HAL Id: jpa-00210849
https://hal.archives-ouvertes.fr/jpa-00210849
Submitted on 1 Jan 1988
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Transfer matrix theory of leaky guided waves
Didier Sornette, Louis Macon, Jean Coste
To cite this version:
Didier Sornette, Louis Macon, Jean Coste. Transfer matrix theory of leaky guided waves. Journal de Physique, 1988, 49 (10), pp.1683-1689. �10.1051/jphys:0198800490100168300�. �jpa-00210849�
Transfer matrix theory of leaky guided waves
Didier Sornette
(1,2),
Louis Macon(1)
and Jean Coste(1)
(1)
Laboratoire de Physique de la Matière condensée, CNRS UA 190, Faculté des Sciences, Parc Valrose, 06034 Nice Cedex, France(2)
also at Centre de Physique Théorique, CNRS LP 014, Ecole Polytechnique, F-91128 Palaiseau Cedex, France(Reçu le 14 avril 1988, accepté le 22 juin 1988)
Résumé. 2014 On présente une approche générale en terme de matrice de transfert pour décrire la propagation
d’ondes guidées en présence d’inhomogénéités et de diffuseurs. Nous nous focalisons particulièrement sur le problème du couplage avec les modes de rayonnement conduisant à des fuites de l’onde guidée dans le milieu environnant. Des exemples sont fournis par les ondes acoustiques de surface, les excitations electromagnéti-
ques et acoustiques ou les ondes évanescentes proches d’une frontière, les ondes guidées dans les fibres
optiques... A partir de considérations de symétrie et de lois de conservation, nous obtenons, dans le cas de diffuseurs symétriques, la forme générale de la matrice de transfert comme fonction de quatre paramètres indépendants. Pour des réseaux périodiques unidimensionnels, les fuites disparaissent aux bords de bandes : cet effet cohérent provient des interférences destructives entre les rayonnements émis dans le volume par
chaque diffuseur. Ce résultat montre clairement que le problème de fuite est profondément différent d’un effet de dissipation usuel. Finalement, nous discutons la compétition entre la localisation d’Anderson et les effets de fuites cohérents, en présence de désordre. Proche des bords de bandes, la longueur d’atténuation due aux
fuites est beaucoup plus grande que la longueur de localisation. Nous suggérons une situation expérimentale
où ces effets pourraient être observés.
Abstract. 2014 We present a general transfer matrix approach for the propagation of guided waves in presence of
inhomogeneities or « scatterers ». We particularly address the problem of the coupling with radiation modes
leading to a leakage of the guided wave to the surrounding bulk medium at each scattering. Examples are
surface acoustic waves, electromagnetic or acoustic excitations or evanescent electromagnetic waves near a
boundary, guided waves in optical fibers... From symmetry and conservation laws, we obtain, in the case of symmetric scatterers, the general form of the transfer matrix in terms of four independent real parameters. For 1D-periodic lattices of identical scatterers, we show that leakage vanishes at the band edge : this coherent effect
stems from the complete destructive interference between the converted radiations at each scatterer. This result demonstrates that coherent leakage is deeply different from a usual « dissipation » effect. Finally, we
discuss the competition between Anderson localization and coherent leakage in the presence of disorder. Near the band edges and in the presence of disorder, the « attenuation length » due to the leakage is much larger
than the localization length. We suggest an experimental situation where these effects could be observed.
Classification
Physics Abstracts
03.40K - 71.55J - 84.40T
1. Introduction.
Wave propagation in arbitrary 1D-systems has been
the focus of intense researches
[1]
and is still thesubject of interesting developments especially when
one considers this problem in novel contexts
[2]
which can raise new experimental and theoretical
questions.
In this respect, we address here the general
problem of the competition between 1 D-wave propa-
gation and coherent leakage to the bulk. Experimen- tally, no real 1D system exists rigorously : 1D- scattering is often bound to lead to a coupling with
bulk radiation modes in addition to the pure 1D- reflection. The interest in this problem is also suggested by the practical fact that guided waves are
often used to develop devices with many industrial
applications. For example, real time signal proces-
Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:0198800490100168300
1684
sing in radars or in TV sets use Rayleigh acoustic
waves propagating at the surface of a piezoelectric
solid
[3].
In particular, surface acoustic waves(SAW),
propagating on reflector arrays constituted of parallel grooves, have numerous applications(reflective
array compressors, bandpass filters, filter banks, high performance SAW resonators, oscil-lators...) [3].
