PHYSICAL REVIEW 8 VOLUME 45, NUMBER 6 1 FEBRUARY 1992-II

Transmission through a Thue-Morse chain

Y. Avishai
Service de Physique Theorique, Centre d'Etudes Nucleaires de Saciay, F91191Gif sur Yv-ette, CEDEXFrance;
Labor atoire des Physique Theorique des Liquides, Universite Paris VI, Paris, France;
Department ofPhysics, Ben Gu-rion University of the Negev, Beer Sh-eva, Israe!
D. Berend
Department ofMathematics and Computer Science, Ben Gu-rion University
of
the Negev, Beer Sheva-, Israel;
Mathematical Sciences Research Institute, Berkeley, California 94720
{Received 27 June 1991)
We study the reflection ~rn ~
of a plane wave (with wave number k )0) through a one-dimensional ar-
ray of N 5-function potentials with equal strengths v located on a Thue-Morse chain with distances d&
and dq. Our principal results are: (1) If k is an integer multiple of m /~d,

d2 ~,
then there is a threshold
value vo for v; if v
)
vo,
then
~ r~
~1as N~ ~, whereas if v (
vo, then ~rN ~~1. In other words, the sys-
tem exhibits a metal-insulator transition at that energy. {2j For any k, if v is su5ciently large, the se-
quence of reflection coefficients ~r~ has a subsequence ~r ~,
which tends exponentially to unity. (3)
Theoretical considerations are presented giving some evidence to the conjecture that if k is not a multi-
ple of n l~d~

dq ~, actually ~r ~
~1for any v )0 except for a
"small"
set (say, of measure 0). However,
this exceptional set is in general nonempty. Numerical calculations we have carried out seem to hint
that the behavior of the subsequence
~
r
~
is not special, but rather typical of that of the whole sequence
2
~rn~ (4) A. n instructive example shows that it is possible to have ~r~~l for some strength v while
~ rz ~
~1 for a larger value of v. It is also possible to have a diverging sequence of transfer matrices with a
bounded sequence of traces.
The experimental advance in submicrometer physics
that enables the fabrication of nearly ideal one-
dimensional wires' naturally leads to increasing interest
in their physical features, especially their Fourier spec-
trum and their transport properties. The quantum-
mechanical relation between electrical conductance at
zero temperature and the transmission probability indi-
cates that some measurable physical quantities can be ac-
curately explained on the microscopic level once a one-
dimensional wire is modeled as an infinite array of poten-
tials. Systems consist of infinite one-dimensional array of
potentials are of course extensively studied in the litera-
ture in connection with Bloch theory (if they are period-
ic), and Anderson localization (if they are completely
disordered). Following the discovery of quasicrystals,
interest has been focused on the mathematical and physi-
cal nature of quasiperiodic sequences and com-
mensurate-incommensurate systems which are the first
class of structures on the way from periodic to random
matter.
In order to study transport properties of such systems
one needs a theory of scattering from a semi-infinite one-
dimensional array of potentials. One such a theory has
been developed earlier for the study of transport in a ran-
dom potential. The basic technique is to express the
transmission and reflection amplitudes through %+1
scatterers in terms of the amplitudes for N scatterers (a
combination of Mobius transformation and multiplica-
tion by a phase equivalent to the transfer matrix) and to
let N~ao (the so-called thermodynamic limit). If the
system is not random and a trace map is available, this
procedure is quite powerful. In two earlier publica-
tions
'
we concentrated on scattering from an infinite
system of 5-function potentials located on the Fibonacci
numbers x=Fand on the Fibonacci chain
x=n+u
[nlrb], n =1,
2, . . .
, N, as N~ao, where u is
a real number, v=(1 +~5) /2, and
[
.
]
denotes integer
value. To this end we have developed special mathemati-
cal tools, basically analytical and number-theoretical
techniques.
In the present work we use the same mathematical
framework and report the results of our study on scatter-
ing of a plane wave with wave number k )0 from a one-
dimensional sequence of 5-function potentials of strength
U located on a Thue-Morse chain.
"
To be more specific,
let
gbe
the Thue-Morse sequence i.e.
, (=0or 1 ac-
cording to the number of
1's
in the binary expansion of
the integer number n being odd or even, respectively.
Then the Thue-Morse chain
[x]
is constructed such
that
y=x+,

x=d, or
d2
when (=0or 1, respec-
tively, where dd2)0 and
d, &d2. This is a prototype
of a sequence generated by substitution, in this case
d, ~d, d2
and
dz ~dzd
&,
with highly nontrivial features.
The basic difference between the Thue-Morse chain and
the Fibonacci chain
'
is expressed in terIns of their
Fourier transforms. In the first case the Fourier trans-
form is singular continuous and the sequence is termed as
aperiodic. In the second case the Fourier transform is
discrete (or atomic), and the system is said to be quasi-
periodic and exhibits Bragg peaks. These aspects of
45 2717
2718 Y. AVISHAI AND D. BEREND 45
quasiperiodic and aperiodic structures have recently been
investigated by
Luck.
'
The spectrum and the nature of
states pertaining to the Thue-Morse sequence have been
thoroughly investigated by several authors.
'
We point
out that in the works of Ref. 13, the aperiodic structure is
introduced by a deterministic aperiodic sequence of diag-
onal site potentials
[ VJ
within the context of the
discrete Schrodinger equation in the tight-binding ap-
proximation
(f
+
&+it'&) +Vg=E
f.
Here, on the other hand, we use the ordinary Schrodinger
equation and the aperiodic structure is introduced
through the sequence
[y)
of distances between scatter-
ing centers. It affects only the phase of the wave func-
tion.
The central question which will be addressed in this
study does not concern the spectral properties but, rath-
er, the question of transmission and reflection. We want
to find out whether a one-dimensional Thue-Morse chain
is a conductor ((rz~~l) or an insulator ()rz~-+1). To be
more specific, is there a curve in the (u, k) plane separat-
ing the conductor and insulator "phases"? The finer de-
tails pertaining to the behavior of
~ rN ~
as a function of N
are also discussed albeit briefly. We point out that this
problem is related to the study of the spectrum in the fol-
lowing heuristic manner: If an energy E
=k
does not
belong to the spectrum we expect that
~r~~
~1 exponen-
tially. On the other hand, if the energy is in the spec-
trum, then the behavior of
~rz~
for large N determines
the nature of the pertinent eigenstate (localized, extend-
ed, or critical). In the present work we simply study the
transmission for all energies, both in the spectrum and
outside it. Any result which applies to "almost every en-
ergy" will then
apply
to the spectrum as well.
Although every sequence has its own characteristics we
believe that the mathematical framework developed here
is capable of solving the scattering problems encountered
in other one-dimensional arrays, especially if a trace map
exists. This is the case for two-letters substitution se-
quences.
'
Recently, trace maps have also been con-
structed for general substitution sequences, with applica-
tions to the circle sequence and the Rudin-Shapiro se-
quence.
'
It has been recognized that in these kinds of problems,
mathematical rigor is essential. That is the reason why
the presentation below is mathematically oriented. Yet,
the reader should be aware of the physical basis for the
present study. For the sake of smooth reading we present,
the proofs of most of our statements in the Appendix.
With that notion of apology we start our study proper.
Consider a one-dimensional array of N 5-function po-
tentials
N
V~(x)=v
g
5(x

x),
n=1
where U &0 and
jx)
is an infinite deterministic real se-
quence whose difference sequence
x+&

xassumes two
possible (positive) values
d,
and d2. In fact, due to
translation invariance what really counts is the sequence
which satisfy unitarity
/r/'+/t/'=I, tr*+t'r =0,
and continuity at the point x =xo,
t=1+r .
(4a)
(4b)
The unitarity relation (4a) is valid of course for any N.
For N & 1 scattering centers, the reflection and transmis-
sion amplitudes can be determined from a recursion rela-
tion as follows: We introduce the dimensionless parame-
ter
q
= u/2k and define
1+iq iq

iq 1

iq
rt
'
(t r)t-

'ky
e
with
0
iky
e
D=CA,
det( C)
=
det( A)
=
det(D)
=
1 .
The matrix
D
is recognized as the transfer matrix at
site n. We then have
a+
be+&
a
=CAb
n
Anticipating the use of product of n transfer matrices we
also define
M=ACA).
. .
CA)C .
The conductance (at zero temperature and in units of
e /h) of this system is given by the Landauer formula
g =~t~/r~~ Therefore, we . need to study the limit of
(t~~ =1/~aN~ as N~~. Equivalently, we may inspect
the limit of
~r~[
=
~b~/aN ~
and use unitarity. If
[ tz )
~0 (equivalently [rN
[~1) we say that the system is
an insulator. If
~tz~
does not tend to 0 the system may
conduct. Our aim is to find out for which values of the
mom. entum k and the strength v the system is an insula-
of differences y=x+
&

x. Some of our results will ap-

ply
to any sequence yassuming only two values, but the
more detailed study is concentrated on the Thue-Morse
chain, for which some deeper results are obtained. In the
special case of a Thue-Morse chain the sequence
ty]
is
defined by
d,
if $=0
yn
=xn+i xn
=
(2)
where g=[1+(

