Zeitschrlft fiir

2. Wahrscheinlichkeitstheorie venv. Gebiete Wahrscheinlichkeitsthewie

68, 365-382 (1985) und verwandte Gebiete
0 Springer-Verlag 1985

On the Length of the Longest Excursion

E. Csaki, P. Erdos, and P. RCvtsz

Mathematical Institute, Reiltanoda u 13-15, H-1053 Budapest, Hungary

Summary. A lower limit of the length of the longest excursion of a sym-

metric random walk is given. Certain related problems are also discussed. It
is shown e.g. that for any a>0 and all sufficiently large n there are C(E)
loglogn excursions in the interval (0, n) with total length greater than n(l
-a), with probability 1.

1. Introduction

Let x,, x,, . . . be a sequence of i.i.d.r.v.‘s with

lP(Xi=+l)=JP(X,= -I)=$ (i=l,2, .-,)
and consider the random walk S(O)=O, S(II)=X,+X,+...+X, (n=l,2, . ..).
Introduce the following notations :

9(n)=J%.{k: O<kS:n,S(k)>Oj,
z%‘(m)=d’o,{k: O<ks;n, S(k)=O),

(,4%. (...I stands for cardinality of the set in brackets).

po=o.
p,=inf{k: k>O.S(k)=O},
p,=inf{k: kz=-p,, S(k)=O},
. .. . . .. . . .. . .. . .. .
P.+l=inf{k: k>p,, S(k)=O},
. . . . . . . . . . . . . . . , . .
,W)=maxtp,. p2-p1, --., Pa(n)-P9(n)-1, +p2(,,).
Here F(?t) is the length of the longest excursion of the random walk S(O),
S(l), . . . , S(n). The main goal of the present paper is to study the properties of
F(Il).
366 E. Cshki et al.

The properties of .2(n) and .Vp(n) were studied by Chung and Hunt (1949)
and Chung and Erdds (1952) resp. Here we recall the Chung-Erdiis theorem.
Theorem A. Let f(x) be a non-decreasing function for which lim f(x)= s and
x-+33
Put
w,=q $&
Then
JP{l(n)>n (1-h) i.o.}={i z :ti;zz7 (1.1)
and
1 $ I(j-)=cc,
P B(n)&& i.o. (1.2)
0 if Z(f)<m

Studying the proof of Theorem A we can realize that the following stronger
statement is also proved by Chung and Erdds:
Theorem B
(1.3)

provided that f(x)/” cc.

(1.3) gives the best possible upper bound for F(n). For example it implies
that for any E>O
~(n)Sn-(log$+s

except finitely many n with probability one and

infinitely often with probability one. We are interested to find a lower bound
for Y(n). Our main result is
Theorem 1. Let f (x) be a non-decreasing function for which

and let
J?(f)=
z f(M)
n
e-lw)
II= 1
Then

(1.4)

where fl= 0,85403 . . . is the root of the equation

0.5)
On the Length of the Longest Excursion 367

(1.4) says for example that

Remark. Equation (1.5) emerges in a paper by Shepp (1967) (see also Green-
wood and Perkins (1983)).
Beside of studying the properties of the length of the longest excursion, it
looks interesting to say something about the second, third . . . etc. longest
excursion. Consider the sample pi, pz -pl, . . . , pacnj--pwcnl- 1, n-pP,(,) (the
lengths of the excursions) and the corresponding ordered sample Yi(n)
= Y(n) 2 T,(n) 2.. .Z Y&+ 1(n). Now we present our
Theorem 2. FOP any fixed k = 1,2, . . . we have

liminf F $r.YJn)=kfl as.

“iC.2

This Theorem, in some sense, answers the question “How small can be
FZ((n), Y3((n), . ..?” In order to obtain a more complete description of these r.v.‘s
we present the following:
Problem 1. Characterize the set of those non-decreasing functions f(n) (n
=l, 2, . ..) for which
lP{&(n)zt (1-h) i.o.)=l.

(1.3) says that for some n nearly the whole random walk S(O), S(l), . . . , S(n)
is one excursion. (1.4) and (1.6) say that for some n the random walk consists
of at least /3-’ loglogn excursions. These results suggest the question: For
k

what value of k= ii(n) will the sum 1 Y?(U) be nearly equal to n? In fact we
formulate two questions: j= 1

Question 1. For any E>O let F(E) be the set of those functions f(n) (n= 1,2, .,.)
for which
f(n)
1 qn)&l(l-&)
J= 1
with probability one except finitely many n. How can we characterize F(E) for
some E>O?
Question 2. Let Y{D) be the set of those functions f(n) (n = 1,2, . . .) for which

lim n- ’ C Fj(n)= 1 as.

n-cc j= 1

How can we characterize F(G)?

Studying our first question we have
368 I?. CsBki et al.

Theorem 3. For any E> 0 there exists a C = C(E)> 0 such that

c loglog MEF(&).

Concerning our Question 2, we have the following result:

Theorem 4. For my C>O
f(n)= c 10g10gm$~(0) 0.7)

andfor any +jr x (n-t x)

w(n) loglogE~((o). (14

2. Proof of Theorem 1

We recall the following well-known

Theorem C
(k = 1, 2, . . .).

Consequently
b,= ___1 k-” exp(3Jk) (2.1)
2fi
where 13,15 1.
By (2,l) we easily obtain
Lemma 1
(2.2)

where /I is the root of Eq. (1.5).

Proof. Clearly we have

$ bjexp ~~)=j~~b~+j~~bj(exp~~)-‘)
j= 1

and

1 z pk f jk-'
A,=----
2fi k= 1 ak k! j= 1
‘co
B” +o(a-t)=(7la)-r+8(a-:)
=‘“‘)-tkCI k!(2k-1)
On the Length of the Longest Excursion 369

A,=O(d)
what proves (2,2).
Let
IP(F(2n).52a) if nza,
Ptl=PJ4=
i 1 if n<a.
Then we clearly have
Lemma 2.

Pn=iP ,ejbj (n=a,a+l. . ..). (2.3)

j= 1
Now we are looking for the solution of (2.3) satisfying the initial condition
p,=l if II= 1,2, . . . ,a -1. We obtain
Lemma 3. There exist positive constants 0 < C, 5 C, < ,zmsuch that

and
C,sC(n,a)sC,

provided that n 2 a*.

From now on C (with or without index) will stand for an absolute constant
whose actual value may change from line to line.
Proof: Replacing (2.4) in (2.3) we get

C(n, a) = j$l C(n -j, a) exp (P i) bj. (2.5)

In case ns2a our statement is trivial. For n>2a the statement follows from
(2.5) by induction,
Lemma 4. Let f(n) (n = 1,2, . . .) be a non-decreasing positive function for which

f-(n) <a loglog n i.0. w9

and put

Then &=$=m’.
ProoJ: Suppose that f(N)<+ loglog N for a fixed N. Then a simple calculation
gives

(2.7)
370 E Csiki et al.

Thus f2 C(log N)’ for infinitely many N, we have # = co. One can see similarly
that j= x8 by observing that condition (2.6) implies that f(n,)<f loglogn, i.o.
Lemma 5. Let f(n) (n= 1, 2, . ..) b e a non-decreasing, positice function. Then &
=a if and only if j=‘z.
Such a lemma like this and the previous lemma is frequently used in the
proofs of theorems like our Theorem 1, hence its proof is routine. For the
convenience of the reader we present it.
Proof

.,pwe-“““‘=c~*
k log k
Similarly one can obtain that

By Lemma 4 one can assume that f(n) >a loglog n (n= 3,4, . .). Hence we have

$*g‘$
that is 2 = cc implies f = W. In order to see the converse statement let
A = {k: f(nk) ~2 loglog nk}.
Then

sCCepftnk)+C f(%)
-e -She) 5 cj + c
keA k$A logk

what proves the implication: if & = xi then 2 = ixj.

The following lemma is trivial, we give it without proof,
Lemma 6. Let {a,} be a non-increasitzg sequence of positioe numbers for which
cc
C a,<ccj. Then
?I= 1
np.)l-~<;c.

Lemma 7. Let
A, = (F(n,) 2 Ok] k = 2, 3, I..
where
Pnk
ak=fo

and f(n) is a non-decreasing positioe function such that f(n) 5 j?n*. Then

JF(A,A,)~ClP(A,)exp (-fly) (lsk<l<m). (2.8)

On the Length of the Longest Excursion 371

ProajI Let F(a, b) (0 S a < b < cc) be the length of the longest excursion of the
random walk S(a), S(a + l), . . . , s(b). Then

=P(F(n,-nk)~aa,)IP(Ak)5C exp ( -/I 7) WA,)

(the last inequality follows from Lemma 3). Hence we have (2.8)
Lemma 8. Let f(n) be a non-decreasing function for which 9(f)= m. Then for
any 0 <E < 1 there exists a non-decreasing function fsuch that
N f(n) Zf(n) (n = L2, . . .I,
(ii) Y(f)= xi,
(iii) T(n) 2 & loglog n.

Proof of this lemma is based on the same idea as that of Lemma 4 and will be
omitted.
Lemma 9. Let
B, = (F(n) 5 b,)
where
Bn
b”=fo

and f (n) is a non-decreasing function for which f c ~13.Then

lP(B, Lo.) = 0. (2.9)

Proof. Let
fc%,=~f'"k+l'.
Then by Lemmas 5 and 6

kg2eMfcnk)< cc
and by Lemma 3

provided that f(n) 5 pn*.

Now let nk~n~nk+l then

with probability one except finitely many k. Hence we have (2,9), if f (n)s /?n+
(n 2 noI.
312 E. Cs&kl et al.

In the case when this condition does not hold, define f, (n) = min (f(n), on)).
f,(n) is non-decreasing with #(S,)< CC and f, (n)sM, hence (2.9) holds for
f(n) replaced by ii(n). Since fi(n)5f(n), we have also (2.9) with the original
f(n). This proves the first part of Theorem 1.
To show the second part, assume that
+ loglog n 5J(n) 5 2 loglog I? (n=3,4, . ..) (2.10)

The lower inequality can be assumed by Lemma 8, while if the upper in-
equality does not hold for all n large enough, then by eliminating those n’s for
which f(n)>2 loglogn, the whole procedure below can be done for the remain-
ing subsequence and still conclude the second part of (1.4).
Defining nk as in Lemma 4, for large enough k and k < 1 we have

log!?=--I-k k(logI-logk)
nk log/ log I log k
,1-k
-- l-k >!I-k (2.11)
=log1 (logI)(logk)=2 1ogE’

Now for k fixed, split the indices 1 (k<isn) into three parts:

L,=(l: O<I-kslogI}
L,=(l: logI<1-k61og2I}
L,=(l: log21<l-k}.
For MEL, we have from (2.10) and (2.11)

yf(nJZ (1-exp (-i $)) kloglogn,

(2.12)

For MEL, we have from (2.11)

2

Hence by Lemma 7 and (2.10)

(2.13)

For IEL, we have from (2.10) and (2.11)

(2.14)
On the Length of the Longest Excursion 373

Hence
,z P(AkAl) 5 cP(A,) 2 e-c’(l-k) 5 cP(A,), (2.15)
1 iELl

(2.16)

since C 15 c(logk)‘.
1ELZ
By Lemmas 3 and 7, (2.14), (2.15), (2.16)

~p(A3 ==
and

,gl ,$, p(AkA,)s ’ (j IP@k))* + ’ ,$, IPb%)

1

consequently

and by the Borel-Cantelli lemma (cf. Spitzer (1964)) we have

P(A, i.0.) 2 C-l > 0.
Hence we have our Theorem 1 by the O-l law.

3. Proof of Theorem 2

We give the following analogue of Lemma 3.

Lemma 10. Let
IP(F(2??)522a, S(2‘n)=O) if nza,
p”,= hi4 = lP(S(2n)=O) if n<a.
i
Then there exist positive constants 0 < C, 5 C, < ~1 such that

and

provided that 0 5 ni a*.

ProoJ Observe that the statement is trivial if Osn<=2a and we have

ii,= i fin-jbj (nZ2a).

j=l
314 E. Cshki et al.

Now we obtain OUT Lemma 10 using the method of proof of Lemma 3.

Let u1,a2, . . ..uk be a sequence of positive integers and m
=min(a,,a,, ,.., a&. Further let d,,d,, . . . . dkil be a sequence of non-negative
integers such that
d,~O,d,ZO,d,~O ,..., d,~OO,d,+,~O
d,+d,+...+d,+,+a,Jrn,+...+a,=n
(3.1)

Introduce the following notations

B, ={S(d,)=O, F(O,d,)=<m},
A,={S(d,+i)+O((i=1,2 ,..., a,-l),S(d,+a,)=O),
B,={S(d,+a,i-dz)=O, ~-(d,+a,,d,+fl,+d,)~m),

A,=(S(d,+a,+...-ta,_,+d,+i)*O (i=1,2 ,..., a,-l),

S(d,+n,+...+d,fa,)=O)
3 k+l={S(dlfai-t...t-d,+a,)=O, ~(d,+a,+...+d,+a,,n)~:m),
A=A,A,...A,B1B2...Bkil,
A*=A*(a,,u,,...,a,)=~*A

where in the sum c* the indices d,, d,, . . . , dk+l run over all (k + l)-tuples of
integers which satisfy (3.1). Clearly A* is the event that the random walk
S(l), S(2), L* ’ , S(n) consists of excursions of size a1,a2, . . ..ak in this order but all
other excursions are shorter than m.
Lemma 11. Let n smm”. Then

IP(A*)s Cm-%(&+m~exp (-pm-’ (n-i1 ai)) (3.2)

where the constant C may depend on k.

Proo$ By Theorem C we have

P(Ai/A,...Ai_,B,...Bi)~ Ca;* (i=l,...,k).

Since d, 5 n 5 m3, by Lemma 10

P(Bi/A,...Ai~,B,...Bi~l)~C(m~~+(rii+l)~f)exp(-~m~‘d~) (i=l, . . ..k)

and by Lemma 3
P(B,+l/A,...A,B,...Bk)~Cexp(-Bm-’dk+l).
Hence
P(A)sCexp (-/I’m-’ (n-$,a;)) ib(m-‘+(di+l)P”).
On the Length of the Longest Excursion 375

Now

and (3.2) follows.

A trivial consequence of Lemma 11 is
Lemma 12. Let a, za, 1,. . >=aL 2 n’ be a sequence of integers. 7hen

p(T,(n)=a,, . . . . T,(n)=a,)jca;~(,ak.~+a:)kexp j-pa;’ (n-iIazjj. (3.3)

Lemma 13. Let kn+su <YE.
k 3k
c Tj(n)Su, T,(n)& scu5- y(nu-++u:)kexp (-/SF). (3.4)
j= 1

Proof. (3.4) follows from (3.3) by summation for the possible a, (i = 1,2, . . . . k)
observing that nf 5 ak 5 u/k and the fact that a sequence a, 1 a2 2.. .L ak >=ns of
k

integers for which c ai 5 u can be chosen at most uk different ways.

i= 1

Lemma 14. For large enough n we haue

P j~~~(~)~/i(l-i)k~)~c(logn)-(i+~)- (3.5)
i

Proof. By letting
u=u,=b(l-E)kA
loglogn
we obtain from (3.4) that

P q;(n) 5 u,? T,(n) 2 n+ 5 c(log n)-(‘++). (3.6)

Furthermore

P jiI 7;.(n)su,/k, T,(n)<,*) IP(T,(n)~u,/k)~c(logn)-(‘+f).

(

Finally, if u,/k 5 a, + . , . +a,5 u, and uk<n%:, then max di zc,n with some
lsisk+l
constant c2, i.e. there exists an interval of length Lc,n such that longest
excursion within this interval is shorter than n*, the probability of which is less
than c1eacsn’, The number of possible choices of a, . . . ak, d, . . . d,,, is obviously at
most nzk+‘, hence

P un/kj i Tj(n)lu,, Tk(n)<n3 ~cln2k+1e-csn3.

jel

Since for large n the upper bound in (3.8) is less than the upper bound in
(3.9 we have Lemma 14 by combining (3.6), (3.7) and (3.8) with some constant
c (different from that in (3.6) and (3.7)).
376 E. Csaki et al.

(3.5) by well-known methods implies

liminfe j~l~(n)~k~ a.s. (3.9)

n-m

Now Theorem 2 follows from (1.6) and (3.9).

4. Proof of Theorem 3

Instead of proving Theorem 3 we prove the analogue statement (Theorem 3*)

for a Wiener process (IV(t), tzO}. Theorem 3 can be obtained from
Theorem 3” constructing the sequence {Xi] from IV(t) by the Skorohod stop-
ping rule.
Theorem 3*. Let {W(t), t 2 0} be a Wiener process. Then for any E> 0 tkere exist
s((T) = [C, E-I loglog T] excursions gl, &, . . . , Jacacnof W in [O. T] suck that
am
i~llgi12il--E)T

if T is large enough with probability one where Igil is the length of the excursion
c?‘~and C, is an absolute constant.
Introduce the following notations: Let a,>0 be a function of T and

Y,=O,
q=Y,(T)=inf{s: s> y_, +a,, W(s)=O} (i=1,2 .,.. ),
v,=max{k: E’,s T},
Zi=Z,(T)= I’-(~-,+a,),
Mi=Zi/aT.

The following lemma is well-known.

Lemma 15. (i) (2,) is a sequence of i.i.d.r.u.‘s.
(ii) { Ui} = (Zi(W( y_, + aT))-2j is a sequence of i.i.d.r.v.‘s

(iii) IP(U,<x)=Ip(U,<xJ W(x-, +a,)=w)=j27r)+ j u-Ye-l;dc,

3 1
0

(iv) E(exp{ -tU;})=exp{-t*), t>O.

The next lemma is an easy consequence of a theorem of Steinebach (1978).
Lemma 16
lim(P(M,+...+M,~rm)~~=p(a),
m-x
where p(a) = inf(A{t)e”‘), i(t) = E(e-*Mi).
On the Length of the Longest Excurslon 371

Lemma 17. Let

UT= E2 T(lOglOg T)-‘,

mT= [(3rr)+s-lloglog T],

Then
IP(V,>m,)=O((logT)-+) US T+a.
Proof

T
~IP(z,+... +Z&T)=JP M,+...+M,r~-
i 'TmT

It is easy to check that

where d(x) is the standard normal distribution function and hence

P(tl)<exp( --(271~)-~).
Lemma 17 now follows from Lemma 16.
Considering the excursions &Yiaround the points K +a, (i = 1,2, . . . . vT) the
non-covered part of the interval [0, T] will be less than vTuT. Hence Lem-
ma 17 implies
Lemma 18. With C, =(3r)* we have
a(T)
It’ ,~JZJ<(l-sC,)T =G((logT)-‘) (T.-o).
L 1
Lemma 18 via standard methods implies Theorem 3” with C, =(3x)+“.

5. Proof of Theorem 4

It is easy to see that (1.8) is a simple consequence of Theorem 3. Instead of

proving (1.7), we present again the proof of the analogue statement for a
Wiener process, In fact we prove our
Theorem 4”. Let {W(t), t 2 0} be a Wiener process and let q(T) 2 5(T) 2. . be
the lengths of the longest, second longest excursions of W up to 7: Then for any
D > 0 there exists an E= E(D)> 0 such that

IP{~~(T)+~~(~~)+...+~~,,,~T(l-E) i.o.}=l (5.1)

where b(T)= [D loglog T].
Introduce the following notations:
378 E. C&ki et al.

a,=a(T)=fi T
log log T’
Y,=O,
I/,=sup{s: S<QT, W(s)=O},
Y,=inf(s: s>ay, W(s)=O),
... .. ... . ... . .... .. ... .
l$+,=sup{s: sty++, W(s)=O},
x+,=inf(s: s>x+aT, W(s)=O),
. . . . . 4. . . . . . . . . . . . . , ,.
di=K-& zi=~;-(~~,+aT),
q=(W(y-1+aT))-2zi, Ni=a,+W(Y,-,+a*),
v,=min{i: xz T},
Ri=y-x-,,

The next lemma is an easy consequence of Lemma 16.

Lemma 19
IP m-l f? qNF<c1 ZCe-“‘I” 6.2)
i= 1
for any ct> 0 and m big enough.
Lemma 20
Ip(Z, f z, + . . . + z,,,, +a(T)b(T)ST)g C(logT)-’ (5.3)
if s=(D2+D)-1.
Proof
1P(Z,+Z,+...+Z,,,,+a(T)b(T)~T)
=IP(Z,+Z,+...+Z,,,,~(l-5D)T)
=lP(b-l(T)a-l(T)(Z, +Z,+ . . . +Z,&<(l4D)6-‘D-l)
b(T)
=lP b-‘(T) 1 u,N,z<D .
( i=l 1

Hence we have (5.3) by (5.2).

Lemma 2LIP(Z,+Z,+...+Z,,,, + a(T) b(T) 5 T Lo.)= 1 equivalently

IP{v,zb, i.o.}=l or lP(Y&~Ti.o.)=l.

Proof. Let T,=k” and let the events A,, At be defined by

On the Length of the Longest Excursion 379

By Lemma 20, we have

Furthermore, since for large k, T,ta(T,+,), the events A, and A,* are inde-
pendent for k<l. Hence

for any F>O provided k< I and k is large enough, where the last step follows
from Lemma 16. One easily verifies that

i i WA,41
lim inf 1~ 1 k= 1

n-*x (pyA,))? 51

and by the already quoted Borel-Cantelli lemma (cf. Spitzer, 1964) we have
P(A, i.0.) = 1
which proves Lemma 21.
By the above procedure we have chosen b, excursions (vi, YJ (i = 1,2, . . . , br)
which however are not necessarily the b, largest ones. It is possible that some
of them can be replaced by larger from the intervals (y, 5 + 1) (i = 0, . . . , b, - 1).
But it is readily seen that even the largest b, excursions in (0, YbT) can not
bcr
cover more than 1 (Zi +max(R,, a,-R,)). Hence the non-covered part of
i= 1

(0, Y&) is at least 5 min(R,, uT -RJ. Since R,/a, (i= 1, . . . , b,) are i.i.d. random
i= 1
variables having arc sine distribution and

we have by the law of the large numbers

Lemma 22
lim (a,b,)-’ 5 min(R,,a,-Ri)=i--k a.s.
T-.tX i=l

It follows that for large enough T

T
$ min(Ri,
U~-Ri)~~~~b,>---
6(D + 1)
as..

which together with Lemma 21 proves Theorem 4.

380 E. Cshki et al.

6. A Consequence and some Problems

Introduce the following notations:

M,(n)= max S(k), ~,(n)=o~~~nIWl,
Osks;n - -

c~~=~~(j)=O, flo=&(j)=max{i: izOo,Mj(i)=O},

a,=cx,(j)=min{i: i>bO,M,(i)=Mj(i+l)},
j?, =j31(j)=max{i: Mj(i)=Mj(r,(i))],
. . . . . . . . . . . . ..‘...........I..........

a,=a,lj)=min{i: i>Bk--lrMj(i)=M,(i+l)}.
&=&(j)=max{i: Mj(i)=M,(rJj))},

@(n)=.@j(n)=max{k: a,tj)sn>,
~(n)=~(j)(M)=max(Bo-x,,B,-a tr...,Bg(n)-l-a~(n)-lr n--a,,,,>
(j=1,2).

Here f-(j)(,) is the length of the longest flat interval of Mj(i) (Oli sn; j
= 1,2). A famous theorem of Levy (see e.g. Knight (1981) p. 130 and Csaki and
RCvCsz (1983)) says that the limit behaviour of M,(n) is the same as that of
S!(n). Applying this result and Theorems B and 1 one has
Consequence. Let f(x) be a non-decreasing function for which

and

where p is defined by (1S) and I resp. $’ are defined in TheoTerns A resp. 1.

This Consequence gives a complete characterization of y-(‘)(?z) and suggests
our
Problem 2. Characterize the sequence y@)(n).
Let {a,> be a non-decreasing sequence of positive integers and consider the
process
m(n) =m(n, a,) = min (W(k+ a,) - B(k)).
OSkSM-0,
On the Length of the Longest Excursion 381

Theorems B and 1 imply

lim sup m(n, a,) = 0 as.

lim infm(n, a,) = 0 as. and I(f)= co.

Problem 3. Characterize those sequences {a,} for which

lim sup m(n, a,) = K as.

where K is a given positive integer.

Problem 4. For a given sequence {a,} find the normalizing factors i(n) = i(n, a,)
whenever I(f) = and s(n)=s(n, a,) (a, >/3n/f(n) whenever
f(f)< co) such that
lim sup ~4% 4 -_ 1 a.s.
s(n)
and
lim inf 4%
____4 = 1 a.s.
i(n)
Remarks. 1. The properties of
max (C%?(k+a,)-8(k))
O(k~n--o,

were studied by Csaki et al. (1983) and by Csaki and FGldes (1984).
2. Our Theorems were formulated originally for random walks. In order to
get a simpler proof we reformulated some of them for Wiener processes and
noted that the reformulated versions imply the original ones by invariance
principle. Here we wish to mention that Theorems 1 and 2 can be reformulated
for Wiener process as well.

Acknowledgement. The authors wish to thank the referee for careful reading of their manuscript
and for helpful remarks,

References

Chung, K.L., ErdGs, P.: On the application of the Borel-Cantelli lemma. Trans. Amer. Math. Sot.
72, 179-186 (1952)
Chung: K.L., Hunt, G.A.: On the zeros of $ + 1. Ann. of Math. 50, 385-400 (1949)
Csaki, E., Csiirgo, M., Foldes, A., Reves~,~P.: Row big are the increments of the local time of a
Wiener process Ann. Probability 11: 593-608 (1983)
Csaki, E., Fiildes, A.: How big are the increments of the local time of a simple symmetric random
walk? Coll. Math. Sot. J. Bolyai 36. Limit theorems in probability and statistics,Veszprem
(Hungary), 1982. P. Rtviisz (ed.). (To appear)
382 E. C&k1 et al.

CsBki, E., R&&z, P.: A combinatorial proof of a theorem of P. LCvy on the local time. Acta Sci.
Math. (Szeged) 45, 119-129 (1983)
Greenwood, P., Perkins, E.: A conditioned limit theorem for random walk and Brownian local
time on square root boundaries. Ann. Probability 11, 227-261 (1983)
Knight, F.B.: Essentials of Brownian motion and diffusion. Am. Math. Sot. Providence, Rhode
Island, 1981
Shepp, L.A.: A first passage problem for the Wiener process. Ann. Math. Statist. 38, 1912-1914
(1967)
Spitzer, F.: Principles of random walk. Princeton, N.J.: Van Nostrand, 1964
Steinebach, J.: A strong law of ErdBs-R&nyi type for cumulative processes in renewal theory. J.
Appl. Probability 15, 96-111 (1978)

Received July 28, 1983