: the distinctness of the numbers $M(\vu)(x,y,z)$ as $\vu$ varies
2570: is the famous, and still unproved, ``unicity conjecture'' for Markoff numbers.)
2571: %% Give a reference?
2572:
2573: %% Say more about combinatorial meaning of denominators
2574: %% for Markof polynomials. -- 5/28/08
2575:
2576: A slightly different point of view of Markoff numbers
2577: focuses on triangles rather than lax superbases:
2578: Say that points $A$, $B$, and $C$ in the equilateral triangular lattice
2579: form a ``fundamental triangle'' if the area of triangle $ABC$
2580: equals the area of the equilateral triangles of which the lattice is composed.
2581: For example, the points $A$, $B$, and $C$ in Figure 21
2582: are the vertices of a fundamental triangle.
2583: (If four points form a fundamental parallelogram for the lattice,
2584: then any three of the four points form a fundamental triangle.)
2585: By Pick's theorem, a triangle is fundamental
2586: if and only if it has no lattice points in its interior
2587: and no lattice points on its boundary other than its three vertices.
2588: Let $A$, $B$, and $C$ form a fundamental triangle.
2589: Define the ``triangulation distance'' $d(x,y)$ between two vertices $x$ and $y$
2590: as $M(\vu)$ where $\vu$ is the vector from $x$ to $y$.
2591: Then the triangulation distances $d(A,B)$, $d(A,C)$, and $d(B,C)$
2592: form a Markoff triple, and every Markoff triples arises in this way.
2593:
2594: We conclude by mentioning a special sequence of Markoff numbers,
2595: obtained by following the tree
2596: along those branches that give greatest numerical increase:
2597: 1, 1, 2, 5, 29, 433, 37666, ...
2598: This sequence was considered by Dana Scott (see~\cite{G}),
2599: %% Mention Mordell?
2600: and satisfies the recurrence
2601: $f(n) = (f(n-1)^2+f(n-2)^2)/f(n-3)$.
2602: Using the $A$ and $B$ matrices
2603: from Section~\ref{sec-snake},
2604: we see that we can alternately characterize the numbers
2605: as the upper-left entries in the sequence of matrices
2606: $$
2607: \left( \begin{array}{rr}
2608: 1 & 1 \\ 1 & 0 \end{array} \right) ,
2609: \left( \begin{array}{rr}
2610: 1 & 0 \\ 0 & 1 \end{array} \right) ,
2611: \left( \begin{array}{rr}
2612: 2 & 1 \\ 1 & 1 \end{array} \right) ,
2613: \left( \begin{array}{rr}
2614: 5 & 2 \\ 2 & 1 \end{array} \right) ,
2615: \left( \begin{array}{rr}
2616: 29 & 12 \\ 12 & 5 \end{array} \right) ,
2617: \left( \begin{array}{rr}
2618: 433 & 179 \\ 179 & 74 \end{array} \right) , ...
2619: $$
2620: satisfying the multiplicative recurrence relation
2621: $$M(n) = M(n-1) M(n-3)^{-1} M(n-1)$$
2622: (note that the Fibonacci numbers satisfy
2623: the additive version of this recurrence).
2624: Andy Hone~\cite{Ho}
2625: has shown that $\log \frac{\log f(n)}{n}$
2626: approaches $\log \frac{1+\sqrt{5}}{2}$
2627: as $n \rightarrow \infty$.
2628:
2629: %% Mention other stuff on two-by-two matrices, such as
2630: %% Tom Ace's formula, or the formula for square snakes
2631: %% that involved raising matrices to the -1 power and
2632: %% cancelling parts of snakes. -- 6/2/08
2633:
2634: \section{Other directions for exploration} \label{sec-other}
2635:
2636: \subsection{Non-integer frieze-patterns}
2637:
2638: Given that the original geometric context of frieze patterns
2639: gives rise to arrays containing numbers that are not integers,
2640: it seems fairly natural to try to extend
2641: the Conway-Coxeter theory to this broader setting.
2642: Enumerative questions would be a good place to start.
2643: One might for instance try to count all the frieze patterns of order $n$
2644: whose entries are either (positive) integers or half-integers,
2645: and see if the enumerating sequence
2646: is any sort of known analogue of the Catalan sequence.
2647: Also, since many geometric frieze patterns
2648: involve (irrational) algebraic numbers,
2649: it might also be natural to enumerate
2650: frieze patterns with entries in a given number ring
2651: (though this might not be so very natural after all:
2652: consider that, in its original geometric context,
2653: positivity of the entries of the frieze pattern
2654: is a consequence of their metric interpretation,
2655: whereas for algebraic number rings
2656: positivity is not a very robust notion
2657: since it depends on the embedding of the ring in $\R$).
2658:
2659: %% I chose not to mention here the result concerning triangulations mod 2
2660: %% (see O07a/jjonsson.to for a statement and proof), once I realized that,
2661: %% just because the first two rows of two frieze patterns agree mod 2,
2662: %% it doesn't follow that the entire arrays agree mod 2. -- 5/29/08
2663:
2664: \subsection{Non-fundamental triangles}
2665:
2666: Suppose $A$, $B$, and $C$ are points in the lattice
2667: such that line segments $AB$, $AB$, and $BC$
2668: contain no lattice-points other than their endpoints,
2669: so that the triangulation distances $d(A,B)$, $d(A,C)$, $d(B,C)$
2670: are well-defined.
2671: We have seen that if triangle $ABC$ contains no lattice points
2672: in its interior,
2673: then these distances satisfy the Markoff equation.
2674: Can anything be said if this condition does not hold?
2675: For instance, in a lattice made of equilateral triangles of side-length 1,
2676: consider an equilateral triangle $ABC$ of side-length $\sqrt{3}$
2677: containing one interior point.
2678: The triangulation distances are all equal to 2,
2679: and $(2,2,2)$ is not a Markoff triple.
2680: Nevertheless, perhaps there is a different algebraic equation
2681: that this triple satisfies.
2682: More precisely, there may be an algebraic relation
2683: satisfied by the triangulation distances
2684: $x=d(B',C')$, $y=d(A',C')$, $z=d(A',B')$
2685: where $A'B'C'$ is any image of $ABC$
2686: under the joint action of $SL_2(\Z)$ (change of lattice-base)
2687: and $\Z^2$ (translation).
2688: Indeed, the whole numbers $x,y,z$ satisfy the condition
2689: that there exist other whole numbers $x',y',z'$
2690: (namely, the triangulation distances
2691: from the interior point to $A'$, $B'$, and $C'$ respectively)
2692: such that $x'^2+y^2+z^2 = 3x'yz$, $x^2+y'^2+z^2 = 3xy'z$,
2693: and $x^2+y^2+z'^2 = 3xyz'$,
2694: and perhaps some sort of quantifier elimination procedure
2695: would permit us to write this as a condition on just $x$, $y$, and $z$.
2696: More broadly, perhaps each orbit of triangles
2697: under the action of $SL_2(\Z)$ and $\Z^2$
2698: gives rise to triples satisfying
2699: a particular algebraic condition specific to that orbit.
2700:
2701: \subsection{Other ternary cubics}
2702:
2703: Neil Herriot (another member of REACH)
2704: %% Find a graceful way to remind readers what REACH was. -- 5/28/08
2705: showed~\cite{He} that if we replace the triangular lattice used above
2706: by the tiling of the plane by isosceles right triangles
2707: (generated from one such triangle by repeated reflection in the sides),
2708: fundamental triangles give rise to triples $x,y,z$ of positive integers
2709: satisfying either $$x^2+y^2+2z^2=4xyz$$ or $$2x^2+2y^2+z^2=4xyz.$$
2710: (Note that these two Diophantine equations are essentially equivalent,
2711: as the map $(x,y,z) \mapsto (x,y,2z)$ gives a bijection between
2712: solutions to the former and solutions to the latter.)
2713: For instance, if for any two vertices $X,Y$
2714: we define the triangulation distance $d(X,Y)$
2715: in analogy with the definition used before
2716: (now using the isosceles right triangle lattice
2717: in place of the equilateral triangle lattice),
2718: then the points $O,A,B,C$ shown in Figure 22 satisfy
2719: $d(A,B) = 1$, $d(O,A) = 1$, $d(B,C) = 2$,
2720: $d(O,B) = 3$, $d(A,C) = 3$, and $d(O,C) = 11$,
2721: corresponding to the solution triples
2722: $(11)^2 + (3)^2 + 2(1)^2 = 4(11)(3)(1)$
2723: and
2724: $2(11)^2 + 2(3)^2 + (2)^2 = 4(11)(3)(2)$.
2725: \begin{center}
2726: \begin{pspicture}(1,1)(9,7)
2727: \psline(2,2)(8,2)
2728: \psline(2,4)(8,4)
2729: \psline(2,6)(8,6)
2730: \psline(2,2)(2,6)
2731: \psline(4,2)(4,6)
2732: \psline(6,2)(6,6)
2733: \psline(8,2)(8,6)
2734: \psline(2,2)(6,6)
2735: \psline(6,2)(8,4)
2736: \psline(6,2)(2,6)
2737: \psline(8,4)(6,6)
2738: \rput(1.8,2.2){{\Large $O$}}
2739: \rput(3.8,4.2){{\Large $A$}}
2740: \rput(5.8,4.2){{\Large $B$}}
2741: \rput(7.8,6.2){{\Large $C$}}
2742: \end{pspicture}
2743: \end{center}
2744: \begin{center}
2745: Figure 22. Herriot's theorem.
2746: \end{center}
2747: More specifically, Herriot showed that
2748: if $ABC$ is a fundamental triangle,
2749: then the triangulation distances
2750: $d(A,B)$, $d(A,C)$, $d(B,C)$ satisfy
2751: $$d(A,B)^2 + 2d(A,C)^2 + 2d(B,C)^2 = 4d(A,B)d(A,C)d(B,C)$$
2752: or
2753: $$2d(A,B)^2 + d(A,C)^2 + d(B,C)^2 = 4d(A,B)d(A,C)d(B,C)$$
2754: according to whether the vertices $A,B,C$
2755: have respective degrees 4,4,8 or 8,8,4.
2756: (One can check that a fundamental triangle
2757: cannot have all three vertices of degree 4
2758: or all three vertices of degree 8.)
2759: A related observation is that
2760: if $OACB$ is a fundamental parallelogram
2761: with $O$ and $A$ of degree 8 and $B$ and $C$ of degree 4,
2762: then $d(B,C) = 2d(O,A)$.
2763:
2764: Herriot's result,
2765: considered in conjunction with the result on Markoff numbers,
2766: raises the question of whether
2767: there might be some more general combinatorial approach
2768: to ternary cubic equations of similar shape.
2769:
2770: Rosenberger~\cite{R}
2771: showed that there are exactly three ternary cubic equations
2772: of the shape $ax^2 + by^2 + cz^2 = (a+b+c)xyz$
2773: for which all the positive integer solutions
2774: can be derived from the solution $(x,y,z)=(1,1,1)$ by means
2775: of the exchange operations
2776: $(x,y,z) \rightarrow (x',y,z)$,
2777: $(x,y,z) \rightarrow (x,y',z)$, and
2778: $(x,y,z) \rightarrow (x,y,z')$, with
2779: $x'=(by^2+cz^2)/ax$,
2780: $y'=(ax^2+cz^2)/by$, and
2781: $z'=(ax^2+by^2)/cz$.
2782: These three ternary cubic equations are
2783: $$x^2+y^2+z^2=3xyz,$$
2784: $$x^2+y^2+2z^2=4xyz,$$ and
2785: $$x^2+2y^2+3z^2=6xyz.$$
2786:
2787: Note that the triples of coefficients that occur here ---
2788: (1,1,1), (1,1,2), and (1,2,3) ---
2789: are precisely the triples that occur in the classification
2790: of finite reflection groups in the plane.
2791: Specifically, the ratios 1:1:1, 1:1:2, and 1:2:3
2792: describe the angles of the three triangles ---
2793: the 60-60-60 triangle, the 45-45-90 triangle, and the 30-60-90 triangle ---
2794: that arise as the fundamental domains
2795: of the three irreducible two-dimensional reflection groups.
2796:
2797: Since the solutions to the ternary cubic $x^2+y^2+z^2=3xyz$
2798: describe properties of the tiling of the plane by 60-60-60 triangles,
2799: and solutions to the ternary cubic $x^2+y^2+2z^2=4xyz$
2800: describe properties of the tiling of the plane by 45-45-90 triangles,
2801: the solutions to the ternary cubic $x^2+2y^2+3z^2=6xyz$
2802: ``ought'' to be associated with some combinatorial model
2803: involving the reflection-tiling of the plane by 30-60-90 triangles.
2804: Unfortunately, the most obvious approach
2805: (based on analogy with the 60-60-60 and 45-45-90 cases) does not work.
2806: So we are left with two problems that may or may not be related:
2807: first, to find a combinatorial interpretation
2808: for the integers (or, more generally, the Laurent polynomials)
2809: that arise from solving the ternary cubic $x^2+2y^2+3z^2=6xyz$;
2810: and second, to find algebraic recurrences
2811: that govern the integers (or, more generally, the Laurent polynomials)
2812: that arise from counting (or summing the weights of)
2813: perfect matchings of graphs
2814: derived from the reflection-tiling of the plane by 30-60-90 triangles.
2815:
2816: If there is a way to make the analogy work,
2817: one might seek to extend the analysis to other ternary cubics.
2818: It is clear how this might generalize on the algebraic side.
2819: On the geometric side, one might drop the requirement
2820: that the triangle tile the plane by reflection,
2821: and insist only that each angle be a rational multiple of 360 degrees.
2822: There is a relatively well-developed theory
2823: of ``billiards flow'' in such a triangle
2824: (see e.g.~\cite{KS})
2825: where a particle inside the triangle
2826: bounces off the sides following the law of reflection
2827: (angle of incidence equals angle of reflection)
2828: and travels along a straight line in between bounces.
2829: The path of such a particle can be unfolded
2830: by repeatedly reflecting the triangular domain
2831: in the side that the particle is bouncing off of,
2832: so that the unfolded path of the particle
2833: is just a straight line in the plane.
2834: Of special interest in the theory of billiards
2835: are trajectories joining a corner to a corner
2836: (possibly the same corner or possibly a different one);
2837: these are called saddle connections.
2838: The reflected images of the triangular domain
2839: form a triangulated polygon,
2840: and the saddle connection is a combinatorial diagonal of this polygon.
2841: It is unclear whether the combinatorics of such triangulations
2842: might contain dynamical information about the billiards flow,
2843: but if this prospect were to be explored,
2844: enumeration of matchings on the derived bipartite graphs
2845: would be one thing to try.
2846:
2847: \subsection{More variables}
2848:
2849: Another natural variant of the Markoff equation $x^2+y^2+z^2=3xyz$
2850: is the equation $w^2+x^2+y^2+z^2=4wxyz$
2851: (one special representative of a broader class
2852: called Markoff-Hurwitz equations; see~\cite{Ba}).
2853: The Laurent phenomenon applies here too:
2854: the four natural exchange operations
2855: convert an initial formal solution $(w,x,y,z)$
2856: into a quadruple of Laurent polynomials.
2857: (This is a special case of Theorem 1.10 in~\cite{FZL}.)
2858:
2859: Furthermore, the coefficients of these Laurent polynomials
2860: appear to be positive, although this has not been proved.
2861:
2862: The numerators of these Laurent polynomials ought to be weight-enumerators
2863: for some combinatorial model, but it is unclear
2864: how to reverse-engineer the combinatorial model from the Laurent polynomials.
2865:
2866: %use either amsalpha bibliography style or acm
2867:
2868: \bibliographystyle{amsalpha}
2869: \begin{thebibliography}{20}
2870:
2871: \bibitem{Ba}
2872: A. Baragar,
2873: Integral solutions of the Markoff-Hurwitz equations,
2874: {\it J. of Number Theory} {\bf 49} (1994), 27--44.
2875:
2876: \bibitem{BM}
2877: K. Baur and R.J. Marsh,
2878: Ptolemy relations for punctured discs,
2879: arXiv:{\tt 0711.1443}.
2880:
2881: \bibitem{BQ}
2882: A. Benjamin and J. Quinn,
2883: Proofs That Really Count.
2884: Mathematical Association of America, Washington, D.C. (2003).
2885:
2886: \bibitem{BQS}
2887: A. Benjamin, J. Quinn, and F. Su,
2888: Counting On continued fractions,
2889: {\it Math. Mag.} {\bf 73} (2000), 98--104.
2890:
2891: \bibitem{BZ}
2892: A. Benjamin and D. Zeilberger,
2893: Pythagorean primes and palindromic continued fractions,
2894: {\it Integers} {\bf 5} (2005), A30.
2895:
2896: \bibitem{BCI}
2897: D. Broline, D.W. Crowe, and I.M. Isaacs,
2898: The geometry of frieze patterns,
2899: {\it Geom. Ded.} {\bf 3} (1974), 171--176.
2900:
2901: \bibitem{Bu}
2902: P. Buneman,
2903: A note on metric properties of trees,
2904: {\it J. Comb. Theory, Ser. B} {\bf 17} (1974), 48--50.
2905:
2906: \bibitem{CaCh}
2907: P. Caldero and F. Chapoton,
2908: Cluster algebras as Hall algebras of quiver representations,
2909: arXiv:{\tt math.RT/0410187}.
2910:
2911: \bibitem{CILP}
2912: G. Carroll, A. Itsara, I. Le, J. Propp,
2913: Markov numbers, Farey sequences, and the Ptolemy recurrence,
2914: unpublished memo (2003).
2915:
2916: \bibitem{CP}
2917: G. Carroll and G. Price,
2918: Two new combinatorial models for the Ptolemy recurrence,
2919: unpublished memo (2003).
2920:
2921: \bibitem{C}
2922: J. Conway,
2923: The Sensual (Quadratic) Form,
2924: Mathematical Association of America, Washington, D.C. (1997).
2925:
2926: \bibitem{CoCo}
2927: J.H. Conway and H.S.M. Coxeter,
2928: Triangulated polygons and frieze patterns,
2929: {\it Math. Gaz.} {\bf 57} (1973), 87--94.
2930:
2931: \bibitem{CG}
2932: J.H. Conway and R.K. Guy, The Book of Numbers,
2933: Springer-Verlag,
2934: New York (1996).
2935:
2936: \bibitem{C1}
2937: H.S.M. Coxeter, Regular Polytopes,
2938: Macmillan, New York (1963);
2939: reprinted by Dover.
2940:
2941: \bibitem{C2}
2942: H.S.M. Coxeter, Frieze patterns,
2943: {\it Acta Arithmetica} {\bf 18} (1971), 297--310.
2944:
2945: \bibitem{C3}
2946: H.S.M. Coxeter,
2947: Cyclic sequences and frieze patterns,
2948: {\it Vinculum} {\bf 8} (1971), 4--7.
2949:
2950: \bibitem{C4}
2951: H.S.M. Coxeter, Regular Complex Polytopes,
2952: Cambridge University Press, London (1974).
2953:
2954: \bibitem{FST}
2955: S. Fomin, M. Shapiro, and D. Thurston,
2956: Cluster algebras and triangulated surfaces, Part I:
2957: Cluster complexes,
2958: arXiv:{\tt math/0608367}.
2959:
2960: \bibitem{FZL}
2961: S. Fomin and A. Zelevinsky,
2962: The Laurent phenomenon, arXiv:{\tt math.CO/0104241}.
2963:
2964: \bibitem{FZY}
2965: S. Fomin and A. Zelevinsky,
2966: Y-systems and generalized associahedra, arXiv:{\tt hep-th/0111053}.
2967:
2968: \bibitem{G}
2969: D. Gale,
2970: The Strange and Surprising Saga of the Somos Sequences,
2971: \emph{Math.\ Intell.} {\bf 13}, 40--42 (1991), and
2972: Somos Sequence Update,
2973: \emph{Math.\ Intell.} {\bf 13,} 49--50 (1991)).
2974: Republished (see pages 2--5, 22--24) in
2975: D. Gale, Tracking the Automatic Ant.
2976: Springer-Verlag, 1998.
2977:
2978: \bibitem{GV}
2979: I.M. Gessel and X. Viennot,
2980: Binomial determinants, paths, and hook length formulae,
2981: {\it Advances in Mathematics} {\bf 58} (1985), 300-321.
2982:
2983: \bibitem{GD}
2984: M. Gordon and W.H.T. Davison,
2985: Theory of resonance topology of fully aromatic hydrocarbons,
2986: {\it I. J. Chem. Phys.} {\bf 20} (1952), 428--435.
2987:
2988: \bibitem{He}
2989: N. Herriot, personal communication; preliminary write-up at
2990: {\tt http://jamespropp.org/reach/Herriot/ptolemywriteup.html}.
2991:
2992: \bibitem{Hi}
2993: D. Hickerson, personal communication.
2994:
2995: \bibitem{Ho}
2996: A. Hone,
2997: Diophantine nonintegrability of a third order recurrence
2998: with the Laurent property,
2999: \emph{J.\ Physics A} {\bf 39} (2006), L171--L177.
3000: {\tt arXiv:math.NT/0601324}.
3001:
3002: \bibitem{I}
3003: A. Itsara, G. Musiker, J. Propp, and R. Viana,
3004: Combinatorial interpretations for the Markoff numbers,
3005: memo dated May 1, 2003;
3006: {\tt http://jamespropp.org/reach/Itsara/markovversion3.pdf}.
3007:
3008: \bibitem{JS}
3009: P. John and H. Sachs,
3010: Wegesysteme und Linearfaktoren in hexagonalen und quadratischen Systemen
3011: (Path systems and linear factors in hexagonal and square systems), in:
3012: Graphen in Forschung und Unterricht (Festschrift K. Wagner),
3013: Barbara Franzbecker Verlag Bad Salzdetfurth 1985; pp. 85--101.
3014:
3015: \bibitem{KS}
3016: R. Kenyon and J. Smillie,
3017: Billiards on rational-angled triangles,
3018: {\it Comment. Math. Helv.} {\bf 75} (2000), 65--108.
3019:
3020: \bibitem{Kuo}
3021: E. Kuo,
3022: Applications of graphical condensation for enumerating matchings and tilings,
3023: {\it Theoret. Comput. Sci.} {\bf 319} (2004), 29--57;
3024: arXiv:{\tt math.CO/0304090}.
3025:
3026: \bibitem{Kup}
3027: G. Kuperberg,
3028: Kasteleyn cokernels,
3029: {\it Electron.\ J.\ Combin.} {\bf 9} (2002), article R29;
3030: arXiv:{\tt math.CO/0108150}.
3031:
3032: \bibitem{L}
3033: B. Lindstr\"om,
3034: On vector representations of induced matroids,
3035: {\it Bull. London. Math. Soc.} {\bf 5} (1973) 85--90.
3036:
3037: \bibitem{M}
3038: G. Musiker,
3039: A conjectured combinatorial interpretation for Markov numbers,
3040: memo dated June 27, 2002;
3041: {\tt http://jamespropp.org/reach/Musiker/NewResults.pdf}.
3042:
3043: \bibitem{R}
3044: G. Rosenberger,
3045: Uber die diophantische Gleichung $ax^2+by^2+cz^2=dxyz$,
3046: {\it J. Reine Angew. Math.} {\bf 305} (1979), 122--125.
3047:
3048: \bibitem{S}
3049: Perfect matchings in hexagonal systems,
3050: {\it Combinatorica} {\bf 4} (1984), 89-99.
3051:
3052: %\bibitem{STT}
3053: %D. Sleator, R. Tarjan, and W. Thurston,
3054: %Rotation distance, triangulations, and hyperbolic geometry,
3055: %{\it J. Amer. Math. Soc.} {\bf 1} (1988), 647--681.
3056:
3057: \bibitem{SS}
3058: D. Speyer and B. Sturmfels,
3059: Tropical mathematics,
3060: arXiv:{\tt math.CO/0408099}.
3061:
3062: \bibitem{ST}
3063: R. Schiffler and H. Thomas,
3064: On cluster algebras arising from unpunctured surfaces,
3065: arXiv:{\tt 0712.4131}.
3066:
3067: \bibitem{T}
3068: D. Thurston, personal communication.
3069:
3070: \bibitem{W}
3071: K. Wagner,
3072: Bemerkungen zum Vierfarbenproblem,
3073: {\it J. Deutschen Math.-Verein.} {\bf 46} (1936), 26--32.
3074:
3075: \bibitem{Z1}
3076: D. Zeilberger,
3077: A combinatorial approach to matrix algebra,
3078: {\it Discrete Math.} {\bf 56} (1985), 61--72.
3079:
3080: \bibitem{Z2}
3081: D. Zeilberger,
3082: Dodgson's determinant-evaluation rule
3083: proved by two-timing men and women,
3084: {\it Elec. J. Comb.} {\bf 4}:2 (1997), R22;
3085: arXiv:{\tt math.CO/9808079}.
3086:
3087: \end{thebibliography}
3088:
3089: \end{document}
3090:
3091: %-----------------------------------------------------------------------
3092: % End of article.tex
3093: %-----------------------------------------------------------------------
3094:
3095: That is, the $(n-1)$st row of the resulting table
3096: will not consist of all 1's.
3097: Putting this differently:
3098: the $n$-tuples $a_1,\dots,a_n$
3099: of positive integers
3100: with the property that the $n$ continuants
3101: $[a_1,a_2,\dots,a_{n-1},a_n]$,
3102: $[a_2,a_3,\dots,a_{n},a_1]$,
3103: \dots
3104: $[a_n,a_1,\dots,a_{n-2},a_{n-1}]$
3105: all equal 1
3106: correspond precisely to the $(2n-4)!/(n-2)!(n-1)!$
3107: ways to triangulate a convex $n$-gon.
3108:
3109: CALCULATIONS
3110:
3111: 1: cos 170 = -.985, sin 170 = .174 (call it a)
3112: 1: cos 210 = -.866, sin 210 = -.500 (call it b)
3113: 1: cos 130 = -.643, sin 130 = .766 (call it c)
3114: 2: cos 90 = .000, sin 90 = 1.000
3115: 3: cos 50 = .643, sin 50 = .766
3116: 4: cos 10 = .985, sin 10 = .174
3117: 5: cos 250 = -.342, sin 250 = -.940
3118: 9: cos 290 = .342, sin 290 = -.940
3119: 13: cos 330 = .866, sin 330 = -.500 (call it x)
3120:
3121: a := 170;
3122: b := 210;
3123: c := 130;
3124: p2 := 90;
3125: p3 := 50;
3126: p4 := 10;
3127: p5 := 250;
3128: p9 := 290;
3129: x := 330;
3130:
3131: Cos := proc(t) evalf(cos(t*2*Pi/360)); end;
3132: Sin := proc(t) evalf(sin(t*2*Pi/360)); end;
3133: X := proc(p,q,r) evalf((Cos(p)+Cos(q)+Cos(r))/3); end;
3134: Y := proc(p,q,r) evalf((Sin(p)+Sin(q)+Sin(r))/3); end;
3135: C := proc(p,q,r) RETURN(X(p,q,r),Y(p,q,r)); end;
3136:
3137: Center of a,b,c: C(a,b,c) = -0.830, 0.146
3138: connect this to b and c
3139: Center of b,c,2: C(b,c,p2) = -0.502, 0.421
3140: connect this to b, c, and p2
3141: Center of b,2,3: C(b,p2,p3) = -0.074, 0.421
3142: connect this to b, p2, and p3
3143: Center of b,3,4: C(b,p3,p4) = 0.254, 0.146
3144: connect this to b, p3, and p4
3145: Center of b,4,5: C(b,p4,p5) = -0.074, -0.422
3146: connect this to b, p4, and p5
3147: Center of 4,5,9: C(p4,p5,p9) = 0.328, -0.568
3148: connect this to p4, p5, and p9
3149: Center of 4,9,x: C(p4,p9,x) = 0.730, -0.422
3150: connect this to p4 and p9
3151:
3152: \psline(abcx,abcy)(bx,by)
3153: \psline(abcx,abcy)(cx,cy)
3154: \psline(cx,cy)(bc2x,bc2y)
3155: \psline(bx,by)(bc2x,bc2y)
3156: \psline(p2x,p2y)(bc2x,bc2y)
3157: \psline(bx,by)(b23x,b23y)
3158: \psline(p2x,p2y)(b23x,b23y)
3159: \psline(p3x,p3y)(b23x,b23y)
3160: \psline(bx,by)(b34x,b34y)
3161: \psline(p3x,p3y)(b34x,b34y)
3162: \psline(p4x,p4y)(b34x,b34y)
3163: \psline(bx,by)(b45x,b45y)
3164: \psline(p4x,p4y)(b45x,b45y)
3165: \psline(p5x,p5y)(b45x,b45y)
3166: \psline(p4x,p4y)(459x,459y)
3167: \psline(p5x,p5y)(459x,459y)
3168: \psline(p9x,p9y)(459x,459y)
3169: \psline(p4x,p4y)(49xx,49xy)
3170: \pscircle[fillstyle=solid,fillcolor=white](ax,ay){.1}
3171: \pscircle[fillstyle=solid,fillcolor=black](bx,by){.1}
3172: \pscircle[fillstyle=solid,fillcolor=black](cx,cy){.1}
3173: \pscircle[fillstyle=solid,fillcolor=black](p2x,p2y){.1}
3174: \pscircle[fillstyle=solid,fillcolor=black](p3x,p3y){.1}
3175: \pscircle[fillstyle=solid,fillcolor=black](p4x,p4y){.1}
3176: \pscircle[fillstyle=solid,fillcolor=black](p5x,p5y){.1}
3177: \pscircle[fillstyle=solid,fillcolor=black](p9x,p9y){.1}
3178: \pscircle[fillstyle=solid,fillcolor=black](xx,xy){.1}
3179: \pscircle[fillstyle=solid,fillcolor=white](abcx,abcy){.1}
3180: \pscircle[fillstyle=solid,fillcolor=white](bc2x,bc2y){.1}
3181: \pscircle[fillstyle=solid,fillcolor=white](b23x,b23y){.1}
3182: \pscircle[fillstyle=solid,fillcolor=white](b34x,b34y){.1}
3183: \pscircle[fillstyle=solid,fillcolor=white](b45x,b45y){.1}
3184: \pscircle[fillstyle=solid,fillcolor=white](459x,459y){.1}
3185: \pscircle[fillstyle=solid,fillcolor=white](49xx,49xy){.1}
3186:
3187: 1 1
3188: 2 1 1 1
3189: 7 3 2 1 1 1
3190: 26 11 7 3 2 1
3191:
3192: 1
3193: 1 2
3194: 1 3
3195: 1 7
3196: 1 2 11
3197: 1 3 26
3198: 1 7
3199: 1 2
3200: 1
3201:
3202: o
3203: / \
3204: o o
3205: / \ /
3206: o o
3207: \ / \
3208: o o
3209: \ / \
3210: o o
3211: / \ /
3212: o o
3213: \ / \
3214: o o
3215: \ / \
3216: o o
3217: / \ /
3218: o o
3219: \ / \
3220: o o
3221: \ /
3222: o
3223:
3224: 1
3225: / \
3226: 1 2
3227: \ / \
3228: 1 3
3229: \ / \
3230: 1 7
3231: / \ /
3232: 1 2 11
3233: \ / \
3234: 1 3 26
3235: \ / \
3236: 1 7
3237: / \ /
3238: 1 2
3239: \ /
3240: 1
3241:
3242: Relate the shape of the snake to the shape of the frieze zig-zag.
3243:
3244: 1
3245: / \
3246: 1 2
3247: \ / \
3248: 1 3
3249: \ / \
3250: 1 7
3251: / \ /
3252: 1 2
3253: \ /
3254: 1
3255:
3256: Stress that not every triangulation is of this form.
3257:
3258: Then go on to give the paths model.
3259:
3260: Then mention G-V.
3261:
3262: 8 1_a
3263: / \
3264: 7 1_b 5
3265: / \ /
3266: 6 1 4 9
3267: / \ /
3268: 5 1 3 7 13
3269: / \ /
3270: 4 1 2 5 10
3271: \ / \
3272: 3 1 3 7
3273: \ / \
3274: 2 1_c 4
3275: \ /
3276: 1 1_d
3277:
3278: 1 2 3 4 5 6 7 8
3279:
3280: $$
3281: \pspicture(0,0)(9,9)
3282: \rput(4,1){1_d}
3283: \rput(3,2){1_c}
3284: \rput(5,2){4}
3285: \rput(2,3){1}
3286: \rput(4,3){3}
3287: \rput(6,3){7}
3288: \rput(1,4){1}
3289: \rput(3,4){2}
3290: \rput(5,4){5}
3291: \rput(7,4){10}
3292: \rput(2,5){1}
3293: \rput(4,5){3}
3294: \rput(6,5){7}
3295: \rput(8,5){13}
3296: \rput(3,6){1}
3297: \rput(5,6){4}
3298: \rput(7,6){9}
3299: \rput(4,7){1_b}
3300: \rput(6,7){5}
3301: \rput(5,8){1_a}
3302: \psline(3.8,1.2)(3.2,1.8)
3303: \psline(1.8,2.2)(2.2,2.8)
3304: \psline(1.8,3.2)(1.2,3.8)
3305: \psline(4.8,2.2)(4.2,2.8)
3306: \psline(2.8,3.2)(3.2,3.8)
3307: \psline(2.8,4.2)(2.2,4.8)
3308: \psline(3.8,5.2)(3.2,5.8)
3309: \psline(4.8,6.2)(4.2,6.8)
3310: \psline(5.8,7.2)(5.2,7.8)
3311: \psline(4.2,1.2)(4.8,1.8)
3312: \psline(3.2,2.2)(3.8,2.8)
3313: \psline(2.2,3.2)(2.8,3.8)
3314: \psline(3.2,4.2)(3.8,4.8)
3315: \psline(4.2,5.2)(4.8,5.8)
3316: \psline(5.2,6.2)(5.8,6.8)
3317: \psline(1.2,4.2)(1.8,4.8)
3318: \psline(2.2,5.2)(2.8,5.8)
3319: \psline(3.2,6.2)(3.8,6.8)
3320: \psline(4.2,7.2)(4.8,7.8)
3321: \endpspicture
3322: $$
3323:
3324: [9]
3325:
3326:
3327: [11] S. Karlin and G. McGregor. Coincidence probabilities.
3328: Pacific Journal of Mathematics,
3329: 9:1141-1164, 1959.
3330: [12] S. Karlin and G. McGregor.
3331: Coincidence probabilities of birth-and-death processes.
3332: Pacific Journal of Mathematics, 9:1109-1140, 1959.
3333: