HAL Id: hal-01258190
https://hal.archives-ouvertes.fr/hal-01258190
Submitted on 18 Jan 2016
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Statistical properties of intermittent maps with
unbounded derivative
Giampaolo Cristadoro, Nicolai Haydn, Philippe Marie, Sandro Vaienti
To cite this version:
Giampaolo Cristadoro, Nicolai Haydn, Philippe Marie, Sandro Vaienti. Statistical properties of
inter-mittent maps with unbounded derivative. Nonlinearity, IOP Publishing, 2010, 23 (5), pp.1071-1096.
�hal-01258190�
arXiv:0812.0555v2 [math.DS] 16 Dec 2008
derivative
Giampaolo Cristadoro, Ni olai Haydn, Philippe Marie, Sandro Vaienti
De ember 16, 2008
Abstra t
We study the ergodi and statisti al properties of a lass of maps of the ir le and of theinterval ofLorenz typewhi h present indierent xedpointsand pointswithunbounded derivative. Thesemapshavebeen previously investigatedinthephysi s literature. We prove inparti ular that orrelationsde aypolynomially,andthatsuitableLimitTheorems ( onver-gen eto StableLawsor Central LimitTheorem) holdfor Hölder ontinuous observables. We moreovershowthatthe returnand hittingtimes areinthelimit exponentiallydistributed.
1 Introdu tion
TheprototypeforintermittentmapsoftheintervalisthewellknownPomeau-Mannevillemap
T
dened on the unit interval[0, 1]
and whi h admits a neutral xed point at0
with lo al behaviorT (x) = x + cx
1+α
;otherwise itis uniformly expanding. The onstant
α
belongs to(0, 1)
toguaranteetheexisten eofaniteabsolutely ontinuousinvariantprobabilitymeasure and the onstantc
ould be hosen in su h a way that the mapT
has a Markov stru ture. This map enjoy polynomial de ay of orrelations and this property still persists even ifthe mapisnot anymoreMarkov [30 ℄.Another interesting lass of maps of the interval are the one-dimensional uniformly ex-panding Lorenz-like maps (see [15 , 29, 11℄ for their introdu tion and for the study of their topologi alproperties),whose featuresarenowthepresen eofpointswithunboundedderiv a-tivesandthela kofMarkovstru ture: inthis aseone ouldbuilduptowersandndvarious rates for the de ay of orrelations depending on the tail of the return time fun tion on the base of the tower, see, for instan e [7℄ and [8 ℄. The latter paper deals in parti ular with one-dimensional maps whi h admit riti al points and, eventually, points with unbounded derivatives,but it leavesopenthe asewhere there ispresen eof neutral xedpoints.
In this paper we are interested in maps whi h exhibit the last two behaviors, namely neutralxedpointsandpointswithunbounded derivatives. Su h mapshavebeenintrodu ed
a polynomial de ayof orrelations and they also studied other statisti al properties like the sus eptibility andthe
1/f
-noise. Another ontributionbyA.Pikovsky [25 ℄showed, stillwith heuristi arguments, that these maps produ e anomalous diusionwith squaredispla ement growing faster than linearly. R. Artuso and G. Cristadoro [3℄ improved the latter result by omputingthemomentsofthedispla ementontheinniterepli asofthefundamentaldomain andshowed aphasetransitionintheexponent ofthemomentsgrowth. Re entlyLorenz usp maps arose to des ribe the distribution of the Casimir maximum in theKolmogorov-Lorenz model of geouid dynami s [24 ℄. Despite this interesting physi al phenomenology, we did not nd any rigorous mathemati al investigation of su h maps. These maps are dened on the torusT
= [
−1, 1]/ ∼
and depend on the parameterγ
(see below); whenγ = 2
the orresponding map was taken as an example of the non-summability of therst hyperboli timeby Alves andAraujoin[2 ℄. This mapsreads:˜
T (x) =
(
2
√
x
− 1
ifx
≥ 0
1
− 2
p
|x|
otherwise (1)and it was proved in [2℄ that it is topologi al mixing, but no other ergodi properties were studied.
A tually, the Grossmann and Horner maps are slightly dierent from those investigated in[25 ℄and [3 ℄,thedieren e beingsubstantiallyinthefa tthatthelatter aredened onthe ir le instead than on the unit interval. We will study in detail the ir le version of these maps in Se tions 2 to 5, and we will show in Se tion 6 how to generalize our results to the interval version: sin e both lasses of maps are Markov, the most important information, espe iallyin omputingdistortion,will omefromthelo albehavioraroundtheneutralxed points and the points with unbounded derivatives and these behaviors will be the same for both versions. There isneverthelessaninterestingdieren e. The ir leversionintrodu edin Se tion1iswritten insu hawaythattheLebesguemeasureisautomati allyinvariant. This is not the ase ingeneral for the intervalversion quoted in Se tion 6. However the strategy thatwe adopttoprove statisti alproperties (Lai-SangYoungtowers) will give us aswellthe existen eof anabsolutely ontinuous invariant measureand wewill ompleteit byproviding informations on the behavior of the density. It is interesting to observe that in the lass of maps onsidered by Grossmann and Horner on the interval
[
−1, 1]
(see Se t. 6), the analog of(1 ) isgiven bythefollowing map:˜
S(x) = 1
− 2
p
|x| .
(2)This mapwas investigated byHemmer in 1984 [19 ℄: he also omputed by inspe tion the in-variant density whi h is
ρ(x) =
1
2
(1
− x)
and theLyapunov exponent (simply equal to1/2
), butheonlyarguedabout aslowde ayof orrelations. WewillshowinSe t. 6howtore over thequalitative behavior of this density(and of all the others in the Grossmann and Horner lass).C
1
on
T
/
{0}
,C
2
on
T
/(
{0} ∪ {1})
and areimpli itly dened onthe ir le bytheequations:x =
1
2γ
(1 + T (x))
γ
if0
≤ x ≤
1
2γ
T (x) +
1
2γ
(1
− T (x))
γ
if1
2γ
≤ x ≤ 1
and for negative values of
x
by puttingT (
−x) = −T (x)
. We assume that parameterγ > 1
. Note that whenγ = 1
the map is ontinuous with onstant derivative equal to2
and is the lassi al doubling map. The point 1 is a xed point with derivative equal to1
, while at 0 the derivative be omes innite. The map leaves the Lebesgue measurem
invariant (it is straightforward to he k that the Perron-Frobenius operator has1
as a xed point). We will prove inthe nextse tions theusual bun h of statisti alproperties: de ayof orrelations (whi h, due to the paraboli xed point, turns out to be polynomial with the rate found in [14℄); onvergen e to Stable Laws and large deviations; statisti sof re urren e. All these results will follow from existing te hniques, espe ially towers, ombined with the distortion bound proved in the next se tion. Distortion will in fa t allows us to indu e with the rst return map on ea h ylinder of a ountable Markov partition asso iated toT
. A tually one ould indu e on a suitable interval only ( alledI
0
in the following): the proof we give is intentedto providedisortion onall ylindersof the ountable Markovpartition overing mod0
thewhole spa e[
−1, 1]
, sin e this is ne essary inorder to apply theindu ing te hnique of [4 ℄ whi h will give us thestatisti alfeaturesof re urren estudied inSe t. 5: distributions of rst return and hitting times, Poissonian statisti s for the number of visits, extreme values laws.2 Distortion
Notations: With
an
≈ bn
we mean that there exists a onstantC
≥ 1
su h thatC
−1
bn
≤
a
n
≤ Cbn
for alln
≥ 1
; witha
n
.
b
n
we mean thatthere existsa onstantC
≥ 1
su h that∀n ≥ 1
,a
n
≤ Cbn
;witha
n
∼ bn
we mean thatlim
n→∞
a
n
b
n
= 1
. We will also usethe symbol "
O
" intheusual sense. Finallywe denote with|A|
thediameter of thesetA
.There is a ountable Markov partition
{Im}m∈Z
asso iated to this map; the partition is builtmod m
as follows:Im
= (am−1, am)
for allm
∈ Z
∗
and
I0
= (a0−, a0+)/
{0}
, where, denoting withT
+
= T
|(0,1)
and withT
−
= T
|(−1,0)
:a
0+
=
1
2γ
, a
0−
=
−
1
2γ
anda
i
= T
−i
+
a
0+
, a
−i
= T
−
−i
a
0−
,
i
≥ 1 .
Thenwedene
∀i ≥ 1
:b
−i
= T
−
−1
a
i−1
andb
i
= T
−1
+
a
−(i−1)
.
We now state without proof a few results whi h are dire t onsequen es of thedenition ofthe map.
-1
0
x
1
-1
0
1
T(x)
a
-2
a
-1
a
0-
b
2
b
1
b
-1
b
-2
a
0+
a
1
a
2
Lemma 1. 1. Whenx
→ 1
−
:T (x) = 1
− (1 − x) −
1
2γ
(1
− x)
γ
+
O ((1 − x)
γ
)
2. Whenx
→ 0
+
:T (x) =
−1 + (2γ)
1
γ
x
1
γ
.
Lemma 2. We have for all
n
≥ 0
,a
±(n+1)
= a±n
+
1
2γ
(1
− a±n)
γ
and:a
n
∼
1
−
2γ
γ
− 1
1
γ−1
1
n
γ−1
1
a−n
∼ −1 +
2γ
γ
− 1
1
γ−1
1
n
γ−1
1
l
n
:= length[a
n−1
, a
n
]
∼
1
2γ
2γ
γ
− 1
γ
γ−1
1
n
γ−1
γ
n > 1
|b±(n+1)| ∼
2γ
1
2γ
γ
− 1
γ
γ−1
1
n
γ−1
γ
,
n > 1 .
We now indu e on the interval
I
m
:= (a
−m
, a
m
)/
{0}
and provide a bounded distortion estimatefor the rstreturn map. WedeneZ
m,p
= Z
+
m,p
∪ Z
m,p
−
,where:Z
+
m,1
:= (b
m+1
, a
m
)
,Z
m,1
−
:= (a
−m
, b
−(m+1)
)
andZ
+
m,p>1
:= (b
m+p
, b
m+p−1
)
,Z
−
m,p>1
:= (b
−(m+p−1)
, b
−(m+p)
)
. Notethat
I
m
=
∪p≥1
Z
m,p
andthattherstreturnmapT = I
b
m
→ Im
a tsonea hZ
m,p
asT = T
b
p
and inparti ular:T
p
(Z
m,p
+
) =
(
(a
−m
, a
m−1
)
p = 1
(a−m, a
−(m−1)
)
p > 1
T
p
(Z
m,p
−
) =
(
(a
−(m−1)
, a
m
)
p = 1
(am−1, am)
p > 1 .
We nallyobserve thattheindu ed map
T
b
isuniformlyexpandinginthesense thatfor ea hm
andp
there existsβ > 1
su h that|D b
T (x)
| > β
,∀x ∈ Im
. 1Proposition3(Boundeddistortion). Letusindu eon
I
m
;thenthereexistsa onstantK > 0
that depends on m, su h that for ea hm
andp
and for allx, y
∈ Zm,p
, we have:DT
p
(x)
DT
p
(y)
≤ e
K|T
p
(x)−T
p
(y)|
≤ e
2K
.
Remark 1. The ylinder
Z
m,p
isthe disjoint union of the two openintervalsZ
+
m,p
andZ
−
m,p
sitting on the opposite sides of
0
(see above). Wheneverx
andy
belongs to dierent om-ponents, we pro eed by rst noti ing thatDT
p
(x) = DT
p
(
−x)
and
−x
sits now in the same omponent asy
. By exploitingthe fa t thatT
p
is oddwe getDT
p
(x)
DT
p
(y)
=
DT
p
(
−x)
DT
p
(y)
≤ e
K|T
p
(−x)−T
p
(y)|
≤ e
K|−T
p
(x)−T
p
(y)|
≤ e
K|T
p
(x)−T
p
(y)|
.
andwe an thus on entrate on the ase when
x
andy
are takeninthe sameopen omponent (see below).Proof. We denote with
l
m
thelength of the interval(a
m−1
, a
m
)
(whenm = 0
,l
0
= length of(0, a
0+
)
). We start byobserving thatDT
p
(x)
DT
p
(y)
= exp
p−1
X
q=0
(log
|DT (T
q
x)
| − log |DT (T
q
y)
|)
= exp
p−1
X
q=0
D
2
T (ξ)
DT (ξ)
|T
q
x
− T
q
y
|
,
(3)where
ξ
is apoint betweenT
q
x
and
T
q
y
. Wedivide the ases
p = 1
andp > 1
.• p = 1
For
(x, y)
∈ Z
−
m,1
(seeRemark(1 )above), using|x − y| < |T (x) − T (y)|
,we dire tlyget:DT (x)
DT (y)
≤ exp [K
1|T (x) − T (y)|] ,
whereK
1
= sup
(Z
−
m,1
)
D
2
T = D
2
T (a
m
)
. 1Usingthe hainrulewe ansee that
β
≡ inf
x∈Z
m,1
|DT (x)| > 1
.• p > 1
Startwith
x, y
∈ Z
−
m,p
(seeRemark(1 )above),thenT x, T y
∈ (am+p−2, am+p−1)
;T
2
x, T
2
y
∈
(a
m+p−3
, a
m+p−2
)
;. . .
;T
p−1
x, T
p−1
y
∈ (am
, a
m+1
)
,we have:(
3)
≤ exp
sup
(Z
m,p
−
)
|D
2
T
|
inf
(Z
−
m,p
)
|DT |
|x − y| +
p−1
X
q=1
sup
(a
m+p−q−1
,a
m+p−q
)
|D
2
T
|
inf
(a
m
+p−q−1
,a
m
+p−q
)
|DT |
|T
q
x
− T
q
y
|
≤ exp
sup
(Z
−
m,p
)
|D
2
T
|
|x − y| +
p−1
X
q=1
sup
(a
m+p−q−1
,a
m+p−q
)
|D
2
T
|
|T
q
x
− T
q
y
|
.
(4)To ontinue we need thefollowing Lemma 4. For
x, y
∈ Z
−
m,p
(see Remark(1 ) above) we have:(i)
Pp−1
q=1
sup
(a
m+p−q−1
,a
m+p−q
)
|D
2
T
|
|T
q
x
− T
q
y
| ≤ C1|T
p−1
Z
|
(ii)sup
Z
−
m,p
|D
2
T
|
|x − y| ≤ C2
|T
p−1
Z|
l
m+1
,
where we set for onvenien e
Z
the intervalwithendpointsx
andy
. Proof. (i) DenoteT
p−1
x = z
x
andT
p−1
y = z
y
; sin e the derivative is de reasing on(0, m)
we have:|T
q
x
− T
q
y
| ≤
1
DT
p−1−q
(a
m+p−q
)
|zx
− zy| .
(5)
Let'snow onsidertheterm:
DT
p−1−q
(a
m+p−q
) = DT (a
m+p−q
)DT (T a
m+p−q
) . . . DT (T
p−2−q
a
m+p−q
) .
(6)Sin efor
q
≥ 1
andξ1
∈ (aq, aq+1)
:DT (a
q
)
≥ DT (ξ1
) =
T (aq+1)
− T (aq)
a
q+1
− aq
=
aq
− aq−1
a
q+1
− aq
itfollows that(
6)
≥
a
m+p−q
− am+p−q−1
a
m+p+1−q
− am+p−q
·
a
m+p−q−1
− am+p−q−2
a
m+p−q
− am+p−q−1
. . .
a
m+2
− am+1
a
m+3
− am+2
≥
a
m+2
− am+1
a
m+p+1−q
− am+p−q
andthus: 21
DT
p−1−q
(a
m+p−q
)
≤
a
m+p+1−q
− am+p−q
a
m+2
− am+1
.
2We have just proved that if
ξ
is any point in(a
m+p
, a
m+p+1
)
(and the same result holds for its negative ounterpart(a
−(m+p+1)
, a
−(m+p)
)
aswell)thenDT
p
(ξ)
≥
a
m+2
−a
m+1
a
m+p+1
−a
m+p
. In asimilar waywe anprovethelower bound:
DT
p
(ξ)
≤
a
0+
Moreover:
|zx
− zy| ≤ |T
p−1
Z
|
. Finally:(
5)
≤
a
m+p+1−q
− am+p−q
a
m+2
− am+1
|T
p−1
Z
| .
(7)Usinglemmas1and2weseethatthereexistsa onstant
C0
dependingonlyonthemapT
su h that:sup
(a
m+q−1
,a
m+q
)
|D
2
T
|
(a
m+q+1
− am+q
)
≤ C0
·
1
(q + m)
γ−2
γ−1
(q + m)
γ
γ−1
= C
0
·
1
(q + m)
2
.
Therefore the sum over
q = 1, 2, . . .
is summable and there exists a onstantC
1
su h thatforx, y
∈ Z
−
m,p
:p−1
X
q=1
sup
(a
m+p−q−1
,a
m+p−q
)
|D
2
T
|
|T
q
x
− T
q
y
| ≤ C1|T
p−1
Z
| .
(8)(ii)Inthis ase we need to ontrol thebehaviorof themap lose to
0
. Inparti ular, by usinglemmas 1 and2 (and thesymmetryofb
±i
) we start bynoti ing thatsup
(b
i+1
,b
i
)
|D
2
T
|
|bi
− bi+1| = O(
i
2γ−1
γ−1
i
2γ−1
γ−1
) = 1.
(9)Combining (8) and (9) with (4) we get that there exists a onstant
D
2
so that for allj
≤ p − 1
1
D
2
≤
DT
j
(x)
DT
j
(y)
≤ D
2
.
(10)Let's all
α = b
−(m+p−1)
, β = b
−(m+p)
the end points ofZ
−
m,p
. Forj
1
, j
2
≤ p − 1
there existη1
∈ (x, y)
andη2
∈ (α, β)
su hthat:|T
j
1
x
− T
j
1
y
| = DT
j
1
(η
1
)
|x − y|,
|T
j
2
α
− T
j
2
β
| = DT
j
2
(η2)
|α − β|.
Thedistortionbound(10) yields
T
j
1
x
− T
j
1
y
|T
j
1
α
− T
j
1
β
|
≤ D
2
2
T
j
2
x
− T
j
2
y
|T
j
2
α
− T
j
2
β
|
.
Ifwe now hoose
j
1
= 0
andj
2
= p
− 1
thensup
(α,β)
|D
2
T
|
|x − y| ≤ D
2
2
sup
(α,β)
|D
2
T
|
|α − β| · |T
p−1
x
− T
p−1
y
|
|T
p−1
α
− T
p−1
β
|
.
Sin e|T
p−1
α
− T
p−1
β
| = lm+1
= a
m
− am+1
andx
andy
to belong toZ
weget:sup
(α,β)
|D
2
T
|
|x − y| ≤ D
2
2
sup
(α,β)
|D
2
T
|
|α − β| · |T
p−1
Z
|
lm+1
and using distortion bound (9 ) on emore we have that there exist a onstant
C
2
su h that:sup
(α,β)
|D
2
T
|
|x − y| ≤ C2
|T
p−1
Z
|
l
m+1
.
By olle ting lemma 4(i) and 4(ii) we see that the ratio
|DT
p
(x)/DT
p
(y)
|
,(x, y
∈ Z)
is bounded as:DT
p
(x)
DT
p
(y)
≤ exp
C2
|T
p−1
Z
|
l
m+1
+ C1|T
p−1
Z
|
≤ exp
K2|T
p−1
Z
|
(11) withK
2
= C
1
+ C
2
/l
m+1
.Wenish the proofof the Propositionby hoosing
K = max(K
1
, K
2
)
3 De ay of orrelations
In this se tion and inthe next we prove several statisti al properties for our map: they are basi ally onsequen es of the distortion inequality got inthe previous se tion mat hed with establishedte hniques.
Proposition 5. The map
T
enjoys polynomial de ay of orrelations (w.r.t. the Lebesgue measurem
), for Hölder ontinuous fun tions onT
. More pre isely, for all Hölderϕ : T
→ R
andallψ
∈ L
∞
(T, m)
, we have:Z
(ϕ
◦ T
n
) ψ dm
−
Z
ϕ dm
Z
ψ dm
= O
1
n
γ−1
1
.
Proof. WewilluseLai-SangYoung'stowerte hnique[30 ℄. Webuildthetowerovertheinterval
I
0
andwedene the return timefun tion astherstreturn time: for allx
∈ I0
, R(x) := min
{n ∈ N
+
; T
n
x
∈ I0} := τI
0
(x) .
Thetower isthus dened by:
∆ =
{(x, l) ∈ I0
× N ; l ≤ τI
0
(x)
− 1}
andthepartition ofthebase
I
0
isgiven bythe ylindersZ
0,p
dened intheprevious se tion. Re allthatthe dynami s on thetowerisgiven by:F (x, l) =
(
(x, l + 1)
ifl < τ
I
0
(x)
− 1
(T
τ
I0
(x)
(x), 0)
ifl = τ
I
0
(x)
− 1
A ordingto[30℄,thede ayof orrelationsisgovernedbytheasymptoti sof
m
{x ∈ I0
; τ
I
0
(x)
≥
n
}
namelym
{x ∈ I0
; τ
I
0
(x) > n
} = m(b−n
, b
n
)
∼
1
γ
2γ
γ
− 1
γ
γ−1
1
(n
− 1)
γ−1
γ
.
useful in the next se tion about limit theorems. Let us rst introdu e the separation time
s(x, y)
between two pointsx
andy
inI
0
. PutT
ˆ
the rst return map onI
0
; we denes(x, y) = min
n≥0{( ˆ
T
n
(x), ˆ
T
n
(y))
lie indistin t
Z
0,p
, p
≥ 1}
. We ask that∃C > 0, δ ∈ (0, 1)
su h that∀x, y ∈ Z0,p
, p
≥ 1
,we haveD ˆ
T (x)
D ˆ
T (y)
≤ exp[Cδ
s( ˆ
T (x), ˆ
T (y))
] .
(12)Let us prove this inequality. Remember that the ylinder
Z
0,p
is the disjoint union of two open omponents,Z
+
0,p
andZ
−
0,p
,whi hsitontheoppositesidesof0
. Supposerstthatx
andy
stay in the same open omponent of someZ
0,p
,p
≥ 1
, and thats( ˆ
T (x), ˆ
T (y)) = n
; then sin etheorbits(underˆ
T
)ofthetwo pointswillbeinthesame ylinderuptotimen
− 1
,and on these ylindersT
ˆ
is monotone and uniformly expanding,|D ˆ
T
| ≥ β > 1
(see footnote 1), we have| ˆ
T (x)
− ˆ
T (y)
| ≤ β
−(n−1)
. Thereforebythedistortioninequalitywe get
D ˆ
T (x)
D ˆ
T (y)
≤ exp
h
Kβ
−(n−1)
i
≤ exp[Cδ
s( ˆ
T (x), ˆ
T (y))
] ,
(13) whereC = Kβ
andδ = β
−1
. Ifinstead
x, y
lieinthe two dierent open omponentsof someZ
0,p
,p
≥ 1
,andagains( ˆ
T (x), ˆ
T (y)) = n
,thismeansthat−x
andy
willhavethesame oding upton
;hen eD ˆ
T (x)
D ˆ
T (y)
=
D ˆ
T (
−x)
D ˆ
T (y)
≤ exp[K| ˆ
T (
−x) − ˆ
T (y)
|] ≤ exp
h
Kβ
−(n−1)
i
≤ exp[Cδ
s( ˆ
T (x), ˆ
T (y))
] .
A ordingto[30℄the orrelationsde aysatises
R
(ϕ
◦ T
n
) ψ dm
−
R
ϕ dm
R
ψ dm
= O(
P
k>n
m
{x ∈
I
0
; τ
I
0
(x)
≥ k}
andtherighthand side ofthis inequality behaveslikeO n
−
1
γ−1
.
Optimal bounds The previous result on the de ay of orrelations ould be strengthened to produ e a lowerbound for the de ayof orrelations for integrable fun tionswhi h vanish in a neighborhood of theindierent xed point. We will usefor that therenewal te hnique introdu ed by Sarig [28 ℄ and su esively improved by Gouëzel [13 ℄. We rst need that our original map is irredu ible: this is a onsequen e of the already proved ergodi ity, but one ouldshowndire tlybyinspe tionthatthe ountableMarkovpartitiongivenbythepreimages ofzerohassu haproperty. Wemoreoverneedadditionalpropertiesthatwedire tlyformulate inour setting:
•
Suppose we indu e onIm
= (a−m, am)/
{0}
and allZm
the Markov partition into the re tanglesZ
m,p
with rst returnp
. A ylinder[d
0
, d
1
,
· · · , dn−1
]
withd
i
∈ Zm
will be theset∩
n−1
l=0
T
ˆ
−i
d
l
.We rst need that the ja obian of the rst return map is lo ally Hölder ontinuous, namelythatthere exists
θ < 1
su h that:wherethe supremumistaken overall ouples
x, y
∈ [d0
, d
1
,
· · · , dn−1
]
,d
i
∈ Zm
andC
is apositive onstant. Butthisis animmediate onsequen eofformula(13)withθ = β
−1
and
C = Kβ
. Usingtheseparationtimes(
·, ·)
,wedeneD
m
f = sup
|f(x)−f(y)|/θ
s(x,y)
, where
f
is an integrable fun tion onI
m
and the supremum is taken over all ouplesx, y
∈ Im
. We thenput||f||L
θ,m
≡ ||f||∞
+ Dmf
. We allLθ,m
the spa e ofθ
-Hölder fun tionsonI
m
.•
We need the so- alled big image property, whi h means that the Lebesgue measure of theimages, underT
ˆ
, all the re tanglesZ
m,p
∈ Zm
are uniformly bounded from below byastri tlypositive onstant. Inour ase,seese tion3,theseimagesareboundedfrom below bythelength ofthe interval(a
−m
, a
m
)
.•
Wenallyneedthatm(x
∈ Im|τ(x) > n) = O(n
−χ
)
,forsome
χ > 1
(thisisGouëzel's as-sumption,whi himprovesSarig'sone,askingforχ > 2
). Inour asebythe onstru tion developed inSe t. 3 we immediately get thatm(x
∈ Im|τ(x) > n) = m(∪p>n
Z
m,p
) =
(b
−(m+n)
, b
m+n
)
∼ C(n + m)
−b
= Cn
−b
(1 + m/n)
−b
∼ Cn
−b
, where the onstantsC
andb
arethe sameasthose given inthe proof of Th. 4,pre iselyC =
1
γ
2γ
γ−1
γ
γ−1
andb =
γ−1
γ
.Undertheseassumptions,SarigandGouëzelprovedalowerboundforthede ayof orrelations whi hwedire tly spe ialize toour map:
Proposition 6. There exists a onstant
C
su h that for allf
whi h areθ
-Hölder andg
integrable andboth supported inI
m
we haveCorr( f, g
◦ T
n
)
− (
∞
X
k=n+1
m(x
∈ Im|τ(x) > n))
Z
g dm
Z
f dm
≤ CFγ
(n)
||g||∞||f||L
θ,m
whereF
γ
(n) =
1
n
γ
γ−1
ifγ < 2
,(log n)/n
2
ifγ = 2
and1
n
γ−1
2
ifγ > 2
. Moreover, ifR
f dm = 0
, thenR
(g
◦ T
n
) f dm =
O(
1
n
γ
γ−1
)
. Finally the entral limit theorem holdsfor the observable
f
.Remark2. (i)Sin ewhen
m
→ ∞
,I
m
oversmod-0
alltheinterval(
−1, 1)
wegetanoptimal de ayof orrelationsoforderO(
1
n
γ−1
1
)
forallintegrablesmoothenoughfun tionswhi hvanish in a neighborhood of
1
.(ii) The last senten e about the existen e of the entral limit theorem will be also obtained, using a dierent te hnique, in Proposition 5, part 2, (a).
4 Limit theorems
Letus re all the notion of stablelaw (see [9, 12℄): a stablelaw isthe limit ofa res aled i.i.d pro ess. Morepre isely,the distribution ofa random variable
X
is said to be stable ifthere existan i.i.dsto hasti pro ess(X
i
)
i∈N
and some onstantsA
n
∈ R
andB
n
> 0
su hthatin distribution:1
B
n
n−1
X
i=0
Xi
− An
−→ X .
The kind of laws we are interested in an be hara terized by their index
p
∈ (0, 1) ∪ (1, 2)
, dened asfollowed:m(X > t) = (c
1
+ o(1))t
−p
,
m(X <
−t) = (c2
+ o(1))t
−p
,
where
c
1
≥ 0
andc
2
≥ 0
aretwo onstantssu hthatc
1
+ c
2
> 0
,andbyothertwoparameters:c =
(c
1
+ c
2
)Γ(1
− p) cos(
pπ
2
)
p
∈ (0, 1) ∪ (1, 2)
1
2
p = 2
,
β =
c
1
− c2
c1
+ c2
.
We will denoteby
X(p, c, β)
the lawwhose hara teristi fun tion isE(e
X(p,c,β)
) = e
−c|t|
p
1−iβ
sgn(t) tan(
pπ
2
)
.
Proposition 7. Letus denote
S
n
ϕ =
Pn−1
k=0
ϕ
◦ T
k
, whereϕ
is anν
-Hölder observable, withR
ϕ(x) dx = 0
.1. If
γ < 2
thenthe Central LimitTheoremholdsfor anyν > 0
. Thatistosaythere exists a onstantσ
2
su h thatS
n
ϕ
√
n
tends in distribution toN (0, σ
2
)
. 2. Ifγ > 2
then: (a) Ifϕ(1) = 0
andν >
1
2
(γ
− 2)
then theCentral LimitTheorem stillholds. Moreoverσ
2
= 0
i there exists a measurable fun tion
ψ
su h thatφ = ψ
◦ T − ψ
(b) Ifϕ(1)
6= 0
thenS
n
ϕ
n
γ−1
γ
onverges in distribution tothe stable law
X p, c, β
with:p =
γ
γ
− 1
c =
1
2γ
2γϕ(1)
γ
− 1
γ
γ−1
Γ(
1
(1
− γ)
) cos(
πγ
2(γ
− 1)
)
β =
sgnϕ(1)
3. Ifγ = 2
then:(a) If
ϕ(1) = 0
then the Central LimitTheorem holds. (b) Ifϕ(1)
6= 0
thenthere exista onstantb
su h thatS
n
ϕ
√
n log n
tends in distributiontoN (0, b)
. Proof.1. Asaby-produ tof the tower's theory we gettheexisten eof the entral limit theorem wheneverthe rateofde ayof orrelationsissummable([30 ℄,Th.4);thishappensinour asefor
γ < 2
. Asusual we shouldavoidthatφ
isa o-boundary.paraboli maps of the interval. We defer the reader to Gouëzel's paper for the preparatory theory; weonly proveherethe ne essary onditions forits appli ation. Weindu e againon
I0
andwe putϕI
0
(x) :=
Pτ
I0
−1
i=0
ϕ(T
i
x)
. We need:i.
φ
mustbelo allyθ
-HölderonI
0
(resp.T
),withθ < 1
,whi h meansthatthere existsa onstantC
su h that|φ(x) − φ(y)| ≤ Cθ
s(x,y)
∀x, y ∈ I0
(resp.
T
)withs(x, y)
≥ 1
. We extend the separation times(x, y)
to the ambient spa e as follows: ifx, y
∈ T
, allx, ˆ
ˆ
y
their rstreturnstoI
0
. WheneverT
i
x, T
i
y
stayin the same element oftheMarkovpartition
{Im}m∈Z
until therst returntoI0
, we puts(x, y) = s(ˆ
x, ˆ
y) + 1
;otherwises(x, y) = 0
.ii.
m
{x ∈ I0
; τ
I
0
(x) > n
} = O(1/n
η+1
)
,for someη > 1
iii.ϕI
0
∈ L
2
I0
.
Re all that the indu ed map
T
ˆ
onI
0
is uniformly expanding with fa torβ > 1
; therefore for any ouple of pointsx, y
∈ T
we have|x − y|
T
≤ Bβ
−s(x,y)
,where
B
is asuitable onstant and| · |
T
denotes the distan eonthe ir le. UsingtheHölder assumption onφ
weget|φ(x) − φ(y)| ≤ D|x − y|
ν
T
≤ Eβ
−νs(x,y)
,whi hshows thatφ
islo ally Hölderwithθ = β
−ν
< 1
.
Thequantityinthese onditemaboveisexa tly
(b
n
, b
−n
)
for whi h weobtainedin the previous se tiona boundof ordern
−(
γ−1
γ
)
. Hen e
η = γ/(γ
− 1) − 1
. Toprove thethird itemdenoteC
ϕ
=
R
I
0
|ϕ(x)|
2
dx
we obtain:Z
I
+
0
|ϕ
I
+
0
(x)
|
2
dx = C
ϕ
+
+∞
X
p=2
Z
Z
0,p
p−1
X
i=0
ϕ(T
i
x)
2
dx
.
C
ϕ
+ 2
+∞
X
p=2
Z
b
p−1
b
p
p−1
X
i=0
|T
i
x
− 1|
ν
T
2
dx
.
C
ϕ
+ 2
+∞
X
p=2
Z
b
p−1
b
p
p−1
X
i=0
|ai
− 1|
ν
2
dx
.
C
ϕ
+ 2
+∞
X
p=2
m(b
p
− bp−1
)p
2(−
ν
γ−1
+1)
dx
.
C
ϕ
+ 2
+∞
X
p=2
p
−(
γ−1
γ
+1)
p
2(−
ν
γ−1
+1)
dx .
Finally if2(−ν+γ−1)
γ−1
−
γ
γ−1
− 1 < −1
(i.e.ν >
1
2
(γ
− 2)
) thenϕI
0
∈ L
2
I0
. (b) Using the fa tthatm[u > nϕ(
−1)] = m(bn
, b
−n
)
∼
1
2γ
2γ
γ
− 1
γ
γ−1
1
n
γ−1
γ
and the proof in2.(a),theresult easily follows alongthe samelines oftheproof of Th. 1.3in[12 ℄.
Large deviations.
The knowledgeof the measure ofthe tail for the rst returnson thetower(in our ase built over
I
0
), will allows us to apply the results of Melbourne and Ni ol [22 ℄ to get the large deviations property for Hölder observables. Applied to our framework, their theorem states thatifm(x; τ
I
0
> n) =
O(n
−(ζ+1)
)
,with
ζ > 0
,thenfor allobservablesφ : [
−1, 1] → R
whi h areHölder and whi hwe takeof zeromean, we have thelargedeviations bounds:Proposition 8. If
γ < 2
then the mapT
veriesthe followinglarge deviationsbounds: (I)∀ǫ > 0
andδ > 0
, there existsa onstantC
≥ 1
(depending onφ
) su h thatm
1
n
n−1
X
j=0
φ(T
j
(x))
> ǫ
≤ Cn
−(ζ−δ)
.
(II) For an open and dense set of Hölder observables
φ
, and for allǫ
su iently small, we havem
1
n
n−1
X
j=0
φ(T
j
(x))
> ǫ
≥ n
−(ζ−δ)
for innitely many
n
andeveryδ > 0
.Remark3. TheMelbourneandNi olresult hasbeenre ently strenghtened byMelbourne[23℄; byadopting thesamenotationasabove, heproved thatwheneverthe observable
φ
isL
∞
(with respe tto theLebesgue measure
m
),andζ + 1 > 0
,then foranyǫ
there existsa onstantC
φ,ǫ
su h thatm
1
n
n−1
X
j=0
φ(T
j
(x))
> ǫ
≤ Cφ,ǫ
n
−ζ
for all
n
≥ 1
. Translated to our map, this means that we have the large deviation property wheneverγ > 1
. SimilarresultshavebeenobtainedbyPolli otandSharp[26℄forthe Pomeau-Manneville lassofmaps;hopefullythey ould begeneralized inthepresen eofunboundedrst derivaties.5 Re urren e Firstreturns.
In the past ten years the statisti s of rst return and hitting times have been widely used as new and interesting tools to understand the re urren e behaviors in dynami al systems. Surveys of the latestresults andsome histori al ba kground an be foundin[20, 17,1 ℄.
Takeaball
B
r
(x)
orradiusr
aroundthepointx
∈ T
and onsidertherstreturnτ
B
r
(x)
(y)
we askwhether thereexiststhe limit of thefollowingdistribution when
r
→ 0
3 :F
r
e
(t) = m
r
y
∈ Br
(x); τ
B
r
(x)
m(B
r
(x)) > t
.
The distributionF
h
r
(t)
for the rst hitting time (intoB
r
(x)
) is dened analogously just takingy
and the probabilitym
onthewhole spa eT
.A powerful toolto investigate su h distributions for non-uniformly expandingand hyper-boli systemsis given bythe onjun tion of thefollowing results, whi h redu ethe omputa-tions to indu edsubsets.
•
Suppose(T, X, µ)
is an ergodi measure preserving transformation of a smooth Rie-mannian manifoldX
; takeX
ˆ
⊂ X
an open set and equip it with the rst return mapˆ
T
and with the indu ed (ergodi ) measureµ
ˆ
. Forx
∈ ˆ
X
we onsider the ballB
r
(x)
(Br(x)
⊂ ˆ
X
)arounditandwe writeτ
ˆ
B
r
(x)
(y)
fortherstreturnofthepointy
∈ Br(x)
underT
ˆ
. We now onsider the distribution of the rst return time for the two vari-ablesτ
B
r
(x)
andτ
ˆ
B
r
(x)
in the respe tive probability spa es (B
r
(x), µ
r
) and (B
r
(x), ˆ
µ
r
) (whereagainthesubindexr
means onditioning totheballB
r
(x)
),as:F
e
r
(t) = µ
r
(y
∈
B
r
(x)); τ
B
r
(x)
(y)µ(B
r
(x)) > t)
andF
ˆ
e
r
(t) = ˆ
µ
r
(y
∈ Br
(x)); ˆ
τ
B
r
(x)
(y)ˆ
µ(B
r
(x)) > t)
.In [4℄ it is proved the following result: suppose that for
µ
-a.e.x
∈ ˆ
X
the distributionˆ
F
r
e
(t)
onverges pointwise to the ontinuous fun tionsf
e
(t)
when
r
→ 0
(remember that the previous distribution depend onx
via the lo ation of the ballB
r
(x)
); then we have as wellF
e
r
(t)
→ f
e
(t)
and the onvergen e is uniform int
4. We should note thatwheneverwe havethedistribution
f
e
(t)
for therstreturn timewe an insurethe existen e of the weak-limit distribution for therst hitting time
F
h
r
(t)
→ f
h
(t)
wheref
h
(t) =
R
0
t
(1
− f
e
(s))ds, t
≥ 0
[16 ℄.Note: From now on we will saythat we have
f
e,k
(t)
as limit distributions for balls, if we gettheminthelimit
r
→ 0
and forµ
-almost all the entersx
oftheballsBr(x)
.•
Thepreviousresultisusefulifweareabletohandlewithre urren eonindu edsubsets, see [5, 6℄ for a few appli ations. Indu tion for one-dimensional maps often produ es pie ewise monotoni maps with ountably many pie es. An interesting lass of su h mapsare the Ry hlik'smaps [27 ℄ : in[4 ℄ Def. 3.1theunderlying measure is onformal. Whenthe onformalmeasureistheLebesguemeasurem
,thenRy hlik'smaps ouldbe hara terized inthe following way:Let
T : Y
→ X
be a ontinuous map,Y
⊂ X
open and dense,m(Y ) = 1
andX
isthe unitinterval or the ir le. Suppose there existsa ountable family of pairwise disjoint open intervalsZi
su h thatY =
S
i≤1
Zi
andT
is: (i)C
2
on ea h
Zi
; (ii) uniformly expanding:inf
Z
i
inf
x∈Z
i
|DT (x)| ≥ β > 1
;(iii) Var(g) <
∞
,whereg = 1/
|DT (x)|
whenx
∈ Y
and0
otherwise(Varg
denotes thetotal variation of thefun tiong : R
→ ∞
). 3We allitdistributionwithabuseoflanguage;inprobabilisti terminologyweshouldrathertake
1
minusthat quantity.4
The result proved in [4℄ is slightly more general sin e it doesn't require the ontinuity of the asymptoti distributions over all
t
≥ 0
. We should note instead that we ould relax the assumption thatX
ˆ
is open just removingfromitasetofmeasurezero,whi h willhappenonourindu edsetsI
m
.balls(i.e.
f
e
(t) = f
k
(t) = e
−t
),whenevertheinvariantmeasureisabsolutely ontinuous w.r.t.
m
andmoreoverthis invariant measure ismixing.Beforewe formulateour nextresultforthemaps
T
investigated inthis paper letusprovethe following lemma.Lemma 9. The map
T
ˆ
is Ry hlik on the ylindersI
m
,m
∈ Z
and the variation of|D ˆ
T
|
is nite on ea h of them.Proof. (see[4℄). Letus onsiderthe ylinder
I
m
and partitionitinto the ylindersZ
m,p
with rst returnp
≥ 1
,aswedidinthese ond se tion; then we have for thevariationonI
m
Var
1
|D ˆ
T
|
≤
X
Z
m,p
Z
Z
m,p
|D
2
T (t)
ˆ
|
|D ˆ
T (t)
|
2
dt + 2
X
Z
m,p
sup
Z
m,p
1
|D ˆ
T
|
.
Bythe distortionbound proved inthese ondse tion we have that
e
2K
≥
D ˆ
T (x)
D ˆ
T (y)
≥
Z
y
x
D
2
T (t)
ˆ
D ˆ
T (t)
dt
≥
Z
y
x
|D
2
T (t)
ˆ
|
D ˆ
T (t)
dt
for any
x, y
∈ Zm,p
, sin e the rst derivative is always positive and the se ond derivative hasthe same sign for all thepoints inthe same ylinder. But this immediately impliesthatR
Z
m,p
|D
2
T (t)|
ˆ
|D ˆ
T (t)|
2
dt
≤ supZ
m,p
1
|D ˆ
T |
e
2K
. Using Remark(1) we an restri t to
Z
−
m
. Sin eˆ
T
mapsZ
m,p>1
−
dieomorphi ally onto(a
m−1
, a
m
)
andZ
−
m,1
onto(a
−(m−1)
, a
m
)
⊃ (am−1
, a
m
)
there will be apointξ
for whi hD ˆ
T (ξ)m(Z
m,p
)
≥ m(am−1
, a
m
)
. Applying thebounded distortion estimateone more time,we getsup
Z
m,p
1
|D ˆ
T |
≤
e
2K
m(Z
m,p
)
m(a
m−1
,a
m
)
. We nallyobtain Var1
|D ˆ
T
|
≤
e
2K
(2 + e
2K
)
m(a
m−1
, a
m
)
X
Z
m,p
m(Z
m,p
) <
∞ .
Thefollowing resultnowfollows by[4℄ Theorem 3.2.
Proposition10. Themap
T
hasexponentialreturnandhittingtimedistributionswithrespe t to themeasurem
providedγ > 1
.Number of visits.
Letus ome ba k to the general framework introdu ed in Se t. 5.1 withthetwo probability spa es
(X, T, µ)
and( ˆ
X, ˆ
T , ˆ
µ)
. Wenowintrodu ethe randomvariablesξ
e
r
andˆ
ξ
r
e
whi h ount thenumber of visitsof the orbits of a pointy
∈ Br
(x)
to the ball itself and up to a ertain res aledtime. Namely:ξ
r
e
(x, t)
≡
h
t
µ
(Br (x))
i
X
j=1
χ
B
r
(x)
T
j
(y)
,
where
χ
stands for the hara teristi fun tion andx
∈ X
. If we takex
∈ ˆ
X
we an dene in the same manner the variableˆ
ξ
r
e
(x, t)
by repla ing the a tion ofT
with that ofˆ
T
. We now introdu e the two distributionsG
e
r
(t, k) = µ
r
(x; ξ
r
e
(x, t) = k), ˆ
G
e
r
(t, k) = ˆ
µ
r
(x; ˆ
ξ
r
e
(x, t) = k) ,
where again the index
r
for the measures means onditioning onBr(x)
. It is proved in [4 ℄ that whenever the distributionG
ˆ
e
r
(t, k)
onverges weakly (int
) to the fun tiong(t, k)
and for almost allx
∈ ˆ
X
, the same happens, with the same limit, to the distributionG
e
r
(t, k)
. For systems with strong mixing properties the limit distribution is usually expe ted to be Poissonian [20 , 17,18 , 1℄:t
k
e
−t
k!
.In[10 ℄itwasshownthatRy hlikmapsenjoyPoissonstatisti sforthelimitdistributionof thevariables
ξ
e
r
and wheneverthe enter oftheball istakena.e.. Hen e we getthefollowing result.Proposition 11. Let
γ > 1
. Then form
-almost everyx
the number of visits to the ballsB
r
(x)
onverges to the Poissoniandistribution asr
→ 0
.Extreme Values.
Thelastquotedpaper[10℄ ontainsanotherinteresting appli ationofthestatisti softherst hitting time that we ould apply to our map
T
too. Let us rst briey re all the Extreme ValueTheory. Given theprobability measurepreservingdynami al system(X, T, µ)
andthe observableφ : X
→ R ∩ {±∞}
, we onsider the pro essY
n
= φ
◦ T
n
for
n
∈ N
. Then we dene the partial maximumM
n
≡ max{Y0
,
· · · , Yn−1}
and we look ifthere are normalising sequen es{an}n∈N
⊂ R
+
and
{bn}n∈N
⊂ R
su hthatµ(
{x : an(Mn
− bn)
≤ y}) → H(y)
for some non-degenerate distribution fun tion
H
: in this ase we will say that an Extreme Value Law (EVL) holds forM
n
. If the variablesY
n
were i.i.d., the lassi al extreme value theorypres ribesthe existen eofonlythreetypesofnon-degenerateasymptoti distributions for themaximumM
n
and underlinearnormalisation, namely:•
Type 1:EV
1
= e
−e
−y
for
y
∈ R
,whi h is alledtheGumbel law.•
Type2:EV
2
= e
−y
−α
for
y > 0
,EV
2
= 0
,otherwise,whereα > 0
isaparameter, whi h is alledFre het law.•
Type 3:EV
3
= e
−(−y)
α
for
y
≤ 0
,EV
3
= 1
, otherwise, whereα > 0
is a parameter, whi h is alled Weibull law.From now on we will take
X
asa Riemannianmanifold withdistan ed
andµ
an absolutely ontinuous(w.r.t. Lebesgue)probabilityinvariantmeasure. Moreover onsidertheobservableφ
of the formφ(x) = g(d(x, ξ))
, whereξ
is a hosenpoint inX
. The fun tiong : [0,
∞) →
R
∪ {+∞}
is a stri tly de reasing bije tion ina neighborhood of0
and it has0
as a global maximum(eventually+
∞
). The fun tiong
ould be taken inthree lasses; we defer to [10 ℄ for the pre ise hara terization. Important representatives of su h lasses (denoted by theindi es 1,2,3) are
g
1
(x) =
− log(x)
;g
2
(x) = x
−1/α
for some
α > 0
;g
3
(x) = D
− x
−1/α
, for some
D
∈ R
andα > 0
. We also remind the distribution of the rst hitting timeF
h
r
(t)
into the ball
B
r
(x)
introdu ed above; we say that a system enjoys exponential hitting time statisti s(EHTS)ifF
h
r
(t)
onverges point wisetoe
−t
for
µ
-a.e.x
∈ X
(wesawbeforethatit is equivalent to getthe exponential limit distribution for the rst return time). We are now ready to state the resultin [10℄whi h establishes an equivalen e between theEHTS andthe EVL; we will be in parti ular on erned with the following impli ation: suppose the system(X, T, µ
) hasEHTS;then it satisesan EVL forthe partial maximumMn
onstru ted on the pro essφ(x) = g(d(x, ξ))
, whereg
is taken in one of the three lasses introdu ed above. In parti ular ifg = g
i
we have an EVL forM
n
of typeEV
i
.Of ourse thisresult an beimmediatelyapplied to themapping
T
underinvestigationin this paper.6 Generalizations
As mentioned in the Introdu tion the original paper by Grossmann and Horner [14 ℄ dealt withdierent Lorenz-like maps
S
whi h map[
−1, 1]
onto itself with two surje tive symmet-ri bran hes dened on the half intervals[
−1, 0]
and[0, 1]
. They have the following lo al behaviour:S(x)
∼ 1 − b|x|
κ
, x
≈ 0, b > 0
S(x)
∼ −x + a|x − 1|
γ
, x
≈ 1−
, a > 0
S(x)
∼ x + a|x + 1|
γ
, x
≈ −1+
where
κ
∈ (0, 1)
andγ > 1
aretwoparameters. We also requirethat(i) inall points
x
6= −1, 1
theabsolutevalueof thederivativeis stri tly biggerthan1
. (ii)S
is stri tly in reasing on[
−1, 0]
, stri tly de reasing on[0, 1]
and onvex on the two in-tervals(
−1, 0), (0, 1)
Themaphasa uspatthe originwhere theleftandrightrstderivativesdivergeto
±∞
and thexed point−1
is paraboli (Fig. 2). Although the mapS
is Markov withrespe tto the partition{[−1, 0], [0, 1]}
itwillbemore onvenient tousea ountableMarkovpartitionwhose endpointsaregiven bysuitable preimages of0
(see below).The reexion symmetry of the map
T
in Se t. 2 was related to the invarian e of the Lebesgue measure. We do not really need that the mapS
is symmetri with respe t to the origin. We didthis hoi eto getonly two s aling exponents (κ
andγ
) in0
and in±1
. This impliesinparti ular thesame s alings forthe preimages of0
on(
−1, 0)
and(0, 1)
. Iftheleft and rigt bran hes arenot anymoresymmetri , still preserving theMarkov stru ture andthe presen eofindierentpointsandofapoint withunbounded derivative,oneshould playwith at most four s alingexponentsgiving the lo albehaviorofS
in0
and±1
.We denote by
S
1
(resp.S
2
) the restri tion ofS
to[
−1, 0]
(resp.[0, 1]
) and denea
0+
=
-1
0
x
1
-1
0
1
S(x)
a
-2
a
-1
a
0-
a
0+
a
1
a
2
b
-1
b
-2
b
2
b
1
Sa
−p
= Sa
p
= a
−(p−1)
. In the same way as we did in the rst se tion we dene the se-quen eb
p
, p
≥ 1
as:Sb
±p
= a
p−1
. The ountable Markov partition,mod m
, will be(a
−p
, a
−(p−1)
) : p
≥ 1
∪ {(ap
, a
p+1
) : p
≥ 1} ∪ {I0}, I0
≡ (a0−
, a
0+
)/
{0}
. From thelo albehaviors one getsthefollowing s aling relationsa
p
=
−a−p
∼ 1 −
1
a(γ
− 1)
1
γ−1
1
p
γ−1
1
a
p
− ap+1
∼ a
1
a(γ
− 1)
γ
γ−1
1
p
γ−1
γ
b
p
=
−b−p
∼
1
ab
(γ−1)
(γ
− 1)
1
k
(γ−1)
1
p
k
(γ−1)
1
b
p
− bp−1
∼
1
k ab
(γ−1)
k(γ−1)
1
1
γ
− 1
k
(γ−1)+1
k(γ−1)
1
p
k
(γ−1)+1
k(γ−1)
•
Therole of Remark1 isplayed here bythe monotoni ityof the right bran h: wheneverx, y
sit on dierent omponents we an just note that|DS(−x)| = |DS(x)|
and that afterone iterationS(x) = S(
−x)
.5
•
Letus onsideragainthestepfromthersttothese ondupperboundin(4 ): wesimply dis ardedthedenominatorgivenbytheinmumoftherstderivativeoverthesetswith given rst return time, sin e itwas ininuent for the mapT
. Instead itwill now plays animportantrolesin eitmakesboundedthefollowingratiosin e,asitiseasyto he k, :|bn+1
− bn| sup(b
n+1
,b
n
)
|D
2
S
|
inf
(b
n+1
,b
n
)
|DS|
=
O(
1
n
) .
Invariant measure and de ay of orrelations. An important dieren e with themap on the ir leisthatwearenotguaranteedthattheLebesguemeasure
m
isanymoreinvariant;sowe havetobuildanabsolutely ontinuousinvariantmeasureµ
. Fortunatelythetower'ste hniques helpsus again. Ifthe tailof thereturn timeonthebaseofthetower ism
-summableandthe distortionis bounded, itfollows the existen eof su hµ
. Tobemore pre ise letus indu e on the ylinderI0
. A sub ylinderZp
ofI0
with rstreturn timep
will have theform6
Z
1
= (a
0−
, b
−1
)
∪ (b1
, a
0+
)
(14)Z
p
= (b
−(p−1)
, b
−p
)
∪ (bp
, b
p−1
) p > 1 .
Consequently the Lebesgue measure of the points in
I
0
with rst return bigger thann
s aleslikem(x
∈ I0
; τ
I
0
(x) > n)
≈
1
n
κ
(γ−1)
1
We an thus invoke Th.1 inLai-Sang Young's paper[30 ℄ to get:
Proposition 12. Let us onsider the map
S
depending upon the parametersγ
andκ
. Then for0 < κ <
1
γ−1
(or for0 < κ < 1
, whenγ
≤ 2
), we get the existen e of an absolutely ontinuous invariant measureµ
whi h mixes polynomially fast on Hölder observables with rateO n
−
1−κ(γ−1)
κ
(γ−1)
.
The map has exponential return and hitting times distributions and Poissonian statisti for the limit distributionof thenumber of visits in balls.
5
In the asymmetri ase
|DS(−x)| 6= |DS(x)|
but still after one iterateS(x)
andS(y)
sit on the same side. This imply that multiplying bythe appropriatefa tor we antreatthe asymmetri ase in the sameway asthe symmetri one.6
We would like to note that, ontrarily to the map