1)'"']/2
and s(n) the number of
1's
in the binary expansion of n.
A plane wave at momentum k, e
',
coming from the
right will have reflection and transmission amplitudes rN
and
tN, respectively. For N=1, these amplitudes are
given by
v 2ik
r
=
2ik
v'
2ik

v
'
45 TRANSMISSION THROUGH A THUE-MORSE CHAIN 2719
tor and for which it may conduct. It is also of interest to
know when the sequence IrN I
has a subsequence converg-
ing to 1 (or a subsequence converging to 0).
Now, we point out that in general (independently of
the sequence x),
all the matrices C, A,
and
M, belong
to the following multiplicative group of 2 X2 matrices:
SU(1,
1)=
'
Denote
=
l~l+ IPI
(10)
Following our previous work' the following proposition
is implied.
Proposition 1. For any sequence
n. we have
lr I
.
+1 if and only if
IIM II
+ oo.
J J
Actually, using basically the same ideas, one can prove
the following proposition.
Proposition 2. There exist positive constants
c, &c2
(depending on k, v, and the sequence
y)
such that
c, /IIMII'1

Irl
c,
/IIMII',
n &1.
For a fixed k it seems plausible that the larger v is, the
better are the "chances" that
ITIN
~1.Surprisingly,
however, it is possible that the re6ection coemcient tends
to unity for some value of u but not for a larger value of v
(see also Example 1). Nevertheless, our next result shows
that the set of
v's
for which
IrNIN
~1is quite nice.
Recall that a set in a metric space is called an F set if it
is a countable union of closed sets, and an F
&
set if it is a
countable intersection of F sets.
Proposition 3. For any sequence [yJ",
and k &0,
the set
[
u:
I r~ I N
~1
l
is an F
s
set. The proof is given
in the Appendix. One actually has the more general re-
l
Ir~l~ 1 v
)
vo( &0),
r
2ktg

sink &0
k
2
vp=

2kctg

sink &0 .
2
(12)
(2) If IrNI~1 then the sequence r
itself lies on a circle in
the complex plane whose diameter is
q/Isink

q
coskl( &1) passing through the origin, and in
particular
limlrl &1;limlrl
=0
. (13)
The proof of Proposition 4 is identical to the one we gave
in Ref. 10 where we have studied the Fibonacci chain. It
is to be noted that the proof does not rely on the nature
of the sequence but merely on the fact that the sequence
of differences assumes only two values.
Remarks (1) Usi.ng Proposition 2 one can improve
somewhat the first part to obtain
suit specified below.
Proposition
3'.
In the setup of Proposition 3, for any
sequence
[N~ J
J"
1
of positive integers the set
[u:Irz
I1v
~
1[
is an F
s
set. The Proof is identical
J J
with that of Proposition 3.
Our main problem is to decide, given the sequence
[yJ
"=1, for which values of k and u do we have
Ir~l1v
+1. Before studying the Thue-Morse chain we
present a statement concerning any sequence ywhich
assumes only two values d
&
and dz.
Proposition 4. Suppose yassumes only two values
d,
and dz. Then for every k which is an integer multiple of
n.
/(d,

dz ),
there is a threshold value
uo
for v such that
v =vti c,
/n1

Irl
cz/n, cz&c,
&0 const,
v &vo=cie
"
1

Irl cze
",
cz&ci )0, and a)0 const dePending on u .
(14)
(2) In the second part of Proposition 4, the sequence r
is
"usually" dense in the circle it lies on, but sometimes is
periodic and so concentrated on a finite subsect of the cir-
cle (see Ref. 10 for more details).
(3) If k is also an integer multiple of m (which is possible
if
di

dz
is rational), then
Irl
converges to 1 for every
u &0
(IIMII
grows linearly with n, so that 1

lrl
behaves as c/n for some constant c).
We now turn to study in detail the Thue-Morse chain
for which the sequence
y
is defined through Eq. (2}.
Re-
call that the Thue-Morse sequence
g
is defined as fol-
lows:
P=kdP=kdz, (17)
and define two sequences of matrices
[Pl
", and
[QJ
",in SU(1, 1) by
where, as we have already indicated, d,
and
d2
are posi-
tive numbers. The case
d&
=d2 amounts to an ordered
system for which
y
is constant (a trivial case in the
present context) and hence it is assumed that
d, Adz.
Proposition 4 covers values of k which are integer multi-
ples of
m.
/(d,

dz)
and we henceforth assume
k 6m /(d,

dz )Z unless stated otherwise. We set
(1=0, gzm+;=1

g';, m &0, 1i 2
(15)
Po=
0
Qo=

ig
0
lf
(18a)
The sequence
[yJ,
Eq. (2}can be defined also as n Qn

1 n

I & Qn Pn
1Qn

1
(18b)
y=d, +(dz

d, )g,
n &1, (16) where C, the transfer matrix through the 5-function po-
2720 Y. AVISHAI AND D. BEREND 45
tential, is defined in Eq. (5). Then direct calculation
yields the following.
Lemma 1. For the sequence of matrices
IM~jg
defined in Eq. (8) one has
M =P,
n~0. (19)
X=tr(P), n
&0
. (20)
At present we do not have the tools to study the whole
sequence
IM& jN,
and our strategy will be to study its
subsequence
tM j"&.
Numerical simulations indicate
that any statement concerning the behavior of the subse-
quence is valid for the whole sequence. We shall investi-
gate the sequence
[ Pj
"=,using the trace map. Denote
2
FIG. 1. Decomposition of the right half plane into the three
sets
S',
2), and 8following Lemma 4.
Note that tr(Q)=x for n &1. Our task is to find out
for which values of k and U the sequence of norms
t ~~P~~
j",tends to infinity and for which it does not. To
this end we shall consider the sequence of traces
Ixj
",. If
~X~~
~ then
)~P~~~
ao, whereas
if
~x~~~
then we have a clue hinting that perhaps
~(P~~~
~ (however, it is possible to have
Ixlmao but
[(P))~
~, as is demonstrated in
Example 1 below). The following Lemma is well known.
Lemma 2. If IPj"
0
and
IQj"
0
are any two se-
quences of matrices in SU(1, 1) satisfying the recursion re-
lations (18.b) and X=tr(P),then
K(x,y) =(y, xy

2x+2), (x,y)EE+ XE . (25)
We then have the following lemma.
Lemma 3. The following equality holds:
(26) yoH =Koy,
which is proved in the Appendix.
In the terminology of dynamical systems we say that
the transformations H and K are semiconjugate via
y.
Clearly, the Lemma implies that we may as well study K
instead of H. In fact, we may replace Eq. (23) by
the fol-
lowing relation:
X +2 X (X
+1
2)+2 (21)
For completeness we present the proof of Lemma 2 in the
Appendix.
To study the behavior of the sequence IXj"=~
it is
thus natural to consider the mapping H: IR ~R defined
by
(Xx,+t)
=K"
(X( X2),
(27)
Thus, our main objective now is to find out which points
in IR+ XIR have K orbits going to infinity and which do
not.
Lemma 4. The following three subsets of IR+ XR are
invariant under K:
H(x, y) =(y, x
y

2x +2), (x,y)
eE
In fact, in view of Lemma 2 we have
X +1) H(X
I X
) H
(Xl X2)
(22)
(1)
6=
I(x,y):x
&O,
y
&2j .
(2)
$=
I(x,y):x
&O,
y
~2,
y
~x

2j
.
(3) T=I(x,y):x &O, x

2~y
2j
.
(23)
so that the H orbit of
(X&,X2)
contains all information re-
garding the traces of the matrices P.
The dynamical system defined by H has been studied
by
various authors.
'
The questions we are interested in,
as well as our approach, are somewhat different, though.
We shall now replace H by
a somewhat simpler transfor-
mation which is adapted to our future discussion. Let
E+=
Ix
EE:x&Oj. Define maps y:R ~R+
XIR and K:IR+XIR ~R+ XIR by
The proof of Lemma 4 (for 7only) is given in the Appen-
dix.
In particular, the borderline between
'9
and g)U7;
namely, the half line
I(x, 2):x
Oj,
is K invariant; in fact
it is carried to the single point (4,2). The decomposition
of the right half plane E+ XR into the sets Vl, S, and 7
is illustrated in Fig. 1.
emma 5. We have
(28)
y(x, y)
=(x,
y), (x,y)
&E (24)
I
(1) K"(x,
y)
~ (~, ao), (x,y)Ella=I(x, y):x&O,y&2j,
(2) K (x,y)
~ (ao,

oo
), x
&
l,
y
+
1, (x,y)A(1,

1) . (29)
The second part of Lemma 5 is proved in the Appendix.
It is now possible to study the behavior of the set of po-
tential strengths
Iv
j
for which the reflection coefficient
tends to 1 in the special case of a Thue-Morse chain. Re-
call that, according to Proposition
3',
for any k &0 the
set
I
v:~r ~~1
j
is an F
s
set. The following propo-
sition, which we prove in the Appendix, may hint that in
our case this set might be even more we11 behaved.
45 TRANSMISSION THROUGH A THUE-MORSE CHAIN
2721
proposition 5. For any k & 0, the set
[v:lg, I,
~~
j
is open. Note that the set considered in Proposition 5 is
just a subset of the set
[
v:
~
r
~~
1
j;
not the whole of
this set (see Example 1).
In Lemma 5 we found two unbounded sets in la+ XR
consisting of points whose orbits under K
go
to infinity.
As T is a compact K-invariant set, its points have bound-
ed Eorbits. The two shaded regions in Fig. 2 depict the
set of points the behavior of whose K orbits is not
covered
by
Lemmas 4 and 5.
Remarks (1) .It is inconsequential for the problem at
hand, but it would be of interest to better understand the
behavior of K on the triangle V. On the boundary of 7;
this is relatively simple. The top side
[(x,
2):x 4j
is car-
ried to the K-fixed point (4,2), whereas the left side
[(O,y):

2~y
~
j
is carried to the top side. To describe
the action of K on the bottom right side of T, namely,
the line interval cP
=
V'A2), we first note that
cF=[(4cos a, 2cos2a):aC[0, 2~)
j
. (30}
A straightforward calculation gives
K(4cos a, 2cos2a)=(4cos 2a, 2cos4a), aE[0,2m) .
(31)
+A,
,
n &0. (34)
This sequence is real if and only if k is either real or lies
on the unit circle in the complex plane. In the first case
the sequence (y2,
y+,
) lies on the ray
I
(x, x

2):x
&
4j,
while in the second case it lies on the interval
d'
(note that
both sets are portions of the same line}. This shows with
more clarity why K acts on
8'
(almost) as multiplication
by
2 on T, and on [(x,
x

2):x
&
4j
as squaring on E+.
There exist points with bounded Eorbits in addition to
those of 8. In fact, any point entering cP after finitely
many iterates of K is such. To describe this set
U,
".
,
K
'(d") we first note that
that of
pz
on 8, where it is also well understood. It would
then be worthwhile to explore the behavior of E on the
whole of 8.
(2) The family of solutions of the trace recursion rela-
tion (21) is two-parametric. A one-parameter subfamily
can be written down explicitly. Start with any matrix
R &SU(l, l), put PO=QO=R and define [Pj",
and
[Qj",by Eq. (18b). (Taking just commuting Po and
Qo
does not provide more generality. ) Then Eq. (21) still
holds. The sequence of traces is easily found out to be of
the form
p2(t)=2t, t ET, (32)
Thus, letting
p2
denote multiplication by
2 on the circle
group T
=
IR /2m Z,
K '(/}=SU
I(x,
2):x &Oj,
and therefore
K
'(/)=K '(8)
U
I(0 y) y
&Ej .
(35)
(36)
and P:T~8 be defined
by
P(i)=(4cos t2c os2 r), t FT,
we find that the following diagram
(33)
Letting P:R ~lR be the projection on the first coordinate
we obtain
K
'+
(cP)=K '(8)
U
I(x,
y):PoK' '(x,y)=0j, i
&1
.
(37)
T
Pl
P2
P$
2 I
is commutative. Now
p2
is a mixing algebraic endomor-
phism of T, and its action has been extensively studied.
Furthermore, the action of K on 8 is semiconjugate to
Some of the one-dimensional sets we encounter in the
process .have simple representations. Thus, to get
from K (/) to K 5(8) we add successively the
sets t(x, 0):x &0j, [(x,
2

2/x):x &0j and

[(2/y (2

y), y):OAy &2j. However, as P

Kj
is a poly-
nomial of degree 2~, these curves become more and more
cumbersome.
Siinilarly, all points entering (1,
ao ) X(

~,

1) under
some power of E have orbits going to infinity. It is
therefore easy to reduce the shaded regions in Fig. 2 by
taking out of them the inverse images of this set under
higher and higher powers of K. However, we could not
answer the following questions.
(1) Are there any points outside
Up,
K '(8) having
bounded orbits?
(2) t (x,y) &E+XIR be a point with an unbounded K
orbit and let
(x,y)
=K"(x,
y). Is it necessarily the case
that xao (and
iy i
ao
)'?
Lemma 6. Let
[sj
",be a sequence of positive num-
bers satisfying
s+2=pss+&,
n
&
1 . (38)
for some
p)
0. Then
FIG. 2. The set of points the behavior of whose E orbits is
not described by Lemmas 4 and 5.

1/2[

1/6

1/3 1/3
](

)
)"
[(
g )
i/6 j2" )
1
(39)
2722 Y. AVISHAI AND D. BEREND
45
The Lemma is easily proved by passing to logarithms and
solving the resulting inhomogeneous linear recurrence
with constant coefficients.
Lemma T. Let (x,y)HR+ XR and
(x,y)=E"(x,
y)
for n 0. If x~~, then
x&a), n )0, (40)
for some a )0 and
g
& l.
Lemma 7 is proved in the Appendix.
Lemma 8. For every
q
)0 (recall that
q
=v/2k), we
have
yz&y,

2. The proof of Lemma 8 is given in the
Appendix.
Thus, all possible pairs
(g, g2}
lie, in our case, to the
I
right of the line
[
(x, x

2):x
)
0],
on which they lie if we
start with commuting
Po, QoESU(1, 1}(see the second
remark following Lemma 5). However, since points out-
side 8 can be carried to 8 by K, we may still get bounded
traces. The following Lemma shows that in this case
bounded traces actually mean bounded matrices.
Lemma 9. If
y
=0
for some
j
1, then P=I for
n
)
j+2, and therefore the sequence [M]",defined in
Eq. (8) takes only finitely many values. For the proof of
Lemma 9, see the Appendix.
It is possible to have
y
=0
already for
j
=1. In fact,
from (A8) it follows that
y,
vanishes for the following
two values of
q
=v/2k [recall the definitions of
P
and
g
Eq. (17)]:
q & 2
=
[

sin(p+ g)+[sin (p+ g)

sing sing cos(p+

p)
]'~
]
/2 sing sing
=[

sin(P+g)+(sin /+sin
g)'~
]/2sing sing .
(41)
Depending on
P
and
g,
none, exactly one, or both
q,
and
qz
may be positive. Thus, the case of bounded traces is
possible. The question whether solutions q) 0 exist for
the equation y~(q)=0
for j)2 involves the solution of
equations of higher degrees than (41). One would expect,
though, that infinitely many of the solutions would turn
out to be positive and thus relevant in the present con-
text.
If
y
=0
for some
j
0
then the answer to our basic
question is known for any N, not only for 1V
=2".
Indeed,
since
Pj+2=QJ+2=I,
we have [see Eq. (8)] M~=I for
every
N:
0 (mod
21+
),
where rtv=0 for each such X.
Note that the sequence
[rz]g,
is nonperiodic. Rather,
it moves from 0 to 0 on two different paths, the order cor-
responding to the rule of the Thue-Morse sequence. Ob-
viously, in this case, max& l rz l
& 1.
Lemma IO (1) For e.very sufficiently large
q
we have
(2) If
lyl~~
for some
q,
then
lg l
&
ag
for some a & 0 and g) 1. Lemma 10 is proved
in the Appendix.
We are now in a position to state our central result.
Propositions 1, 2, 4, and Lemma 10 now yield the follow-
ing theorem.
Theorem 1. For every k, if v is sufficiently large, then
l
r
l ~1.Moreover, the convergence is exponential,
namely,
1

lr l ag,
n &0, a )0, 0&(&1 . (42}
Recall that we assumed k(d,

dz) not to be an integer
multiple of ~. The theorem actually holds in that case as
well, but in the particular case of k itself being also a
multiple of n (which is possible if
d&

d2
is rational), the
convergence is not exponential, and the right-hand side
of (42) is replaced by
a /4".
Example 1. We shall see here that it is possible to have
l r~ l z
~1 for a certain value of v and yet
limz lrzl
&1 for some larger value of v. In some
sense, this result is counterintuitive since for a positive
potential, increasing v means stronger repulsion. In the
context of one-dimensional barrier penetration it can be
viewed as a resonance tunneling but we are unaware of its
I
presence for aperiodic systems.
We shall also encounter in this example a nontrivial
case where the sequence of traces
Iy]",
is bounded
but
llP ll~
~. Thus, the consequences drawn from
the use of the trace map should be taken with some care.
Consider the specific case in which we take k, dd2,
so that P=kd, =5m/6 an.
d
g=kd2=3'/4. By (41),
g, (q) vanishes for
q,
=(~3+1+@6)/2 and q2=(~3+1

v'6)/2. In view of Lemma 9, P3=I and

lrzl
&1 for
these values of q. We mention that, taking q3=1, we
may verify that
y3(q3)=0, so that Ps=I and hence
max& l r& l
& 1 for
q3
as well N.ow consider
q&=(v'3+1)/2. Denoting R
=[] and
p
=i
(&3 1)/4
,
we find after lengthy calculations that
(43) P2=I+2pR, Q~=I+pR, P3=Q3=I+3pR .
It follows that y=2for n
3
and
M~=P3 =I+3EpR, X
0
. (44)
~~, that
lr~l~
~1 r
q
=q4.
Notice incidentally that in view of Proposition 2 the con-
vergence in this special case is not exponential.
ACKNOWLEDGMENTS
Proof of Proposition 3. Denote
(Al)
Since
M~ depends continuously on v, each L.
z
is a closed
set. Obviously
We are grateful to H. Furstenberg, T. Spencer, J. M.
Luck, and R. M. Redheffer for valuable discussions on
this subject and to D. Glaubman for verifying the calcu-
lations in Example 1 using the MATHEMATICAO software.
This work was sponsored in part by a Grant from the
United States-Israel Binational Science Foundation and
in part by
an NSF Grant No. 8505550. This work was
partially supported by
the Laboratoire de la Direction
des Sciences de la Matiere du Commissariat a 1'Energie
Atomique and by the Israel Ministry of Science.
APPENDIX
45 TRANSMISSION THROUGH A THUE-MORSE CHAIN 2723
=n
n nL, ,
j
=
1 i =1 %=i
(A2)
which shows that our set is indeed an F
&
set. This
proves the proposition.
Proof of
Lemma 2. Employing the Cayley-Hamilton
theorem we readily obtain
y+2=tr(P+2)=tr(PQQP)=tr(PQ)=tr[(yPI
)
(gQI)]
=ytr(PQ)

ytr(P)

ytr(Q)+
tr(I) =yy+,

2y+2
=y(y+,

2)+2 . (A3)
Proof of
Lemma 3. For any (x,y)
CR
(yo H)(x, y)
=
y(y,
x'y

2x'+2)
=(y, x
y

2x +2)
and
(Koy)(x, y)=K(x, y)=(y, x
y

2x +2) .
(A4)
(A5)
and
xy

2x +2
+
2x

2x+2=2 (A6)
Proof of
Lemma 4. We shall only show that, say, T is
K invariant. In fact, if (x,
y
}G Tthen
where K(x,y) GQ, .
Thus, starting with (x,y) EQand denoting
(xy,
)=K"(x,
y)
for n &0, we have
(x,y
)Qg)i for
each n A. s in Eq. (A8), we have
yi
&y
=yo, and
by the
same token the sequence
[y]"
o
is nonincreasing. It
then follows that the sequence
Ix]
"
o
is nondecreasing.
Set L
=
limy. Then
x=y& ~L . The fore-
going discussion implies that, if L &

00, then the point
(L,L) must be a fixed point of K. It is, however, easily
2
verified that the only fixed points of K are (1,1), (1,

1),
and (4,2). Consequently, L
=
ao unless (x,y)
=(1,

1).
This proves part (2) of the Lemma.
Proof ofProposition 5. In view of Lemma 5 we have
y

2 =xy

2x +2

[y

(x

2) ](2

y)
~xy

2x +2, (A7)
(A9)
xy

2x+2&y+2(1

x) &y
&
1, (A8)
so that K(x,y)
G C.
Proof of
Lemma 5. We prove the second part Eq. (27}.
Let us first show that the set
S,
=
[1,
ao
) X( ~, 1]
is K in-
variant. In fact, if (x,y) &2)i, then
y
&
1 and
Now, each trace
g;
considered as a function of v, is clear-
ly
continuous. Hence, each of the sets in the union on
the right-hand side of (A9) is open, where so is their
union. This proves the proposition.
Proof of
Lemma 7. Since x~
~ we have
y~
Do as
well. Hence, for all suSciently large n
ly.
+21=lx. +iy. + i

2x. +i+2I
=
ly.
'y.
+i

2y.
'+2I
&y.'ly.
~i

2I

2&y'ly+
iI
&2 . (A10)
tr(Po)=2(cosg+q sing),
tr(Qo)=2(cosf+q sing),
(Al 1)
(A12)
pi=tr(Pi ) =4q sing sinP+4q sin(P+P)+2cos(P+g),
(A13)
X2

tr(PoQoQoPo }
=tr(PoQo )
=tr[tr(Po)Po

I][tr(Qo )Qo
I]
=4(cosg+q sing)(cosf+q cosP)gi
4(cosg+q sing)2

4(cosg+q sing) +2,
(A14)
By
Lemma 6 we then get ~y~
&
bg
for some b & 0 and
g& 1, and hence
x=y,
&
b
g,
which proves the
Lemma.
Proof of Lemma 8. Recall the definitions of
P
and
f
[Eq. (17)] and of the matrices [P]"
i
and [Q]"
[Eqs. (18)]. We routinely get
y2

pi+
2
=
4q
sin (
P

P)
. (A15)
The right-hand side of Eq. (A15) is strictly positive since
the case
P

g=k (d
~

d2 ) being an integer multiple of

m.
has been excluded. Hence the Lemma is proved.
Proof of Lemma 9. First, note that if tr(A)=0 for
some A ESU(1, 1), then A =tr(A)A
I=
I. Conse-
-
quently, if tr( A )
=
tr(B)
=
0, then ABBA
=
A ( I) A
=I
In our case, since
y,
=0, namely, tr(P, }=tr(Q,
.
)=0, we
get P;+2=P,
.
Q, Q,
.
P;=I, and similarly Q;+2=I. By
in-
duction this implies P=Q=I for n
&i
+2, which
proves Lemma 9.
Proof of
Lemma 10 (1) From (A1.3) and (A14) we see
that, in general,
y,
(q)
is a quadratic polynomial in
q
and
gz(q)
is a quartic. Their leading coefficients are
4 sing sing and (4 sing sing), respectively. Hence, if
both sing%0 and sing&0, the Lemma follows from (25)
and Lemma 5. Now suppose that exactly one of the sines
is nonzero, say sing%0 but sing
=0. Then
sin(P+g)=+sing%0, so that, by (A13), yi(q)
as a func-
tion of
q
is nonconstant. Furthermore, according to
(A14), y2(q)
is a quadratic polynomial with leading
2724 Y. AVISHAI AND D. BEREND 45
coefBcient
16sing cos1b sin(P+ 1b)

4 sin
P
=
16 sin (I) cos
1b

4 sin $=12sin /%0,

and we conclude as in the former case. Finally, the
remaining case sing
=
sing
=
0 is excluded since we have
assumed that k(d&

dz)=P

f
is not an integer multi-
ple of m.
(2) The proof of this part follows straightforwardly
from Lemma 7.
'A. A. Gogolin, Phys. Rep 86, 1 (1982); 166, 269 (1988); Science
and Engineering of
I and 0Di-mensional Semi Conductors,
NATO Workshop, 1989, edited by S. P. Beaumont and C. M.
Sotomayor-Torres (Plenum, New York 1989); M. Pepper, in
One Dimensional Transport Phenomena in Semiconductor
Structure, Proceedings of the VI International Conference on
Recent Progress in Many Body Theories, edited by Y. Av-
ishai (Plenum, New York, in press).
2R.
Landauer, IBM J. Res. Dev.
, 1, 223 (1957).
F. Bloch, Z. Phys. 52, 555 (1928).
4P. W. Anderson, Phys. Rev. 109, 1492 (1958); 124, 41 (1961);E.
Abrahams, P. W. Anderson, D. C. Licciardello, and T. V.
Ramakrishnan, Phys. Rev. Lett. 42, 673 {1979).
5D.
Shechtman, I. Blech, D. Gratias, and J. W. Chan, Phys.
Rev. Lett. 53, 1951 (1984).
M. Kohmoto, L. P. Kadanoff, and Chao Tang, Phys. Rev. Lett.
50, 1870 (1983); S. Ostlund, R. Pandit, D. Rand, H. J.
Schellnhuber, and E. D. Siggia, ibid. 50, 1873 (1983); D.
Levine and P. J. Steinhardt, ibid. 53, 2477 (1984); M. Fujita
and K. Machida, Solid State Commun. 59, 61 (1986); D.
Levine and P. J. Steinhardt, Phys. Rev. B34, 596 (1986); J. E.
S. Socoper and P. J. Steinhardt, ibid. 34, 617 (1986); M.
Kohmoto and B. Sutherland, ibid 34, 384.9 (1986); Q. Niu and
F. Nori, Phys. Rev. Lett. 57, 2057 (1986); R. Birklund and M.
Severin, J. Phys. C 21, 3217 (1988); P. Hu and C. S. Ting,
Phys. Rev. B 34, 8331 (1986); M. Kohmoto, B. Sutherland,
and K. Iguchi, Phys. Rev. Lett. 58, 2436 (1988); T. Schneider,
A. Politi, and D. Wurtz, Z. Phys. B66, 469 (1987); J. Goda, J.
Phys. Soc. Jpn. 56, 1924 (1987).
B. Simon, Adv. Apl. Math. 3, 463 (1982); J. B. Sokoloff, Phys.
Rep. 126, 189 (1985).
8P. Erdos and R. C. Herndon, Adv. Phys. 31, 65 (1982); Y. Band
and Y. Avishai, Phys. Rev. B34, 3429 (1986).
Y. Avishai and D. Berend, Phys. Rev. B41, 5492 (1990).
'
Y. Avishai and D. Berend, Phys. Rev. B43, 6873 (1991).
A. Thue, in Selected Mathematica/ Papers, edited by Trygve
Nagell, Arte Selberg, Sigmund Selberg, and Knut Thalberg
(Universitetsforlaget, Oslo, 1977); M. Morse, Trans. Am.
Math. Soc. 22, 84 (1921); M. Queffelec, Substitution Dynami
cal Systems-Spectral Analysis, Lecture Notes in Mathematics,
Vol. 1294 (Springer, Berlin, 1987).
J. M. Luck, Phys. Rev. B39, 5834 (1989).
F. Axel, J. P. Allouche, M. Kleman, M. Mendes-France, and
J. Peyiere, J. Phys. (Paris) Colloq. 47, C3-181 (1986); J. Bel-
lissard, in Statistical Mechanics and Field Theory: Mathemat-
ical Aspects, Proceedings of the International Conference,
Groningen, No. 257 of Lecture Notes in Physics (Springer,
Berlin, 1986); R. Riklund, M. Severin, and Y. Liu, Int. J.
Mod. Phys. B1, 121 (1987); F. Axel and J. Peyriere, C. R.
Acad. Sci. Ser. II, 306, 179 (1988); J. Stat. Phys. 57, 1013
(1989); F. Axel, in Quasicrystals and Incommensurate Struc
tures in Condensed Matter, Third International Meeting on
Quasicrystals (Vista Hermosa, Mexico, 1989) edited by
M.
Jose Yacaman, D. Romeu, V. Castano, and A. Gomes
(World Scientific, Singapore, 1989) p. 243; J. Bellissard, in
Number Theory and Physics, Proceedings of the Winter
School, edited by J. M. Luck, P. Moussa, and M. Wald-
schmidt, Les Houches, France, 1989 Springer Proceedings in
Physics, Vol. 47 (Springer, Berlin, 1990); J. Bellissard, A. Bo-
vier, and J. M. Ghez, Commun. Math. Phys. 135, 379 (1991);
F. Axel and H. Terauchi, Phys. Rev. Lett. 66, 2223 (1991};M.
Combescure (unpublished).
J. P. Allouche and J. Peyriere, C. R. Acad. Sci. Ser. II, 302,
1135 (1986).
~5M.
Kolar and F. Nori, Phys. Rev. B42, 1062 (1990}.