1: \section{Pictures }\label{pictures}
2:
3: Let $A$ and $B$ be groups and $m\geq 2$ an integer and set $H=A*B$.
4: Let $X^\prime_A$
5: and $X_B^\prime$ be CW--complexes with $\pi_1(X_A^\prime,x_A)=A$ and
6: $\pi_1(X_B^\prime,x_B)=B$, for some $x_A\in X_A^\prime$ and $x_B\in X_B^\prime$.
7: Let $O=A$ or $O=B$ and let $e_O$
8: be a copy of the
9: unit interval $[0,1]$.
10: Form the quotient space $X_O$ of $X^\prime_O\dcup e_O$
11: by identifying $x_O$ and $1\in e_O$.
12: Form the quotient space $Y$ of $X_A\dcup X_B$
13: by identifying $0\in e_A$ and $0\in e_B$ to a point $p$.
14: Then $\pi_1(Y,p)=H$.
15:
16: Fix a word $r\in H$ with
17: $l (r)\geq 2$, such that $r$ is not a proper power in $H$. Let
18: $s=r^m$, $N(s)=\langle\langle s\rangle\rangle^H$ and
19: $G=H/N(s).$ Suppose that $s=a_1b_1\cdots a_nb_n$, when written in normal form for the free product.
20: Let $\D_s$ be a closed unit disk with oriented boundary
21: divided into $2n$ intervals. Label the intervals consecutively $a_1,b_1,\ldots, a_n,b_n$,
22: so that $\pd\D_s$, read from a suitable starting point in the direction of orientation, is
23: labelled $s$. Form the quotient space
24: $Z$ of $Y\dcup \D_s$ by identifying $\pd\D_s$ with a path in $(Y,p)$ representing
25: $s\in H$, according to the label of $\pd\D_s$. Then $\pi_1(Z,p)=G$.
26: Alternatively, let $G=H$ and define $\D_s=p\in Y$, so that $Z=Y$ and $\pi_1(Z,p)=G$ again.
27:
28: Let $\S$ be a compact surface and let $f:\S\maps Z$ be a continuous map satisfying the
29: following properties.
30: \be[$S$1.]
31: \item \label{sketch1} $f$ is transverse to the midpoint $c$ of $\D_s$ and
32: %if $x\in f^{-1}(c)$ then
33: %the connected component
34: %of $f^{-1}(\D_s)$ containing $x$ is a regular neighbourhood of $x$
35: the closure of $f^{-1}(\inr(\D_s))$ is a regular neighbourhood of $f^{-1}(c)$.
36: \item \label{sketch2} $f$ is transverse to $p$ on
37: $\S_0=\S\bl f^{-1}(\inr(\D_s))$ and $f^{-1}(e_A\cup e_B)$ is a
38: regular neighbourhood of $f^{-1}(p)$ in $\S_0$.
39: \ee
40:
41: Let $\G= f^{-1}(\D_s)$. Then $(f,\G)$ is a {\em sketch} on $\S$ over $G$ (with respect
42: to $Z$) which we refer to merely as $\G$, when the meaning is clear. A connected
43: component of the closure of $f^{-1}(\inr(\D_s))$ is called a {\em vertex}
44: of $\G$ and the set of all vertices is denoted $\cV(\G)$ or just $\cV$.
45: Condition
46: $S$\ref{sketch1} implies that $\cV$
47: consists of a collection of
48: embedded disks in $\inr(\S)$, each of which is a regular neighbourhood of a point
49: of $f^{-1}(c)$.
50: If $G=H$ then $\cV=\emptyset$.
51: A connected component of $f^{-1}(p)$ is called an {\em arc} of $\G$ and the
52: set of all arcs is denoted $\cA=\cA(\G)$. Condition $S$\ref{sketch2} implies
53: that $\cA$ consists of a set of properly embedded $1$--submanifolds of $\S_0$.
54:
55: A connected component of $\pd v\bl \{\cup e: e\in \cA\}$, where $v\in \cV$, is called a {\em vertex corner}
56: of $\G$.
57: % and the
58: %set of vertex corners is denoted $\cC_{\cV}(\G)$ or $\cC_{\cV}$.
59: A component of $\pd\S\bl \G$ is called a {\em boundary corner} of $\G$. %and the
60: %set of boundary corners is denoted $C_{\pd}(\G)$ or $C_{\pd}$.
61: The closure of
62: a connected component of $\S\bl \G$ is called a {\em region} of $\G$ and the
63: set of regions is denoted $\cR(\G)$ or $\cR$.
64: Every vertex and boundary corner belongs to a unique region of $\G$.
65:
66: Now suppose that $\S^\prime$ is a connected component of $\S$
67: and define $\G^\prime=\S^\prime \cap \G$. If $\S^\prime$ is
68: orientable we may choose consistent
69: orientations of each region and vertex of $\S^\prime$. We call such a
70: choice an {\em orientation} of $\G^\prime$. If $\bt$ is a boundary component
71: of $\S^\prime$ then an orientation $\z$ of $\G^\prime$ gives
72: rise to a unique orientation $\z(\bt)$ of $\bt$. Similarly the orientation of
73: regions induces an orientation on $\pd v$, which we denote
74: $\z(\pd v)$, for all $v\in \cV$. Thus $\z(\pd v)$ is the opposite of
75: the orientation induced on $\pd v$ by the orientation of $v$.
76: If $l$ is a sub--interval of a corner $c$ of $\G$ then there is a unique
77: orientation $\z(l)$ of $l$ induced from the orientation of the
78: boundary component (of $\S^\prime$ or $v\in \cV$) to which $c$ belongs.
79:
80: If $\S^\prime$ is non--orientable then we choose an
81: orientation
82: for each connected component of $\pd\S^\prime$,
83: for each connected component of $\pd \D$, for all regions
84: $\D$ of $\G^\prime$,
85: and for $\pd v$, for all vertices $v$ of $\G^\prime$.
86: To each vertex $v$ we assign the orientation opposite to that chosen for
87: $\pd v$.
88: %$\S^\prime\bl(\bigcup_{\D\in \cR}\inr(\D)\cup\bigcup_{v\in \cV}\inr(v))$
89: %of $\pd v$, for all vertices $v\in \cV$, on all boundary
90: %components of $\bt$ of $\S^\prime$ and on $\pd(\D)$, for all regions
91: %$\D\in \cR(\G)$
92: Call the collection of these orientations $\z$
93: an {\em orientation} of $\G^\prime$.
94: Let $l$ be a sub--interval of a corner $c$ of $\G^\prime$. Then
95: $c$ belongs to a boundary component $\bt$ of $\S^\prime$ or to the boundary $\bt$ of
96: a vertex. The orientation $\zeta(\bt)$ induces an
97: orientation of $l$ which we call $\zeta(l)$.
98:
99: We now fix an orientation for $\G^\prime$ for all connected
100: components $\S^\prime$ of $\S$. We call these orientations an {\em orientation}
101: of $\G$ and denote this by $\z$. We refer
102: to the orientation induced on $y$ by $\z$ as $\z(y)$, whenever $y$ is an appropriate
103: subset of $\S$.
104: %The sketch $\G$ together with the orientation $\z$ is called a
105: %{\em directed} sketch.
106:
107: Let $(f,\G)$ be a %directed
108: sketch on $\S$
109: with orientation $\z$. We define a labelling function $f^l$ from corners of $\G$ to $H$
110: as follows. Let $c$ be a corner of $\G$ with
111: orientation $\z(c)$ and let
112: $q$ be the closed path $f(c)$ in $Y$. % with orientation induced from $\z(c)$.
113: If $c$ meets an arc of $\G$ then $q$ is a directed path based at $p$.
114: In this case we
115: define the {\em label} $f^l(c)$ {\em of} $c$ {\em read in the direction} $\zeta(c)$
116: to be the element of $H$
117: represented by the based directed path $q$.
118: If $c$ does not meet an arc of $\G$ then we choose a base point
119: $x$ on $c$ and regard $q$ as a directed path based at $f(x)$.
120: Since $c$ meets no arc of $\G$, the path $q$ lies in $Y\backslash p$ and we may
121: choose a path $\al$ from $p$ to $x$ such that int$(\al)$ lies in $X_A$ or $X_B$.
122: Then the directed path $\al q\al^{-1}$ represents an element $g$ of $A\dcup B$.
123: The label of $c$, read in the direction of orientation $\z(c)$, is defined to
124: be $g$.
125: %(In the latter case
126: %$c$ is contained in an $A$-- or $B$--region of $\G$.)
127: If $c$ read in the direction $\zeta(c)$
128: has label $w$ then
129: the label of $c$ {\em read in the direction opposite to} $\zeta(c)$ is $w^{-1}$.
130: \begin{defn}\label{pic_def}
131: The sketch $\G$ endowed with the orientation $\z$ and the labelling function $f^l$ is
132: called a {\em picture} on $\S$ over $G$.
133: \end{defn}
134:
135: A picture over $G=H$ may be regarded as a picture over $G=H/N(s)$, having no vertices. Conversely
136: a picture over $G=H/N(s)$ with no vertices may be regarded as a picture over $H$.
137: If $v$ is a vertex of $\G$ and $x$ is a point of
138: $\pd v\cap \{\cup e:e\in \cA\}$ then the {\em label} of $v$ with {\em base point} $x$
139: is the word of $H$ obtained
140: by reading off the labels of corners of $v$ in the order they occur whilst travelling
141: once round $\pd v$, from $x$ in the direction $\z(\pd v)$: where the label of
142: each corner $c$
143: is read in the direction of orientation induced from that of
144: $\zeta(\pd v)$, namely $\zeta(c)$.
145: Since $f(\pd v)=\pd \D_s$ we have:
146: \be[$P$1.]
147: \item \label{pic_cond0} the label of $v$ (with any base point)
148: is a cyclic permutation of $s^{\e}$, where
149: $\e=\pm 1$.
150: \ee
151: The orientation $\z$ on $\pd v$ induces an orientation on
152: $f(\pd v)=\pd \D_s$. If this is the same as the fixed orientation of $\pd \D_s$
153: then we may take $\e = 1$ and we
154: say that $v$ is a {\em positive} vertex (with respect to $\z$). Otherwise we take $\e=-1$ and
155: say $v$ is a {\em negative} vertex (with respect to $\z$), as shown in Figure \ref{vertex-pos-neg}.
156: \begin{figure}
157: \psfrag{a}{$a$}
158: \psfrag{b}{$b$}
159: \psfrag{c}{$c$}
160: \psfrag{d}{$d$}
161: \psfrag{a^-1}{$a^{-1}$}
162: \psfrag{b^-1}{$b^{-1}$}
163: \psfrag{c^-1}{$c^{-1}$}
164: \psfrag{d^-1}{$d^{-1}$}
165: \begin{center}
166: \mbox{
167: \subfigure[Positive\label{vertex-pos}]%\label{diag1-2-1}
168: { \includegraphics[scale= 0.4,clip]{fig/vertex-pos.eps} }\qquad
169: \subfigure[Negative\label{vertex-neg} ]
170: { \includegraphics[scale=0.4,clip]{fig/vertex-neg.eps} }
171: }
172: \caption{The orientation of the boundary is shown outside and the orientation of the
173: vertex inside: $s=abcd$.}\label{vertex-pos-neg}
174: \end{center}
175: \end{figure}
176: Similarly, if $\bt$ is a connected component of
177: $\pd\S$ or a connected component $\pd\D$, for some region
178: $\D$ of $\G$, let $x\in \bt\cap \{\cup e:e\in \cA(\G)\}$ or, if $\bt$ consists of
179: a boundary corner meeting no arc of $\G$, let $x$ be the chosen base point of this
180: corner. Then
181: the {\em label} of $\bt$ with {\em base point} $x$ is the word obtained by reading
182: once round $\bt$, from $x$ in the direction of $\z(\bt)$. When reference
183: is made to the label of such $\bt$ it is to be assumed that a base
184: point has been chosen. The {\em boundary labels} of
185: a region $\D$ are the labels of the connected components of $\pd\D$.
186:
187: As a consequence of properties $S$\ref{sketch1} and $S$\ref{sketch2}
188: and the definitions above a
189: picture $\G$ on $\S$ satisfies the following properties
190: $P$\ref{pic_cond1}--$P$\ref{pic_cond3}.
191: \be[$P$1.]\label{piccond}
192: \addtocounter{enumi}{1}
193: \item \label{pic_cond1} Let $\D$ be a region of $\G$ with boundary labels
194: $d_1,\ldots, d_n$. Then either $d_i\in A$, for $i=1,\ldots , n$, in which case
195: $\D$ is called an $A${\em --region} or
196: $d_i\in B$, for $i=1,\ldots , n$, in which case
197: $\D$ is called a $B${\em --region}.
198: \item \label{pic_cond2}
199: Let $O=A$ or $B$ and let $\D$ be an $O$--region, with boundary labels
200: $d_1,\ldots , d_n$, which is homeomorphic to the connected sum of an $n$-punctured
201: sphere with $t$ torii and $p$ projective planes. Then there exist elements
202: $w_1,\ldots , w_n, x_1,\ldots, x_t,y_1,\ldots, y_t,z_1,\ldots,z_p\in O$ such that
203: \begin{equation}\label{region_cond}
204: \prod_{i=1}^{n} w_i^{-1}d_iw_i\prod_{i=1}^t[x_i,y_i]\prod_{i=1}^p z_i^2=1,
205: \end{equation}
206: in $O$.
207: \item \label{pic_cond3}
208: If $a$ is an arc of $\G$ then $a$ separates an $A$--region from a $B$--region.
209: \ee
210:
211: Conversely properties $P$\ref{pic_cond0} to $P$\ref{pic_cond3} characterise pictures.
212: \begin{lemma}\label{pic_comb}
213: Let $\S$ be a compact surface and let $\G$ consist of
214: \be
215: \item a finite collection $\cV$ of disjoint
216: closed disks in int$(\S)$ (called vertices);
217: \item a
218: finite collection $\cA$ of disjoint connected
219: properly embedded $1$--submanifolds of
220: $\S_0=\S\backslash\cup_{v\in \cV}\textrm{int}(v)$ (called arcs);
221: \item
222: regions and corners and
223: an orientation $\z$ defined as
224: in the case of the sketch above;
225: \item a labelling function $f^l$
226: from corners of $\G$ to $A\cup B$, which respects orientation, and a collection
227: of labels defined
228: as for pictures above.
229: \ee
230: If $\G$ satisfies $P$\ref{pic_cond0} to $P$\ref{pic_cond3} then $\G$ is a picture on $\S$
231: over $G$.
232: \end{lemma}
233: We define the {\em genus} of a compact connected surface
234: $\S$, with $k$ boundary components, to be
235: genus$(\S)=(2-\chi(\S)-k)/2$. The {\em genus} of a disconnected compact surface
236: is defined to be the sum of the genera of its connected components.
237:
238: An arc of a picture, which is not a closed curve, both ends of which
239: meet $\pd \S$ is called an arc of {\em type} $I$.
240: Let $\D$ be a region of $\G$. Then we define
241: \begin{align*}
242: t(\D)&= \mbox{ number of arcs of } \G \mbox{ in }\pd\D,\\
243: \bt(\D)&= \mbox{ number of components of }[\pd\S\backslash \{\mbox{arcs of }\G\}]\cap\pd\D\\
244: &=\textrm{ number of boundary corners of } \D,\\
245: \rho(\D)&=
246: \mbox{ number of components of }[\G\backslash \{ \mbox{arcs of } \G\}]\cap\pd\D\\
247: &=\textrm{ the number of vertex corners of }\D,\\
248: \g(\D)&= \mbox{ number of components } \pd\D \textrm{ meeting no
249: % vertex or component of }\pd\S \textrm{ and }\\
250: %& \textrm{\bf should this
251: % be no of bdry comps of } \D \textrm{ not meeting}
252: arc of }\G\\
253: %$$a(\D) \mbox{ or }\phi(\D)=\mbox{ angle on all corners of }\D
254: %\mbox{ (not well defined for boundary regions)}\\
255: %and
256: \e(\D)&= \textrm{ number of arcs of $\G$ in }\pd\D \mbox{ meeting no vertex of }\G\\
257: &= \textrm{the number of arcs which are closed curves or of type } I \textrm{ in }\pd\D.
258: \end{align*}
259: If $\x(\D )=1$ and $t(\D )=2$ then we call $\D$ {\em collapsible}.
260:
261: \subsection{Boundary intervals}
262: Let $\bt$ be a connected component of $\pd\S$, with orientation $\z$, and let
263: $b_1,\ldots ,b_n$ be $n$ points occuring consecutively around
264: $\bt$ (read in the direction of $\z(\bt)$). We call $(b_1,\ldots ,b_n)$ a
265: {\em boundary partition} of $\bt$. The closed intervals $[b_{i},b_{i+1}]$ of $\bt$,
266: for $i=1,\ldots ,n$ (subscripts modulo $n$) are called {\em boundary intervals} of
267: $\bt$. The boundary interval $[b_{i},b_{i+1}]$ has {\em length} $|[b_{i},b_{i+1}]|$ equal
268: to the number of connected components of $[b_{i},b_{i+1}]\bl \G$.
269: (The $b_i$'s do not have to belong to $\G\cap \bt$.)
270:
271: Let $\cL$ be a consistent system of parameters
272: and $z=\prod_{i=1}^n(h_i,m_i)\in F(H^\Lm)$. The boundary partition $(b_1,\ldots ,b_n)$ is said to
273: $\cL${\em --admit} $z$ if (with subscripts modulo $n$)
274: \be
275: \item $|[b_{i},b_{i+1}]|=1$, for all $i$ such that $l(h_i)=1$,
276: \item $|[b_{i},b_{i+1}]|=l(h_i\^m_i)$, for all $i$ such that $(h_i,m_i)$ is degenerate and
277: \item $|[b_{i},b_{i+1}]|=q_il(h_i)+k_i$, for some $k_i,q_i\in\ZZ$ such that $q_i\ge 0$ and $\cL$ implies that
278: $m_i\equiv k_i (\mod l(h_i))$, for all $i$ such that $l(h_i)>1$ and $(h_i,m_i)$ is non--degenerate.
279: \ee
280: If $\al$ is a solution to $\cL$, the boundary partition $(b_1,\ldots ,b_n)$ $\cL$--admits $z$
281: and $[b_{i},b_{i+1}]$ maps to a path in $Y$ representing
282: $\hat\al(h_i,m_i)$,
283: then we say that
284: $[b_{i},b_{i+1}]$ has {\em label} $\hat\al(h_i,m_i)$ and
285: {\em prime label} $(h_i,m_i)$,
286: for $i=1,\ldots ,n$.
287: In this case the label of $\bt$ is $\hat\al(z)$ and we say
288: that $\bt$ has {\em prime label} $z$ with {\em partition}
289: $(b_1,\ldots ,b_n)$.
290:
291: Let $(\mbf n,\mbf t,\mbf p)$
292: be a positive 3--partition of $(n,t,p)$ of
293: length $k$, with $\mbf n=(n_1,\ldots, n_k)$ etc..
294: We shall say that a surface has {\em type}
295: $(\mbf n,\mbf t,\mbf p)$
296: if it is a disjoint union of compact
297: surfaces $\S_i$, $i=1,\ldots ,k$, such that $\S_i$ has
298: $n_{i}$ boundary components, genus $t_{i}+p_{i}/2$ and is orientable if $p_{i}=0$.
299: In this case, if $\G$ is a picture on $\S_1\dcup \cdots \dcup \S_k$ we say that $\G$ has
300: {\em partitioned boundary labels list}
301: $(u_1,\ldots,u_n)$
302: if
303: the boundary labels of $\G\cap \S_i$ are $u_{\mu_i+1},\ldots ,u_{\mu_i+n_{i}}$, for
304: $i=1,\ldots , k$, where
305: $\mu_i$ is defined as above.
306:
307: Now with $H=A*B$ and $G=H/N(s)$ or $G=H$ as above, let $(\mbf h,\mbf n,\mbf t,\mbf p)$
308: be a positive 4--partition,
309: $\mbf z$ a special
310: element of $(H^\Lm,\cL,\mbf n,\mbf h)$ and $\al$ a solution to $\cL$. Suppose that $\G$ is a picture
311: on a surface
312: of type $(\mbf n,\mbf t,\mbf p)$ with partitioned boundary labels list
313: $(\hat\al(z_1),\ldots,\hat\al(z_n))$.
314: Then we may choose a boundary partition $\mbf b_i$ for the boundary component $\bt_i$
315: labelled $\hat\al(z_i)$
316: such that $\mbf b_i$ $\cL$--admits $z_i$.
317: Then $\bt_i$ has prime label $z_i$ with partition $\mbf b_i$, for all $i$.
318: \begin{defn}\label{boundary_partition}
319: Given such boundary partitions we set $\mbf b= \mbf b_1,\ldots,\mbf b_n$ and say that
320: $\G$ has {\em boundary partition} $\mbf b$ and {\em prime labels} $\mbf z$ with
321: {\em partition} $\mbf b$. For ease of notation we write $\hat\al(\mbf z)$ for
322: $(\hat\al(z_1),\ldots,\hat\al(z_n))$.
323: \end{defn}
324: %From now on we shall assume that a picture comes equipped with
325: %fixed
326: %boundary partitions $\mbf b_i$ of all connected compents
327: %of $\pd\S$, satisfying these properties.
328: %In this case we say that $\G$ has {boundary partition} $\mbf b=\mbf
329: %b_1,\ldots \mbf b_n$, {\em prime labels} $\mbf z$ and
330: %is {\em labelled} by $\hat\al(\mbf z)$.
331:
332: \begin{comment}
333: Let $w$ be a word of $A\ast B$ which has no cyclic subword $s$
334: such that $s$ is also a cyclic subword of
335: $r^{\pm m}$ and $l (s) \ge {ml / 2} + 1$. Then we say that $w$ is an
336: $r^m${\em --reduced} word.
337: A picture {\em label--reduced}
338: if the boundary label of every boundary interval $[b,c]$ is an $r^m$--reduced word of
339: $A\ast B$. {\bf is this necessary if} $\mbf z$ {\bf is special?}
340: \begin{defn}
341: \label{r-vertex}
342: Suppose that $[b_i,b_{i+1}]$ is a boundary interval, of some boundary component of $\S$,
343: with prime label an $(\cL,r)$--letter of $H^\Lm$. Then we say that $[b_i,b_{i+1}]$ is an
344: $r${\em --boundary interval}.
345: Let $v$ be a vertex of
346: $\G $ incident to an $r$--boundary interval. Then we say that $v$ is
347: an $r${\em --vertex}.
348: The numbers of $r$-boundary intervals and $r$--boundary vertices of
349: $\G$ are denoted by $\cD_R$ and $\cV_R$, respectively.
350: \end{defn}
351: \end{comment}
352: \begin{defn}
353: \label{h-vertex}
354: Let $a$ be a boundary arc and let $x$ be an endpoint of $a$ on $\pd\S$.
355: Then $x$ meets 2 distinct boundary corners: denote the closures of these corners
356: by $c_1$ and $c_2$. If int$(c_1\cup c_2)$ contains a point of a boundary partition
357: then $a$ is called an $H^\Lm${\em --arc}.
358: A vertex incident to an $H^\Lm$--arc is
359: called an $H^\Lm${\em --vertex}. The sets of $H^\Lm$-arcs and $H^\Lm$--vertices of
360: $\G$ are denoted by $\cA_B$ and $\cV_B$, respectively.
361: \end{defn}
362:
363: \begin{defn}
364: A boundary interval with prime label a minor $H^\Lm$ letter
365: is called a {\em minor} boundary interval. A minor boundary interval with
366: label of exponential length at least $1$ is called a {\em partisan} boundary
367: interval.
368: The boundary components of $\S$
369: %meeting no arc
370: containing partisan boundary intervals
371: are called {\em partisan}
372: boundary components. A region of $\G$ containing a partisan boundary interval in
373: its boundary is called a {\em partisan} region.
374: \end{defn}
375: %Note that all the boundary intervals in the boundary partition of
376: %a degenerate boundary component are minor.
377: %
378: %Now suppose that $\G$ is a picture with prime labels $\mbf z
379: %=(z_1,\ldots ,z_n)$,
380: %labelled $(\mbf z, \al)$, where $\mbf z\in (F(H^\Lm),\cL,\mbf n,\mbf h)$,
381: %as above.
382: For future use we record the following result.
383: \begin{lemma}\label{HLbound}
384: The number of $H^\Lm$--vertices of $\G$ is at most $2W_1$ (see (\ref{W1})).
385: \end{lemma}
386:
387: %\medskip
388: \noindent
389: {\em Proof.} Traverse each boundary component of $\pd\S$ once in the
390: direction of orientation $\z$ counting $H^\Lm$--vertices as they occur.
391: \medskip
392:
393: %Recall that we have defined $R$ to be
394: %the number of $r$--boundaries of $\G $,
395: %$M = \max \{\mbox{\rm lcm}(t,m)/ t - 1 :
396: %r^{\pm t} \mbox{ is the base-label of an $r$--boundary}\,\}$,
397: %$W = \max \{ l (w_i) : i=1,\cdots ,n \}$ and $K = \max \{ml ,W^2\}$.
398: %
399: %Let $K$ and $M$ be defined as in Section \ref{isoperimetry}
400: %and
401: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%555
402: \subsection{Bridge moves and vertex cancellation}
403: Let $u_1$ and $u_2$ be vertices of a picture $\G$,
404: with orientation $\z$, and let $e$ be an arc with end--points $x_1$ and $x_2$
405: on $u_1$ and $u_2$, respectively.
406: Let $N$ be a regular
407: neighbourhood of $u_1\cup u_2\cup e$. If $N$ is orientable
408: let $\z_e$ be an orientation of $N$.
409: If $\z_e(u_i)=\z(u_i)$, for both $i=1$ and $2$, or $\z_e(u_i)\neq\z(u_i)$, for both $i=1$ and $2$,
410: then set $\de=1$. Otherwise set $\de=-1$. If $N$ is non--orientable set $\de=1$.
411: The latter occurs if $u_1=u_2$ and $N$ is
412: homeomorphic to a M\"obius band (see Figure \ref{twisted_loop}).
413: \begin{figure}
414: \psfrag{e}{$e$}
415: \psfrag{u}{$u$}
416: \psfrag{N}{$N$}
417: \begin{center}
418: \mbox{
419: \subfigure[$N$ non--orientable\label{twisted_loop}]%\label{diag1-2-1}
420: { \includegraphics[scale= 0.3,clip]{fig/twisted_loop.eps} }\qquad
421: \subfigure[$N$ orientable\label{flat_loop}]
422: { \includegraphics[scale=0.3,clip]{fig/flat_loop.eps} }
423: }
424: \caption{}\label{loop}
425: \end{center}
426: \end{figure}
427: For $i=1,2$ set $\e_i=1$ if $u_i$ is positive and $\e_i=-1$ if $u_i$ is negative.
428: \begin{defn}\label{e_confluent}
429: If $\de\e_1\e_2=-1$ we say that $u_1$ and $u_2$ are $e${\em --confluent}.
430: \end{defn}
431: Thus $u_1$ and $u_2$ are $e$--confluent if $f$ preserves orientation $\z_e$
432: on precisely one of $\pd u_1$ and
433: $\pd u_2$.
434: Figure \ref{e-confluent-y} shows a pair of $e$--confluent vertices, with $\e_1=\de=1$ and $\e_2=-1$.
435: Figure \ref{e-confluent-n} shows a pair of vertices which are not $e$--confluent,
436: with $\e_2=\de=-1$ and $\e_1=1$.
437: \begin{figure}
438: \psfrag{e}{$e$}
439: \psfrag{r^m}{$r^m$}
440: \psfrag{~r^m}{$\sim r^{-m}$}
441: \begin{center}
442: \mbox{
443: \subfigure[Confluent\label{e-confluent-y}]%\label{diag1-2-1}
444: { \includegraphics[scale= 0.37,clip]{fig/e-confluent-y.eps} }\qquad
445: \subfigure[Not confluent\label{e-confluent-n} ]
446: { \includegraphics[scale=0.37,clip]{fig/e-confluent-n.eps} }
447: }
448: \caption{$\sim r^{-m}$ denotes a cyclic permutation of $r^{-m}$}\label{e-confluent}
449: \end{center}
450: \end{figure}
451:
452: If $u_1=u_2$ and a regular neighbourhood $N$ of $u_1\cup e$ is non--orientable then $u_1$ is $e$--confluent
453: to itself, as in Figure \ref{twisted_loop}. On the other hand if $N$ is orientable, as in Figure \ref{flat_loop},
454: then $u_1$ is not $e$--confluent to itself.
455:
456: In the above setting let $w_i$ be the label of $u_i$ with base point $x_i$ (so
457: $w_i$ is obtained by reading in the direction opposite to $\z(u_i)$).
458: If $w_1w_2^{\de}=1$
459: in $H$ then $u_1$ and $u_2$ are said to be vertices which {\em cancel}
460: (see Figure \ref{vertex_cancel}).
461: \begin{figure}
462: \psfrag{a}{$a$}
463: \psfrag{b}{$b$}
464: \psfrag{c}{$c$}
465: \psfrag{d}{$d$}
466: \psfrag{a^-1}{$a^{-1}$}
467: \psfrag{b^-1}{$b^{-1}$}
468: \psfrag{c^-1}{$c^{-1}$}
469: \psfrag{d^-1}{$d^{-1}$}
470: \psfrag{e}{$e$}
471: \psfrag{w1}{$w_1$}
472: \psfrag{w1-1}{$w_1^{-1}$}
473: \begin{center}
474: {\mbox{
475: \subfigure[$\de=1$\label{vertex_cancel-cy}]%\label{diag1-2-1}
476: { \includegraphics[scale=0.37]{fig/vertex_cancel.eps} }\qquad
477: \subfigure[$\de=-1$\label{vertex_cancel-cn} ]
478: {\includegraphics[scale=0.37]{fig/vertex_cancel-cn.eps} }
479: }}
480: \caption{Cancelling vertices}\label{vertex_cancel}
481: \end{center}
482: \end{figure}
483: A picture
484: with no pair of distinct vertices which cancel is said to be {\em reduced}.
485:
486: Suppose that $u_1$ and $u_2$ cancel. If $u_1$ and $u_2$ are $e$--confluent then we may
487: assume, without loss of generality, that $\e_1=\de=1$ and $\e_2=-1$, that $w_1=r^m$ and $w_2$ is
488: a cyclic permutation of $r^{-m}$, so $w_2^{-1}$ is a cyclic permutation of $r^m$.
489: Suppose that $r=r_0r_1$ and $w_2^{-1}=(r_1r_0)^m$. As $u_1$ and $u_2$ cancel and $\de=1$ we have
490: \[(r_0r_1)^m=w_1=w_2^{-1}=(r_1r_0)^m.\]
491: It follows that $r^m$ has periods $\lm=\textrm{min}\{|r_0|,|r_1|\}$ and $l$. From
492: \cite[Section 3, Proposition 1]{\Ha} it follows that $r^m$, and hence $r$, has period $d=\gcd(\lm,l)$.
493: By assumption $r$ is not a proper power so this cannot happen unless $d=l$ and $\lm=0$, in which case
494: $w_2^{-1}=r^m$.
495:
496: Now suppose that $u_1$ and $u_2$ cancel but are not $e$--confluent. Without loss of generality
497: we may assume that $\e_1=\e_2=\de=1$, that $w_1=r^m$ and that $w_2=r_1r^{m-1}r_0$, where $r=r_0r_1$.
498: We have
499: \[(r^{-1}_1r_0^{-1})^m=w_1^{-1}=w_2=r_1r^{m-1}r_0,\]
500: so $r_1^2=r_0^2=1$. It follows that $r$ has a cyclic permutation $xUyU^{-1}$, where
501: $x,y\in A\dcup B$, with $x^2=y^2=1$. That is $G$ is of type $E(2,2,m)$.
502:
503: Now suppose that $a$ and $b$ are distinct arcs of $\G$ and that $\g$ is
504: the image of an embedding of $[0,1]$
505: in $\S$ with one end--point on $a$ and the other on $b$ such that $\g\cap \G=\{a,b\}$. Then
506: $f(\g)$ is a closed path in $(X_O,p)$, for $O=A$ or $B$. If $f(\g)$ is nullhomotopic, then
507: after a homotopy (constant outside a regular neighbourhood of $\g$) we may assume that
508: $f(\g)=p$. A further homotopy can be effected to make $f(\g)=x_O$, resulting in
509: a new picture, as shown in Figure
510: \ref{bridge_move}. We call the transformation from the original to the
511: new picture a {\em bridge move}.
512: \begin{figure}
513: \psfrag{a}{$a$}
514: \psfrag{b}{$b$}
515: \psfrag{'g}{$\g$}
516: \begin{center}
517: {
518: \includegraphics[scale=0.4]{fig/bridge_move.eps}
519: }\\
520: \caption{A bridge move}\label{bridge_move}
521: \end{center}
522: \end{figure}
523:
524: If $u$ and $v$ are a pair of distinct vertices which cancel then we can perform
525: bridge moves on arcs incident $u$ and $v$, as shown in Figure \ref{bridge_pair},
526: \begin{figure}
527: \psfrag{a}{$a$}
528: \psfrag{b}{$b$}
529: \psfrag{u}{$u$}
530: \psfrag{v}{$v$}
531: \psfrag{a^-1}{$a^{-1}$}
532: \psfrag{b^-1}{$b^{-1}$}
533: \begin{center}
534: {
535: \includegraphics[scale=0.4]{fig/bridge_pair.eps}
536: }\\
537: \caption{Bridge moves on a pair of cancelling vertices}\label{bridge_pair}
538: \end{center}
539: \end{figure}
540: to increase the number of arcs joining $u$ to $v$. Repeating such
541: bridge moves we obtain a picture in which all arcs incident to $u$ are incident to $v$,
542: in which case we call $u$, $v$ and their incident arcs a {\em floating dipole}: see Figure
543: \ref{floating_dipole}.
544: \begin{figure}
545: \psfrag{a}{$a$}
546: \psfrag{b}{$b$}
547: \psfrag{u}{$u$}
548: \psfrag{v}{$v$}
549: \psfrag{a^-1}{$a^{-1}$}
550: \psfrag{b^-1}{$b^{-1}$}
551: \begin{center}
552: {
553: \includegraphics[scale=0.4]{fig/floating_dipole.eps}
554: }\\
555: \caption{A floating dipole}\label{floating_dipole}
556: \end{center}
557: \end{figure}
558: An arc $c$ of $\G$ such that $c$ is a simple closed curve bounding a
559: disk $D$ with int$(D)\cap\G=\nul$ is called a {\em floating arc}.
560:
561: If $C$ is a simple closed arc in $\S\bl \G$ and one component of $\S\bl C$ is
562: a disk $C_D$ such that $\G\cap C_D$ is a floating dipole or floating
563: arc then $f(C)$ is null--homotopic in $Y$. Performing a homotopy, constant outside
564: $C_D\cup C$, we may replace $f$ by a map such that $f(C_D)=x_O$, where $O=A$ or $B$.
565: Therefore we may form a
566: new picture on $\S$ by removing all arcs and vertices of $C_D\cap\G$ from $\G$.
567: We call this operation {\em removal} of a floating dipole or arc.
568: We consider the equivalence relation $\sim$ generated by the relations
569: (\ref{pic_equiv1})--(\ref{pic_equiv3}) below
570: on the set of pictures on
571: a surface $\S$,
572: %on compact surfaces,
573: with $n$ boundary components
574: with fixed boundary labels $u_1,\ldots ,u_n\in H$.
575: Suppose $\G$ and $\G^\prime $ are such pictures on $\S$ over $G$. Then
576: $\G \sim \G^\prime$ if $\G^\prime$ is obtained from $\G$ by
577: \be
578: \item\label{pic_equiv1} an isotopy of $\S$,
579: \item\label{pic_equiv2} a bridge move or
580: \item\label{pic_equiv3} removal of a floating dipole or floating arc.
581: \ee
582:
583: \begin{defn}\label{pic_effic}
584: A picture is {\em efficient} if it has fewest arcs in its equivalence class.
585: \end{defn}
586: It
587: follows that an efficient picture is reduced, since otherwise we can perform
588: bridge moves on arcs incident to a pair of vertices which cancel until a floating
589: dipole is produced. Removal of the floating dipole results in a picture with
590: fewer arcs, contrary to efficiency of the original picture.
591:
592: Let $g$ and $n$ be fixed integers and $u_1,\ldots ,u_n$ fixed elements
593: of $H$. Consider the set $T$ of all pictures
594: over $G$ on surfaces
595: with $n$ boundary components and genus at most $g$,
596: with boundary labels $u_1,\ldots ,u_n$.
597: \begin{defn}\label{pic_min}
598: A
599: picture in $T$ is said to
600: be {\em minimal} if no other picture in $T$ has fewer arcs.
601: \end{defn}
602: Minimal pictures are efficient and efficient pictures are reduced.
603: In general it is not the case that reduced pictures
604: are efficient or that efficient pictures are minimal.% {\bf but in case ...}
605: %Let $A$, $B$ and $s$ be defined as at the begining of Section
606: %\ref{pictures}.
607: \begin{prop}[\cite{\DHa}]\label{no_closed_arc}
608: Let $\G$ be a minimal picture over $G$ on a %compact
609: %connected
610: surface $\S$. Then no arc of $\G$ is closed curve.
611: \end{prop}
612: \medskip
613:
614: \noindent
615: {\em Proof.}
616: Suppose that $c$ is an arc of $\G$ and a closed curve. The closure
617: of $\S\backslash c$ has $2$ boundary components $c_1$ and $c_2$ corresponding to $c$.
618: Form a new surface $\S_1$ by capping $c_i$ with a disk, for $i=1,2$. Then $\G\backslash c$
619: is a picture on $\S_1$, since $f(c)= p\in Z$. The genus of $\S_1$ is no more than that of $\S$
620: and $\G\backslash c$ has the same boundary labels as $\G$, but fewer arcs.
621: This contradicts the minimality of
622: $\G$.
623: \medskip
624:
625: \begin{prop}[\cite{\DHa}]\label{genreg}
626: Let $\G$ be a minimal picture over $G$ on a compact
627: connected surface $\S$. Then
628: $\chi(\D)\ge\chi(\S)$, for all regions $\D$ of $\G$.
629: \end{prop}
630: \medskip
631:
632: \noindent
633: {\em Proof.} Suppose that $\D$ is a region such that
634: $\chi(\D)<\chi(\S)$. Let $\S_0=\S\bl\textrm{int}(\D)$ and let
635: $\Omega_0,\ldots,\Omega_k$ be the connected components of $\S_0$. Then
636: $\chi(\D)<\chi(\S)=\chi(\S_0)+\chi(\D)=\sum_{i=0}^k\chi(\Omega_i)+\chi(\D).$
637: Therefore $\chi(\Omega_i)>0$, so $\Omega_i$ is a disk or a sphere, for some $i$.
638: As $\S$ is connected this means that $\Omega_i$ is a disk with
639: $\Omega_i\cap \D=C$, where $C$ is homeomorphic to $S^1$, for such $i$.
640: Let $\Omega^\prime$ be a regular neighbourhood of $\Omega_i$
641: and let $\G^\prime=\G\cap \Omega^\prime$. Then $\Omega^\prime$ is a
642: disk and the boundary of $\Omega^\prime$ lies in $\D$. Hence the
643: boundary label of $\G^\prime$ is an element $u$ of $A\dcup B$ such
644: that $u=1$ in $G$. As the Freiheitssatz holds for $G$
645: (\cite{Howie89}) it follows that $u=1$ in $A$ or $B$, as appropriate. Thus we may
646: replace $\G^\prime$ with the empty picture on $\Omega^\prime$ without
647: altering $\S$ or the boundary labels of $\G$. This contradicts the
648: minimality of $\G$.
649: %%%%%%%%%%%%%%5
650: \subsection{Parallel arcs and the graph of a picture}
651: Given two arcs $a_1$ and $a_2$ of a picture $\G$ such that either
652: \be
653: \item $a_1$ and $a_2$ bound an annulus $S$ with int$(S)\cap \G=\nul$ or
654: \item there are corners $c_1$ and
655: $c_2$ of $\G$ such that $a_1$, $a_2$, $c_1$ and $c_2$ bound a disk
656: $\D$ with int$(\D)\cap \G=\nul$,
657: \ee
658: then $a_1$ and $a_2$ are said to be {\em parallel} arcs.
659: There is an equivalence relation on the set of arcs of $\G$ generated
660: by the parallel relation on arcs. The equivalence class of an arc $a$ is
661: called an {\em edge} of $\G$, denoted $\bar{a}$. If an edge consists of
662: $d$ parallel arcs we say that it has {\em width} $d$. We form the {\em
663: graph} $\bar\G$
664: of $\G$
665: on $\S$ as follows. Collapse each vertex of $G$ to a point to give a
666: vertex of $G$.
667: Collapse each class of arcs $\bar a$ onto $a$: the resulting arcs are
668: the edges of $\bar \G$, except for any that are closed loops (meeting
669: no vertex of $\G$) and these are discarded.
670: In addition there is one vertex on $\pd\S$ wherever an edge of
671: $\bar\G$ meets $\pd\S$. If $e^+$ is one end of an edge $e$ of $\bar \G$ then
672: the end points of arcs of $\G$ which are identified to form $e^+$ comprise an
673: {\em end} of the class of arcs corresponding to $e$. If $\bar a$ is a class of arcs with at least
674: one end on the boundary then we say that $\bar a$ is a {\em boundary} class. Otherwise
675: $\bar a$ is an {\em interior} class.
676: An arc of $\G$ with both ends on $\pd\S$ is called an {\em arc of type} $I$.
677: An edge of $\bar\G$ which is the image of a class of arcs of type $I$
678: is called an {\em edge of type} $I$.
679: We define the {\em degree} of a vertex $v$ of $\G$ to be the degree of the
680: corresponding vertex in the graph $\bar\G$, that is
681: $\deg(v)=$ number of classes of arc incident to $v$.
682:
683: If $a$ is an arc of an edge $e$ and $a$ is incident to vertices $u$ and $v$ then
684: $u$ and $v$ are $a$--confluent if and only if they are $b$--confluent for all arcs
685: $b$ of $e$. In this case we say that $u$ and $v$ are $e${\em --confluent}.
686:
687: The following facts are established in \cite{\Ha}. A word $x_1\cdots x_n$ is said to have
688: {\em period} $\lm$ if $x_{i+\lm}=x_i$, for all $i$ such that $1\le i\le n-\lm$.
689: \begin{lemma}\label{bound_edge_width}
690: Assume $G$ is not of type $E(2,*,m)$.
691: Let vertices $u_1$ and $u_2$ of a picture $\G$ over $G$ be joined by an edge $e$ of
692: width $k$. Assume either that $\G$ is reduced and $u_1$ and $u_2$ are distinct or that
693: $\G$ is minimal. Then $k\le l-2$.
694: %Moreover, if $k>l-2$ then $u_1$ and $u_2$ are not $e$--confluent
695: %and $G$ is of type $E(2,*,m)$.
696: \end{lemma}
697: \medskip
698:
699: \noindent{\it Proof.} The ends of $e$ determine subwords $s_1$ and $s_2$ of the boundary labels
700: $w_1$ and $w_2$ of $u_1$ and $u_2$, respectively, with $l(s_1)=l(s_2)=k-1$. Suppose first that
701: $u_1$ and $u_2$ are $e$--confluent. Without loss of generality we may assume that $\e_1=\de=1$ and
702: $\e_2=-1$ (with the notation of Definition \ref{e_confluent}).
703: Then $s_1$ is a cyclic subword of $r^m$, $s_2$ is a cyclic subword of $r^{-m}$ and $s_2^{-1}=s_1$
704: (see Figure \ref{confluent_edge}).
705: \begin{figure}
706: \psfrag{rm}{$r^m$}
707: \psfrag{r-m}{$r^{-m}$}
708: \psfrag{s1}{$s_1$}
709: \psfrag{s2}{$s_2$}
710: %\psfrag{a^-1}{$a^{-1}$}
711: %\psfrag{b^-1}{$b^{-1}$}
712: \begin{center}
713: {
714: \includegraphics[scale=0.4]{fig/confluent_edge.eps}
715: }\\
716: \caption{}\label{confluent_edge}
717: \end{center}
718: \end{figure}
719: Therefore $r^m$ has identical subwords $s_1$ and $s_2^{-1}$. We may assume that $s_1$ is an initial
720: subword of $r^m$. Then $r^m$ begins $r_0s_2^{-1}$, where $r_0$ is an initial subword of $r$ of length
721: $l(r_0)=\lm\ge 0$. As $r$ is cyclically reduced in $A*B$ it follows that $\lm$ is even.
722: We may also assume that $\lm\le l/2$ (otherwise reverse the roles of $s_1$ and $s_2$).
723: If $k\ge l-1$ this implies that the union of $s_1$ and $s_2$ is an initial subword of $r^m$ of period
724: $\lm$. If $\lm>0$ then $\lm\ge 2$ so $l(s_2)+\lm\ge l$ and
725: then it follows that $r^m$ has periods $\lm$ and $l$. Hence, from
726: \cite[Section 3, Proposition 1]{\Ha}, $r^m$ and also $r$ have period $d=\gcd(\lm,l)$. This cannot occur
727: as $r$ is not a proper power. Therefore $\lm=0$ and $s_1$ and $s_2^{-1}$ coincide as subwords of
728: $r^m$. In this case, if $u_1\neq u_2$ then $u_1$ and $u_2$ cancel.
729: % and $\G$ cannot be reduced.
730: %we may perform bridge moves on arcs incident to $u_1$ and $u_2$ to
731: %form a floating dipole (see Figure \ref{form_floating_dipole}).
732: As $\G$ is reduced this is a contradiction.
733: \begin{comment}
734: \begin{figure}
735: \psfrag{s1}{$s_1$}
736: \psfrag{s2}{$s_2$}
737: \begin{center}
738: {
739: \includegraphics[scale=0.3]{fig/form_floating_dipole.eps}
740: }\\
741: \caption{}\label{form_floating_dipole}
742: \end{center}
743: \end{figure}
744: \end{comment}
745:
746: On the other hand, if $\lm=0$, $u_1=u_2$ and $u_1$ is $e$--confluent to itself
747: then a regular neighbourhood $N$ of $u_1\cup e$ is homeomorphic to a M\"obius band. In this case
748: we can perform bridge moves on arcs incident to $u_1$ and outside $e$ to form a class containing
749: all arcs incident to $u_1$. We illustrate this diagrammatically: suppose $N$ is as shown in
750: Figure \ref{mobius_band}.
751: \begin{figure}
752: \psfrag{a}{$a$}
753: \psfrag{b}{$b$}
754: \psfrag{c}{$c$}
755: \psfrag{s1}{$s_1$}
756: \psfrag{s2}{$s_2$}
757: \psfrag{N}{$N$}
758: \begin{center}
759: {
760: \includegraphics[scale=0.4]{fig/mobius_band.eps}
761: }\\
762: \caption{$N$ after cutting along arc $ab$}\label{mobius_band}
763: \end{center}
764: \end{figure}
765: Cutting along $c$ and pasting together both copies of $a$ and $b$ we obtain the diagram
766: of $N$ shown in Figure \ref{cut_mobius_band}.
767: \begin{figure}
768: \psfrag{a}{$a$}
769: \psfrag{b}{$b$}
770: \psfrag{c}{$c$}
771: \psfrag{N}{$N$}
772: \begin{center}
773: {
774: \includegraphics[scale=0.4]{fig/cut_mobius_band.eps}
775: }\\
776: \caption{$N$ after cutting along $c$ and pasting $a$ and $b$}\label{cut_mobius_band}
777: \end{center}
778: \end{figure}
779: Clearly the number of arcs incident to the upper and lower halves of the vertex in this diagram
780: are equal and so we may perform bridge moves as claimed. The result is shown in Figure \ref{floating_projective}.
781: \begin{figure}
782: \psfrag{a}{$a$}
783: \psfrag{b}{$b$}
784: \psfrag{c}{$c$}
785: \psfrag{N}{$N$}
786: \begin{center}
787: {
788: \includegraphics[scale=0.4]{fig/floating_projective.eps}
789: }\\
790: \caption{$N$ after performing bridge moves}\label{floating_projective}
791: \end{center}
792: \end{figure}
793: Since the dotted boundary of this diagram has trivial label we may remove $u_1$ and all its incident
794: arcs from $N$, leaving a picture on $\S$ with the same boundary labels as $\G$ but fewer arcs. This contradicts
795: minimality of $\G$.
796: Thus if $u_1$ and $u_2$ are $e$--confluent then
797: %$\lm>0$ and so $l(s_2)+\lm<l$.
798: % and so, in this case,
799: %$l(s_2)\le l-3$ and
800: $k\le l-2$.
801:
802: Now suppose that $u_1$ and $u_2$ are not $e$--confluent and that $k\ge l-1$. Without loss of
803: generality we may assume that $\e_1=\e_2=\de$ and that $s_1$ is an initial subword of $r^m$.
804: Then $s_2$ is a cyclic subword of $r^m$ and $s_1$ is identically equal to $s_2^{-1}$, as a word in $A*B$.
805: Again we may assume that $r^m$ begins $r_0 s_2=U$, with $l(r_0)=\lm\le l/2$. If $s_1$ has even length
806: then $s_1$ and $s_2$ %, which is identically equal to $s^{-1}_1$,
807: begin with letters of different groups,
808: so $\lm$ must be odd. Similarly if $l(s_1)$ is odd then $\lm$ must be even. As $k\ge l-1$ we have
809: $l(s_i)\ge l-2$ and $l(U)=l(s_2)+\lm\ge l-1$. We may write $s_1=r_0t$, where $s_2=tr_1$, since
810: $\lm\le l-2\le l(s_i)$. Hence $s_1=s_2^{-1}$ implies $t=t^{-1}$ and $r_0=r_1^{-1}$.
811: Hence $U=r_0s_2=r_0tr_0^{-1}$, where $t=t^{-1}$, so
812: $t$ is of odd length and its
813: middle letter has order $2$. Therefore $U=VyV^{-1}$, for some $V$. If $l(U)=l-1$ then
814: $r=VyV^{-1}x$ and $G$ is of type $E(2,*,m)$, a contradiction. Suppose $l(U)\ge l$. We have $r^m=VyV^{-1}W$ and
815: we may write $V=V_0xU_0$, where $x\in A\dcup B$, $l(U_0yU_0^{-1})=l-1$ and $l(V_0)\ge 0$.
816: Hence $r^m$ has a cyclic permutation $xU_0yU_0^{-1}x^{-1}V_0^{-1}WV_0$ which must have period $l$.
817: Again a contradiction arises as $x=x^{-1}$ and
818: $G$ is of type $E(2,2,m)$.
819: \begin{comment}
820: Furthermore we have the configuration of Figure \ref{not_confluent_edge}.
821: \begin{figure}
822: \psfrag{a}{$a$}
823: \psfrag{a-1}{$a^{-1}$}
824: \psfrag{e}{$e$}
825: \psfrag{s1}{$s_1$}
826: \psfrag{s2}{$s_2$}
827: \psfrag{u1}{$u_1$}
828: \psfrag{u2}{$u_2$}
829: \psfrag{rm}{$r^{m}$}
830: \begin{center}
831: {
832: \includegraphics[scale=0.4]{fig/not_confluent_edge.eps}
833: }\\
834: \caption{}\label{not_confluent_edge}
835: \end{center}
836: \end{figure}
837: %If $u_1\neq u_2$ then
838:
839: Bridge moves may be performed on arcs incident to $u_1$ and $u_2$ to form a
840: class containing all arcs incident to $u_1$ and $u_2$. If $u_1\neq u_2$ then
841: this results in a
842: floating dipole, contrary to efficiency of $\G$. On the other hand if $u_1=u_2$ then a regular
843: neighbourhood $N$ of $u_1\cup e$ is homeomorphic to an annulus containing $u_1$ and $e$
844: (see Figure \ref{annulus_edge}).
845: \begin{figure}
846: \psfrag{x}{$x$}
847: \psfrag{x-1}{$x^{-1}$}
848: \begin{center}
849: {
850: \includegraphics[scale=0.4]{fig/annulus_edge.eps}
851: }\\
852: \caption{}\label{annulus_edge}
853: \end{center}
854: \end{figure}
855: Furthermore it follows from arguments similar to those above that the upper and lower corners of
856: $u_1$ in Figure \ref{annulus_edge} have mutually inverse labels, $x$ and $x^{-1}$, say, with
857: $x\in A\dcup B$. Since both boundary components of $N$ (with orientations shown)
858: are mapped by $f$ to a path representing $x$, we may remove
859: $u_1$ and $e$ from $\G$ leaving a picture with the same boundary labels as $\G$ but fewer arcs, contrary
860: to minimality of $\G$.
861:
862: Therefore $l(U)\le l-1$ and it follows that either $l(s_i)=l-2$ and $\lm=1$ or $l(s_i)=l-1$ and $\lm=0$.
863: In both cases $k\le l$ and $G$ is of type $E(2,*,m)$.
864: \end{comment}
865: %%%%%%%%%%
866: \begin{defn}\label{minimalistic}
867: A picture $\G$ on a surface $\S$ is called {\em minimalistic} if the following conditions hold.
868: \be
869: \item No arc of $\G$ is a closed curve.
870: \item $\chi(\D)\ge \chi(\S_\D)$, whenever $\D$ is a region of $\G$ and $\S_\D$ is the connected
871: component of $\S$ containing $\D$.
872: \item No interior edge of $\G$ has width more than $l$. If $G$ is not of type $E(2,*,m)$ then no
873: interior edge has width more than $l-2$.
874: \ee
875: \end{defn}
876: From Proposition \ref{no_closed_arc}, Proposition \ref{genreg}
877: and Lemma \ref{bound_edge_width}, minimal pictures are minimalistic. The converse
878: does not hold in general.
879: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
880: \subsection{Routes and distance}
881: If $u$ and $v$ are vertices incident to a region $\D$ satisfying
882: $\rho(\D)=\bt(\D)=2$, $\chi(\D)=1$, $t(\D)=4$ and $\e(\D)=0$, as shown in
883: Figure \ref{PV}, then $u$ and $v$ are said to be {\em paired}. If $e$
884: is a class of arcs having both end points on $\pd \S$ and $e\cap \pd
885: \D \neq \nul$,
886: where $\D$ is a region satisfying $\rho(\D)=1$, $\bt(\D)=2$, $\chi(\D)=1$, $t(\D)=3$
887: and $\e(\D)=0$, as shown in
888: Figure \ref{PE}, then $u$ and $e$ are said to be {\em paired}.
889: \begin{figure}
890: \psfrag{v}{$v$}
891: \psfrag{u}{$u$}
892: \psfrag{e}{$e$}
893: \psfrag{'D}{$\D$}
894: \begin{center}
895: \mbox{
896: \subfigure[paired vertices $u$ and $v$\label{PV}]
897: {\includegraphics[scale= 0.6,clip]{fig/PV.eps} }\qquad
898: \subfigure[vertex $u$ paired to class of arcs $e$\label{PE}]
899: { \includegraphics[scale=0.6,clip]{fig/PE.eps} }
900: }
901: \caption{}\label{paired}
902: \end{center}
903: \end{figure}
904:
905: A {\em route} in $\G$, from $x_1$ to $x_n$,
906: is a sequence $x_1,\ldots,x_n$, where $x_i$ is a vertex or
907: a class of arcs such that the set $\{x_i,x_{i+1}\}$ consists either of
908: \be
909: \item two vertices which are joined by an arc or
910: \item a vertex and a boundary class of arcs paired to that vertex or
911: \item two vertices which are paired,
912: %or
913: %\item a vertex and a class of arcs which are paired, {\bf is this required/not?}
914: \ee
915: for $i=1,\ldots ,n$.
916: %Furthermore if $x_1$ or $x_n$ are classes of arcs then they must be boundary classes.
917: Let $l_1=\{i: x_i\textrm{ is a boundary class of arcs}\}$ and
918: $l_2=\{i:x_i\textrm{ and } x_{i+1}\textrm{ are both vertices}\}$. The {\em length} of the
919: route is then $|l_1|+|l_2|$. A {\em route} in $\bar\G$ is defined to be the
920: image of a route in $\G$ under the quotient map $\G\maps \bar\G$.
921: % so
922: %it follows that routes in $\G$ correspond to routes in $\bar\G$.{\bf ????????????}
923: A route is {\em simple} if $x_i\neq x_j$, for
924: all $i\neq j$.
925:
926: If $S$ is a subset of $\cV$ then a vertex $v$ is said to be {\em distance}
927: $d$ from $S$ if the length of a shortest route from $v$ to a vertex
928: of $S$ is $d$. If there is no route from
929: $v$ to a vertex of $S$ then the distance of $v$ from $S$ is defined to
930: be $\infty$. If $e$ is a class of arcs that is paired to a vertex then
931: the {\em distance} of $e$ from $S$ is $d+1$, where $d$ is the minimum of
932: the distances from $S$ of the vertices paired to $e$.
933: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
934: \subsection{Pictures and equations}
935:
936: The correspondence between solutions of quadratic equations and pictures is given
937: in the next three results.
938: \begin{theorem}\label{fundpic}
939: Let $u_1,\ldots, u_k$ be elements of $H$ and let $t,p$ be non--negative
940: integers. Then there exist elements $w_1,\ldots , w_{k+2t+p}\in H$ such that
941: \begin{equation}\label{pic_eqn}
942: w=\prod_{i=1}^k w_i^{-1}u_iw_i
943: \prod_{i=1}^t [w_{i+k},w_{i+k+t}]\prod_{i=1}^p w_{i+k+2t}^2=1 \in G
944: \end{equation}
945: if and only if there exists a picture $\G$ over $G$ on a surface $\S$,
946: %homeomorphic
947: %to the connected sum of
948: %a $k$--punctured sphere with
949: %at most
950: %$t$ torii and $p$ projective
951: %planes with $k$ punctures,
952: of genus $t+p/2$, orientable if $p=0$,
953: with $k$ boundary components which have
954: labels $u_1,\ldots ,u_k$ (with respect to some
955: orientation of $\pd \S$).
956: \end{theorem}
957: \medskip
958:
959: \noindent
960: {\em Proof.}
961: In the notation of Section \ref{qwords}, let $F=F(D\dcup X)$ and
962: $q=q(0,k,t,p)$.
963: Then $\S(q)$ is the connected sum of a $k$--punctured sphere with
964: $t$ torii and $p$ projective planes, and so has genus $t+p/2$ and is orientable
965: if $p=0$, and has boundary components
966: $\bt_1,\ldots, \bt_k$ with labels $d_1,\ldots, d_k$, respectively.
967:
968: Suppose a picture $(f,\G$) with orientation $\z$
969: exists on a surface $\S$ of genus $t+p/2$, orientable if $p=0$, with boundary
970: components $\wt\bt_1,\ldots ,\wt{\bt_k}$ having labels $u_1,\ldots ,u_k$.
971: %Let $f:\S\maps Z$ be the corresponding map.
972: %Without loss of
973: %generality we may assume that the number of torii and projective planes in the
974: %connected sum $\S$ are $t$ and $p$, respectively.
975: Assume first that $\S$ is connected.
976: Then $\S$ is homeomorphic to $\S(q)$ and we may choose this
977: homeomorphism $\theta :\S(q)\maps \S$ so
978: that the boundary component $\bt_i$ is mapped to
979: $\wt\bt_i$, so that the base point of $\bt_i$ is mapped to a base point of $\wt{\bt_i}$ and so that
980: the orientation of $\wt{\bt_i}$ is the same as that induced by $\theta(\bt_i)$.
981: By construction $\S(q)$ is a quotient of the disk $D$ with boundary
982: $\pd D$ labelled $q$, so $f\circ \theta$ can be regarded as a map from
983: $D$ to $Z$. Then $f\circ\theta|_{\pd D}$ maps each labelled interval
984: of $\pd D$ to a path in $Y$ representing an element of $H$.
985: Let $\phi$ be the map from $F$ to $H$ induced by
986: $f\circ \theta|_{\pd D}$ on the labelled intervals of $\pd D$.
987: Then $\phi$ maps the boundary label $q$ of $D$
988: to an element of $H$ representing the identity of $G$.
989: Furthermore $\phi(d_i)=u_i$, for $i=1,\ldots, k$ and
990: $\phi(x_i)=w_i\in H$, for
991: $i=1,\ldots,k+2t+p$, and it follows that $w=1$ in $G$, as required.
992:
993: Suppose now that $\S$ has connected components
994: $\S_1,\ldots, \S_n$, for some $n\ge 2$. We may assume that $\S_j$ is
995: the connected sum of a $k_j$--punctured sphere with $t_j$ torii and
996: $p_j$ projective planes where $k_j>0$, $\sum_{j=1}^n k_j=k$,
997: $\sum_{j=1}^n t_j=t$ and $\sum_{j=1}^n p_j=p$.
998: Let the boundary labels of $\G_i=\S_i\cap \G$ be $u_{i_1},\ldots ,u_{i_{k_j}}$,
999: where $i_1<\cdots <i_{k_j}$. The argument above implies the existence
1000: of elements $w_{ij}$, $x_{ij}$, $y_{ij}$, $z_{ij}\in H$ such that
1001: \[
1002: \prod_{i=1}^{k_j} w_{ij}^{-1}u_{i_j}w_{ij}
1003: \prod_{i=1}^{t_j} [x_{ij},y_{ij}]\prod_{i=1}^{p_j} z_{ij}^2=1 \in G,
1004: \]
1005: for $j=1,\ldots ,n$.
1006: The equality (\ref{pic_eqn}) follows after some rearrangement of these equations.
1007:
1008: Conversely, suppose that there exist $w_i$ in $H$ satisfying
1009: (\ref{pic_eqn}). Let $F$, $q$ and $\S(q)$ be defined as above, where
1010: $\S(q)$ is a quotient of a disk $D$ with boundary label $q$. Let $\hat d_i$ and
1011: $\hat x_i$ be
1012: the oriented sub--intervals of $\pd D$ labelled $d_i$ and $x_i$, respectively.
1013: For $g\in H$ let $\wt g$ be a closed path in $(Y,p)$ representing $g$. Define a map
1014: $f_\pd:\pd D\maps Z$ by $f_\pd(\hat x_i)= \wt w_i$ and $f_\pd(\hat d_i)=\wt u_i$.
1015: As $w=1$ in $G$ the map $f_\pd$ maps $\pd D$ to
1016: a path which is nullhomotopic in $Z$ and so extends to a map
1017: $f:D\maps Z$. As $f_\pd$ is identical on any two intervals of $\pd D$ which are identified
1018: in forming $\S(q)$, $f$ factors through $\S(q)$ and we may
1019: regard $f$ as a map from $\S(q)$ to $Z$. We may make
1020: $f$ transverse to the mid--point $c$ of $\D_s$ with a homotopy which fixes
1021: $\pd\S(q)$. A further homotopy ensures that each component of the closure
1022: of $f^{-1}(\inr(\D_s))$ is a regular neighbourhood of $f^{-1}(c)$. Let $\S_0=\S(q)\bl
1023: f^{-1}(\inr(\D_s))$. Then $f$ restricted to $\pd\S_0$ is transverse to $p$ and so
1024: there is a homotopy of $f$, relative to $\pd\S_0$, to a map which is transverse to
1025: $p$ on $\S_0$. Therefore we may assume that $f$ gives rise to a sketch $\G$
1026: on $\S(q)$.
1027: Extend the orientation of the $\hat{d_i}$'s to an orientation of regions and vertices of
1028: $\S(q)$. (This can be done consistently if $\S(q)$ is orientable.) Let $f^l(d_i)=u_i$ and
1029: define $f^l$ on vertex corners as in the definition of a picture. Then $(f,\G)$ is a picture
1030: on $\S$ over $G$ with boundary labels $u_1,\ldots ,u_k$, as required.
1031: %%%%%%%%%%%%%%
1032: % over $G$ with boundary labels $u_1,\ldots ,u_k$ (under an appropriate
1033: %orientation). Let $\D$ be a region of $\G$. Then
1034: %$\inr(\D) \cap f^{-1}(p)=\nul$ so $f(\D)\subset X_O$, for $O=A$ or $B$. It
1035: %follows that $\D$ is an $O$--region. The argument of the first half of the proof
1036: %shows that condition $P$\ref{pic_cond2} of the definition of picture holds, for
1037: %all regions $\D$ of $\G$. Finally, if $a$ is an arc of $\G$ and $x\in a$ then, as
1038: %$f$ is transverse to $p$ and $f^{-1}(e_A\cup e_B)$ is a regular neighbourhood of
1039: %$f^{-1}(p)$ it follows that $x$ meets an $A$--region and a $B$--region. Hence
1040: %$\G$ is a picture, as required.
1041: %%%%%%%%%%%%%%%%%%
1042: \begin{comment}
1043: \begin{corol}\label{pic_qpe_con}
1044: Let $u_1,\ldots, u_k$ be elements of $H$, $t,p$ be non--negative
1045: integers, $F=F(D\dcup X)$, $q=q(0,k,t,p)$ and define $\bt(d_i)=u_i$,
1046: for $i=1,\ldots, k$.
1047: Then there exists a solution $\phi$ to the equation $(q=1,\bt)$ if and
1048: only if
1049: there exists a minimalistic picture $\G$ over $G$ on a surface $\S$,
1050: of genus $t+p/2$, orientable if $p=0$,
1051: with $k$ boundary components which have
1052: labels $u_1,\ldots ,u_k$ (with respect to some
1053: orientation of $\S$). Furthermore if the $\cQ((n^\prime),(t^\prime),(p^\prime))$--problem is
1054: solvable in $A$ and $B$, for all
1055: $n^\prime$, $t^\prime$, $p^\prime$ such that
1056: \begin{gather}
1057: t^\prime+p^\prime/2 \le t+p/2 \textrm{ and }\nonumber\\
1058: n^\prime \le n+2(t-t^\prime)+(p-p^\prime),\nonumber %\label{genus_cond}
1059: \end{gather}
1060: then, given such a minimalistic picture $\G$ a
1061: solution to $(q=1,\bt)$ may be effectively constructed.
1062: \end{corol}
1063: \medskip
1064:
1065: \noindent{\em Proof.} That a solution to $(q=1,\bt)$ exists if and only if
1066: a picture exists, on a surface of the given genus, with boundary components labelled as described,
1067: follows directly from the theorem. If such a picture exists then there is a minimal picture satisfying
1068: the same conditions. To prove the final
1069: statement of the Corollary suppose that
1070: the hypotheses on $A$ and $B$ hold and that we're given
1071: such a minimalistic picture $\G$ over $G$ on a surface $\S$. We assume that $\S$ is connected, the
1072: argument in the general case following as in the proof of the Theorem. We use the notation
1073: of the proof of the theorem. The $w_i$'s are labels of the arcs $\theta(x_i)$ and we shall
1074: show how to construct these labels. First note that $\th$ may be chosen so that $\th(\pd D)$
1075: is transverse to $\pd \S$ and to arcs of $\G$ and meets no vertex of $\G$. Let $\D$ be a region
1076: of $\G$ which meets $\th(\pd D)$. Let $\D$ be the connected sum of $t_\D$ torii and $p_\D$ projective
1077: planes with $k_\D$ boundary components. These
1078: boundary components are unions of vertex and boundary corners of $\G$ and
1079: arcs of $\G$. We have $2t_\D+p_\D\le 2t+p$ and, since $\G$ is minimalistic,
1080: $\chi(\D)\ge \chi(\S)$. Therefore $k_\D\le k_\D+2(t-t_\D)+(p-p_\D)$ and it follows that the
1081: $\cQ(k_\D,t_\D,p_\D)$--problem is solvable in $A$ and $B$.
1082:
1083: Choose a base point on each boundary component of $\D$. On boundary components which are also boundary
1084: components of $\S$ the base point chosen is a point of the boundary partition. On all other boundary
1085: components the base point chosen is in the interior of an arc of $\G$. Choose a point $*$ in the interior
1086: of $\D$. Let $q_\D=q(0,k_\D,t_\D,p_\D)$. Then as in the proof of the theorem there is a homeomorphism $\th_\D$ from
1087: $\S(q_\D)$ to $\D$. The surface $\S(q_\D)$ is a quotient of a disk $D_\D$ with boundary labelled $q_\D$.
1088: Regarding $\th_\D$ as a map from $D_\D$ to $\D$ we may choose $\th_\D$ so that the end points of intervals
1089: labelled $d_i$ (boundary components of $\S(q_\D)$) map to base points of the images $\th_\D(d_i)$ and $\th_\D(x_i)$
1090: is a path from $*$ to the base point of $\th_\D(d_i)$, for $i=1,\ldots, k_\D$. We may in addition choose $\th_\D$
1091: so that $\th_\D(x_i)$ is a closed loop based at $*$, for $i>k_\D$. Finally we may choose $\th_\D$ so that
1092: $\th_\D(\pd D_\D)$ is transverse to $\th(\pd D)\cap \D$. As the $\cQ((k_\D),(t_\D),(p_\D))$--problem
1093: is solvable in $A$ and $B$ we may find a solution $\phi_\D$ to $(q_\D=1,\bt)$, where $\bt_\D(d_i)=\th_\D(d_i)$.
1094: We now label the oriented arc $\th_\D(x_i)$ with $v_i=\phi_\D(x_i)$. We also label the boundary of $D_\D$
1095: assigning $u_i$ to the interval $d_i$ and $v_i$ to the intervals $x_i$. Suppose $\D$ is an $O$--region, where
1096: $O=A$ or $B$. By construction there exists a map
1097: $f_\D$ from $\D$ to $X_O$ such that $f_\D\circ \th_D=\phi_D$. Repeating this construction over all regions
1098: which meet $\th(\pd D)$ we obtain a picture $(f^\prime,\G)$, where
1099: \[f^\prime|_\D =\left\{\begin{array}{ll}
1100: f_\D, & \textrm{ if } \th(\pd D)\cap \D\neq \emptyset\\
1101: f|_\D, & \textrm{ if } \th(\pd D)\cap \D = \emptyset
1102: \end{array}\right.
1103: .\]
1104: The picture $(f^\prime,\G)$ has the same corner labels as $(f,\G)$ but the image of $\th(\pd D)$ under $f$ may
1105: not be the same as under $f^\prime$.
1106:
1107: Now let $\D$ be a region of $\G$ which meets $\th(\pd D)$ as before.
1108: Let $p$ be a connected component of $\th(x_i)\cap \D$. Then $\th^{-1}_\D(p)$ consists of a disjoint union of
1109: properly embedded $1$--submanifolds of $D_\D$. Let $p^\prime$ be a connected component of $\th^{-1}_\D(p)$
1110: and suppose that $p^\prime \pd D_\D=\{x_1,x_2\}$. Then we assign the label $g$ to the arc
1111: $\th_\D(p^\prime)$, oriented
1112: from $\th_D(x_1)$ to $\th_\D(x_2)$,
1113: where $g$ is the label of one of the components of $\pd D_\D\backslash \{x_1,x_2\}$, read
1114: from $x_1$ to $x_2$ (and where the shorter label is chosen if this makes sense). The label of the arc $p$ is
1115: formed by taking the product of the
1116: labels of the sub--arcs $\th_\D(p^\prime)$ for each connected component $p^\prime$
1117: of $\th^{-1}_\D(p)$ as they occur in turn along $p$. Similarly the arc $\th(x_i)$ is formed by taking the
1118: product of labels of components $\th(x_i)\cap \D$ over all regions through which $\th(x_i)$ passes.
1119: Since $(f^\prime ,\G)$ is a picture
1120: we have elements $w_i=\th(x_i)$, $i=1,\ldots ,k+2t+p$, giving (\ref{pic_eqn}). These may not
1121: be the same $w_i$'s as found in the Theorem but we can effectively construct these elements.
1122: \medskip
1123:
1124: \end{comment}
1125: %a surface constructed by taking a sphere,
1126: %at most $t$ torii and at most $p$ projective planes, forming
1127: %connected sums of some, all or none of these surfaces and then removing $n$ points.
1128: \begin{corol}\label{pic_qpe}
1129: %Let $H=A*B$ and $G=H/N(s)$ as in the theorem above.
1130: Let $(\mbf h,\mbf n,\mbf t,\mbf p)$
1131: be a positive 4--partition,
1132: $\cL$ a consistent system of parameters and $\mbf z$ an
1133: element of $(H^\Lm,\cL,\mbf n,\mbf h)$. Then
1134: the triple $(\mbf n,\mbf t,\mbf p)\in \cL$--genus$(\mbf z)$ if and only if
1135: there exists a solution $\al$ to $\cL$ and a minimalistic picture $\G$ on a surface
1136: of type $(\mbf n,\mbf t,\mbf p)$ with partitioned boundary labels list
1137: $(\hat\al(z_1),\ldots,\hat\al(z_n))$.
1138: \begin{comment}
1139: Furthermore, if the $\cQ(\mbf n^\prime,\mbf t^\prime,\mbf p^\prime)$--problem is
1140: solvable in $A$ and $B$, for all
1141: $(\mbf 0,\mbf n^\prime, \mbf t^\prime,\mbf p^\prime)\in \cN(\mbf h,\mbf n,\mbf t,\mbf p)$
1142: then, given such a solution $\al$ to $\cL$ and minimalistic picture $\G$ a
1143: solution to $Q(\mbf z,\cL,\mbf n,\mbf t,\mbf p)$ may be effectively constructed.
1144: \end{comment}
1145: \end{corol}
1146: \medskip
1147:
1148: \noindent
1149: {\em Proof.}
1150: Given a solution $(\phi,\al)$ to $Q=Q(\mbf z,\cL,\mbf n,\mbf t,\mbf p)$
1151: we have $\phi(q(\xi_j,n_j,t_j,p_j))=1$ in $G$, for $j=1,\ldots ,k$.
1152: It follows, from Theorem \ref{fundpic}, that there exists a picture
1153: $\G_j$ over $G$ on a surface $\S_j$ of genus $t_j+p_j/2$, with $n_j$ boundary
1154: components labelled $\phi(d_{\xi_j+1}),\ldots ,\phi(d_{\xi_j+n_j}).$
1155: If $\G_j$ exists then there also exists a minimalistic picture satisfying the same
1156: conditions, so we may assume that $\G_j$ is minimalistic.
1157: As $(\phi,\al)$
1158: is a solution to $Q$, $\phi(d_{\xi_j+i})=\hat\al\bt(d_{\xi_j+i})=\hat\al (z_{\mu_j+i})$,
1159: for all $i,j$. The picture $\G=\G_1\cup\cdots\cup\G_k$ on $\S_1\dcup\cdots\dcup\S_k$
1160: therefore satisfies the given requirements.
1161:
1162: Conversely, given a solution $\al$ to $\cL$ and a picture $\G$ on a surface
1163: $\S$ as described, suppose that $\S$ has connected components $\S_1,\ldots,\S_k$, where $\S_j$ has
1164: genus $t_j+p_j/2$ and $n_j$ boundary components labelled
1165: $\hat\al(z_{\mu_j+1}),\ldots,\hat\al(z_{\mu_j+n_j}).$
1166: Let $\G_j=\G\cap \S_j$ and $q_j=q(\xi_j,n_j,t_j,p_j)$ and
1167: define $\bt_j:F(L_D(q_j))\maps H$ by $\bt_j(d_{\xi_j+i})=\hat\al(z_{\mu_j+i})$,
1168: for $j=1,\ldots ,k$ and $i=1,\ldots, n_j$. Then, from Theorem \ref{fundpic},
1169: $\G_j$ gives rise to a solution $\phi_j$ to $(q_j=1,\bt_j)$. Define $\phi :
1170: F(L(\mbf q))\maps H$ by $\phi(x_i)=\phi_j(x_i)$, for
1171: $\xi_j+1\le i \le \xi_j+n_j+2t_j+p_j$ and $\phi(d_{\xi_j+i})=\hat\al(z_{\mu_j+i})$,
1172: for $1\le i\le n_j$. Then $(\phi,\al)$ is a solution to $Q$.
1173: \begin{comment}
1174: Under the additional hypotheses
1175: on $A$ and $B$ the solutions $\phi_j$ above may be effectively constructed and so therefore
1176: may $\phi$.
1177:
1178: {\bf the above results need some thought, as solutions may obviously
1179: be constructed once $\G$ and $\al$ are known to exist}
1180: \end{comment}
1181:
1182: \begin{defn}\label{pic_soln}
1183: Let $(\mbf h,\mbf n,\mbf t,\mbf p)$
1184: be a positive 4--partition,
1185: $\cL$ a consistent system of parameters and $\mbf z$ an
1186: element of $(H^\Lm,\cL,\mbf n,\mbf h)$.
1187: Let $\al$ be a solution to $\cL$ and $\G$ a picture on a surface
1188: of type $(\mbf n,\mbf t,\mbf p)$ with partitioned boundary labels list
1189: $(\hat\al(z_1),\ldots,\hat\al(z_n))$. Then we say that $(\al,\G)$ {\em
1190: corresponds} to the solution $(\phi,\al)$ of $Q(\mbf z,\cL,\mbf n,\mbf t,\mbf p)$
1191: constructed in Corollary \ref{pic_qpe}. The pair $(\al,\G)$ is
1192: called {\em exponent--minimal} if it corresponds to an
1193: exponent--minimal solution
1194: $(\phi,\al)$ to $Q(\mbf z,\cL,\mbf n,\mbf t,\mbf p)$.
1195: \end{defn}
1196:
1197: When there is no ambiguity we refer to $\G$, rather than $(\al,\G)$, as exponent--minimal.
1198: Observe that if an exponent--minimal picture with prime labels $\mbf z$ exists then
1199: a reduced (or efficient or minimal) exponent--minimal picture with the same prime
1200: labels also exists, because all pictures in an equivalence class and in $T$ have the same
1201: boundary labels.
1202: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1203: \section{Corridors}\label{corridor}
1204:
1205: Let $(\mbf h,\mbf n,\mbf t,\mbf p)$ be a positive 4--partition, $\cL$
1206: a consistent system of parameters, $\mbf z$ a special element of
1207: $(H^\Lm,\cL,\mbf n,\mbf h)$ and $\al$ a solution to $\cL$. From now
1208: until further notice we assume that $\G$ is a picture on a surface of
1209: type $(\mbf n,\mbf t,\mbf p)$ with prime
1210: labels $\mbf z$, labelled by $\hat\al(\mbf z)$ and with boundary partition
1211: $\mbf b=(b_1,\ldots ,b_{W_1})$.
1212:
1213: Let $D$ be the disk $[a,a+1]\+[0,1]\subset\RR^2$, where $a\ge 0$ is an
1214: integer, and let $x_0,y_0,x_1$
1215: and $y_1$ be non--negative integers with $x_i>0$, $i=1,2$. Define a
1216: set $D(x_0,x_1,y_0,y_1)$ of {\em distinguished} points of $D$ to be
1217: %disk $D$ together with
1218: \be
1219: \item $(a,i/(y_0+1))$, for $i=1,\ldots,y_0$;
1220: \item $(a+1,i/(y_1+1))$, for $i=1,\ldots,y_1$;
1221: \item $(a+(i/(x_0+1)),0)$, for $i=1,\ldots,x_0$ and
1222: \item $(a+(i/(x_1+1)),1)$, for $i=1,\ldots,x_1$. \ee A picture $\G_D$ over $G$
1223: on $D$ is said to have {\em type square--}$0$ (with {\em base}
1224: $D(1,1,0,0)$) if $\G_D$ has no vertices and one arc
1225: $\{1/2\}\+[0,1]$. A picture $\G_D$ on $D$ is said to have {\em type
1226: square--}$1$ with {\em base} $D(x_0,x_1,y_0,y_1)$ if $x_0,x_1,y_0$
1227: and $y_1$ are non--negative integers such that $x_0,x_1>0$,
1228: $x_0+x_1+y_0+y_1=ml$, $\G_D$ has one vertex $v$ and $\G_D$ has $ml$
1229: arcs each beginning at a point of $\pd v$ and ending at one
1230: of the distinguished points of $D(x_0,x_1,y_0,y_1)$. A picture of
1231: type square--$1$ with base $D(2,3,1,2)$ is shown in Figure
1232: \ref{square-1}.
1233: \begin{figure}
1234: \begin{center}
1235: { \includegraphics[scale=0.4]{fig/square-1.eps}
1236: }\\
1237: \caption{A picture of type square--$1$ with base $D(2,3,1,2)$}\label{square-1}
1238: \end{center}
1239: \end{figure}
1240:
1241: A picture $\G_D$ over $G$ on $D$ is said to have {\em type square--}$2$ with
1242: {\em base} $D(x_0,x_1,y_0,y_1)$ if the following conditions hold. \be
1243: \item $x_0,x_1,y_0$ and $y_1$ are positive integers such that
1244: $y_0,y_1\ge 3$ and $x_0+x_1+y_0+y_1\le 2ml-1$.
1245: \item $\G_D$ has $2$ vertices $v_0$ and $v_1$ and
1246: $ml+(x_0+x_1+y_0+y_1)/2$ arcs. $\G_D$ has one non--empty class of
1247: arcs joining $v_0$ to $v_1$ and $\G_D$ has $x_i$ arcs
1248: joining $v_i$ to distinguished points of $D$ on $[a,a+1]\+\{i\}$,
1249: for $i=1,2$.
1250: \item There are integers $r_i$, $i=0,1$ such that $0<r_i<y_i$ and
1251: $\G_D$ has $r_i$ arcs joining $v_i$ to distinguished points
1252: of $\{a+i\}\+[0,1]$ and $y_i-r_i$ arcs joining $v_{1-i}$ to
1253: distinguished points of $\{a+i\}\+[0,1]$.
1254: \ee
1255: A picture of type
1256: square--$2$ with base $D(4,5,7,6)$, $r_0=3$ and $r_1=2$ is shown in Figure
1257: \ref{square-2}.
1258: \begin{figure}
1259: \begin{center}
1260: { \includegraphics[scale=0.4]{fig/square-2.eps}
1261: }\\
1262: \caption{A picture of type square--$2$ with base $D(4,5,7,6)$}\label{square-2}
1263: \end{center}
1264: \end{figure}
1265: A picture of type square--$j$ is called a {\em
1266: square--picture} of type $j$, where $j=0,1$ or $2$. An arc $e$ of a square--picture
1267: meeting $[a,a+1]\+\{i\}$, for $i=0$ or $1$ is called a {\em lateral}
1268: arc of the picture. An arc $e$ of a square--picture of type 2 incident
1269: to both vertices is called an {\em internal--lateral} arc.
1270:
1271: A square--picture $(f_D,\G_D)$ comes equipped with a labelling
1272: function $f_D^l$, from corners of $\G_D$ to $A\dcup B$. Since we wish
1273: to map square--pictures into arbitrary pictures in such a way that the
1274: intervals $([a+(k/(x_i+1),a+((k+1)/(x_i+1)]\+\{i\})$ correspond to
1275: boundary corners of $\G$ we now refine this labelling function. We
1276: redefine the label of the corner $c_0=(\{a\}\+[0,1/(y_0+1)])\cup
1277: ([a,a+(1/(x_0+1))]\+\{0\})$ setting
1278: $f^l_D([a,a+(1/(x_0+1))]\+\{0\})=1$ and $f^l_D(\{a\}\+[0,1/(y_0+1)])=$
1279: the original label of $c_0$. Similarly, we redefine the labelling
1280: function on the corner $c_1=(\{a\}\+[y_0/(y_0+1),1])\cup
1281: ([a,a+(1/(x_1+1))]\+\{1\})$ so that the subinterval
1282: $[a,a+(1/(x_1+1))]\+\{1\}$ has trivial label and the defining property
1283: of boundary labels of regions of pictures is preserved (see
1284: Figure \ref{left_trivial}).
1285: \begin{figure}
1286: \psfrag{a}{$a$}
1287: \psfrag{1}{$1$}
1288: \psfrag{c0}{$c_0$}
1289: \psfrag{c1}{$c_1$}
1290: \begin{center}
1291: { \includegraphics[scale=0.4]{fig/left_trivial.eps}
1292: }\\
1293: \caption{}\label{left_trivial}
1294: \end{center}
1295: \end{figure}
1296: We say such a
1297: labelling function on $\G_D$ is {\em left--trivial}. If $\G_D$ is of
1298: type $0$ all labels are trivial so any labelling function is
1299: left--trivial. If $\G_D$ is not of type $0$ then, since a regular
1300: neighbourhood of an arc of $\G_D$ maps to a contractible subset of
1301: $Z$, we may achieve a left--trivial labelling by a homotopy of $f_D$
1302: fixing all of $D$ except for the components of $D\bl\G_D$ meeting
1303: $c_0$ and $c_1$. We shall henceforward assume that square--pictures
1304: have left trivial labelling functions.
1305:
1306: If $D$ is the disk $[a,b]\+[0,1]$, where $a,b\in \ZZ$, $0\le a<b$, we define $D_c$ to be
1307: the disk $[c,c+1]\+[0,1]$, for $c\in\ZZ$, $a\le c<b$. We define a {\em sum} of
1308: pictures of {\em type} 1 to be a picture $\G_D$ on the disk
1309: $D=[a,b]\+[0,1]$, where
1310: \be
1311: \item $b-a\ge 1$,
1312: \item\label{sum-square} $\G_D\cap D_c$ is a picture of type square--0
1313: or square--1, for $a\le c\le b-1$, and there is at least one $c$ in
1314: this range such that $\G_D\cap D_c$ is a picture of type square--1
1315: and
1316: \item if $\G_D\cap D_c$ has base $D_c[x_0,x_1,y_0,y_1]$ then $\G_D\cap
1317: D_{c+1}$ has base $D_{c+1}[x^\prime_0,x^\prime_1,y_1,y^\prime_1]$,
1318: for $a\le c\le b-2$.
1319: \ee
1320:
1321: For $j=0$ or $2$ we define a sum of pictures of {\em type} $j$
1322: similarly except that in condition (\ref{sum-square}) all pictures
1323: must be of type square--$j$ and if $j=2$ then the following condition
1324: must also hold.
1325: \be
1326: \setcounter{enumi}{3}
1327: \item\label{square_2_cond}
1328: If $b-a>1$ then $\G_D$ has the form shown in either Figure \ref{square_j_2_cond_a} or
1329: \ref{square_j_2_cond_b} (where all the edges shown contain at least one arc).
1330: \begin{comment}
1331: Assume $\G_D\cap D_c$ has base $D_c[x_0,x_1,y_0,y_1]$ and
1332: $\G_D\cap D_{c+1}$ has base $D_{c+1}[x_0^\prime,x_1^\prime,y_1,y_1^\prime]$. Let
1333: $T$ be the interval $\{c\}\times[0,1]$ and let $v_c$ and $v_{c+1}$ be
1334: the vertices of $\G_D\cap D_c$ and $\G_D\cap D_{c+1}$, respectively,
1335: incident to $[a,b]\times \{0\}$. Then the number of arcs incident to $v_c$ meeting
1336: $T$ is not equal to the number of arcs incident to $v_{c+1}$ meeting $T$
1337: (see Figure \ref{square_j_2_cond}).
1338: \end{comment}
1339: \ee
1340: \begin{figure}
1341: \psfrag{a}{$a$}
1342: \psfrag{b}{$b$}
1343: \begin{center}
1344: \includegraphics[scale= 0.4,clip]{fig/square_j_2_cond_a.eps}
1345: \caption{}\label{square_j_2_cond_a}
1346: \end{center}
1347: \end{figure}
1348: \begin{figure}
1349: \psfrag{a}{$a$}
1350: \psfrag{b}{$b$}
1351: \begin{center}
1352: \includegraphics[scale= 0.4,clip]{fig/square_j_2_cond_b.eps}
1353: \caption{}\label{square_j_2_cond_b}
1354: \end{center}
1355: \end{figure}
1356: Given a sum of square--pictures $\G_D$ on
1357: $D=[a,b]\+[0,1]$ we set $\G_D(c)=\G_D\cap D_c$, for $a\le c\le b-1$. A
1358: sum of square--pictures on $[a,b]\+[0,1]$ is said to have {\em width}
1359: $b-a$. The lateral and internal--lateral arcs of the square--pictures
1360: making up a
1361: sum of square--pictures are called {\em lateral} and {\em
1362: internal--lateral} arcs, respectively, of the sum.
1363:
1364: Let $D=[a,a+1]\+[0,1]$ and let $\G_D$ be a square--picture of type $j$
1365: with base $D(x_0,x_1,y_0,y_1)$ on $D$. Suppose $j>0$. If $j=1$ let
1366: $v_0$ be the vertex of $\G_D$. If $j=2$ let $v_i$ be the vertex of
1367: $\G_D$ incident to $[a,a+1]\+\{i\}$, for $i=0,1$. Let $a_0$ be the arc
1368: joining $v_0$ to $(a+1/(x_0+1),0)$. Travelling clockwise around $\pd
1369: v_0$, starting from the endpoint of $a_0$, let $a_1$ be the first arc
1370: encountered which does not meet $\{a\}\+[0,1]$. If $j=1$ then $a_1$
1371: connects $v_0$ to $(a+1/(x_1+1),1)$. If $j=2$ then $a_1$ connects $v_0$
1372: to $v_1$. In the latter case, travelling clockwise around $\pd v_1$,
1373: starting from $a_1$, let $a_2$ be the first arc encountered which does
1374: not meet $\{a\}\+[0,1]$: so $a_2$ joins $v_1$ to $(a+1/(x_1+1),1)$. The
1375: label of $v_i$, read from an appropriate starting point in a clockwise
1376: direction, is $r^{\de_i m}$, where $\de_i=\pm 1$, for $0\le i\le j-1$.
1377: The label $u_i$ on the subinterval of $\pd v_i$, read clockwise, from
1378: $a_i$ to $a_{i+1}$ is then a subword of $r^{\de_i m}$, for $0\le i\le
1379: j-1$. Let $o_i$, $\rho_i$ and $\s_i$ be
1380: non--negative integers such that $\rho_i<m$, $o_i<l(r)$, $\s_i<l(r)$
1381: and $u_i=\t(r^{\de_i},o_i)(r^{\de_i})^{\rho_i}\io(r^{\de_i},\s_i)$,
1382: for $0\le i\le j-1$. In
1383: cases $j=0$ or $1$ define $\de_i=o_i=\rho_i=\s_i=0$, for $j\le i\le 1$.
1384: \begin{defn}
1385: The {\em left interior marking} of $\G_D$ is the sequence
1386: \[\de_0,o_0,\rho_0,\s_0,\de_1,o_1,\rho_1,\s_1.\]
1387: \end{defn}
1388:
1389: For example suppose that, in Figure \ref{square-1_exx},
1390: $r=ab$, $m=4$ and the label on $v_0$ read clockwise from $a_0$ is $(ab)^4$.
1391: Then $\G_D$ has left interior marking $1,0,1,0,0,0,0,0$.
1392: \begin{figure}
1393: \psfrag{a0}{$a_0$}
1394: \psfrag{a1}{$a_1$}
1395: \psfrag{v0}{$v_0$}
1396: \begin{center}
1397: { \includegraphics[scale=0.4]{fig/square-1_exx.eps}
1398: }\\
1399: \caption{}\label{square-1_exx}
1400: \end{center}
1401: \end{figure}
1402: Again, in Figure \ref{square-2_exx} suppose $r=ab$, $m=7$, the
1403: label of $v_0$ read from $a_0$ is $(ba)^7$ and the label of
1404: $v_1$ read from $a_1$ is $(a^{-1}b^{-1})^7$. Then $\G_D$ has left interior
1405: marking $1,1,1,1,-1,1,2,0$.
1406: \begin{figure}
1407: \psfrag{a0}{$a_0$}
1408: \psfrag{a1}{$a_1$}
1409: \psfrag{v0}{$v_0$}
1410: \psfrag{a2}{$a_2$}
1411: \psfrag{v1}{$v_1$}
1412: \begin{center}
1413: { \includegraphics[scale=0.4]{fig/square-2_exx.eps}
1414: }\\
1415: \caption{}\label{square-2_exx}
1416: \end{center}
1417: \end{figure}
1418:
1419: Let $D=[a,b]\+[0,1]$ and suppose that there exists a map $\mu:D\maps
1420: \S$ and boundary intervals $P_0$ and $P_1$ of $\G$ satisfying the following
1421: properties.
1422: \be
1423: \item $\mu$ is injective on $D\bl (\{0,1\}\+[0,1])$;
1424: \item either $\mu$ is injective on $\{0,1\}\+[0,1])$ or
1425: $\mu(a,t)=\mu(b,t)$ or $\mu(a,t)=\mu(b,1-t)$, for all $t\in [0,1]$;
1426: \item $\pd \S\cap \mu(D)=\mu([a,b]\+\{0,1\})$ and
1427: $\mu([a,b]\+\{i\})\subset P_i$, for $i=0,1$;
1428: \item $\mu(\{c\}\+[0,1])$ is transverse to $\G$ and meets no vertex of
1429: $\G$, for $c\in\ZZ$ with $a\le c\le b$.
1430: \ee
1431: Then we call $\mu$ a {\em binding map}
1432: (on $D$) between $P_0$ and $P_1$.
1433: Now suppose that $P_i$ is not a partisan
1434: boundary interval, for $i=0,1$, and that no arc of $\G$ contained in
1435: $\mu(D)$ is an $H^\Lm$--arc. Let $(f_D,\G_D)$ be a sum of
1436: square--pictures of type $j$ on $D$, with left--trivial labelling
1437: function $f^l_D$, and let $\mu(D)=B$. Let $c_0,c_1$ be the left--hand
1438: corners of $D$ defined above and in addition let
1439: $c_2=(\{b\}\+[0,1/(y_1+1)])\cup ([b-1+(x_0/(x_0+1)),b]\+\{0\})$ and
1440: $c_3=(\{b\}\+[y_1/(y_1+1),1])\cup ([b-1+(x_1/(x_1+1)),b]\+\{1\})$.
1441: Recall that $f$ is the map from $\S$ to $Z$ described in the
1442: definition of picture.
1443:
1444:
1445: Assume that $\mu(\G_D)=\G\cap B$ and
1446: \be
1447: \item $f^l(\mu(c))=f^l_D(c)$, for all corners $c$ of $\G_D$ except
1448: $c_0,c_1, c_2$ and $c_3$,
1449: \item $f(\mu([a,a+(1/(x_i+1))]\+\{i\}))$ is nullhomotopic, for
1450: $i=0,1$, and
1451: \item $d_i=\mu([b-1+(x_i/(x_i+1)),b]\+\{i\})\subset c^\prime_i$, where
1452: $c^\prime_i$ is a boundary corner of $\G$ and $f(c_i^\prime\bl d_i)$
1453: is nullhomotopic, for $i=0,1$.
1454: %$f(d_i)$ maps to a path in $Z$ representing the element
1455: %$f^l(c^\prime_i)\in A\cup B$, for $i=0,1$.
1456: \ee
1457: Then $C=(D,\G_D,\mu)$ is called a $j${\em --corridor} of {\em
1458: width} $b-a$ with binding map $\mu$ {\em binding} $P_0$ and
1459: $P_1$. In this case $\G_B=\G\cap \mu(D)$ is a picture on $B$ and we
1460: call $B$ the {\em image} of $C$.
1461: The images $\mu(x)$ of arcs $x$ of the picture $\G_D$ are called
1462: {\em arcs} of $C$.
1463: If $\mu(a,t)=\mu(b,t)$ or $\mu(a,t)=\mu(b,1-t)$, for all $t\in [0,1]$,
1464: then all arcs of $C$ are arcs of $\G$. Otherwise, given an arc
1465: $x$ of $\G_D$, the arc $\mu(x)$ of
1466: $C$ is an arc of $\G$ if and only if $x$ does not meet
1467: $\{a,b\}\+[0,1]$. Nonetheless it is convenient to refer to $\mu(x)$ as
1468: an arc of $C$, for all arcs $x$ of $\G_D$.
1469: A
1470: corridor of width 1 is called a {\em compartment}. The {\em left
1471: interior marking of rank} $s$ of the corridor $C$ is defined to be
1472: the left interior marking of $\G_D(s)$, for all $s$ such that $a\le
1473: s\le b-1$.
1474:
1475:
1476: If $e$ is a lateral arc, or internal--lateral arc, of $\G_D$
1477: %which meets $[a,b]\+\{i\}$, for $i=0$ or 1,
1478: then $\mu(e)$ is called a {\em lateral} arc, or {\em
1479: internal--lateral} arc, of the corridor $(D,\G_D,\mu)$. If
1480: $C=(D,\G_D,\mu)$ and $C^\prime=(D^\prime,\G_D^\prime,\mu^\prime)$ are
1481: corridors with identical sets of lateral arcs then we say that $C$ and
1482: $C^\prime$ are {\em equivalent} and this is an equivalence relation on
1483: the set of corridors. If $C$ and $C^\prime$ are equivalent then
1484: $\mu^\prime$ is homotopic to a map $\mu^{\prime\prime}$ such that
1485: $\mu^{\prime\prime}(D^\prime)=\mu(D)$ and
1486: $(D^\prime,\G_D^\prime,\mu^{\prime\prime})$ is a corridor equivalent
1487: to $C$. If $C^\prime=(D^\prime,\G_D^\prime,\mu^\prime)$ is a corridor
1488: such that $\{$lateral arcs of
1489: $C^\prime\}\subseteq\{$lateral arcs of $C\}$
1490: then $C^\prime$ is called a {\em sub--corridor} of $C$. If $a\le
1491: c<d\le b$ and $D^\prime = [c,d]\+[0,1]$ then the map
1492: $\mu^\prime=\mu|_{D^\prime}$ is called the binding map {\em induced}
1493: from $\mu$ on $D^\prime$ and the corridor $(D^\prime,
1494: \G_{D^\prime},\mu^\prime)$ is called the sub--corridor {\em induced}
1495: from $(D,\G_D,\mu)$, denoted $C[c,d]$ or $C[c]$, if $c+1=d$.
1496: If $C$ is corridor and is not a sub--corridor of
1497: any corridor of greater width then we say that $C$ is a {\em maximal}
1498: corridor. Obviously two corridors are equivalent if and only
1499: if each one is a sub--corridor of the other.
1500: % and a homotopy of the binding map of the second,
1501: % relative to vertices and lateral arcs, makes their images identical.
1502: If $C=(D,\G_D,\mu)$ and
1503: $C^\prime=(D^\prime,\G_{D^\prime},\mu^\prime)$ are corridors, with
1504: images $B=\mu(D)$ and $B^\prime=\mu^\prime(D^\prime)$, we
1505: call $B\cap B^\prime$ the {\em intersection} of $C$
1506: and $C^\prime$, denoted $C\cap C^\prime$, and if
1507: $B\cap B^\prime\neq \nul$ we shall say that $C$ and
1508: $C^\prime$ {\em overlap}.
1509:
1510: \begin{defn}
1511: Choose one representative from each equivalence class of maximal
1512: $j$--corridors of $\G$, where $j=0$ or $1$.
1513: %These representatives are called {\em maximal} $\G${\em --corridors}.
1514: These corridors and the sub--corridors induced from them are called
1515: {\em designated} corridors of $\G$ (of type 0 and 1).
1516: \end{defn}
1517: (We defer the definition of designated 2--corridor until later.)
1518: We may always choose the maximal designated corridor $(D,\G_D,\mu)$
1519: such that
1520: $D=[0,c]\+[0,1]$, for some integer $c$. In the absence of any
1521: explicit statement to the contrary it is to be assumed that all
1522: corridors are designated corridors.
1523:
1524: If $(D,\G_D,\mu)$ is a maximal $0$--corridor then $\G_D$ consists
1525: of arcs from a class of
1526: %parallel arcs which meets no vertex.
1527: arcs of type $I$. If two distinct designated maximal 0--corridors
1528: overlap then we may choose different representatives of the two
1529: equivalence classes of corridors (see Figure \ref{Corridor0-i})
1530: \begin{figure}
1531: \begin{center}
1532: { \includegraphics[scale=0.5]{fig/corridor0-i.eps}
1533: }\\
1534: \caption{}\label{Corridor0-i}
1535: \end{center}
1536: \end{figure}
1537: which do not overlap. We assume then that maximal designated
1538: 0--corridors do not overlap. If two distinct designated maximal
1539: 1--corridors overlap but the intersection contains no arc or vertex
1540: then, as in the type 0 case, we designate different representatives
1541: which do not overlap. We may assume then that the intersection
1542: contains an arc or vertex. If the intersection contains a vertex then
1543: %the two designated corridors have in common
1544: it contains the lateral arcs incident
1545: to this vertex. After performing a homotopy if necessary we have two
1546: maximal corridors whose intersection contains a compartment. Therefore we may
1547: extend each corridor to contain all compartments of the other. Since
1548: the original corridors are maximal they must be equivalent,
1549: contradicting uniqueness of representatives. If the intersection
1550: contains an arc but no vertex then this arc must be the unique lateral
1551: arc of a compartment of type 0. The same argument applies in this
1552: case. Hence we may assume that maximal designated 1--corridors do not
1553: overlap.
1554:
1555: Maximal 2--corridors require further analysis. To begin with we
1556: shall, from now on, call the 2--corridors defined above {\em
1557: simple}
1558: 2--corridors. For ease of reference we shall also refer to corridors
1559: of type 0 and 1 as {\em simple}: so simple $j$--corridors and
1560: $j$--corridors are the same thing, for $j=0,1$. Let $C=(D,\G_D,\mu)$ be
1561: a simple 2--corridor and let $(B,\G_B)=(\mu(D),\G\cap\mu(D))$. Then the
1562: vertices of $\G_B$ together with those arcs of $\G$ whose closure is
1563: contained in $B$ map to a subgraph $\bar C$ of $\bar \G$ on $\S$,
1564: which we call the {\em graph} of $C$. The images of lateral and
1565: internal--lateral arcs of $C$ are called {\em internal} and {\em
1566: internal--lateral} edges, respectively, of $\bar C$. Let $C^\prime
1567: =(D^\prime,\G_{D^\prime},\mu^\prime)$ be a simple 2--corridor. Then $C$ is a
1568: sub--corridor of $C^\prime$ if and only if $\bar C\subset \bar
1569: C^\prime$. Moreover, if $\mu(D)\cap\mu^\prime(D^\prime)$
1570: contains an arc $a$ then every arc equivalent to $a$ also belongs to
1571: $\mu(D)\cap\mu^\prime(D^\prime)$. Hence
1572: $\mu(D)\cap\mu^\prime(D^\prime)$ contains an arc if and only
1573: if $\bar C\cap \bar C^\prime$ contains and edge. Suppose now that $C$
1574: and $C^\prime$ both bind boundary intervals $P_0$ and $P_1$. If $\bar
1575: C\subset \bar C^\prime$ and both $C$ and $C^\prime$ have width $1$
1576: then clearly $C$ and $C^\prime$ are equivalent. Suppose next that $\bar
1577: C\subset \bar C^\prime$, the
1578: width of $C$ is 1 and the width of $C^\prime$ is at least 2.
1579: Assume further that the graph $\bar C$ meets $P_i$ at a point $x_i$,
1580: let $e_i$ be the edge of $\bar C$ incident to $x_i$ and let $u_i$ be
1581: the vertex of $\bar C$ incident to $e_i$, for $i=0,1$. (See Figure
1582: \ref{2corridor-1-2}.)
1583: \begin{figure}
1584: \psfrag{x0}{$x_0$}
1585: \psfrag{x1}{$x_1$}
1586: \psfrag{u0}{$u_0$}
1587: \psfrag{u1}{$u_1$}
1588: \psfrag{e0}{$e_0$}
1589: \psfrag{e1}{$e_1$}
1590: \psfrag{P0}{$P_0$}
1591: \psfrag{P1}{$P_1$}
1592: \begin{center}
1593: { \includegraphics[scale=0.4]{fig/2corridor-1-2.eps}
1594: }\\
1595: \caption{}\label{2corridor-1-2}
1596: \end{center}
1597: \end{figure}
1598: As $\bar C\subset \bar C^\prime$ the vertices $u_i$ also belong to
1599: $\bar C^\prime$. There are two possibilities. \be
1600: \item\label{2corr-align} The internal--lateral edge of $\bar C$
1601: joining $u_0$ and $u_1$ is an internal--lateral edge of $\bar
1602: C^\prime$. In this case $C$ is equivalent to a compartment of
1603: $C^\prime$.
1604: \item\label{2corr-diag} The internal--lateral edge of $\bar C$ joining
1605: $u_0$ and $u_1$ is not an internal--lateral edge of $\bar C^\prime$.
1606: Therefore either $u_0$ or $u_1$ is joined to a vertex of $\bar
1607: C^\prime$ by an internal--lateral edge of $\bar C^\prime$ which is
1608: not an internal--lateral edge of $\bar C$. Suppose that $u_0$ is
1609: joined to vertex $v_1$ of $\bar C^\prime$ by an internal--lateral
1610: edge of $\bar C^\prime$ which is the image of a class of arcs in
1611: a compartment $C^\prime[j]$ of $C^\prime$. Then $u_1$ is a vertex of
1612: compartment $C^\prime[j-1]$ or $C^\prime[j+1]$
1613: of $C^\prime$
1614: (see Figure \ref{diag1-2}). In this case $C$ is not equivalent to
1615: any compartment of $C^\prime$. The case $u_1$ joined to a vertex of
1616: $\bar C^\prime$ by an internal--lateral edge of $\bar C^\prime$ is
1617: similar.
1618: \ee
1619: \begin{figure}
1620: \psfrag{v1}{$v_1$}
1621: \psfrag{u0}{$u_0$}
1622: \psfrag{u1}{$u_1$}
1623: \begin{center}
1624: \mbox{
1625: \subfigure[$u_1\in C^\prime \lbrack j-1\rbrack$\label{diag1-2-1}]%\label{diag1-2-1}
1626: { \includegraphics[scale= 0.4,clip]{fig/diag1.eps} }\qquad
1627: \subfigure[$u_1\in C^\prime \lbrack j+1 \rbrack$ ]
1628: { \includegraphics[scale=0.4,clip]{fig/diag2.eps} }
1629: }
1630: \caption{}\label{diag1-2}
1631: \end{center}
1632: \end{figure}
1633:
1634: In general if $\bar C\subset \bar C^\prime$ then
1635: without loss of generality we may assume that
1636: $D=[0,c]\+[0,1]$ and $D^\prime=[0,c^\prime]\+[0,1]$, for some integers
1637: $c$ and $c^\prime$.
1638: \begin{comment}
1639: Furthermore if ($C$ has width at least 2 and )
1640: $u$ is the vertex of the square--picture $\G_D(0)$ of $\G_D$
1641: incident to $[0,c]\+\{0\}$ then we may
1642: assume that
1643: $u$ is joined to the vertex of $\G_D(1)$
1644: which is incident to $[0,c]\+\{1\}$ (as well as the vertex of $\G_D(0)$
1645: incident to $[0,c]\+\{1\}$).
1646: \end{comment}
1647: %Let $D_0\subset D$
1648: %and $\G_0\subset\G_D$ be defined as above
1649: %and let $\mu_0=\mu|_{D_0}$. Then $C_0=(D_0,\G_0,\mu_0)$ is a compartment
1650: %of $C$.
1651: If the compartment $C[0]$ is equivalent to a compartment of $C^\prime$ (as in
1652: case (\ref{2corr-align}) above) then $C$ is equivalent to a
1653: sub--corridor of $C^\prime$. If, on the other hand, $C[0]$ is not
1654: equivalent to a compartment of $C^\prime$ (as in case (\ref{2corr-diag})
1655: above) then no compartment of $C$ is equivalent to a compartment of
1656: $C^\prime$. (This follows by induction starting with the case (\ref{2corr-diag}) above.)
1657: In this case $C$ is a sub--corridor of $C^\prime$ but equivalent
1658: to no sub--corridor of $C^\prime$ and we say that $C$ is {\em diagonal} to $C^\prime$. By
1659: definition, if $C$ has width at least 2 and is diagonal to $C^\prime$
1660: then $C^\prime$ has a sub--corridor which is diagonal to $C$. For
1661: example, consider Figure \ref{sub-diagonal}, which shows the graphs of
1662: two corridors $C\subset C^\prime$: the edges of $\bar C$ are shown solid,
1663: whilst those of $\bar C^\prime\bl\bar C$ are dashed. The
1664: internal--lateral edges of $\bar C$ are shown diagonal whilst those of $\bar
1665: C^\prime$ are vertical. Thus the left hand compartment of $C$ meets
1666: the boundary at $a$ and $d$ whilst the right hand compartment meets it at $c$ and
1667: $e$. The sub--corridor of $C^\prime$ between $b$ and $c$ is diagonal
1668: to $C$.
1669: \begin{figure}
1670: \psfrag{a}{$a$}
1671: \psfrag{b}{$b$}
1672: \psfrag{c}{$c$}
1673: \psfrag{d}{$d$}
1674: \psfrag{e}{$e$}
1675: \begin{center}
1676: { \includegraphics[scale=0.6]{fig/sub-diagonal.eps}
1677: }\\
1678: \caption{}\label{sub-diagonal}
1679: \end{center}
1680: \end{figure}
1681: We say that simple 2--corridors $C_0$ and $C_1$, both binding boundary intervals $P_0$ and
1682: $P_1$ and both of width $c$,
1683: are {\em diagonally--equivalent} if there is a sub--corridor of $C_i$ of width
1684: $c-1$ which is diagonal to $C_{1-i}$, for $i=0,1$.
1685:
1686: %%%%%%%%%%B goes here
1687: If $C_2$ is a corridor of width $2$ then there is a unique equivalence class of
1688: compartments diagonal to $C_2$. This is a consequence of condition
1689: (\ref{square_2_cond}) for sums of pictures of type $2$.
1690: It follows that if $C_0$ is a corridor diagonally--equivalent to a
1691: corridor $C_1$ and if $C_1$ is
1692: diagonally--equivalent to a corridor $C_2$ then $C_0$ and $C_2$ are equivalent.
1693: In particular if $C_0$
1694: and $C_1$ are corridors
1695: diagonally--equivalent to a corridor $C_3$ then $C_0$ and $C_1$ are equivalent.
1696: %%%%%%%%%%%%%%%%%
1697: \begin{lemma}\label{2corr-maxint}
1698: Let $C%=(D,\G_D,\mu)
1699: $ and $C^\prime%=(D^\prime,\G_{D^\prime},\mu^\prime)
1700: $
1701: be maximal simple 2--corridors.
1702: \be
1703: \item\label{2corr-maxint-1} If a simple sub--corridor of $C$
1704: is equivalent to a sub--corridor of $C^\prime$
1705: then $C$ and $C^\prime$ are equivalent.
1706: \item\label{2corr-maxint-2} If a simple sub--corridor of $C$ is diagonal
1707: to $C^\prime$ then $C$ and $C^\prime$ are diagonally--equivalent.
1708: \ee
1709: %then $E$ and $E^\prime$ are equivalent $2^+$-corridors.
1710: %Let $E=C_0+C_1$ and $E^\prime=C_0^\prime+C_1^\prime$ be maximal extended
1711: \end{lemma}
1712: \medskip
1713:
1714: \noindent
1715: {\em Proof.}
1716: \be
1717: \item
1718: If some sub--corridor of $C$ is
1719: equivalent to a
1720: sub--corridor of $C^\prime$ then there is a corridor $C^\pprime$
1721: containing all compartments of $C$ and $C^\prime$.
1722: Then both $C$ and $C^\prime$ are sub--corridors of $C^\pprime$
1723: and, as $C$ and $C^\prime$ are maximal, $C^\pprime$ is a sub--corridor
1724: of them both. Hence $C$ and $C^\prime$ are equivalent to $C^\pprime$.
1725: \item
1726: Assume that
1727: some sub--corridor
1728: $C_s$ of $C$ is diagonal to $C^\prime$. Choose $p<q$ such that
1729: $C_s=C[p,q]$ is maximal
1730: in this respect and write $C_s=(D_s,\G_s,\mu_s)$. As $C_s$ is maximal among
1731: sub--corridors of $C$ diagonal to $C^\prime$, either there is a vertex
1732: of $\mu(\G_D({p-1}))$ which is not a vertex of $\mu^\prime(\G_{D^\prime})$
1733: or $p=0$. Similarly, either there is a vertex of $\mu(\G_D(q))$ which is
1734: not a vertex of $\mu^\prime(\G_{D^\prime})$ or $q=c$. If both $p=0$ and
1735: $q=c$ then $C$ is a sub--corridor of $C^\prime$ and the width of $C^\prime$
1736: is necessarily greater than that of $C$, contrary to maximality of $C$.
1737: Thus either $p>0$ or $q<c$. Assume first that $p>0$ and that there is a vertex
1738: of $\mu(\G_D({p-1}))$ which is not a vertex of $\mu^\prime(\G_{D^\prime})$.
1739: Since the vertices of $\mu(\G_D(p))$ belong to different compartments of $C^\prime$,
1740: it follows that one of the vertices of $\mu(\G_D({p-1}))$ belongs to a compartment
1741: of $C^\prime$ (and is joined to a vertex of $\mu(\G_D(p))$ by an internal--lateral
1742: arc of $C^\prime$).
1743: %If both
1744: %vertices of $\mu(\G_D({p-1}))$ lie outside $\mu^\prime(\G_{D^\prime})$ then
1745: %it follows that the compartment $\mu(\G_D(p))$ of $C$
1746: %is equivalent to a compartment of $C^\prime$,
1747: %a contradiction.
1748: Hence one vertex of $\mu(\G_D({p-1}))$ is contained in
1749: $\mu^\prime(\G_{D^\prime})$. If $p\ge 2$ it is therefore
1750: possible to extend $C^\prime$ to a wider corridor by adjoining one vertex of
1751: $\mu(\G_D({p-1}))$
1752: and one vertex of $\mu(\G_D({p-2}))$. This is contrary to the maximality of $C^\prime$ and
1753: so if $p>0$ then $p=1$. A similar argument shows that if $q<c$ then $q=c-1$.
1754: Now if $p=1$ and $q=c-1$ then it follows that $C^\prime$ is
1755: a sub--corridor of $C$, so $C$ and $C^\prime$ are equivalent, contrary to hypothesis.
1756: Hence either
1757: $p=1,q=c$ or $p=0,q=c-1$. It follows that the width of $C_s$ is one less than that
1758: of $C$ and that the compartment of $C$ not in $C_s$ has precisely
1759: one vertex in $C^\prime$. As $C^\prime$ has a sub--corridor diagonal to $C$ the same
1760: holds on reversing the roles of $C$ and $C^\prime$. Thus
1761: $C$ and $C^\prime$ are diagonally--equivalent.
1762: \ee
1763: %We define {\em strong equivalence}
1764: %of 2--corridors to be the equivalence relation generated by equivalence and
1765: %diagonal--equivalence.
1766: %Hence corridors $C_s$ and $C_t$ are strongly equivalent if and only if $C_s$ is
1767: %either equivalent or diagonally--equivalent to $C_t$.
1768:
1769: From the proof of the second part of the above Lemma it follows that if $C$ and
1770: $C^\prime$ are diagonally equivalent then $\bar{C^\prime}$ contains exactly one
1771: vertex and one lateral edge not in $\bar C$, and vice--versa.
1772:
1773: Given that $C=(D,\G_D,\mu)$ and
1774: $C^\prime=(D^\prime,\G_{D^\prime},\mu^\prime)$ are
1775: diagonally--equivalent simple 2--corridors we define the sum
1776: $C + C^\prime$ to be $(D\dcup D^\prime,\G_D\dcup \G_{D^\prime},\mu\dcup\mu^\prime)$
1777: %$\mu(\G_D)\cup\mu^\prime(\G_{D^\prime})$ on $\mu(D)\cup \mu^\prime(D^\prime)$
1778: and call this an {\em extended} 2--corridor, {\em extending} $C$ and
1779: $C^\prime$. The set of {\em lateral} arcs of an extended 2--corridor $C+C^\prime$ is the
1780: union $\{$lateral arcs of $C\}\cup \{$lateral arcs of $C^\prime\}$.
1781: We define the set of {\em sub--corridors} of the extended 2--corridor
1782: $E=C+C^\prime$ to be the union of $\{E\}$ with the sets of
1783: sub--corridors of $C$ and $C^\prime$.
1784: (We shall not need to consider extended corridors as sub--corridors of other
1785: extended corridors.)
1786: If $E=C+C^\prime$ is an extended 2--corridor then we define the {\em graph}
1787: of $E$ to be $\bar E=\bar C\cup\bar C^\prime$ and the {\em image} of
1788: $E$ to be $B\cup B^\prime$, where $B$ and $B^\prime$ are the images of
1789: $C$ and $C^\prime$, respectively. By the remark following the Lemma above
1790: $\bar E$ has one more lateral edge than both $\bar C$ and $\bar C^\prime$.
1791:
1792: We define the set of 2--corridors to be the union of the
1793: set of all simple 2--corridors and the set of all extended 2--corridors.
1794: We extend the
1795: equivalence relation defined on simple 2--corridors by defining 2--corridors
1796: $C$ and $C^\prime$ to be {\em equivalent} if they have the same lateral arcs.
1797: As graphs of
1798: simple $2$--corridors have an even number of lateral edges, graphs of
1799: extended $2$--corridors
1800: have an odd number of lateral edges and so no simple $2$--corridor can be
1801: equivalent to an extended $2$--corridor.
1802: A
1803: 2--corridor $C$ is {\em maximal} in the set of 2--corridors if
1804: $C$ is equivalent to $C^\prime$, for all 2--corridors $C^\prime$ such that
1805: $\{$lateral edges of $C\}\subset \{$lateral edges of $C^\prime\}$.
1806: If $E$ and $E^\prime$ are 2--corridors with images $B$ and
1807: $B^\prime$, respectively, then we say that $E$ and $E^\prime$ have
1808: {\em intersection} $B\cap B^\prime$ and that they {\em
1809: overlap}
1810: if $B\cap B^\prime\neq\nul$.
1811:
1812: \begin{lemma}\label{maxl-ext}
1813: Let $E=C_0+C_1$ be an extended 2--corridor
1814: where $C_0$ and $C_1$ are diagonally--equivalent simple $2$--corridors.
1815: Then the following hold.
1816: \be
1817: \item\label{maxl-ext-1} If $E$ is maximal as a $2$--corridor then
1818: $C_0$ and $C_1$ are maximal simple 2--corridors.
1819: \item\label{maxl-ext-2} if $E=C_0+C_1^\prime$ then $C_1$ is equivalent to $C_1^\prime$.
1820: %\item\label{maxl-ext-3} if $E^\prime=C_0+C_2$ is a maximal $2+$--corridor then
1821: %$C_1$ is equivalent to $C_2$.
1822: \ee
1823: \end{lemma}
1824: \medskip
1825:
1826: \noindent
1827: {\em Proof.}
1828: Let $C_0$ and $C_1$ have width $c$.
1829: \be
1830: \item
1831: Let $C^\prime_i$ be a maximal simple $2$--corridor containing $C_i$, $i=0,1$.
1832: Then $C_0$ is a sub--corridor of $C_0^\prime$ diagonal to $C^\prime_1$. From
1833: Lemma \ref{2corr-maxint}.(\ref{2corr-maxint-2}),
1834: $C_0^\prime$ and $C^\prime_1$ are diagonally--equivalent.
1835: Hence we may form the sum $C_0^\prime+C_1^\prime$. As $E$ is maximal it is
1836: equivalent to $C_0^\prime+C_1^\prime$. If $e$ is a lateral edge of
1837: $\bar C^\prime_0\backslash \bar C_0$ this implies that $e$ is a lateral edge of
1838: $\bar C_1\backslash \bar C_0$.
1839: As $C_0$ and $C_1$ are diagonally equivalent there is only one
1840: such edge. However, if $\bar C_0^\prime$ is wider than $\bar C_0$ then it must contain
1841: at least $2$ more lateral edges than $\bar C_0$. Hence $C_0^\prime$ and $C_0$ are
1842: equivalent. A similar argument applies to $C_1$.
1843: \item There is a unique vertex $v$ of $\bar E$ which belongs to $\bar C_1$ but not
1844: $\bar C_0$. Therefore $v$ is the unique vertex of $\bar E$
1845: which belongs to $\bar C_1^\prime$ but not $\bar C_0$. Hence $C_1$ and
1846: $C_1^\prime$ must have the same vertices which implies that they are equivalent.
1847: %\item If $\bar E^\prime$ contains a vertex not in $E$ then this vertex belongs to
1848: %$\bar C_2$ and not $\bar C_0$ or $\bar C_1$. As both $C_1$ and $C_2$ are
1849: %diagonally-equivalent to $C_0$ it follows that the graph
1850: %$T=\bar C_0\cup\bar C_1 \cup\bar C_2$ strictly contains the graph $\bar E$. However the
1851: %graph $T$ corresponds to a 2--corridor or extended 2--corridor, contrary to maximality
1852: %of $E$.
1853: %\item
1854: \ee
1855:
1856: \begin{lemma}\label{2+corr-maxint}
1857: Let $E$ and $E^\prime$ be maximal 2--corridors.
1858: \be
1859: \item \label{2+corr-maxint-1}If a simple sub--corridor of $E$ is
1860: equivalent to a simple sub--corridor of $E^\prime$ or
1861: \item\label{2+corr-maxint-2} a simple sub--corridor of $E$ is diagonal
1862: to a simple sub--corridor of $E^\prime$
1863: \ee
1864: then $E$ and $E^\prime$ are equivalent 2--corridors.
1865: %Let $E=C_0+C_1$ and $E^\prime=C_0^\prime+C_1^\prime$ be maximal extended
1866: \end{lemma}
1867: \medskip
1868:
1869: \noindent
1870: {\em Proof.}
1871: If $E$ and $E^\prime$ are simple 2--corridors then, from Lemma \ref{2corr-maxint},
1872: (\ref{2+corr-maxint-1}) follows immediately and (\ref{2+corr-maxint-2}) is not possible.
1873: If $E$ is a simple 2--corridor and $E^\prime$ an extended 2--corridor then it follows from
1874: Lemma \ref{maxl-ext} and Lemma \ref{2corr-maxint} with either (\ref{2+corr-maxint-1}) or (\ref{2+corr-maxint-2}),
1875: that $E^\prime$ is equivalent to $E+C^\prime$, where $C^\prime$ is a simple sub--corridor of $E^\prime$,
1876: contrary to maximality
1877: of $E$. If $E^\prime$ is a simple 2--corridor and $E$ an extended 2--corridor then again a
1878: contradiction arises.
1879: %it follows
1880: %from Lemma \ref{maxl-ext} and Lemma \ref{2corr-maxint} that $E^\prime$ is not maximal,
1881: %a contradiction.
1882: Hence we assume that both $E$ and $E^\prime$ are extended 2--corridors,
1883: say $E=C_0+C_1$ and $E^\prime=C_0^\prime+C_1^\prime$.
1884: \be
1885: \item If a simple sub--corridor of $E$ is equivalent to a simple sub--corridor of $E^\prime$ then,
1886: for some $i,j\in\{0,1\}$ a sub--corridor of $C_i$ is equivalent to a sub--corridor
1887: of $C^\prime_j$. From Lemmas \ref{maxl-ext}(\ref{maxl-ext-1}) and
1888: \ref{2corr-maxint}(\ref{2corr-maxint-1}) it follows that
1889: $C_i$ is equivalent to $C^\prime_j$. Thus $E=C_i+C_{1-i}$ is equivalent to
1890: $C^\prime_j+C_{1-i}$.
1891: Hence $C_{1-i}$ is diagonally--equivalent to $C^\prime_j$ which is
1892: diagonally--equivalent to $C^\prime_{1-j}$. It follows that $C_{1-i}$ and $C^\prime_{1-j}$ are
1893: equivalent and therefore so are $E$ and $E^\prime$.
1894: \item If a simple sub--corridor of $E$ is diagonal to a simple sub--corridor of $E^\prime$ then,
1895: for some $i,j\in\{0,1\}$ a sub--corridor of $C_i$ is diagonal to
1896: $C^\prime_j$. From Lemma \ref{maxl-ext}(\ref{maxl-ext-1}) and Lemma
1897: \ref{2corr-maxint}(\ref{2corr-maxint-2}) it follows that $C_i$ and $C^\prime_j$
1898: are diagonally--equivalent. As $C_i$ is diagonally--equivalent to $C_{1-i}$ it follows
1899: that $C^\prime_j$ is equivalent to $C_{1-i}$. As in the previous case it follows
1900: that $E$ and $E^\prime$ are equivalent.
1901: \ee
1902:
1903: \begin{lemma}\label{ext2corr-int}
1904: Let $E=C_0+C_1$ and $E^\prime=C_0^\prime+C_1^\prime$
1905: be maximal extended 2--corridors. Then either $E$ and $E^\prime$ are
1906: equivalent or $\bar E\cap \bar E^\prime\subset \bar C_i\cap \bar C^\prime_j$, for
1907: some $i,j\in\{0,1\}$.
1908: \end{lemma}
1909: \medskip
1910:
1911: \noindent
1912: {\em Proof.}
1913: If $\bar{E} \cap \bar{E}^\prime=\emptyset$ then the result holds trivially.
1914: Suppose then that $E$ and $E^\prime$ are not equivalent and that
1915: $\bar{E} \cap \bar{E}^\prime\neq\emptyset$. Note that this implies that
1916: the images of $E$ and $E^\prime$ are disks (and not annulii or M\"obius bands).
1917: There is a unique vertex
1918: $u_0$ of $\bar C_0$ not in $\bar C_1$ and
1919: a unique vertex
1920: $u_1$ of $\bar C_1$ not in $\bar C_0$. Similarly,
1921: there is a unique vertex
1922: $v_0$ of $\bar C^\prime_0$ not in $\bar C^\prime_1$ and
1923: a unique vertex
1924: $v_1$ of $\bar C^\prime_1$ not in $\bar C^\prime_0$. Fix $j=0$ or $1$: if the
1925: vertex $v_j$ is not in $\bar E$ then
1926: $\bar E\cap \bar E^\prime\subset \bar E \cap \bar C^\prime_{1-j}$.
1927: Now if $u_i$ is not in $\bar C^\prime_{1-j}$, for
1928: $i=0,1$, then
1929: $\bar E\cap \bar E^\prime\subset \bar C_{1-i} \cap \bar C^\prime_{1-j}$.
1930: On the other hand if $u_i\in \bar C^\prime_{1-j}$, for $i=0$ and $1$, then
1931: $\bar C^\prime_{1-j}$ contains $\bar E$, contrary to maximality of $E$. Finally suppose
1932: that $v_j$ belongs to $\bar E$, for $j=0$ and 1. As $\bar E$ and $\bar E^\prime$ are
1933: both graphs of extended 2--corridors on a disk this implies that $\bar E^\prime$
1934: is a subgraph of $\bar E$. As $E$ and $E^\prime$ are not equivalent it follows that
1935: $\bar E^\prime$ is a proper subgraph of $\bar E$ and so $E^\prime$ is not maximal,
1936: a contradiction.
1937:
1938: \begin{defn}
1939: Choose one representative $E$ from each equivalence class of maximal
1940: 2--corridors of $\G$.
1941: If $E$ is the representative of an equivalence class of maximal
1942: extended 2--corridors then we call $E$ a {\em designated} extended
1943: 2--corridor. Corresponding to each designated extended 2--corridor
1944: $E$ choose an ordered pair
1945: $(C,C^\prime)$ of diagonally--equivalent simple 2--corridors such that
1946: $E=C+C^\prime$ and say that $E$ has {\em foundation} $C$ and
1947: {\em extension} $C^\prime$. If $C$ is the chosen representative of an
1948: equivalence class of maximal simple 2--corridors or $C$ is the foundation of
1949: a designated extended 2--corridor
1950: then we call $C$ and all its induced sub--corridors {\em designated}
1951: simple 2--corridors. We say that a 2--corridor is {\em designated} if it
1952: is either a designated extended 2--corridor or a designated simple 2--corridor.
1953: \end{defn}
1954: We assume from now on that all 2--corridors are designated. In particular if
1955: $C_s$ is a proper sub--corridor of the extended 2--corridor $E$ with foundation
1956: $C$ then, by definition, $C_s$ is a sub--corridor induced from $C$.
1957: (Note that if $E$ is a designated extended 2--corridor with
1958: foundation $C$ and extension $C^\prime$ then $C$ is a designated
1959: simple 2--corridor but $C^\prime$ is not.)
1960:
1961: Suppose now that $E=C_0+C_1$ and $E^\prime=C_0^\prime+C_1^\prime$
1962: are maximal designated
1963: 2--corridors, with foundations $C_0$ and $C_0^\prime$,
1964: respectively. Assume further that $E$ and $E^\prime$ are not
1965: equivalent. If $E$ and $E^\prime$ overlap then we may assume
1966: that the images of $E$ and $E^\prime$ are not annulii or M\"obius bands and, as
1967: in the case of simple 2--corridors,
1968: that the graphs $\bar E$ and $\bar E^\prime$ have at least an edge in common.
1969: Furthermore, from Lemma \ref{2+corr-maxint}, no sub--corridor of $E$ is equivalent or
1970: diagonal to a sub--corridor of $E^\prime$ and, from Lemma
1971: \ref{ext2corr-int}, $\bar E\cap \bar E^\prime\subset \bar C_i\cap \bar
1972: C_j^\prime$, for some $i,j\in \{0,1\}$.
1973: It follows that no compartment of $C_i$ is equivalent or diagonal to
1974: a sub--corridor of $C_j^\prime$, and vice--versa. Therefore $\bar
1975: C_i\cap \bar C_j^\prime$ contains a unique edge. Hence the
1976: intersection of $E$ and $E^\prime$ contains one vertex and one class
1977: of lateral arcs (corresponding to the edge of $\bar C_i\cap \bar
1978: C_j^\prime$). Therefore the union of the images of $E$ and $E^\prime$
1979: form a configuration of the form shown in Figure \ref{keel}: in which
1980: case we say $E$ and $E^\prime$ form
1981: a {\em keel}.
1982: \begin{figure}
1983: \psfrag{C}{$C$}
1984: \psfrag{C'}{$C^\prime$}
1985: \begin{center}
1986: { \includegraphics[scale=0.6]{fig/keel.eps}
1987: }\\
1988: \caption{The unique edge of the intersection is shown as a dashed line}\label{keel}
1989: \end{center}
1990: \end{figure}
1991:
1992: We collect together our conclusions on the intersection of corridors
1993: in the following Lemma.
1994: \begin{lemma}
1995: If $C$ and $C^\prime$ are maximal designated $j$--corridors, $j=0,1$
1996: or 2, then either $C$ and $C^\prime$ do not overlap or $j=2$ and
1997: $C$ and $C^\prime$ form a keel.
1998: \end{lemma}
1999: From now on all corridors are assumed to be designated.
2000: \subsection{$Z$--cancellation}
2001: Let $C=(D,\G_D,\mu)$ be a $j$--corridor, where $j=0,1$ or 2 and
2002: $D=[0,c]\+[0,1]$, for some integer $c$. Recall that $\G$ has boundary
2003: partition $\mbf b=(b_1,\ldots ,b_{W_1})$ and suppose that $C$ binds
2004: boundary intervals $P_0=b_{p_0}$ and $P_1=b_{p_1}$, where $1\le p_1\le
2005: W_1$, that $P_i$ has prime label
2006: $(h_i,f_i)$ and that $\al(f_i)=m_i$: so the label of $P_i$ is
2007: $h_i\^m_i$. Taking the point $\mu((0,i))$ to precede $\mu((c,i))$
2008: gives an orientation $\zeta_i$ of $P_i$. If $\zeta_i$ coincides with the fixed
2009: orientation $\zeta(P_i)$ of $P_i$ set $\e_i=1$ and if not set
2010: $\e_i=-1$.
2011: Fix an integer $s$ with $0\le s\le c-1$ and assume that $\G_D(s)$ has
2012: base $D_s[x_0,x_1,y_0,y_1]$.
2013: The label $w_i$ of the sub--interval of $P_i$, between the point
2014: $\mu((s+\frac{1}{x_i+1},i))$ and the end of the interval $P_i$, read
2015: in the direction of $\zeta_i$, therefore begins with a terminal
2016: segment of the word $h_i^{\e_i}$ of uniquely determined length $r_i$
2017: such that $0\le r_i<l(h_i)$. We define the {\em left boundary marking
2018: of rank} $s$ of $C$ to be the sequence
2019: \[p_0,p_1,\e_0,\e_1,r_0,r_1.\]
2020: The {\em left boundary marking
2021: of rank} $s$ of an extended 2--corridor $E$ is defined to be the
2022: left boundary marking of rank $s$ of its foundation.
2023:
2024: Now assume that $C$ above is a maximal simple (designated) $j$--corridor. Let
2025: $0<p<q< c$ and
2026: consider the sub--corridor $C[p,q]$
2027: of
2028: $C$
2029: of width $p-q\le c-1$.
2030: % (and $D^\prime =[0,c^\prime]\+[0,1]$).
2031: %Then the left boundary marking of rank $s$ of $C$ and of $C^\prime$
2032: %are, by definition, the same, for $s=0,\ldots, c-1$.
2033: Suppose that the left boundary marking $p_0,p_1,\e_0,\e_1,r_0,r_1$ of
2034: rank $p$ of $C$ is
2035: equal to the left boundary marking of rank $q$ of $C$. Let
2036: $\G_{D}(p)$ have base $D_p(x_0,x_1,y_0,y_1)$ and
2037: $\G_{D}(q)$ have base
2038: $D_{q}(x^\prime_0,x^\prime_1,y^\prime_0,y^\prime_1)$
2039: and let $I_i$ be the sub--interval
2040: \[I_i=\left[p+\frac{1}{x_i+1},q+\frac{1}{x^\prime_i+1}\right]\+\{i\}\]
2041: of $[0,c]\+\{i\}$, for $i=0,1$ (see Figure \ref{Z-canc}).
2042: \begin{figure}
2043: \psfrag{(p+1/(x0+1),0)}{$\left(p+\frac{1}{x_0+1},0\right)$}
2044: \psfrag{(p+1/(x1+1),1)}{$\left(p+\frac{1}{x_1+1},1\right)$}
2045: \psfrag{(q+1/(x1+1),1)}{$\left(q+\frac{1}{x^\prime_1+1},1\right)$}
2046: \psfrag{(q+1/(x0+1),0)}{$\left(q+\frac{1}{x_0^\prime+1},0\right)$}
2047: \psfrag{Dp}{$D_p$}
2048: \psfrag{Dq}{$D_q$}
2049: \psfrag{q+1}{$q+1$}
2050: \psfrag{I0}{$I_0$}
2051: \psfrag{I1}{$I_1$}
2052: \psfrag{p}{$p$}
2053: \psfrag{q}{$q$}
2054: \begin{center}
2055: { \includegraphics[scale=0.6]{fig/Z-canc.eps}
2056: }\\
2057: \caption{}\label{Z-canc}
2058: \end{center}
2059: \end{figure}
2060: Then the label of
2061: $\mu(I_i)$ read in the direction $\zeta_i$ is a cyclic permutation of
2062: $(h^{\e_i})^{\al_i}$,
2063: for some positive integer $\al_i$.
2064: Assume further that the left
2065: interior markings of $\G_{D}(p)$ and $\G_{D}(q)$ are
2066: equal (so in particular $y_0=y_0^\prime$). Under these circumstances we
2067: form a new picture $\G^\prime$ from $\G$ as follows.
2068: Let $T:\RR^2\maps\RR^2$ be the translation defined by
2069: \begin{equation}\label{T-translation}
2070: T(x,y)=(x-(q-p),y), \textrm{ for }(x,y)\in \RR^2.
2071: \end{equation}
2072: Let $S$ be a homeomorphism
2073: $S:D_p\maps [p,q+1]\+[0,1]$ such that $S$ fixes \[\{p\}\+[0,1]\cup
2074: [p,p+\frac{1}{x_0+1}]\+\{0\}
2075: \cup [p,p+\frac{1}{x_1+1}]\+\{1\}\] and $S((p+1,t))=(q+1,t)$, for all
2076: $t\in [0,1]$. Define $D^l=[0,p]\+[0,1]$ and
2077: $D^r=[p+1,c-(q-p)]\+[0,1]$.
2078: %and (as usual) $D^\prime_p=[p,p+1]\+[0,1]$.
2079: Consider the triple $C^\prime=(D^\prime,\G_{D^\prime},\mu^\prime)$
2080: where $D^\prime=[0,c-(q-p)]\+[0,1]$,
2081: \[\G_{D^\prime}(i)=\left\{
2082: \begin{array}{ll}
2083: \G_D(i) & \textrm{ if } 0\le i\le p-1\\
2084: T(\G_D(i+(q-p))) & \textrm{ if } p\le i\le c-(q-p)-1
2085: \end{array}
2086: \right.
2087: ,
2088: \]
2089: $\mu^\prime|_{D^l}=\mu|_{D^l}$, $\mu^\prime|_{D^r}=(\mu\circ
2090: T^{-1})|_{D^r}$
2091: and $\mu^\prime|_{D^\prime_p}=(\mu\circ S)|_{D^\prime_p}$.
2092: %
2093: %\left\{
2094: %\begin{array}{ll}
2095: %\mu|_{D_i} & \textrm{ if } 0\le i\le p-1\\
2096: %\mu|_{D_{i+(q-p)}} & \textrm{ if } p+1\le i\le c-(q-p)-1
2097: %\end{array}
2098: %\right.
2099: %.
2100: %\[
2101: Then $\G_{D^\prime}$ is a sum of square--pictures with left--trivial
2102: labelling on the disk $D^\prime$.
2103: Remove $\mu(\G_D)$ from $\G$ and replace it with
2104: $\mu^\prime(\G_{D^\prime})$. Under the given
2105: hypotheses this results in a picture $\G^\prime$ over $G$ on
2106: $\S$. The triple $C^\prime$ is, by construction, a maximal simple corridor
2107: of $\G^\prime$. Furthermore
2108: $\mu([p,q+1]\+[0,1])=\mu^\prime([p,p+1]\+[0,1])$ and $\G\cap (\S\bl \mu([p,q+1]\+[0,1]))=
2109: \G^\prime\cap (\S\bl \mu^\prime ([p,p+1]\+[0,1]))$. Hence all
2110: corridors of $\G$, except $C$ are also corridors of $\G^\prime$.
2111: We
2112: designate $C^\prime$ as the representative for its equivalence class of
2113: maximal corridors in $G^\prime$.
2114: %%%%%%%%%%%%%%%%cancel-extension in here
2115: %Moreover if $C$ is the foundation
2116: %of a designated maximal extended 2--corridor of $\G$ then $C^\prime$
2117: %is chosen as the foundation of the corresponding maximal extended
2118: %2--corridor of $\G^\prime$.
2119: %%%%%%%%%%%%%%%%%
2120:
2121: Now assume that $C$ is the foundation of a designated maximal extended $2$--corridor $E$
2122: with extension $C_e$. For example the graphs $\bar C$ and $\bar C_e$ may be as shown in
2123: Figure \ref{diag_cancel_1}, where the vertical edges $e_1,\ldots ,e_n$ are composed of internal--lateral
2124: arcs of compartments of $C$ and the diagonal edges $d_1,\ldots ,d_n$ are composed of internal--lateral
2125: arcs of compartments of $C_e$. Thus $u$ is the vertex of $C$ not in $C_e$ and $v$ is the vertex of
2126: $C_e$ not in $C$.
2127: \begin{figure}
2128: \psfrag{u}{$u$}
2129: \psfrag{v}{$v$}
2130: \psfrag{e1}{$e_1$}
2131: \psfrag{en}{$e_n$}
2132: \psfrag{d1}{$d_1$}
2133: \psfrag{dn}{$d_n$}
2134: \psfrag{a}{$a$}
2135: \psfrag{b}{$b$}
2136: \psfrag{a'}{$a^\prime$}
2137: \psfrag{b'}{$b^\prime$}
2138: \psfrag{p}{$p$}
2139: \psfrag{q}{$q$}
2140: \psfrag{p'}{$p^\prime$}
2141: \psfrag{q'}{$q^\prime$}
2142: \psfrag{...}{$\cdots$}
2143: \begin{center}
2144: { \includegraphics[scale=0.34]{fig/diag_cancel_1.eps}
2145: }\\
2146: \caption{}\label{diag_cancel_1}
2147: \end{center}
2148: \end{figure}
2149:
2150: Cancellation from $p$ to $q$ along $C$ results in corridors $C^\prime$ and $C^\prime_e$
2151: which have graphs $\bar C^\prime$ and $\bar C^\prime_e$, as shown in Figure \ref{diag_cancel_2}.
2152: \begin{figure}
2153: \psfrag{u}{$u$}
2154: \psfrag{v}{$v$}
2155: \psfrag{e1}{$e_1$}
2156: \psfrag{en}{$e_n$}
2157: \psfrag{d1}{$d_1$}
2158: \psfrag{dn}{$d_n$}
2159: \psfrag{a}{$a$}
2160: \psfrag{b}{$b$}
2161: \psfrag{a'}{$a^\prime$}
2162: \psfrag{b'}{$b^\prime$}
2163: \psfrag{p}{$p$}
2164: \psfrag{q}{$q$}
2165: \psfrag{p'}{$p^\prime$}
2166: \psfrag{q'}{$q^\prime$}
2167: \psfrag{...}{$\cdots$}
2168: \begin{center}
2169: { \includegraphics[scale=0.34]{fig/diag_cancel_2.eps}
2170: }\\
2171: \caption{}\label{diag_cancel_2}
2172: \end{center}
2173: \end{figure}
2174: The compartments of $C_e$ containing $a^\prime$ and all those to its left become compartments of
2175: $C_e^\prime$, as does the compartment of $C_e$ containing $b^\prime$ and all those to its right.
2176: In addition $C^\prime_e$ has one compartment containing $a$ and $q^\prime$ which replaces all the
2177: compartments of $C_e$ between those containing $a^\prime$ and $b^\prime$. Therefore the only compartment
2178: of $C^\prime_e$ which is not a compartment of $C_e$ is that containing vertices $a$ and $q^\prime$. Note
2179: that in $C$ the left interior markings of the compartments containing $p$ and $q$ are equal. It follows
2180: that the left interior markings of the compartments of $C_e$ and $C_e^\prime$ containing $a$ are equal.
2181: From the graph it is clear that $C^\prime$ and $C^\prime_e$ are diagonal and, since $C$ and $C_e$ are maximal
2182: so are $C^\prime$ and $C^\prime_e$. Thus $E^\prime=C^\prime+C^\prime_e$ is a maximal extended $2$--corridor.
2183: We choose $E^\prime$ as the designated representative of its class in $\G^\prime$ and $C^\prime$ and $C^\prime_e$
2184: as its foundation and extension, respectively.
2185: %%%%%%%%%%%%%%%%%%%%
2186:
2187: The formation of $\G^\prime$ from $\G$ along a corridor $C$ in this way is called $Z$--{\em
2188: cancellation along} $C$ {\em from} $p$ to $q$. If $C$ is of type $j$
2189: we refer to $Z(j)$--cancellation along $C$, $j=0,1,2$.
2190: We remark that if $\G^\prime$ is formed from $\G$ by
2191: $Z$--cancellation then the left interior and left boundary markings of
2192: compartments of corridors of $\G^\prime$ are
2193: left interior and left boundary markings of
2194: compartments of corridors of $\G$.
2195: Also, the boundary partition $\mbf b$ is unaffected by
2196: $Z$--cancellation, the prime labels of boundary intervals are the same
2197: in $\G^\prime$ as in $\G$
2198: and for $j\neq p_i$ the boundary interval $b_j$ has the
2199: same label in $\G^\prime$ as in $\G$. With the above notation, the
2200: label of $P_i=b_{p_i}$ in
2201: $\G^\prime$ is $h_i\^(m_i-l(h_i)\al_i)$, for $i=0,1$.
2202:
2203: \begin{theorem}\label{z-canc-poss}
2204: If $C$ is a $j$--corridor, for $j=0,1$ or 2 and $C$ has more than
2205: $M(j)$ arcs then $Z$--cancellation is possible along some
2206: sub--corridor of $C$. ($M(j)$ is defined in (\ref{M_j}).)
2207: \end{theorem}
2208: \medskip
2209:
2210: \noindent{\em Proof.}
2211: Assume first that $C$ is a simple corridor of width $c$,
2212: $C=(D,\G_D,\mu)$. Then $Z$--cancellation is possible if
2213: there are integers $q_0,q_1$, with $0< q_0<q_1<c$, such that
2214: the left interior marking of $C$ of rank $q_0$ equals the left interior
2215: marking of $C$ of rank $q_1$ and
2216: the boundary marking of $C$ of rank $q_0$ equals the boundary
2217: marking of $C$ of rank $q_1$.
2218:
2219: Let $p_0,p_1,\e_0,\e_1,r_0,r_1$ be the left boundary marking of rank
2220: $q_0$ and let $(h_{p_i},m_{p_i})$ be the prime label of the boundary
2221: interval $b_{p_i}$, for $i=0,1$. Since $p_i$ and $\e_i$ are determined
2222: by $C$ then, for all $q_1$ with $0<q_1<c$, the left boundary
2223: marking of rank $q_1$ is $p_0,p_1,\e_0,\e_1,s_0,s_1$, for some
2224: integers $s_0$ and $s_1$. By definition $0\le r_i,s_i<l(h_i)\le
2225: W_2(\mbf z)$. Hence there are at most $W_2^2$ distinct left boundary
2226: markings of $C$. If $j=0$ the left interior
2227: markings of all ranks of $C$ are identical and the number of arcs of
2228: $C$ is $c$. If $c>W_2^2+1=M(0)$ then from the above it follows that
2229: $Z$-cancellation along $C$ is possible, as claimed.
2230:
2231: If $j=1$ then the left interior marking of rank $s$ of $C$ is
2232: $\de,o,\rho,\s,0,0,0,0$, where
2233: $\de\in\{-1,0,1\}$, $o,\s\in\{0,\ldots ,l(r)-1\}$, $0\le \rho<m$ and $\de=o=\rho=\s=0$ if and
2234: only if $\G_D(s)$ is of type square--$0$. Therefore there are at most
2235: $3ml(r)^2$ distinct left interior markings of $C$. It follows that if
2236: $C$ has width $c>W_2^2\cdot 3ml(r)^2+1$ then $Z$--cancellation along $C$
2237: is possible. A square--picture of type 0 has 1 arc while a
2238: square--picture of type 1 has $ml(r)$ arcs. Hence if $C$ has more that
2239: $(W_2^2\cdot 3ml(r)^2+1)ml(r)\le M(1)$ arcs then $Z$--cancellation along $C$
2240: is possible.
2241:
2242: If $j=2$ then the left interior marking of rank $s$ of $C$ is
2243: $\de_0,o_0,\rho_0,\s_0,\de_1,o_1,\rho_1,\s_1$, where
2244: $\de_i=\pm 1$, $o_i,\s_i\in\{0,\ldots ,l(r)-1\}$ and $0\le \rho_i<m$.
2245: Therefore there are at most
2246: $4m^2l(r)^4$ distinct left interior markings of $C$.
2247: It follows that if
2248: $C$ has width $c>W_2^2\cdot 4m^2l(r)^4+1$ then $Z$--cancellation along $C$
2249: is possible. A square--picture of type 2 has less than $2ml(r)$ arcs
2250: so if $C$ has more that
2251: $(W_2^2\cdot 4m^2l(r)^4+1)ml(r)$ arcs then $Z$--cancellation along $C$
2252: is possible.
2253:
2254: If $C$ is not simple then $j=2$ and $C$ has foundation a simple
2255: 2--corridor $C^f$. Then $C$ has at most $ml(r)$ more arcs than
2256: $C^f$. Thus if $C$ has more than $(W_2^2\cdot 4m^2l(r)^4+2)ml(r)=M(2)$
2257: arcs then, from the above $Z$--cancellation is possible along $C^f$
2258: and therefore along $C$, as claimed.
2259: \medskip
2260:
2261: From this Theorem it follows that a $Z$--cancellation along a
2262: $j$--corridor may be achieved
2263: by a sequence of $Z$--cancellations along sub--corridors of width at
2264: most $M(j)$.
2265:
2266: \begin{defn}\label{z_reduced}
2267: A picture on which no $Z$--cancellation is possible is called $Z${\em --reduced}.
2268: \end{defn}
2269:
2270: \begin{corol}\label{corridor-size}
2271: If $C$ is a $j$--corridor of a $Z$--reduced picture then $C$ has at
2272: most $M(j)$ arcs.
2273: \end{corol}
2274:
2275: \begin{lemma}\label{canc-effic}
2276: Let $\G$ be a picture on a surface $\S$ of
2277: type $(\mbf n,\mbf t,\mbf p)$ with prime
2278: labels $\mbf z$, labelled by $\hat\al(\mbf z)$ and with boundary partition
2279: $\mbf b=(b_1,\ldots ,b_{W_1})$. Assume that $\G^\prime$ is obtained from $\G$
2280: by $Z$--cancellation. Then $\G^\prime$ is reduced if and only if $\G$ is reduced. Furthermore
2281: $\G^\prime$ is reduced and minimalistic if and only if $\G$ is reduced and minimalistic.
2282: \end{lemma}
2283: \medskip
2284:
2285: \noindent{\em Proof.} Suppose first that $\G$ is reduced.
2286: If $\G^\prime$ has a pair of cancelling vertices
2287: $u,v$ then these must be adjacent in $\G^\prime$ but not in
2288: $\G$. Hence $u,v$ belong to consecutive compartments of some corridor
2289: $C=(D,\G_D,\mu)$ of $\G^\prime$
2290: which results from
2291: $Z$--cancellation in $\G$. It follows
2292: that $C$ must have type $j=1$ or $2$. As $u$ and $v$ cancel in
2293: $\G^\prime$ they are the images of vertices which
2294: cancel in $\G_D$; so we restrict attention to $\G_D$ and assume $u$
2295: and $v$ belong to $\G_D$. Suppose $u$ belongs to $\G_D(p)$, $v$
2296: belongs to $\G_D(p+1)$ and that $D_p$ and $D_{p+1}$ have bases
2297: $D_p(x_0,x_1,y_0,y_1)$ and
2298: $D_p(x_0^\prime,x_1^\prime,y_1,y_1^\prime)$, respectively. If $j=1$
2299: then it must be possible to perform a bridge move on arcs
2300: joining $u$ and $v$ to $[p,p+1]\+\{i\}$, for $i=0$ or 1. Assume that
2301: this occurs for $i=0$ and let $a_0$ and $a_1$ be the arcs joining $u$
2302: to $q_0=(p+1-\frac{x_0}{1+x_0},0)$ and $v$ to
2303: $q_1=(p+1+\frac{1}{1+x^\prime_0},0)$, respectively: then a bridge move
2304: is possible transforming $a_0$ and $a_1$ into arcs $b_0$, joining $u$
2305: to $v$, and $b_1$ with end points $q_0$ and $q_1$. However this
2306: means that the interval $[q_0,q_1]$ of $\pd D$ has trivial label,
2307: contrary to the standing hypothesis that $\mbf z$ is special. Hence if
2308: $\G_D$ is of type 1 no such cancellation is possible.
2309:
2310: Now assume that $j=2$. If $u$ and $v$ are both incident
2311: to $[p,p+2]\+\{i\}$ then, as in the case $j=1$, no bridge move is
2312: possible on arcs joining $u$ and $v$ to $\pd D$. Hence $u$ and $v$ do
2313: not cancel. We assume therefore that $u$ is incident to
2314: $[p,p+1]\+\{0\}$ and that $v$ is incident to
2315: $[p+1,p+2]\+\{1\}$. However no bridge moves are possible now because
2316: both regions incident to the edge of $\ovr{\G}_D$ joining $u$ to $v$
2317: are triangular: a bridge move on the sides of a triangle would result
2318: in a trivial label on the corner of the triangle opposite the edge joining $u$
2319: to $v$. Thus no cancellation is possible if $j=2$.
2320: It follows that if $\G$ is reduced then $\G^\prime$ is reduced.
2321: A similar argument shows that the converse also holds.
2322:
2323: Now suppose that $\G$ is reduced and minimalistic.
2324: Then $\G^\prime$ is reduced.
2325: Regions involved in $Z$--cancellation all have Euler characteristic $1$
2326: and both $\G$ and $\G^\prime$ are pictures on the surface $\S$, so the
2327: condition on Euler characteristic of regions is satisfied by $\G^\prime$.
2328: No arc of $\G^\prime$ is a closed curve since no such arcs
2329: are created in the process of $Z$--cancellation.
2330: %(or $G$ is
2331: %not of type $E(2,*,m)$ and $\G^\prime$ has an internal edge of width greater than
2332: %$l-2$) then, as in the proof of Lemma \ref{bound_edge_width},
2333: %the vertices incident to this edge cancel.
2334: Suppose that $\G^\prime$ has interior edge of width greater than $l-2$.
2335: Since $\G^\prime$ is reduced it follows from Lemma \ref{bound_edge_width}
2336: that both ends of this edge are incident to the same vertex. No such edge
2337: results from the process of $Z$--cancellation so this edge also belongs
2338: to $\G$, which cannot therefore be minimalistic. It follows that
2339: $\G^\prime $ is reduced and minimalistic.
2340: The converse follows similarly.
2341:
2342: \subsection{Corridor--sections}\label{corridor-section}
2343:
2344: Let $\G_D$ be a sum of square--pictures of type $j$ on the disk $D=[0,c]\+[0,1]$
2345: and let $w_i$ denote the label of $[0,c]\+\{i\}$, for $i=0,1$, $j=0,1$
2346: or $2$.
2347: Assume that the following conditions hold.
2348: \be
2349: \item There exist integers $p_0,p_1, \e_0,\e_1,t_0$ and $t_1$ such that $1\le p_i\le
2350: W_1$, the $p_i$th letter $(h_i,f_i)$ of $\mbf z$ is not %an $(\cL,r)$--letter,
2351: a degenerate letter or
2352: a minor letter of $H^\Lm$, $\e_i=\pm 1$, $0\le
2353: t_i<l(h_i)$ and $w_i=g_i^{\al_i}$, for some $\al_i>0$, and
2354: $g_i=\t(h_i^{\e_i},t_i)\io(h^{\e_i},l(h_i)-t_i)$, for
2355: $i=0,1$.
2356: \item $\G_D$ has at most $M(j)$ arcs.
2357: \item Let $D^e$
2358: be the disk $[c,c+1]\+[0,1]$.
2359: \begin{comment}
2360: and let the base of $\G_D(0)$ is $D_0[x_0,x_1,y_0,y_1]$. Then the base of
2361: $\G_D(c-1)$ is $D_{c-1}[x_0^\prime, x_1^\prime, y_0^\prime,y_0]$ and there
2362: exists a
2363: picture $\G_{D^e}$ of type square--$j$ on $D^e$, with base
2364: $D^e[x_0^\pprime,x_1^\pprime,y_0,y_1^\pprime]$ and left interior marking equal
2365: to that of $\G_D(0)$,
2366: such that
2367: \end{comment}
2368: Let $T_c$ be the translation $T(x,y)=(x+c,y)$, for $(x,y)\in \RR^2$, and
2369: let $\G_{D^e}=T_c(\G_D(0))$, a picture of type square--$j$ on $D_e$. Then
2370: $\G_D\cup \G_{D^e}$ is a sum of square--pictures
2371: on the disk $[0,c+1]\+[0,1]$.
2372: \ee
2373: Then $(\G_D,p_0,p_1,\e_0,\e_1,t_0,t_1)$
2374: is called a {\em corridor--section} on $\G_D$, of width $c$, binding
2375: the $p_0$th and $p_1$th letters of $\mbf z$, and of {\em type} $j$ if
2376: $\G_D$ is a sum of square--pictures of type $j$.
2377:
2378: %Suppose that $(\G_D,p_0,p_1,\e_0,\e_1,t_0,t_1)$ is a corridor--section.
2379: %, binding
2380: %the $p_0$th and $p_1$th letters of $\mbf z$.
2381: % and that $\G_D(0)$ is a
2382: %square--picture of type $j$.
2383: The corridor--section $(\G_D,p_0,p_1,\e_0,\e_1,t_0,t_1)$ is said to have
2384: {\em left interior marking} equal to the left interior marking of $\G_D$,
2385: {\em left boundary marking}
2386: %(binding the $p_0$th and $p_1$th letters of $\mbf z$)
2387: the sequence
2388: \[p_0,p_1,\e_0,\e_1,t_0,t_1\] and
2389: {\em extent}
2390: %(on $p_0$ and $p_1$)
2391: the pair$(\al_0,\al_1)$.
2392:
2393: Let $C$ be a corridor section of width $c$.
2394: The conditions on a corridor--sections imply that if $0< p<c-1$ then a new
2395: corridor--section $C_p$ may be formed by cutting $C$ along $\{p\}\+[0,1]$ and then
2396: identifying $(0,t)$ with $(c,t)$, for all $t\in [0,1]$. We set $C=C_0$.
2397: The left interior and left
2398: boundary markings of $C_p$ are called the {\em left interior} and {\em left boundary}
2399: markings, respectively, of $C$ of {\em rank} $p$. Thus the left interior marking and
2400: left boundary marking of $C$ coincide with the left interior and boundary markings
2401: of rank 0 of $C$.
2402:
2403: Let $L$ denote the set of all corridor sections. As the number of arcs
2404: of a corridor section is bounded $L$ is finite. Let $L$ have $|L|$
2405: elements indexed by the integers $1,\ldots, |L|$ and denote these
2406: elements $L(1),\ldots,L(|L|)$. For $i=1,\ldots, |L|$, define the quintuple
2407: \[L_5(i)=(L(i),s(i),t(i),\al(i,0),\al(i,1))\] where $s(i)$ and $t(i)$ are
2408: the left boundary marking and left interior marking, respectively, of
2409: $L(i)$ and $(\al(i,0),\al(i,1))$ is the extent of $L(i)$.
2410: Furthermore write
2411: \[s(i)=p(i,0),p(i,1),\e(i,0),\e(i,1),r(i,0),r(i,1)\]
2412: for the left boundary marking of $L(i)$.
2413: For fixed $i$ with $1\le i\le |L|$ and each integer $p$ with $1\le p\le W_1$ define
2414: \[
2415: \zeta(p,i)=\left\{
2416: \begin{array}{ll}
2417: 0,& \textrm{ if } p\neq p(i,0) \textrm{ and } p\neq p(i,1)\\
2418: \al(i,0), & \textrm{ if } p=p(i,0) \textrm{ and } p\neq p(i,1)\\
2419: \al(i,1), & \textrm{ if } p\neq p(i,0) \textrm{ and } p=p(i,1)\\
2420: \al(i,0)+\al(i,1), & \textrm{ if } p=p(i,0)=p(i,1)
2421: \end{array}
2422: \right.
2423: .
2424: \]
2425:
2426:
2427: The following lemma follows directly from the
2428: definitions above and the remarks preceding Theorem
2429: \ref{z-canc-poss}.
2430: \begin{lemma}\label{corr-sec-remove}
2431: Let $C$ be a $j$--corridor of $\G$%, binding boundary intervals $b_{p_0}$ and
2432: %$b_{p_1}$,
2433: and let $C[p,q]=(D,\G_D,\mu)$ be a sub--corridor (of
2434: width $q-p$) with at most $M(j)$ arcs.
2435: If $Z$--cancellation takes place along $C$ from $p$ to
2436: $q$ and the left boundary marking of $C$ of rank $p$ is $t$
2437: then $(\G_D,t)$ is a corridor--section of width $q-p$ and
2438: type $j$. Let $(\G_D,t)=L(i)$, for some $i$ with $1\le
2439: i\le |L|$, let $\G^\prime $ be the picture formed by this
2440: $Z$--cancellation and let
2441: the boundary interval $b_p$ of $\G$ have label
2442: %$p$th $H^\Lm$ letter of $\mbf z$ be
2443: %$(h_p,m_p)$
2444: $h_p\^m_p$, for $p=1,\ldots,W_1$. Then the boundary interval of $b_p$
2445: in $\G^\prime$ has label
2446: \[h_p\^(m_p-l(h_p)\zeta(p,i)),\]
2447: for $p=1,\ldots,W_1$.
2448: \end{lemma}
2449:
2450:
2451: The $Z$--cancellation of Lemma \ref{corr-sec-remove}, along $C$ from $p$ to
2452: $q$, where the left boundary marking, of rank $p$, of $C$ is $t$ is
2453: called $Z$--cancellation of corridor--section $(\G_D,t)$. As remarked
2454: (above Definition \ref{z_reduced}) any $Z$--cancellation may be effected by a sequence of
2455: $Z$--cancellations of corridor--sections.
2456:
2457: \subsection{$Z$--insertion}
2458: Let $C=(D,\G_D,\mu)$ be a maximal simple $j$--corridor, with
2459: $D=[0,c]\+[0,1]$, binding $b_{p_0}$ and $b_{p_1}$, for $p_i$ with
2460: $1\le p_i\le W_1$. Let the left interior marking and left boundary marking of
2461: of $C$, of rank $p$, be $s$ and $t$, respectively.
2462: Let $(\G_\D,t)$ be a corridor section with left interior marking $s$, where $\D=[0,q]\+[0,1]$.
2463: Let $T_p:\RR^2\maps\RR^2$ and $T_q:\RR^2\maps\RR^2$ be the translations defined by
2464: $T_p(x,y)=(x+p,y)$ and $T_q(x,y)=(x+q,y)$, for $(x,y)\in \RR^2$ and positive integers $p$ and $q$.
2465: Let $D_p$ have base $D_p(x_0,x_1,y_0,y_1)$ and $\D_0$ have base
2466: $\D_0(x_0^\prime,x_1^\prime,y_0^\prime,y_1^\prime)$. Let $S$ be a homeomorphism
2467: $S:[p,p+q+1]\+[0,1]\maps [p,p+1]\+[0,1]$ such that $S$ fixes
2468: $\{p\}\+[0,1]$,
2469: \[S([p,p+\frac{1}{x^\prime_i+1}]\+\{i\})=[p,p+\frac{1}{x_i+1}]\+\{i\},\]
2470: for $i=0,1$,
2471: and $S((p+q+1,t))=(p+1,t)$, for all
2472: $t\in [0,1]$. Define $D^l=[0,p]\+[0,1]$, $D^c=[p,p+q+1]\+[0,1]$ and
2473: $D^r=[p+q+1,c+q]\+[0,1]$.
2474:
2475: Consider the triple $C_I=(D_I,\G_{D_I},\mu_I)$, where
2476: $D_I=[0,c+q]\+[0,1]$,
2477: \[
2478: \G_{D_I}(i)=\left\{
2479: \begin{array}{ll}
2480: \G_D(i) & \textrm{ for } i=0,\ldots , p-1\\
2481: T_p(\G_{\D}(i-p)) &\textrm{ for } i=p,\ldots ,p+q-1\\
2482: T_q(\G_{D}(i-q)) &\textrm{ for } i=p+q,\ldots, c+q-1
2483: \end{array}
2484: \right.,
2485: \]
2486: $\mu_I|_{D^l}=\mu|_{D^l}$,
2487: $\mu_I|_{D^c}=(\mu\circ S)|_{D^c}$ and
2488: $\mu_I|_{D^r}=(\mu\circ T^{-1}_q)|_{D^r}$.
2489:
2490: Then $\G_{D_I}$ is a sum of square--pictures with left--trivial
2491: labelling on the disk $D_I$.
2492: %Form a new picture $\G_I$ from $\G$ as follows.
2493: Remove $\mu(\G_D)$ from $\G$ and replace it with
2494: $\mu_I(\G_{D_I})$. This results on a picture $\G_I$ over $G$ on
2495: $\S$. The triple $C_I$ is, by construction, a maximal corridor
2496: of $\G_I$. Furthermore
2497: $\mu([p,p+1]\+[0,1])=\mu_I([p,p+q+1]\+[0,1])$ and $\G\cap (\S\bl \mu([p,p+1]\+[0,1]))=
2498: \G_I\cap (\S\bl \mu_I([p,p+q+1]\+[0,1]))$. Hence all
2499: corridors of $\G$, except $C$, are also corridors of $\G_I$.
2500: We
2501: designate $C_I$ as the representative for its equivalence class of
2502: maximal corridors in $\G_I$. Moreover if $C$ is the foundation
2503: of a designated maximal extended 2--corridor of $\G$ then $C_I$
2504: is a maximal simple sub--corridor of a maximal extended $2$--corridor of
2505: $\G_I$ (as can be seen by reversing the argument for cancellation along
2506: extended $2$--corridors). In this case $C_I$
2507: is chosen as the foundation of the corresponding maximal extended
2508: 2--corridor of $\G_I$, which becomes a designated $2$--corridor of $\G_I$.
2509:
2510: The formation of $\G_I$ from $\G$ in this way is called $Z$--{\em
2511: insertion of} $(\G_\D,t)$ {\em into} $C$ {\em at position} $p$. If $C$ is of type $j$
2512: we refer to $Z(j)${\em --insertion}, $j=0,1,2$.
2513: %We remark that
2514:
2515: \begin{lemma}\label{one-insertion}
2516: If $\G_I$ is formed from $\G$ by
2517: $Z$--insertion of $(\G_\D,t)$ then the left interior and left boundary markings of
2518: compartments of corridors of $\G_I$ are
2519: left interior and left boundary markings of square--pictures of $\G_\D$ or of
2520: compartments of corridors of $\G$.
2521: Also, the boundary partition $\mbf b$ is unaffected by $Z$--insertion
2522: and, given that
2523: the boundary interval $b_p$ of $\G$ has label
2524: $h_p\^m_p$, for $p=1,\ldots,W_1$, then the boundary interval of $b_p$
2525: in $\G_I$ has label
2526: \[h_p\^(m_p+l(h_p)\zeta(p,i)),\]
2527: for $p=1,\ldots,W_1$.
2528: \end{lemma}
2529:
2530: The operations of $Z$--cancellation and $Z$--insertion are inverse to
2531: each other in an obvious way. To be more precise, let $\G_I$ be formed
2532: from $\G$
2533: by $Z$--insertion of $(\G_\D,t)$ into $C$ at position $p\ge 1$ and let the
2534: resulting corridor of $\G_I$ be $C_I$ as above. Then $Z$--cancellation
2535: of corridor--section $(\G_\D,t)$ along $C_I$ from $p$ to $q$ is
2536: possible and results in $\G$. A similar statement holds starting with
2537: $Z$--cancellation.
2538: \subsection{The $Z$-graph}
2539:
2540: Let $L$ denote the set of corridor--sections of
2541: $\mbf z$
2542: (indexed by the integers $1,\ldots ,|L|$ as before)
2543: and %, with the notation of Section \ref{corridor-section},
2544: define the {\em left marking}, of {\em rank} $p$, of $L(i)\in L$ to be
2545: $m(i,p)=s(i,p),t(i,p)$, where $s(i,p)$ and $t(i,p)$ are the left interior and boundary
2546: markings, respectively, of rank $p$ of $L(i)$. Define
2547: the {\em left marking} of $L(i)$ to be $m(i)=m(i,0)$,
2548: let \[m_{L(i)}=\{m(i,p): 0\le p< \textrm {width of } L(i)\}\]
2549: and let
2550: \[m_{\mbf z}=\bigcup_{i=1}^{|L|}m_{L(i)}.\]
2551:
2552: Let $C$ be a corridor of the picture $\G$ and define the {\em left marking},
2553: of rank $p$,
2554: of $C$ to be $s,t$, where $s$ and $t$ are the left interior and
2555: boundary markings, respectively, of rank $p$, of $C$.
2556: Define the {\em active} left markings of $\G$ to be those left markings of
2557: corridors of $C$ which are elements of $m_{\mbf z}$.
2558:
2559: The $Z${\em --graph} $G(\mbf z)$ of $\mbf z$ is defined as follows.
2560: The vertices of $G(\mbf z)$ are
2561: the subsets of $m_{\mbf z}$. There is one directed edge labelled $L(i)$
2562: from vertex $R$ to vertex $T$ corresponding to each corridor--section $L(i)$ such that
2563: $L(i)$ has left marking in $R$ and $T=R\cup m_{L(i)}$. Corresponding to the picture
2564: $\G$ is the vertex
2565: \[v(\G)=\{\textrm{active left markings of } \G\}\]
2566: of $G(\mbf z)$ which we call the {\em initial} vertex of $\G$.
2567:
2568: Now, for $i=0,\ldots n$, let $\G_i$ be a picture, over $G$ on $\S$, such
2569: that $\G_{i+1}$ is obtained from $\G_i$ by $Z$--cancellation of a
2570: corridor--section $C_i$ along a
2571: corridor of $\G_i$. Assume that $\G=\G_n$.
2572: %and that if $C_i$ has type $j$
2573: %then it has at most $M(j)$ arcs.
2574: Then this sequence of $Z$--cancellations uniquely
2575: determines a path $S_n,\ldots,S_0$ in $G(\mbf z)$, where $S_i\in
2576: V(G(\mbf z))$, $S_n=v(\G)$, $S_{i}=S_{i+1}\cup \{\textrm{left markings
2577: of } C_i\}$ and the edge of the path
2578: joining $S_{i+1}$ to $S_{i}$ is labelled $C_i$. Then $\G_i$ is
2579: obtained from $\G_{i+1}$ by $Z$--insertion of $C_i$. Hence the set of active left
2580: markings of $\G_{i}$ is the union of the left markings of $C_i$ with
2581: the active left markings of $\G_{i+1}$. It follows, by induction, that
2582: the set of active left markings of $\G_0$ is $S_0$.
2583:
2584: Conversely, let $S_n,\ldots,S_0$ be a directed path in $G(\mbf z)$,
2585: where $S_n=v(\G)$ and the edge of the path joining $S_{i+1}$ to $S_{i}$
2586: is labelled $C_i$. Then $S_{i}=S_{i+1}\cup \{\textrm{left markings
2587: of } C_i\}$. Let $\G_n=\G$ and let $m$ be an integer with $n\ge
2588: m>0$. Suppose inductively that we have
2589: constructed pictures $\G_i$ with active left markings $S_i$, for
2590: $i=n,\ldots , m$ such that $\G_{i+1}$ is obtained from $\G_{i}$ by
2591: $Z$--cancellation of the corridor--section $C_i$, for $i=n-1,\ldots m$.
2592: As $C_{m-1}$ is the label of the edge joining $S_{m}$ to $S_{m-1}$ it
2593: follows that $S_{m-1}=S_{m}\cup\{\textrm{left markings of }C_{m-1}\}$ and that
2594: the left marking of $C_{m-1}$ is in $S_{m}=\{\textrm{active left
2595: markings of } \G_m\}$. Hence we may form a
2596: picture $\G_{m-1}$ from $\G_m$ by $Z$--insertion of $C_{m-1}$. The
2597: active left markings of $\G_{m-1}$ are then the left markings of
2598: $C_{m-1}$ together with the active left markings of $\G_m$: that is
2599: $S_{m-1}$. Moreover as $\G_{m-1}$ is formed from $\G_m$ by
2600: $Z$--insertion of $C_{m-1}$ it follows that $\G_m$ is formed from
2601: $\G_{m-1}$ by $Z$--cancellation of $C_{m-1}$. Hence the path
2602: $S_n,\ldots,S_0$ determines a sequence of $Z$--cancellations
2603: \[\G_0\maps\cdots\maps\G_n=\G,\]
2604: with $\G_{i+1}$ obtained from $\G_{i}$ by $Z$--cancellation of $C_i$ and
2605: the set of active left markings of $\G_0$ equal to $S_0$.
2606:
2607: We have therefore proved the following.
2608: \begin{lemma}
2609: Let $\G$ and $\G^\prime$ be pictures over $G$ on $\S$. Then there
2610: exists a sequence
2611: \[\G^\prime=\G_0\maps\cdots\maps\G_n=\G,\]
2612: with $\G_{i+1}$ obtained from $\G_{i}$ by $Z$--cancellation of the
2613: corridor--section $C_i$ if and only if there exists a directed path
2614: \[v(\G)=S_n,\ldots,S_0=v(\G^\prime)\] in $G(\mbf z)$ such that the
2615: edge joining $S_{i+1}$ to $S_i$ is the corridor section $C_i$.
2616: \end{lemma}
2617:
2618: \begin{defn}
2619: Let $p=S_n,\ldots,S_0$ be a directed path in $G(\mbf z)$ such that the
2620: edge joining $S_{i+1}$ to $S_i$ is the corridor section $C_i$. Let
2621: \[\G_0\maps\cdots\maps\G_n,\] be a sequence of pictures
2622: over $G$ on $\S$ with $\G_{i+1}$ obtained from $\G_{i}$ by $Z$--cancellation of the
2623: corridor--section $C_i$. If $v(\G_n)=S_n$ then we say that $\G_0$ is
2624: the picture obtained from $\G_n$ by $Z$--insertion {\em following} $p$.
2625: If $v(\G_0)=S_0$ then we say that $\G_n$ is the picture obtained from
2626: $\G_0$ by $Z$--cancellation {\em following} $p$.
2627: \end{defn}
2628:
2629: Let $Q$ denote the quadratic exponential equation $Q(\mbf z,\cL,\mbf n,\mbf t,\mbf
2630: p)$ and $Q_\cH=Q(\ovr{\mbf z},\cH,\mbf n,\mbf t,\mbf
2631: p)$ the homogeneous equation corresponding to $\mbf z$.
2632: Let the $q$th letter of $\mbf z$ be $(h_q,f_q)$, for $1\le q\le W_1$.
2633:
2634: \begin{lemma}\label{follow-path}
2635: Let $p=S_n,\ldots,S_0$ be a directed path in $G(\mbf z)$ with initial
2636: vertex $v(\G_n)$. Let
2637: $e_1,\ldots,e_t$ be the edges of $G(\mbf z)$ occuring in $p$,
2638: let $L(s_i)$ be the label of $e_i$ and let
2639: $n_i$ be the number of occurrences of $e_i$ in $p$, for $i=1,\ldots, t$.
2640: Let $\G^\prime$ be the picture obtained from $\G_n$ by
2641: $Z$--insertion following $p$. Then the
2642: following hold.
2643: \be
2644: \item\label{follow-path-1} Let $\al_n$ be a solution to $\cH$ such that $(\al_n,\G_n)$
2645: corresponds to a solution to $Q_\cH$ and write $m_q=\al_n(\lm_q)$, for $1\le q\le W_1$.
2646: The boundary interval $b_q$ of
2647: $\G^\prime$ has label
2648: \[h_q\^\left(m_q+l(h_q)\left(\sum_{i=1}^{t}n_i\z(q,s_i)\right)\right),\]
2649: for $1\le q\le W_1$.
2650: \item\label{follow-path-2} Let $\al$ be a solution to $\cL$ such that $(\al,\G^\prime)$
2651: corresponds to a solution to $Q$. Let $\al_n:M\maps \ZZ$ be the retraction defined
2652: by setting
2653: \[\al_n(\lm_q)= \al(f_q)-l(h_q)\left(\sum_{i=1}^{t}n_i\z(q,s_i)\right),\]
2654: for all $q\in \{1,\ldots ,W_1\}$ such that the $q$th letter of $\mbf z$ is
2655: a proper exponential $H^\Lm$--letter, and $\al_n(\lm)=0$, for all other $\lm\in\Lm$.
2656: Then $(\al_n,\G_n)$ corresponds to a solution of $Q_\cH$.
2657: \ee
2658: \end{lemma}
2659: \medskip
2660:
2661: \noindent{\em Proof.}
2662: The first part of the Lemma follows from Lemma \ref{one-insertion}.
2663: For the second part note that the boundary label of $b_q$ in $\G_n$ is
2664: $h_q\^\al_n(\lm_q)$, for all $q$ (using Lemma \ref{one-insertion} again).
2665: By definition of $Z$--cancellation therefore $\al_n(\lm_q)>0$. Also
2666: $\al_n(\lm_q)\equiv \al(f_q)\equiv k_q (\mod l(h_q))$.
2667: %If $b_q$ is an
2668: %$r$--boundary interval then $b_q$ is unaffected by $Z$--cancellation
2669: %so $\al_n(\lm_q)=\al(f_q)<ml/2$.
2670: It follows that $(\al_n,\G_n)$
2671: corresponds to a solution to $Q_\cH$.
2672: \medskip
2673:
2674: A picture $\G$ such that there exists a solution $\al$ to $\cH$ and
2675: $(\al,\G)$ corresponds to a solution to $Q_\cH$ will be called a
2676: {\em basic} picture for $(\mbf z,\cL)$.
2677:
2678: %{\bf weed out unnecessary terms from this para}
2679: We call a path in a graph with no repeated vertices a {\em simple
2680: path} and a path with initial and terminal vertices identical but no
2681: other pair of vertices identical a {\em cycle}. A cycle of length 1
2682: is called a {\em loop}. The {\em path--subgraph} of a path is the
2683: subgraph consisting of all edges and vertices of the path.
2684: The directed path $p$ of $G(\mbf z)$ corresponding to a sequence of
2685: $Z$--cancellations contains no cycle of length
2686: greater than 1 but may contain loops.
2687: We call a
2688: directed path $p$ in $G(\mbf z)$ a $Z${\em --path} if $p$
2689: has no cycle of length greater than 1. % and call the subgraph of a
2690: %$Z$--path a $Z${\em --subgraph}. Thus a $Z$--subgraph
2691: The path--graph of a $Z$--path
2692: consists of the
2693: path--subgraph of a directed simple path together with a (possibly
2694: empty) collection of loops. Given a directed path $p$
2695: assign weight $w_p(e)$ to each edge of $G(\mbf z)$, where $w_p(e)$ is
2696: the number of times the edge appears in $p$.
2697: We call the edge weighting $w_p$ the {\em weighting} of the path $p$.
2698: Then there are finitely
2699: many directed paths in $G(\mbf z)$ which have
2700: %the same initial and terminal vertices as $p$ and
2701: the same weighting as $p$.
2702: We call a weighting of the edges of $G(\mbf z)$ a
2703: $Z${\em --weighting} if it equals the weighting of a $Z$--path.
2704: Given a $Z$--weighting $w$ of $G(\mbf z)$ the subgraph $P$
2705: consisting of all edges with positive weight
2706: (and their incident vertices) is the path--subgraph of a $Z$--path $p$,
2707: every edge of $P$ has positive weight
2708: and if $e$ has weight greater than 1 then $e$ is a loop. Hence any path $p^\prime$
2709: with weighting $w$ is a $Z$--path with the same initial and terminal vertices as
2710: $p$.
2711: Given a sequence of $Z$--cancellations the corresponding path $p$
2712: gives rise to a $Z$--weighting of $G(\mbf z)$.
2713: %Conversely given a
2714: %$Z$--weighting $w$ of $G(\mbf z)$ there are finitely many directed paths
2715: %$p^\prime$ such that $w_{p^\prime}$ is equal to $w$. As the simple path
2716: %contained in a $Z$--subgraph is directed all the paths $p^\prime$ have
2717: %the same initial and terminal vertices and they are all $Z$--paths.
2718:
2719: \begin{lemma}\label{label-subgraph}
2720: Let $p_0$ and $p_1$ be $Z$--paths in $G(\mbf z)$ with weightings $w_{p_0}$ and $w_{p_1}$,
2721: respectively, and initial vertex $v=v(\G_s)$, for some basic picture $\G_s$. Let
2722: $\G_i$ be the picture obtained from $\G_s$ by
2723: $Z$--insertion following $p_i$, $i=0,1$. If $w_{p_0}=w_{p_1}$ then the boundary
2724: interval $b_q$ has the same label in $\G_0$ and $\G_1$, for $1\le q\le W_1$.
2725: \end{lemma}
2726: \medskip
2727:
2728: \noindent{\em Proof.} From Lemma \ref{follow-path} the boundary labels
2729: of $\G_i$ depend only on the number of occurrences of edges of $P$ in
2730: $p_0$ and $p_1$. As $w_{p_0}=w_{p_1}$ every edge of $G(\mbf z)$
2731: occurs the same number of times in $p_0$ as in $p_1$.
2732: \medskip
2733:
2734: Let $P$ be the path--subgraph of a $Z$--path in $G(\mbf z)$, let
2735: $ e_1,\ldots,e_t$ be the edges of $P$ and let $L(s_i)$ be the label of
2736: $e_i$. Let $\G_s$ be a basic picture for $(\mbf z,\cL)$ with prime labels
2737: ${\mbf{\ovr z}}$, labelled by $\hat\al_s({\mbf{\ovr z}})$, for some solution
2738: $\al_s$ to $\cH$. Let $\lm_1^\prime,\ldots ,\lm_t^\prime$ be elements
2739: of $\Lm$ not occuring in ${\mbf z}$, $\mbf{\ovr z}$ or $\cL$. Let $(h_q,f_q)$ be the
2740: $q$th letter of $\mbf z$: so the label of the boundary interval $b_q$
2741: of $\G_s$ is $h_q\^\al_s(\lm_q)$, for $q\in\{1,\ldots, W_1\}$.
2742: Define the system of parameters $\cL(\G_s,P)$ as follows.
2743: \[f_q=\al_s(\lm_q)+l(h_q)\left(\sum_{i=1}^t \lm^\prime_i\z(q,s_i)
2744: \right),\textrm{ for }1\le q\le W_1,\]
2745: \[\lm_i^\prime=1, \textrm{ if } e_i\textrm{ is not a loop},\]
2746: \[\lm_i^\prime\ge 1, \textrm{ for } i=1,\ldots ,t.\]
2747:
2748: \begin{theorem}\label{basic-reduce}
2749: Let $P$ be the path--subgraph of a $Z$--path with initial vertex
2750: $v(\G_s)$ and let
2751: $\al_s$ be a solution to $\cH$ such that
2752: $(\al_s,\G_s)$ corresponds to a solution of $Q_\cH$. Then the following
2753: are equivalent.
2754: \be
2755: \item\label{basic-reduce-1}
2756: There exists a solution $\al$ to $\cL$ and a picture $\G$ such
2757: that $(\al,\G)$ corresponds to a solution of $Q$ and such that $\G_s$
2758: is obtained from $\G$ by $Z$--cancellation, along some $Z$--path with path--subgraph
2759: $P$ and initial vertex $v(\G_s)$.
2760: \item\label{basic-reduce-2} There exists a solution $\al_0$ to $\cL\cup \cL(\G_s,P)$.
2761: %, where
2762: % $P$ is the path--subgraph of some $Z$--path in $G(\mbf z)$ with
2763: % initial vertex $v(\G_s)$.
2764: \ee
2765: Furthermore $\al$ and $\al_0$ may be chosen so that $\al_0(\lm)=\al(\lm)$, for all $\lm$
2766: occuring in $\mbf z$ or $\cL$.
2767: \end{theorem}
2768: \medskip
2769:
2770: \noindent{\em Proof.}
2771: Assume (\ref{basic-reduce-1}) holds and let $\G_s$ be obtained from
2772: $\G$ by $Z$--cancellation following the $Z$--path $p$, with path--subgraph $P$. Assume $p$ has
2773: edges $e_1,\ldots, e_t$, let $L(s_i)$ be the label of $e_i$
2774: and let $n_i$ be the number of occurrences of
2775: edge $e_i$ in $p$, $i=1,\ldots,t$.
2776: Then, using Theorem \ref{follow-path}(\ref{follow-path-1}), the label of the boundary interval $b_q$ of $\G$ is
2777: \begin{equation}\label{exp-lab}
2778: h_q\^\al(f_q)=h_q\^\left(m_q+l(h_q)\left(\sum_{i=1}^{t}n_i\z(q,s_i)\right)\right),
2779: \end{equation}
2780: where $m_q=\al_s(\lm_q)$.
2781: Define a retraction $\al_0:M\maps \ZZ$ by $\al_0(\lm)=\al(\lm)$, for
2782: $\lm\in\Lm\bl\{\lm_1^\prime,\ldots,\lm_t^\prime\}$ and
2783: $\al_0(\lm_i^\prime)=n_i$, for $i=1,\ldots,t$. Then $\al_0(f_q)=\al(f_q)$,
2784: for $1\le q\le W_1$, since $\lm_i^\prime$ does not occur in $\mbf
2785: z$. Hence, from (\ref{exp-lab}),
2786: \[\al_0(f_q)=m_q+l(h_q)\left(\sum_{i=1}^{t}\al_0(\lm^\prime_i)\z(q,s_i)\right)\]
2787: and, in the light of the above remarks about weightings of $Z$--paths,
2788: \[\al_0(\lm_i^\prime)\ge 1,\] for all $i$, with equality if $e_i$ is not
2789: a loop. Hence $\al_0$ is a solution of $\cL(\G_s,P)$. As
2790: $\al_0(\lm)=\al(\lm)$, when $\lm\notin\{\lm_1^\prime,\ldots,\lm_t^\prime\}$,
2791: it follows that $\al_0$ is a solution to $\cL$ and that the final statement
2792: of the Theorem holds.
2793:
2794: Conversely, suppose there exists a solution $\al_0$ to $\cL\cup \cL(\G_s,P)$, where
2795: $P$ is the path--subgraph of some $Z$--path in $G(\mbf z)$ with
2796: initial vertex $v(\G_s)$. Let $P$ have edges $e_1,\ldots ,e_t$ and
2797: let
2798: $L(s_i)$ be the corridor--section labelling $e_i$.
2799: %If $\al_0$ is a solution to $\cL(\G_s,p)$ t
2800: Then assigning weight
2801: $n_i=\al_0(\lm^\prime_i)$ to edge $e_i$ of $P$ (and 0 to all other edges of $G(\mbf
2802: z)$) gives a $Z$--weighting $w$ of $G(\mbf z)$. Let $p$ be a $Z$--path
2803: with path--subgraph $P$ and weighting $w$: so edge $e_i$ occurs
2804: $w(e_i)=\al_0(\lm^\prime_i)$ times in $p$. Then $p$ has initial vertex
2805: $\G_s$. Let $\G$ be the picture obtained from $\G_s$ by $Z$--insertion
2806: following $p$ and let $m_q=\al_s(\lm_q)$, for $q=1,\ldots ,W_1$.
2807: Then the label of the boundary interval $b_q$ in $\G$
2808: is
2809: \begin{align*}
2810: & h_q\^\left(m_q+l(h_q)\left(\sum_{i=1}^t n_i\z(q,s_i)\right)\right)\\
2811: &=h_q\^\al_0(f_q),
2812: \end{align*}
2813: as $\al_0$ is a solution to $\cL(\G_s,P)$. Thus $\al_0$ is a solution to
2814: $\cL$ and the boundary label $b_q$ of $\G$ has label
2815: $\hat\al_0(h_q,f_q)$: that is $(\al_0,\G)$ corresponds to a solution
2816: of $Q$.
2817: %%% Local Variables:
2818: %%% mode: latex
2819: %%% TeX-master: "qpe"
2820: %%% End:
2821: