HAL Id: hal-00013282
https://hal.archives-ouvertes.fr/hal-00013282v5
Preprint submitted on 24 Jun 2009
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
Density of paths of iterated Levy transforms of
Brownian motion.
Marc Malric
To cite this version:
Marc Malric.
Density of paths of iterated Levy transforms of Brownian motion..
2007.
�hal-00013282v5�
Brownian motion
Mar Malri
June 24, 2009
Abstra t : The Lévy transform of a Brownian motion B is the Brownian motion
B
′
t
=
Z
t
0
sgn (B
s
) dB
s
. CallT
the orresponding transformation onthe Wiener spa e
W
. We establishthat a. s. the orbit ofw(∈ W )
underT
is dense inW
for the ompa t uniform onvergen e topology.related tools.
Let
(Ω, F , P)
betheprobabilityspa ewhereallrandomelementsaredened, and(W, W, π)
the Wiener spa e. Any measurable mapfromΩ
toany mea-sure spa e, denedP
−
a.e., will be alled a random variable. IfX
is a r.v. withvaluesinsome measurablespa eA
,theprobabilitymeasureP
◦ X
−1
on
A
is alled the law ofX
, and denoted byL(X)
. Forinstan e, aW
−
valued r.v. with lawπ
is aBrownian motion.The Lévytransform
T
: W −→ W
isdenedπ−
a.e. and preservesπ
. Given a Brownian motionB
, we denote byB
n
its
n−
Lévy iterate, that is, the Brownian motionB
n
= T
n
◦ B
.
From nowon,
T > 0
,ϕ ∈ C([0, T ], R)
andε > 0
arexed, andB
aBrownian motion. ThegoalistoprovethattheeventE = E
ε
= {∀n ≥ 0, kB
n
−ϕk
∞
>
ε}
isnegligible,wherekf k
∞
= kf
|[0,T ]
k
∞
. Itsu esinfa ttoshowP
(E) < ε
, be auseE
ε
1
⊂ E
ε
2
when
ε
2
< ε
1
.The idea is to onstru t from B another sto hasti pro ess
Γ : Ω −→ W
, whi hdepends onT
,ϕ
andε
, and has the following threeproperties :(i) The law of the pro ess
Γ
, i.e., the probabilityF −→ P(Γ
−1
F )
on
(W, W)
, isabsolutely ontinuous w.r.t. the lawπ
ofB
.(ii) Forsomedeterministi
r ≥ 0
,onehasΓ
r
= B
r
,thatis,T
r
◦Γ = T
r
◦B
. (iii)P
(∀n ≥ 0kΓ
n
− ϕk
∞
> ε) < ε
.Property (i) impliesthat
T
◦ Γ
an be (almosteverywhere) dened, inspite ofT
not being everywhere dened. Indeed, ifT
′
: W −→ W
is another version of
T
,that isifT
′
= T
a.e., the set
{T
′
6= T}
isπ−
negligible;hen e, by (i),Γ
−1
{T
′
6= T}
isP
−
negligible,andT
′
◦ Γ = T ◦ ΓP−
a.s.. Similarly, one andenethe sto hasti pro essesΓ
n
= T
n
◦ Γ
,whi hverify
Γ
0
= Γ
and
T
◦ Γ
n
= Γ
n+1
.Proposition 1. For xed
G ∈ W
andε > 0
, let us suppose that there exists a sto hasti pro essΓ : Ω −→ W
satisfying properties (i), (ii) and(1)
P
(∀n ≥ 0, Γ
n
∈ G) < ε.
Then :
Proof Take
G ∈ W
and putF =
T
n≥0
T
−n
G
. Then, forr ≥ 0
, (3)T
−r
F =
\
n≥r
T
−n
G ⊃
\
n≥0
T
−n
G = F.
But these two sets,
F
andT
−r
F
, in luded in one another, have the same
π−
probabilitybyT
−
invarian e;soequalityF = T
−r
F
holdsupto
π−
negligibility. As the laws ofΓ
andB
are absolutely ontinuous w.r.t.π
(this is where (i) is used), we haveΓ
−1
(F ) = Γ
−1
(T
−r
F )
and
B
−1
F = B
−1
(T
−r
F )
up to
P
−
negligible events. In other words, almost surely, we have{Γ ∈ F } =
{Γ
r
∈ F }
and
{B ∈ F } = {B
r
∈ F }
. Consequently, hoosing
r
given by (ii) and usingΓ
r
= B
r
, we have
{Γ ∈ F } = {B ∈ F }
a.s.. That is tosay : (4)P
(∀n ≥ 0, Γ
n
∈ G) = P (∀n ≥ 0, B
n
∈ G)
Spe ializing
G = {w ∈ W
;kw − ϕk
∞
> ε}
, we obtain:P
(E) < ε
Proposition 1 redu es the proof of the approximation theorem to the on-stru tion of a pro ess
Γ
verifying (i), (ii), and (iii). We shall rst hooser
in asuitableway, then work ba kwards, inr
steps, fromΓ
r
= B
r
to
Γ = Γ
0
; ea h step ( alled a Lévy raise) will onstru t
Γ
n−1
from its Lévy transform
Γ
n
= T ◦ Γ
n−1
. The sequen e
(Γ
r
, Γ
r−1
, ..., Γ
0
)
is given aname : Denition 1. Given
r ∈ N
, a sequen e(Γ
r
, Γ
r−1
, ..., Γ
0
)
is alled a sequen e of
B−
raisedBrownianmotionsof indexr
ifea hΓ
n
isa
W
−
valuedr.v. with law absolutely ontinuous w.r.t.π
, ifΓ
r
= B
r
, and ifwe have
Γ
n
= T ◦ Γ
n−1
for
0 < n ≤ r
.Infa t,for onvenien eof exposition,let usenlarge thelteredprobability spa e
Ω
, wesuppose it ontainsthe whole sequen e(B
′n
)
n∈N
of the Brownian iterates ofB
′
, B.M. independent fromB
. So we an assert :Corollary 1. To prove the approximation theorem, it su es to exhibit a sequen e
(Γ
r
, Γ
r−1
, ..., Γ
0
)
of
B−
raised Brownian motions of indexr
su h that(5)
P
(kΓ
n
− ϕk
of a sequen e of
B−
raised Brownian motions,and (iii)is implied by (1).A Lévy raise starts with a given
W
−
valued r.v.Γ
n
, and yields some r.v.
Γ
n−1
with Lévy transform
Γ
n
. Given a
W
−
valued r.v.V
, how an one nd a r.v.U
su h thatV = T ◦ U
? KnowingV
is equivalent toknowing|U|
, so todeneU
one onlyneedstode idewhi hsign isassignedtoea hex ursion of|U|
away fromzero. To makethis rigorous, weneed aformaldenition of the ex ursions of a path and of their signs.Notation 1. For
w ∈ W
andq > 0
, denote byZ(w) = {s ≥ 0/w(s) = 0}
the set of zeros ofw
, and deneg
q
(w) = sup ([0, q] ∩ Z(w)) ≥ 0
(last zero before q) andd
q
(w) = inf ([q, ∞] ∩ Z(w)) 6= ∞
(rst zero after q).Fix a dense sequen e
(q
n
)
in[0, ∞]
. To ea hw ∈ W
, we an atta h the sequen e(e
p
)
of disjoint,open intervalsobtained from the sequen e(6)
((g
q
1
, d
q
1
), (g
q
2
, d
q
2
), ..., (g
q
n
, d
q
n
), . . . )
bydeletinganintervalwheneveritalreadyo ursearlierinthesequen e. The
e
p
are the ex ursionintervals ofw.π−
almost surely,there are innitely many of them, and they are the onne ted omponents of the open set[0, ∞]\Z(w)
. The intervale
p
(w)
will be alled the p-th ex ursion interval ofw
;e
p
is an interval-valued measurable map, dened on(W, W)
up toπ−
negligibility.Sin e
w
does not vanish one
p
(w)
, its sign is onstant on this interval; this sign willbe denoted byS
p
(w)
, and the sequen e(S
p
)
will be alledS
. IfB
isaBrownianmotion,the sequen e ofr.v.S
◦ B = (S
p
◦ B)
isa oin-tossing; thismeans,itisani.i.d. sequen e,withea hr.v.S
p
◦B
uniformlydistributed onthe set{−1, +1}
. Moreover,S
p
◦ B
and|B|
are independent. (SeeChap. XII of [R,Y℄).Lemma 1. Dene
I
: W −→ W
byI
(w)(s) = inf
[0,s]
w
; that is,|w| =
T
w − ITw
forπ−
a.a.w
.Proof Fix
s ≥ 0
. On[0, s]
,B
1
= |B| − L ≥ −L
s
= L
g
s
= B
1
g
s
. SoB
1
g
s
= inf
[0,s]
B
1
, and|B
s
| = B
1
s
+ L
s
= B
s
1
− inf
[0,s]
B
1
.Notation 2. If
f
is a ontinuous fun tion on[u, v]
, we allarg min
[u,v]
f
, resp.arg max
[u,v]
f
, the largestt ∈ [u, v]
su h thatf (t) = min
[
u, v]f
, resp.f (t) = max
[
u, v]f
.Lemma 2. Let
A
,A
′
and
A
′′
be three measure spa es; let
µ
1
andµ
2
be two measures on A,f
a measurable map fromA
toA
′
, andν
a measure onA
′′
. Ifµ
1
≪ µ
2
, then (i)µ
1
◦ f
−1
≪ µ
2
◦ f
−1
; (ii)µ
1
⊗ ν ≪ µ
2
⊗ ν
. Proof (i)IfF ⊂ A
′
ismeasurableandif(µ
2
◦ f
−1
)(F ) = 0
,thenµ
2
(f
−1
F ) =
0
, soµ
1
◦ f
−1
(F ) = µ
1
(f
−1
F ) = 0
. (ii)IfameasurablesubsetF
ofA×A
′′
isnegligiblefor
µ
2
⊗ν
,thenν−
almost all itsse tionsF
y
verifyµ
2
(F
y
) = 0
. Hen e they alsoverifyµ
1
(F
y
) = 0
, and onsequently(µ
1
⊗ ν)(F ) =
R
µ
1
(F
y
)ν(dy) = 0
.Lemma 3. Let
τ = (τ
p
)
be oin-tossing,τ
′
= (τ
′
p
)
a r.v. with values in{−1, 1}
N
su h that
τ
′
p
= τ
p
for all but a.s. nitely manyp
, andX
a r.v. independent ofτ
.Then
L(X, τ
′
) ≪ L(X, τ )
.
Thislemmasaysthat hangingnitelymanyvaluesof
τ
doesnotperturb too mu h the joint law ofX
andτ
. For instan e, it implies that a pro ess obtained from a Brownian motion by hanging the signs of nitely many ex ursions has alawabsolutely ontinuous w.r.t.π
. This is alled 'prin ipe de retournement des ex ursions' in [M℄.Proof If
u = (u
1
, u
2
, ...)
is an innite sequen e, denote byu
p]
the nite sequen e(u
1
, ..., u
p
)
and byu
[p+1
the innite sequen e(u
p+1
, u
p+2
, ...)
. We have(x, u) = f
p
(x, u
p]
, u
[p+1
)
forsome fun tionf
p
.We have to show that if
F
is measurable set su h thatP
[(X, τ ) ∈ F ] = 0
, thenP
[(X, τ
′
) ∈ F ] = 0
. So assumeP
[(X, τ ) ∈ F ] = 0
. Forp ∈ N
, sin eτ
p]
takesvalues in{−1, 1}
p
, we an write (7)X
σ∈{−1,1}
p
P
f
p
(X, σ, τ
[p+1
) ∈ F and τ
p]
= σ
= P [(X, τ ) ∈ F ] = 0.
Usingthe independen e of
τ
p]
and(X, τ
[p+1
)
, this be omes :(8)
X
σ∈{−1,1}
p
2
−p
P
f
p
(X, σ, τ
[p+1
) ∈ F
= 0;
Sofor ea h
p ∈ N
and ea hσ ∈ {−1, 1}
p
, the event
{f
p
(X, σ, τ
[p+1
∈ F }
is negligible. Sin eτ
′
p
(ω) = τ
p
(ω)
for allp
larger than someN(ω)
, one hasP
[(X, τ
′
) ∈ F ] =
lim
p→∞
P
(X, τ
′
) ∈ F and τ
′
[p+1
= τ
[p+1
=
lim
p→∞
X
σ∈{−1,1}
p
P
f
p
(X, σ, τ
[p+1
) ∈ F and τ
p]
′
= σ and τ
[p+1
′
= τ
[p+1
.
This is null be ause the event
f
p
(X, σ, τ
[p+1
) ∈ F
is negligible, as shown above.Proposition 2. (me hanism of a Lévy raise) Suppose given the followingthree r.v. :
(i)
V
, aW
−
valued r.v., su hthatL(V ) ≪ π
; (ii)τ = (τ
p
)
p∈N
, a oin-tossing independent ofV
; (iii)τ
′
= (τ
′
p
)
p∈N
, a r.v. valued in{−1, 1}
N
, su h that the random set
p ∈ N
τ
p
′
(ω) 6= τ
p
(ω)
isa.s. nite.Thenthere exists a unique
w
−
valued r.v.U
su h that (9)|U| = V − I ◦ V and S
p
◦ U = τ
′
p
f or each p.
It is measurable w.r.t. the
σ−
eldσ(V, τ
′
)
and we have
L(U) ≪ π
andT
◦U = V
. Foranyn ≥ 0
,wehaveU = B
n
ontheevent
{V = B
n+1
and τ
′
=
Proof We start from
L(V ) ≪ π = L(B) = L(B
1
)
. Using Lemma 2 (i) we write
L(V, V −I◦V ) ≪ L(B
1
, B
1
−I◦B
1
)
. ByLemma2(ii),the oin-tossing
τ
(resp.S
◦ B
)whi hisindependentofV
(resp.B
1
) anbeaddedontheleft (resp. right), and we obtain
L(V, V − I ◦ V, τ ) ≪ L(B
1
, B
1
− I ◦ B
1
, S ◦ B)
; by Lemma1,the right-handside is
L(B
1
, |B|, S ◦ B)
. Lemma 3allows us to repla e
τ
byτ
′
in the left-hand side, sowe nallyhave
(10)
L(V, V − I ◦ V, τ ) ≪ L(B
1
, |B|, S ◦ B)
Now,we all
W
+
thesetofnon-negativepathsand
f : W
+
×{−1, 1}
N
−→
W
the measurable fun tion su h thatw = f (|w|, S(w))
. We remark thatB = f (|B|, S ◦ B)
andwe deneU = f (V − I ◦ V, τ
′
)
;this isthe unique r.v.
U
su h that|U| = V − I ◦ V
andS
◦ U = τ
′
. To verify that
L(U) ≪ π
andT
◦U = V
,weapplyLemma2(i)to(2)withthefun tionsg(x, y, σ) = f (y, σ)
andh(x, yσ) = (f (y, σ), x)
. Withg
weobtainL(U) ≪ L(B)
,the rst laim. Withh
we obtainL(U, V ) ≪ L(B, B
1
)
; this implies
T
◦ U = V
sin e the jointlawL(B), (B
1
)
is arried by the graphof
T
. Last, onthe event{V = B
n+1
andτ
′
= s ◦ B
n
}
,using the denition of
U
and Lemma 1 wehave(11)
U = f (V −I◦V, τ
′
) = f (B
n+1
−I◦B
n+1
, S◦B
n
) = f (|B
n
|, S◦B
n
) = B
n
.
Proposition 3. Denote by
P
f
(N)
the set of all nite subsets ofN
. Fixr
inN
. For ea hn ≤ r
, let be givenN
n
, a r.v. with values in
P
f
(N)
, andΣ
n
= (Σ
n
p
, p ∈ N
n
)
, a r.v. taking its values inS
M∈P
f
(N)
{−1, 1}
M
, su h thatΣ
n
(ω) ∈ {−1, 1}
N
n
(ω)
. Starting withΓ
r
= B
r
, we an dene a sequen e
(Γ
n
)
n≤r
su h thatΓ
n−1
is the
W
−
valued r.v.U
obtained in Proposition2 fromV = Γ
n
,
τ = S ◦ B
′n−1
,
τ
′
p
=
Σ
n
p
(ω)
ifp ∈ N
n
(ω)
τ
p
(ω)
elseThen the sequen e
(Γ
n
)
n≤r
is aB
-raised Brownian motions of indexr
.Proof First, we verify that the
Γ
n
an be onstru ted stepwise. Assuming
Γ
n
Proposition 2appliesto
V = Γ
n−1
and
τ = S ◦ B
′n−1
(they are independent. The r.v.
Γ
n−1
= U
yielded by Proposition 2 alsosatises
L(Γ
n−1
) ≪ π
, and is measurable in
σ(F , τ
′
)
.
The rest ofthe proof willexhibit asequen e
(Γ
n
)
n≥r
of B-raised motions su hthatΓ
J−n
= C
n
. StartingwithΓ
r
= B
r
,theotherΓ
m
willbeindu tively dened : ifm < r
, supposeΓ
m+1
has been dened, is
σ(B
m+1
)−
measurable, and veriesL(Γ
m+1
≪ π
;deneΓ
m
as the r.v.
U
obtained in Proposition2 from (12)V = Γ
m+1
,
τ = S ◦ B
m
,
τ
p
′
=
Σ
J−m−1
(ω)
ifp ∈ N
J−m−1
(ω)
τ
p
(ω)
elseThis is possible sin e
V
andτ
are independent andN
J−m−1
is a.s. nte; the resultΓ
m
veriesL(Γ
n
) ≪ π
andT
◦ Γ
m
= Γ
m+1
. To show thatΓ
m
isσ(B
m
)−
measurable, it su es to show that so is
τ
′
; this may be done separatelyonea hoftheevents
{J ≤ m}
,{J = m + 1}
,...,{J = r}
,be ause they formaσ(B
m
)−
partitionofΩ
. On{J ≤ m}
, wehaveτ
′
= S ◦ B
m
; this isσ(B
m
)−
measurable. To see what happens for other values of
j
, introdu eϕ
n
andψ
n
su h thatN
n
= ϕ
n
(B
J−n
)
andΣ
n
= ψ
n
(B
J−n
)
for0 ≤ n < k
. Forj ∈ {m + 1, ..., m + k}
, wehave on{J = j}
(13)τ
′
= ψ
n
(B
J−n
)1
ϕ
n
(B
J −n
)
+ τ 1
Ω
− 1
ψ
n
(B
J −n
)
Thisisσ(B
m
)−
measurable too. We have established that
Γ
r
, ..., Γ
0
exist and formasequen e of
B−
raisedmotions;itremainstosee thatΓ
J−n
= C
n
. This is done in two steps. Firstly, by indu tion on
m
, we haveΓ
m
= B
m
on
{J ≤ m}
: this holds form = r
, and if it holds form + 1
, it holds form
too, owing to the last statement in Proposition 1. Consequently,Γ
m
=
B
m
on{J = m}
,thatis
Γ
J
= B
J
= C
0
. Se ondly,topro eedbyindu tion on
n
,wewillassumethatΓ
J−n
= C
n
forsome
n ≥ 0
,and showΓ
J−n−1
= C
n+1
. It su es to show this equality on the event
{J = j}
; on this event, using the denitionofΓ
m
with
m = j − n − 1
andtheinequalitym = j − n − 1 < j
, the r.v.Γ
J−n−1
satisesbothT
(Γ
j−n−1
) = Γ
j−n
= C
n
= T(C
n+1
)and (14)S
◦ Γ
j−n−1
=
Σ
n+1
onN
n
S
◦ B
j−n−1
else= S ◦ C
n+1
.
These twoequalities entail
Γ
J−n−1
= C
n+1
a.s. on
{J = j}
.Lemma 4. Let
(j, k) ∈ N
2
and
Q
andR
be two r.v. su hthatk ≤ j − 1
and0 ≤ Q ≤ R
. On the event{∀n ∈ {k, ..., j − 1}Z ◦ B
n
∩ (Q, R) = ∅}
that the rst iterates of
B
do not vanish betweenQ
andR
, there exists a (random) isometryi : R −→ R
su h thatB
j
= i ◦ B
k
on the interval
(Q, R)
.Proof By indu tion, it su es toshow that if
B
j−1
doesnot vanish on the interval
(Q, R)
, thenB
j
= i ◦ B
j−1
on
(Q, R)
, for some random isometryi
. This is just Lemma 5 withj = k + 1
andB
j−1
instead of
B
k
, so we may suppose that
j = 1
.On theevent
{Q = R}
,theresultistrivial. On{Q < R} ∩ {Z ◦ B ∩ (Q, R) =
∅}
, the lo al timeL
is onstant on[Q, R]
be ause its support isZ ◦ B
, and the signofB
is onstanton(Q, R)
;soB
1
= |B| − L = i(B)
on
(Q, R)
,wherei
is the randomisometryx 7→ xsgn B
(Q+R)/2
− L
(Q+R)/2
.Notation 3. For
w ∈ W
, the p-th ex ursion intervale
p
(w)
was dened earlier; the numberh
p
(w) = max
s∈e
p
(w)
|w(s)|
will be alled the height of theorresponding ex ursion.
Lemma 5. Let
X
be a pro ess whose law is absolutely ontinuous w.r.t. Wiener measure. Almost surely,• lim
p→∞
h
p
(X)1
{e
p
(X)⊂[0,t]}
= 0
;•
P
p∈N
h
p
(X)1
{e
p
(X)⊂[0,t]}
= ∞
;•
the set(
P
p∈M
h
p
(X)1
{e
p
(X)⊂[0,t]}
, M ∈ P
f
(N)
)
is dense in [0, ∞)
;•
betweenanytwo dierentex ursionsofX
, thereexistsathirdone,with height smaller than any given random variableη > 0
.Proof By a hange of probability, we may suppose that
X
is a Brownian motion. It is known (see Exer ise (VI.1.19) of [RY℄) that whenη → 0
+
, the number
P
p
1
{e
p
(X)⊂[0,t]}
1
{h
p
(X)>η}
of down rossings of the interval[0, η]
by
|X|
beforet
is a.s. equivalent toη
−1
L
at
0
. This easily implies(i) and (ii),wherefrom (iii)follows.Last, between any two ex ursions of
X
there are innitely many other ones (be auseX
hasnoisolatedzeroes)and,by(i),onlynitelymanywithheights aboveη
, when e (iv).Notation 4. An ex ursion whose interval is in luded in
[0, t]
will be alled at−
ex ursion.Itremains todes ribe the
N
n
and
Σ
n
,i.e., to hoose the signs of nitely manyex ursions whenLévy-raisingfrom
Γ
n
to
Γ
n−1
. Thiswillbedone soon; we rst need some notation and alemma.
Notation 5. If
e
′
and
e
′′
are two ex ursions of a path (or of a pro ess),
e
′
≺ e
′′
means thate
′
is anterior toe
′′
:s
′
< s
′′
for alls
′
∈ e
′
ands
′′
∈ e
′′
. For an ex ursione
ofw
, we denote byi
w
e := inf{w
s
; s ∈ [0, d
e
]}
.Denition 2. An ex ursion
e
of a pathw ∈ W
is said to be tall if it is positive (this implies that the pro essI
w
remains onstant duringe
); and if for any ex ursione
′
of
w
su hthati
w
e
′
= i
w
e
and higherthane
, thene
′
= e
. Formally,
e
istall if it is positive and if(15)
max (w(s); s ≥ 0, (Iw)(s) = i
w
e) = max (w(s); s ∈ e) .
Lemma 6. Let
η
be a positive number,m ≥ 1
be an integer andw ∈ W
a path. Lete
1
, . . . , e
m+1
bem + 1
dierentt−
ex ursions ofw
, numbered in hronologi al order :e
1
≺ · · · ≺ e
m+1
; allh
1
, . . . , h
m+1
their respe tive heights. Letf
1
, . . . , f
p
denote all ex ursions ofw
whi h are anterior toe
m+1
and whose heights are≥ min(η, h
1
, . . . , h
m+1
)
, numbered in reverse hrono-logi al order : letg
1
, . . . , g
p
bep
ex ursions ofw
verifyingf
p
≺ g
p
≺ · · · ≺
f
1
≺ g
1
≺ e
m+1
. Suppose that•
the ex ursione
m+1
is negative, and allt−
ex ursions higher thane
m+1
are positive;•
theex ursionsg
1
, . . . , g
p
arenegative; andeverynegative ex ursion an-terior tog
p
is smaller thang
q
.We all the
g
i
's the plug-ex ursions, and thee
j
's the prote ted ex ursions. Thene
1
, . . . , e
m
are tall, and|i
w
e
1
| < |i
w
e
2
| < · · · < |i
w
e
m
| < η
.Proof Firstly,
|i
w
f
1
| < η
be ausef
1
≺ e
m+1
and any ex ursion anterior toe
m+1
and havingheight≥ η
is one of thef
q
, hen e positive.Se ondly, for
1 ≤ q ≤ p
, the ex ursiong
q
is negative and higher than any negative ex ursion, anterior to it; soI
w
is not onstant duringg
q
, and on-sequently we have(16)
|i
w
f
p
| < |i
w
f
p−1
| < · · · < |i
w
f
1
| < height of g
1
,
where ea h<
sign is due toI
w
varying on the orrespondingg
q
.thirdly, ombining (20) with
|i
w
f
1
| < η
(rst step), and noti ing that, by denition of thef
q
,(e
1
, . . . , e
m
)
is asub-sequen e of(f
p
, . . . , f
1
)
, weobtain (17)|i
w
e
1
| < · · · < |i
w
e
m
| < η.
Last, it remains to establish that
e
l
is tall for1 ≤ l ≤ m
. Lete
′
denote a positiveex ursion of
w
with heighth
′
≥ h
l
and su h thati
w
e
′
= i
w
e
l
. From (13), we have|i
w
e
′
| = |i
w
e
l
| < height of g
1
; soe
′
is anterior tog
1
and a fortiori anterior toe
m+1
. Ash
′
≥ h
l
,e
′
must be one of the
f
q
(see their denition). Bute
l
isalsoone ofthef
q
and, dueto(13), alli
w
f
q
aredierent; soe
′
= e
l
. This means thate
l
is tall.In the proof ofLemma6, thenegativeex ursions
g
q
areused toseparate thef
q
fromea hother. Yet, inthe end,we are not interested inthe behavior of allf
q
but onlyin thee
l
. It is possible to repla e this lemma with a variant, where2m
ex ursions (instead ofp
ones, theg
q
) are made negative, ea he
l
being anked by two of them.Lemma 7. Let
X
be a pro ess with law absolutely ontinuous w.r.t.π
, andE
a tall ex ursion ofT
◦ X
with heightH
. There exists an ex ursion ofX
, with interval{s; (I ◦ T ◦ X)(s) = i
T◦X
E}
, and with heightH + |i
T◦X
E|
.Proof First,re all a.s., Brownianmotion
B
doesnot rea h its urrent min-imumI
◦ B
in the interior of a time-interval whereI
◦ B
is onstant. (Thisis a onsequen e of (
I
◦ B)(s) < 0
fors > 0
and of the Markov property at the rst time thatB = I ◦ B
after some rational).Put
Y = T ◦ X
and allF
the interval{s ≥ 0; (I ◦ Y )(s) = i
X
E}
;Y
rea hes its urrent minimumI
◦ Y
at both endpoints ofF
but not in the interior ofF
(see above). Sin e|X| = Y − I ◦ Y
by Lemma 1, we have thatF
is the supportof some ex ursion ofX
. The height of that ex ursionismax(|X
s
|; s ≥ 0, and (I ◦ Y )(s) = i
Y
E)
=
max (Y
s
− (I ◦ Y )(s); s ≥ 0, (I ◦ Y )(s) = i
Y
E)
=
max (Y
s
; s ≥ 0, (I ◦ Y )(s) = i
Y
E) − i
Y
E
=
max (Y
s
; s ∈ E) − i
Y
E
because E is tall
=
H + |i
Y
E|.
Lemma 8. Let
(0,
−
→
i ,
−
→
j )
be an orthonormal basis of the plan in whi h we represent paths. Letτ
a+
b
be the verti al translation of ve tor(b − a)
−
→
j
andτ
b
a−
the ree tion along the horizontal axis of equation :y =
a + b
2
. Consider(t, k, p) ∈ R
+
∗
× N
2
su h thatw
k
t
= a
andw
k+p
t
= b
and denoteγ
t
the rst time posterior tot
when at least one of the iterated Lévy transformsw
s
,
k ≤ s ≤ k + p − 1
, vanishes. Then we have :w
k+p
|[t,γ
t
]
=
τ
b
a+
o w
k
|[t,γ
t
]
ifk+p−1
Π
i=k
w
i
t
> 0
τ
b
a−
o w
−k
|[t,γ
t
]
else We will denoteτ
k
k+p
(w)
the plan transformation, whi h transformsw
k
|[t,γ
t
]
inw
k+p
|[t,γ
t
]
.Proof It is animmediate onsequen e of Tanaka'sLemma, when
p = 1
. In general ase, we break up the displa ementτ
whi h transformsw
k
|
[t,γ
t
]
inw
k+p
|
[t,γ
t
]
under the form
τ = τ
p
◦ τ
p−1
◦ ... ◦ τ
1
whereτ
i
transformsw
k+i−1
|
[t,γ
t
]
inw
k+i
|
[t,γ
t
]
. From the pre eding remark, ea h
τ
i
is a verti al translation or a ree tion alongan horizontalaxis, a ordingto thesign ofw
k+i−1
t
. Thenwe dedu e the laim.To onstru t the desired pro ess
Γ
, we will pro eed by indu tion on dis- retized time, andsowe willperform,fromthe levelr
,twotypesofraisings. In a rst type, the so- alled horizontal raisings, at a level when the path vanishes on[t
d
, t
d+1
]
, we prote t the material furnished by the indu tion hy-pothesis on[0, t
d
]
, namelyΓ
e
. And we prepare the path on[t
d
, t
d+1
]
to give it the formit ought to have at this level for being near toϕ
on the interval whenΓ
is near toϕ
on[0, t
d
]
.So we make positive the signi ative ex ursions of the path alled here the prote ted ex ursions, and insert between them small ex ursions alled the plug-ex ursions (see lemma 6). And on
[t
d
, t
d+1
]
, we prepare ex ursions , the building ones, whi h we prote t, and they will a t, one by one, during a su ession of horizontalraisings, to give the path the previewed form (see Lemma 13 and Proposition 14), at the ondition the path, after that, will not vanishedon[t
d
, t
d+1
]
.Inase ondtype,theso alledverti alraisings,wegiveanewtotheprote ted ex ursions the signs they have before the horizontal raisings. Then we get up while the path doesn't vanish on
[t
d
, t
d+1
]
. In fa t we must distinguish the last raising of a su ession of horizontal raisings, the so alled terminal horizontal raising, when we leave to prote t the prote ted ex ursions and givethem the goodsigns.It is important to know the real level of the path, ie. the level without the horizontal raisings. Pre isely,wedene the r.v. indu tively:
RL
r
(w) = r
and f or all integer n ≤ r
RL
n−1
(w) =
(
RL
n
(w)
if thestepn − 1 → n
orresponds to an horizontal raisingRL
n
(w) − 1
if it orresponds to a verti al raising.Forourneeds, wewill allmap-ex ursion,orsimplyex ursion,ea hmap
e : R
+
→ R
whosesupportisanot empty segmentand whi hdoesn't vanish at any point of the interior of the support. In parti ular, for
w ∈ W
andt > 0
, we will all ex ursion straddlingt
, and denote itby :e
t
(w)
, the map so dened :e
t
(w) : R
+
→ R, ∀u ∈ R
+
, e
t
(w)(u) =
0
ifu ∈ [0, g
t
(w)] ∪ [d
t
(w), +∞[
w
u
elseWewill introdu e the map
de
t
(w) : R
+
→ R
∀u ∈ R
+
, de
t
(w)(u) =
0
ifu ∈ [0, t] ∪ [R
t
(w), +∞[
where
R
t
(w) = sup{u > t, ∀s ∈]t, u], w
s
6= w
t
}
w
u
− w
t
elseand we allitadierentialex ursionof
w
wheneveritssupportisnotempty, ie.∃ε > 0, ∀u ∈ (t, t + ε), w
u
6= w
ε
We denote them
e
t
andde
t
when there is no ambiguity, ande
s
t
andde
s
t
in the ase of ex ursions ofw
s
.
Lemma 9. Let
w ∈ W
ande
a negativem−
ex ursionofT
w
, lowerthanall pre eding it. Letγ
be its beginningandδ
its end. We set :γ
1
= arg min Tw
[0,γ]
, γ
2
= inf{t ∈ supp(e); e(t) = Tw
γ
1
}, γ
3
= arg min e
[γ,δ]
Then :
de
γ
1
oin ides with an ex ursion of|w|,
and its support is[γ
1
, γ
2
]
de
γ
3
oin ides with an ex ursion of|w|,
whi h begins atγ
3
and whose support ontains[γ
3
, δ]
Furthermore,
∀u ∈ [γ
2
, γ
3
]
,de
u
oin ides with an ex ursion of|w|
, if, and only if :de
u
is a positive ex ursione(u) = inf{e(t), t ∈ [γ, u]}
Itisthe ase inparti ular when
de
u
istherst positiveex ursionof theformde
v
,v ∈ [γ
2
, γ
3
]
to overow a given value.Proof
From Tanaka's formula:
|w
t
| = Tw
t
+ sup{−Tw
u
, u ∈ [0, t]}
Therefore,
|w
γ
1
| = Tw
γ
1
− Tw
γ
1
= 0,
while, for all
t > γ
1
, su iently small:T
w
t
> Tw
γ
1
.
So,
de
γ
of
|w|
beginning atγ
3
whose support ontains[γ
3
, δ]
. Lete
′
be an ex ursion of
w
with support in luded in[γ
2
, γ
3
]
. Its beginningu
, and its endv
verify:u = arg
[0,v[
min Tw
andv = arg
[0,v]
min Tw.
So wededu e :
de
u
= |e
′
|
.
Re ipro ally,let
u ∈ [γ
2
, γ
3
]
su h thatde
u
is a positive ex ursionandu = arg
[0,u]
min Tw
.Then,
u = arg
[0,v[
min Tw
,wherev
is the end ofde
u
, be ausede
u
is positive.Thus,
w
u
= w
v
= 0
, and for allt ∈]u, v[
,w
t
6= 0
. Consequently,de
u
isan ex ursionof|w|
.Let
h > 0
be su h that there existsu ∈ [γ
2
, γ
3
]
verifyingde
u
is the rst positive ex ursion of the formde
v
,v ∈ [γ
2
, γ
3
]
, whose height overowsh
. Then, for allv < u
, the support ofde
v
an't ontain this ofde
u
without denying the minimality ofu
.As an immediate onsequen e,we observe :
Corollary 2. The ex ursions of
w
oin ide with the positive dierential ex- ursions ofT
w
, beginning atarg min
[0,t]
(Tw)
for allt ∈ [0, +∞[
.In thisparagraph,wewanttheraisedpathtoapproa hthemap
ϕ
uniformly on[0, T ]
. Pre isely :Whatever
ε
stri tlypositive, andϕ ∈ W
|
[0,T ]
, thereexistsΓ
aB−
raised Brow-nian motion su h that :P
kΓ
|
[0,T ]
− ϕ
|
[0,T ]
k
∞
< ε
> 1 − ε
We onsider a modulus of uniform ontinuity
α
0
asso iatedtoε
4
, ϕ, [0, T ]
and a real number
α
1
su h thatP (A
0 ε
) > 1 −
ε
2
whereA
0 ε
=
h
sup{|B
t
− B
u
|, (t, u) ∈ [0, T ]
2
and|t − u| < α
1
} <
ε
2
i
,
then weset
α := min(α
0
, α
1
)
,d
0
:=
T
α
+ 1
,and foralld ∈ N
,t
d
= (dα) ∧ T
. Weset again, for allintegerd ∈ [1, ..., d
0
]
,A
d
ε
:=
h
sup{|B
t
− B
u
|, (t, u) ∈ [t
d
, T ]
2
, |t − u| < α} <
ε
2
i
.
Our aimis to show, by indu tion on
d
,the following propertyP
d
: "For allε > 0
,thereexists anintegerr
d
andΓ
aB−
raisedBrownianmotionof indexr
d
su h that :P
h
kΓ
|
[0,td]
− ϕ
|
[0,td]
k
∞
< ε
i
∩ [|Γ
t
d
− ϕ(t
d
)| < ε
1
] ∩ A
d
ε
> 1 − ε
1 +
d
d
0
”
Noti e that
P
0
immediately yields from the hoi e ofϕ
whi h vanishes at0
. We suppose nowP
d
true. We are going to apply this hypothesis to the BrownianmotionB
s
0
, foraninteger
s
0
whi h,asthe realnumberε
1
,willbe later spe ied. AsA
d
ε
⊂
sup{|B
t
− B
u
|, (t, u) ∈ [t
d
, t
d+1
]
2
} <
2
ε
∩ A
d+1
ε
,and fromthe inde-penden eofthein rementsofBrownianmotion,we andedu etheexisten e of adisjointedsumΓ
e
ofB
s
0
−
raised Brownian motionsof index
r
d
su h that :P
(A
ε
0
) > 1 − ε
1 +
d
d
0
,
whereA
ε
0
:=
h
ke
Γ
|
[0,td]
− ϕ
|
[0,td]
k
∞
<
ε
2
i
∩
h
|e
Γ
t
d
− ϕ(t
d
)| < ε
1
i
...
...
∩
h
sup{|B
s
0
+r
− B
s
+r
|, (t, u) ∈ [t
, t
]
2
} <
ε
i
∩ A
d+1
(It su es toapply
P
d
withε
2
instead ofε
).We willdenote :
∀i ∈ N, e
Γ
i
= e
w
i
. By denition,
∀i > r
d
, ˜
w
i
= w
s
0
+i
. From the theorem of density of zeroes ([M℄), there exists a.s. an integerℓ
su h thatw
e
ℓ
vanishes atleast one time on
[t
d
, t
d+1
]
.Let
L(w)
be the smallest of these integersℓ
.L
is a r.v. almost surely nite. Then there exists anintegerℓ
0
whi hwe will hoose> r
d
su hthat :P
(A
ε
1
) > 1 − ε
1 +
d
d
0
−
ε
2d
0
oùA
ε
1
:= A
ε
0
∩ [L ≤ ℓ
0
] .
Weset :r = s
0
+ max(r
d
, l
0
)
.Denition 3. Theprote ting tree of
e
Γ
. LetN
be a set of ex ursionsofw
|[0,d
t
(w)]
,w ∈ W
, andε > 0
. We denote byT
t,ε
w
(N )
the set of ex ursionsofT
w
|[0,d
t
(Tw)]
so dened :e
′
∈ T
t,ε
w
(N )
ie
′
is an ex ursion of
T
w
|[0,d
t
(Tw)]
whi h appears in the dif-ferential ex ursionsdE
ofT
w
orresponding to an ex ursionE ∈ N
, andh(e
′
) >
ε
4
(we all them ex ursions of the rst type) or the support ofe
′
ontains
arg max
of the dierential ex ursiondE
, we all them ex ursions of the se ond type, and all the ex ursionse
′′
of
T
w
of height belonging to[h(e
′
), h(e
′
−
)]
,e
′
−
being the pre eding ex ursion ofT
w
of height>
ε
4
, withe
′
−
< e
′′
< e
′
, we allthese ex ursionse
′′
the third type ex ursions.
Lemma 10. When
N
is a nite set, so isT
t,ε
w
(N )
.Proof We remark rstly that the number of distin t dierential ex ur-sionsof
T
w
orrespondingtotheelementsofN
isnite,equaltothe ardinal ofN
. Thenumberofex ursionsofthersttypeisnitebe ausetheirheight isgreaterthanε
4
. Thenumberofex ursionsofthese ondtypeisnite follow-ingthe rst remark. And forea h ex ursionof the se ond type,the number of ex ursions of the third typeis nite too.Thenwe all prote tingtree of
e
Γ
,the tree onstituted by :•
at the0
-generation : the elements ofN
e
0
, the set of
d
t
( e
w
0
)
-ex ursions of height
> ε
.•
and, for alln ∈ N
,if we denotebyN
e
n
, the set of prote tedex ursions of
w
e
n
whose elements, given in hronologi al order, onstitute the
n
generation of the tree, then at the
(n + 1)
th
generation, the elements of
N
e
n+1
:= T
t
d
,ε/4
n
e
w
n
e
N
n
. As we have ordered ea h
N
e
n
, whi h is now a nite sequen e of ex ursions of
w
e
n
, we an onsider
Σ
e
n
the nite sequen e of their signs. And our next task is to denite
N
′n
the sequen e of prote ted
t
d
-ex ursions ofΓ
n
and
Σ
′n
the asso iated sequen e of their signs. We have dened the
N
e
n
's by getting down in the iterations. Wewilldenethe
N
′n
's by gettingup fromlevel
r
. We putN
′r
(ω) = e
N
r−s
0
(ω)
, andΣ
′r
(ω) = e
Σ
r−s
0
(ω)
. IfΓ
n+1
−→ Γ
n
is an horizontal raising, then
N
0,n+1
is onstituted by the familyofthe ex ursions of
Γ
n
whoseabsolutevalueshavethe same
arg max
asthe ex ursions ofN
0,n+2
. Then we pro eed as in lemma6, onsidering the elements of
N
0,n+1
asthe
e
i
,1 ≤ i ≤ m
. We havenow todenethe r.v.η
: letη
i
betheminimaldistan e between theheights of distin t prote ted ex ursions ofw
e
i
for
i ∈ {0, . . . , r − s
0
}
andη
m
=
r−s
V
0
i=0
η
i
, e
ε
i
= ε ∧ η
m
4
i+1
1
, fori ∈ {0, . . . , r − s
0
}
, andε
i
= e
ε
RL(i)−s
0
1
4
i′ −i
, wherei
′
= sup{j > i/RL(j) = RL(i)}
. Then we hoose for
η
the r.v.ε
RL(n)−s
0
. And we ( an) dene the setN
′n
as the set of elements the
f
j
's and theg
j
's. AndΣ
′n
give sign
−1
to theg
j
's and+1
tothef
j
's ifΓ
n+1
−→ Γ
n
is not a terminalhorizontalraising. In the ase of aterminalhorizontalraising,weput simply
N
′n+1
= N
0,n+1
. IfΓ
n+1
−→ Γ
n
isa verti al raising,N
n+1
is onstituted by the family of the ex ursions of
Γ
n
orrespondingtothe positivedierentialex ursions of
Γ
n+1
whose support en ounters at least the supportof an element of
N
0,n+2
and beginning at an
arg min
ofΓ
n+1
. In these two ases, to spe ify
Σ
′n+1
, we need the following lemma. Let us all argext of an ex ursion the
arg max
(resp.arg min
) of this ex ursion if itis positive (resp. negative).Lemma 11. If for all
i
from levelr
ton
(withr ≥ n
) the elements ofN
0,i
and
N
e
RL(i)−s
0
have the same argext, whi h implies
|N
0,i
| =
e
N
RL(i)−s
0
, andthe same signs, then the elements of
N
0,n−1
and
N
e
RL(n−1)−s
0
have the same argext, hen e these sets have the same ardinality, and the same hierar hy, ie. the order of the heights between the respe tive ex ursions of ea h set is the same. So we put
Σ
0,n−1
= e
Σ
RL(n−1)−s
0
.Proof The distin tion between
Γ
andΓ
e
is due to the introdu tion of horizontal raises. IfΓ
n+1
−→ Γ
n
is an horizontal raising, ea h prote ted ex ursion gains in height the height of a plug ex ursion, and its support enlarges. So, duringsu haraising,the"error"between
Γ
andΓ
e
in reasedofon the other side,
Γ
n+1
−→ Γ
n
is a verti al raising, a prote ted ex ursion of
Γ
n
is onstituted by atmost two prote ted ex ursions of
Γ
n+1
. Soduring su h a raise, the error between
Γ
andΓ
e
doubles. And it is easy, by means of our hoi eof the heightsof the plug ex ursions, toshowthat this erroris always majorized byε ∧ η
. Hen e the lemmafollows immediately.Now we an onsider (see Proposition 2) that
Γ
|[0,t
d
]
is orre tly dened : for on eRL
w
(n) = s
0
, we end the raises, if ne essary, by horizontal ones : prote ting allthe material a quainted. But isΓ
near frome
Γ
? To answer this question, we have to onsider the error betweenΓ
ande
Γ
now, as beingkΓ
|[0,t
d
]
− e
Γ
|[0,t
d
]
k
∞
.We have just built
Γ
|[0,t
d
]
. We have now to ontrol :kΓ
N
|[0,t
d
]
− e
Γ
0
|[0,t
d
]
k
∞
, where
N = sup {n ∈ {0, . . . , r}/RL(n) − s
0
= 0}
(withN = −∞
if the set is empty).N
is ar.v. Lemma 12.kΓ
N
|[0,t
d
]
− e
Γ
0
|[0,t
d
]
k
∞
≤ 2ε
, on the event[N ≥ 0]
.Proof Letusrst remark,fromthe pre edinglemma,that, onthe event
[N ≥ 0]
,wehave prote tedex ursions with thesame argext, thesame signs, and heights near atε
. The supports of the prote ted ex ursions ofΓ
N
on-taining the supports of the orresponding prote ted ex ursions of
e
Γ
0
. F ur-thermore, onthe dieren eof their supports,
Γ
N
and
Γ
e
0
dierfrom atmost
2ε
, and likelyoutside the union of their supports.The purpose of the following lemma is to prepare, at level
s
, when the iteratedBrownianmotionvanishes on]t
d
, t
d+1
[
,the ex ursions whi hwill al-lowthe orre tly raised path toapproa hϕ
at level0
on]t
d
, t
d+1
[
.Lemma 13. Full planing. Let
w
belongtoW
, andt, t
′
, ε
′
∈ R
+
∗
be su hthatt < t
′
. We suppose there is no interval in whi h
w
is onstant, andw
vanishes in(t, t
′
)
. The following r.v. are fun tionalsof
|w|
:• t
0
(w) = g
inf{s>d
t
;|w
s
|≥ε
′
}
∧ t
′
, withinf ∅ = +∞
.• ∀n ∈ N
, whilet
n
< t
′
, we set :t
n+1
:=
inf{u ∈ [t
n
; arg max |de
t
n
|[, h(de
u
) > ε
′
and sgn
(de
u
) = −
sgn(de
t
n
)}
if this set is not empty,else :
arg max |de
t
n
| ∧ t
The sequen e
(t
n
)
n∈N
is stri tly in reasing and nite. Let1 + K(w)
be its ardinality.Proof By onstru tion, the sequen e
(t
n
)
is stri tly in reasing and lower thant
′
. Suppose the number of its terms is innite. In this ase, it would admit a limit
t
∗
≤ t
′
, and the os illation of
w
att
∗
would be innite, so ontradi ting the ontinuity ofw
. Then(t
n
)
n
isnite.The measurability and the niteness of
K
are immediate.Let us remark that this Lemma gives us the possibility of planing the path after
d
t
inK
raises.For, duringthe rst raise, we put negative the ex ursion beginningat
t
0
(w)
and positive all the other ex ursions in(t
0
(w), t
′
)
of height greater than
ε
′
. Then during the se ond raise, we put negative the ex ursion whose support ontains this of
de
t
1
, and so on. At the end of su hK
raises, the path on[t
0
, t
′
]
has an absolute value whi h doesn't ex eedε
′
+ Kε
′′
, andε
′
on the ex ursion straddlingt
′
.Sowe are going now to analyze its behavior on
(t
d
, t
d+1
)
. Letus denoteγ
t
d
the rst time aftert
d
at whi h one of theΓ
σ
,
0 ≤ σ ≤ s
0
+ l
0
, vanishes on(t
d
, t
d+1
)
, andσ
0
the orresponding level.Proposition 4. There exists a
σ(Γ
σ
′
0
+1
)−
measurable,
N
−
valued r.v.,K
′
σ
0
su h that there exists
K
′
σ
0
− 1
r.v.P
1
, . . . , P
K
′
σ0
−1
themselves with values in
N
andσ(Γ
σ
0
+1
)−
measurable, su h that : (i) the
K
′
σ
0
− 1
ex ursion intervalse
P
1
(Γ
σ
0
), . . . , e
P
K
′
σ0 −1
(Γ
σ
0
)
are disjoint and in luded in(d
t
d
(Γ
σ
0
), t
0
(Γ
σ
0
))
(ii) the heights
H
1
, . . . , H
K
′
−1
of theseK
′
− 1
ex ursions ofΓ
σ
0
satisfy on[σ
0
≥ 0]
:τ
RL
0
σ0
−s
0
)
( e
w)(ϕ(t
d+1
))
−ε
′
< H
1
+· · ·+H
K
′
σ0
−1
<
τ
RL
0
σ0
−s
0
)
( e
w)(ϕ(t
d+1
))
+ε
′
.
Proof Noti ing that the pro ess :
X =
X
n∈N
Γ
n
1
[σ
0
=n]
onse-
Ofteninthesequel,wewilldenote
K
′
σ
0
simplybyK',ifthereisnoambiguity. Set :σ
0
′
= σ
0
− K(Γ
σ−1
) + K
σ
′
0
.
Proposition 5. For all
(n, ω)
su h thatσ
0
(ω) ≤ n < σ
0
(ω) − K
σ
0
(ω) −
K
σ
′
0
(ω)
,N
n
(ω)
andΣ
n
(ω)
an be hosen so that : (i)0 ≤ Γ
σ
0
−K
t
d+1
< ε
′
andke
Γ
RL(σ
0
−1)−s
0
|[0,t
d
]
− Γ
σ
0
|[0,t
d
]
k
∞
< Kε
′′
. (ii)−(I ◦ Γ
σ
0
)(t
d+1
) = H
K
′
σ0
−1
(iii)Γ
σ
0
hasK
′
− 2
tall ex ursions in luded in
(t
d
, t
d+1
)
:E
1
< E
2
<
· · · < E
K
′
−2
with respe tive heightsH
1
, . . . , H
K
′
−2
verifying|i
Γ
σ0
E
1
| <
|i
Γ
σ0
E
2
| < · · · < |i
Γ
σ0
E
K
′
−2
|
. (iv)H
n+1
+ · · · + H
K
′
σ0
−1
≤ Γ
σ
0
−K−n
t
d+1
< H
n+1
+ · · · + H
K
σ0
′
+ ε
′
+ nK
′
ε
′′
. (v)H
K
′
σ0
−n
< −(I ◦ Γ
σ
0
−K−n
)(t
d+1
) < H
K
′
σ0
−n
+ K
′
ε
′′
. (vi)Γ
σ
0
−K
σ0
−n
hasK
′
σ
0
− n − 2
tall ex ursions in luded in(t
d
, t
d+1
)E
n
1
<
· · · < E
n
K
′
σ0
−n−2
su hthat:|i
Γ
σ0−Kσ0 −n
E
1
| < · · · <
i
Γ
σ0−Kσ0 −n
E
K′σ0−n−2
<
ε
′′
, and whose heights
H
n
1
, . . . , H
K
n
′
J0
−n−2
satisfy :
H
l
≤ H
l
n
< H
l
+ nε
′′
f or n + 1 ≤ l < K
σ
′
0
− 1.
In our pursuitof the pro edure, we an state : Proposition6. Itispossibleto hoose
N
n
(ω)
and
Σ
n
(ω)
, forall
(n, ω)
su h thatn = σ
′
0
, in order to have :Γ
σ
′
0
t
d+1
− τ
0
RL(σ
′
0
)−s
0
(ϕ(t
d+1
))
< ε
′
+
1
2
(K
′
σ
0
− 1)(K
′
σ
0
− 2)ε
′′
Proof Wenoti ethatthebuildingex ursionswhi happearinProposition5, the
e
P
0
(Γ
σ
0
+K
σ0
)
's, are su essively prote ted. On eprote ted, ea hof them re eivesasmallex ursionofheightlowerthan
ε
′′
atea hraise. Sowededu e the result.