1: \documentclass[10pt, titlepage]{article}
2: \usepackage{epsf,here,amssymb}
3: \pagestyle{plain}
4: \def\commentout#1{}
5: \newtheorem{fact}{Fact}
6: \newtheorem{obs}{Observation}
7: \def\proof{\noindent{\bf Proof.\ }}
8: %\def\endproof{\begin{flushright}$\Box$\end{flushright}}
9: \def\endproof{\hfill $\Box$}
10: \input epsf
11: %\newenvironment {proof} {{\bf Proof.}}{\hspace*{\fill}$\Box$\par\vspace{4mm}}
12: \newtheorem{theorem}{Theorem}
13: \newtheorem{lemma}[theorem]{Lemma}
14: \newtheorem{corollary}[theorem]{Corollary}
15: %\newcommand\QED{\ifhmode\allowbreak\else\nobreak\fi
16: %\quad\nobreak$\Box$\medbreak}
17: %\newenvironment{proof}{\par\noindent{\bf Proof:}\rm\enspace}{\QED\par}
18: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
19:
20: \begin{document}
21: \begin{titlepage}
22: \title{\Large Probabilistic Analysis of Rule 2 }
23: \author{Jennie C. Hansen\\
24: \small Actuarial Math and Statistics Department\\
25: \small Herriot-Watt University\\
26: \small J.Hansen@ma.hw.ac.uk
27: \and
28: Eric Schmutz\\
29: \small Department of Mathematics\\
30: \small Drexel University\\
31: \small Philadelphia, Pa. 19104 \\
32: \small Eric.Jonathan.Schmutz@drexel.edu\\
33: \\
34: Li Sheng\\
35: \small Department of Mathematics\\
36: \small Drexel University\\
37: \small Philadelphia, Pa. 19104\\
38: \small lsheng@cs.drexel.edu\\
39: %\and
40: %Jie Wu??????\\
41: %\small Department of Computer Science and Engineering\\
42: %\small Florida Atlantic University \\
43: %\small Boca Raton, FL. 33431\\
44: %\small jie@cse.fau.edu
45: }
46:
47: \date{\today}
48: \maketitle \abstract{ Li and Wu proposed Rule 2,
49: a localized
50: approximation
51: algorithm that attempts to find a small connected dominating set
52: in a graph. Here we study the asymptotic performance of Rule $2$
53: on random unit disk graphs formed from $n$ random points in an
54: $\ell_{n}\times \ell_{n}$ square region of the plane. If
55: $\ell_{n}=O(\sqrt{n/\log n})$, Rule 2 produces a dominating set
56: whose expected size is $O(n/(\log\log n)^{3/2}).$
57:
58:
59:
60:
61: \noindent {\bf keywords and phrases}: {\em coverage process,
62: dominating set, localized algorithm, performance analysis,
63: probabilistic analysis, Rule k, unit disk graph} }
64: \end{titlepage}
65:
66:
67: \section{ Introduction}
68: Suppose random points $V_{1},V_{2},\dots ,V_{n}$ are selected from
69: a connected region ${\cal Q}$ in $\Re^{2}.$ For each $i$, let
70: $D_{1}(V_{i})$ be the unit disk centered at $V_i$. There is a
71: large literature on coverage processes\cite{Hall} that enables one
72: to answer questions such as whether or not the random disks are
73: likely to cover all of ${\cal Q}$, i.e. whether ${\cal Q}\subseteq
74: \bigcup\limits_{i=1}^{n}D_{1}(V_{i}).$ A variant question asks
75: whether there is small {\em subset} of the disks whose union
76: already covers ${\cal Q}$: given $k<n$, are there indices
77: $i_1<i_2<\dots i_{k}$ such that ${\cal Q}\subseteq
78: \bigcup\limits_{j=1}^{k}D_{1}(V_{i_{j}}).$
79: For this variant, there are several interesting ways to
80: modify the meaning of \lq\lq coverage.\rq\rq For example: is
81: there a small subset of the disks whose union is connected and
82: contains all $n$ points $V_1,V_2,\dots ,V_n$ (but not necessarily
83: all of ${\cal Q}$)? These questions are a bit vague, but specific
84: examples arise naturally in connection with probabilistic models
85: for wireless networks. In particular, they are central to the
86: probabilistic analysis of Rule 2 in this paper.
87:
88: Rule 2 is a well known algorithm that was proposed by Wu and Li
89: \cite{WuLi} as a means of increasing the efficiency of routing in
90: ad hoc wireless networks. To describe the algorithm and a
91: probabilistic model, we need some graph theoretic terminology. A
92: {\em unit disk graph } has for its vertex set ${\cal V}$ a finite
93: set of points in $\Re^{2}.$ Given the vertex set ${\cal V}$, the
94: edge set ${\cal E}$ is determined as follows: an {\it undirected}
95: edge $e\in {\cal E}$ connects vertices $u,v\in {\cal V}$ (and in
96: this case we say that $u$ and $v$ are adjacent) iff $d(u,v)$, the
97: Euclidean distance between them, is less than one. Unit disk
98: graphs have been used by many authors
99: as mathematical
100: models for the interconnections between nodes in a wireless
101: network, and random unit disk graphs have been used as
102: probabilistic models for these networks
103: \cite{CCJ},\cite{Gilbert},
104: \cite{GuptaKumar},\cite{Hale},\cite{HS},\cite{M},\cite{M2}. A {\sl
105: dominating set} in any graph $G=({\cal V},{\cal E})$ is a subset
106: ${\cal C}\subseteq {\cal V}$ such that every vertex $v\in {\cal
107: V}$ either is in the set ${\cal C}$, or is adjacent to a vertex in
108: ${\cal C}.$ We say ${\cal C}$ is a {\em connected dominating set }
109: if ${\cal C}$ is a dominating set and the subgraph induced by
110: ${\cal C}$ is connected. Of course it is not possible for $G$ to
111: have a connected dominating set if $G$ itself is not connected. We
112: use the acronym \lq\lq CDS\rq\rq for a dominating set ${\cal C}$
113: such that the subgraph induced by ${\cal C}$ has the same number
114: of components that $G$ has. This paper deals with a {\em random}
115: unit disk graph model, ${\cal G}_{n}$, which is connected with
116: asymptotic probability one. Thus any CDS for ${\cal G}_{n}$ will
117: also be connected with high probability. We assume that each
118: vertex has a unique identifier taken from a totally ordered set.
119: For convenience, when $|{\cal V}|=n$, we will use the numbers
120: $1,2,\dots ,n$ as IDs, and will number the vertices accordingly.
121: If $v_i$ is any vertex (with ID $i$), define the neighborhood
122: ${\cal N}(v_i)$ to be
123: the set consisting of $v_i$ and any vertices in ${\cal V}$ that are
124: adjacent to $v_{i}.$
125: The CDS constructed by the Rule $2$ algorithm is denoted ${\cal
126: C}({\cal V})$, and its cardinality is $C({\cal V})=|{\cal C}({\cal
127: V})|.$ The elements of ${\cal C}({\cal V})$ are called \lq\lq
128: gateway nodes\rq\rq. ${\cal C}({\cal V})$ consists of all vertices
129: $v_{i}\in {\cal V} $ that are not excluded under the following
130: version of Rule 2:
131:
132: \vskip.5cm \noindent {\bf Rule 2: } {\sl Vertex
133: $v_{i}$ is excluded from ${\cal C}({\cal V})$ iff
134: ${\cal N}(v_i)$ contains at least one set of two vertices
135: $v_{i_1},v_{i_2}$ such that
136: \begin{itemize}
137: \item $i_{1}>i_{2}>i $ and
138: \item ${\cal N}(v_i)\subseteq {\cal N}(v_{i_1})\cup {\cal N}(v_{i_2})$ and
139: \item $v_{i_{1}}$ is adjacent to $ v_{i_{2}}.$
140: \end{itemize}}
141:
142: Wu and Li showed that this algorithm produces a CDS. They also
143: conjectured, based on simulation data, that it is effective in
144: the sense that it selects a CDS that is small relative
145: to $n$ \lq\lq in the average
146: case\rq\rq. In this paper we treat the analysis of Rule 2
147: mathematicially by considering its performance when
148: it is applied to a random unit disk
149: graph ${\cal G}_{n}.$
150: Specifically, let $\ell_{1}\leq
151: \ell_{2}\leq \dots $ be a sequence of real numbers such that
152: $\ell_{n}=O(\sqrt{n/\log n})$ as $n\rightarrow\infty,$ but
153: $\ell_{n}\geq \log n$ for all $n$. Let ${\cal Q}_{n}$ be an
154: $\ell_n\times \ell_{n}$ square region in $\Re^{2}$.
155: Select
156: $n$ points $V_1,V_2,\dots ,V_n$ independently and uniform
157: randomly from an ${\cal Q}_{n}$, and use these $n$ points as the
158: vertex set for a unit disk graph ${\cal G}_{n}$. With this
159: probabilistic model, the size of the Rule 2 dominating set is a
160: random variable. We prove asymptotic estimates for
161: the expected size of the Rule 2 dominating set. The proof
162: involves some interesting problems in elementary geometry and
163: geometric probability.
164:
165:
166:
167:
168:
169:
170:
171:
172: \section{A Geometric Lemma}
173: As observed in
174: \cite{JLMV}, a unit disk centered at a point $o$ cannot be completely covered with
175: two unit disks having centers at points $u$ and $w$
176: ($u\not=o\not=w$): \ $(D_{1}(u)\bigcup D_{1}(w))^{c}\bigcap
177: D_{1}(o)\not= \emptyset.$
178: One might infer that a typical vertex $o$ is not likely to be be
179: pruned under Rule 2
180: because no two points in ${\cal N}(o)$ will cover all the vertices in ${\cal N}(o).$
181: This reasoning suggests that Rule 2 will be ineffective.
182: But such reasoning is not sound. Typically there are points $u$ and $w$
183: that cover all but a negligible fraction of
184: the disk centered at $o$. The uncovered region is small enough
185: so that it usually does not include any nodes.
186: A more precise version of this statement is proved in the next
187: section, but first we need to look carefully at the area of regions
188: such as $(D_{1}(u)\bigcup D_{1}(w))^{c}\bigcap
189: D_{1}(o).$ In particular, we need Lemma
190: \ref{notrunc}, which is the main result in this section.
191:
192:
193:
194:
195: To state Lemma \ref{notrunc} we adopt some notation. Throughout
196: this section $b>1$ will be a parameter and in terms of $b$
197: we let $L=\lfloor b^{1/3}(\log b)^{2}\rfloor, \delta={1\over
198: \sqrt[3]{b}\log b}$, and $\theta_{b}={\pi/L}.$ We fix
199: $o=(x_{o},y_{o})\in\Re^{2}$ %such that $x_{o},y_{o}>\delta.$
200: and for any $r>0$, %and any $o\in \Re^{2}$,%_{+},$
201: let $D_{r}(o)$ be the
202: closed disk centered at $o$ with radius $r$.
203: We
204: are going to partition the small disk $D_{\delta}(o)$ into $2L$
205: sectors as follows. Choose a new coordinate system centered at $o$, and for
206: $0\leq i< L,$ let
207: $Q_{i}$ be the sector consisting of those points
208: $(x,y)=(r\cos\theta,r\sin\theta)$ whose polar coordinates satisfy
209: $0<r\leq \delta $ and $(i-{1\over 2})\theta_{b}\leq \theta\leq
210: (i+{1\over 2}) \theta_{b} $.
211: Similarly let $R_{i}$ be the sector
212: that is obtained by reflecting $Q_{i}$ about $o$, namely the
213: points with $0<r<\delta $ and $(i-{1\over 2})\theta_{b}<
214: \theta-\pi < (i+{1\over 2})\theta_{b} $.
215: The analysis of Rule 2 depends on a geometric lemma about these sectors.
216: For any $i$, and any points
217: $q_i\in Q_{i},u_i\in R_{i}$, let
218: $X(q_i,u_i)$ be the area of $(D_{1}(q_i)\bigcup
219: D_{1}(u_i))^{c}\bigcap D_{1}(o)$, i.e. the area of the omitted
220: region in $D_{1}(o)$ that is not covered by $(D_{1}(q_i)\bigcup
221: D_{1}(u_i)$ . Let $\tilde{q}_i$
222: and
223: $\tilde{u}_i$ be the extreme points whose polar coordinates are
224: respectively
225: $(r,\theta)=(\delta,(i-{1\over
226: 2})\theta_{b})$ and $(r,\theta)=(\delta,(i+{1\over
227: 2})\theta_{b}+\pi).$
228: %Finally, define the \lq\lq truncated unit
229: %disk\rq\rq\ $\hat{D}_{1}(o):=D_{1}(o)\cap \Re_{+}^{2}.$ Thus
230: %$\hat{D}_{1}(o) \subseteq D_{1}(o) ,$ and $\hat{D}_{1}(o)=
231: %D_{1}(o)$ iff $x_{o},y_{o}\geq 1.$
232: We prove:
233:
234: \begin{lemma}
235: \label{notrunc}
236: There is a uniform constant $C>0$ such that, for $0\leq
237: i<L$, and for all $q_i\in Q_{i},u_i\in R_{i},$ we have
238: $%\hat
239: {X}(q_{i},u_{i})\leq X(\tilde{q}_{i},\tilde{u}_{i}) \leq
240: {C\over b\log^{3}b}$.%, where $\hat{X}(q,u)$ is the area of
241: %$(D_{1}(q)\cap D_{1}(u))^{c}\cap \hat{D}_{1}(o).$
242: \end{lemma}
243:
244:
245: \begin{proof}
246: %Clearly $\hat{X}(q_i,u_i)\leq X({q}_{i},{u}_{i})$ since
247: %$\hat{D}_{1}(o) \subseteq D_{1}(o).$ It therefore suffices to
248: %prove that
249: %\begin{equation}
250: %\label{notrunc}
251: %X({q}_{i},{u}_{i})\leq X(\tilde{q}_{i},\tilde{u}_{i}) \leq {C\over b\log^{3}b}.
252: % \end{equation}
253:
254: We prove four facts which together imply Lemma \ref{notrunc}. In
255: the first fact, we observe that omitted area $X(q,u)$ gets larger
256: if we move one (or both) of the two points $q,u$ away from the
257: origin along a radial line.
258:
259: \begin{fact}
260: Let $q_1,q_2$ and $u_1,u_2$ be four points in $D_{1}(o)$ such
261: that $q_1$ lies on the line segment $\overline{o,q_{2}}$ and $u_1$
262: lies on the line segment $\overline{o,u_2}$. Then $X(q_2,u_2)\ge
263: X(q_1,u_1)$.
264: \end{fact}
265:
266: \proof It suffices to show that $D_{1}(q_{2})\cap D_{1}(o)
267: \subseteq D_{1}(q_1)\cap D_{1}(o)$ and that $D_{1}(u_{2})\cap
268: D_{1}(o) \subseteq D_{1}(u_1)\cap D_{1}(o).$
269: Suppose $p\in D_{1}(q_{2})\cap D_{1}(o).$ Since $q_1$
270: lies on the line segment from $o$ to $q_2$, we have $d(q_1,p)\leq
271: \max(d(o,p),d(q_2,p)) \leq 1.$
272: Hence $p\in D_{1}(q_1)\cap D_{1}(o).$
273: By a similar same argument,
274: $D_{1}(q_{2})\cap D_{1}(o) \subseteq D_{1}(q_1)\cap
275: D_{1}(o).$
276:
277:
278: \endproof
279:
280: \begin{fact}\label{f2}
281: Let $a,b$ be the two points where the circles $\partial
282: D_{1}(p),
283: \partial D_{1}(q)$ intersect. Then, $\overline{a,b}\ \bot \
284: \overline{p,q}$, and the two line segments $\overline{a,b}$ and
285: $\overline{p,q}$ intersect at their midpoints.
286: \end{fact}
287:
288: \proof This follows immediately from the fact that $d(p,a)=d(p,b)=
289: d(q,a)=d(q,b)=1$.
290: \endproof
291:
292: \begin{fact}
293: Let $o_1,o_2$ be two points on the circle $x^2+y^2=\delta^2$.
294: Then, $X(o_1,o_2)$ is a decreasing function of $\angle o_1oo_2$.
295: \end{fact}
296:
297: \proof For convenience, we will use polar coordinates. Without
298: loss of generality, let $o_1$ be the point with polar coordinates
299: $(r_{o_1},\phi_{o_1})=(\delta,\pi)$. Let $o_2$ be an arbitrary
300: point on the circle with the polar coordinates $(\delta,\phi_2)$.
301: By symmetry, we only need to consider the case when $o_2$ is in
302: the first or second quadrant; we may, without loss of generality,
303: assume that $0\le \phi_2 \le \pi$. We will show that $X(o_1,o_2)$
304: is an increasing function of $\phi_2$, then the result follows
305: from the fact that $\angle o_1oo_2=\pi-\phi_2$.
306:
307: Let $a_1,b_1$ be the two points where the circles $\partial
308: D_{1}(o_1)$ and $\partial D_{1}(o)$ intersect, with $a_1$ in the
309: second quadrant and $b_1$ in the third quadrant.
310:
311: Let $o^*$ be a point on the circle $x^2+y^2=\delta^2$ so that
312: $\partial D_{1}(o^*)$ meets with both $\partial D_{1}(o)$ and
313: $\partial D_{1}(o_1)$ at $a_1$. Let $b^*,d^*$ be the other
314: intersection points of $\partial D_{1}(o^*)$ with $\partial
315: D_{1}(o)$ and $\partial D_{1}(o_1)$, respectively. For
316: convenience, let's denote $\phi_{o^*}$ by $\phi^*$.
317: Figure~\ref{fig:phi} illustrates the position of $\partial
318: D_{1}(o_1),\partial D(o)$, and $\partial D_{1}(o^*)$ and their
319: intersections.
320:
321: \begin{figure}[H]
322: \begin{center}
323: \leavevmode \epsfysize=2in \epsfbox{./phi.eps}
324: \end{center}
325: \caption{The position of the circle $\partial D_{1}(o^*)$}
326: \label{fig:phi}
327: \end{figure}
328:
329: As in the proof of Fact~\ref{f2},we have $\overline{a_1,d^*} \
330: \bot \ \overline{o_1,o^*}$, $\overline{a_1,b^*}\ \bot \
331: \overline{o,o^*}$. Notice also that $o$ is on the line segment
332: $\overline{a_1,d^*}$. So,
333: \begin{equation}~\label{e1}
334: \angle b^*a_1o=\angle oo^*o_1=\angle o^*o_1o=\frac{\phi^*}{2}.
335: \end{equation}
336: It follows that
337: \begin{equation}\label{sin-phi*}
338: 0< \phi^*/2 <\pi /2,\ \mbox{and},\ \sin \frac{\phi^*}{2}=
339: \frac{\delta}{2}
340: \end{equation}
341: Now, for the point $o_2$ with polar coordinates $(\delta,\phi_2)$,
342: let $a_2,b_2$ denote the two points where $\partial D_{1}(o_2)$
343: and $\partial D_{1}(o)$ intersect, and let $c_2,d_2$ denote the
344: two points where $\partial D_{1}(o_2)$ and $\partial D_{1}(o_1)$
345: intersect. There are two cases to consider: $\phi_2 \leq \phi^*$,
346: and $\phi_2 \geq \phi^*$
347:
348:
349: Case 1. $\phi_2 \le \phi^*$.
350:
351: \begin{figure}[H]
352: \begin{center}
353: \leavevmode \epsfysize=2in \epsfbox{./case1.eps}
354: \end{center}
355: \caption{The case when $\phi_2\le \phi^*$} \label{case1}
356: \end{figure}
357:
358: Notice that $a_1,b_1$ partitions the circle $\partial D_{1}(o)$
359: into two arcs: the right section and the left section. When,
360: $\phi_2 \le \phi^*$, as illustrated in Figure~\ref{case1}, $a_2,
361: b_2$ are both on the right section of the circle $\partial
362: D_{1}(o)$ between $a_1,b_1$. Similarly, $c_2,d_2$ are both on the
363: right section of the circle $\partial D_{1}(o_1)$ between
364: $a_1,b_1$. Clearly,
365: \[
366: X(o_1,o_2)=B_1-(B_2-B_3)=B_1-B_2+B_3,\] where
367: \begin{itemize}
368: \item $B_1=area(D_{1}(o_1)^c\cap D_{1}(o))$
369: \item $B_2=area (D_{1}(o)\cap D_{1}(o_2))$
370: \item $B_3=area(D_{1}(o_1)\cap D_{1}(o_2))$, the shaded area in
371: Figure~\ref{case1}
372: \end{itemize}
373: Notice that $B_3$ is the only area that depends on $\phi_2$. We
374: shall now give an expression for $B_3$.
375:
376: Let's denote $\angle c_2o_1o_2=y$. Since $\angle
377: o_2o_1o=\frac{\phi_2}{2}$, we have
378: \begin{equation}~\label{ey}
379: 0< y< \frac{\pi}{2},\ \mbox{and}, \ \cos y=\delta \cos
380: \frac{\phi_2}{2}
381: \end{equation}
382:
383: %By that $length(\overline{c_2,o_1})=length(\overline{c_2,o_2})$, and
384: %$length(\o%verline{d_2,o_1})=length(\overline{d_2,o_2})$,
385: By symmetry, one can see that the shaded region is partitioned
386: equally by the line $\overline{c_2,d_2}$. So,
387: \[B_3= 2(\frac{2y}{2\pi} \pi -\frac{1}{2} (2\sin y)(\cos
388: y))=2y-\sin 2y. \] Here, the first term is the area of the sector
389: $D_{1}(o_1)$ that extends from $c_2$ to $d_2$, and the second term
390: is the area of the triangle($c_2,o_1,d_2$).
391:
392: From the above two equations, we have
393: \[\frac{d X(o_1,o_2)}{d \phi_2}=\frac{d B_3}{d \phi_2}
394: =\frac{d B_3}{d y} \cdot \frac{d y}{d \phi_2}= (1-\cos 2y)\cdot
395: {\delta\sin{\phi_{2}\over 2}\over 2\sin y} > 0.\]
396: %2(1-\cos 2y)\cdot \frac{1}{2}\delta \sin \frac{\phi_2}{2}/\sin
397: %y>0\]
398: Here the last inequality follows from the fact that
399: $0<\frac{\phi_2}{2},y < \frac{\pi}{2}$. Thus
400: $X(o_1,o_2)$ is an increasing function in $\phi_2$.
401:
402:
403: Case 2. $\phi_2 > \phi^*$.
404: \begin{figure}[H]
405: \begin{center}
406: \leavevmode \epsfysize=2in \epsfbox{./case2.eps}
407: \end{center}
408: \caption{The case when $\phi_2> \phi^*$} \label{case2}
409: \end{figure}
410:
411: One can see from Figure~\ref{case2} that
412: \[
413: X(o_1,o_2)=B_1-(B_2-B_3)=B_1-B_2+B_3\] Where $B_1,B_2$ are defined
414: the same as those in the case 1, but
415: \[
416: B_3=area(D_{1}(o_1)\cap D_{1}(o_2)\cap D_{1}(o)), \ \mbox{the
417: shaded area in
418: Figure~\ref{case2} }\]
419: Again, $B_3$ is the only area that depends on $\phi_2$. We will
420: now give an expression of $B_3$.
421:
422: We show first that $\angle c_2oa_1 = \angle a_2oc_2$ by showing
423: that $\phi_{c_2}-\phi_{a_1}=\phi_{a_2}-\phi_{c_2}$. Then, it
424: follows that $B_3$ is split in half by the line segment
425: $\overline{c_2,d_2}$.
426:
427:
428: >From Figure~\ref{fig:phi}, one can see that
429: \begin{equation}
430: \phi_{a_1}=\phi^*+(\frac{\pi}{2}-\angle
431: b^*a_1o)=\phi^*+(\frac{\pi}{2}-\frac{\phi^*}{2})=\frac{\pi}{2}+\frac{\phi^*}{2}
432: \end{equation}
433:
434: To find $\phi_{a_2}$, observe that, as in the proof of
435: Fact~\ref{f2},we have $\overline{a_2,b_2}\ \bot \
436: \overline{o,o_2}$. So, $\sin \angle b_2a_2o =\frac{\delta}{2}$.
437: Comparing with (\ref{sin-phi*}), we see that $\sin \angle b_2a_2o
438: = \sin \frac{\phi^*}{2}$. This implies that $\angle b_2a_2o=
439: \frac{\phi^*}{2}$. Thus,
440: \begin{equation}
441: \phi_{a_2}=\phi_2+(\frac{\pi}{2}-\angle
442: b_2a_2o)=\phi_2+(\frac{\pi}{2}-\frac{\phi^*}{2})
443: \end{equation}
444:
445: Now, for $c_2$, using the fact that $\overline{c_2,o} \ \bot\
446: \overline{o_1,o_2}$,
447: \begin{equation}
448: \phi_{c_2}=\pi-(\frac{\pi}{2}-\angle
449: o_2o_1o)=\pi-(\frac{\pi}{2}-\frac{\phi_2}{2})=\frac{\pi}{2}+\frac{\phi_2}{2}
450: \end{equation}
451:
452: It follows that
453: $\phi_{c_2}-\phi_{a_1}=\phi_{a_2}-\phi_{c_2}=\frac{\phi_2}{2}-\frac{\phi^*}{2}$.
454: Now, using that the circle $\partial D_{1}(o_1)$ in the polar
455: system is
456: \[r=\sqrt{1-\delta^2\sin ^2 \phi} -\delta \cos \phi\]
457: and that
458: \begin{equation}
459: \phi_{d_2}=-(\pi-\phi_{c_2})=-(\frac{\pi}{2}-\frac{\phi_2}{2})
460: \end{equation}
461: we get
462: \[\begin{array}{ll}
463: B_3&=2(
464: \int_{-(\frac{\pi}{2}-\frac{\phi_2}{2})}^{\frac{\pi}{2}+\frac{\phi^*}{2}}\int_{0}^{\sqrt{1-\delta^2\sin
465: ^2 \phi} -\delta \cos \phi}
466: r\,drd\phi+\frac{\frac{\phi_2}{2}-\frac{\phi^*}{2}}{2\pi}\cdot
467: \pi
468: ) \vspace{0.2cm}\\
469: &=
470: \int_{-(\frac{\pi}{2}-\frac{\phi_2}{2})}^{\frac{\pi}{2}+\frac{\phi^*}{2}}
471: 1-\delta^2\sin^2 \phi+\delta^2\cos^2 \phi -2\delta \cos \phi
472: \sqrt{1-\delta^2\sin^2 \phi}
473: d\phi + \frac{{\phi_2}-{\phi^*}}{2}
474:
475: \end{array}
476: \]
477:
478: Thus,
479: \[\begin{array}{lll}
480: \frac{d X(o_1,o_2)}{d
481: \phi_2}=\frac{d B_3}{d
482: \phi_2}&=&-\frac{1}{2}[1-\delta^2\sin^2
483: (-\frac{\pi}{2}+\frac{\phi_2}{2})+\delta^2\cos^2
484: (-\frac{\pi}{2}+\frac{\phi_2}{2}) \vspace{0.2cm} \\
485: && -2\delta\cos
486: (-\frac{\pi}{2}+\frac{\phi_2}{2})\sqrt{1-\delta^2\sin^2
487: (-\frac{\pi}{2}+\frac{\phi_2}{2})}] +\frac{1}{2}
488: \vspace{0.2cm} \\
489: &=&\frac{1}{2}[\delta^2\cos^2\frac{\phi_2}{2}-\delta^2\sin^2\frac{\phi_2}{2}+
490: 2\delta \sin \frac{\phi_2}{2}
491: \sqrt{1-\delta^2\cos^2\frac{\phi_2}{2}}]\vspace{0.2cm} \\
492: &=&\frac{1}{2}[-(\delta\sin\frac{\phi_2}{2}
493: -\sqrt{1-\delta^2\cos^2\frac{\phi_2}{2}})^2+1] \vspace{0.2cm} \\
494: &\ge &0
495: \end{array}
496: \]
497: The last inequality follows because $0\le \delta
498: \sin\frac{\phi_2}{2}\le 1$, $0\le
499: \sqrt{1-\delta^2\cos^2\frac{\phi_2}{2}}\le 1$, and thus
500: $(\delta\sin\frac{\phi_2}{2}
501: -\sqrt{1-\delta^2\cos^2\frac{\phi_2}{2}})^2 <1$.
502: %Where, the last inequality follows from the fact that $\sin
503: %\frac{\phi_2}{2} >\sin \frac{\phi^*}{2}=\frac{\delta}{2}$, which
504: %implies that $\sqrt{1-\delta^2\cos^2\frac{\phi_2}{2}}\ge 1-\delta
505: %\sin\frac{\phi_2}{2}$. Thus
506: % $X(o_1,o_2)$ is an increasing function in $\phi_2$.
507:
508: \end{proof}
509: \begin{fact}
510: \label{omitted}Uniformly for all $i$, we have
511: $X(\tilde{q}_i,\tilde{u}_i)=O({1\over
512: b\log^{3}b}).$
513:
514: \end{fact}
515: \begin{proof}
516: Without loss of generality, let $i=0$ and $v=(0,0).$ To simplify
517: notation, define $x_{b}=\delta\cos(-{1\over 2}\theta_{b})$,
518: $y_{b}=\delta\sin(-{1\over 2}\theta_{b}).$
519: Let $(\xi,\eta)$ be the point in the first quadrant where
520: the circles $x^2+y^2=1$ and $(x-x_{b})^{2}+(y-y_{b})^2=1$ meet.
521: Then
522:
523: $$X(\tilde{q}_0,\tilde{u}_0)\leq 4\int\limits_{0}^{\xi}
524: \sqrt{1-x^{2}}-(y_{b}+\sqrt{1-(x-x_{b})^{2}})dx$$
525: $$= -4y_{b}\xi+4\int\limits_{0}^{\xi}{
526: {-2xx_{b}+x_{b}^{2}}\over
527: \sqrt{1-x^{2}}+\sqrt{1-(x-x_{b})^{2}}}dx$$ Hence we have
528: \begin{equation}
529: \label{xbound} X(\tilde{q}_0,\tilde{u}_0)=O(\xi
530: y_{b})+O(x_{b}\xi^{2})+O(x_{b}^{2}\xi).
531: \end{equation}
532: Note that $x_{b}^{2}+y_{b}^{2}=\delta^{2}={1\over
533: b^{2/3}\log^{2}b},$ that $\xi^{2}+\eta^{2}=1$, that
534: $(\xi-x_{b})^{2}+(\eta-y_{b})^{2}=1$, that
535: $x_{b}= \delta(1+O(\theta_{b}^{2}))$, and that $y_{b}=
536: {-\delta\theta_{b}\over 2}(1+O(\theta_{b}^{2})).$ Combining these
537: equations, we get
538: $\xi =O(\delta)$.
539: Putting this estimate back into (\ref{xbound}), we get
540:
541: \begin{equation}
542: \label{xarea} X(\tilde{q}_0,\tilde{u}_0)=O({1\over b\log^{3}b}).
543: \end{equation}
544: \end{proof}
545:
546: In the analysis of Rule 2 it is necessary to consider
547: vertices in ${\cal G}_n$ which are close to the boundary
548: of the square ${\cal Q}_n$. For this reason we
549: define, for $o\in \Re_+^2$, the ``truncated unit disk''
550: $\hat{D}_1(o):=D_1(o)\cap\Re_+^2$ and we note that
551: $\hat{D}_{1}(o) \subseteq D_{1}(o) ,$ and $\hat{D}_{1}(o)=
552: D_{1}(o)$ iff $x_{o},y_{o}\geq 1.$
553: Then for $L$ and $\delta$ as defined above,
554: we have the following corollary to Lemma \ref{notrunc}:
555:
556: \begin{corollary}
557: \label{trunc}
558: There is a uniform constant $C>0$ such that,
559: for all $o\in\Re_+^2$ such that $D_{\delta}(o)\subseteq
560: \Re_+^2$, for $0\leq
561: i<L$, and for all $q_i\in Q_{i},u_i\in R_{i},$ we have
562: $\hat
563: {X}(q_{i},u_{i})\leq X(\tilde{q}_{i},\tilde{u}_{i}) \leq
564: {C\over b\log^{3}b}$, where $\hat{X}(q,u)$ is the area of
565: $(D_{1}(q)\cap D_{1}(u))^{c}\cap \hat{D}_{1}(o).$
566: \end{corollary}
567:
568:
569: \begin{proof}
570: Clearly $\hat{X}(q_i,u_i)\leq X({q}_{i},{u}_{i})$ since
571: $\hat{D}_{1}(o) \subseteq D_{1}(o)$.
572: So the result follows from Lemma \ref{notrunc} (since
573: $\tilde{q}_i, \tilde{u}_i\in D_{\delta}(o)
574: \subseteq \Re_+^2$).
575: \end{proof}
576:
577:
578: \section{Local Coverage by Two Discs}
579:
580: Recall that under Rule 2 a vertex $v_i$ is excluded from ${\cal C}({\cal V})$
581: if there are two adjacent vertices, $v_{i_1}, v_{i_2}\in {\cal N}(v_i)$,
582: with higher IDs than
583: $v_i$ which also `cover' $v_i$, i.e. ${\cal N}(v_i)\subseteq {\cal N}(v_{i_1})
584: \cup {\cal N}( v_{i_2})$). In the analysis of Rule 2 we will distinquish vertices
585: in ${\cal N}(v_i)$ with higher ID than $v_i$ by coloring them blue; all other
586: vertices in ${\cal N}(v_i)$ are colored white. With this in mind, we consider in this
587: section a two-colored random unit disk graph and prove a local coverage
588: result.
589:
590:
591:
592: Let $w$ and $b$ be positive integers such that $w<b(\log b)^{2}$
593: and, as before, let $L=\lfloor b^{1/3}(\log b)^{3/2}\rfloor$ and $\delta=
594: {1\over b^{1/3}\log b}.$
595: Fix $o\in\Re^{2}_{+}$ such that $D_{\delta}(o)\subseteq
596: \Re_+^2$ and select $w+b$ points
597: independently and uniform randomly from
598: the truncated disk $\hat{D}_{1}(o).$ Color
599: the first $w$ points white, and the remaining $b$ points blue.
600: Form a random (improperly colored) unit disk graph $\hat{{\cal
601: H}}_{w,b}$ by putting an edge between two of the $w+b$ colored
602: points iff the distance between them is one or less.
603: Our goal in this section is to prove that,
604: with high probability, $\hat{{\cal H}}_{w,b}$ contains a
605: dominating set consisting of two blue vertices that are adjacent
606: to each other.
607:
608: For $0\leq i<L$, let $Q_i, R_i$ denote the sectors
609: of $D_{\delta}(o)$ as defined in the previous section
610: and let $N(Q_{i}), N(R_{i})$ respectively be the number of blue
611: vertices of $\hat{{\cal H}}_{w,b}$ that lie in $Q_{i}$ and $R_{i}$.
612: Let $\tau_{b}=\sum\limits_{i=0}^{L-1}I_{i}$ where, in this section only, the $I_{i}=1$
613: if and only if $N(R_{i})=N(Q_{i})=1$ (and otherwise $I_i=0$.) We note
614: that the distribution of $\tau_b$ depends on the position of $o$
615: and we indicate this dependence by using the notation
616: ${\Pr}_o(\tau_b\in \cdot)$. Provided $o$ is not too close to the boundary of $\Re_+^2$,
617: we can obtain uniform bounds on the tail of the distribution
618: of $\tau_b$:
619: \begin{lemma}
620: \label{taubound}%If $o=(x_o,y_o)$ satisfies $\delta \leq x_o,y_o$,
621: %then
622: ${\Pr}_o\left(\tau_b <{b^{1/3}\over 16\log^{6}b}\right) =
623: O({\log ^{6}b\over b^{1/3}})$ uniformly for all $o\in \Re_+^2$
624: such that $D_{\delta}(o)\subseteq \Re_+^2$.
625: \end{lemma}
626: \begin{proof}
627: %Recall that $L=\lfloor b^{1/3}(\log b)^{2}\rfloor$ and $\delta=
628: %{1\over b^{1/3}\log b}.$
629: %Define\ $\hat{\alpha}(o)={\rm Area}(\hat{D}_{1}(o)),$ and
630: Let $|\hat{D}_{1}(o)|$ denote the area of $\hat{D}_{1}(o)$, let
631: $\hat{\lambda}=\hat{\lambda}(o)={\pi\over |\hat{D}_{1}(o)|},$
632: %$\hat{\lambda}=\hat{\lambda}(o)={\pi\over \hat{\alpha}(o)},$
633: and define
634: \begin{equation}
635: \hat{p}={{\rm Area}(Q_{i})\over |\hat{D}_{1}(o)|} =
636: {\pi\delta^{2}/2L\over |\hat{D}_{1}(o)| }= {\hat{\lambda}\over
637: 2b\log^{4}b}\left(1+O({1\over b^{1/3}\log^{2}b})\right).
638: \end{equation}
639: The expected value of $I_{i}$ depends on $o$:
640: \begin{equation} E_{o}(I_{i})=b(b-1)\hat{p}^{2}(1-2\hat{p})^{b-2}= {\hat{\lambda}^{2}\over
641: 4\log^{8} b}\left(1+O({1\over \log^{4}b})\right).
642: \end{equation}
643: Hence
644: \begin{equation}
645: \label{Etaub}
646: E_{o}(\tau_{b})=LE_{o}(I_{i})={b^{1/3}\hat{\lambda}^{2}\over
647: 4\log^{6}b}\left(1+O({1\over \log^{4}b})\right).
648: \end{equation}
649: We likewise have, for $i\not=j$,
650: \begin{equation}
651: \label{crossterms}
652: E_{o}(I_{i}I_{j})=b(b-1)(b-2)(b-3)\hat{p}^{4}(1-4\hat{p})^{b-4}=
653: {\hat{\lambda}^{4}(o)\over 16\log^{16} b}\left(1+O({1\over
654: \log^{4}b})\right).
655: \end{equation}
656: Note that \begin{equation} \label{hatalphabnd}
657: \pi \ge |\hat{D}_{1}(o)| \ge {\pi\over
658: 4},\end{equation} and therefore \begin{equation}
659: \label{hatlambdabnd}
660: 1\leq
661: \hat{\lambda}(o)\leq 4.
662: \end{equation}
663: Therefore we have uniformly for all $o\in\Re_+^2$ such that
664: $D_{\delta}(o)\subseteq \Re_+^2$
665: \begin{equation}\label{variance} Var(\tau_{b})=O\left({b^{1/3}\over \log^{6}b}
666: \right).
667: \end{equation}
668: Observe that
669: \begin{equation}
670: \label{chebyshev} {\Pr}_o\left(\tau_b <{b^{1/3}\over 16\log^{6}b}\right)\leq
671: {\Pr}_o\left(\tau_b \leq {1\over 2}E_o(\tau_{b})\right) \leq
672: {\Pr}_o\left(|\tau_{b}-E_o(\tau_{b})|>{1\over 2}E_o(\tau_{b})\right).
673: \end{equation}
674: The lemma now follows from (\ref{variance}), (\ref{chebyshev}) and
675: Chebyshev's inequality.
676: \end{proof}
677: \vskip.5cm Recall our assumptions that $w<b(\log b)^{3/2},$ that
678: $\delta={1\over b^{1/3}\log b},$ and that $x_{o},y_{o}\geq
679: \delta.$ With these assumptions, we have:
680:
681: \begin{theorem}
682: \label{localCDS}
683: There is a constant $c>0$, independent of the position of $o$, such that with probability at least $1-{c\over ( \log b)^{3/2}},$
684: the random graph
685: $\hat{{\cal H}}_{w,b}$ has a connected dominating set that
686: consists of two blue vertices in $D_{\delta}(o).$
687: \end{theorem}
688:
689:
690:
691:
692:
693: \begin{proof}
694:
695: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
696:
697:
698: Let ${\cal T}_{b}\subseteq \bigl\lbrace 0,1,2,3,\dots
699: ,L-1\bigr\rbrace $ be the random subset of indices such that $i\in
700: {\cal T}_{b}$ iff $N(Q_{i})=N(R_{i})=1.$ If
701: ${\cal T}_{b}\not= \emptyset,$
702: define $Y=\min {\cal T}_{b}$ to be the smallest of the indices in
703: ${\cal T}_{b}$; otherwise, if ${\cal T}_{b}=\emptyset,$ set $Y=-1$.
704: %Let $\ell =\lfloor \log b\rfloor $, and
705: %define ${\cal Y}_{b}\subseteq
706: %{\cal T}_{b}$, a random subset of indices,
707: %as follows. If $\tau_{b}=
708: %|{\cal T}_{b}|< \ell ,$ then ${\cal Y}_{b}=\emptyset.$ Otherwise,
709: %if $\tau_b \geq \ell $ , let ${\cal Y}_{b}$ be the $\ell$ smallest
710: %indices in ${\cal T}_{b}.$ In other words, if
711: %${\cal T}_{b}=\bigl\lbrace i_1,i_2,\dots
712: %i_{\tau_{b}}\bigr\rbrace$ and
713: %$i_1<i_2<\dots <i_{\tau_{b}}, $
714: %then ${\cal
715: %Y}_{b}=\bigl\lbrace i_1,i_2, \dots ,i_{\ell}\bigr\rbrace $.
716:
717: Define the random variable $X_b$ as follows: If $\tau_{b}=|{\cal
718: T}_{b}|=0 $ then $X_{b}=0$; otherwise, if
719: ${\cal T}_{b}=\bigl\lbrace i_1,i_2,\dots
720: i_{\tau_{b}}\bigr\rbrace$ and $i_1<i_2<\dots <i_{\tau_{b}}, $
721: then
722: $X_{b}=1$ iff $Q_{i_{1}}\cup R_{i_{1}}$ contains a blue connected
723: dominating set for $\hat{{\cal H}}_{w,b}.$
724:
725: Let ${\cal B}=\bigl\lbrace g_{1},g_{2},\dots ,g_{b}\bigr\rbrace $
726: be the set of blue nodes, selected independently and uniform
727: randomly from $\hat{D}_{1}(o).$ Define ${\cal Z}={\cal
728: B}\bigcap D_{\delta}(o)$ to be set of blue points that fall near
729: the origin $o$, and let $Z=|{\cal Z}|$ be the number of these points.
730: Then
731:
732: \begin{equation}
733: {\Pr}_o(X_{b}=0)\leq {\Pr}_o\left(X_{b}=0,\tau_{b}\not=0, Z\leq {2\hat{\lambda}
734: b^{1/3}\over (\log b)^{2}}\right)
735: +{\Pr}_o(\tau_{b}=0)+{\Pr}_o\left(Z>{2\hat{\lambda}b^{1/3}\over (\log b)^{2}}
736: \right).
737: \end{equation}
738: Note that $Z$ has a binomial distribution: $ Z{\buildrel d\over
739: =}Bin(b,\hat{\lambda}\delta^{2})$ where $\hat{\lambda}$ is as defined in the
740: proof of Lemma \ref{taubound}. If
741: $\beta={2\hat{\lambda}b^{1/3}\over (\log b)^{2}}$, then by
742: Chernoff's inequality,
743: \begin{equation}
744: \label{Binomial} {\Pr}_o(Z\geq \beta)
745: \leq \exp(-b^{1/3}/4(\log b
746: )^{2}).
747: \end{equation}
748: By Lemma \ref{taubound}, ${\Pr}_o( \tau_{b}=0)= O({\log^{6}b\over
749: b^{1/3}}).$
750: Therefore
751: \begin{equation}
752: \label{RHS}
753: \label{Xr } {\Pr}_o(X_{b}=0)\leq {\Pr}_o(X_{b}=0,\tau_{b}\not=0,Z\leq
754: \beta) +O({\log^{6}b\over b^{1/3}}).
755: \end{equation}
756:
757: Now we decompose the first term on the right side of (\ref{RHS}) according to
758: the value of $Y$.
759: \begin{equation}
760: \label{onestar}
761: {\Pr}_o(X_{b}=0,\tau_{b}\not= 0, Z\leq \beta)=
762: \sum\limits_{k=0}^{L-1} {\Pr}_o( X_{b}=0| Y=k, Z\leq \beta){\Pr}_o(Y=k,
763: Z\leq \beta).
764: \end{equation}
765: (The redundant condition $\tau_{b}\not=0$ need not be included on
766: the right side of (\ref{onestar}) because it a consequence of
767: the condition $Y\geq 0.$) We have
768: \begin{equation}
769: {\Pr}_o( X_{b}=0| Y=k, Z\leq \beta) =\sum\limits_{S} {\Pr}_o(X_{b}=0|
770: {\cal Z}=S,Y=k) {\Pr}_o( {\cal Z}=S\bigl| Y=k, Z\leq \beta)
771: \end{equation}
772: where the sum is over subsets $S\subseteq [b]$ such that
773: $2\leq |S|\leq \beta .$
774:
775: \begin{equation}
776: \Pr(X_{b}=0| {\cal Z}=S,Y=k)=1-\Pr(X_{b}=1|{\cal Z}=S,Y=k),
777: \end{equation}
778: so it is enough to find a lower bound for $\Pr(X_{b}=1|{\cal
779: Z}=S,Y=k).$
780:
781: \commentout{ By Boole's inequality
782: \begin{equation}
783: \label{Boole2}
784: {\Pr}_o(X_{b}=0| {\cal Z}=S,{\cal Y}_{b}=Y)\leq
785: \ell(1-{\Pr}_o(J_{i_{1}}=1|{\cal Z}=S,{\cal Y}_{b}=Y)).
786: \end{equation}
787: }
788:
789:
790: To simplify notation, let $\gamma=X(\tilde{q}_0,\tilde{u}_0),$
791: and recall that $\gamma= O({1\over b\log^{3}b}).$
792: %%%%%%%%%%%%%%
793: %\commentout{In
794: %this section of the paper, define $\alpha={\pi \over b^{2/3}(\log
795: %b)^{2}}$ to the area of the disk $D_{\delta}(v)$, and
796: %$\hat{\alpha}(v)=Area(\hat{D}_{1}(v)).$ }%%%%%%%%%%%%%%
797: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
798: In this section of the paper, define $|{D}_{\delta}(o)|={\pi \over
799: b^{2/3}(\log b)^{2}}$ to be the area of the disk $D_{\delta}(o)$,
800: and let $|\hat{D}_{1}(o)|=Area(\hat{D}_{1}(o)).$ An important
801: observation is that, once we have specified $b-|S|=$ the {
802: number} of blue points that fall {\em outside} $D_{\delta}(o)$,
803: the locations in $D_{\delta}(o)^{c}\cap\hat{D}_1(o)$ of these $b-|S|$ points are
804: independent of the locations of the $|S|$ blue points {\sl in}
805: $D_{\delta}(o),$ and are also independent of the locations of the
806: white points. Hence
807: \begin{equation}
808: {\Pr}_o(X_{b}=1|{\cal Z}=S,Y=k)\geq {(1-{|{D}_{\delta}(o)|\over
809: |\hat{D}_{1}(o)|}-{\gamma\over |\hat{D}_{1}(o)| } )^{b-|S|} \over
810: (1-{|{D}_{\delta}(o)|\over |\hat{D}_{1}(o)|
811: })^{b-|S|}}\left(1-{\gamma\over |\hat{D}_{1}(o)| }\right)^{w}
812: \end{equation}
813: \begin{equation}
814: \geq \left(1-{C\over b(\log b)^{3}}\right)^{b-|S|+w}
815: \end{equation}
816: for some constant $C$ that is independent of $o$. With our
817: assumption $w<b(\log b)^{3/2}$
818: %and $|S|\leq \beta\leq
819: %{8b^{1/3}\over (\log b)^{2}}$ (since $\hat{\lambda}\leq 4$)
820: we get, for all sufficiently large
821: $b$, the lower bound
822: \begin{equation}
823: \label{oops}
824: {\Pr}_o(X_{b}=1|{\cal Z}=S,Y=k) \geq
825: \left(1-{C^{'}\over b(\log b)^{3}}\right)^{b(\log b)^{3/2}} \geq 1-{C^{''}\over
826: (\log b)^{3/2}}
827: \end{equation}
828: for some constants $C'$ and $C''$ which are independent of ${\cal Z}, Y$, and $o$.
829: Hence
830: \begin{equation}
831: {\Pr}_o(X_{b}=0)\leq {c\over (\log
832: b)^{3/2}}
833: \end{equation}
834: for some constant $c$ that is {\em independent of the point $o$.}
835:
836:
837:
838: \end{proof}
839:
840: \section{Analysis of Rule 2}
841: Let ${U}$ be the number of nodes that become non-gateways when
842: Rule 2 is applied to the random graph ${\cal G}_{n}$: ${
843: U}=\sum\limits_{i}I_{i}$ where (in this section)
844: the indicator variable $I_i=1$ %is an
845: %indicator that is one
846: iff the node with ID $i$ becomes a non-gateway under
847: Rule $2$. Assume that there is a positive constant $\bar{c}$ such that,
848: for all $n>1$, $\log n \leq \ell_{n}\leq \bar{c}\sqrt{{n\over \log n}}$.
849: Let $\xi_{n}={\alpha_{n}\over \ell_{n}^{2}}$, where $\langle
850: \alpha_{n}\rangle$ is any sequence of real numbers satisfying
851: the following three conditions:
852: \begin{itemize}
853: \item
854: $\alpha_{n}=o(n)$ as $n\rightarrow\infty.$
855: \item $\xi_{n}={\alpha_{n}\over
856: \ell_{n}^{2}}\rightarrow\infty$ as $n\rightarrow\infty.$
857: \item For all sufficiently large $n$,
858: ${16n\over \log^{3/2}\xi_{n}} < \alpha_{n}.$
859: \end{itemize}
860: For example, if $\ell_{n}=\Theta(\sqrt{n/\log n}),$
861: then the sequence
862: %$\alpha_{n}={n\over (\log\log n)^{\varrho}}$
863: $\alpha_{n}={32n\over (\log\log n)^{3/2}}$
864: satisfies the three
865: conditions. On the other hand, if $\ell_{n}=\Theta(({n/\log
866: n})^t)$ for some fixed positive $t<1/2$, then $\alpha_{n}={n\over
867: \log n}$ satisfies the three conditions above. With these three
868: assumptions, our goal is to prove
869: \begin{theorem}
870: \label{main}
871: $E({ U})\geq n -O(\alpha_{n}).$
872: \end{theorem}
873:
874: \begin{proof}
875: The idea of the proof is to use Theorem
876: \ref{localCDS} to bound the probability that a typical vertex $V_{i}$ is
877: pruned by Rule 2. In this case the
878: blue vertices correspond to nodes in $D_{1}(V_{i})$ with IDs {\em
879: higher} than $i$, and the white vertices correspond to nodes in
880: $D_{1}(V_{i})$ with lower IDs. Let $r={1\over \log^{3/2}
881: \xi_{n}},$ and let ${\cal A}_{i}$ be the event that
882: $D_{r}(V_{i})\subseteq {\cal Q}_{n}.$ Then
883: \begin{equation}
884: \label{calAi} \Pr({\cal A}_{i})={(\ell_{n}-2r)^{2}\over
885: \ell_{n}^{2}}\geq 1-{4r\over \ell_{n}}. \commentout{1-{4\over
886: \ell_{n}(\log \xi_{n})^{3/2}}.}
887: \end{equation}
888: Let $\hat{D}_{1}(V_{i})=D_{1}(V_{i})\cap {\cal
889: Q}_{n}$ be the set of points in ${\cal Q}_{n}$ whose distance from
890: $V_{i}$ is one or less, and let $|\hat{D}_{1}(V_{i})|$ be the area
891: of $\hat{D}_{1}(V_i).$ Let $\rho_{i}^{(b)}$ denote the number of nodes in $\hat{D}_{1}(V_{i})$
892: having a label that is larger than $i$, and let $\rho_{i}^{(w)}$
893: be the number of nodes in $\hat{D}_{1}(V_{i})$ having a label that is
894: smaller than $i$. Then, given the location of the $i$'th vertex
895: $V_{i}$,
896: $\rho_{i}^{(b)}$ has a Binomial$(n-i,
897: {|\hat{D}_{1}(V_{i})|\over \ell_{n}^{2}})$ distribution. Define
898: $\mu_{b}=\mu_{b}(i)$ to be the expected value of $\rho_{i}^{(b)}$
899: given the location of the $i$'th point:
900: \begin{equation}
901: \label{mub} \mu_{b}= E(\rho_{i}^{(b)}| V_{i}
902: )=
903: {(n-i)|\hat{D}_{1}(V_{i})|\over \ell_{n}^{2}}.
904: \end{equation}
905: Similarly
906: $\rho_{i}^{(w)}$ has a Binomial$(i-1,
907: {|\hat{D}_{1}(V_{i})|\over \ell_{n}^{2}})$ distribution, and we
908: define $\mu_{w}=\mu_{w}(i)$ to be the expected value:
909: \begin{equation}
910: \label{muw}\mu_{w}=
911: E(\rho_{i}^{(w)}|V_{i})={(i-1)|\hat{D}(V_i)|\over \ell_{n}^{2}}.
912: \end{equation}
913: If ${\cal A}_i$ occurs, then by Chebyshev's inequality,
914:
915: \begin{equation}
916: \label{concrhob}
917: \Pr(|\rho_{i}^{(b)}-\mu_{b}(i)|< {\mu_{b}\over 2}|{\cal A}_i)\geq
918: 1-{16\ell_{n}^{2}\over n-i},
919: \end{equation}
920: and similarly for $\rho_{i}^{(w)}.$
921:
922: If we let ${\cal D}_{i}$ be the event that {\em both} of the
923: inequalities $|\rho_{i}^{(b)}-\mu_{b}(i)|< {\mu_{b}\over 2}$ and
924: $|\rho_{i}^{(w)}-\mu_{w}(i)|< {\mu_{w}\over 2}$ are satisfied,
925: then
926: \begin{equation}
927: \label{DgivenA} \Pr({\cal D}_{i}|{\cal A}_{i})\geq
928: 1-{16\ell_{n}^{2}\over n-i}-{16\ell_{n}^{2}\over i-1}.
929: \end{equation}
930: Combining (\ref{DgivenA}) and (\ref{calAi}), we get
931: \begin{equation}
932: \label{DandA} \Pr({\cal D}_{i}\cap {\cal A}_{i})\geq
933: \left(1-{16\ell_{n}^{2}\over n-i}-{16\ell_{n}^{2}\over i-1}\right)
934: \left(1-{4r\over \ell_n}\right).
935: \commentout{(1-{4\over \ell_{n}(\log \xi_{n})^{3/2}}).}
936: \end{equation}
937:
938:
939:
940:
941: Now let $\lambda_{n}=n-\alpha_{n},$ then clearly
942: \begin{equation}
943: \label{sumoni}
944: E({U})\geq
945: \sum\limits_{i=\alpha_{n}}^{\lambda_{n}}\Pr(I_{i}=1) \geq
946: \sum\limits_{i=\alpha_{n}}^{\lambda_{n}}\Pr(I_{i}=1|{\cal
947: D}_{i}\cap {\cal A}_{i})\Pr({\cal D}_{i}\cap {\cal A}_{i})
948: \end{equation}
949: To obtain a lower bound for the right hand side of
950: inequality (\ref{sumoni}), we prove
951: %%%%%%%%%%%%%%%
952: %For each term in the sum, we can decompose the first factor
953: %according to the number of higher-labelled neighbors that $V_{i}$
954: %has. We obtain a lower bound rather than equality if we restrict
955: %the sum to $J=\bigl\lbrace k: {\xi_{n}\pi\over 8}\leq k \leq
956: %{3(n-\alpha_{n})\pi\over 2\ell_{n}^{2}}\bigr\rbrace$:
957: %\begin{equation}
958: %\label{restricted}
959: % \Pr(I_{i}=1|{\cal D}_{i}\cap {\cal A}_{i})\geq
960: % \sum\limits_{k\in J}\Pr(\rho_{i}^{(b)}=k)
961: % \Pr(I_{i}=1|\rho_{i}^{(b)}=k,{\cal D}_{i}\cap {\cal
962: %A}_{i}).
963: %\end{equation}
964: %Each term in (\ref{restricted}) has two factors. To estimate the
965: %second of these two factors, we prove
966: %%%%%%%%%%%%%%%
967: \begin{lemma}
968: \label{termtwo} There is a constant $\tilde{c}>0$ such that
969: for all sufficiently large $n$ and all $\alpha_n\leq i<\lambda_n$,
970: $\Pr(I_{i}=1| {\cal D}_{i}\cap {\cal A}_{i})\geq
971: 1-{\tilde{c}\over (\log \xi_{n})^{3/2}}.$
972: \end{lemma}
973: \begin{proof}We begin by noting that {\it given} the event
974: ${\cal D}_i\cap{\cal A}_i$ and $\alpha_n\leq i<\lambda_n=
975: n-\alpha_n$, we have
976: %To apply the local results from the preceding section, we must
977: %verify the hypotheses of Theorem \ref{localCDS}. If ${\cal
978: %D}_{i}\cap {\cal A}_{i}$ occurs, and if $i<\lambda_{n},$ then
979: \begin{equation}
980: \label{35}
981: \rho_{i}^{(w)}< {3\over 2}\mu_{w}(i)
982: ={3(i-1)|\hat{D}_1(V_i)|\over 2\ell_n^2}\leq
983: {3\pi n\over 2\ell_{n}^{2}}.
984: \end{equation}
985: %for some positive constant $c$.
986: Similarly
987: \begin{equation}
988: \label{**}
989: \rho_{i}^{(b)}> {1\over 2}\mu_{b}(i)={(n-i)|\hat{D}_1(V_i)|
990: \over 2\ell_n^2}> { \alpha_n\pi\over
991: 8\ell_{n}^{2}}={\xi_{n}\pi\over 8}
992: \end{equation}
993: It follows from inequalities (\ref{35}) and (\ref{**})
994: and from the conditions on the sequences
995: $\langle\xi_n\rangle$ and $\langle\alpha_n\rangle$ that, {\it given} ${\cal D}_i\cap
996: {\cal A}_i$ and $\alpha_n\leq i<\lambda_n$,
997: \begin{equation}
998: \rho_{i}^{(b)}(\log
999: \rho_{i}^{(b)})^{3/2}\geq \rho_{i}^{(w)}.
1000: \end{equation}
1001: Next we consider the conditional probability
1002: $\Pr(I_i=1 | \rho_i^{(b)}, \rho_i^{(w)},V_i, {\cal D}_i\cap {\cal A}_i)$
1003: where the values of $\rho_i^{(b)}$ and $\rho_i^{(w)}$ and
1004: the location of $V_i$ are consistent with the event
1005: ${\cal D}_i\cap {\cal A}_i$. In this case, it follows from
1006: inequality (\ref{**}) that
1007: \begin{equation}
1008: \label{***}
1009: \delta(\rho_i^{(b)}) :=
1010: {1\over (\rho_i^{(b)})^{1/3}\log
1011: (\rho_i^{(b)})}\leq {1\over
1012: (\xi_n/3)^{1/3}\log (\xi_n/3)}
1013: \leq {1\over (\log (\xi_n))^{3/2}}=r.
1014: \end{equation}
1015: Since the event ${\cal A}_i$ implies $D_r(V_i)\subseteq\Re_+^2$,
1016: it follows from (\ref{***}) that
1017: $D_{\delta(\rho_i^{(b)})}(V_i)\subseteq\Re_+^2$. Finally, it
1018: follows from Theorem 4 that for some fixed positive constant
1019: $\tilde{c}$
1020: \begin{equation}
1021: \label{!}
1022: \Pr(I_i=1 | \rho_i^{(b)}, \rho_i^{(w)},V_i, {\cal D}_i\cap {\cal A}_i)
1023: \geq 1-{c\over (\log(\rho_i^{(b)}))^{3/2}}
1024: \geq 1 -{\tilde{c}\over (\log(\xi_i^{(b)}))^{3/2}}
1025: \end{equation}
1026: for all sufficiently large $n$ and all $\alpha_n\leq i<\lambda_n$.
1027: The lemma now follows from (\ref{!}).
1028: \end {proof}
1029: %We also need to check that the distance from\ $V_{i}$ to the
1030: %boundary of ${\cal Q}_{n}$ is at least $\delta.$ But this follows
1031: %from the assumption that ${\cal A}_{i}$ occurs.
1032: %\end{proof}
1033: %\vskip0cm\noindent
1034: % Now, putting Lemma \ref{termtwo} into
1035: %(\ref{restricted}), we get
1036: %\begin{equation}
1037: % \Pr(I_{i}=1|{\cal D}_{i}\cap {\cal A}_{i})\geq
1038: %\sum\limits_{k\in J}\Pr(\rho_{i}^{(b)}=k) (1-O({1\over \log \xi_{n}}))=
1039: %(1-O({1\over \log \xi_{n}}))\Pr(\rho_{i}^{(b)}\in J).
1040: %\end{equation}
1041: %By (\ref{concrhob}), $\rho_{i}^{(b)}$ is concentrated on $J$: for
1042: %all $i\in [\alpha_{n},\lambda_{n}],$ $$\Pr(\rho_{i}^{(b)}\in
1043: %J)=1-O({\ell_{n}^{2}\over n-i })=1-O({\ell_{n}^{2}\over
1044: %\alpha_{n}}).$$ Hence
1045: %\begin{equation}
1046: %\label{penultimate}
1047: % \Pr(I_{i}=1|{\cal D}_{i}\cap {\cal A}_{i})=(1-O({1\over \log \xi_{n}}))
1048: %(1-O({\ell_{n}^{2}\over \alpha_{n} })).
1049: %\end{equation}
1050:
1051: Recall that $\lambda_{n}=n-\alpha_{n}$, that $\alpha_{n}=o(n),$
1052: that $\xi_{n}={\alpha_{n}\over \ell_{n}^{2}}\rightarrow\infty$ as
1053: $n\rightarrow\infty,$ and that for all sufficiently large $n$,
1054: $\alpha_{n}> {16n\over (\log\xi_n)^{3/2}}.$ \commentout{ ${4n\over
1055: \log^{3/2}\xi_{n}} <\alpha_{n}.$ } So it follows from
1056: Lemma \ref{termtwo}
1057: and (\ref{DandA}) and (\ref{sumoni}), that
1058:
1059: $$
1060: E(U)\geq n-2\alpha_{n}+o(\alpha_{n}). $$
1061: \end{proof}
1062:
1063:
1064: \section{Discussion}
1065:
1066:
1067: In this final section, assume $\ell_{n}=\Theta(({n\over \log
1068: n})^{t})$ for some fixed positive $t\leq {1\over 2}$.
1069: For all sufficiently large $n$, the
1070: expected size of the Rule 2 dominating set is at least
1071: $\ell_{n}^{2}/4$ (See Theorem 5 of \cite{HS}).
1072: There is a gap between this lower
1073: bound and the $O(\alpha_{n})$ upper bound in Theorem \ref{main}.
1074: For example, when $t=1/2$, the lower and upper bounds for the
1075: expected size of the Rule 2 dominating set are respectively
1076: $\Theta(n/\log n)$ and $\Theta(n/(\log\log n)^{3/2})$. For
1077: $t<{1/2}$ the gap is even wider: the lower and upper bounds are
1078: respectively $\Theta( ({n\over \log n})^{2t})$ and
1079: $\Theta({n\over \log n}).$ We conjecture that, in fact, the
1080: expected size of the Rule 2 dominating set is
1081: $\Theta(\ell_{n}^{2}).$
1082:
1083:
1084:
1085: \vfill\eject
1086: \noindent
1087: {\bf Acknowledgement} We thank Jie Wu for introducing us
1088: to this problem. Professor Wu made helpful suggestions regarding
1089: an early version of this paper.
1090:
1091:
1092: \begin{thebibliography}{9}
1093:
1094: \bibitem{AJV}C.Adjih,P.Jacquet, L.Viennot,Computing Connected
1095: Dominating Sets with Multipoint Relays, {\sl INRIA RR-4597}
1096: (2002).
1097: \bibitem{AWF} K.M.Alzoubi,P.J.Wan,and O.Frieder, Distributed
1098: Heuristics for Connected Dominating Sets in Wireless Ad Hoc
1099: Networks, {\sl Journal of Communications and Networks},{\bf 4 (1)}
1100: (2002) 1--8.%OK
1101: \bibitem{AppelRusso} M.J.B.Appel and R.P.Russo, The minimum vertex
1102: degree of a graph on uniform points in $[0,1]^{d},$ {\sl Advances
1103: in Appl.Probab} {\bf 29 (3)} (1997) 582--594.
1104: \bibitem{BJ}E.Baccelli and P.Jacquet, Flooding Techniques in
1105: Mobile Ad Hoc Networks, {\sl INRIA RR-5002} (2003).
1106: \bibitem{CW} T.Camp and B.Williams, Comparison of Broadcasting
1107: Techniques for Mobile Ad Hoc Networks, Proceedings of the 3'rd
1108: ACM International Symposium on Mobile Ad Hoc Networking and
1109: Computing, MobiHoc 2002 (2002) 194--205.
1110: %DBLP Bibliography Server
1111: \bibitem{CCCD} M.Cardei,X.Cheng,X.Cheng, D.Du,Connected Domination
1112: in Multihop Ad Hoc Wireless Networks, {\sl JCIS} (2002) 251--255.
1113: %DBLP Bibliography Server
1114: \bibitem{Span}B. Chen, K.
1115: Jamieson, H. Balakrishnan, R. Morris , Span: An Energy-Efficient
1116: Coordination Algorithm for Topology Maintenance in Ad Hoc Wireless
1117: Networks {\sl Proc. of the 6th ACM MOBICOM Conf.}, Rome, Italy,
1118: July 2001.
1119: \bibitem{CCJ} B.N. Clark,C.J. Colburn, and D.J.Johnson, Unit Disk
1120: Graphs,{\sl Discrete Mathematics} {\bf 86(1-3)} (1990) 165--177.%MATHSCI
1121: \bibitem{BD} B.Das and V. Bharghavan Routing in Ad-Hoc Networks
1122: Using Minimum Connected Dominating Sets, International Conference
1123: on Communications {\bf 1} (1997) 376--380. %DBLP Bibliography Server
1124: \bibitem{DaiWu} F.Dai and J.Wu, An Extended Localized Algorithm for Connected
1125: Dominating Set Formation in Ad Hoc Wireless Networks, {\sl IEEE
1126: Transactions on Parallel and Distributed Systems}, {\bf 15} (10)
1127: 2004. %AW
1128: \bibitem{DaiWu2} F.Dai and J.Wu, Performance analysis of broadcast
1129: protocols in ad hoc networks based on self pruning, {\sl IEEE
1130: Trans.on Parallel and Distributed Systems} {\bf 15 (11)} (2004). %AW
1131: \bibitem{Gilbert} E.N.Gilbert, Random Plane Networks,
1132: {\sl J.Soc.Indust.Appl.Math.} {\bf 9} (1961) 533.
1133: \bibitem{GW} A.Godbole and B.Wielund, On the Domination Number of
1134: a Random Graph, {\sl Electronic Journal of Combinatorics} {\bf 8}
1135: \#R37 (2001)
1136: \bibitem{GK} S.Guha and S.Khuller, Approximation algorithms for
1137: connected dominating sets, {\sl Algorithmica} {\bf 20 (4)} (1998)
1138: 374--387. %MATHSCI
1139: \bibitem{GuptaKumar} P.Gupta and P.R.Kumar,
1140: \lq\lq Critical power for asymptotic connectivity in wireless
1141: networks\rq\rq, in Stochastic Analysis, Control, Optimization and
1142: Applications, Birkhauser (1999) 547--566.%MATHSCI
1143: \bibitem{Hale} W.K.Hale, Frequency Assignment: Theory and Applications,{\sl
1144: Proc. IEEE} {\bf 68} (1980) 1497--1514. %
1145: \bibitem{HS} J.C.Hansen and E.Schmutz, The Expected size of the
1146: Rule k dominating set, {\sl submitted.}
1147: \bibitem{Hall} P.Hall, Introduction to the Theory of Coverage
1148: Processes, Wiley (1988).
1149: \bibitem{J} P.Jacquet, Analytical Results on
1150: Connected Dominating Sets in Mobile Ad Hoc Networks, {\sl INRIA
1151: RR-5173} (2004).%AW
1152: \bibitem{JLMV} P.Jacquet, A.Laouiti,P.Minet,L.Viennot, Performance
1153: of Mutltipoint Relaying in Ad Hoc Mobil Routing Protocols, In
1154: \lq\lq Networking 2002\rq\rq\, {\sl
1155: Lecture Notes in Computer Science} {\bf 2345} (2002) 387--398. %Springer
1156: \bibitem{Licht} D.Lichtenstein, Planar formulae and their uses,
1157: {\sl SIAM J.Comput.} {\bf 11(2)} (1982) 329--343.
1158: \bibitem{MBHRJ}M.V.Marathe,H.Breu,H.B.Hunt,S.S.Ravi,and
1159: D.J.Rosenkrantz, Simple Heuristics for Unit Disk Graphs, {\sl
1160: Networks} {\bf 25} no.2 (1995) 59--68. %Mathsci
1161: \bibitem{M} C.McDiarmid, Discrete mathematics and radio channel
1162: assignment, {\sl Recent Advances in algorithms and combinatorics,}
1163: 27--63, CMS Books Math (2003).%MATHSCINET
1164: \bibitem{M2}C.McDiarmid, Random channel assignment in the plane,
1165: {\sl Random Structures and Algorithms} {\bf 22 (2)} 187--212.%MATHSCI
1166: \bibitem{PL}W.Peng and X.Lu,On the reduction of broadcast
1167: redundancy in mobile ad hoc networks, {\sl Proceedings of the 1'st
1168: ACM International Symposium on Mobile Ad Hoc Networking}
1169: (2000)129--130.%ACM portal, dblp
1170: \bibitem{Penrose} Random Geometric Graphs, Oxford Studies in Probability {\bf 5},
1171: Oxford University Press, (2003)
1172: ISBN 0-19-850626-0.
1173: \bibitem{BDS}R.Sivakumar, B.Das, and V. Bharghavan, , Spine-based
1174: routing in ad hoc networks, {\sl Cluster Computing} {\bf 1 (2)}
1175: (1998) 237--248.%DBLP Bibliography Server
1176: \bibitem{Solomon} Solomon, Geometric Probability, {\sl CBMS-NSF Conference Series}
1177: {\bf 28} SIAM, 1978.
1178: \bibitem{SSZ} I. Stojmenovic, M. Seddigh, J. Zunic, Dominating sets and
1179: neighbor elimination based broadcasting algorithms in wireless
1180: networks, {\sl IEEE Transactions on Parallel and Distributed
1181: Systems }, Vol. {\bf 13, No. 1},(2002), 14-25.%AW
1182: \bibitem{NTCS} Y.C.Tseng,S.Y.Ni, Y.S.Chen, J.P.Sheu,The broadcast
1183: storm problem in a mobile ad hoc network, {\sl Wireless Networks}
1184: {\bf 8(2-3)} (2002) 153-167. %ACM Digital library
1185: \bibitem{WuLi} J. Wu and H.Li, On calculating connected dominating
1186: set for efficient routing in ad hoc wireless networks, {\sl
1187: Workshop on Discrete Algorithms and Methods for MOBILE Computing
1188: and Communications} (1999) 7--14.%ACM Digital library, AW, dblp
1189: \bibitem{Yao} A.C.C.Yao, On constructing spanning trees in $k$
1190: dimensional spaces and related problems, {\sl SIAM J.Computing}
1191: {\bf 11 (4)} (1982) 721--736.%MATHSCINET
1192: \end{thebibliography}
1193: \end{document}
1194:
1195:
1196:
1197:
1198:
1199:
1200:
1201:
1202: