1: \documentclass[a4paper]{amsart}%
2: \usepackage{amssymb}
3: \usepackage{amsfonts}
4: \usepackage{graphicx}
5: \usepackage{breakcites}%
6: \usepackage{amsmath}%
7: \setcounter{MaxMatrixCols}{30}
8: %TCIDATA{OutputFilter=latex2.dll}
9: %TCIDATA{Version=4.10.0.2363}
10: %TCIDATA{CSTFile=amsartci.cst}
11: %TCIDATA{Created=Wednesday, April 30, 2003 10:04:39}
12: %TCIDATA{LastRevised=Wednesday, April 30, 2003 16:46:34}
13: %TCIDATA{<META NAME="GraphicsSave" CONTENT="32">}
14: %TCIDATA{<META NAME="DocumentShell" CONTENT="Articles\SW\AMS Journal Article">}
15: \newtheorem{theorem}{Theorem}
16: \theoremstyle{plain}
17: \newtheorem{acknowledgement}{Acknowledgement}
18: \newtheorem{algorithm}{Algorithm}
19: \newtheorem{axiom}{Axiom}
20: \newtheorem{case}{Case}
21: \newtheorem{claim}{Claim}
22: \newtheorem{conclusion}{Conclusion}
23: \newtheorem{condition}{Condition}
24: \newtheorem{conjecture}{Conjecture}
25: \newtheorem{corollary}{Corollary}
26: \newtheorem{criterion}{Criterion}
27: \newtheorem{definition}{Definition}
28: \newtheorem{example}{Example}
29: \newtheorem{exercise}{Exercise}
30: \newtheorem{lemma}{Lemma}
31: \newtheorem{notation}{Notation}
32: \newtheorem{problem}{Problem}
33: \newtheorem{proposition}{Proposition}
34: \newtheorem{remark}{Remark}
35: \newtheorem{solution}{Solution}
36: \newtheorem{summary}{Summary}
37: \numberwithin{equation}{section}
38: \setlength{\oddsidemargin}{0cm}
39: \setlength{\evensidemargin}{0cm}
40: \setlength{\footskip}{1cm}
41: \setlength{\textwidth}{6.5in}
42: \setlength{\textheight}{8.25in}
43: \begin{document}
44: \title[On a composition of digraphs]{On a composition of digraphs}
45: \author{Simone Severini }
46: \address{Department of Computer Science, University of Bristol, BS8 1UB, Bristol,
47: United Kingdom}
48: \email{severini@cs.bris.ac.uk}
49: \date{March, 2003}
50: \subjclass[2000]{Primary 05C20; Secondary 68R10.}
51: \keywords{Digraphs; composition; de Bruijn digraph; interconnection networks; quantum mechanics.}
52:
53: \begin{abstract}
54: Many \textquotedblleft good\textquotedblright\ topologies for interconnection
55: networks are based on line digraphs of regular digraphs. These digraphs
56: support unitary matrices. We propose the property \textquotedblleft being the
57: digraph of a unitary matrix\textquotedblright\ as additional criterion for the
58: design of new interconnection networks. We define a composition of digraphs,
59: which we call diagonal union. Diagonal union can be used to construct digraphs
60: of unitary matrices. We remark that diagonal union digraphs are state split
61: graphs, as defined in symbolic dynamics. Finally, we list some potential
62: directions for future research.
63:
64: \end{abstract}
65: \maketitle
66:
67:
68: \section{Introduction}
69:
70: Point-to-point interconnection networks for parallel and distributed systems
71: are usually modeled by (di)graphs. Survey papers on interconnection networks
72: are, \emph{e.g.}, (in chronological order)\ Feng \cite{F81}, Heydemann
73: \cite{H97} and Ferrero \cite{Fe99}. The vertices of the (di)graph correspond
74: to the nodes of the network, that is processors, nodes with local memory,
75: switches, \emph{etc}.; the arcs correspond to communication wires. Performance
76: of parallel and distributed systems is significantly determined by the choice
77: of the network topology. The basic requirements are: (i) large numbers of
78: nodes; (ii)\ small distance between any two nodes (small communication delay);
79: (iii) limited number of wires. These requirements result in the need of
80: finding (di)graphs, with, respectively: (i) large order; (ii) small diameter;
81: (iii) bounded degree. Another important requirement is fault-tolerance, which
82: is measured in the terms of several parameters, two of which are
83: vertex-connectivity and arc(edge)-connectivity.
84:
85: Many \textquotedblleft good\textquotedblright\ topologies for interconnection
86: networks are based on line digraphs of regular digraphs (\emph{e.g.} de Bruijn
87: digraphs, Reddy-Pradhan-Kuhl digraphs, butterflies digraphs, \emph{etc.}),
88: hypercubes (that are not line digraphs), and a number of their
89: generalizations. Line digraphs have many remarkable properties. Line digraphs
90: are used as tools in algorithmic applications (see, \emph{e.g.}, Gusfield
91: \cite{GKWS98}), as underlying digraphs of coined quantum random walks
92: \cite{S03} (see Kempe \cite{K03}, for a survey paper on quantum random walks),
93: in the study of spectral statistics and random matrix theory (see,
94: \emph{e.g.}, Pakonski \emph{et al.} \cite{PTZ02}), in the study of growth
95: functions of free groups (see, \emph{e.g.}, Rivin \cite{R99}).
96:
97: Recall that, given an $n\times n$ matrix $M$ (over any field), a digraph $D$
98: is said to \emph{support} $M$, or, equivalently, to be \emph{the digraph of
99: }$M$, if $D$ is on $n$ vertices, and its adjacency matrix, $M\left( D\right)
100: $, has $ij$-th element
101: \[
102: M\left( D\right) _{i,j}=\left\{
103: \begin{tabular}
104: [c]{cc}%
105: $1$ & if $M_{i,j}\neq0,$\\
106: $0$ & otherwise.
107: \end{tabular}
108: \right.
109: \]
110: Recall that a complex $n\times n$ matrix $U$ is \emph{unitary} if $U^{\dagger
111: }U=I_{n}$, where $U^{\dagger}$ is the adjoint of $U$, and $I_{n}$ is the
112: identity matrix of size $n$. The digraph of a unitary matrix is without
113: cut-vertices and it is bridgeless, then it is $2$-vertex-connected and
114: $2$-arc-connected. Note that this properties insure a certain degree of
115: fault-tolerance when thinking about the digraph as a model of an
116: interconnection network. Both, line digraphs of regular digraphs and
117: hypercubes support unitary matrices \cite{S03I}. This observation motivates
118: the present paper, which is structured as follows.
119:
120: A \emph{composition of digraphs} is a method to construct digraphs
121: \textquotedblleft putting together\textquotedblright\ smaller digraphs and
122: adding arcs according to specific rules. In \S 2 we define a composition of
123: digraphs, which we call \emph{diagonal union}. The composition can be used to
124: construct digraphs of unitary matrices. We observe: simple conditions under
125: which a diagonal union digraph supports unitary matrices; that line digraphs
126: of regular digraphs are diagonal union digraphs (\S 3); that diagonal union
127: digraphs are state split graphs, as defined in symbolic dynamics (\S 3).
128: Finally, in \S 4, we propose a few potential direction for future research.
129:
130: Although the content of the paper is elementary, we wish to write it as
131: self-contained as possible. For terms of graph theory not defined here, we
132: refer the reader to the monograph by Bang-Jensen and Gutin \cite{B-JG01}.
133:
134: \section{A composition of digraphs}
135:
136: \subsection{Set-up}
137:
138: A (finite) \emph{directed graph}, for short \emph{digraph}, consists of a
139: non-empty finite set of elements called \emph{vertices} and a (possibly
140: empty)\ finite set of ordered pairs of vertices called \emph{arcs}. Let us
141: denote by $D=\left( V,A\right) $ a digraph with vertex-set $V\left(
142: D\right) $ and arc-set $A\left( D\right) $. The \emph{adjacency matrix} of
143: a digraph $D$ on $n$ vertices, denoted by $M\left( D\right) $, is the
144: $n\times n$ $\left( 0,1\right) $-matrix with $ij$-th entry
145: \[
146: M_{i,j}\left( D\right) =\left\{
147: \begin{tabular}
148: [c]{ll}%
149: $1$ & if $\left( v_{i},v_{j}\right) \in A\left( D\right) ,$\\
150: $0$ & otherwise.
151: \end{tabular}
152: \ \right.
153: \]
154:
155:
156: A digraph $H$ is a \emph{subdigraph} of a digraph $D$ if $V\left( H\right)
157: \subseteq V\left( D\right) $ and $A\left( H\right) \subseteq A\left(
158: D\right) $. A subdigraph $H$ of a digraph $D$ is a \emph{spanning subdigraph}
159: of $D$ if $V\left( H\right) =V\left( D\right) $;\ in such a case, we say
160: that $H$ \emph{spans} $V\left( D\right) $. A \emph{decomposition} of a
161: digraph $D$ is a set $\left\{ H_{1},H_{2},...,H_{k}\right\} $ of subdigraphs
162: whose arc-sets are exactly the classes of a partition of $A\left( D\right)
163: $. A subdigraph of $D$ is called a \emph{factor} of $D$ if it spans $V\left(
164: D\right) $. A \emph{factorization} of $D$ is a decomposition of $D$ into factors.
165:
166: The \emph{Hadamard product} (in the literature, sometimes called \emph{Schur
167: product} or \emph{entry-wise product}) of $n\times m$ matrices $N$ and $M$,
168: denoted by $N\circ M$, is defined as follows:
169: \[
170: \left( N\circ M\right) _{i,j}=N_{i,j}\cdot M_{i,j}.
171: \]
172:
173:
174: \begin{definition}
175: [Generalized Hadamard product]Let $M$ and $N$ be respectively an $m\times m$
176: and an $n\times n$ matrix (over any field). Let $m=n\cdot r$. The
177: \emph{generalized Hadamard product} of $N$ with $M$, denoted by $N\circ_{G}M$,
178: is defined as follows:
179:
180: \begin{itemize}
181: \item[(i)] If $r=1$ ($n=m$ and $N$ has the same size of $M$) then
182: \[
183: N\circ_{G}M=N\circ M.
184: \]
185:
186:
187: \item[(ii)] If $r>1$ then we look at $M$ as a block-matrix of the form
188: \[
189: \left[
190: \begin{array}
191: [c]{ccc}%
192: A_{1,1} & \cdots & A_{1,n}\\
193: \vdots & \ddots & \vdots\\
194: A_{n,1} & \cdots & A_{n,n}%
195: \end{array}
196: \right] ,
197: \]
198: where the $ij$-th block is $r\times r$. Then
199: \[
200: N\circ_{G}\left[
201: \begin{array}
202: [c]{ccc}%
203: A_{1,1} & \cdots & A_{1,n}\\
204: \vdots & \ddots & \vdots\\
205: A_{n,1} & \cdots & A_{n,n}%
206: \end{array}
207: \right] =\left[
208: \begin{array}
209: [c]{ccc}%
210: N_{1,1}A_{1,1} & \cdots & N_{1,n}A_{1,n}\\
211: \vdots & \ddots & \vdots\\
212: N_{n,1}A_{n,1} & \cdots & N_{n,n}A_{n,n}%
213: \end{array}
214: \right] .
215: \]
216:
217: \end{itemize}
218: \end{definition}
219:
220: \begin{example}
221: Let
222: \[
223: \sigma_{x}=\left[
224: \begin{array}
225: [c]{cc}%
226: 0 & 1\\
227: 1 & 0
228: \end{array}
229: \right] .
230: \]
231: Then
232: \[%
233: \begin{tabular}
234: [c]{lll}%
235: $\sigma_{x}\circ_{G}\left[
236: \begin{array}
237: [c]{cc}%
238: A & B\\
239: D & C
240: \end{array}
241: \right] =\left[
242: \begin{array}
243: [c]{cc}%
244: \mathbf{0} & B\\
245: D & \mathbf{0}%
246: \end{array}
247: \right] $ & and & $I_{2}\circ_{G}\left[
248: \begin{array}
249: [c]{cc}%
250: A & B\\
251: D & C
252: \end{array}
253: \right] =\left[
254: \begin{array}
255: [c]{cc}%
256: A & \mathbf{0}\\
257: \mathbf{0} & C
258: \end{array}
259: \right] .$%
260: \end{tabular}
261: \]
262:
263: \end{example}
264:
265: \subsection{Definition}
266:
267: Let $G$ be a finite group of order $n$. Let $W$ be a complex vector space of
268: finite dimension. Let $GL\left( W\right) $ be the group of the bijective
269: linear maps on $V$. A \emph{linear representation} of $G$ is a homomorphism
270: $\rho:G\longrightarrow GL\left( W\right) $. To each $g_{i}\in G$, we
271: associate the unit column vector $e_{i}\in W$ with $j$-th entry
272: \[
273: e_{i_{j}}=\left\{
274: \begin{tabular}
275: [c]{ll}%
276: 1 & if $j=i,$\\
277: 0 & otherwise.
278: \end{tabular}
279: \ \right.
280: \]
281: Let $GL\left( n,F\right) $ be the group of invertible $n\times n$ matrices
282: over the field $F=GF\left( 2\right) $. The (left) \emph{regular permutation
283: representation} of $G$, denoted by $\rho_{reg}$, is a homomorphism $\rho
284: _{reg}:G\longrightarrow GL\left( n,F\right) $. Then $\rho_{reg}\left(
285: g_{l}\right) $ is a permutation matrix with $ij$-entry
286: \[
287: \rho_{reg}\left( g_{l}\right) _{i,j}=\left\{
288: \begin{tabular}
289: [c]{ll}%
290: 1 & if $g_{l}g_{i}=g_{j},$\\
291: 0 & otherwise.
292: \end{tabular}
293: \ \right.
294: \]
295:
296:
297: Let us denote by $\mathbb{Z}_{n}$ the additive abelian group of the integers
298: modulo $n$. Let us denote by $J_{n}$ the $n\times n$ all-ones matrix, and by
299: $M\oplus N$ the direct sum of matrices $M$ and $N$.
300:
301: \begin{definition}
302: [Diagonal union]Let $D$ be a digraph on $n$ vertices and let $\mathcal{F}%
303: =\left\{ H_{1},H_{2},...,H_{k}\right\} $ be a factorization of $D$. The
304: digraph $D_{\mathcal{F}}$ is defined as follows:
305: \[
306: M\left( D_{\mathcal{F},1}\right) =M\left( D_{\mathcal{F}}\right) =\left(
307: J_{k}\otimes I_{n}\right) \cdot\bigoplus_{i=1}^{k}M\left( H_{i}\right) .
308: \]
309: Iteratively,
310: \[
311: M\left( D_{\mathcal{F},d}\right) =\left( J_{k}\otimes I_{n^{d-1}}\right)
312: \cdot\bigoplus_{j=0}^{k-1}\left[ \rho_{reg}\left( j\right) \circ
313: _{G}M\left( D_{\mathcal{F},d-1}\right) \right] ,
314: \]
315: $\rho_{reg}$ is the (left)\ regular permutation representation of
316: $\mathbb{Z}_{k}$. The digraph $D_{\mathcal{F},d}$ is called $d$\emph{-diagonal
317: union}\footnote{On the choice of the terminology:\ The term \textquotedblleft
318: diagonal union\textquotedblright\ may be justified by the structure of
319: $M\left( D_{\mathcal{F},d}\right) $. In fact, $M\left( D_{\mathcal{F}%
320: ,d}\right) $ is a block-matrix, with the blocks on the diagonal being the
321: adjacency matrices of the factors of $D_{\mathcal{F},d-1}$.} of $D$ in respect
322: to $\mathcal{F}$.
323: \end{definition}
324:
325: \begin{remark}
326: If the factors $H_{1},H_{2},...,H_{k}$ are digraphs of unitary matrices then
327: $M\left( D_{\mathcal{F},d}\right) $ is also the digraph of a unitary matrix.
328: This insures that $M\left( D_{\mathcal{F},d}\right) $ is $2$%
329: -vertex-connected and $2$-arc-connected \cite{S03I}.
330: \end{remark}
331:
332: A concrete example might be useful to clarify Definition 2. Let us denote by
333: $\overleftrightarrow{K}_{n}^{+}$ the complete symmetric digraph with a
334: self-loop at each vertex.
335:
336: \begin{example}
337: Let $\mathcal{F}=\left\{ H_{1},H_{2}\right\} $ be a factorization of a
338: digraph $D$. In Figure \ref{fig1}, below, are represented the matrices
339: $M\left( D_{\mathcal{F}}\right) ^{\intercal}$, $M\left( D_{\mathcal{F}%
340: ,2}\right) ^{\intercal}$, $M\left( D_{\mathcal{F},3}\right) ^{\intercal}$
341: and $M\left( D_{\mathcal{F},4}\right) ^{\intercal}$. The light and the dark
342: shaded blocks represent $M\left( H_{1}\right) $ and $M\left( H_{2}\right)
343: $, respectively; the white areas are of zeros only.
344:
345: Let
346: \[%
347: \begin{tabular}
348: [c]{llll}%
349: $D=\overleftrightarrow{K}_{2}^{+},$ & $M\left( H_{1}\right) =I_{2}$ & and &
350: $M\left( H_{2}\right) =\sigma_{x}.$%
351: \end{tabular}
352: \]
353: Then
354: \[
355: M\left( D_{\mathcal{F},k}\right) =B\left( 2,k\right) ,
356: \]
357: the binary $k$-dimensional de Bruijn digraph.%
358: %TCIMACRO{\FRAME{ftbpF}{4.0857cm}{6.0297cm}{0pt}{}{\Qlb{fig1}}{deb1.eps}%
359: %{\special{ language "Scientific Word"; type "GRAPHIC";
360: %maintain-aspect-ratio TRUE; display "USEDEF"; valid_file "T";
361: %width 4.0857cm; height 6.0297cm; depth 0pt; original-width 11.5081cm;
362: %original-height 17.0788cm; cropleft "0"; croptop "1"; cropright "1";
363: %cropbottom "0"; tempfilename 'DEB1.EPS';tempfile-properties "XNPR";}}}%
364: %BeginExpansion
365: \begin{figure}
366: [ptb]
367: \begin{center}
368: \includegraphics[
369: natheight=17.078800cm,
370: natwidth=11.508100cm,
371: height=6.0297cm,
372: width=4.0857cm
373: ]%
374: {DEB1.EPS}%
375: \label{fig1}%
376: \end{center}
377: \end{figure}
378: %EndExpansion
379:
380: \end{example}
381:
382: \section{Remarks}
383:
384: \subsection{Extremal cases}
385:
386: Let $D$ be a digraph on $n$ vertices and $m$ arcs. Suppose that $\mathcal{F}$
387: is the trivial factorization of $D$, that is $\mathcal{F}=\left\{ D\right\}
388: $. In such a case, $D_{\mathcal{F}}=D$. A \emph{cycle factor} of a digraph $D$
389: is a collection of pairwise vertex-disjoint dicycles spanning $D$. In other
390: words, a cycle factor is a spanning $1$-regular subdigraph of $D$. Note that
391: the adjacency matrix of a cycle factor is a permutation matrix. A
392: \emph{factorization in cycle factors} is a factorization whose members are
393: cycle factors. The \emph{line digraph} of a digraph $D$, denoted by
394: $\overrightarrow{L}D$, is defined as follows (see, \emph{e.g.}, \cite{P96}%
395: ):\ the vertex-set of $\overrightarrow{L}D$ is $A\left( D\right) $; $\left(
396: v_{i},v_{j}\right) ,\left( v_{k},v_{l}\right) \in A\left( \overrightarrow
397: {L}D\right) $ if and only if $v_{j}=v_{k}$. Suppose that $\mathcal{F}$ is a
398: factorization in cycle factors. In such a case,
399: \[
400: D_{\mathcal{F},d}=\overrightarrow{L}^{d-1}D.
401: \]
402: So, note that the line digraph of a regular digraph is a diagonal union digraph.
403:
404: \subsection{State split graphs}
405:
406: The notion of \emph{state split graph}, introduced by Williams \cite{W74}, is
407: important in symbolic dynamics and coding. Lind and Marcus \cite{LM95} is
408: comprehensive for a monograph on symbolic dynamics. Given a digraph $D$, let
409: $\mathcal{E}_{i}$ be the set of the arcs in which $v_{i}$ is tail. For each
410: vertex $v_{i}\in V\left( D\right) $, let
411: \[%
412: \begin{tabular}
413: [c]{lll}%
414: $\mathcal{E}_{i}=\biguplus_{k=1}^{m\left( i\right) }\mathcal{E}_{i}^{k},$ &
415: & $m\left( i\right) \geq1.$%
416: \end{tabular}
417: \]
418: Let $\mathcal{P}$ be the resulting partition of $\mathcal{E}$, and let
419: $\mathcal{P}_{i}$ the resulting partition restricted to $\mathcal{E}_{i}$. The
420: \emph{state split graph} $D^{\left[ \mathcal{P}\right] }$ \emph{formed from}
421: $D$ \emph{using} $\mathcal{P}$ has vertices $v_{i}^{1},v_{i}^{2}%
422: ,...,v_{i}^{m\left( i\right) }$, for all $v_{i}\in V\left( D\right) $. The
423: arc $\left( v_{h},v_{l}\right) ^{j}\in A\left( D^{\left[ \mathcal{P}%
424: \right] }\right) $, where $\left( v_{h},v_{l}\right) \in A\left(
425: D\right) $ and $1\leq j\leq m\left( v_{l}\right) $. The arc $\left(
426: v_{h},v_{l}\right) \in\mathcal{E}_{h}^{k}$, for some $k$. The tail and the
427: head of $\left( v_{h},v_{l}\right) ^{j}$ are respectively $v_{h}^{k}$ and
428: $v_{l}^{j}$. If the construction of a state split graph uses a partition of
429: the outgoing (ingoing) arcs (as above) then the obtained digraph is called
430: \emph{out-split (in-split) graph}.
431:
432: \begin{remark}
433: A digraph $M\left( D_{\mathcal{F}}\right) $ is a state split graph. In
434: particular, $M\left( D_{\mathcal{F}}\right) $ is a in-split graph formed
435: from $D$ using the partition of $A\left( D\right) $ arising from the
436: factorization $\mathcal{F}$.
437: \end{remark}
438:
439: \begin{example}
440: Consider the Cayley digraph $D=Cay\left( \mathbb{Z}_{4},\left\{
441: 1,2,3\right\} \right) $:
442: \[
443: M\left( D\right) =\sum_{i=1}^{3}\rho_{reg}\left( i\right) =\left[
444: \begin{array}
445: [c]{ccccc}%
446: \emph{0} & 0 & 1 & 1 & 1\\
447: \emph{1} & 1 & 0 & 1 & 1\\
448: \emph{2} & 1 & 1 & 0 & 1\\
449: \emph{3} & 1 & 1 & 1 & 0
450: \end{array}
451: \right] ,
452: \]
453: where in the first column are the labels of the vertices.
454:
455: Let $\mathcal{F}=\left\{ D_{1},D_{2}\right\} $, where
456: \[%
457: \begin{tabular}
458: [c]{lll}%
459: $M\left( D_{1}\right) =\rho_{reg}\left( 1\right) +\rho_{reg}\left(
460: 3\right) $ & and & $M\left( D_{2}\right) =\rho_{reg}\left( 2\right) .$%
461: \end{tabular}
462: \]
463: Then
464: \begin{align*}
465: M\left( D_{\mathcal{F}}\right) & =\left[ \rho_{reg}\left( 1\right)
466: \otimes\rho_{reg}\left( 3\right) \right] \cdot\left( J_{2}\otimes
467: I_{4}\right) \\
468: & =\left[
469: \begin{array}
470: [c]{ccccccccc}%
471: \emph{0}^{\emph{1}} & 0 & 1 & 0 & 1 & 0 & 0 & 1 & 0\\
472: \emph{1}^{\emph{1}} & 1 & 0 & 1 & 0 & 0 & 0 & 0 & 1\\
473: \emph{2}^{\emph{1}} & 0 & 1 & 0 & 1 & 1 & 0 & 0 & 0\\
474: \emph{3}^{\emph{1}} & 1 & 0 & 1 & 0 & 0 & 1 & 0 & 0\\
475: \emph{0}^{\emph{2}} & 0 & 1 & 0 & 1 & 0 & 0 & 1 & 0\\
476: \emph{1}^{\emph{2}} & 1 & 0 & 1 & 0 & 0 & 0 & 0 & 1\\
477: \emph{2}^{\emph{2}} & 0 & 1 & 0 & 1 & 1 & 0 & 0 & 0\\
478: \emph{3}^{\emph{2}} & 1 & 0 & 1 & 0 & 0 & 1 & 0 & 0
479: \end{array}
480: \right] .
481: \end{align*}
482: Then $D_{\mathcal{F}}$ $\ $is an in-split graph formed from $D$ using the
483: partition arising from $\mathcal{F}$.
484: \end{example}
485:
486: \section{What's next?}
487:
488: The following seem to be natural questions:
489:
490: \begin{itemize}
491: \item[(i)] To what extent the property ``being the digraph of a unitary
492: matrix'' makes a digraph a ``good'' network topology?
493:
494: \item[(ii)] Can we make use of diagonal union digraphs as bases of new network topologies?
495:
496: \item[(iii)] We have seen that diagonal union digraphs are obtained via
497: state-splitting. Are tools from symbolic dynamics and coding helpful in the
498: design and analysis of interconnection networks, and in broadcasting problems?
499: \end{itemize}
500:
501: There is also space for a legitimate speculation:
502:
503: \begin{itemize}
504: \item[(i)] The time-evolution of a quantum mechanical system, in a pure state,
505: is induced by unitary matrices. Then interconnection networks based on
506: digraphs of unitary matrices have something to do with quantum mechanical
507: systems. What? In some way, these networks model certain classical systems,
508: whose dynamics share some combinatorial structure with quantum systems. So,
509: they are (partially?) combinatorially quantum. Can this idea be exploited in
510: the context of quantum computation and communication?
511: \end{itemize}
512:
513: \bigskip
514:
515: \textbf{Acknowledgement.} The author wishes to thank Yaokun Wu for introducing
516: him to symbolic dynamics, Cecilia Bebeacua and Dhiraj Pradhan for showing
517: interest in this paper. The author is supported by a University of Bristol
518: Research Scholarship.
519:
520: \begin{thebibliography}{999999} %
521: \bibitem[B-JG01]{B-JG01}J. Bang-Jensen and G. Gutin, \emph{Digraphs. Theory,
522: algorithms and applications}, Springer Monographs in Mathematics,
523: Springer-Verlag, London, 2001.
524:
525: \bibitem[F81]{F81}T. Y. Feng, A survey of interconnection networks,
526: \emph{Computer} \textbf{14} (1981), 960-965.
527:
528: \bibitem[Fe99]{Fe99}D. Ferrero, Introduction to interconnection network
529: models, \emph{Publ.Mat. Urug.}, 99/25 (1999).
530:
531: \bibitem[GKWS98]{GKWS98}D. Gusfield, R. Karp, L. Wang and P. Stelling, Graph
532: traversal, genes and matroids:\ an efficient case of the travelling salesman
533: problem, \emph{Discrete Appl. Math.} \textbf{88} (1998), 167-180.
534:
535: \bibitem[H97]{H97}M.-C. Heydemann, Cayley graphs and interconnection networks,
536: in: Graph Symmetry, Algebraic Methods and Applications (eds. G. Hahn and G.
537: Sabidussi), 1997.
538:
539: \bibitem[LM95]{LM95}D. Lind and B. Marcus, \emph{An introduction to symbolic
540: dynamics and coding}, Cambridge Univ. Press, 1995.
541:
542: \bibitem[K03]{K03}J. Kempe, Quantum random walks - an introductory overview,
543: \emph{Contemporary Physics}, to appear (2003). quant-ph/0303081
544:
545: \bibitem[PTZ02]{PTZ02}P. Pakonski, G. Tanner and K. \.{Z}yczkowski, Families
546: of line-graphs and their quantization, \emph{preprint} (2002). nlin.CD/0110043.
547:
548: \bibitem[P95]{P95}E. Prisner, Line graphs and generalizations---a survey.
549: Surveys in graph theory (San Francisco, California, 1995) \emph{Congr. Numer.}
550: \textbf{116} (1995), 193-229.
551:
552: \bibitem[R99]{R99}I. Rivin, Growth in free groups (and other stories),
553: \emph{Pr\'{e}publications de l'IH\'{E}S} (1997) \ math.CO/9911076.
554:
555: \bibitem[S03]{S03}S. Severini, The underlying digraph of a coined quantum
556: random walk, submitted to \emph{Journal of Quantum Computation and
557: Information}. preprint (2002). quant-ph/02100055.
558:
559: \bibitem[S03I]{S03I}S. Severini, Graph of unitary matrices, \emph{preprint}
560: (2003). math-CO/0303084 (draft).
561:
562: \bibitem[W74]{W74}R. F. Williams, Classification of subshifts of finite type,
563: \emph{Annals of Math.} \textbf{98} (1973), 120-153; erratum, \emph{Annals of
564: Math.} \textbf{99} (174), 380-381.
565: \end{thebibliography}
566:
567:
568: \end{document}