Scribd
Upload
61 views

Short Proof of A "Radon-Nikodym Theorem" For Additive Functionals

This document provides a short proof of a Radon-Nikodym theorem for additive functionals. It summarizes the key assumptions and definitions needed, including the definition of an additive functional and Palm measures. It then states two theorems: Theorem 2 shows that two additive functionals are equal if their Palm measures agree, and Theorem 3 shows they are equal if they agree on the predictable sigma-algebra. The proof uses these two theorems along with properties of Palm measures and time changes to show that if one additive functional is absolutely continuous with respect to another, there exists a random variable such that the integral of this variable with respect to one functional equals the other functional.

Uploaded by

Jacek Podlewski
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
61 views

Short Proof of A "Radon-Nikodym Theorem" For Additive Functionals

This document provides a short proof of a Radon-Nikodym theorem for additive functionals. It summarizes the key assumptions and definitions needed, including the definition of an additive functional and Palm measures. It then states two theorems: Theorem 2 shows that two additive functionals are equal if their Palm measures agree, and Theorem 3 shows they are equal if they agree on the predictable sigma-algebra. The proof uses these two theorems along with properties of Palm measures and time changes to show that if one additive functional is absolutely continuous with respect to another, there exists a random variable such that the integral of this variable with respect to one functional equals the other functional.

Uploaded by

Jacek Podlewski
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Z. Wahrscheinlichkeitstheorieverw. Geb.

25, 159-162 (1973)


9 by Springer-Verlag 1973

Short Proof of a "Radon-Nikodym Theorem"


for Additive Functionals
Joseph Horowitz

w1. By a flow O=(Ot), t~R, on a probability space (~2,~ o , p) we mean a one-


parameter group of bimeasurable, measure preserving bijections 0t: s ~ ~2, where
0 o =identity, and the mapping (t, co)--, Or(co) is measurable, R being given the
usual Borel o-field. We assume f2 is endowed with a right-continuous, increasing
family of o-fields {4~ t~R, whose union generates ~ o and with the property
that, if X is an ~s%measurable random variable, Xo 0 t is ~+t-measurable.
~o Let
be the P-completion o f ~ ~ and let 4 be obtained by adjoining to 4 o all sets in
of measure zero. The following conditions are assumed to hold:
(i) If Y is an ~,~o~ random variable, the process (Yo Ot) is progres-
sively measurable relative to {4}.
(ii) For any non-negative (respectively, bounded) random variable X on
(~2,~ P), there exists a non-negative (respectively, bounded) ~o~
function Jf such that
(,) 2oOr=E(XoOrl~r) a.s.
for each bounded stopping time T o f the family {4}, t~R.
These assumptions are all satisfied by the "canonical spaces" constructed in
[3]; in particular see Theorem 4.1 of [3] concerning (i) and (ii).
w An additive functional is a random process ~ = {~(t, co): t~R, co~f2} which
satisfies
(1)
(2) almost every path is right-continuous, nondecreasing,
(3) for each s, t e R there is a set Ns, t e ~ of measure zero such that
(4) e(t+s, co)=e(t, co)+c~(s, Otco), coCN;,t,
(5) Ee(1)< o% and finally,
(6) e(t)-c~(s) is 4-measurable i f s < t .
Our purpose is to give an easy proof of Theorem 5.1 of [3] based on the
theory of "Palm measures" as developed by Geman and the author [2]. This
theorem may be stated as follows:
Theorem 1. Let ~ and fi be additive functionals such that, for almost every coEf2,
the measure fl(dt, co) on R is absolutely continuous with respect to c~(dt, co) (written
fi(dt, co)~ c~(d t, co)). There exists a random variable Y > O, ~~ such that,
160 J. Horowitz:

for almost every co,

(7) fi(t, co) i YoOs(co) d~(s, co) for every t~ R.


0

This is the analogue of a well-known theorem of Motoo (see [1]) for Markov
additive functionals. We omit an additional assertion in [3] concerning predictable
additive functionals. The proof there requires the decomposition of supermartin-
gales, the decomposition of an additive functional into discrete and continuous
parts, the weak topology a(L ~, L1), and the theory of time changes.
w 3. For an additive functional c~ we define the corresponding Palm measure
by ([2], [3, p. 127])
1
(8) ~(A)=E.[ IAoOtd~(t), A~ ~
0

A monotone class argument yields


1
(9) E~(X) = E ~ X o 0t d~(t)
0

for any ~ ~ variable X > 0, where E~ denotes integration with respect to ~.


Because of (5), ~ is a finite measure.
Our proof of Theorem 1 is based on the following two results.
Theorem 2. For additive functionals ~ and fl, ~ = ~ iff, for almost every co~(2,
~(t, co)=fi(t, co) for all t~R.
Theorem 3. For additive functionals ~ and fi, ~ = ~ iff ~ and ~ agree on ~o ~
Theorem 2 was announced (in a less precise form) in [2]. We need one bit of
preparation.
Lemma. Let ~ be an additive functional.
a) E~(t)=tE~(1), t~R. 1
b) I f X is a non-negative, ffo~ function and E ~ XoOsd~(s)< ~ ,
t 0
then 7 (t) = ~ X o Osdc~(s) is an additive functional.
0
c) For any non-negative ~ • ~~ function u(t, co),

(10) E ~ u(s, O~co) d~(s, co)= ~ ~ u(s, co) ds.


R R

(~ is the Borel a-field on R.)


t
Parts (a) and (b) are trivial (for (b) use [5, p. 109]). As for (c), ify (t)= ~ IAo Ofla(s)
for A e ~ -~ then ~ satisfies (1)-(5). By (4), o
t

(~ 1) t P~(A) = E S I~o O~d~ (s),


0

so (10) holds whenever u is the indicator of a ~ • ~-~ rectangle, hence


for all u as stated. (Eq. (10) is essentially due to Mecke [4].)
S h o r t P r o o f of a " R a d o n - N i k o d y m T h e o r e m " for A d d i t i v e F u n c t i o n a l s 161

We can now prove Theorems2 and 3. Choosing u(s, co)=Ito.tl(s)XoO_~(co)


in (10), with X being non-negative and f~ we obtain

(12) e(x :(t))= ~ i x oo_~ d~.


0

If ~ = ~ , we conclude ~(t)=p(t) a.s. for each t, and Theorem2 follows from


the a.s. right continuity of a and ft.
Denote by ~(s) the time change [5] associated with ~: ~(s)=inf{t: cr
By (6), ~(s) is a stopping time of {o~} for each s>0. Let X be a non-negative
fr~ function and J) as in (ii). Since s<c~(1)~(s)__< 1, and {s<e(1)}~
~ ( s ) / x 1 ' we have

1 co

E .[XoOsdo~(s)=E ~ Ito,~(1))(s) XoOsc~)ds [5, p. 108]


0 0

co

= [. E[XoO~(~)~; s<c~(1)] ds
0

co
= ~ E[f[oOs(~)A1; s<c~(1)] ds by (*)
0

or

(13) E~(X) = E;~(2)

from which Theorem 3 is immediate.

w Let c~ and fl be as described in Theorem 1. We shall show first that ~ P ~ .


1 1
Indeed, if A afr ~ P~(A)= 0 implies ~ IAO 0, de (t) = 0 P-a.s., hence ~ IAO 0t dfi (t) = 0
0 0
P-a.s., and ~(A)=0. Next, by the Radon-Nikodym theorem there is an Yo~
measurable function Y > 0 such that, on fo ~ dR = YdS. Thus, from (8),

1
~(A) = E j" IAoOt YoOtdc~(t)
0

=~(A), AEffo~

where 7(t)= i YoO~dc~(s). Since Y is fo~ and ET(1)=~(f2)< o0,7 is


0
an additive functional (Lemma (b)), and from Theorem 3, applied to ? and fl, we
obtain Theorem 1.
162 J. Horowitz: Short Proof of a "Radon-Nikodym Theorem" for Additive Functionals

References
1. Getoor, R. K.: Additive functionals and excessive functions. Ann. Math. Statistics 36, 409-422 (1965).
2. Geman, D., Horowitz, J.: Abstract 71T-Fll, Notices AMS 81, no. 6, p. 972 (1971).
3. Lazaro, J. de Sam, Meyer, P.A.: M6thodes de martingales et th~orie des flots. Z.Wahrscheinlich-
keitstheorie verw. Geb. 18, 116-140 (1971).
4. Mecke, J.: Stationiire zuNllige Mage auf lokalkompakten Abelschen Gruppen. Z. Wahrscheinlich-
keitstheorie verw. Geb. 9, 36-58 (1967).
5. Meyer, P.A.: Probability and Potentials. Waltham-Toronto-London: Blaisdetl 1966.

Joseph Horowitz
Mathematics Department
University of Massachusetts
Amherst, Massachusetts 01002
USA

(Received August 1, 1972)

You might also like