1:
2: \documentclass[12pt]{article}
3:
4: \usepackage{amsfonts,amsmath}
5: \usepackage{latexsym}
6: \usepackage{times}
7: %\usepackage{mathptm}
8: \usepackage{calc}
9: \usepackage{psfrag}
10: %\usepackage[dvips]{epsfig}
11: \usepackage[dvips]{graphicx}
12: \usepackage[T1]{fontenc} % íslenskir stafir
13: \usepackage[utf8]{inputenc} % íslenskir stafir
14: \usepackage[icelandic,english]{babel}
15: \usepackage{epsfig}
16: \usepackage{psfrag}
17: %\usepackage{amsthm}
18: %\usepackage{amscd}
19: \usepackage{epsfig}
20: \setlength{\textwidth}{430pt}
21: \setlength{\textheight}{46\baselineskip}
22: \setlength{\topmargin}{10pt} \setlength{\oddsidemargin}{10pt}
23: %\usepackage{psfrag}
24: \selectlanguage{icelandic}
25:
26:
27: \textwidth 150mm
28: \textheight 230mm
29: \setlength{\unitlength}{0.01in}
30: \def\sepand{\rule{14cm}{0pt}\and}
31:
32: %\usepackage{epsf}
33:
34:
35: %\input{psfig}
36:
37: \input eqmacros
38: \input totalmacroe
39: %\English
40:
41: \begin{document}
42:
43: \selectlanguage{english}
44:
45: \topmargin 0pt
46: \oddsidemargin 5mm
47: \headheight 0pt
48: \topskip 0mm
49:
50:
51: \addtolength{\baselineskip}{0.5\baselineskip}
52:
53: \pagestyle{empty}
54:
55:
56: \hfill
57:
58:
59:
60: \vspace{1cm}
61:
62: \begin{center}
63:
64:
65: {\Large \bf The spectral dimension of random brushes}
66:
67:
68: \medskip
69: \vspace{1.5 truecm}
70:
71: %{\large \bf \today}
72:
73: \vspace{1.5 truecm}
74:
75: {\bf Thordur Jonsson and Sigurður Örn Stefánsson}
76:
77: \vspace{0.4 truecm}
78:
79: %\selectlanguage{english}
80:
81: The Science Institute, University of Iceland
82:
83: Dunhaga 3, 107 Reykjavik
84:
85: Iceland
86:
87: \vspace{.7 truecm}
88:
89: \vspace{1.3 truecm}
90:
91: \end{center}
92:
93: \noindent {\bf Abstract.} We consider a class of random graphs,
94: called random brushes, which are constructed by adding linear graphs
95: of random lengths to the vertices of $\bbZ^d$ viewed as a graph. We
96: prove that for $d=2$ all random brushes have spectral dimension
97: $d_s=2$. For $d=3$ we have ${5\over 2}\leq d_s\leq 3$ and for
98: $d\geq 4$ we have $3\leq d_s\leq d$.
99:
100:
101:
102: \newpage
103: \pagestyle{plain}
104:
105: \section{Introduction}
106: The generic structure of random geometrical objects is of interest
107: in many branches of physics ranging from condensed matter physics to
108: quantum gravity, see e.g.\ \cite{bookben} and \cite{book}. One of
109: the methods used to analyze such objects is to study diffusion or
110: random walk. Diffusion allows us to define a notion of dimension,
111: the spectral dimension, for random geometrical objects. In recent
112: years the spectral dimension of triangulations has been studied
113: numerically in quantum gravity
114: \cite{earlynumsim,numerics,scaling,specdim1,specdim2} and
115: analytically for certain classes of random trees \cite{tjjw,DJW,GT}.
116: In \cite{DJW} the spectral dimension of various ensembles of random
117: combs was calculated. In this article we generalize the monotonicity
118: results of \cite{DJW} which allows us to find bounds on the spectral
119: dimensions of a class of graphs which we call brushes and define below.
120:
121: Let $G$ be a connected, locally finite (i.e.\ each vertex has finitely many
122: nearest neighbours) rooted graph. All graphs that we consider will be
123: assumed to have this property. Let $p_G(t)$ be the probability
124: that a simple random walk on $G$ which starts at the root is back at
125: the root after $t$ steps. If \beq{specdim0} p_G(t) \sim t^{-d_s/2}
126: \eeq as $t\to\infty$ then we say that $d_s$ is the spectral
127: dimension of the graph $G$. The existence of $d_s$ is not guaranteed
128: for individual graphs but its ensemble average can be shown to be
129: well defined in many cases \cite{DJW,GT}. It is easy to see
130: that if the spectral
131: dimension exists then it is independent of the starting site of the
132: random walk.
133:
134: Let us view $\bbZ^d$ as a graph with $j,k\in\bbZ^d$ neighbours if
135: their distance is 1 and let the origin of $\mathbb{Z}^{d}$ be the
136: root.
137: %The probability of a random walk returning to the root on
138: %$\mathbb{Z}^{d}$ after $t$ steps behaves as
139: %$\begin {equation} \label {specdim}
140: %$p_{\bbZ^d}(t) \sim t^{-d/2}
141: %$\end {equation}
142: It is well known that the spectral dimension of $\mathbb{Z}^{d}$
143: is $d$. Let $N_l$ be a linear chain of length $\ell$, i.e., the graph
144: obtained be connecting nearest neighbours in $\{0,1,{\ldots}
145: ,\ell\}$ with a link. Let $0$ be the root of $N_\ell$. Similarly,
146: let $N_\infty$ be the infinite linear chain with root at $0$. A
147: $d$-brush is a graph constructed by attaching one of the graphs
148: $N_\ell$ to each vertex of $\bbZ^d$ by identifying the root of
149: $N_\ell$ with a vertex in $\bbZ^d$, $\ell\in \bbN_0\cup \{\infty\}$,
150: $\ell =0$ corresponding to the empty chain. In a brush $B$ we will
151: refer to $\bbZ^d$ as the base and the linear chains as bristles. A
152: random brush is defined by letting the length of the bristles be
153: identically and independently distributed by a probability measure
154: on $\bbN_0\cup \{\infty\}$. We see that the case $d=1$ corresponds
155: to the combs studied in \cite{DJW} which were shown to have a
156: spectral dimension in the interval $[1,{3\over 2}]$.
157:
158: For $d>1$
159: we will show that the spectral dimensions of random brushes satisfy
160: the following:
161: \bea \label{result} & d_s & = 2,~~~{\rm
162: if} ~~~d=2,\nonumber \\ {5\over 2} \leq & d_s & \leq 3,~~~{\rm if}
163: ~~~d=3,\\ 3 \leq & d_s &\leq d,~~~{\rm if} ~~~d\geq 4.\nonumber
164: \eea
165: Some
166: comments are in order.
167: We see that when $d\geq 3$, attaching the bristles to the base serves to lower
168: the spectral dimension since the spectral dimension of $\bbZ^d$ is
169: equal to $d$. This is opposite to the case of combs where the linear
170: chains tended to increase the spectral dimension. Intuitively this
171: can be understood in the following way. If there is a very long
172: bristle somewhere, a random walk can go up it and spend a long
173: time there before returning to the base which it must do eventually since
174: the bristles are recurrent. Once it returns to the base
175: it will go back to the root with nonzero probability. We will indeed
176: see below that adding a single infinite bristle to $\bbZ^d$ with
177: $d\geq 4$ will bring the spectral dimension down to $3$.
178: The two dimensional case is special because
179: $\bbZ^2$ is only marginally recurrent and the generating function
180: for $p_{\bbZ^2}(t)$ has a logarithmic singularity which is not changed by the
181: presence of bristles. Assuming that the spectral dimension of random
182: brushes can be calculated by mean field theory we show that the full
183: range of exponents in \rf{result} is realized.
184:
185: The paper is organized as follows. In the next section we define
186: the generating functions used to analyze the spectral dimension.
187: We then establish generalized monotonicity lemmas which are shown to
188: imply the stated bounds on $d_s$ in Section 4. Section 5 contains a
189: discussion of mean field theory for brushes. A final section contains
190: some comments.
191:
192: \section{Generating Functions}
193: Let $G$ be a graph and $p_G^1(t)$
194: the probability that a random walk is at the root at time $t$ for the
195: first time after $t = 0$. We define the return generating function
196: \begin {equation}
197: Q_G(z) = \sum_{t=0}^{\infty}p_G(t)z^t
198: \end {equation}
199: and the first return generating function
200: \begin {equation}
201: P_G(z) = \sum_{t=0}^{\infty}p^1_G(t)z^t.
202: \end {equation}
203: The generating functions are related by
204: \begin {equation} \label{QPrelation}
205: Q_G(z) = \frac{1}{1-P_G(z)}.
206: \end {equation}
207: If $G$ has a spectral dimension $d_s$ then
208: \begin {equation} \label {singrelation}
209: Q_G^{(n)}(z) \sim \left\{ \begin{array}{ll}
210: 1 & \textrm{if $n = d_s/2-1$}\\
211: (1-z)^{d_s/2-1-n} & \textrm{otherwise}\\
212: \end{array} \right.
213: \end {equation}
214: where $n$ is the smallest nonnegative integer for which
215: $Q_G^{(n)}(z)$ diverges as $z \rightarrow 1$. Similarly, the behaviour (\ref{singrelation}) implies that the spectral
216: dimension is $d_s$. Here
217: $f(y) \sim y^{\alpha}$ as $y \rightarrow 0$ means that for any $\epsilon > 0$ there exist
218: positive constants $c_1$ and $c_2$, which may depend on $\epsilon$,
219: such that
220: \begin {equation} \label {assym}
221: c_1y^{\alpha + \epsilon} \leq f(y) \leq c_2y^{\alpha - \epsilon}
222: \end {equation}
223: for $y$ small enough. Note that $f(y) \sim 1$ allows $f$ to have a logarithmic singularity at $0$.
224:
225: The function $P_G(z)$ is analytic in the unit disc and $|P(z)| < 1$
226: for $|z| < 1$. If $P_G(z) \rightarrow 1$ as $z \rightarrow 1$ then $Q_G(z)$ clearly
227: diverges in which case the random walk is recurrent and $d_s \leq 2$. If $P_G(z)
228: \not\rightarrow 1$ as $z\rightarrow 1$ then the random walk is
229: transient and $d_s \geq 2$. In the latter case we see that if some derivative
230: $Q^{(n)}(z)$ diverges as $z\rightarrow 1$ then $Q_G^{(n)}(z) \sim
231: P_G^{(n)}(z)$ as $z \rightarrow 1$.
232:
233: If a graph has the property that every random walk which begins and ends at the root has an even number of steps, as is the case for brushes and bristles, we have to replace $p_G(t)$ with $p_G(2t)$ in (\ref{specdim0}) and $z$ with $z^2$ on the right hand side of (\ref{singrelation}). Then it is convenient to introduce a variable $x = 1 - z^2 \in [0,1]$. We will use the variable $z$ for general graphs but the variable $x$ when dealing with brushes and bristles.
234:
235: We will need the following first return generating functions for the graphs $N_l$ and $N_\infty$ \cite{DJW}
236:
237: \begin{equation}\label{linear}
238: P_l(x) = 1-\sqrt{x}\frac{(1+\sqrt{x})^l - (1-\sqrt{x})^l}{(1+\sqrt{x})^l + (1-\sqrt{x})^l}
239: \end{equation}
240: and
241: \begin {equation} \label {geninfty}
242: P_\infty(x) = 1-\sqrt{x}.
243: \end {equation}
244:
245: Let $\mu$ be a probability measure on
246: $\mathbb{N}_{0}\cup\{\infty\}$. Let $\mathcal{B}^d$ be the set of
247: all $d$-brushes. We define a probability measure $\pi$ on
248: $\mathcal{B}^d$ by letting the measure of the set of $d$-brushes
249: $\Omega$ which have bristles at $n_1,n_2,...,n_k\in\bbZ^d$ of length
250: $\ell_1,\ell_2,...,\ell_k$ be
251: \begin {equation}
252: \pi(\Omega) = \prod_{i=1}^{k} \mu(l_i).
253: \end {equation}
254: The set $\mathcal{B}^d$ together with $\pi$ defines a random brush
255: ensemble. We
256: define the averaged generating functions
257: \begin {equation}
258: \overline{P}(x) = \langle P_B(x)\rangle_{\pi}
259: \end {equation}
260: and
261: \begin {equation}
262: \overline{Q}(x) = \langle Q_B(x)\rangle_{\pi}
263: \end {equation}
264: where $\langle \cdot \rangle_{\pi}$ denotes expectation with respect
265: to $\pi$. We say that a random brush has spectral dimension
266: $d_s$ if $\overline{Q}(x)$ obeys the relation (\ref{singrelation}) (after replacing $z$ with $z^2$ on the right hand side).
267:
268:
269: \section{Monotonicity} \label {chMon}
270:
271: \newtheorem{theo}{Lemma}
272:
273: Here we present the monotonicity results in a slightly more general setting
274: than is needed for the applications below. This is both for clarity
275: and potential applications to random graphs different from
276: the brushes.
277:
278: Let $G_1$ and $G_2$ be graphs such that $G_1$ can be
279: constructed from $G_2$ by attaching rooted graphs $F(i)$ by their
280: roots to sites $i\neq r$ of $G_2$. Let the roots of $G_1$ and $G_2$
281: be the same vertex (regarding $G_2$ as a subgraph of $G_1$).
282: The following result is a generalization of the Monotonicity Lemma of
283: \cite{DJW}.
284:
285:
286: \begin {figure}[h]
287: \begin {center}
288: \includegraphics[width=0.6\textwidth]{graf}
289: \caption{An example of a graph $G_1$ constructed from $G_2$ and the
290: $F(i)$'s.}
291: \end {center}
292: \end {figure}
293:
294: \begin {theo} \label{thmon1}
295:
296: With $G_1$ and $G_2$ defined as above and $G_1 \neq G_2$ we have
297: \begin {equation} \label {mon1}
298: P_{G_1}(z) \leq P_{G_2}(z)
299: \end {equation}
300: with equality if and only if all the $F(i)$'s are recurrent and z =
301: 1.
302: \end {theo}
303: Proof: For any graph $G$ we can write $P_{G}(z)$ as a weighted sum over all random walks
304: $\omega$ on $G$ which start and end at the root without intermediate visits
305: to the root (this condition is denoted '$\omega$: FR on $G$'). Each walk $\omega$ has a weight
306: \begin {equation}
307: W_G(\omega)=\prod_{t=0}^{|\omega|-1}\sigma_{G}(\omega_t)^{-1}
308: \end {equation}
309: where $\sigma_{G}(\omega_t)$ is the order of the vertex $\omega_t$
310: on $G$ where the walk is located at time $t$ and $|\omega |$ is
311: the number of steps in $\omega$. Each step of a walk has a factor $z$ associated with it so
312: \begin {equation} \label{f1}
313: P_{G}(z) = \sum_{\text{\scriptsize{$\omega$: FR on $G$}}}
314: W_G(\omega)z^{|\omega|}.
315: \end {equation}
316:
317: Now consider a random walk $\omega'$ on $G_1$ which starts at the
318: root. Let $\omega$ be the subwalk of $\omega'$ which only travels on
319: $G_2$. If we look at the walk $\omega$ at time $t$ and location
320: $\omega_t$ then $\omega$ can be a subwalk of many different walks
321: $\omega'$ which correspond to all possible excursions into the graph
322: $F(\omega_t)$ before returning back to the walk on $G_2$. The weight
323: of these excursions is
324: \begin {equation} \label{factor1}
325: \sum_{n=0}^{\infty}\Big(\frac{\sigma_{F(\omega_t)}(\omega_t)}{\sigma_{G_1}(\omega_t)}P_{F(\omega_t)}(z)\Big)^n
326: =
327: \frac{1}{1-\Big(\frac{\sigma_{F(\omega_t)}(\omega_t)}{\sigma_{G_1}(\omega_t)}P_{F(\omega_t)}(z)\Big)}
328: \end {equation}
329: where $n$ counts the number of visits to $\omega_t$ before
330: the walk leaves $\omega_t$ for another vertex on $G_2$
331: and the factor in front of
332: $P_{F(\omega_t)}(z)$ changes the order of the root of $F(\omega _t)$ to
333: $\sigma_{G_1}(\omega_t) = \sigma_{G_2}(\omega_t) +
334: \sigma_{F(\omega_t)}(\omega_t)$. The weight of the first step back
335: into $G_2$ after all the visits to $F(\omega_t)$ is
336: \begin {equation} \label {factor2}
337: \frac{z}{\sigma_{G_1}(\omega_t)}.
338: \end {equation}
339: Now replace the original weight
340: $\sigma_{G_2}(\omega_t)^{-1}z$ of $\omega$ at each point
341: $\omega_t\neq \omega_0$ by the product of the factors
342: (\ref{factor1}) and (\ref{factor2}). This newly weighted $\omega$
343: then accounts for every random walk on $G_1$ which has $\omega$ as a
344: subwalk on $G_2$. Thus we can write
345: \begin {eqnarray} \nonumber \label {rep}
346: P_{G_1}(z) &=& \sum_{\text{\scriptsize{$\omega$: FR on $G_2$}}}
347: \sigma_{G_2}(\omega_0)^{-1}z\prod_{t=1}^{|\omega|-1}\Big(\frac{z}{\sigma_{G_2}(\omega_t)+
348: \sigma_{F(\omega_t)}(\omega_t)(1
349: -P_{F(\omega_t)}(z))}\Big) \\
350: &=& \sum_{\text{\scriptsize{$\omega$: FR on $G_2$}}}
351: K_{G_1,
352: G_2}(z;\omega)W_{G_2}(\omega)z^{|\omega|}
353: \end {eqnarray}
354: where in the last step we defined
355: \begin {equation}\label{KK}
356: K_{G_1,G_2}(z;\omega) =
357: \prod_{t=1}^{|\omega|-1}\Big(\frac{\sigma_{G_2}(\omega_t)}{\sigma_{G_2}(\omega_t)+
358: \sigma_{F(\omega_t)}(\omega_t)(1 - P_{F(\omega_t)}(z))}\Big).
359: \end {equation}
360: Since $P_{F(\omega_t)}(z) \leq 1$ with equality if and only if $F(\omega_t)$
361: is recurrent and $z=1$ it is clear that $K_{G_1,G_2}(z;\omega) \leq
362: 1$ for all $z$ with equality if and only if all the graphs
363: $F(\omega_t)$ for a given $\omega$ on $G_2$ are recurrent and $z =
364: 1$. The inequality
365: (\ref{mon1}) follows.
366: \begin{flushright} $\square$
367: \end{flushright}
368:
369: \begin {theo} \label {thmon2}
370: Let $n\in\mathbb{Z}^+$ be such that $P^{(n-1)}_{G_2}(z)$
371: is continuous on the closed interval $[0,1]$.
372: If all the $F(i)$'s are recurrent then
373: for a given $z\in]0,1[$ there exists a $\xi\in ]z,1[$ such that
374: \begin {equation}
375: P^{(n)}_{G_1}(\xi) \geq P^{(n)}_{G_2}(\xi).
376: \end {equation}
377: \end {theo}
378: Proof: We define
379: \begin {equation}
380: H_{G_1,G_2}(z;n) = \sum_{\text{\scriptsize{$\omega$: FR on $G_2$}}}
381: K_{G_1,G_2}(z;\omega)W_{G_2}(\omega)
382: \frac{d^{n-1}}{dz^{n-1}}z^{|\omega|}
383: \end {equation}
384: where $K_{G_1,G_2}$ is defined as above. Every derivative of a (first) return generating function is a positive increasing function of $z\in[0,1[$ since the power series have no negative coefficients. It is easy to verify that the function $K_{G_1,G_2}$
385: has the same property. Therefore we get by
386: differentiating (\ref{rep}) $n$ times
387: \begin {eqnarray} \nonumber \label {long}
388: P^{(n)}_{G_1}(z) &=&
389: \sum_{i=0}^{n}\binom{n}{i}\sum_{\text{\scriptsize{$\omega$: FR on
390: $G_2$}}}K_{G_1,G_2}^{(i)}(z;\omega)W_{G_2}(\omega)
391: \Big(z^{|\omega|}\Big)^{(n-i)}
392: \\ \nonumber &\geq& \sum_{\text{\scriptsize{$\omega$: FR on
393: $G_2$}}}K_{G_1,G_2}(z;\omega)W_{G_2}(\omega)
394: \Big(z^{|\omega|}\Big)^{(n)}
395: \\ \nonumber &+& \text{\quad} n\sum_{\text{\scriptsize{$\omega$: FR on
396: $G_2$}}}K_{G_1,G_2}'(z;\omega)W_{G_2}(\omega)
397: \Big(z^{|\omega|}\Big)^{(n-1)}
398: \\ \nonumber &\geq& \sum_{\text{\scriptsize{$\omega$: FR on
399: $G_2$}}}K_{G_1,G_2}(z;\omega)W_{G_2}(\omega)
400: \Big(z^{|\omega|}\Big)^{(n)}
401: \\ \nonumber&+& \sum_{\text{\scriptsize{$\omega$: FR on
402: $G_2$}}}K_{G_1,G_2}'(z;\omega)W_{G_2}(\omega)
403: \Big(z^{|\omega|}\Big)^{(n-1)}
404: \\ &=& H'_{G_1,G_2}(z;n).
405: \end {eqnarray}
406:
407: With the same argument as in the proof of Lemma \ref{thmon1} it
408: holds that \\$H_{G_1,G_2}(z;n) \leq
409: P_{G_2}^{(n-1)}(z)$ with equality when $z=1$ since all
410: the $F(i)$'s are recurrent and because $P_{G_2}^{(n-1)}(z)$ and
411: therefore also $H_{G_1,G_2}(z;n)$ are continuous on $[0,1]$. Since $H_{G_1,G_2}(z;n)$ and
412: $P_{G_2}^{(n-1)}(z)$ are positive and increasing functions
413: of $z$ we find that
414: \begin {equation}
415: \frac{H_{G_1,G_2}(1;n)-H_{G_1,G_2}(z;n)}{P_{G_2}^{(n-1)}(1) -
416: P_{G_2}^{(n-1)}(z)} \geq 1.
417: \end {equation}
418: By a generalized mean-value theorem there exists a $\xi \in ]z,1[$
419: such that
420: \begin {eqnarray} \label {hospital}
421: \frac{H_{G_1,G_2}(1;n)-H_{G_1,G_2}(z;n)}{P_{G_2}^{(n-1)}(1) -
422: P_{G_2}^{(n-1)}(z)} = \frac{H'_{G_1,G_2}(\xi;n)}{P_{G_2}^{(n)}(\xi)}.
423: \end {eqnarray}
424: In view of (\ref{long}) the Lemma follows.
425: \begin{flushright} $\square$
426: \end{flushright}
427:
428: \newtheorem {cor} {Theorem}
429:
430: \begin {cor} \label {corr}
431: Assume that all the $F(i)$'s are recurrent and that $G_1$ and $G_2$
432: have spectral dimensions $d_{s_1}$ and $d_{s_2}$ respectively. If
433: $G_2$ is recurrent then $G_1$ is recurrent and $d_{s_1} \geq
434: d_{s_2}$. If $G_2$ is transient then $G_1$ is transient and $d_{s_1}
435: \leq d_{s_2}$.
436: \end {cor}
437: Proof: Since all the $F(i)$'s are recurrent we have $P_{G_1}(1) = P_{G_2}(1)$ and therefore if $G_2$ is transient/recurrent then so is $G_1$. First assume that $G_2$ is recurrent. Then by using Lemma 1 and Equations (\ref{QPrelation}), (\ref{singrelation}) and (\ref{assym}) we find that for any $\epsilon > 0$ there exist positive constants $c_1$ and $c_2$ which may depend on $\epsilon$ such that
438: \begin{equation}
439: c_1 (1-z)^{d_{s_1}/2-1+\epsilon} \leq Q_{G_1}(z) \leq Q_{G_2}(z) \leq c_2 (1-z)^{d_{s_2}/2 - 1 - \epsilon}
440: \end{equation}
441: for $z$ close to 1. If $d_{s_1} \neq d_{s_2}$ we choose $\epsilon < \frac{1}{4} |d_{s_2}-d_{s_1}|$ and send $z \rightarrow 1$ to conclude that $d_{s_1} > d_{s_2}$. When $G_2$ is transient we use Lemma 2 and similar arguments as above to show that $d_{s_1} \leq d_{s_2}$.
442: \begin{flushright} $\square$
443: \end{flushright}
444:
445:
446:
447: \section {The Spectral Dimension}
448: The $d$-brush where every bristle is $N_\infty$ we call
449: the full $d$-brush and denote it $\ast d$.
450: We can relate the generating function of the full $d$-brush to the
451: generating functions of $\mathbb{Z}^{d}$ and $N_{\infty}$. We use
452: the same argument as in the proof of Lemma 1.
453: Replacing all the graphs $F(i)$ with
454: $N_{\infty}$ and noting that the order of every point in
455: $\mathbb{Z}^{d}$ is $1/2d$ we get
456: \begin{equation} \label{f2}
457: P_{\ast d}(x) =
458: \Big(1+\frac{1-P_{\infty}(x)}{2d}\Big)P_{\mathbb{Z}^{d}}(x_{\text{ren}}(x))
459: \end {equation}
460: where $x_{\text{ren}}$ is defined by
461:
462: \begin {equation} \label{f3}
463: \sqrt{1-x_{\text{ren}}} =
464: \frac{\sqrt{1-x}}{1+\frac{1-P_{\infty}(x)}{2d}}.
465: \end {equation}
466: We see that $x_{\text{ren}} = \sqrt{x}/d + O(x).$ By differentiating
467: (\ref{f2}) once and comparing with (\ref{singrelation}) we find the
468: spectral dimension of the full brush
469: \begin {equation} \label {BS}
470: d_{\ast} = \left\{ \begin{array}{ll}
471: \frac{d}{2} + 1 & \quad\textrm{if $1 \leq d \leq 4$}\\
472: 3 & \quad\textrm{if $d \geq 4$.}\\
473: \end{array} \right.
474: \end {equation}
475: If we replace the infinite bristles with finite ones, all of which
476: have the same length, then with the same calculation we see that the
477: spectral dimension remains equal to $d$. These are special cases of a
478: more general result obtained in \cite{italir} for so called bundled
479: structures. There, the base $\mathbb{Z}^{d}$ can be replaced by any
480: graph $B$ and the infinite bristle (fiber) can also be replaced by
481: any fixed graph $F$.
482:
483: Using the above calculation and Theorem 1 we can
484: find bounds on the spectral dimensions of fixed and random
485: $d$-brushes. Any fixed $d$-brush $B$ can be constructed from
486: $\mathbb{Z}^{d}$ by attaching (recurrent) bristles to it and the
487: full $d$-brush can be constructed from $B$ by attaching (recurrent)
488: bristles to it. Therefore, by Theorem 1,
489: the spectral dimension of any fixed $d$-brush, if it
490: exists, lies between $d$ and $d_{\ast}$. This also holds for random
491: brushes as is clear from equations (\ref{ramon1}) and
492: (\ref{ramon2}) below and the proof of Theorem 1.
493: The spectral dimension for any fixed or random
494: $d$-brush, if it exists, therefore obeys the inequalities
495: (\ref{result}).
496:
497: The spectral dimension of random
498: 2-brushes always equals 2.
499: Indeed it follows from the fact that $Q_{\mathbb{Z}^{2}}(x)$ is asymptotic to
500: $|\ln (x)|$ as $x\to 0$ and Lemma 1
501: that there exist positive constants $c_1$ and $c_2$ such that
502: \begin {equation}
503: c_1|\ln(x)| \leq \overline{Q}(x) \leq c_2|\ln(x)|
504: \end {equation}
505: when $x$ is small enough . This is a stronger
506: condition on the asymptotic behavior of $\overline{P}(x)$
507: than $\overline{P}(x)\sim 1$ as
508: $x\rightarrow 0 $.
509:
510: It is interesting that for $d \geq 4$ the lower bound on the
511: spectral dimension always equals 3. In fact it is easy to see that
512: attaching a single infinite bristle to $\mathbb{Z}^d$ with $d \geq
513: 4$ reduces the spectral dimension to 3. We can show this by
514: attaching an infinite bristle to the root of $\mathbb{Z}^d$ since
515: the spectral dimension is independent of the starting site of the
516: random walks. Let us call the resulting brush
517: $\bot d$. The first return generating function is simply
518: \begin {equation} \label{add}
519: P_{\bot d}(x) = \frac{2d}{2d+1}P_{\mathbb{Z}^d}(x) +
520: \frac{1}{2d+1}P_{\infty}(x).
521: \end {equation}
522: Since
523: $d \geq 4$ equation (\ref{singrelation}) shows
524: that $Q'_{\mathbb{Z}^{d}}(x)$ diverges slower than any negative
525: power of $x$ as $x
526: \rightarrow 0$ but
527: $Q'_{\infty}(x) \sim x^{-1/2}$.
528: Therefore by differentiating (\ref{add}) we get
529: \begin {equation} \label{Qprime}
530: Q'_{\bot d}(x) \sim x^{-1/2}
531: \end {equation}
532: as $x \rightarrow 0$ and therefore by (\ref{singrelation}) the
533: spectral dimension equals $3$. It follows
534: that if a random $d$-brush with $d\geq 4$ has a nonzero probability
535: of having one or more infinite bristles its spectral dimension
536: equals 3.
537:
538: We find with similar arguments that adding a single (or finitely
539: many) infinite bristles to $\mathbb{Z}^3$ gives the spectral dimension 3.
540: However, if we add infinitely many bristles the spectral dimension
541: of $\mathbb{Z}^3$
542: can be lowered as is seen e.g.\ in the case of the full 3-brush.
543:
544: We now use the notation of Section 3 and
545: consider the case when $G_2 = \mathbb{Z}^d$ and instead
546: of having a fixed $G_1$ we take a random $d$-brush.
547: We would like to get bounds for the spectral dimension of
548: random brushes similar to those in Theorem 1. First
549: we note that by Lemma \ref{thmon1} we have for any
550: $B\in\mathcal{B}^d$ that
551: \begin {equation}
552: P_{\ast d}(x) \leq P_B(x) \leq P_{\mathbb{Z}^d}(x)
553: \end {equation}
554: and averaging we get
555: \begin {equation} \label {ramon1}
556: P_{\ast d}(x) \leq \overline{P}(x) \leq P_{\mathbb{Z}^d}(x).
557: \end {equation}
558: In order to generalize Lemma
559: \ref{thmon2} to random brushes we consider the case $d > 2$ and define the
560: functions
561: \begin {equation}
562: \overline{H}_a(x;n) = \langle H_{B,\mathbb{Z}^d}(x;n)\rangle_\pi \quad
563: \quad \text{and} \quad \quad \overline{H}_b(x) = \langle H_{\ast
564: d,B}(x;1)\rangle_\pi
565: \end {equation}
566: where $n=[{d-1\over 2}]$ is the smallest positive integer for which
567: $P_{\mathbb{Z}^d}^{(n)}(x)$ diverges as $x \rightarrow 0$. With the
568: same calculation as in (\ref{long}) we get
569: \begin {equation}
570: \frac{\overline{H}'_a(x)}{\overline{P}^{(n)}(x)} \leq 1 \quad \quad
571: \text{and} \quad \quad\frac{\overline{H}'_b(x)}{P'_{\ast d}(x)} \leq
572: 1.
573: \end {equation}
574: We clearly have $(-1)^{n-1}\overline{H}_a(x) \leq
575: (-1)^{n-1}P_{\mathbb{Z}^d}^{(n-1)}(x)$ and $\overline{H}_b(x) \leq
576: \overline{P}(x)$ both with equality when $x=0$. Since the functions
577: $(-1)^{n-1}\overline{H}_a(x)$,$(-1)^{n-1}P_{\mathbb{Z}^d}^{(n-1)}(x)$,
578: $\overline{H}_b(x)$
579: and
580: $\overline{P}(x)$ are all decreasing functions of $x$ we get with
581: the same argument as in the proof of Lemma 2 that for a given
582: $x\in]0,1[$ there exists a $\xi\in]0,x[$ such that
583: \begin {equation} \label {ramon2}
584: 1 \leq \frac{\overline{P}^{(n)}(\xi)}{P_{\mathbb{Z}^d}^{(n)}(\xi)}
585: \quad \quad \text{and} \quad \quad 1 \leq \frac{P'_{\ast
586: d}(\xi)}{\overline{P}'(\xi)}.
587: \end {equation}
588: This extends Theorem 1 to random brushes and establishes the bounds
589: \rf{result}.
590:
591: \section{Mean Field Theory}
592: It is an obvious question to ask whether the full range of spectral
593: dimensions allowed by \rf{result} is realized for some random brushes.
594: We do not have an answer to this question. However, in \cite{DJW} the
595: spectral dimensions for different classes of random combs were
596: calculated exactly and shown to take the same values as in mean field
597: theory \cite{mft}. By mean field theory we mean that the walk on the base (spine
598: in the case of combs)
599: always sees a new bristle drawn from the
600: probability distribution $\mu$ whenever it is located at the root of a
601: bristle. Since mean field theory is exact in one dimension we find it
602: likely that it is also exact in higher dimensions where the walks are
603: less likely to visit the same points on the base often. Mean field theory
604: allows us to evaluate the spectral dimension very easily as we now
605: explain.
606:
607: The ensemble average of the function $K_{G_1,G_2}$ defined in
608: \rf{KK} can be written
609: \begin {eqnarray} \nonumber
610: \langle K_{B,\mathbb{Z}^d}(x;\omega) \rangle_\pi &=&
611: \Big\langle \prod_{t=1}^{|\omega|-1}\frac{2d}{2d+
612: 1 - P_{F(\omega_t)}(x)}\Big\rangle_\pi \\ &\stackrel{\text{\scriptsize{m.f.t.}}}{
613: =}& \Big(\Big\langle\frac{2d}{2d+
614: 1 - P_{l}(x)}\Big\rangle_\mu\Big)^{|\omega| - 1}.
615: \end {eqnarray}
616: where the second equality is the mean field theory approximation.
617: The mean field theory approximation to the
618: first return generating function is
619: \begin {equation} \label{mft}
620: \overline{P}_{\text{\scriptsize{m.f.t.}},d}(x) =
621: \Big\langle\frac{2d}{2d+1-P_{l}(x)}\Big\rangle_\mu^{-1}
622: P_{\mathbb{Z}^{d}}(x_{\text{ren}}(x))
623: \end {equation}
624: where $x_{\text{ren}}(x)$ is defined through
625: \begin {equation}
626: \sqrt{1-x_{\text{ren}}(x)} = \Big\langle\frac{2d}{2d+
627: 1 - P_{l}(x)}\Big\rangle_\mu \sqrt{1-x}.
628: \end {equation}
629: Now choose $\mu(l) = c_a l^{-a}$ with $a>1$. The cases $d=1$ and $d=2$ we
630: understand. Therefore
631: consider the case $d\geq 3$. It is straightforward to calculate
632: the asymptotic behaviour of the following derivatives:
633: \begin {equation}
634: \langle P_l^{(n)}(x)\rangle_\mu \sim x^{a/2-n} ~~~~~~~~~~~~ \text{for $n\geq 1$},
635: \end {equation}
636: \smallskip
637: \begin {equation}
638: x_{\text{ren}}(x) \sim \left\{ \begin{array}{ll}
639: x^{a/2} & \\
640: x & ,\\
641: \end{array} \right.\text{}~~~~~~~ ~~~~~~~ x_{\text{ren}}'(x)
642: \sim \left\{ \begin{array}{ll}
643: x^{a/2-1} & ~~~~~~\quad\textrm{if $1 < a \leq 2$}\\
644: 1 & ~~~~~~\quad\textrm{if $a > 2$}\\
645: \end{array} \right.
646: \end {equation}
647: \smallskip
648: and
649: \smallskip
650: \begin {equation}
651: x_{\text{ren}}^{(n)}(x) \sim x^{a/2-n} ~~~~~~~~~~~~ \text{for $n\geq 2$}
652: \end {equation}
653: \smallskip
654: when $x \rightarrow 0$. We also see that the leading behaviour of the
655: $n$-th derivative of (\ref{mft}) is
656: \begin {equation}
657: \overline{P}^{(n)}_{\text{\scriptsize{m.f.t.}},d}(x) \sim \langle
658: P_l^{(n)}(x)\rangle_\mu + P^{(n)}_{\mathbb{Z}^d}(x_{\text{ren}}(x))(x'_{
659: \text{ren}}(x))^n.
660: \end {equation}
661: First consider the case $d=3$, when we only have to look at the first
662: derivative. Then $P'_{\mathbb{Z}^3}(x) \sim x^{-1/2}$ as
663: $x\rightarrow0$ and therefore
664: \begin {equation}
665: \overline{P}'_{\text{\scriptsize{m.f.t.}},3}(x) \sim \left\{ \begin{array}{ll}
666: x^{a/4 - 1} & ~~~~~~\quad\textrm{if $1 < a \leq 2$}\\
667: x^{-1/2} & ~~~~~~\quad\textrm{if $a > 2$}\\
668: \end{array} \right.
669: \end {equation}
670: which gives
671: \begin {equation}
672: d_s = \left\{ \begin{array}{ll}
673: \frac{a}{2} + 2 & ~~~~~~\quad\textrm{if $1 < a \leq 2$}\\
674: 3 & ~~~~~~\quad\textrm{if $a > 2$}.\\
675: \end{array} \right.
676: \end {equation}
677: Doing the same for $d\geq4$ we get the result
678: \begin {equation}
679: d_s = \left\{ \begin{array}{ll}
680: a + 2 & ~~~~~~\quad\textrm{if $1 < a \leq d-2$}\\
681: d & ~~~~~~\quad\textrm{if $a > d-2$}.\\
682: \end{array} \right.
683: \end {equation}
684: It is
685: easy to see that putting a single bristle on $\mathbb{Z}^d$ with probability
686: distribution $\mu$ for $d\geq 4$ gives the same spectral dimension as mean
687: field theory.
688:
689: Now consider the random brush defined by $\mu(\infty) = p > 0$ and $\mu(0) = 1-p$. It was shown in \cite{DJW} that for $d=1$ the spectral dimension of this random brush equals the spectral dimension of the full brush. The same is of course true for $d=2$ and as well for $d\geq 4$, as was noted in the discussion below (\ref{Qprime}). Using mean field theory and similar analysis as above, we find that in any dimension the resulting random brush has also the same spectral dimension as the full brush. It is therefore clear that for this class of random brushes, if $d\neq3$, mean field theory gives the correct spectral dimension. Settling the case $d=3$ would require some extra work.
690:
691: \section{Conclusions}
692: We have established bounds on the spectral dimensions of random graphs
693: constructed by attaching linear graphs to ${\mathbb Z}^d$ and argued
694: that mean field theory is likely to give the right value for the
695: spectral dimension. The main monotonicity results are in fact valid
696: for a much larger class of graphs as explained in Section 3; the base
697: can be arbitrary and the bristles need only be recurrent graphs.
698:
699: While our random brushes do contain loops, they are all on the base
700: which is nonrandom and therefore do not yield much insight into how
701: one might hope to bound or evaluate the spectral dimension of random
702: graphs that contain loops like e.g.~random surfaces. For such graphs we need to develop new techniques.
703:
704: \noindent {\bf Acknowledgment.} This work is supported in part by
705: Marie Curie grant MRTN-CT-2004-005616, the Icelandic Science Fund
706: and the University of Iceland Research Fund. We would like to thank
707: Bergfinnur Durhuus and John F.\ Wheater for many discussions and helpful comments on the manuscript.
708:
709: \bigskip
710:
711: \begin{thebibliography}{99}
712:
713: \bibitem{bookben}D.~ben-Avraham and S.~Havlin, {\it Diffusion and
714: reactions in fractals and disordered systems}, Cambridge University
715: Press, Cambridge, (2000)
716:
717: \bibitem{book}J.~Ambj\o rn, B.~Durhuus and T.~Jonsson, {\it Quantum
718: geometry: a statistical field theory approach,} Cambridge University
719: Press, Cambridge (1997)
720:
721: %\bibitem{coulhon}T.~Coulhon, {\it Random walks and geometry
722: %on infinite graphs}, Lecture notes on analysis on metric spaces,
723: %Trento, C.I.M.R., 1999, Ed. L.~Ambrosio and F.~S.~Cassano. Scuola
724: %Normale Superiore di Pisa (2000)
725:
726: \bibitem{DJW}B.~Durhuus, T.~Jonsson and J.~Wheater, {\it Random walks
727: on combs,} J. Phys. {\bf A39} (2006) 1009-1038
728:
729: \bibitem{GT}B.~Durhuus, T.~Jonsson and J.~Wheater, {\it The spectral
730: dimension of generic trees,} [math-ph/0607020] J. Stat. Phys. in press.
731:
732: \bibitem{earlynumsim}J.~Ambj\o rn, J.~Jurkiewicz and Y.~Watabiki,
733: {\it On the fractal structure of two-dimensional quantum gravity},
734: Nucl. Phys. {\bf B 454} (1995) 313-342
735:
736: \bibitem{numerics}J.~Ambj\o rn, K.~N.~Anagnastopoulos, T.~Ichihara,
737: T.~Jensen and Y.~Watabiki, {\it Quantum geometry and diffusion},
738: JHEP {\bf 11} (1998) 022
739:
740: \bibitem{scaling}J.~Ambj\o rn, D.~Boulatov, J.~L.~Nielsen, J.~Rolf and
741: Y.~Watabiki, {\it The spectral dimension of 2D quantum gravity,}
742: JHEP 9802 (1998) 010
743:
744: \bibitem{specdim1}J.~Ambj\o rn, J.~Jurkiewicz and R.~Loll, {\it Spectral
745: dimension of the universe,} Phys.~Rev.~Lett. {\bf 95} (2005) 171301
746: [hep-th/0505113]
747:
748: \bibitem{specdim2}J.~Ambj\o rn, J.~Jurkiewicz and R.~Loll, {\it
749: Reconstructing the universe,} Phys.~Rev.~D {\bf 72} (2005) 064014
750: [hep-th/0505154]
751:
752: \bibitem{tjjw}T.~Jonsson and J.~Wheater, {\it The spectral dimension of the
753: branched polymer phase of two-dimensional quantum gravity}, Nucl.
754: Phys. {\bf B 515} (1998) 549-574
755:
756: %\bibitem{jwc}J.~D.~Correia and J.~F.~Wheater, {\it The spectral dimension of
757: %non-generic branched polymer ensembles,} Phys. Lett. B {\bf 422}
758: %(1998) 76-81
759:
760: \bibitem{italir} D. Cassi and S. Regina. \emph{Random
761: walks on bundled structures,} Phys. Rev. Lett. \textbf{76},
762: 2914-2917 (1996).
763:
764: \bibitem{mft} S. Havlin, J. E. Kiefer and G.H. Weiss. {\it Anomalous diffusion on a random comblike structure,} Phys. Rev. A {\bf 36} (1987) 1403-8.
765:
766: %\bibitem{donetti2}L.~Donetti and C.~Destri, {The statistical-
767: %geometry of
768: %scale-free random trees}, J. Phys. A: Math. Gen. {\bf 37} (2004) 6003-6025
769: %[cond-mat/0308624]
770:
771: %\bibitem{exotic}Z.~Burda, J.~Erdmann, B.~Petersson and M.~Wattenberg, {\it
772: %Exotic trees}, Phys.\ Rev.\ E {\bf 67} (2003) 026105
773:
774: %\bibitem{grigoryan}A.~Grigoryan and T.~Coulhon, {\it Pointwise estimates for
775: %transition probabilities of random walks in infinite graphs}, in: Trends in
776: %mathematics: Fractals in Graz 2001, Ed.\ P.~Grabner and W.~Woess.
777: %Birkh\"{a}ueser (2002)
778:
779:
780: \end{thebibliography}
781: \end{document}
782: