On a class of embedded Markov processes and recurrence
Citation for published version (APA):
Simons, F. H. (1976). On a class of embedded Markov processes and recurrence. (Memorandum COSOR; Vol. 7610). Technische Hogeschool Eindhoven.
Document status and date: Published: 01/01/1976 Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
EI~~HOVEN UNIVERSITY OF TECHNOLOGY
Department of Mathematics
PROBABILITY THEORY, STATISTICS AND OPERATIONS RESEARCH GROUP
Memorandum COSOR 76-10
On a class of embedded Markov processes and recurrence
by
F.H. Simons
Eindhoven, July 1976 The Netherlands
On a class of embedded Markov processes and recurrence
by
F.R. Simons
Abstract. By means of a general type of embedded process we shall give a short deduction of some recurrence properties of the Markov shift.
1. Preliminaries
Let (X,L,m) be a a-finite measure space. Let M+ be the space of (equivalence classes of almost everywhere equal) nonnegative extended real valued measurable functions on X. A Markov operator is a mapping P of M+ into itself such that PI :s; I, and P(
I
ex f )=
n=1 n n 00I
a Pf n=1 n n + f E: H , ex n n ~o.
The domain of P can be extended to Loo such that P is a positive linear contraction in Loo Such an operator is always the adjoint of a positive linear contraction in L1, which we shall also denote by P, but now written to the right of the function symbol. The action of this positive linear contraction P on L1 can,
+
by means of monotone approximation, be extended to the space M • It follows that <fP, g> = <f ,Pg> for all f, g E:
M
+ •Here <f,g> stands for !fgdm.
With respect to a given Markov operator P on (X,!:,m) we can decompose the
space X into a conservative par.t C and a dissipative part D. This decomposition is __e to E. Hopf, and can e.g. be found in [1], chapter II, [5],
chap-ter 4, § 2. For later use we collect the results which we shall need in two
lemma's and some corollaries.
Lemma 1. The following statements are equivalent.
i. The conservative part of X with respect to P is C.
ii. The set C is the (mod m) largest set such that for all subsets A we have
on A.
2
-Lemma 2. The following statements are equivalent.
1. The dissipative part of X with respect to P is D.
ii. The set D is the (mod m) largest set such that there exists a function g ~ 0, with {g > O} = D and I:=opng is bounded.
(This equivalence can be obtained e.g. from (2,5) in [1] and the maximum principle, chapter 2, theorem 1.12 in [5J).
Corollaries. I • 2. PI = PI = I on C, and therefore PI D = o on C. C 00 +
L
pnfFor every f E M we have = 0 or 00 on C. n=1
00 co
I f
Co
= {I
pnf = O} n C, and CI = {I
pnf = oo} n C, then PIC. =o
onn"'l n=l 1
C
I-1., i=O,I.
3. If A c C and m(A) > 0, then < Ic,PI
A> > O. It follows that {lcP > O} = c. 4. For every s ~ I the conservative part of X with respect to pS equals C.
2. The embedded processes
s-I
Let P, Hand H' be Markov operators, and assume H + H' • P for some s ~ I. We define the operator Q by
00
Q =
L
(PH')npH n=OObviously Q is a a-additive mapping of M+ onto itself. The next lemma implies by substituting f = ) that Q is a Markov operator.
+
L~maa 3. If for f E
M
we have Pi $ i, then we also have Qf $ f.Proof. Using pSf $ f, we easily verify by writing out
n-I
I
(PH,)rpHf + (PH,)nf $ f.r=O
3
-If H is the multiplication by the characteristic function of a set A~ H' the multiplication by lA" then Q is the embedded process on the set A. The situation that H is multiplication by a function f, 0 ~ f
s
1, and H'multiplication by the function 1 - f. is studied in [2J, [4J.
o
In both cases we have H + H' = P , hence s = I. The situation with s > I
occurs when we are investigating recurrence properties of the Markov shift, as We shall see in the next section.
Theorem 1. For every f E
M
+ we have00 00
I
Qnf=
I
pns+lHf. n=1 n=O Proof. p(n-l)s+I Hf=
P(H' + H)P P(H' + H)PHf n n =L
(PH') IpH{PH ' ) 2pH nl+···+nk+k=n and therefore 00 n =L
L
(PH') IpH n=l nO+ ••• +nk+k=n n (PH') k pHf n (PH') k pHf..
00 00 00 n ~L
(PH I) IpH ••• (PH') "'pHf ~=OTheorem 2. The conservative part of X with respect to Q is {leH > O} n C, where C is the conservative part of X wi th respect to P.
4
-Proof.
L Suppose Ac {I H > o} n C. c
By theorem 1 we have
It follows by corollary 4 and corollary 2 that this sum is 0 or ~ on C. Put 00 Co
= {
I
Qn1A=
o} n C, n=l 00 C 1= {
I
Qn1A = oo} n C, • n=l then pSICo = 0 on C1' and since PHI A
o
we obtain00 ~
I
pns pH1A=
o on Co n=O ~I
pns pH1A n=O 0 co ~I
pnspslc
= 0 on CI,
n=O 0hence in particular PHI
A
o
= 0 on C.This means <lCPH1A >
=
0 , and therefore by corollary 3o
S~nce A
Oc A, we have m(Ao) = 0, and
on A.
Hence by lemma 1 the set {lCH > O} n C is a subset of the conservative part of X with respect to Q.
1.1.. 1£ A
that
- 5
-O} n C, then <ICHI
A> = 0, hence RIA = 0 on C. It follows
o
on C,hence A is a subset of the dissipative part of X with respect to
Q.
iii. Let g be a function with {g > O} == D and
r
n=OPng is bounded. Then0:> 00 00 00
L
Qng =L
pnspHg ~L
pnsg ::;L
png < 00,
n=1 n=O n=1 n=O
and D is a subset of the dissipative part of Xwith respect to
Q.
Corollary. If s
=
1 and H is multiplication by a function h with 0 ~ h ~ I,then {lcH > O} n C
=
{h > O} n C and we obtain a result of Lin [4].3. Recurrence properties for the Markov shift
In this section we shall give, with the aid of the operator
Q
of the previous section, a fast deduction of some recurrence properties of the Markov shift. These results go back to a paper of Harris and Robbins [3J.Let S be a measurable transformation on a measure space (n,F,M). A set A is
said to be uandering under S if no points of A return to Aunder the action of S,
and recurrent if M-almost all points of A return to A under the action of S.
A set is said to be dissipative if it is (mod M) the union of countablymany wandering sets, and conservative if every subset is recurrent.
Obviously, if An is the subset of A of the points which return exactly n times under S to A then A is wandering. Hence, if almost all points of Areturn
n
finite ,. many times to A, then A is dissipative.
Now let (~,F,M) be the realization space of Markov proces P on
(x,I>
with initial probability m, i.e. (n,F)n:=o(X,L),
and... PIA I>
6
-the i-th coordinate. X-IE and let S be the
n '
for all AO, ••• ,A E E, where X. denotes projection on
n 1 -I
Let Fn denote the a-algebra generated byX
o
L: , ••• ,shift transformation in (n,F).
Suppose that the initial measure is such that F can also be con'sidered as a
+
Markov operator on M (X,E,m). (This is the case if and only if meA)
=
0 ~ - P(·,A) = 0 m-a.e.) Let C be the conservative part of X with respect to P,00
and define C = {X ~ C for all n}.
n
Theorem 3.
00
i) The set n\C is dissipative.
ii) For every n and every A E
F ,
the set An
COO is recurrent.n
Proof. Because of Pin = 0 on C, we have
00
n\Coo = {X
OE D} = u {XOE Di } ,
i=1 where D
I,D2,. .. is a partition of D such that
tence of such a partition easily follows from
00 00
L
M(X E D. ) :=I
<IPnI D.> < 00,
n 1 n=O n=O 1. 00I
pnl is bounded. The exis~O D.
n= 1.
lemma 2. We then have
hence by the Borel-Cantelli lemma we obtain M{X E D. i.o.}
=
O. Almost alln 1.
points of {X
OE Di} return to this set under S only finitely many times, so
00
{X
o
E Di} is dissipative, and therefore {XOE D}
=
u {XOE Di} is also dis-i=1dipative.
00
ii) Without loss of generality we may assume that X = C, and therefore C n. Ch00 ~ A E F I' and d f ·e lne for every f c~ M+
s-where we consider the FO~easurablefunctions on nin the right-hand side as functions on X.
The operators Hand H' are Markov operators on (X,E,m) and satisfy (H + H')f = E(f(X
7
-Let A
Obe the set of points of A which return to A under SS at least once. Then u n=l ns } i ~ n, S W E: A <Xl M(A O) =
l
<IH(PH,)n-l pH1 > = <IHQ1> • n=1since by theorem 2 Q ~s conservative on {IH >
a},
we have QI=
I on {IH > OJ,and therefore M(A
O)
=
<IH1>=
M(A) < <Xl, hence AO=
A (mod M). Hence the set. s
A ~s recurrent under S , and therefore under S. Remark I. Let P be conservative. If A E F
s' then A E: Ft for every t
z
s, and we have actually shown that A is recurrent under st for every t. Hence almost all points of A return to A infinitely many times underS.Remark 2. The crucial point in the paper of Harris and .Robbins is that if there exists an algebra of recurrent sets generating
F
and a finite or a-finite equi-valent invariant measure, thenn
must be conservative. Therefore, if there exists a function u with 0 < u < <Xl on C, u=
0 on D and uP=
u~ then themea-dM'
sure M' defined by ---
=
u(XO) is (o-)finite and invariant under S, and the
dM
00
set C is conservative.
References
[IJ Foguel, S.R.: The ergodic theory of Markov processes. New York: Van
Nos-tran~ Reinhold Company, 1969.
[2J Foguel, S.R., Lin, M.: Some ratio limit theorems for Markov operators. Z. Wahrsch. verw. Geb. ~, 55-66 (1972).
[3J Harris, F.E., Robbins, H.: Ergodic theory of Markov chains admitting an infinite invariant measure. Proc. Nat. Acad. Sci. U.S.A. ~, 860--864 (1953).
[4J Lin, M.: On quasi-compact Markov operators. The Annals of Probability
l,
464-475 (1974).[5J Revuz, D.: Markov Chains, Amsterdam: North-Holland Publishing Company, 1975.