O tober WSPC Pro eedings Trim Size: 75in x 5in suquet revision

Description

Niveau: Supérieur, Master, Bac+5
O tober 19, 2006 18:17 WSPC - Pro eedings Trim Size: 9.75in x 6.5in suquet-revision 1 REPRODUCING KERNEL HILBERT SPACES AND RANDOM MEASURES ChARLES SUQUET Laboratoire P. Painlevé, UMR CNRS 8524, Bât M2, Cité S ientique, Université Lille I F59655 Villeneuve d'As q Cedex, Fran e We show how to use Guilbart's embedding of signed measures into a R.K.H.S. to study some limit theorems for random measures and sto hasti pro esses. Key words: Mathemati s Subje t Classi ation: 1. R.K.H.S. and metri s on signed measures In the late seventies, C. Guilbart [4, 5? introdu ed an embedding into a reprodu - ing kernel Hilbert spa e (R.K.H.S.) H of the spa e M of signed measures on some topologi al spa e X. He hara terized the inner produ ts on M indu ing the weak topology on the subspa e M+ of bounded positive measures and established in this setting a Glivenko-Cantelli theorem with appli ations to estimation and hypothesis testing. In this ontribution we present a onstru tive approa h of Guilbart's em- bedding following [20?. This embedding provides a Hilbertian framework for signed random measures.

• wspc - pro eedings


• let µ•1


• measure


• positive measure


• random measures


• topologi al dual

Sujets

• Measure
• Measure (mathematics)

Informations

• Publié par: rasum
• Nombre de lectures: 16
• Langue: English

Extrait

H M
X M
+M
X M
σ X μ
+ −(μ ,μ ) |μ| =
+ −μ +μ
Z
K(x,y) = r(x,u)r(y,u)ρ(du), x,y∈X,
U
ρ (U,U)
r :X×U→C
supkr(x, .)k 2 <∞.L (ρ)
x∈X
H K
r(.,u) μ X ρ
e
In
measures.
the

late
ariation
sev
Painlev?,
en
[20
ties,
osition
C.
wing
Guilbart
Universit?
[4
CES
,
ert
5]
t
in
sto
tro
to

repro
an
em
em
q
b
measure
edding
ab
in
1
to
y
a
is
repro
Octob

ositiv
ing
b
k
for
ernel
theorems
Hilb
in
ert

space
ha
(R.K.H.S.)
signed
and
w
of
r
the
F59655
space
a
R.K.H.S.
space
of
function
signed
satises
measures
HILBER
on
9.75in
some
W
top

ological
with
space
k
1.
(2),
.
dierence
He
o

b

e
the
hastic
inner
measures
pro
Hahn-Jordan

b
on
some
Classication:
a

total
the
W
w

eak
k
top
the
ology
tation
on
edding
the
use

w

e
Sub
dex,
of
leneuve
b
le
ounded
Cit?
p
ositiv
ositiv
some
e
CNRS
measures
and
and
atoir
established
SUQUET
in
RANDOM
this
SP
setting
ODUCING
a
6.5in
Gliv
rim
enk
-
o-Can
denote
telli
the
theorem
ernel
with
asso
applications
.
to

estimation
(Prop.2
and
that
h
19,
yp
the
othesis
of
testing.
w
In
p
this
e

ounded
tribution
W
w
denote
e
y
presen

t
and
a
random

its
e


and
h
y
of
limit
Guilbart's
study
em-
R.K.H.S.
b
to
edding
its
follo
v
wing
measure.
[20
e
℄
the
This
of
em

b
ernels
edding
ving
pro
follo
vides
represen
a
measures
Hilb
of
ertian
b
framew
Guilbart's
ork
to
for
ho
signed
sho
random
W
measures.

Mathematics
F
measures
Ce
signed

tegrable
Vil
v
I
on
Lil
metrics
Scientique,
(1)
M2,
B?t
where
this
is

p
to
e
limit
on
theorems
measurable
for
8524,
random
UMR
measures
P.
and
the
partial
e
sums
or
pro
L

ChARLES
Let
MEASURES
ords:
AND
b
A
e
T
a
KERNEL
metric
REPR
space
suquet-revision
and
x
let
Size:
w
T
denote
Pro
the
(2)
space
e
of
b
signe
WSPC
d
repro
me
k
asur
Hilb
es
space
on

the
18:17
Borel
It
Key
easily
-eld

of
ed

in
.
℄
A
under
signed
2006
measure
er
pro
is
W
is
e
-in
shall
o
discuss
er
some
for
applications
-almost
ofu∈U
Z
μ∈M r(x,u)μ(dx) = 0 ρ u μ = 0
X
2M H L (ρ)
2E L (ρ) {r(x,.), x ∈X}
2h :X→C H g∈L (ρ)
Z
h(x) = g(u)r(x,u)ρ(du), x∈X.
U
Ψ :H→E h →g
K M
Z
hμ,νi := K(x,y)μ⊗ν(dx,dy), μ,ν ∈M.K
2X
(M,h.,.i ) HK
Z
I :M→H, μ −→I := K(x,.)μ(dx).μ
X
Z Z
hh,I i = hdμ, hI ,hi = hdμ, h∈H,μ∈M.μ μ
X X
2ζ = Ψ◦I :μ →ζ M L (ρ)μ
Z
ζ (u) = r(x,u)μ(dx), u∈U.μ
X
ρ U =N r r(x,i) :=
f (x) x ∈X i ∈ N f : X → Ri iR
μ ∈ M f dμ = 0 i ∈ NiX P
2kfk <∞ii∈N ∞
X
2K(x,y) = f (x)f (y), x,y∈X .i i
i∈N
R R P
2μ ℓ (N) ζ = f dμ H I = f dμ fμ i μ i ii∈NX Xi∈N
f Hi
d dX = U = R r(x,u) := exp(ihx,ui) x,u ∈ R
dρ R
Z
dK(x,y) = exp(ihx−y,ui)ρ(du), x,y∈R .
dR
-almost
measure
Under
b
L
y

ting
anne

that
the
follo
for
The
e
rim
ak
y
T
the
1.1.
(2)
Example
ed
applies.
of
1.1
b
Theorem
uous
where
W
,
19,
examples
and
some
the
examine
op
,
follo
us
and
Let
T
(8)
[20
,
h
where
are
the
em

ose
of
e
functions
es
satises
(3)
into
v
of
6.5in
dding
-
e
by
emb
of
separates
ac
the
d
measures,
b
i.e.
hold.
the
lowing
only
It

from
The
ery
d)
elongs

Example
h
e
that
Theorem
and
with
have
pro
we
theorem
over
in
e
to
Mor
,
(6)
out
by

for
as
all
p
of
on
e
This
ac

is
k
the
then
n
and
dene
ull
assume
measure.
2
T
9.75in
o

ha
18:17
v
Octob
e
d
a
sp
b
e
ounded
in
k
b
ernel
subsp
w

e
e
also
et
assume
a)
that
erties
subsp
pr
dense
fol
a
.
to
easily

ws
is
(4)

ev
(5)
(3),
formula
b
the
to
by
.
on
1.2.
.
ak
Then
(1),

1.1.
o
℄
pr
,
inner
in
an
v
es
is

whic
b)
wing
.
the
,
gathered
es
and
ac
in
sp
eddings
ert
b
Hilb
the
of
ab
isometry
and
an
ho
denes
tial
(4)
a
esentation
ounded
epr
ositiv
r
measure
The
essen
(4)
.
that
giv

the
unique
tin
a
stationary
is
ernels
is
.
represen
,
ted
for
in
if
on
that
e
er
ther
moreo
if
e
only
.
and
suquet-revision
b
x
y
Size:
if
T
to
Pro
elongs
WSPC
b
2006
function
er
A
.
(7)R
ζ (u) = exp(ihx,ui)μ(dx) =: μˆ(u) μμ dRR
I (x) = exp(−ihx,ui)μˆ(u)ρ(du).μ dR
X =U = [0,1] ρ = λ+δ λ1
δ 1 r(x,u) := 1 (u) K(x,y) =1 [x,1]
2−max(x,y) ζ (u) =μ([0,u])μ
R
M f → f dμ
X
f ∈ F F X X
F = C(X) X
F = C (X)0
ε A X
|f(x)| < ε x ∈X\A X
C(X) = C (X) C (X)0 0
M fi
C (X) H C (X)0 0
(μ ) M sup |μ |(X)<∞n n≥1 nn≥1
H I Iμ μn
M μ μn
•μ M
σ G
• ω
μ : (Ω,F,P)−→ (M,G), ω →μ .
•(Ω,F,P) μ P
• −1P ◦(μ ) G
Pn• −1μ = n δ XX in ii=1PN
X δ NYii=1
Y Xi
M
σ X M σ G
M H
to
for

ev
the
ery
oin
p
in
ositiv
the
e
w
v
app
there
the
is
examples
a
's

are
subset
of

framew
of
setting
the
edding


h
(under
that
while
study
the
to

℄
space
[20
,
in
Kallen
used
b
are
When
ernels
of
for
sucien
ev
random
ery

k
p
These
y
and

of
probabilit
function
distribution
.
the
In
Example
the
w
sp
let
ecial
ology

on
where
where

ts
is
oin

e

the
the

is
is
,
table
Here
is
3
measures
suquet-revision
and
6.5in
Borel
x
top
9.75in
ositiv
.

Endo
and
w
ab
ed

with
Still
the
y
suprem
℄
um
an
norm,
at
Size:
Here
rim
uous
T
is

space
Pro
w
is
s