Using this characterization of the
ids one can prove the following adaption of the duality of the ids given in [
GJLS97]
.
Theorem 8 ([GJLS97] )
Let
${k}_{V,\omega}^{L}$
and
${k}_{V,\omega}^{H}$
be the integrated density of states of
${H}_{V,\omega ,\phi}$
and
${L}_{V,\omega ,\phi}$
respectively, for some real analytic
$V:\mathbb{T}\to \mathbb{R}$
and nonresonant frequency
$\omega $
. Then
$${k}_{V,\omega}^{L}\left(a\right)={k}_{V,\omega}^{H}\left(a\right)$$
for all
$a\in \mathbb{R}$
.
Proof: Let
$$g(\theta ,n)={\delta}_{n,0}$$
which belongs to
$\mathcal{\mathscr{H}}$
. Then
$Ug=g.$
Moreover by ( 9 ) and the unitary equivalence between
${\stackrel{~}{H}}_{V,\omega}$
and
${\stackrel{~}{L}}_{V,\omega}$
we have that, for any continuous
$f,$
$$\langle g,f\left({\stackrel{~}{H}}_{V,\omega}\right)g{\rangle}_{\mathcal{\mathscr{H}}}=\langle Ug,Uf\left({\stackrel{~}{H}}_{V,\omega}\right)g{\rangle}_{\mathcal{\mathscr{H}}}=\langle Ug,Uf\left({\stackrel{~}{H}}_{V,\omega}\right){U}^{1}Ug{\rangle}_{\mathcal{\mathscr{H}}}=\langle g,f\left({\stackrel{~}{L}}_{V,\omega}\right)g{\rangle}_{\mathcal{\mathscr{H}}}.$$
Therefore, since
${n}_{V,\omega}^{H}$
and
${n}_{V,\omega}^{L}$
are the Borel measures such that
$$\langle g,f\left({\stackrel{~}{H}}_{V,\omega}\right)g{\rangle}_{\mathcal{\mathscr{H}}}=\int f\left(\lambda \right)d{n}_{V,\omega}^{H}\left(\lambda \right)$$
and
$$\langle g,f\left({\stackrel{~}{L}}_{V,\omega}\right)g{\rangle}_{\mathcal{\mathscr{H}}}=\int f\left(\lambda \right)d{n}_{V,\omega}^{L}\left(\lambda \right)$$
for every continuous
$f$
the two measures must coincide (and also their distribution functions,
${k}_{V,\omega}^{L}$
and
${k}_{V,\omega}^{H}$
).
$\square $
Let us end this section summing up some facts useful in the sequel.
Proposition 9
Let
$V$
be real analytic,
$\omega $
nonresonant and
${\mu}_{\phi}$
a spectral measure of
${L}_{V,\omega ,\phi}$
. Assume that there is a measurable set
$A$
such that
$${\mu}_{\phi}\left(A\right)=0$$
for almost every
$\phi \in \mathbb{T}$
. Then
${n}_{V,\omega}^{L}\left(A\right)=0$
and
${n}_{V,\omega}^{H}\left(A\right)=0.$
2.3 The rotation number and Eliasson's theorem revisited
We have seen in the previous section that it is possible to assign an ids for quasiperiodic Schrödinger cocycles using its associated operator. Here we will see that it is possible to define an extension of this object, the fibered rotation number, for more general quasiperiodic cocycles. This object, introduced originally by Herman [
Her83]
in this discrete case (see also Johnson & Moser [
JM82]
, Delyon & Souillard [
DS83b]
), allows us to give a version of Eliasson's theorem for these cocycles. Let us follow the presentation by Krikorian [
Kri]
.
Let
$(A,\omega )$
be a quasiperiodic cocycle on
$SL(2,\mathbb{R})\times \mathbb{T}$
which is homotopic to the identity. That is
$A:\mathbb{T}\to SL(2,\mathbb{R})$
is a continuous map (although we will later assume that it is real analytic) that is homotopic to the identity (for example a Schrödinger cocycle). The fibered rotation number, which we now introduce measures how solutions wind around the origin in
${\mathbb{R}}^{2}$
in average.
Let
${\mathbb{S}}^{1}$
be the set of unit vectors of
${\mathbb{R}}^{2}$
and let us denote by
$p:\mathbb{R}\to {\mathbb{S}}^{1}$
the projection given by the exponential
$p\left(t\right)={e}^{it}$
, identifying
${\mathbb{R}}^{2}$
with
$\mathbb{C}$
. Because of the linear character of the cocycle and the fact that it is homotopic to the identity, the continuous map
$$\begin{array}{c}\begin{array}{cccc}F:& {\mathbb{S}}^{1}\times \mathbb{T}& \mathbb{\u27f6}& {\mathbb{S}}^{1}\times \mathbb{T}\\ & (v,\theta )& \mapsto & \left(\frac{A\left(\theta \right)v}{\parallel A\left(\theta \right)v\parallel},\theta +2\pi \omega \right)\end{array}\end{array}$$ 
(10)

is also homotopic to the identity. Therefore, it admits a continuous lift
$\stackrel{~}{F}:\mathbb{R}\times \mathbb{T}\to \mathbb{R}\times \mathbb{T}$
of the form:
$$\stackrel{~}{F}(t,\theta )=\left(t+f(\theta ,t),\theta +2\pi \omega \right)$$
such that
$$f(t+2\pi ,\theta +2\pi )=f(t,\theta )\text{and}p\left(t+f(t,\theta )\right)=\frac{A\left(\theta \right)p\left(t\right)}{\parallel A\left(\theta \right)p\left(t\right)\parallel}$$
for all
$t\in \mathbb{R}$
and
$\theta \in \mathbb{T}$
. The map
$f$
is independent of the choice of
$\stackrel{~}{F}$
up to the addition of a constant
$2\pi k$
, with
$k\in \mathbb{Z}$
. Since the iteration
$\theta \mapsto \theta +2\pi \omega $
is uniquely ergodic on
$\mathbb{T}$
for all
$(t,\theta )\in \mathbb{R}\times \mathbb{T}$
, one has that the limit
$${lim}_{N\to \infty}\frac{1}{2\pi N}{\sum}_{n=0}^{N1}f\left({\stackrel{~}{F}}^{n}(t,\theta )\right)$$
exists modulus
$\mathbb{Z}$
and it is independent of
$(t,\theta )$
, see Herman [
Her83]
. This object is called the fibered rotation number of
$(A,\omega )$
, and it will be denoted by
$ro{t}_{f}(A,\omega )$
.
The fibered rotation number of a Harperlike equation is defined as the fibered rotation number of the associated Schrödinger cocycle on
$SL(2,\mathbb{R})\times \mathbb{T}$
and will be denoted as
$ro{t}_{f}(a,V,\omega )$
.
The rotation number of a Harperlike equation can be linked to its
ids . Indeed, using a suspension argument (see Johnson [
Joh83]
) it can be seen that
$$ro{t}_{f}(a,V,\omega )=\frac{1}{2}{k}_{V,\omega}\left(a\right)(\text{mod.}\mathbb{Z}).$$
The rotation number is not invariant under conjugation, but one has the following.
Proposition 10 (cf. [Kri] )
Let
$\omega $
be nonresonant and
$({A}_{1},\omega )$
and
$({A}_{2},\omega )$
be two quasiperiodic cocycles on
$SL(2,\mathbb{R})\times \mathbb{T}$
homotopic to the identity. If they are conjugated for some continuous
$Z:\mathbb{T}\to SL(2,\mathbb{R}),$
then
$$ro{t}_{f}({A}_{1},\omega )=ro{t}_{f}({A}_{2},\omega )+\langle \mathbf{k},\omega \rangle \text{modulus}\mathbb{Z},$$
where
$\mathbf{k}\in \mathbb{Z}$
is the degree of the map
$Z:\mathbb{T}\to SL(2,\mathbb{R})$
. If the conjugation
$Z$
is not defined on
$\mathbb{T}$
but on
$(\mathbb{R}/(4\pi \mathbb{Z}\left)\right)$
and it has degree
$\mathbf{k}\in \mathbb{T}$
, then
$$ro{t}_{f}({A}_{1},\omega )=ro{t}_{f}({A}_{2},\omega )+\frac{1}{2}\langle \mathbf{k},\omega \rangle .$$
Keeping in mind this result, we can define two classes of rotation numbers which are preserved under conjugation. An important class is that of resonant rotation numbers. A number of the form
$$\alpha =\frac{1}{2}\langle \mathbf{k},\omega \rangle ,(\text{mod}\frac{1}{2}\mathbb{Z})$$
for some
$k\in \mathbb{Z}$
is called resonant with respect to
$\omega $
. We can also define the class of fibered rotation numbers which are Diophantine with respect to
$\omega $
. Its elements are the numbers
$\alpha $
such that the bound
$$\leftsin\left(\pi \left(2\alpha \langle k,\omega \rangle \right)\right)\right\ge \frac{K}{k{}^{\sigma}},$$
holds for all
$k\in \mathbb{Z}\left\{0\right\}$
and suitable fixed positive constants
$K$
and
$\sigma $
. Both classes of rotation numbers are constant under conjugation.
With these definitions we can give a more precise version of Eliasson's reducibility theorem for general quasiperiodic cocycles on
$SL(2,\mathbb{R})\times \mathbb{T}$
homotopic to the identity. Again the result is valid for more than one frequency, but we restrict ourselves to this onedimensional case.
Theorem 11 ([Eli92] )
Let
$\rho >0$
,
$\omega \in DC(c,\tau )$
be Diophantine and
${A}_{0}$
be a matrix in
$SL(2,\mathbb{R})$
. Then there is a constant
$C=C(c,\tau ,\rho ,{A}_{0}\left\right)$
such that, if
$A\in {C}_{\rho}^{a}(\mathbb{T},SL(2,\mathbb{R}\left)\right)$
is real analytic with
$${\leftA{A}_{0}\right}_{\rho}<C$$
and the rotation number of the cocycle
$(A,\omega )$
is either resonant or Diophantine with respect to
$\omega $
, then
$(A,\omega )$
is reducible to constant coefficients of a quasiperiodic (perhaps with frequency
$\omega /2$
) and analytic transformation.
Remark 12
The proof of this theorem was originally given in [
Eli92]
in the continuous case and for Schrödinger operators (instead of cocycles), although it extends to the setting of Theorem 11 .
Applied to Schrödinger cocycles one obtains the perturbative version of Theorem 1 with the additional characterization of the set of reducible energies in terms of its rotation number. More precisely, the theorem above implies that the set of “reducible” rotation numbers is of full measure in
$\mathbb{T}$
. To obtain a fullmeasure condition on the energies it is necessary to use some facts on the growth of the rotation number at these reducible points which will be also used in Section 3 and which are due to Deift & Simon [
DS83a]
.
3 Proof of Theorem 1
We are now ready to show that Theorem 1 is a direct consequence of the following result by Bourgain & Jitomirskaya [
BJ02b]
, which we restate in a convenient way:
Theorem 13 ([BJ02b] )
Let
$\rho >0$
be a positive number. Then there is a constant
${\varepsilon}_{0}={\varepsilon}_{0}\left(\rho \right)$
such that, for any real analytic
$V\in {C}_{\rho}^{a}(\mathbb{T},\mathbb{R})$
with
$${\leftV\right}_{\rho}<{\varepsilon}_{0},$$
and Diophantine
$\omega $
there is a set
$\Phi \subset \mathbb{T}$
, of zero (Lebesgue) measure such that, if
$\phi \notin \Phi $
, the operator
${L}_{V,\omega ,\phi}$
has pure point spectrum with exponentially decaying eigenfunctions.
Remarks 14

1.
In [BJ02a] , the bound
${\varepsilon}_{0}$
depends on
$\parallel V{\parallel}_{1},$
$\parallel V{\parallel}_{2},$
$\parallel V{\parallel}_{\infty}$
and
$\rho $
. Since
$V$
belongs to
${C}_{\rho}^{a}(\mathbb{T},\mathbb{R})$
, all these previous norms can be controlled by
$V{}_{\rho}$
.

2.
The set
$\Phi $
consists of those phases
$\phi $
for which the relation
$$\begin{array}{c}\leftsin\left(\phi +\pi k\omega \right)\right<exp\left(k{}^{\frac{1}{2\tau}}\right)\end{array}$$ 
(11)

holds for infinitely many values of
$k$
, where
$\omega \in DC(c,\tau )$
. For any Diophantine
$\omega $
, this is a set of zero Lebesgue measure.
Our strategy to prove Theorem 1 will be, first of all, to show that Corollary 3 is a simple consequence of Theorem 13 and the duality of the ids . Then, in Section 3.1 it will be shown that Corollary 3 actually implies Theorem 1 .
Let
$\rho >0$
and
$V$
,
$\omega $
and
$\Phi $
be as in the Theorem 13 . As a consequence of Proposition 9 , the set
$$A={\sigma}^{L}(V,\omega )\backslash {\bigcup}_{\phi \notin \Phi}{\sigma}_{pp}^{L}(V,\omega ,\phi ),$$
where
${\sigma}_{pp}^{L}(V,\omega ,\phi )$
is the set of point eigenvalues of
${L}_{V,\omega ,\phi}$
given by Theorem 13 satisfies that
${n}_{V,\omega}^{L}\left(A\right)=0$
. Indeed, according to Proposition 9 we only need to show that
${\mu}_{\phi}\left(A\right)=0$
for all
$\phi \notin \Phi $
, where
${\mu}_{\phi}$
are the spectral measures of the longrange operators
${L}_{V,\omega ,\phi}$
. This is a consequence of the fact that the spectral measures
${\mu}_{\phi}$
, for
$\phi \notin \Phi $
are supported on the set of point eigenvalues of the corresponding operator.
Therefore also
${n}_{V,\omega}^{H}\left(A\right)=0$
due to Proposition 9 . To prove Corollary 3 it only remains to show that also the Lebesgue measure of
$A$
is zero. To do so, one can invoke Deift & Simon [
DS83a]
. For almost periodic discrete Schrödinger operators they prove that for Lebesgue almost every
$a$
in the set where the Lyapunov exponent is zero, one has the inequality
$$\begin{array}{c}2\pi sin\pi {k}_{V,\omega}^{H}\left(a\right)\frac{d{k}_{V,\omega}^{H}}{da}\ge 1.\end{array}$$ 
(12)

Thus, under the additional assumption that the Lyapunov exponent vanishes in the spectrum, the inequality ( 12 ) implies that if
$A$
is a subset of
${\sigma}^{H}(\omega ,V)$
with
${n}^{H}\left(A\right)=0$
then also the Lebesgue measure of
$A$
is zero.
As a consequence of Bourgain & Jitormirskaya [
BJ02a,
BJ02b]
, for any
$a$
in
${\sigma}^{H}(V,\omega ),$
(with
$V{}_{\rho}<\varepsilon $
) the Lyapunov exponent is zero. Therefore, the set
$A$
has Lebesgue measure zero and for the values of
$a$
in its complement in the spectrum,
$$a\in {\sigma}_{V,\omega}^{H}\backslash A,$$
which is a total measure subset of
${\sigma}^{H}(V,\omega )$
, the corresponding Harperlike equation
$$\begin{array}{c}{x}_{n+1}+{x}_{n1}+V\left(2\pi \omega n\right){x}_{n}=a{x}_{n},n\in \mathbb{Z}\end{array}$$ 
(13)

has an analytic quasiperiodic Bloch wave, using the argument of duality in the beginning of Section 2.1 . Indeed, we saw that if
$a$
is a point eigenvalue of the operator
${L}_{V,\omega ,\phi}$
whose eigenfunction decays exponentially then the Harperlike equation ( 13 ) has an analytic quasiperiodic Bloch wave with Floquet exponent
$\phi $
.
This completes the proof of Corollary
3 .
3.1 From Bloch waves to reducibility
In this section we will see how Corollary 3 (which we proved in the previous section) implies our main result, Theorem 1 . By this corollary we know that if
$V$
,
$\omega $
and
$\Phi $
are as in Theorem 13 then, for almost all
$a\in {\sigma}^{H}(V,\omega )$
, the equation ( 13 ) has an analytic quasiperiodic Bloch wave with Floquet exponent
$\phi \notin \Phi $
. Since we only want to prove a result for almost every
$a$
, it is sufficient to show that if
$\phi \notin \Phi $
is such that
$$\begin{array}{c}\phi \pi k\omega \pi j\ne 0\end{array}$$ 
(14)

for all
$k,j\in \mathbb{Z}$
and ( 13 ) has an analytic quasiperiodic Bloch wave with this Floquet exponent
$\phi $
, then the corresponding Schrödinger cocycle
$({A}_{a,V},\omega )$
is reducible to constant coefficients.
Remark 15
If
$\phi /2\pi $
is resonant with respect to
$\omega $
,
$$\phi =\pi k+\pi j\omega ,$$
for some integers
$k,j$
, then one can also prove reducibility [
Pui04a]
. In Section 4.3 we will consider the case of
$\phi =2\pi k$
which will be used for the Cantor structure of the spectrum.
The existence of a Bloch wave for Equation ( 13 ) implies that the Schrödinger cocycle has the following quasiperiodic solution
$$\begin{array}{c}\left(\begin{array}{c}\stackrel{~}{\psi}(4\pi \omega +\theta )\\ {e}^{i\phi}\stackrel{~}{\psi}(2\pi \omega +\theta )\end{array}\right)={e}^{i\phi}\left(\begin{array}{cc}aV\left(\theta \right)& 1\\ 1& 0\end{array}\right)\left(\begin{array}{c}\stackrel{~}{\psi}(2\pi \omega +\theta )\\ {e}^{i\phi}\stackrel{~}{\psi}\left(\theta \right)\end{array}\right)\end{array}$$ 
(15)

for all
$\theta \in \mathbb{T}$
. Moreover, writing
$$\begin{array}{c}v\left(\theta \right)={\left(\stackrel{~}{\psi}(\theta +2\pi \omega ),{e}^{i\phi}\stackrel{~}{\psi}\left(\theta \right)\right)}^{T}\end{array}$$ 
(16)

and
$$\begin{array}{c}Y\left(\theta \right)=\left(\begin{array}{cc}{v}_{1}\left(\theta \right)& {\overline{v}}_{1}\left(\theta \right)\\ {v}_{2}\left(\theta \right)& {\overline{v}}_{2}\left(\theta \right)\end{array}\right),\end{array}$$ 
(17)

where the bar denotes complex conjugation, one always has the relation
$$\begin{array}{c}{A}_{a,V}\left(\theta \right)Y\left(\theta \right)=Y(\theta +2\pi \omega )\Lambda \left(\phi \right),\end{array}$$ 
(18)

where
$$\Lambda \left(\phi \right)=\left(\begin{array}{cc}{e}^{i\phi}& 0\\ 0& {e}^{i\phi}\end{array}\right).$$
Obviously,
$Y$
will only define a conjugation between the cocycles
$({A}_{a,V},\omega )$
and
$(\Lambda (\phi ),\omega )$
if it is nonsingular. Because of ( 18 ), the determinant of
$Y$
is constant as a function of
$\theta $
and it is purely imaginary. In particular,
$v\left(\theta \right)$
and
$\overline{v}\left(\theta \right)$
are linearly independent for all
$\theta $
if, and only if, they are independent for some
$\theta $
. In the case that
$v$
and
$\overline{v}$
are linearly independent, it is not difficult to prove reducibility to constant coefficients of the cocycle.
Lemma 16
Let
$A:\mathbb{T}\to SL(2,\mathbb{R})$
be a real analytic map and
$\omega $
be nonresonant.
Assume that there is an analytic map
$v:\mathbb{T}\to {\mathbb{R}}^{2}\backslash \left\{0\right\}$
, with
$v$
and
$\overline{v}$
linearly independent, such that
$$v(\theta +2\pi \omega )={e}^{i\phi}A\left(\theta \right)v\left(\theta \right)$$
holds for all
$\theta \in \mathbb{T}$
, where
$\phi \in [0,2\pi )$
. Then the cocycle
$(A,\omega )$
is reducible to constant coefficients by means of a real analytic transformation. Moreover, the Floquet matrix can be chosen to be of the form
$$\begin{array}{c}B=\left(\begin{array}{cc}cos\phi & sin\phi \\ sin\phi & cos\phi \end{array}\right).\end{array}$$ 
(19)

Proof: Let
${Z}^{1}\left(\theta \right)=Y\left(\theta \right)$
as in ( 17 ),
${B}^{1}=\Lambda \left(\phi \right)$
and
$$d\left(\theta \right)={v}_{1}\left(\theta \right){\overline{v}}_{2}\left(\theta \right){\overline{v}}_{1}\left(\theta \right){v}_{2}\left(\theta \right).$$
be the determinant of
${Z}^{1}$
. Therefore
${Z}^{1}$
defines a conjugation between
$({A}_{a,V},\omega )$
and
$({B}^{1},\omega )$
because
$v$
and
$\overline{v}$
are linearly independent,
${Z}^{1}$
is real analytic and, for every
$\theta \in \mathbb{T}$
,
${Z}^{1}\left(\theta \right)$
is nonsingular.
Moreover, from the conjugacy (
18 ) and the nonresonance of
$\omega $
,
$d\left(\theta \right)$
is constant as a function of
$\theta $
. By the linearity of our system, we choose this constant value to be
$i/2$
(recall that, due to the form of
${Z}_{1},$
its determinant must be purely imaginary).
To obtain the real rotation consider the composition
$$Z\left(\theta \right)={Z}^{1}\left(\theta \right){Z}^{2}$$
where
${Z}^{2}$
is the constant matrix
$${Z}^{2}=\left(\begin{array}{cc}1& i\\ 1& i\end{array}\right),$$
Then
$Z$
satisfies the desired conjugation
$$A\left(\theta \right)Z\left(\theta \right)=Z(\theta +2\pi \omega )B$$
being
$B$
the rotation of angle
$\phi $
given by ( 19 ). Thanks to the construction
$Z$
is real and with determinant one.
$\square $
To complete the proof of Theorem 1 , it only remains to rule out the possibility that
$\phi $
satisfies ( 14 ) and
$v$
and
$\overline{v}$
are linearly dependent at the same time. Recall that these two vectors are linearly independent for all
$\theta $
if, and only if, they are linearly independent for some
$\theta .$
Note that both
$v\left(\theta \right)$
and
$\overline{v}\left(\theta \right)$
are different from zero for all
$\theta \in \mathbb{T}$
by construction. Assume that
$v\left(\theta \right)$
and
$\overline{v}\left(\theta \right)$
were linearly dependent for all
$\theta $
. Since these vectors depend analytically on
$\theta $
, there would exist an analytic
$h:\mathbb{T}\to \mathbb{R}$
and an integer
$k\in \mathbb{Z}$
such that
$$\overline{v}\left(t\right)={e}^{i\left(h\left(t\right)+kt\right)}v\left(t\right)$$
for all
$t\in \mathbb{R}$
. Using that
$v$
and
$\overline{v}$
are quasiperiodic solutions of
$(A,\omega )$
, this would imply that
$${e}^{i\left(h\left(t\right)+kt\right)}{e}^{i\phi}={e}^{i\left(h(t+2\pi \omega )+kt+2\pi k\omega \right)}{e}^{i\phi}.$$
Therefore,
$h$
must satisfy the following small divisors equation
$$h(\theta +2\pi \omega )h\left(\theta \right)=2\phi 2\pi k\omega 2\pi j$$
for all
$\theta \in \mathbb{T},$
where
$j$
is some fixed integer. Clearly, such analytic
$h$
cannot exist unless
$$\phi =\pi \left(j+k\omega \right),$$
which is a contradiction with the nonresonance condition ( 14 ). This ends the proof of Theorem 1 .
$\square $
4 Applications
In this section we will prove several consequences of the main theorem which are summarized in theorems 4 and 5 . In Section 4.1 we will present the setting of this section. In 4.2 we adapt MoserPöschel perturbation arguments to the discrete case. This is applied in Section 4.3 to the proof of nonperturbative genericity of Cantor spectrum. Finally, in Section 4.4 we prove that, in our situation, Cantor spectrum implies nonreducibility for a
${G}_{\delta}$
set of energies.
4.1 Reducibility at gap edges
In previous sections, we discussed the reducibility of a quasiperiodic Schrödinger cocycle
$({A}_{a,V},\omega )$
when
$a$
is a point eigenvalue of the dual operator
${L}_{V,\omega ,\phi}$
and
$\phi $
satisfies a nonresonance condition of the form ( 14 ), which was enough to prove the main result. The “resonant” values of
$\phi $
:
$$\begin{array}{c}\phi =\pi j+\pi \omega k,j,k\in \mathbb{Z}\end{array}$$ 
(20)

are particularly important for the description of the spectrum of these operators because the corresponding point eigenvalues lie at endpoints of spectral gaps. Let us prove the reducibility at these endpoints. What follows mimics the proof of the “Ten Martini Problem” given in [
Pui04a]
.
Bourgain & Jitomirskaya [
BJ02a]
also prove that, provided
$\leftV\right(\theta \left)\right<\varepsilon $
and
$\omega $
is Diophantine
${L}_{V,\omega ,\phi}$
has purepoint spectrum with exponentially localized eigenfunctions if
$\phi $
is of the form ( 20 ), see Remark 8.2 after Theorem 7 in [
BJ02a]
.
Taking into account the symmetries of the operators, the study reduces to the four cases
$\phi =0,\pi ,\pi \omega ,\pi \omega +\pi $
. For the sake of simplicity we consider here
$\phi =0$
. As a direct consequence of Aubry duality and BourgainJitomirskaya result for the dual operators
${L}_{V,\omega ,0}$
, the set of purepoint eigenvalues
${\sigma}_{pp}^{L}(V,\omega ,0)$
is a dense subset of
${\sigma}^{H}(V,\omega )$
and any energy
$a\in {\sigma}_{pp}^{L}(V,\omega ,0)$
has a quasiperiodic Bloch wave with
$0$
as a Floquet exponent for the dual eigenvalue equation. Let
$a$
be one of these eigenvalues. The condition above means that there is an analytic map
$\stackrel{~}{\psi}:\mathbb{T}\to \mathbb{C}$
such that
$$x=({x}_{n}{)}_{n\in \mathbb{Z}}={\left(\stackrel{~}{\psi}\left(2\pi \omega n+\theta \right)\right)}_{n\in \mathbb{Z}}$$
is a nonzero solution of
${H}_{V,\omega ,\theta}x=ax$
. Clearly, due to the symmetry of the eigenvalue equation for
${L}_{V,\omega ,0}$
, the function
$\stackrel{~}{\psi}$
can be chosen real analytic. In terms of the cocycle we have that the relation
$$\left(\begin{array}{c}\stackrel{~}{\psi}(4\pi \omega +\theta )\\ \stackrel{~}{\psi}(2\pi \omega +\theta )\end{array}\right)=\left(\begin{array}{cc}aV\left(\theta \right)& 1\\ 1& 0\end{array}\right)\left(\begin{array}{c}\stackrel{~}{\psi}(2\pi \omega +\theta )\\ \stackrel{~}{\psi}\left(\theta \right)\end{array}\right)$$
holds for all
$\theta \in \mathbb{T}$
. Instead of Lemma 16 we now have the following (see [
Pui04a]
for the proof, which is a simple triangularization and averaging argument).
Lemma 17
Let
$A\in {C}_{\delta}^{a}(\mathbb{T},SL(2,\mathbb{R}\left)\right)$
be a real analytic map and
$\omega $
be Diophantine. Assume that there is a nonzero real analytic map
$v\in {C}_{\delta}^{a}(\mathbb{T},{\mathbb{R}}^{2})$
such that the relation
$$v(\theta +2\pi \omega )=A\left(\theta \right)v\left(\theta \right)$$
holds for all
$\theta \in \mathbb{T}.$
Then the quasiperiodic cocycle
$(A,\omega )$
is reducible to constant coefficients by means of a quasiperiodic transformation which is analytic in
$\left\text{Im}\theta \right<\delta $
. Moreover the Floquet matrix can be chosen to be of the form
$$\begin{array}{c}B=\left(\begin{array}{cc}1& c\\ 0& 1\end{array}\right)\end{array}$$ 
(21)

for some
$c\in \mathbb{R}$
.
In the Almost Mathieu case Ince's argument [
Inc44,
Pui04a]
shows that
$c\ne 0$
.
Otherwise the dual model (which is also a Schrödinger operator) would have a point eigenvalue with two linearly independent eigenvectors in
${l}^{2}(\mathbb{Z})$
, and this is a contradiction with the limitpoint character of Schrödinger operators (or just the preservation of the Wronskian in this discrete case). The fact that
$c\ne 0$
is important for the description of the spectrum, because if a Schrödinger cocycle is reducible to a Floquet matrix of the form ( 21 ) with
$c\ne 0$
then the corresponding energy is at the endpoint of an open gap in the spectrum, as it will be seen in the next section.
For general potentials
$V$
, however, we cannot use Ince's argument and it may happen that some of these are collapsed. In fact, there are examples of quasiperiodic Schrödinger operators (with
$V$
small,
$\omega $
Diophantine) for which some
$c$
are zero [
BPS03]
or even do not display Cantor spectrum (see De Concini & Johnson [
DCJ87]
).
Nevertheless, even if
$c$
can be zero, Moser & Pöschel [
MP84]
showed that, in this reducible setting, a closed gap can be opened by means of an arbitrarily small and generic real analytic perturbation of the potential. In the next section we give an adaption of their proof to the discrete case together with some extra properties which will be needed later.
4.2 MoserPöschel perturbation argument
In this section we prove the following adaption of MoserPöschel argument to the discrete case, which deals with cocycles which are perturbations of constant matrices of the form ( 21 ).
Proposition 18
Let
$V$
be real analytic,
$\omega $
Diophantine and
$({A}_{a,V},\omega )$
, for some
$a\in \mathbb{R}$
be a quasiperiodic Schrödinger cocycle. Assume that
$({A}_{a,V},\omega )$
is analytically reducible to the constant coefficients cocycle
$(B,\omega )$
with
$$B=\left(\begin{array}{cc}1& c\\ 0& 1\end{array}\right)$$
for some
$c\in \mathbb{R}$
. Let
$W:\mathbb{T}\to \mathbb{R}$
be real analytic and
$\alpha $
real. If
$Z:\mathbb{T}\to SL(2,\mathbb{R})$
is the real analytic reducing matrix and the conditions
$$\begin{array}{c}c\ne 0\text{and}\left[W{z}_{11}^{2}\right]\ne 0\end{array}$$ 
(22)

or
$$\begin{array}{c}c=0\text{and}[W{z}_{11}{z}_{12}{]}^{2}+[W{z}_{12}^{2}\left]\right[W{z}_{11}^{2}]>0\end{array}$$ 
(23)

are satisfied, then the quasiperiodic cocycle
$({A}_{a,V+\alpha W},\omega )$
has an exponential dichotomy provided
$\left\alpha \right>0$
is small enough and
$$\begin{array}{c}c\alpha \left[W{z}_{11}^{2}\right]<0\text{if (eq:conditioncne022) holds.}\end{array}$$ 
(24)

Moreover in the case
$c\ne 0$
, the Lyapunov exponent of
$({A}_{a,V+\alpha W},\omega ),$
$\gamma (a,V+\alpha W,\omega ),$
and its rotation number,
$ro{t}_{f}(a,V+\alpha W,\omega ),$
satisfy
$$\begin{array}{c}{lim}_{\begin{array}{c}\alpha \to 0,\\ c\alpha \left[W{z}_{11}^{2}\right]>0\end{array}}\frac{\left\gamma \right(\alpha \left)\right}{\alpha {}^{1/2}}=c\left[W{z}_{11}^{2}\right]={lim}_{\begin{array}{c}\alpha \to 0,\\ c\alpha \left[W{z}_{11}^{2}\right]<0\end{array}}\frac{ro{t}_{f}(\alpha )ro{t}_{f}(0\left)\right}{\alpha {}^{1/2}}\end{array}$$ 
(25)

Remark 19
This type of perturbation arguments have been used in a variety of contexts, c.f. Moser & Pöschel [
Mos81,
MP84]
, Johnson [
Joh91]
, Nún͂ez [
Nn95]
, Broer, Puig & Simó [
BPS03]
and Puig & Simó [
PS04]
.
Proof: Since
$Z$
is the reducing transformation of
$({A}_{a,V},\omega )$
to
$(B,\omega )$
, it also renders the perturbed cocycle
$({A}_{a,V+\alpha W},\omega )$
to
$(B+\alpha WP,\omega )$
where
$$P\left(\theta \right)=\left(\begin{array}{cc}{z}_{11}{z}_{12}c{z}_{11}^{2}& c{z}_{11}{z}_{12}+{z}_{12}^{2}\\ {z}_{11}^{2}& {z}_{11}{z}_{12}\end{array}\right).$$
After one step of averaging this cocycle can be analytically conjugated to
$$\left(B+\alpha \left[WP\right]+{\alpha}^{2}{R}_{2},\omega \right)$$
where
$[\cdot ]$
denotes the average of a quasiperiodic function and
${R}_{2}$
depends analytically on
$\alpha $
and
$\theta $
in some open neighbourhoods of
$0$
and
$\mathbb{T}$
. Moreover, a computation shows that
$$B+\alpha \left[WP\right]+{\alpha}^{2}{R}_{2}=exp\left({\stackrel{~}{B}}_{0}+\alpha {\stackrel{~}{B}}_{1}+{\alpha}^{2}{\stackrel{~}{R}}_{2}\right),$$
being
$${\stackrel{~}{B}}_{0}=\left(\begin{array}{cc}0& c\\ 0& 0\end{array}\right),{\stackrel{~}{B}}_{1}=\left(\begin{array}{cc}\left[W{z}_{11}{z}_{12}\right]\frac{c}{2}\left[W{z}_{11}^{2}\right]& c\left[W{z}_{11}{z}_{12}\right]+\left[W{z}_{12}^{2}\right]\\ \left[W{z}_{11}^{2}\right]& \left[W{z}_{11}{z}_{12}\right]+\frac{c}{2}\left[W{z}_{11}^{2}\right]\end{array}\right)$$
and
${\stackrel{~}{R}}_{2}\in sl(2,\mathbb{R})$
depending analytically on
$\alpha $
and
$\theta $
. Let
$$D=\left(\begin{array}{cc}{d}_{1}& {d}_{2}\\ {d}_{3}& {d}_{1}\end{array}\right)={\stackrel{~}{B}}_{0}+\alpha {\stackrel{~}{B}}_{1},$$
whose determinant is
$d={d}_{1}^{2}{d}_{2}{d}_{3}$
. Now let us distinguish between the cases
$c\ne 0$
and
$c=0$
.
If (
22 ) holds then the expression for the determinant becomes
$$d=c\alpha \left[W{z}_{11}^{2}\right]+O\left({\alpha}^{2}\right)$$
so that it is negative if, in addition ( 24 ) holds. In this case, the matrix
$$Q=\left(\begin{array}{cc}{d}_{2}& {d}_{2}\\ {d}_{1}+\sqrt{d}& {d}_{1}\sqrt{d}\end{array}\right),$$
which is welldefined, has determinant
$$2c\sqrt{c\alpha \left[W{z}_{11}^{2}\right]}+O\left(\alpha \right).$$
and satisfies
$DQ=Q\Delta ,$
where
$$\Delta =\left(\begin{array}{cc}\sqrt{d}& 0\\ 0& \sqrt{d}.\end{array}\right).$$
Therefore the change of variables defined by
$Q$
transforms the cocycle
$$\begin{array}{c}\left(exp\left({\stackrel{~}{B}}_{0}+\alpha {\stackrel{~}{B}}_{1}+{\alpha}^{2}{\stackrel{~}{R}}_{2}\right),\omega \right)\end{array}$$ 
(26)

into
$$\begin{array}{c}\left(exp\left(\left(\Delta +{\stackrel{~}{S}}_{2}\right)\right),\omega \right)\end{array}$$ 
(27)

where
$${\stackrel{~}{S}}_{2}(\alpha ,\theta )={\alpha}^{2}{Q}^{1}{R}_{2}(\alpha ,\theta )Q$$
which is
$O\left(\right\alpha {}^{3/2})$
uniformly in
$\theta $
. Note that
$$\Delta +{\stackrel{~}{S}}_{2}=\sqrt{c\alpha \left[W{z}_{11}^{2}\right]}\left(\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)+O\left(\right\alpha \left\right)\right).$$
so that if ( 24 ) holds and
$\left\alpha \right>0$
is small enough the cocycle
$({A}_{a,V+\alpha W},\omega )$
has an exponential dichotomy and the Lyapunov exponent satisfies
$$\frac{1}{\sqrt{\left\alpha \right}}\left\gamma \left({A}_{a,V+\alpha W},\omega \right)\right\to \sqrt{\leftc\right[W{z}_{11}^{2}\left]\right},$$
see Coppel [
Cop78]
. To obtain the asymptotics of the rotation number, we can consider the transformation
$Q$
defined for
$c\alpha \left[W{z}_{11}^{2}\right]>0$
. This, although complex, is a welldefined conjugation between
$D$
and
$\Delta $
, which is now a complex rotation of angle
$\sqrt{\leftd\right}.$
Therefore ( 26 ) is conjugated to ( 27 ), a perturbation of a complex rotation. Using the definition of the fibered rotation number given in Section 2.3 the result follows.
Let us now consider the situation when (
23 ) holds. In this case, the matrix
$D$
becomes
$$D=\alpha \left(\begin{array}{cc}\left[W{z}_{11}{z}_{12}\right]& \left[W{z}_{12}^{2}\right]\\ \left[W{z}_{11}^{2}\right]& \left[W{z}_{11}{z}_{12}\right]\end{array}\right)=:\alpha \stackrel{~}{D}$$
Condition ( 23 ) is equivalent to the hyperbolicity of
$\stackrel{~}{D}$
whose determinant is
$$\stackrel{~}{d}=[W{z}_{11}{z}_{12}{]}^{2}+[W{z}_{12}^{2}\left]\right[W{z}_{11}^{2}].$$
Therefore there is a change of variables
$Q$
, independent of
$\alpha $
and
$\theta ,$
which renders it to a diagonal form
$\stackrel{~}{\Delta}$
with
$\sqrt{d}$
and
$\sqrt{d}$
as diagonal entries. This conjugation transforms the cocycle
$$\left(exp\left(\alpha \stackrel{~}{D}+{\alpha}^{2}{\stackrel{~}{R}}_{2}\right),\omega \right)$$
into
$$\left(exp\left(\left(\alpha \stackrel{~}{\Delta}+{\alpha}^{2}{\stackrel{~}{S}}_{2}\right)\right),\omega \right)$$
where
$${\stackrel{~}{S}}_{2}(\alpha ,\theta )={Q}^{1}{\stackrel{~}{R}}_{2}(\alpha ,\theta )Q.$$
Since,
$$\alpha \Delta +{\alpha}^{2}{\stackrel{~}{S}}_{2}=\alpha \sqrt{\stackrel{~}{d}}\left(\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)+\frac{\alpha}{\sqrt{\stackrel{~}{d}}}{\stackrel{~}{S}}_{2}\right).$$
the cocycle
$({A}_{a,V+\alpha W},\omega )$
has an exponential dichotomy when
$\alpha \ne 0$
is small enough (see, again Coppel [
Cop78]
).
$\square $
The perturbation argument in the previous proposition can applied to the reducible cocycles at endpoints of gaps as we do next.
Corollary 20
Let
$V$
,
$a$
and
$\omega $
as in Proposition 18 and assume that
$c\ne 0$
.
Then
$a$
is at the endpoint of a noncollapsed spectral gap
$I$
of
$\sigma (V,\omega )$
(the right one if
$c>0$
and the left one if
$c<0$
). Moreover, the limits
$$\begin{array}{c}{lim}_{\begin{array}{c}\alpha \to 0,\\ a+\alpha \in I\end{array}}\frac{\gamma (a+\alpha ,V,\omega )}{\sqrt{\left\alpha \right}}={lim}_{\begin{array}{c}\alpha \to 0,\\ a+\alpha \in I\end{array}}\frac{\leftro{t}_{f}(a+\alpha ,V,\omega )ro{t}_{f}(a,V,\omega )\right}{\sqrt{\left\alpha \right}}\end{array}$$ 
(28)

exist and are different from zero.
Proof: Take
$W=1$
in Proposition 18 . Then, the cocycle
$({A}_{a+\alpha ,V},\omega )$
has an exponential dichotomy if
$c\alpha <0$
and
$\left\alpha \right$
is small enough. This means that there is an open spectral gap besides
$a$
(to the left if
$c>0$
and to the right otherwise).
Moreover the asymptotics of formula (
25 ) imply ( 28 ).
$\square $
Finally we consider the variation of the rotation number in the case
$c=0$
in a more general setting which will be needed in the next section.
Proposition 21
Let
$V$
be continuous and
$\omega $
nonresonant. Assume that the Schrödinger cocycle
$({A}_{{a}_{0},V},\omega )$
is reducible to the cocycle
$(B,\omega )$
with
$B\in SO(2,\mathbb{R})$
a constant matrix. Then the map
$$a\in \mathbb{R}\mathbb{\mapsto}ro{t}_{f}(a,V,\omega )$$
is differentiable at
${a}_{0}.$
Proof: Let
$\rho $
be the angle of the rotation,
$$B=\left(\begin{array}{cc}cos\rho & sin\rho \\ sin\rho & cos\rho \end{array}\right).$$
The cocycle
$({A}_{a,V},\omega )$
is conjugated to
$(B+\alpha R,\omega )$
where
$$R\left(\theta \right)=Z(\theta +2\pi \omega {)}^{1}\left(\begin{array}{cc}1& 0\\ 0& 0\end{array}\right)Z(\theta )$$
and
$\alpha =a{a}_{0}$
. The cocycle
$(B+\alpha R,\omega )$
induces a lift
$\stackrel{~}{F}$
from
$\mathbb{R}\times \mathbb{T}$
to itself of the form
$$\stackrel{~}{F}(t,\theta )=\left(t+\rho +\alpha f(t,\theta ,\alpha ),\theta +2\pi \omega \right),$$
where
$f$
is continuous and
$2\pi $
periodic in both
$t$
and
$\theta $
. Therefore,
$$ro{t}_{f}(B+\alpha R,\omega )ro{t}_{f}(B,\omega )={lim}_{N\to \infty}\frac{1}{2\pi N}{\sum}_{n=0}^{N1}\left(\alpha f\left({\stackrel{~}{F}}^{n}(t,\theta )\right)\right)=O\left(\alpha \right)$$
as we wanted to show.
$\square $
Remarks 22

1.
A computation shows that the derivative of the rotation number above is nonzero. In particular, when a Schrödinger cocycle is reducible to the identity, the corresponding energy lies at the endpoint of a collapsed gap.

2.
Similar results have been obtained when
$Z,{Z}^{1}:\mathbb{T}\to SL(2,\mathbb{R})$
are square integrable and
$B:\mathbb{T}\to SO(2,\mathbb{R})$
is measurable, compare with Moser [Mos81] and Deift & Simon [DS83a] .
4.3 Genericity of Cantor spectrum
In the previous section we have seen that if a Schrödinger cocycle is reducible to a matrix with trace
$2$
then the corresponding energy is at the endpoint of a spectral gap which is collapsed if the Floquet matrix is the identity. The next consequence of Proposition 18 is that when the Floquet matrix is the identity (a similar statement holds for minus the identity) one can “open up” the collapsed gap by means of a generic perturbation.
Corollary 23
Let
$V$
,
$a$
,
$\omega $
and
$Z$
be as in Proposition 18 and assume that
$c=0$
. If
$W$
is a generic real analytic potential then for
$\left\beta \right\ne 0$
small enough the spectrum
$\sigma (V+\beta W,\omega )$
has an open spectral gap with ids
$k(a,V,\omega )$
.
Proof: In Proposition 18 we proved that for a perturbation
$\stackrel{~}{W}$
satisfying ( 23 )
$$[\stackrel{~}{W}{z}_{11}{z}_{12}{]}^{2}+[\stackrel{~}{W}{z}_{12}^{2}\left]\right[\stackrel{~}{W}{z}_{11}^{2}]<0$$
the cocycle
$({A}_{a,V+\beta \stackrel{~}{W}},\omega )$
has an exponential dichotomy if
$\left\beta \right>0$
small enough.
This means that
$a$
lies in a spectral gap of
$\sigma (V+\beta \stackrel{~}{W},\omega )$
which, by continuity must satisfy that
$$k(a,V+\beta \stackrel{~}{W},\omega )=k(a,V,\omega )$$
for
$\left\beta \right$
small enough. Let us now show that if
$W$
is a generic potential, then for every
$\left\beta \right\ne 0$
small enough there is a value of
$\alpha $
such that
$a+\alpha $
lies in a spectral gap of
$\sigma (V+\beta W,\omega )$
with
$$k(a+\alpha ,V+\beta W,\omega )=k(a,V,\omega ).$$
Note that the condition ( 23 ) can be rewritten as
$$[\stackrel{~}{W}{y}_{1}{]}^{2}[\stackrel{~}{W}{y}_{2}{]}^{2}[\stackrel{~}{W}{y}_{3}{]}^{2}<0$$
where
$${y}_{1}=\frac{1}{2}\left({z}_{11}^{2}+{z}_{12}^{2}\right),{y}_{2}=\frac{1}{2}\left({z}_{11}^{2}{z}_{12}^{2}\right),{y}_{3}={z}_{11}{z}_{12}.$$
Let
$\alpha $
be such that
$$\left[\right(\alpha +W\left){y}_{1}\right]=\alpha \left[{y}_{1}\right]+\left[W{y}_{1}\right]=0,$$
(this determines
$\alpha $
since
$\left[{y}_{1}\right]\ne 0$
). Then the shifted perturbation
$\alpha +W$
satisfies condition ( 23 ) unless
$$\left[W{y}_{1}\right]\left[{y}_{2}\right]+\left[W{y}_{2}\right]\left[{y}_{1}\right]=0\text{and}\left[W{y}_{1}\right]\left[{y}_{3}\right]+\left[W{y}_{3}\right]\left[{y}_{1}\right]=0,$$
which is clearly a generic condition. Then, if
$\left\beta \right>0$
is small enough, the spectrum
$\sigma (V+\beta W,\omega )$
has an open gap with
$$k(a+\alpha \beta ,V+\beta W,\omega )=k(a,V,\omega )$$
as we wanted to show.
$\square $
Remark 24
As Moser & Pöschel show, when
$c=0$
it is always possible to choose the reducing transformation such that
$\left[{z}_{11}^{2}\right]=\left[{z}_{12}^{2}\right]=1$
and
$\left[{z}_{11}{z}_{12}\right]=0$
so that
$\left[{y}_{1}\right]=1$
,
$\left[{y}_{2}\right]=0$
,
$\left[{y}_{3}\right]=0$
and the generic
$W$
must satisfy
$$\left[W\right({z}_{11}+{z}_{12}{)}^{2}]\ne 0\text{or}[W({z}_{11}{z}_{12}{)}^{2}]\ne 0.$$
Let us now summarize the situation. Using the two past sections we have seen that if
$V$
is a real analytic potential on
${C}_{\rho}^{a}(\mathbb{T},\mathbb{R})$
, with
$V{}_{\rho}<\varepsilon $
and
$\omega $
is Diophantine, there is a countable dense subset of energies in the spectrum where the system is reducible to a Floquet matrix with trace
$2$
. These lie at endpoints of gaps. Although these can be collapsed, a generic and arbitrarily small perturbation opens them as Corollary 23 says. Since there is a countable number of gaps Theorem 4 follows.
4.4 Cantor spectrum implies nonreducibility
In [
Eli92]
it was seen that, for a generic real analytic Schrödinger cocycle (with Diophantine frequencies) besides the almost everywhere reducibility there was a set of zero measure of energies for which the cocycle was not reducible to constant coefficients. The proof relies on the kam procedure developed there, but the Cantor structure of the spectrum is seen to play a key role. In this section we prove irreducibility for a
${G}_{\delta}$
set of energies assuming only Cantor structure of the spectrum and Theorem 1 . This argument is reminiscent of some techniques in circle maps, see Arnol'd [
Arn61]
. We state here a slightly more general version than that of Theorem 5 . More applications will be given elsewhere.
Theorem 25
Let
$\rho >0$
. There is a constant
$\varepsilon >0$
such that if
$V\in {C}_{\rho}^{a}(\mathbb{T},\mathbb{R})$
is real analytic with
$V{}_{\rho}<\varepsilon $
,
$\omega $
is Diophantine and
$I$
is an open interval such that
$$K=\sigma (V,\omega )\cap \overline{I}$$
is a nonvoid Cantor set, then there is a
${G}_{\delta}$
dense set of energies in
$K$
for which the corresponding Schrödinger cocycle is not reducible to constant coefficients by means of a continuous transformation.
Proof: Consider, for any
${a}_{1},{a}_{2}\in K$
with
${a}_{1}\ne {a}_{2}$
,
$$\delta ({a}_{1},{a}_{2})=\left\frac{k({a}_{1},V,\omega )k({a}_{2},V,\omega )}{{a}_{1}{a}_{2}}\right.$$
Now, for any
$a\in \mathbb{R}$
we can define
$$m\left(a\right)={sup}_{\lambda \ne a,\lambda \in K}\delta (a,\lambda )$$
which is either a positive real number or
$+\infty .$
If
$a\in \sigma (V,\omega )$
is reducible to constant coefficients then we have two situations.
Either the Floquet matrix
$B$
has trace
$\pm 2$
, in which case
$m\left(a\right)=\infty $
(see Corollary 20 ) or
$B\in SO(2,\mathbb{R})$
and then
$m\left(a\right)<\infty $
(see Proposition 21 ). Due to the fact that
$V{}_{\rho}<\varepsilon ,$
$\omega $
is Diophantine and the Cantor structure of the spectrum there is a dense set of endpoints of gaps,
${\mathcal{G}}_{K}$
, where the system is reducible to constant coefficients because of Eliasson's Theorem 11 .
We will show that the set where
$m\left(a\right)=\infty $
is a
${G}_{\delta}$
dense subset of
$K$
.
Excluding the endpoints of gaps where there is reducibility to a Floquet matrix with trace
$\pm 2$
(which are at most countable) we will still have a
${G}_{\delta}$
dense subset of energies in
$K$
whose corresponding cocycle cannot be reducible to constant coefficients.
Let, for any
$n\in \mathbb{N}\cup \left\{0\right\}$
and
${a}_{0}\in K$
,
$$U({a}_{0},n)=\left\{a\in K;\delta (a,{a}_{0})>n\right\}$$
and
$$U\left(n\right)={\bigcup}_{{a}_{0}\in K}U({a}_{0},n).$$
The sets
$U\left(n\right)$
are open in
$K$
because of the continuity of the rotation number.
Moreover they are dense in
$K$
because they contain
${\mathcal{G}}_{K}$
, which is dense in
$K$
.
Therefore
$$U(\infty ){\bigcap}_{n>0}U\left(n\right)=\left\{a\in K;m\left(a\right)=\infty \right\},$$
is a
${G}_{\delta}$
dense subset of
$K$
. If we exclude the endpoints of open gaps the remaining energies, which still form a
${G}_{\delta}$
dense subset of
$K$
, cannot be reducible to constant coefficients by means of a continuous transformation. This proves 25 and also 5 .
$\square $
Acknowledgements The author wishes to thank H. Eliasson, S. Jitomirskaya, C. Simó and B. Simon for many ideas which have been decisive in this paper. The paper arose in a stay at Caltech Mathematics Department and completed mostly in the Departament de Matemàtica Aplicada i Anàlisi at the Universitat de Barcelona as part of his PhD thesis. I am grateful to both institutions. This work has been supported by grants DGICYT BFM200309504C0201 (Spain) and CIRIT 2001 SGR70 (Catalonia).
References

S. Aubry and G. André. Analyticity breaking and Anderson localization in incommensurate lattices. In Group theoretical methods in physics (Proc. Eighth Internat. Colloq., Kiryat Anavim, 1979), pages 133–164. Hilger, Bristol, 1980.

A. Avila and S. Jitomirskaya. The Ten Martini Problem. In preparation, 2005.

A. Avila and R. Krikorian. Reducibility or nonuniform hyperbolicity for quasiperiodic Schrödinger cocycles. Preprint, 2003.

V. I. Arnol
${}^{\prime}$
d. Small denominators. I. Mapping the circle onto itself. Izv. Akad. Nauk SSSR Ser. Mat., 25:21–86, 1961.

J. Avron and B. Simon. Almost periodic Schrödinger operators II. The integrated density of states. Duke Math. J., 50:369–391, 1983.

J. Bourgain and S. Jitomirskaya. Absolutely continuous spectrum for 1D quasiperiodic operators. Invent. Math., 148(3):453–463, 2002.

J. Bourgain and S. Jitomirskaya. Continuity of the Lyapunov exponent for quasiperiodic operators with analytic potential. J. Statist. Phys., 108(56):1203–1218, 2002. Dedicated to David Ruelle and Yasha Sinai on the occasion of their 65th birthdays.

J. Bourgain. On the spectrum of lattice Schrödinger operators with deterministic potential. II. J. Anal. Math., 88:221–254, 2002. Dedicated to the memory of Tom Wolff.

J. Bourgain. Green's function estimates for lattice Schrdinger operators and applications. Princeton University Press, Princeton, N. J., 2004. Annals of Mathematics Studies, to appear.

J. Bourgain. On quasiperiodic lattice Schrödinger operators. Discrete Contin. Dyn. Syst., 10(12):75–88, 2004.

H. W. Broer, J. Puig, and C. Simó. Resonance tongues and instability pockets in the quasiperiodic HillSchrödinger equation. Comm. Math. Phys, 241(2–3):467–503, 2003.

V. Chulaevsky and F. Delyon. Purely absolutely continuous spectrum for almost Mathieu operators. J. Statist. Phys., 55(56):1279–1284, 1989.

V. A. Chulaevsky. Almost periodic operators and related nonlinear integrable systems. Nonlinear Science: Theory and Applications. Manchester University Press, Manchester, 1989. With a foreword by Ya. G. Sinaĭ, Translated from the Russian.

W. A. Coppel. Dichotomies in stability theory. SpringerVerlag, Berlin, 1978. Lecture Notes in Mathematics, Vol. 629.

C. De Concini and R. A. Johnson. The algebraicgeometric AKNS potentials. Ergodic Theory Dynam. Systems, 7(1):1–24, 1987.

S. Datta, T. Jäger, G. Keller, and R. Ramaswamy. On the dynamics of the critical harper map. Nonlinearity, 17(6):2315–2323, 2004.

E. I. Dinaburg and Y.G. Sinai. The onedimensional Schrödinger equation with quasiperiodic potential. Funkt. Anal. i. Priloz., 9:8–21, 1975.

P. Deift and B. Simon. Almost periodic Schrödinger operators III. The absolute continuous spectrum. Comm. Math. Phys., 90:389–341, 1983.

F. Delyon and B. Souillard. The rotation number for finite difference operators and its properties. Comm. Math. Phys., 89(3):415–426, 1983.

L. H. Eliasson. Floquet solutions for the onedimensional quasiperiodic Schrödinger equation. Comm. Math. Phys., 146:447–482, 1992.

A. Y. Gordon, S. Jitomirskaya, Y. Last, and B. Simon. Duality and singular continuous spectrum in the almost Mathieu equation. Acta Math., 178(2):169–183, 1997.

M. R. Herman. Une méthode pour minorer les exposants de Lyapunov et quelques exemples montrant le caractère local d'un théorème d'Arnold et de Moser sur le tore de dimension 2. Comment. Math. Helvetici, 58(3), 1983.

E. L. Ince. Ordinary Differential Equations. Dover Publications, New York, 1944.

C. Janot. Quasicrystals : a primer. Clarendon Press, 1992.

S. Ya. Jitomirskaya. Almost everything about the almost Mathieu operator. II. In XIth International Congress of Mathematical Physics (Paris, 1994), pages 373–382. Internat. Press, Cambridge, MA, 1995.

S. Jitomirskaya. Metalinsulator transition for the almost Mathieu operator. Ann. of Math. (2), 150(3):1159–1175, 1999.

S. Jitomirskaya. Nonperturbative localization. In Proceedings of the International Congress of Mathematicians, Vol. III (Beijing, 2002), pages 445–455, Beijing, 2002. Higher Ed. Press.

R. Johnson and J. Moser. The rotation number for almost periodic potentials. Comm. Math. Phys., 84:403–438, 1982.

R. Johnson. Analyticity of spectral subbundles. J. Diff. Eq., 35:366–387, 1980.

R. Johnson. The recurrent Hill's equation. J. Diff. Eq., 46:165–193, 1982.

R. Johnson. A review of recent work on almost periodic differential and difference operators. Acta Appl. Math., 1(3):241–261, 1983.

R. A. Johnson. On the SatoSegalWilson solutions of the KdV equation. Pacific J. Math., 132(2):343–355, 1988.

R. Johnson. Cantor spectrum for the quasiperiodic Schrödinger equation. J. Diff. Eq., 91:88–110, 1991.

R. Krikorian. Reducibility, differentiable rigidity and Lyapunov exponents for quasiperiodic cocycles on
$T\times SL(2,R)$
. Preprint.

J. A. Ketoja and I. I. Satija. Harper equation, the dissipative standard map and strange nonchaotic attractors: relationship between an eigenvalue problem and iterated maps. Phys. D, 109(12):70–80, 1997. Physics and dynamics between chaos, order, and noise (Berlin, 1996).

Y. Last. Almost everything about the almost Mathieu operator. I. In XIth International Congress of Mathematical Physics (Paris, 1994), pages 366–372. Internat. Press, Cambridge, MA, 1995.

J. Moser. An example of Schrödinger equation with almost periodic potential and nowhere dense spectrum. Comment. Math. Helvetici, 56:198–224, 1981.

J. Moser and J. Pöschel. An extension of a result by Dinaburg and Sinai on quasiperiodic potentials. Comment. Math. Helvetici, 59:39–85, 1984.

C. Nún͂ez. Extension of a MoserPöschel theorem for the Schrödinger equation with ergodic potential. In XIV CEDYA/IV Congress of Applied Mathematics (Spanish)(Vic, 1995), page 10 pp. (electronic). Univ. Barcelona, Barcelona, 1995.

D. Osadchy and J. Avron. Hofstadter butterfly as quantum phase diagram. J. Math. Phys., 42(12):5665–5671, 2001.

J. Puig and C. Simó. Analytic families of reducible linear quasiperiodic equations. Preprint, 2004.

J. Puig. Cantor spectrum for the Almost Mathieu operator. Comm. Math. Phys, 244(2):297 – 309, 2004.

J. Puig. Reducibility of QuasiPeriodic SkewProducts and the Spectrum of Schrödinger Operators (PhD Thesis). University of Barcelona, 2004.

B. Simon. Almost periodic Schrödinger operators: a review. Adv. in Appl. Math., 3(4):463–490, 1982.

B. Simon. Schrödinger operators in the twentieth century. J. Math. Phys., 41(6):3523–3555, 2000.

Ya. G. Sinaĭ. Structure of the spectrum of a Schrödinger difference operator with almost periodic potential near the left boundary. Funktsional. Anal. i Prilozhen., 19(1):42–48, 96, 1985.

E. Sorets and T. Spencer. Positive Lyapunov exponents for Schrödinger operators with quasiperiodic potentials. Comm. Math. Phys., 142(3):543–566, 1991.