Professional Documents
Culture Documents
4, 1993
1. I N T R O D U C T I O N
Recently, Teles 5 used some techniques in electrical network to make an
excellent argument for random walks on infinite graphs. To be precise, let
G= (V, E) be an infinite graph, {Xn}n~>o be the simple random walk on G
starting at a fixed vertex x s V. Furthermore, let dR, d and dn be respec-
tively the random walk (RW) dimension, ffactal dimension and resistance
dimension, which are defined in [Ref. 5, Section 4] (or Section 2). Then,
Telcs 5 proved that
a) d>~dn, i f G i s dense
b) dR=d+2-dn>~2, if G is smooth
where the definitions of dense and smooth are given in Ref. J [Def. 5].
635
2d
d, = ~--s (1.1)
where the exponent d2 is the spectra of the graph G (see Ref. 6 or Section 2).
The second remark is to deduce a reasonable estimate on the hitting
time of random walk in terms of resistance, which is stronger than the
relation b), for some nice graphs (see Proposition 3.2 and Proposition 3.3).
A relation with this type was already used by Barlow and Bass to get a
reasonable estimate for the transition density of Brownian motion on the
Sierpinski carpet in R 2, which leads to the existence of spectral dimension
for this fractal (see Ref. 1). In fact, similar results can also be obtained by
using the reasonable estimate on hitting time for some strongly recurrent
fractals (e.g., see Ref. 3).
Resistance, Random Walk, and Fractal Dimensions 637
2. A G E N E R A L R E S U L T
Assign a resistor with resistance R~y= 1 to the edge (x, y) for each
(x, y)~E, and define the simple random walk {X.}.>~o (or denote it by
(py~)) on G with a fixed starting point X0 = x by
~(deg(y)) 1 (y, z) ~ E
Pyz = P( X. + 1= z IX. = y) = (0, otherwise,
where
deg(y)=#{zeVb(y,z)6E}, VyEV
Further, we let
Tx=inf(keN[X~=x},
Tx, u = i n f { k c N I X k ~ S x , N},
Rx(G)= lim R(x, Sx, N),
N ~ o~
where R(x, Sx, N) is the effective resistance of Bx, N between x and Sx, N
which can be defined as follows (see Ref. 2)
AN=DN--AN,
v(z) p y ~ - V ( y ) = O , VyEB(Nm)\({x}uS(Nm));
x: (y,z) ~ E
where
To prove the main result in this section, we begin with two lemmas.
Lemma 2.1. For any locally finite graph G, the following assertion is
correct
dn<.d
Proof By [Ref. 5, Lemma 1], we easily know that
By definition, we have
where eN and e~v~ 0 as N ~ oo. So we have dn ~<d for the case: Rx(G)= oo.
We now prove dn ~<d for the case: Rx(G)< oo. For this purpose, we
choose Ni = 2 s for i >~0. Then
In Ni
lim - - - 1
i~oo lnNi_l
and for any N e [Ni_l, Ni)
- -b N~ <In -b(Ni)
in - ~ < In- b(Ni)
- in Ni
InN lnNi_x l n N i lnNi 1"
d = li--~ In b(N,)
i ~ oo in Ni
640 Zhou
2 In (NNk--~k
- 1) In b(Nk) lnNk
>~ ~ + 2 - - - x - -
In Ark_ 1 In NK In N k 1
Clearly, we have
2 1 n ( Ark - - 1 )
\Nk- 1
lira -- 0
k~ co In Nk_ 1
Thus,
Therefore
2-da> 2-d
Namely, da ~< d. 9
( 2-1-~i~ ~
In R(x, Sx, Ni)
In N i
if R,( G ) = oo
dn = ~ ln(Rx(G) - R(x, S(N,)))
R~.(G) < O0
(2-1Tmi ~ 0~ In Ni , if
d R = d + 2-d~>~2
R~,(G) = ~ (2.2)
Clearly, (vy) is the potential on the electrical network Bm with a unit total
current flow from x to Sm. The proof of [Ref. 5, Theorem 2] states
dR<~ d+ 2 - d a
On the other hand, one gets that from the hypothesis (H2) and Ohm's law
(see Ref. 2)
m
ET~>~ 2 2 v>,
k ~ 1 y E Bk~\Bk L
m -- kO
>1 ~ ( b i - b i - l ) "~m~
Irn_kO >-Co(bm_ko
~" - nU l I J~r(
r n m~
i=1
for some constants Co, C1 ~ (0, oo). By Eq. (2.2), we easily know that there
exists a subsequence {nt}t~>~ such that
~< ~
ln( Nm,) ln(Nm,)
Obviously, Eq. (2.4) holds if n l = mr (V/~> 1). In other words, we also have
+ 1
R(x, Sm,)~(1 m-~-"~_l)R(x,Smt l)
dR>~d+ 2 - d a
Thus, we complete the proof of Eq, (2.1) under the hypothesis in Eq. (2.2).
We now prove Eq. (2.1) for the case:
Let
It is clear that
lim Pm = 0
m~co
-- P~- 1, Vl >/1
d n = 2- lim In p~:
l . oo in N~I
Thus, there exists a l'/> 1 for any t ~> 1 such that t' 1` oo as l 1"oo, and
nr<.kz<nr+l
So we have
dn = 2- lim In lOm~
~ ~o in Nm.
and
Vl>~ 1
644 Zhou
dR>d+ 2-dn
On the other hand, we let (V'y) and (Vy) be the potentials on the elec-
trical network Bm+~ and Bm respectively with a unit total current flow
from x to S,~+~ and S~. One can also get that from the hypothesis (H2)
and Ohm's law
F(m+l)~
m+ 1 R(X, Rm+ 1) - R ( x , Sm)
Note that (Vy) is the potential o n B m with voltage R(x, Sin) and v~) is the
potential o n O m with the voltage R(x, Sm+ ~). Thus, the maximum principle
deduces that
V y + ~Fm(re+l)'>,,'
+l ~"~ V y ,
Vy@B m
<~ ~ deg(y)(vy+--m+~
F ( m + I)~,+~--m+l
/~ r ' ( m + l ) lWm+~
l, __bin)
y~Bm
~ E T m + ~/m
)/~ F( m+l
--m+l ) ~F(m+l)(h
+~--m+l ~m+l
-bin)
< ET m ~-a-~I)F(
--m+
m+I 1)h
~'rn + 1
m+l
< ET~ + D ~, bkr~~ (2.7)
k--2
By the hypothesis (H~) and Eq. (2.7), one can easily show that
k0 m+ I
ETm+~<ET~+ y" Db~r~+ ~ DColbk(R(x, Sm+l)-R(x, Sk-ko))
k=2 k=k0+l
k0 re+l--k0
ET1+ D ~ 01,
" F k(k) + ~ DC o 1C ,bkpk
k=2 k=l
Resistance, RandomWalk, and Fractal Dimensions 645
It follows that
dn~d+ 2-da
The proof of Eq. (2.1) is complete. 9
Remark 2.1. From this proof, one easily sees that the conclusion of
Theorem 2.1 also holds if (H2) is replaced by the hypothesis (Hi), where
(Hi) There exists a constant ko = ko(a)~ N for any given e e (0, 1),
such that
F(mm)< m~7~"~)ko, Vm>k o
Remark 2.2. Telcs [Ref. 7, Proposition l j himsdf proves that Eq. (2.1)
also holds under some other conditions for strongly recurrent random walk
(i.e., dn <2). Thus, by Lemma 2.1, one easily finds that the conclusion of
Theorem 2.1 also holds if the condition in [Ref. 7, Proposition 1] is
satisfied. In fact, we can improve the relation of Eq. (2.1) for this kind of
graphs (see Proposition 3.2).
3. REMARKS
In the present section, we give two remarks. First, we check that the
hypotheses (H1) and (H2) are satisfied if the graph is moderate in the sense
of Ref. 6.
d = lira In b(ki.___~)
i~ lnk, '
646 Zhou
and
b(ki)
Q~<---< ' Vi~>l
b(k~_ ~) "~ Q'
for some constants Q, Q'E (1, oo), then the hypothesis (H1) is satisfied.
Proof Let {ki}i~o be the sequence as described earlier. By the
hypothesis, one can show that there are constants C1, C2 e (0, oo) such that
In k, ( nk, b(_k;))
ilimo~1 - ~ = l i r a \ l n b(ki) In i J
In b(ki)
= d -1 lira O0
i~o~ In i
and
In k i In ki In b(ki) In b(ki_ 1)
lira lira - - = 1
i~oolnki_l i-.oolnb(ki) lnb(ki_l) lnk~. 1
In b(ki) b(ki)
d = lim - - , Q<<.~<~Q', vi>~2,
i~ co In ki b(ki_ 1)
R(x, S(k,))
Q<~R(x,S(ki_,))<~Q', Vi>~2, if R~(G) = o%
and
d. = 9 12
L 2-1-~i~ ~
In Ki '
l n ( R x ( G ) - R(x, S(K~)))
In Ki '
if R~(G) < cc
= d + 2 - d~ - dR = 0, if G is moderate.
C~F('~'I)~'~
v tl
3l(m'l),
tn -- 1
Wv l<n<m
where
K~e(0, oo) and l<m, then there is a constant K2~(0, oo) which is
independent of K~, I and m such that
Thus, we have
rain v(y) ~7 } re'l)
y E S(Ni)
max v ( y ) = max
y ~ S(Ni) y ~ (B(Nm)\B(Ni)) w S(Ni)
Proof The upper bound is easy to get (see Ref. 5 or the proof of
Theorem 2.1). We now prove the lower bound. Let R(j, k) be the resistance
of B(Nk) between S(Nj) and S(Nk) for j-G<k, and (vx)~a~ul) be the poten-
tial on B(Ni) satisfying
Therefore, there is a fixed io = io(e)>~ 1 for any eE (0, C11) such that
By Lemma 3.1 and Ohm's law again, there is a constant C2 e (0, oo) such
that
860/6/4-3
650 Zhou
for some constant Co E(0, oo). Indeed, this estimate comes immediately
from (H4)(ii) and Eq. (3.1) for i0 = 1.
We now consider the corresponding problem for the transient graph.
We first introduce a hypothesis.
(Hs) There are a subseries {Ni} and constants q, Q s (1, oo) such that
(i) C.qi<~b(N~)-b(N~_~)<~C'q ~, Vi~>I,
(ii) C . Q - ~ R ( i - I , i ) < ~ C ' Q -~, Vi~>I,
(iii) lira ( N i - N i _ l ) = oo
for some constants C, C' ~ (0, oo ), where R ( i - 1, i) has been defined in the
proof of Proposition 3.2.
C3 2 Vy ~ E x r ( N i ) < ~ C 4 2 Vy
y ~ B(Ni) y EB(N,)
By Lemma 3.1, we know that there is a constant C5 e (0, oo) such that
Using the hypothesis (Hs) and noting Rz(G) < o% one has
i
C6 Z b(N,) . R ( I - 1, i)
l=1
Resistance, Random Walk, and Fractal Dimensions 651
for some constant C 6 ~ (0, (30). For any l ~<i, we let (~y) be the potential on
(B( N,) \ B( N t- 1)) w S( N t_ 1) satisfying
<-%R ( l - 1, i - 1) + C 7 rain Vz
zcS(Ni 1)
<. g ( l - 1, i - 1 ) + C 7 R ( i - 1, i)
Therefore
i f
ExT(Ni)<~ C6"C8 ~ b(N,). ~ Q 9
l=1 j=l
i j
= C6" C8 2 2 b(Nl)Q-9
j=l /=1
i
C9 2 Q - i q J, (by (Hs))
J= 1
for some constant C 9 ~ (0, 00). Let lEvi be the number of all edges
c o n n e c t i n g S N_ 1 and SN- Since the graph G is with bounded degree, we
easily see that
C~ol(b(Ni) - b ( X i - t )) - 1 (Ni-- Ni 1 - 1 ).
652 Zhou
lim ( N i - N i _ 1 - 1 ) = ~ .
Thus
Q<q
E x T ( N i ) < ~ C 1 L Q - i q i, Vi>~I
The p r o o f is complete. 9
or (Hs)(ii) is replaced by
C.Q-i<~Rx(G)_R(x,S(N,))<~C,Q i Vi~l
for some constants q, Q E (1, ~ ) and C1, C2~ (0, ~ ) , then the conclusion
of Proposition 3.3 holds as well.
ACKNOWLEDGMENTS
The author thanks to Prof. Shi Jian Yan and Prof. M u Fa Chen for
their encouragements and m a n y helpful suggestions. The a u t h o r also
expresses his sincere thanks to the referee for m a n y corrections and helpful
suggestions.
REFERENCES
1. Barlow, M. T., and Bass, R. F. (1992). Transition densities for Brownian motion on the
Sierpinski carpet, Prob. Th. ReL Fields 91, 307-330.
2. Doyle, P. G., and Snell, J. L. (1984). Random Walks and Electric Networks, Math. Assoc.
Amer., Washington, DC.
3. K_umagai, T., Kusuoka, S., and Zhou, X Y. (1992). Resistance and spectral dimension for
some homogeneous random fractals, preprint.
4. Kusuoka, S., and Zhou, X. Y. (1992). Dirichlet form on fractals: Poincare constant and
resistance, Prob. Th. Rel. Fields 93, 169-196.
5. Telcs, A. (1989). Random walks on graphs, electric networks and fractals, Pro& Th. Rel.
Fields 82, 435449.
6. Telcs, A. (1990). Spectra of graphs and fractal dimensions I, Prob. Th. Rel. Fields 85,
489-497.
7. Teics, A. (1990). A note on recurrent random walks on graphs, J. Stat. Phys. 60, 801 807.