Abdumajid Begmatov

Faculty of Mathematics and Mechanics. Samarkand State University

Boulevard Street 15
140104 Samarkand, Uzbekistan
Akhtam Dzhalilov
Turin Politechnic University in Tashkent
Kichik halqa yuli 17
100095 Tashkent, Uzbekistan
Dieter Mayer
Institut für Theoretische Physik, TU Clausthal
Leibnizstrasse 10
D-38678 Clausthal-Zellerfeld, Germany
(Communicated by the associate editor name)
Let f be an orientation preserving circle homeomorphism with a single break
point xb , i.e. with a jump in the first derivative f 0 at the point xb , and with
irrational rotation number ρ = ρf . Suppose that f satisfies the Katznelson and
Ornstein smoothness conditions, i.e. f 0 is absolutely continuous on [xb , xb + 1] and
f 00 (x) ∈ Lp ([0, 1), d`) for some p > 1, where ` is Lebesque measure. We prove, that
1
the renormalizations of f are approximated by linear-fractional functions in C 1+L ,
1 00 1
that means, f is approximated in C − norm and f is appoximated in L − norm.
Also it is shown, that renormalizations of circle diffeomorphisms with irrational
rotation number satisfying the Katznelson and Ornstein smoothness conditions are
1
close to linear functions in C 1+L - norm.
Introduction Let f be an orientation preserving homeomorphism of the circle
S 1 = R1 /Z 1 , which we identifify with [0, 1), and with lift F : R → R, i.e.
f (x) = F (x) (mod 1), x ∈ S 1 ,
where F is a continuous, strictly increasing function on R and F (x + 1) = F (x) + 1,
x ∈ R. Suppose that the lift F satisfies the following conditions:
(a) there is a point xb ∈ S 1 such that the one-sided derivatives F 0 (xb −0), F 0 (xb +
0) exist, are positive and F 0 (xb − 0) 6= F 0 (xb + 0);
(b) F 0 (x) is absolutely continuous on [xb , xb + 1];
(c) F 00 (x) ∈ Lp ([0, 1], d`), for some p > 1, where ` is Lebesque measure.
Conditions (b), (c) are called the Katznelson and Ornstein smoothness con-
ditions for the homeomorphisms f with a break point xb . For diffeomorphisms
condition (b), is replaced by condition
(b0 ) F 0 (x) is absolutely continuous on the circle;
The ratio p
σ := σf (xb ) = F 0 (xb − 0)/F 0 (xb + 0) 6= 1
is called the jump ratio of f at xb or, for short, f -jump ratio. Notice that the
parameter σ is obviously an invariant under smooth coordinate transformations
and characterizes the type of the singularity. Denote by ρf the rotation number of
f , i.e.(see. [0])
F n (x)
ρf = lim (mod 1),
n→∞ n
n
where here and later F denotes the n-th iteration of the function F . Among
others, the renormalization approach, which since Feigenbaum’s works on interval
maps plays a central role in the modern theory of dynamical systems, is one of
the main tools in establishing rigidity results. Thereby the main goal is to prove
smoothness of the conjugacy between two dynamical systems, which a priori are only
topologically equivalent. For the circle dynamics it means that two maps with the
same irrational rotation number and the same local structure of their singular points
are smoothly conjugate to each other. For the class of circle diffeomorphisms the
corresponding statement is equivalent, in a certain sense, to Herman’s theory(see.
[0], [0], [0], [0], [0]).
Natural generalizations of circle diffeomorphisms are diffeomorphisms f with
break points,where i.e. f has jumps in the first derivative on finitely many points. In
general, their ergodic properties like their invariant measures, their renormalizations
and also their rigidity properties are rather different from those of diffeomorphisms
(see. chapter VI of [0], [0], [0], [0], [0], [0]).
Let the rotation number ρ = ρf of f be irrational and ρf = [k1 , k2 , . . . kn , . . .] be
it’s continued fraction expansion. For n ∈ N denote by pn /qn := [k1 , k2 , . . . , kn ], n ≥
1, the convergents of ρf . Their denominators qn satisfy the well known recursion
relation qn+1 = kn+1 qn + qn−1 , n ≥ 1, with initial conditions q0 = 1, q1 = k1 .
Fix any point x0 ∈ S 1 and consider its iterations, i.e. xi = f i (x0 ), i ≥ 1.
(n)
Define ∆0 (x0 ) as the closed interval on S 1 with endpoints x0 and xqn = f qn (x0 ),
such that for n odd xqn is to the left of x0 and for n even it is to its right. Denote
(n) (n) (n)
by ∆i (x0 ) := f i (∆0 (x0 )), i ≥ 1, the iterates of the interval ∆0 (x0 ) under f .
these had to be replaced by the lift F : R → R respectively subsets of R. Having
this in mind we introduce the renormalized coordinate z on Un given by

x = x0 + z(x0 − xqn−1 ). (1)

It is clear, that the coordinate z varies from -1 to 0, when x is varying from xqn−1
to x0 and it varies from 0 to an when x is varying from x0 to xqn , where an denotes
the new coordinate of the point xqn , i.e.

xqn − x0
an := an,x0 = . (2)
x0 − xqn−1

In the new renormalized coordinate the following pair of maps (fn,x0 , gn,x0 )
corresponds to the map πn,x0 (x):

F qn (x0 + z(x0 − xqn−1 )) − x0 − pn

fn,x0 (z) = , z ∈ [−1, 0), (3)
x0 − xqn−1

F qn−1 (x0 + z(x0 − xqn−1 )) − x0 − pn−1

gn,x0 (z) = , z ∈ [0, an ]. (4)
x0 − xqn−1
The pair of functions (fn,x0 , gn,x0 ) is called the n-th renormalization pair of the
initial homeomorphism f at the point x0 .
The renormalization behaviour of circle diffeomorphisms has been studied by
Sinai and Khanin in [0] and Stark in [0] and plays a key role in their proof of
smoothness of the correponding conjugacies.
From now on we shall denote by K constants that depend only on the original
map f.
Theorem 1(Sinai and Khanin)(see. [0]). For ε > 0 let f be a C 2+ε (S 1 )- circle
diffeomorphism with irrational rotation number ρf . Then for all n ≥ 1

kfn,x0 (z) − (z + an,x0 )kC 1 ([−1,0]) ≤ Kλnε

Kλ(n−1)ε
kgn,x0 (z) − (z − 1)kC 1 ([0,an,x0 ]) ≤
an−1,x0
where 0 < λ < 1 is constant depending only on f.
We define two linear-fractional functions Fan ,bn and Gan ,bn ,cn as follows:

an + (an + bn mn )z −an cn + (cn − bn mn )z

Fan ,bn (z) := , Gan ,bn ,cn (z) := , (5)
1 + (1 − mn )z an cn + (mn − cn )z

where an was defined in (2) and

(q −1 Z )
xm
f 00(y)
n
X
mn := mn,x0 = exp dy , (6)
m=0 xm +qn−1 2f 0 (y)

respectively
x0 − xqn +qn−1 n
bn := bn,x0 = , cn := σ (−1) . (7)
x0 − xqn−1
Next we take for x0 the break point xb and put from now on

an := an,xb , bn := bn,xb , mn := mn,xb , fn = fn,xb , gn = gn,xb , (8)

respectively
Fn := Fan ,bn , Gn := Gan ,bn ,cn . (9)
The renormalization behaviour of circle diffeomorphisms with one break point
xb was first studied by Khanin and Vul in [0]:
Theorem 2(Khanin and Vul). Let f ∈ C 2+ε (S 1 \{xb }), ε > 0 be a circle homeo-
morphism with break point x0 = xb and irrational rotation number ρf . Then for all
n≥1
kfn (z) − Fn (z)kC 2 ([−1,0]) ≤ Kλnε ,
Kλnε
kgn (z) − Gn (z)kC 2 ([0,an ]) ≤ ,
an
0 < λ < 1 is constant depending only on f.
The proofs of Theorems 1 and 2 are based on the analysis of the dynamical
partitions ξn .
The main purpose of our paper is to study the renormalization behaviour of
circle diffeomorphisms with one break point, but with weaker smoothness properties.
The main results of our present paper are the following Theorems:
Theorem 3Suppose the circle homeomorphism f with one break point x0 = xb
satisfies conditions (a), (b), (c) and its rotation number ρf is irrational. Then for
all n ≥ 1

kfn (z) − Fn (z)kC 1 ([−1,0]) ≤ Kηn , kfn00 (z) − Fn00 (z)kL1 ([−1,0],d`) ≤ Kηn , (10)

kgn (z) − Gn (z)kC 1 ([0,an ]) ≤ Kηn , kgn00 (z) − G00n (z)kL1 ([0,an ],d`) ≤ Kηn , (11)
where the sequence of positive numbers {ηn }∞ belongs to l2 and depends only on
n=1
f.
In the proof of Theorem 3 we use martingales which in circle dynamics have been
used also by Swiatek in [0], and Katznelson and Ornstein in [0].
Using Theorem 2 we can generalize a result of Khanin and Vul (see [0]) for the
coefficients of the linear-fractional functions Fn and Gn :
Theorem 4Suppose that a homeomorphism f satisfies the conditions of Theorem
3. Let an , bn , mn , cn be the parameters in the linear-fractional functions Fn (z) and
Gn (z) defined in (5). Then for all n ≥ 1 the following relations hold

|an + bn mn − cn | ≤ Kan · ηn ,

where the sequence of positive numbers {ηn }∞ n=1 belongs to l2 and depends only on
f.
Following the arguments of Sinai and Khanin (see [0]) in the case of C 2+ε − dif-
feomorphisms, it is not difficult to derive from Theorem 3 in the special case of a
diffeomorphism f, where cn in (7) is identical 1, that |mn − 1| ≤ Kηn from which
follows
Theorem 5Suppose the circle diffeomorphism f satisfies the Katznelson and Orn-
stein smoothness conditions on S 1 and its rotation number ρf is irrational and
x0 ∈ S 1 . Then for all n ≥ 1

kfn,x0 (z) − (z + an,x0 )kC 1 ([−1,0]) ≤ Kηn ,

Kηn
kgn,x0 (z) − (z − 1)kC 1 ([0,an,x0 ]) ≤ ,
an−1,x0
where the sequence of positive numbers {ηn }∞
n=1 belongs to l2 and depends only on
f.
Preliminaries and Notations
The following Lemma plays a key role for studying the metrical properties of
the homeomorphism f :
Lemma 6(see [0]) Let f be a circle homeomorphism with one break point xb with
jump ratio σ and irrational rotation number ρ. Suppose that f ∈ C 1 ([xb , xb + 1])
and var[xb ,xb +1] log f 0 = θ < ∞. Put θ = θ + 2|logσ|. If y0 ∈ S 1 and f i (y0 ) 6=
xb , 0 ≤ i < qn , then
n −1
qY
−θ
e ≤ f 0 (f i y0 ) ≤ eθ . (12)
i=0

Inequality (12) is called the Denjoy inequality. The proof of Lemma 6 is as for
circle diffeomorphisms (see for instance [0]). Recall that the partition ξn+1 (x0 ) is
a refinement of the partition ξn (x0 ): indeed the intervals of order n are elements
(n−1)
of ξn+1 (x0 ) and each interval ∆i (x0 ) ∈ ξn (x0 ) 0 ≤ i < qn , is partitioned into
kn+1 + 1 intervals belonging to ξn+1 (x0 ) such that
kn+1 −1
(n−1) (n+1) (n)
[
∆i (x0 ) = ∆i (x0 ) ∪ ∆i+qn−1 +sqn (x0 ).
s=0

Using Lemma 6 it can be shown, that the intervals of the dynamical partition ξn (x0 )
have exponentially small length. Indeed one finds.
Corollary 1Suppose that the circle homeomorphism f satisfies the conditions of
Lemma 6. Let ∆(n) ∈ ξn (x0 ), ∆(k) ∈ ξk (x0 ) and ∆(n) ⊂ ∆(k) ∈ ξn (x0 ), n − k ≥ 2.
Then
`(∆(n) ) ≤ λn−k−1 `(∆(k) ), (13)
where λ = (1 + eθ )−1/2 < 1.
Corollary 1 implies that `(∆(n) ) ≤ Kλn , for all ∆(n) ∈ ξn (x0 ). Then the trajectory
of every point x ∈ S 1 is dense in S 1 . This together with monotonicity of f implies
that the homeomorphism f is topologically conjugate to the linear rotation fρ (x) =
x + ρ(mod 1). We recall the following definition introduced in [0]:
Definition 7An interval I = (τ, t) ⊂ S 1 is qn -small and its endpoints τ, t are
qn -close, if the intervals f i (I), 0 ≤ i < qn are disjoint.
It is clear that the interval (τ, t) is qn -small if, depending on the parity of n,
either t ≺ τ  f qn−1 (t) ≺ t or f qn−1 (τ )  t ≺ τ ≺ f qn−1 (τ ) in the counter clockwise
order on the circle S 1 .
Lemma 8(see [0]). Suppose that f satisfies the conditions of Lemma 2.1. Let I be
qn -small and m < n, then
qm+1 −1
[
`( f j (I)) ≤ Kλn−m .
j=0

Put Df i (t) :== (f i )0 (t).

Lemma 9(see [0]). Suppose, the homeomorphism f has irrational rotation number
ρf , satisfies the conditions of Lemma 2.1 and τ, t are qn -close. Then for any 0 ≤
l ≤ qn − 1 the following inequality holds:
Df l (x)
e−θ ≤ ≤ eθ . (14)
Df l (y)
Take any qn - close points x, y ∈ S 1 and 0 ≤ l ≤ qn . Denote by I the open interval
with endpoints x and y. Because the intervals f l (I), 0 ≤ l ≤ qn − 1 are disjoint, we
obtain
n −1
qX
|logDf l (x) − logDf l (y)| ≤ |logf (f s (x)) − logf (f s (y))| ≤ θ,
s=0
from which inequality (14) follows immediately.
Following Katznelson and Ornstein [0] we define a sequence of piecewise con-
stant functions Φn : S 1 → R1 , n ≥ 1, using the dynamical partitions ξn (x0 ):
Z
1
Φn (x) := f 00 (y)dy, x ∈ ∆(n) , (15)
`(∆(n) )
∆(n)

where ∆(n) is an interval of the n-th partition ξn (x0 ). Using the Katznelson and
Ornstein smoothness condition (c) it can be easily shown that {Φn (x), n ≥ 1} is a
Lp - bounded martingale with respect to ξn (x0 ).
It is obvious that
Z
Φ0 (x) = Φ0 = f 00 (y)dy = δb ,
S1

for all x ∈ S 1 , where δb = f 0 (xb − 0) − f 0 (xb + 0) is the jump of the function f 0 at

the break point xb .
Define
hn (x) := Φn (x) − Φn−1 (x), n ≥ 1.
One can then prove the following Lemma, used also in [0],
Lemma 10Suppose that the circle homeomorphism f satisfies conditions (a)-(c)
and its rotation number ρf is irrational. Then
∞
(1) f 00 (x) − δb = hn (in L2 − norm);
P
n=1
(2) for any interval ∆(n−1) of the dynamical partition ξn−1 and for all n ≥ 1,
we have Z
hn (x)d` = 0.
∆(n−1)

Denote by kf kp the norm of f in Lp (S 1 , d`), p > 1. In the proof of our main

results we’ll use the following theorem by Katznelson and Ornstein in [0].
Theorem 11Suppose that the homeomorphism f satisfies conditions (a) − (c) and
its rotation number ρf is irrational. Let {(Φn (x), n ≥ 1} be a Lp - bounded martin-
gale w.r.t. the partition ξn . Then the sequence {||hn kp , n ≥ 1} belongs to l2 .
We need later also the following
Lemma 12Let {rn , n ≥ 1} ∈ l2 be a sequnce of positive numbers and let λ ∈ (0, 1)
∞ ∞
λj−n rj , n ≥ 1. Then the series ηn2 converges, i.e.
P P
be a constant. Put ηn =
j=n n=1
the sequence {ηn , n ≥ 1} belongs to l2 .
The assertion of this lemma can be checked easily.
The proof of Theorem 3
For the proof of Theorem 3 we need several Lemmas which we fomulate next.
Their proofs will be given later. Fix x0 to be the break point xb of f and consider
the dynamical partition ξn (xb ). Assuming n to be even one has xqn−1 ≺ x0 ≺ xqn .
(n−1)
For 0 ≤ i ≤ qn we introduce relative coordinates zi in the intervals ∆i =
[xqn−1 +i , xi ], respectively for 0 ≤ j ≤ qn−1 relative coordinates z j in the intervals
(n)
∆j = [xj , xqn +j ] by setting:
x = xi − zi (xi − xqn−1 +i ), respectively y = xj − z j (xj − xqn +j ). (16)
(n−1) (n) (n−1)
The points x ∈ ∆i and y ∈ ∆j are mapped by f to the points f (x) ∈ ∆i+1
(n)
and f (y) ∈ ∆j+1 , with relative coordinates zi+1 and z j+1 , respectively. Next define
(n−1)
αi := xi+qn−1 , γi := xi , βi := f i (x) ∈ ∆i , (17)
 Rβi Rγi
1
f 0 (αi )(βi −αi ) f 00 (y)(y − αi )dy + 1
f 0 (αi )(γi −βi ) f 00 (y)(γi − y)dy
αi βi
Ai := − , (18)
1
Rγi
1+ f 0 (αi )(γi −αi ) f 00 (y)(γ i − y)dy
αi

Zγi n −1
qX
f 00 (y)  1+A z
i i

(qn )
Bi := dy, ψ i (z0 ) := −B i − log , τ (z0 ) := ψi (z0 ).
2f 0 (y) 1 + Ai (zi − 1) i=0
αi
(19)
Analogous we define
qn−1 −1
X
qn−1
τ (z 0 ) := ψi (z 0 ).
i=0

In the following Lemmas we extend analogous results of Sinai and Khanin [0] for
C 2+ε − diffeomorphisms to the class of homeomorphisms satifying the Kazznelson
and Ornstein conditions (b), (c). They show that zqn and z qn−1 are almost linear-
fractional functions of z0 and z 0 , respectively.
Lemma 13Suppose the circle homeomorphism f satisfies the conditions of Theorem
1.3. Then the following relation holds
(qn )
z0 mn eτ (z0 )
zqn = (qn )
. (20)
1 + z0 (mn eτ (z0 )
− 1)

Moreover, for τ (qn ) (z0 ) and its derivatives the following estimates hold

dτ (qn ) (z0 )
max |τ (qn ) (z0 )| ≤ Kηn , max |(z0 − z02 ) | ≤ Kηn , (21)
0≤z0 ≤1 0≤z0 ≤1 dz0

Z1 Z1
dτ (qn ) (z0 ) d2 τ (qn ) (z0 )
| |dz0 ≤ Kηn , |(z0 − z02 )| |dz0 ≤ Kηn , (22)
dz0 dz02
0 0

where the sequence of positive numbers {ηn , n = 1, 2, . . .} belongs to l2 and depends

only on the homeomorphism f.
Define mn in analogy to mn in (6) as follows
qn−1 −1 xZ
i+qn
X f 00 (y)
mn = exp{ dy}.
i=0
2f 0 (y)
xi

Lemma 14For the relative coordinate z qn−1 one has the following relation
qn−1
z 0 mn eτ (z 0 )
z qn−1 = qn−1 . (23)
1 + z 0 (mn eτ (z 0 )
− 1)

Moreover, for τ qn−1 (z 0 ) and its derivatives the following estimates hold

dτ qn−1 (z 0 )
max |τ qn−1 (z 0 )| ≤ Kηn , max |(z 0 − z 20 ) | ≤ Kηn , (24)
0≤z 0 ≤1 0≤z 0 ≤1 dz 0
 Z1 Z1
dτ qn−1 (z 0 ) d2 τ qn−1 (z 0 )
| |dz 0 ≤ Kηn , (z 0 − z 20 )| |dz 0 ≤ Kηn , (25)
dz 0 dz 20
0 0

where the sequence of positive numbers {ηn , n = 1, 2, . . .} belongs to l2 and depends

only on the homeomorphism f.
The estimates (21),(22),(24),(25) show that zqn , respectively z qn are approxi-
mated by linear-fractional functions of z0 , respectively z 0 for large n.
To characterize this approximation more precisely define linear-fractional func-
tions Hqn (z0 ), H qn−1 respectively functions Dqn , Dqn−1 as follows
z0 mn
Hqn (z0 ) := , Dqn (z0 ) := zqn (z0 ) − Hqn (z0 ), (26)
1 + z0 (mn − 1)

z 0 mn
H qn−1 (z 0 ) := , Dqn−1 (z0 ) := z qn−1 (z 0 ) − H qn−1 (z 0 ). (27)
1 + z 0 (mn − 1)
Using relation (20) we rewrite Dqn (z0 ) in the following form

mn z0 (1 − z0 )(exp{τ (qn ) (z0 )} − 1)

Dqn (z0 ) = . (28)
(1 + z0 (mn exp{τ (qn ) (z0 )} − 1))(1 + z0 (mn − 1))

A not too hard calculation shows that

Dq0 n (z0 ) = zq0 n (z0 ) − Hq0 n (z0 ), Dq00n (z0 ) = zq00n (z0 ) − Hq00n (z0 ), (29)

where
mn exp{τn (z0 )}(1 + z0 (1 − z0 )τn0 (z0 )) mn
zq0 n (z0 ) = , Hq0 n (z0 ) =
(1 + z0 (mn exp{τn (z0 )} − 1))2 (1 + z0 (mn − 1))2

mn exp{τn (z0 )}(z0 − z02 )τn00 (z0 )

zq00n (z0 ) = +
(1 + z0 (mn exp{τn (z0 )} − 1))2
2mn exp{τn (z0 )}(1 − z02 − (2z0 − z02 )mn exp{τn (z0 )})τn0 (z0 )
+ +
(1 + z0 (mn exp{τn (z0 )} − 1))3
(1 − 2z0 mn exp{τn (z0 )})(z0 − z02 )(τn0 (z0 ))2
+ −
(1 + z0 (mn exp{τn (z0 )} − 1))3
2mn exp{τn (z0 )}(mn exp{τn (z0 )} − 1)
− ,
(1 + z0 (mn exp{τn (z0 )} − 1))3
−2mn (mn − 1)
Hq00n (z0 ) = . (30)
(1 + z0 (mn − 1))3
It is clear that the function 1 + z0 (mn exp{τn (z0 )} − 1) is bounded and

Z1 Z1
|(z0 − z02 )(τn0 (z0 ))2 |dz0 ≤K |τn0 (z0 )|dz0 .
0 0

Using Lemma 3.1 and equations (28)-(29) we get

kDqn (z0 )kC 1 [0,1] ≤ Kηn , kDq00n (z0 )kL1 ([0,1],d`) ≤ Kηn . (31)
 Proof of Theorem 3. First we express the functions fn and gn in (3), (4)
for x0 the break point xb by using Lemmas 13 and 14 in terms of the renormalized
coordinate z. Recall, that this renormalized coordinate z on the interval Un is
defined by
x = x0 + z(x0 − xqn−1 ).
In this new coordinate the points xqn−1 and x0 take the values -1 respectively 0.
Denote by an and −bn the new coordinates of the points xqn respectively xqn +qn−1 ,
, i.e.
xqn − x0 xq +q − x0
an = , −bn = n n−1 .
x0 − xqn−1 x0 − xqn−1
Notice that x ∈ [xqn−1 , x0 ] has the renormalized coordinate z and the relative
coordinate z0 , and x ∈ [x0 , xqn ] has the two new coordinates z and z 0 . Using these
coordinates it is easy to see that
z
z0 = −z, z 0 = 1 − .
an
The definition of the function fn (z) and relation (20) give
f qn (x) − x0 xqn − x0 xqn − f qn (x) xqn − xqn−1 +qn
fn (z) = = − · =
x0 − xqn−1 x0 − xqn−1 xqn − xqn−1 +qn x0 − xqn−1

(−z)mn exp{τ (qn ) (−z)}

= an − (an + bn )zqn (z0 ) = an − (an + bn ).
1 + (−z)(mn exp{τ (qn ) (−z)} − 1)
Analogously it can be shown, that
z qn−1 z
(1 − an )mn exp{τ (1 − an )}
gn (z) = −b − z (qn ) z
(1 − bn ).
1 + (1 − an )(mn exp{τ (1 − an )} − 1)

If τn (z) := τ (qn ) (−z) and τ n (z) := τ qn−1 (1 − azn ), the last expressions for fn (z)
and gn (z) can be rewritten as:
an + z(an + bn mn exp{τn (z)})
fn (z) = , (32)
1 + z(1 − mn exp{τn (z)} − 1)

−an + z(1 − bn m−1

n exp{τ n (z)})
gn (z) = , (33)
an + z(m−1
n exp{τ n (z)} − 1)
which shows that they are almost Mobius functions for large n.
Since
n −1
qX Zxi qn−1 −1 xZ
i+qn
f 00 (y) X f 00 (y)
mn mn = exp{ 0
dy + dy} =
i=0 x
2f (y) i=0
2f 0 (y)
i+qn−1 xi

n −1 Z
qX
f 00 (y)
= exp{ dy} = exp{ 12 (log f 0 (xb − 0) − log f 0 (xb + 0))} = σ,
i=0
2f 0 (y)
S1

we get m−1
n =σ
−1
mn . This together with the second relation for gn in (33) implies
−an cn + z(cn − bn mn exp{τ n (z)})
gn (z) = . (34)
an cn + z(mn exp{τ n (z)} − cn )
As above, using the assertions of Lemma 14 it can be shown, that
00
kDqn−1 (z 0 )kC 1 [0,1] ≤ Kηn , kDqn−1 (z 0 )kL1 ([0,1],d`) ≤ Kηn . (35)

We can now continue the proof of Theorem 3. First we prove the relation
(10). The functions fn (z) and Fn (z) can be obtained from zqn (z0 ) and Hqn (z0 ) by
z − an
expressing the coordinates z0 and zqn by z, namely z0 = −z and zqn = ,
an + bn
which leads to

fn (z) = an − (an + bn )zqn (−z), Fn (z) = an − (an + bn )Hqn (−z).

and hence using the definition of Dqn

fn (z) − Fn (z) = −(an + bn )Dqn (−z). (36)

Lemma 6 then implies the following estimates

(an + bn ) ≤ K, (1 − bn ) ≤ K, (1 − bn )/an ≤ K.

The last relations together with (28)-(36) imly the bounds (10). Next we prove the
bounds (11). It is again easy to see that gn (z) can be obtained from z qn−1 (z 0 ) by
z z + bn
replacing the coordinates z 0 by 1 − and z qn by which gives, respectively
an 1 − bn
z
gn (z) = −bn − (1 − bn )z qn−1 (1 − ). (37)
an

Next we introduce
z
F n (z) = −bn − (1 − bn )H qn−1 (1 − ). (38)
an

This last equation together with (3.13) implies that

z
|gn (z) − F n (z)| ≤ (1 − bn )|Dqn−1 (1 − )| ≤ Kηn , z ∈ [0, an ],
an

d (1 − bn ) dDqn−1 z
| (gn (z) − F n (z))| ≤ | (1 − )| ≤ Kηn , z ∈ [0, an ], (39)
dz an dz an
Zan Zan
d2 (1 − bn ) d2 Dqn−1 z ηn
| 2 (gn (z) − F n (z))|dz ≤ | (1 − )|dz ≤ K .
dz a2n dz 2 an an
0 0

We rewrite F n (z) in the following form

−an mn + z(mn − bn )
F n (z) = . (40)
an mn + z(1 − mn )

It was shown that m−1n = σ

−1
mn . By substituting this into (40) we get F n (z) =
Gn (z). This, together with (38)-(40) implies the bounds (11). Theorem 3 is then
completely proved.
The proofs of Lemmas 13 and 14
 We’ll prove only Lemma 13, which shows that zqn is almost a Möbus function,
(qn )

i.e. zqn = z0 mn eτ (z0 ) . First we prove the last relation for zqn , then
(qn )
1 + z0 (mn eτ (z0 ) − 1)
we will prove the bounds for τn (z0 ) and τn0 (z0 ) respectively τn00 (z0 ) in (20)-(23). For
the proof of these bounds we will essentially use the expansion
∞
X
f 00 (x) − δb = hn (x) (in L2 − norm), (41)
n=1

where hn = Φn − Φn−1 , {Φn , n ≥ 1} is Lp bounded martingale, and Theorem 11.

Lemma 14 can be proved analogously.
Recall the definition of αi , βi and γi ) in (17) αi := xi+qn−1 , γi := xi , βi := f i (x) ∈
[αi , γi ]. Then one has for the relative coordinates zi and zi+1 of the points f i (x)
and f i+1 (x) in the interval ∆in−1 respectively ∆i+1 n−1

γ i − βi γi+1 − βi+1
zi = , zi+1 = , (42)
γi − αi γi+1 − αi+1

where

Zβi
0
αi+1 = f (αi ), βi+1 = f (βi ) = f (αi ) + f (αi )(βi − αi ) + f 00 (y)(βi − y)dy,
αi

Zγi
0
γi+1 = f (γi ) = f (αi ) + f (αi )(γi − αi ) + f 00 (y)(γi − y)dy.
αi

Using this we rewrite zi+1 as

Rγi Rβi
f 0 (αi )(γi − βi ) + f 00 (y)(γi − y)dy − f 00 (y)(βi − y)dy
αi αi
zi+1 = =
Rγi
f 0 (αi )(γi − αi ) + f 00 (y)(γi − y)dy
αi

Rγi Rβi
 
00 00
(βi − αi ) f (y)(γi − y)dy − (γi − αi ) f (y)(βi − y)dy 
γi − βi 
αi αi
= 1 +  = zi (1+Ai (zi −1)),
 
γi − αi Rγi
f 0 (αi )(γi − αi )(γi − βi ) + (γi − βi ) f 00 (y)(γi − y)dy
 
αi

where Ai was defined in (18) This implies, using relations (19)

1 − zi+1 1 − zi 1 + Ai zi 1 − zi
= · = exp(−Bi ) exp(−ψi ).
zi+1 zi 1 + Ai (zi − 1) zi

Iterating this equation we obtain, using once more (19)

n −1
qX n −1
qX
1 − zqn 1 − z0 1 − z0 1
= exp{− Bi } exp{− ψi } = · (qn ) (z ))
.
zqn z0 i=0 i=0
z0 m n exp(τ 0
(43)
Solving for zqn gives finally equation (20).
Next we estimate τn (z0 ). For this we first estimate Ai in (18). Denote by Vi the
second term in the denominator of Ai . Using condition (c) given in the introduction
and Hölder’s inequality we get

Zγi 1
1 00 kf 00 kp (γi − αi )1+ q 1
|Vi | ≤ 0 |f (y)|(γi −y)dy ≤ 0 ≤ K(γi −αi ) q ,
f (αi )(γi − αi ) f (αi )(γi − αi )(1 + 1/q)
αi

where p1 + 1q = 1.
In analogy it can be shown that the absolute values of both terms of the
numerator of Ai are bounded by K(γi − αi )1/q . Since [αi , γi ] is an interval of the
partition ξn its length is not larger than λn−1 . Hence

|Vi |, |Ai | ≤ Kλn/q , 0 ≤ i ≤ qn . (44)

Next we rewrite τn (z0 ) in the following form:

n −1
qX n −1
qX   n −1
qX n −1
qX
1 + Ai zi
τn (z0 ) = − Bi − log = − log mn − Ai + 0( A2i ).
i=0 i=0
1 + Ai (zi − 1) i=0 i=0
(45)
Using expression (18) for Ai we get the following bounds:

2
Zβi

n −1
qX n −1
qX
1 1
A2i ≤ 2 { 0 f 00 (y)(y − αi )dy  +
i=0 i=0
(1 + Vi )2 f (αi )(βi − αi )
αi

 2
Zγi
1
+ f 00 (y)(γi − y)dy  } ≤ (46)
 
f 0 (αi )(γi − βi )
βi

n −1
qX Zβi Zγi
00 00 00
≤K {kf kp (βi − αi ) 1/q
|f (y)|dy + kf kp (γi − βi ) 1/q
|f 00 (y)|dy} ≤
i=0 αi βi

γi
n −1 Z
qX
≤ K max |γi − αi |1/q |f 00 (y)|dy ≤ Kλn/q .
0≤i≤qn
i=0 α
i

qn
P −1
To estimate the sum Ai , we rewrite it in the following form:
i=0

n −1
qX n −1
qX Zβi
1 1
Ai = { f 00 (y)(y − αi )dy
i=0 i=0
1 + Vi f 0 (αi )(βi − αi )
αi
Zγi
1
+ f 00 (y)(γi − y)dy} =
f 0 (αi )(γi − βi )
βi
 γi Zβi Zβi
 
n −1 Z
qX 00 n −1
qX 00
f (y) 1 f (y) 
=− dy −  f 00 (y)(y − αi )dy − dy −
i=0 α
2f 0 (y) i=0
f 0 (αi )(βi − αi ) 2f 0 (y)
i αi αi
 
n −1
qX Zγi Zγi 00
1 f (y) 
− f 00 (y)(γi − y)dy − dy  + (47)

f 0 (αi )(γi − βi ) 2f 0 (y)

i=0 βi βi

n −1
qX Zβi
Vi 1
+ [ f 00 (y)(y − αi )dy+
i=0
1 + Vi f 0 (αi )(βi − αi )
αi

Zγi
1
f 00 (y)(γi − y)dy].
f 0 (αi )(γi − βi )
βi

The first sum after the second equality sign gives (− log mn ). Since |Vi | ≤
Kλn/q , the absolute value of the last sum is bounded by
 
n −1
qX Zβi Zγi
 1 y − αi 1 γi − y 
Kλn/q  0 f 00 (y) dy + 0 f 00 (y) dy  ≤
i=0
f (α i ) β i − α i f (αi ) γ i − βi
αi βi

βi
n −1 Z
qX
≤ Kλ n/q
|f 00 (y)|dy ≤ Kλn/q .
i=0 α
i

This, together with (45), (46) and (47) implies that

n −1
qX n −1
qX
τn (z0 ) = − log mn − Ai + O( A2i ) =
i=0 i=0

Zβi Zβi
 
n −1
qX 00
1 f (y) 
−  f 00 (y)(y − αi )dy − dy − (48)
i=0
f 0 (αi )(βi − αi ) 2f 0 (y)
αi αi
 
n −1
qX Zγi Zγi 00
1 f (y) 
− f 00 (y)(γi − y)dy − dy  + O(λn/q ) ≡

f 0 (αi )(γi − βi ) 2f 0 (y)

i=0 βi βi

n/q
≡ Sn + S n + O(λ ).
Next we prove the following estimates:

|Sn |, |S n | ≤ Kηn , (49)

where the sequence of positive numbers {ηn , n = 1, 2, ...} belongs to l2 . We estimate

only the sum Sn . The second sum S n can be estimated analogously. We rewrite the
sum Sn in the following form:
β
Zβi 00

n −1
qX Z i 00
Sn =  f (y) y − αi dy − f (y)
dy  +
f 0 (α ) β − α 2f 0 (α )
i=0 i i i i
αi αi
 βi Zy
n −1 Z
qX
f 00 (y)dy
+ f 00 (t)dt ≡ Sn(1) + Sn(2) . (50)
i=0 α
2f 0 (y)f 0 (αi )
i αi

As before, using condition (c) and Hölder’s inequality, we can easily obtain the
bound β 2
n −1
qX Zi
|Sn(2) | ≤ K  |f 00 (y)|dy  ≤
i=0 αi
β 1/p
n −1
qX Zi Zβi
00
≤K p
 |f (y)| dy  (βi − αi )1/q
|f 00 (y)|dy ≤
i=0 αi αi
βi
n −1 Z
qX
≤ Kkf 00 kp max |γi − αi |1/q |f 00 (y)|dy ≤ Kλn/q . (51)
0≤i≤qn
i=0 α
i

(1)
Sn we use the martingale approach going back to the work of Katznelson and
Ornstein [0]. According to the first assertion of Lemma 10,
∞
X
f 00 − δb = hn (in L1 − norm).
n=1

(1)
We rewrite Sn as follows
βi βi
n −1 Z
qX n −1 Z
qX
f 00 (y) y − αi 1 f 00 (y) − δb y − αi 1
Sn(1) = 0
( − )dy = 0
( − )dy+
i=0 α
f (αi ) βi − αi 2 i=0
f (αi ) βi − αi 2
i αi

βi
n −1 Z
qX
δb y − αi 1
+ ( − )dy =
i=0 α
f 0 (αi ) βi − αi 2
i

βi
n −1 Z
qX N
1 X y − αi 1
= 0 (f 00 (y) − δb − hk (y))( − )dy+
f (αi ) i=0 βi − αi 2
αi k=1

βi
n −1 Z X
qX N
1 y − αi 1
+ 0 hk (y)( − )dy. (52)
f (αi ) i=0 βi − α i 2
αi k=1

qn βi
−1 R
δb
( y−αi − 21 )dy vanishes. The absolute value of
P
where we used, that 0 (α ) βi −αi
i=0 αi f i
the sum in (52) with f 00 can be estimated as
βi
n −1 Z
qX N
X Z N
X
K |f 00 (y) − δb − hk (y)|dy ≤ K |f 00 (y) − δb − hk (y)|dy ≤
i=0 α k=1 k=1
i S1

N
X
≤ Kkf 00 − δb − hk k1 .
k=1
 N
Since lim kf 00 − δb −
P
hk k1 = 0 we can choose a sufficiently large number N
N →∞ k=1
such that
N
X
00
kf − δb − hk k1 ≤ Kλn/q .
k=1

Hence, the absolute value of this sum in (52) can be bounded by Kλn/q . Recall that
the point βi = f i (x) belongs to the interval [αi , γi ]. Next choose r0 > 0, minimal
such that for [αi , βi ], one has

∆(n+r0 +1) ⊂ [αi , βi ] ⊂ ∆(n+r0 ) ⊂ [αi , γi ],

where ∆(n+r0 +t) ∈ ξn+t , for t = 0, 1.

We need the following two remarks:
A) one has
y − αi 1 y − a(k−1)
− − Lk,i (y) = ≤ 1.
βi − αi 2 βi − αi
for all n + 1 ≤ k ≤ n + r0 and y ∈ [αi , βi ];
B) applying the assertion of Corollary 1 we get

y − αi 1 y − a(k−1)
− − Lk,i (y) = ≤ λk−n−2 .
βi − α i 2 βi − αi

for all k ≥ n + r0 + 1 and y ∈ [αi , βi ].

To estimate the last sum in (52), we divide the sum in the integrand into three
terms corresponding to summation over 1 ≤ k ≤ n + r0 , n + r0 + 1 ≤ k ≤ n + r0 + 4
and n+r0 +5 ≤ k ≤ N . Consider the first sum. By definition, the piecewise constant
function hk (y) takes constant values on the atoms of the dynamical partition ξk . On
the other hand, when passing from partition ξn to ξn+1 , each interval of rank n is
preserved and the ones of rank n − 1 are divided up into kn+1 intervals of ξn+1 . This
together with [αi , βi ] ⊂ ∆(n+r0 ) ⊂ ∆(n) ∈ ξn implies, that the function hk (y) takes
constant values on the intervals [αi , γi ], i.e. hk ([αi , γi ]) = hk (αi ), i = 0, 1, ..., qn .
Recall, that [αi , βi ] ⊂ [αi , γi ] ∈ ξn . Using these remarks we have
βi βi
n −1 Z n+r
qX X0 n −1
X0 qX
n+r
y − αi y − αi
Z
1 1
hk (y)( − )dy = hk (αi ) ( − )dy.
i=0 α
βi − αi 2 i=0
βi − αi 2
k=1 k=1 αi
i

It is easy to see that

Zβi
y − αi 1
( − )dy = 0,
βi − αi 2
αi

and hence, the last sum is zero. Next we consider the sum where n + r0 + 1 ≤ k ≤
n + r0 + 4. Then we have

n+r n −2
0 +4 qX Zβi   n+r n −2 Z
0 +4 qX
γi
X 1 y − αi 1 X
0
hk (y) − dy| ≤ K |hk (y)|dy ≤
f (αi ) βi − αi 2
k=n+r0 +1 i=1 αi k=n+r0 +1 i=1 α
i

n+r+4
X Z n+r
X 0 +4

≤K |hk (y)| ≤ K khk kp . (53)

k=n+r0 +1S 1 k=n+r0 +1
If k takes finally values in [n + r0 + 5, N ] we denote the corresponding sum by Pn ,
i.e.
βi βi
n −1 Z
qX N N n −1 Z
qX
X y − αi 1 X y − αi 1
Pn = hk (y)( − )dy = hk (y)( − )dy.
i=0 α k=n+r0 +5
βi − αi 2 βi − αi 2
k=n+r5 +5 i=0 α
i i

Because k ≥ n + r0 + 5, each atom [αi , γi ] ∈ ξn is the union of intervals of the

partition ξk−1 . Define a piecewise constant function Lk,i (y) on [αi , γi ] which takes
constant values on the atoms of the partition ξk−1 such that

a(k−1) − αi 1
Lk,i |[a(k−1) , b(k−1) ] = − ,
βi − αi 2

if [a(k−1) , b(k−1) ] ∈ ξk−1 and [a(k−1) , b(k−1) ] ⊂ [αi , γi ]. Then we rewrite the sum
Pn in the following form
βi
N n −1 Z
qX  
X y − αi 1 k,i
Pn = hk (y) − − L (y) dy+ (54)
βi − αi 2
k=n+r0 +5 i=0 α
i

βi
N
X n −1 Z
qX
+ hk (y)Lk,i (y)dy
k=n+r0 +5 i=0 α
i

(1) (2)
Denote by Pn and Pn the two sums in (54) respectively. First we estimate the
(2)
sum Pn . Since [αi , βi ] ⊂ [αi , γi ] ∈ ξn , the interval [αi , βi ] is covered by intervals of
the partition ξk−1 . Denote by ∆ ¯ (k−1) the interval of the partition ξk−1 containing
i
the point βi . If there are two such intervals then we consider the left one. Applying
the second assertion of Lemma 10, we obtain:

N n −1
qX
Lk,i (∆(k−1) )
X X Z
|Pn(2) | ≤ hk (y)dy +
f 0 (αi )
k=n+r0 +5 i=0 ∆(k−1) ⊂[αi ,γi ]
∆(k−1)

N n −1
qX (k−1) ∞
X Lk,i (∆ ¯ )
Z X Z
i
+ 0
|hk (y)|dy ≤ K |hk (y)|dy,
f (αi )
k=n+r0 +5 i=0 k=n+r0 +5U
¯ (k−1)
∆ k
i

−1
qnS
where Uk = ¯ (k−1) ). By Lemma 8 `(Uk ) ≤ λk−n−1 . We have
f i (∆ 0
i=0
∞ ∞ 1 ∞
λk−n
P R P P
|hk (y)|dy ≤ khk kp (`(Uk )) q ≤ K 1 khk kp ≡
k=n+r0 +5 Uk k=n+r0 +5 k=n+r0 +5
Kζn .
where λ1 = λ1/q ∈ (0, 1). Finally,

|Pn(2) | ≤ Kζn . (55)

(1)
Next we estimate the sum Pn in (54). Using the estimates (44) we obtain:

|Pn(1) | ≤ Kδn , (56)

 ∞
λk−n khk kp . Applying Theorem 11 and Lemma 12 it can
P
where δn =
k=n+r0 +5
easily be shown, that δn and ζn belong to l2 . Define ηn := max(ζn , δn ), n ≥ 1. By
collecting equations (51)-(56) we get (49). Relations (48) and (49) then imply the
first estimate in (21).
dτn (z0 )
Next we estimate |(z0 − z02 ) | in (21).
dz0
z0 (1 − z0 )
First we estimate . Put Df i (t) = (f i )0 (t). Using (42) we get
zi (1 − zi )

z0 (1 − z0 ) x0 − x f i (x0 ) − f i (xqn−1 )
= i · ·
zi (1 − zi ) f (x) − f i (x0 ) x0 − xqn−1

x − xqn−1 f i (x0 ) − f i (xqn−1 )

· · =
f i (x) − f i (xqn−1 ) x0 − xqn−1
Df i (t0 ) Df i (y0 )
t0 ))−1 · Df i (t0 ) · (Df i (e
= (Df i (e y0 ))−1 · Df i (y0 ) = · ,
t0 ) Df i (e
Df i (e y0 )
where x ∈ [xqn−1 , x0 ], t0 ∈ [xqn−1 , x0 ], e
t0 ∈ [x, x0 ], y0 ∈ [xqn−1 , x0 ], ye0 ∈ [xqn−1 , x].
Applying Lemma 9 we obtain

Df i (t0 ) Df i (y0 )
e−θ ≤ , ≤ eθ ,
t0 ) Df i (e
Df i (e y0 )

and hence
z0 (1 − z0 )
e−2θ ≤ ≤ e2θ . (57)
zi (1 − zi )
dzi
Using (42) we calculate :
dz0

dzi dzi dβi dx d γi − βi df i d

= · · = ( )· (x) · (x0 − z(x0 − xqn−1 )) =
dz0 dβi dx dz0 dβi γi − αi dx dz0
(n−1)
1 `(∆0 ) Df i (x)
= · (γ0 − α0 ) · Df i (x) = · Df i
(x) = , (58)
γi − αi (n−1)
`(∆i ) Df i (t)
where x, t ∈ [xqn−1 , x0 ]. Applying again Lemma 9 we get

dzi
e−θ ≤ ≤ eθ . (59)
dz0
As seen from their definitions in (18) and (19) the functions Ai and ψi depend
on the variable βi which is linear in the variable zi , therefore Ai , ψi themselves
dψi dAi
depend on zi . The functions and are defined almost everywhere on S 1 . By
dzi dzi
calculating the derivatives of ψi and Ai it is easy to see that

dψi A2i − A0i

= , (60)
dzi (1 + Ai zi )(1 + Ai (zi − 1))
where
dAi dAi dβi dAi
A0i = = = (γi − αi ) ,
dzi dβi dzi dβi
 Rβi Rγi
1
f 0 (αi )(βi −αi )2 f 00 (y)(y − αi )dy − 1
f 0 (αi )(γi −βi )2 f 00 (y)(γi − y)dy
dAi αi βi
= . (61)
dβi 1
Rγi
1+ f 0 (αi )(γi −αi ) f 00 (y)(γi − y)dy
αi

Summing (46), (57), (59) (60) and (61) implies

n −1
qX
dτn (z0 ) dψi dzi n
|(z0 − z02 ) | = |(z0 − z02 ) · | ≤ Kλ q + (62)
dz0 i=0
dzi dz0

n −1
qX Zβi Zγi
γi − αi 00y − αi γi − α i γi − y
+K| (zi − zi2 )[ f (y) dy − f 00 (y) dy]|.
i=0
βi − αi βi − αi γi − β i γ i − βi
αi βi

Denote by En the remaining sum in (62). We have

 
n −1
qX Zβi Zγi
 1 y − αi 1 γi − y 
En = (zi − zi2 )  f 00 (y) dy − f 00 (y) dy  ≤
i=0
1 − zi β i − α i zi γ i − βi
αi βi

βi γi
n −1 Z
qX n −1 Z
qX
00y − αi γi − y
≤K f (y) dy + K f 00 (y) dy ≡ En(1) + En(2) , (63)
i=0 α
βi − αi i=0
γ i − β i
i βi

Using the same arguments as in the proof of the first relation in (21) it can be
shown that
En(1) , En(2) ≤ Kηn .
Together with (62)-(63), this implies the second relation in (21).
Next we pass to the proof of the first bound in (22). As in (62) and using (59)
we have
Z1 Z1 qX
n −1 Z1 qX
n −1
dψi dzi n dAi
|τn0 (z0 )|dz0 = dz0 = O(λ ) + K
q dzi ≤
i=0
dzi dz0 i=0
dzi
0 0 0
 
Z1 qX
n −1 Zβi Zγi
y − αi γi − y n
≤K (γi − αi )  f 00 (y) dy − f 00 (y) dy  dzi +O(λ q ).
 
(βi − αi )2 (γi − βi ) 2
i=0
0 αi βi

γ i − βi
Substituting zi = in the last integral we get
γi − αi

Z1 γi Zβi Zγi
n −1 Z
qX
f 00 (y)(y − αi ) f 00 (y)(γi − y) n
|τn0 (z0 )|dz0 ≤K dy − dy dβi + O(λ q ).
i=0 α
(βi − αi )2 (γi − βi ) 2
0 i αi βi
(64)
Denote by In the sum in (64). We will show that In ≤ Kηn . We have
γi Zβi Zγi
n −1 Z
qX
00 y − αi γi − y
In = | f (y) dy − f 00 (y) dy|dβi =
i=0 α
(βi − αi )2 (γi − βi )2
i αi βi
 γi Zβi Zγi
n −1 Z
qX
00 y − αi γi − y
= | (f (y) − δb ) dy − (f 00 (y) − δb ) dy+
i=0 α
(βi − αi )2 (γi − βi )2
i αi βi

Zβi Zγi
y − αi γi − y
+ δb · dy − δb · dy|dβi =
(βi − αi )2 (γi − βi )2
αi βi

γi Zβi Zγi
n −1 Z
qX
00 y − αi γi − y
= | (f (y) − δb ) dy − (f 00 (y) − δb ) dy|dβi ,
i=0 α
(βi − αi )2 (γi − βi )2
i αi βi

where we use that

Zβi Zγi
y − αi γi − y
dy = dy.
(βi − αi )2 (γi − βi )2
αi βi

Using the first assertion of Lemma 10, we rewrite In in the following form:

γi Zβi
n −1 Z
qX N
!
00
X y − αi
In ≤ | f (y) − δb − hk (y) dy−
i=0 α
(βi − αi )2
αi k=1
i

Zγi N
!
00
X γi − y
− f (y) − δb − hk (y) dy|dβi +
(γi − βi )2
βi k=1

γi Zβi X Zγi X
n −1 Z
qX N N
y − αi γi − y
+ hk (y) dy − hk (y) dy dβi . (65)
i=0 α
(βi − αi )2 (γi − βi )2
i αi k=1 βi k=1

(1) (2)
Denote by In and In the last two sums in (65) respectively. Let us first estimate
(1)
In . Using Hölder’s inequality we obtain:

γi βi
n −1 Z Z
qX N
X y − αi
In(1) ≤ [ f 00 (y) − δb − hk (y) dy+
i=0 α α
(βi − αi )2
k=1
i i

Zγi N
X γi − y
+ f 00 (y) − δb − hk (y) dy]dβi ≤
(γi − βi )2
βi k=1

γi
1/2  β 1/2
Zβi 2

n −1 Z
qX N Zi 2
X (y − αi )
≤ { f 00 (y) − δb − hk (y) dy   dy  +
i=0 α
(βi − αi )4
αi k=1 αi
i

 1/2  1/2
Zγi N 2 γi
2
  (γi − y)
X Z
+  f 00 (y) − δb − hk (y) dy   dy  }dβi ≤
 
(γi − βi )4
βi k=1 βi
 1/2 γ
qn −1 Zγi 2

N Z i 
1 X 00
X 1 1
≤√ f (y) − δb − hk (y) dy  √ +√ dy ≤
3 i=0 k=1
βi − αi γ i − βi
αi αi
1/2
qn −1 Zγi 2

N
4 X X √
≤√ f 00 (y) − δb − hk (y) dy  · γi − αi ≤
3 i=0 k=1
αi
 γi 2
1/2 !1/2
n −1 Z
qX N n −1
qX
4 00
X √
≤√  f (y) − δb − hk (y) dy  ( γi − αi )2 ≤
3 i=0 k=1 i=1
αi
 2
1/2 !1/2
Z N n −1
qX
4 X
≤ √  f 00 (y) − δb − hk (y) dy  |γi − αi | ≤
3 k=1 i=0
S1

N
4 X
≤ √ kf 00 − δb − hk k2 .
3 k=1

N
lim kf 00 − δb −
P
The first assertion of Lemma 10 implies that hk k2 = 0. We
N →+∞ k=1
choose N sufficiently large, such that
N
X n
kf 00 − δb − hk k2 ≤ λ q .
k=1
(1) n
Hence In is bounded by Kλ . q
(2)
Next we estimate the sum In . Recall, that the point βi = f i (x) belongs to
the interval [αi , γi ]. Chose r0 minimal such that for [αi , βi ] one has

∆(n+r0 +1) ⊂ [αi , βi ] ⊂ ∆(n+r0 ) ⊂ [αi , γi ],

where ∆(n+r0 +t) ∈ ξn+t , t = 0, 1.

(2)
To estimate In , we split the integrand into two terms with summations over
1 ≤ k ≤ n + r0 respectively n + r0 + 1 ≤ k ≤ N. Consider first the sums over k
from 1 to n + r0 . The piecewise constant function hk (y) takes constant values on the
atoms of the partition ξk . Since [αi , βi ] ⊂ ∆(n+r0 ) ⊂ ∆(n) ∈ ξn the function hk (y)
takes constant values on the intervals [αi , γi ],1 ≤ k ≤ n + r0 , i.e. hk ([αi , γi ]) =
hk (αi ), i = 0, 1, ..., qn . Recall that [αi , βi ] ⊂ [αi , γi ] ∈ ξn .
After these preparations we have

γi Zβi n+r Zγi n+r

n −1 Z
qX X0 X0
y − αi γi − y
| hk (y) dy − hk (y) dy|dβi =
i=0 α
(βi − αi )2 (γi − βi )2
αi k=1 βi k=1
i

n −1 n+r
qX X0 Zγi Zβi Zγi
y − αi γi − y
= hk (αi ) | dy − dy|dβi = 0, (66)
i=0 k=1
(βi − αi )2 (γi − βi )2
αi αi βi

since the difference of the two integrals in the last sum vanishes.
 Next we consider the sum over n + r0 + 1 ≤ k ≤ N, which we denote by Jn .
Notice, that each interval [αi , γi ] ∈ ξn (x0 ) is the union of a finite number of intervals
of the partition ξk−1 (x0 ). Define piecewise constant functions L e k,i (y) and M
fk,i (y)
on [αi , βi ] respectively on [βi , γi ], which are approximations of the integrands Jn
in the corresponding intervals and take constant values on the atoms of ξk−1 , as
follows
(k−1)
e k,i |[c(k−1) ,d(k−1) ] = c − αi
L ,
βi − αi
if [c(k−1) , d(k−1) ] ∈ ξk−1 (x0 ) and [c(k−1) , d(k−1) ] ⊂ [αi , βi ], respectively
(k−1)
fk,i |[c(k−1) ,d(k−1) ] = γi − d
M
γi − βi

if [c(k−1) , d(k−1) ] ∈ ξk−1 (x0 ) and [c(k−1) , d(k−1) ] ⊂ [βi , γi ].

Then we have
γi Zβi
n −1 Z
qX N  
X y − αi
Jn ≤ | hk (y) − Lk,i (y) dy−
e
i=0 α
(βi − αi )2
i αi k=n+r0 +1

Zγi N  
X γi − y
− hk (y) − Mk,i (y) dy|dβi +
f (67)
(γi − βi )2
βi k=n+r0 +1

γi Zβi Zγi
n −1 Z
qX N
X N
X
+ | e k,i (y)dy −
hk (y)L hk (y)M
fk,i (y)dy|dβi
i=0 α αi k=n+r0 +1 βi k=n+r0 +1
i

(1) (2) (2)

Denote by Jn respectively Jn the two sums in (67) over i. To estimate Jn , we
remember that [αi , βi ] ⊂ [αi , γi ], and the interval [αi , βi ] is covered by intervals of
the partition ξk−1 . In the case βi on the boundary of one of the intervals of the
(2)
ξk−1 the second assertion of Lemma 10 shows that Jn vanishes. If βi lies inside of
(k−1)
some interval of ξk−1 we denote this interval by ∆i whereas those which do not
contain βi are denoted by ∆(k−1) . Applying again the second assertion of Lemma
10, we obtain in the limit N → ∞

∞
X n −1
qX X Z
Jn(2) ≤ | e k,i (∆(k−1) )
L hk (y)dy|+
k=n+r0 +1 i=0 ∆(k−1) ⊂[αi ,βi ] ∆(k−1)

∞ n −1
qX Z
e k,i (∆(k−1)
X
+ L i ) |hk (y)|dy+
k=n+r0 +1 i=0 (k−1)
∆i

∞
X n −1
qX X Z
+ | fk,i (∆(k−1) )
M hk (y)dy|+
k=n+r0 +1 i=0 ∆(k−1) ⊂[βi ,γi ] ∆(k−1)

∞ n −1
qX Z ∞ Z
fk,i (∆(k−1) )
X X
+ M i |hk (y)|dy ≤ K |hk (y)|dy,
k=n+r0 +1 i=0 (k−1) k=n+r0 +1U
∆i k
 qn −1
(k−1)
. By Lemma 8 we have `(Uk ) ≤ λk−n−1 . Therefore
P
where Uk = ∆i
i=0

∞ Z ∞
X X 1
|hk (y)|dy ≤ khk kp (`(Uk )) q ≤
k=n+r0 +1U k=n+r0 +1
k

∞
X
≤K λ1k−n−r0 −1 khk kp ≡ Kγn
k=n+r0 +1
1
where λ1 = λ ∈ (0, 1). By Theorem 11 the sequence {khk kp , k = 1, 2, ...} belongs
q
∞
γn2 converges. Summarizing, we
P
to l2 . Applying Lemma 12 we conclude that
k=1
have shown the estimate
|Jn(2) | ≤ Kγn . (68)
(1)
We pass to the estimate of Jn
in (67). For this we consider the integrand which is
a piecewise linear function. Notice, that the intervals [αi , βi ] and [βi , γi ] are covered
by intervals of the partition ξk−1 if k ≥ n + r0 + 1. If y ∈ [αi , βi ] we get
(k−1)
y − αi e k,i (y) ≤ max |∆ |
2
− L 2
,
(βi − αi ) (βi − αi )

respectively if y ∈ [βi , γi ], then we get

(k−1)
γi − y fk,i (y) ≤ max |∆ |
2
− M 2
, with ∆(k−1) ∈ ξk−1
(γi − βi ) (γi − βi )

from which one derives by using Corollary 1 the estimate

∞
X
|Jn(1) | ≤ K λk−n−r0 −1 kgk kp ≡ Kηn . (69)
k=n+r0 +1

∞
ηk2 converges. Relations (65)-(69) then imply the fol-
P
By Lemma 12 the series
k=1
lowing bound:
|In | ≤ Kηn .
This together with (64) implies the first estimate in (22).
Next we prove the second estimate in (22). For this note, that the function
d2 zi
dz 2
is defined almost everywhere. We can then prove the following estimate for
0
d2 zi
dz02
Z1
d2 zi
| |dz0 ≤ Kkf 00 k1 , (70)
dz02
0

where i = 1, 2, ..., qn − 1 and the constant K > 0 depends only on f . Using the
2
relations x = x0 − z(x0 − xqn−1 ), βi = f i (x) and () we calculate ddzz2i :
0

   
(n−1) i−1 (n−1) i−1
d2 zi d  `(∆0 ) Y 0 d `(∆ ) Y dβi dx
= · f (βj ) =  0
· f 0 (βj ) · · =
dz02 dz0 `(∆(n−1) ) j=0 dβi `(∆(n−1) ) j=0 dx dz0
i i
 (n−1) i−1 00
[`(∆0 )]2 X f (βj )
= · Df i (x) · · Df j (x) =
(n−1)
`(∆i ) j=0
f 0 (βj )
 
i−1
dzi X f 00 (βj ) (n−1)
= · 0 (β )
· Df j (x) · `(∆0 )=
dz0 j=0
f j

 
i−1 (n−1)
dzi X f 00 (βj ) (n−1) `(∆0 )
= · 0 (β )
· `(∆j )· (n−1)
Df j (x) =
dz0 j=0
f j `(∆ )
j
 
i−1 00
dzi  X f (βj ) (n−1) dzj 
= · 0 (β )
· `(∆j )· · .
dz0 j=0
f j dz 0

This together with (59) implies that

 
Z1 Z1 i−1 00
d2 zi X f (β j ) (n−1) 
| 2 |dz0 ≤ e2θ 
0 (β )
· `(∆j ) dzi .
dz0 j=0
f j
0 0

γ i − βi
Substituting zi = in the last integral we obtain
γi − αi

Zγi
 
Z1 i−1 00
d2 zi X f (β j ) (n−1) dβi
| 2 |dz0 ≤ e2θ 
0 (β )
· `(∆j ) ≤
dz0 j=0
f j γ i − αi
0 αi

γj
n −1 Z
qX Z
≤K |f 00 (βj )|dβj ≤ K |f 00 (y)|dy ≡ Kkf 00 k1 .
j=0 α
j S1

On the other hand, by differentiating (60) we obtain:

d2 ψi 2Ai A0i − A00i 2(A0i zi + Ai ) dψi dψi 2

2 = − · −( ) , (71)
dzi (1 + Ai zi )(1 + Ai (zi − 1)) 1 + Ai zi dzi dzi

where
d 2 Ai d2 Ai
A00i = 2 = (γi − αi )2 . (72)
dzi dβi2
By differentiating (61) we obtain:

Rβi 2(f 00 (βi )−f 00 (y))(y−αi ) Rγi 2(f 00 (βi )−f 00 (y))(γi −y)dy
f 0 (α i )(βi −αi )2 dy + f 0 (αi )(γi −βi )2
d 2 Ai αi βi
= . (73)
dβi2 1
Rγi
1+ f 0 (αi )(γi −αi ) f 00 (y)(y − αi )dy
αi

Since
n −1  2
qX
d ψi dzi 2 dψi dzi2

τn00 (z0 ) = ·( ) + ·
i=0
dzi2 dz0 dzi dz02
and using the first relation in (22) together with (57), (70)-(73) we get:

Z1 Z1 n −1
qX
d2 Ai
|(z0 − z02 )τn00 (z1 )|dz0 ≤K (zi − zi2 ) (γi − αi )2 + Kηn ≤
i=1
dβi2
0 0

Z1 n −1 
qX 2 Zβi
γi − αi y − αi
≤K (zi − zi2 ) [f 00 (βi ) − f 00 (y)] dy dz0 + (74)
i=1
βi − αi βi − α i
0 αi

Z1 n −1 
qX 2 Zγi
γi − αi γi − y
+K (zi − zi2 ) [f 00 (βi ) − f 00 (y)] dy dz0 + Kηn .
i=1
βi − αi γ i − βi
0 βi

Applying the same arguments as in the proof of the first relation in (22), we can
show, that (74) implies the second relation (22). Lemma 13 is therefore completely
proved.
Proof of Theorem 4
We will show that the parameters an , bn , mn , cn in Theorem 4 get asymptoti-
cally linearly dependent as n → ∞. Notice that the maps fn and gn are commuting.
Let us assume n to be odd. Define
Zxi Zxi
f 00 (y) f 00 (y)
Θi := dy, Θi := dy, 0 ≤ i ≤ qn−1 .
2f 0 (y) 2f 0 (y)
xi+qn−2 xi+qn

By definition  xi+qn−1 
n −1
 qX Z 00(y)
f 
mn = exp − dy .

i=0
2f 0 (y) 
xi

It is easy to see that

qn−1 −1
f 00 (y)
Z
X 1
exp{ Θi } · m−1
n = exp 0
dy = σ, cn =
i=0
2f (y) σ
S1

P−1
qn−1
and consequently, mn = cn · exp{ Θi }. We need the following Lemma.
i=0
Lemma 15The following relation holds for the numbers mn :

mn = cn (1 + an an−1 (mn − 1)) exp(χn ), (75)

where |χn | ≤ Kan an−1 ηn and the sequence {ηn }∞

n=1 belongs to l2 .
Denote by di the relative coordinate of the point xi+qn ∈ [xi+qn−2 , xi ], i.e.

xi − xi+qn
di = , 0 ≤ i ≤ qn−1 .
xi − xi+qn−2

It is obvious that d0 = an an−1 . Applying Lemma 13 we get

d0 mi exp(τi )
di = , (76)
1 + d0 (mi exp(τi ) − 1)
 i−1
P i−1
P
where m(i) = exp( Θj ), τi = ψj as (18). Because
j=0 j=0

qn−1 −1 n −1
qX
X
mi ≤ exp{ Θj + Θj } ≤ σ, and |τi | ≤ Kηn , ηn ∈ l2 ,
j=0 j=0

(76) implies that |di | ≤ Kd0 , 0 ≤ i ≤ qn−1 .

Using Lemmas (10- 11) it can be show that
qn−1 −1
X n
Θi = di Θi + ϑi , 0 ≤ i ≤ qn−1 , where |ϑi | ≤ Kd0 λ q . (77)
i=0

Actually, by substituting y = di t + xi (1 − di ), t ∈ [xqn−2 +i , xi ] into the definition

of Θi , we have
Zxi Zxi
f 00 (y) f 00 (di t + xi (1 − di ))
Θi = dy = dj · dt = (78)
2f 0 (y) 2f 0 (di t + xi (1 − di ))
xqn +i xqn−2 +i

Zxi Zxi
f 00 (t) f 00 (di t + xi (1 − di )) f 00 (t)
 
= di · dt + di · − 0 dt = di · Θi + ζi ,
2f 0 (t) 0
2f (di t + xi (1 − di )) 2f (t)
xqn−2 +i xqn−2 +i

where
qn−1 −1 qn−1 −1 qn−1 −1
f 00 (di t + xi (1 − di )) f 00 (t)
Z
(n)
X X X
|ζi | ≤ d0 · | 0
− 0
|dt = d 0 · Ri .
i=0 i=0
2f (di t + xi (1 − di )) 2f (t) i=0
(n−2)
∆i

P−1
qn−1
(n)
To estimate the sum Ri , we define as in (15) a sequence of piecewise con-
i=0
stant functions Φ̃n : S 1 → R1 , n ≥ 1,

f 00 (y)
Z
1
Φ̃n (x) := dy, x ∈ ∆(n) ,
`(∆(n) ) f 0 (y)
∆(n)

where ∆(n) is an interval of the n-th partition ξn (x0 ). It is obvious that

Z 00
f (y)
Φ̃0 (x) = dy = log σ 2 .
f 0 (y)
S1

We can also show that

∞
f 00 (y) 2
X
− log σ = hn (in L1 − norm),
f 0 (y) n=1

where
hn (x) := Φ̃n (x) − Φ̃n−1 (x), n ≥ 1.
 P−1
qn−1
(n)
Using this, we rewrite the sum Ri as
i=0

qn−1 −1 qn−1 −1 Z N
!
X (n) 1 X f 00 (di t + xi (1 − di )) 2
X
Ri = | − log σ − hk (y) −
i=0
2 i=0 f 0 (di t + xi (1 − di ))
(n−2) k=1
∆i

N
!
f 00 (t) 2
X
− − log σ − hk (y) |dt ≤
f 0 (t)
k=1

qn−1 −1 N
f 00 (t)
X Z X
≤ | 0
− log σ 2 − hk (y)|dt ≤
i=0
f (t)
(n−2) k=1
∆i

N N
f 00 (t) f 00
Z X X
2 2
≤ | − log σ − hk (y)|dt = k − log σ − hk k1 .
f 0 (t) f0
S1 k=1 k=1

00 N
Since lim k ff 0 − log σ 2 −
P
hk k1 = 0, we can choose the number N sufficiently
n→∞ k=1
large such that
N
f 00 X n
k 0
− log σ 2 − hk k1 ≤ Kλ q .
f
k=1

This together with (78) imply the relation (77).

We have

log (1 + d0 (mi+1 exp(τi+1 ) − 1)) − log (1 + d0 (mi exp(τi ) − 1)) =

Θi d0 mi exp(τi ) Θi d0 mi exp(τi )
= log [1+ + (exp(Θi +τi+1 −τi )−1−Θi )] =
1 + d0 mi (exp(τi ) − 1) 1 + d0 mi (exp(τi ) − 1)
Θi d0 mi exp(τi ) Θi d0 mi exp(τi )
= log [1 + + (exp(Θi + ψi ) − 1 − Θi )] =
1 + d0 mi (exp(τi ) − 1) 1 + d0 mi (exp(τi ) − 1)
Θi d0 mi exp(τi )
= + υi , 0 ≤ i ≤ qn−1 ,
1 + d0 (mi exp(τi ) − 1)
P−1
qn−1 n
where |υi | ≤ Kd0 λ q . This implies that
i=0

qn−1 −1 qn−1 −1 qn−1 −1  

1 X X X Θi d0 mi exp(τi )
log ( mn ) = Θi = (di Θi +ζi ) = + ζi
cn i=0 i=0 i=0
1 + d0 (mi exp(τi ) − 1)

qn−1 −1
X
= (log(1 + d0 (mi+1 exp(τi+1 ) − 1)) − log(1 + d0 (mi exp(τi ) − 1)))+ (79)
i=0

qn−1 −1
X
+ (ζi − υi ) = log (1 + d0 (mn−1 exp(τqn ) − 1)) − log (1 + d0 (m0 exp(τ0 ) − 1))+
i=0
 qn−1 −1
X
+ (ζi − υi ) = log(1 + d0 (mn−1 − 1)) + χn ,
i=0

where |χn | ≤ Kd0 ηn , ηn ∈ l2 . Hence Lemma 15 is proved.

Now we can continue to prove Theorem 4. Put rn := an + bn mn − cn . Using
Lemma 15 it can be shown that

rn = cn an rn−1 + δn , where {δn } ∈ l2 , (80)

where the sequence {δn }∞

n=1 belongs to l2 . Actually, the relation

fn−1 (−an an−1 ) Fa ,b (−an an−1 ) νn

bn = = n−1 n−1 + ,
an−1 an−1 an−1

with νn = fn−1 (−an an−1 ) − Fan−1 ,bn−1 (−an an−1 ), implies that

rn = an + bn mn − cn = an + cn [1 − (an−1 + bn−1 mn−1 an ) exp(χn )]+

 
νn 1
+ − cn = cn − an−1 − bn−1 mn−1 + (81)
an−1 cn
νn mn
+[cn an (an−1 + bn−1 mn−1 ) − cn ](1 − exp(χn )) + = −cn an rn−1 + δn ,
an−1
where νn mn
δn = [cn an (an−1 + bn−1 mn−1 ) − cn ](1 − exp(χn )) + .
an−1
With Theorem 3 we get:

νn mn fn−1 (−an an−1 ) − Fan−1 ,bn−1 (−an an−1 )

| | = mn | | ≤ Kan · ηn ,
an−1 an−1

where {ηn }∞
n=1 belongs to l2 . This implies the estimate (80) for δn .
Iterating relation (80) we get
n
Y n
X n
Y
rn = r1 (−cj aj ) + δi (−cj δj ). (82)
j=2 i=2 j=i+1

From Corollary 1 on the other hand it follows that:

n
Y x0 − xqn
| (−cj aj )| = cn an | | = Kan λn−i ,
j=i
x 0 − x qi

n n
λn−i δi , and since λn−i δi ∈ l2 , we have
P P
where λ ∈ (0, 1). Therefore rn ≤ Kan
i=2 i=2

|rn | ≤ Kan ηn and |an + bn mn − cn | ≤ Kan ηn ,

where the sequence {ηn }∞

n=1 belongs to l2 . Theorem 4 is therefore proved.

Acknowledgement
The work of A. D. was partially supported by the Lower Saxony Professorship of
D. M., financed by the Volkswagenstiftung.
9
(MR0140699) V.I. Arnol’d, Small denominators: I. Mappings from the circle onto
itself, Izv. Akad. Nauk SSSR, Ser. Mat., 25 (1961), 21–86.
(MR) H. Akhadkulov, A. Dzhalilov and D. Mayer, On conjugations of circle home-
omorphisms with two break points. Ergod. Theor. and Dynam. Syst. / FirstView
Article / pp 1-17 DOI: 10.1017/etds.2012.
(MR0832433) P. Cornfeld, S.V. Fomin and Ya.G. Sinai, Ergodic Theory, Springer
Verlag, Berlin 1982.
A. Denjoy, Sur les courbes définies par les équations différentielles à la surface du
tore, J. Math. Pures Appl., 11 (1932), 333–375.
(MR) A.A. Dzhalilov and K.M. Khanin, On invariant measure for homeomorphisms
of a circle with a point of break, Funct. Anal. Appl., 32 (1998), 153–161.
(MR2250801) A.A. Dzhalilov and I. Liousse, Circle homeomorphismswith two break
points, Nonlinearity, 19 (2006), 1951–1968 .
(MR2486582) A.A. Dzhalilov, I. Liousse and D. Mayer, Singular measures of piece-
wise smooth circle homeomorphisms with two break points, Discrete and continuous
dynamical systems, 24 (2009), 381–403.
(MR2593628) A.A. Dzhalilov, H. Akin, S. Temir, Conjugations between circle maps
with a single break point, Journal of Mathematical Analysis and Applications, 366
(2010), 1–10.
(MR2951817) A.A. Dzhalilov, D. Mayer and U.A. Safarov, Piecwise-smooth circle
homeomorphisms with several break points, Izvestiya RAN: Ser. Mat., 76 (2012),
95–112, translation of Izvestiya: Mathematics, 76 (2012), 101–120.
(MR1728375) E. de Faria and W. de Melo, Rigidity of critical circle mappings, I.
J. Eur. Math. Soc. (JEMS),1 (1999), 339–392.
(MR0538680) M. Herman, Sur la conjugaison différentiable des difféomorphismes
du cercle à des rotations, Inst. Hautes Etudes Sci. Publ. Math., 49 (1979), 5–233.
(MR1036902) Y. Katznelson and D. Ornstein, The differentability continuity of the
conjugation of certain diffeomorphisms of the circle, Ergod. Theor. Dyn. Syst., 9
(1989), 643–680.
(MR1036903) Y. Katznelson and D. Ornstein, The absolute continuity of the conju-
gation of certain diffeomorphisms of the circle, Ergod. Theor. Dyn. Syst., 9 (1989),
681–690 .
(MR0997684) K.M. Khanin and Ya.G. Sinai, Smoothness of conjugacies of diffeo-
morphisms of the circle with rotations, Russ. Math. Surv., 44 (1989), 69–99,
translation of Usp. Mat. Nauk, 44 (1989), 57–82.
(MR1118158) K.M. Khanin and E.B. Vul, Circle homeomorphisms with weak dis-
continuities, Advances in Soviet Mathematics, 3 (1993), 57–98.
(MR1969721) K.M. Khanin and D. Khmelev, Renormalizations and Rigidity Theory
for Circle Homeomorphisms with Singularities of the Break Type, Commun. Math.
Phys., 235 (2003), 69–124.
(MR2081026) I. Liousse, PL Homeomorphisms of the circle which are piecewise C 1
conjugate to irrational rotations, Bull. Braz. Math. Soc., 35, (2004), 269–280.
(MR0967471) J. Stark, Smooth conjugacy and renormalization for diffeomorfisms
of the circle, Nonlinearity, 1 (1988), 541–575.
(MR0968483) G. Swiatek, Rational rotation number for maps of the circle, Com-
mun. Math. Phys., textbf119 (1988), 109–122.
(MR1094555) M. Stein, Groups of piecewise linear homeomorphisms, Trans. A.M.S.,
332 (1992), 477–514.
(MR2308853) A.Yu. Teplinskii and K.M. Khanin, Robust rigidity for circle diffeo-
morphisms with singularities, Inventiones mathematicae, 169 (2007), 193–218.
(MR0741080) J.C. Yoccoz Il n’y a pas de contre-exemple de Denjoy analytique, C.
R. Acad. Sci. Paris 298 (1984), 141–144.