The behaviour of electromagneticwaves near surfaces and interfaces is important to
the operation of integrated optics devices
[4].
Guided electromagnetic waves in optical fibers are developedfor efficient communication, sensing and signal pro-
cessing purposes.
This problem of the competition between 1D-
wave propagation and coherent leakage in the bulk has been partly addressed for example in the context
of SAW in
[5]
within an approach which integrates afield theory with the theory of two-mode coupling
and also recently for dielectric waveguides within a coupled mode theory
[6].
In the present paper, we extend these studies to full generality within a macroscopic model and show how to incorporate the specific coherent nature of the guided wave to bulkradiation leakage in the usual transfer matrix formal- ism of 1D-wave propagation. From space reversal symmetry and a local energy flux conservation, we
obtain the general form of the single symmetric
scatterer transfer matrix in terms of four independent
real parameters
(compared
to two parameters for symplectic matrices describing space-symmetricalscatterers without leakage corresponding to the pure
one dimensional
(ID) case).
This formalism is relevant for the optimal design of high qualityresonators and oscillators, to non-destructive testing
methods
(where
the leakage of the Rayleigh wavedue to scattering by a defect acts as a signature of its
presence),
to the achievement of high contrasts inultrasonic microscopes
(where
the contrast is largelyincreased by the leakage of excited Rayleigh waves by
discontinuities),
to surface optical devices as wellas to modulated optical fibers...
Two among the four parameters of the transfer matrix have a simple meaning
(similar
to that for symplecticmatrices),
one being related to the localvalue of the 1D-wave velocity and the other to the
typical
longitudinal
scale of the 1D-scatterer. The existence and meaning of two additional parameterscan be understood physically as follows : the coupl- ing of the 1D-wave with the bulk implies that the propagation is no longer purely 1D. This means that
one must therefore specify both the value of a scale transverse to the 1D-direction
(for
example the sizeof the core in an
optical fiber)
which measures a« distance » from 1D-behaviour and a physical prop- erty of the bulk medium
(the
sheath of the core in anoptical
fiber)
such as its wave velocity or index ofrefraction.
From the general transfer matrix expression, we
obtain the main following new results. i) For 1 D-
periodic
lattices of identical symmetric scatterers, we show that the leakage vanishes at the band edges :this coherent effect stems from the complete destruc-
tive interference between the converted radiations at each scatterer. This result demonstrates that cohe- rent leakage is deeply different from a usual « dissi-
pation » effect. It is interesting to note that this
cancellation of the leakage in the bulk rationalizes the observed quality factors
(as
high as105)
ofsurface acoustic wave
(SAW)
resonators[3, 5, 7, 8].
ii)
We discuss the competition between Anderson localization and coherent leakage in presence of disorder. Near the band edges and in the presence ofdisorder, the « attenuation length » due to coherent leakage is much larger than the localization length Çlac. We suggest an experimental situation
[9]
wherethese effects could be observed. In section 2 we
expose the general formalism, apply it to 1D- periodic arrays in section 3 and discuss the competi-
tion between Anderson localization and the coherent
leakage in section 4.
2. Transfer matrix formalism.
Since 1D-systems are always topologically ordered
[1],
the equations for the wave propagation in any1D-system at any energy or frequency can be cast
into essentially the same transfer matrix formalism
[1].
To be concrete, let us consider a waveamplitude
at point x of the general form
Yl (x )
=Yl
eikx + Yi e- ikx before the scatterer andY2 (x ) = Y2+
ei kx + Yi e- ikx after the scatterer. We look for the general expression of the matrix T defined byIn absence of leakage, symmetry under time reversal
imposes d = a * and c = b *
[10].
In our case, thissymmetry does not hold anymore. In the following,
we assume that the scatterers have a center of symmetry. This spatial symmetry is weaker than the usual time reversal symmetry. In this case, one must verify
which shows that T verifies 03C31
Tal
= T-1 whereThis
implies
c = - b and ad - bc = 1.(Note
that fora simplectic matrix, the space symmetry implies
c = b * = - b which shows that b is a pure imaginary
number b=ig
and la 12 _I 1 b 12
=1. This last re-lation is also obtained by the conservation of the
energy flux).
The transfer matrix therefore takes thefollowing form
Det T =1 leads to
We have yet to write the conservation of energy flux in the leakage case. In this goal, we must specify the
nature of the radiation waves. Let us note
Z ( 03B8 )
the amplitude of the bulk wave radiated in the direction characterized by the polar angle 03B8[11].
Due to the linearity of the equation of propagation, we can write, without loss of generality, thefollowing
linearrelation between
Z(0 ), Yi
and Yi :Equation (3)
is the new ingredient characteristic of the leakage problem where the total bulk radiation is written as the sum of the scattering of the two 1D-waves,
Yi
andY2 ,
which are incident on thescatterer. Due to the linearity of the underlying propagation equations, the theory is independent of
the choice
(Yl , Y2 ) :
another choice such as(Yi , Y1)
can be made. Equation(3)
completely encaptures the role of interference effects on the coherent radiation of 1D-waves since it involveswaves amplitudes. The coherent radiation of a single
scatterer has been thoroughly studied theoretically
[12]
and experimentally[13]
for example for SAW impinging upon a single groove and is characterizedby a complicated radiation emissivity diagram for
the
energy le(O )12 (incident
wave from theleft)
orI e’(0 )12 (incident
wave from theright)
dependenton the geometry of the grooves. Note that equation
(3)
is very general and can be applied to many other situations. For symmetrical scatterers, one hase’(0) = e(v - 0 ).
The total energy converted into bulk radiation is
Energy flux conservation imposes
This equation must be valid for all
Yi
and Y1 and is therefore equivalent to the matrix equationwith
and
where
and
p, -- 0
and P2 are both real(we
consider symmetricalscatterers).
Notethat p2
is a crossed term describingthe interference of two bulk waves, one radiated from a
left-propagating
1D-wave and the other froma right-propagating 1D-wave.
Equations
(2)
and(4)
with the form(1)
yield a non-homogeneous system of two complex and tworeal relations between a, b, d, pl
and p2.
Then, onecould expect the system to depend only on two real parameters. In fact, it is both
surprising
and interest-ing to realize
that P, and p2
can be expressed(putting (3)
and(4)
in(2))
in terms of a and b as follows :This shows that the initial equations,
(2)
and(4),
arenot independent, which was already the case in the
absence of leakage leading to symplectic matrices.
Equations (5)
and(6)
permit one to determine the radiative coefficients pland p2
from the coefficients of the transfer matrix(say a
andb).
Therefore, space symmetry and the conservation of energy flux let free four independent real parameters(a, b ).
The knowledge of the 1D-reflection(bid)
and 1D-trans- mission( 1 /d )
over the scatterer completely deter-mines the system and, in particular, from the meas-
ure of the 1D-propagation properties one can deduce
the radiative coefficients pl
and p2 of
the coherent leakage in the bulk.Additional constraints
(inequalities)
apply on thefour free parameters. First, p, is positive which yields
Second, the bulk radiation energy is also positive or
in other words, the quadratic form obtained with the matrix tT* u 3 T - 03C33 must be negative. This implies
that two additional inequalities must be verified
1686
A chain relation insures that if the quadratic form
obtained with the matrix ’T* (73 T - 0-3 is negative
then the quadratic form obtained with the matrix
tTn* 0’3 Tn - 03C33 is negative for all n. If expression
(8)
is verified as an
equality I a 12 _ I b12 _ 1 = 0,
onerecovers the usual simplectic case without leakage
and we are therefore interested in the cases where
inequality
(8)
holds strickly.If inequality
(9)
holds strictly, it is easy to show that the energy C radiated in the bulk is always positive and does not vanish. In this case, the energy loss per scatterer is of order p, in all situations : this is similar to a wave propagation in the presence of adissipation.
A more interesting case happens when
inequality (9)
holds as anequality
i.e. whenIt is easy to verify that this happens if and only if the parameter a has the following structure
with 0 real
(it
is interesting to note that this recoversthe case studied in
[6]).
Let us notef3 = - (a* b + b* d)
which is equal to /3==-
(b + b *) e- icP
using(10).
Then, if in additionkil) with n being an integer, then C reads
This shows that for a. particular value of the fre- quency such that
i) inequality (9)
holds as an equality,ii)
equation(11)
is verified andiii)
I Yi I - (- l)n e- it/> I Yïl ]
= 0, then the total energy radiated in the bulk vanishes. The simultaneousvalidity of conditions
i)-iii)
may seem drastic but wewill see in section 3 that it occurs at the band edge of
a periodic lattice of scatterers leading to expression
(10).
In the following, we restrict our discussion to the class of problems for which
equation (10)
holds. Thisimplies pi
= P2 = -(b
+ b*)/11 - b 12. Conversely,
if pl = P2
(which
occurs for example when the emissivity diagrame (0 )
is symmetric around7r/2:
e (7r - B ) = e ( B )
as occurs for Dirac point-likescatterers),
then a is of the form(10).
This caserecovers the expression of the transfer matrix
(Eq.
(5)
of[6])
derived within a coupled modetheory
ofthe
propagation
of light waves in dielectric waveguides, as a special case of our general ap-proach. Equation
(10)
implies d =-
(1 - b ) e-‘
and we have three free parameters b = ) + I> and 0.It is useful to clarify the meaning of the different parameters b =- 6 + i g and 0. We consider the case
of inefficient or « small » scatterers such that the
typical value of the reflection coefficient for a single
scatterer is small. This is not a restriction since most
situations fall into this class. Since
1/d
={ ( 1- b ) e- icf>} -1 is
the amplitude transmission coef-ficient, one
expects 11 Id I - 1
which impliesI b I
1 and 0 has the meaning of a pure propagation dephasing. In a periodic lattice of scatterers, 0 is the dephasing over one spatial period. Equation(10)
implies that, = p2 = p = -(b+b*)Ill -bl2 - -2 6111 -bl2. Since lbl -c
1,this gives ) = - p /2 (03BE is negative)
to first order. Now,I bld 1 2 _ 6 2 + 03BC 2
is the energy reflection coefficient. Since p(and
therefore
6)
has the meaning of an energy, it is the square of an amplitude and therefore is much smaller than the amplitude reflection coefficient which we identify of the order of JL. In summary, onehas 03BE - p-
u 2.
To leading order, we havea 12 - 1 +g2-p
and d =(1 +p-igp)a*.
The case ofthe absence of any leakage for all values of 0 is
found with p = 0, leading to b = i u which confirms the interpretation of tk as the leading contribution to the reflection coefficient. The presence of leakage brings a negative real part to b of
order JL 2,
inagreement with the specific model studied in
[6].
To be concrete, consider the case of a Rayleigh
SAW of wavelength A = 2
7r Ik
propagating at thesurface of a solid and impinging upon a groove of width w and depth h. A perturbative-type theory of
the single scattering event
[12],
confirmed by exper- imental measurements[13],
shows that a portionJ.L 2
of the energy of the wave is reflected and aportion p of it is converted to bulk longitudinal and
shear waves. u
21p --
1/20 with tL =0.6
(h /A )
sin(7TW I À) [13]
whereh/A
plays therole of the small parameter
(03BC = h/A
and p -(h/A )2).
Typically, h -- 1 Rm and A =3= 10 >m lead-ing to
h/A
and > in the range 10-1-10- 3.3. Case of a periodic array of scatterers.
Consider a
periodic
lattice of such identical scatterers withperiod
A. The general wave structure in the nthunit cell is
[14] Yn (x ) _ { yo eIk(x - nn) + Yo
e- ik(x -nA)}
eiK(x - nA) e- iKx with K given bys e± iKA and s verifies the secular
equation
s2 -(a + d ) s
+ 1 = 0 leading to KA =cos-1 { (a + d )/2} .
The Bloch equations lead toYo
= b andYo
= s - a. In general, K is complex(and
not a pureimaginary)
which reflects theattenuation of the 1D-wave due to leakage in the
bulk. However, for the particular values 0 = n03C0
with n integer, one finds s =
(-1 )n.
This corres-ponds
to the « inferior » band edges at which themodulus of the eigenvalue s raising to unity reflects a vanishing leakage. This striking result can be recov-
ered explicitly by computing the wave energy reflec- tion coefficient of a wave incident from the left upon
a semi-infinite periodic lattice
[14] :
It is then easy to verify that R = 1 for s = ± 1. The vanishing of leakage at the band edges shows that leakage and dissipation are completely different phenomena. Dissipation is not sensitive to coherence
whereas leakage is controlled by the coherent struc- ture of the 1D-wave. We also verify that conditions
i)-iii)
of section 2, for the leakage C to vanish, areverified for the Bloch waves with s = ± 1 provided §
is negative
(which
is truesince 03BE p /2).
Physi- cally, this coherent effect results from the Bragg phase condition which insures that the guided stand- ing wave has the period of the reflective array but is out of phase with it.Let us now analyse more carefully the competition
between leakage and reflection in the stop
(forbid- den)
bands. Let us consider the first stop band which extends from 0 = 0 to cp = - 2 03BC. In the vicinity ofthe band edge n 2 7T + with 0 small, we obtain
The leakage loss
(plg2) (2 g 10 I )1/2
exhibits asquare root singularity as 0 --+ 0. This result allows
us to understand a result obtained previously in
[5]
for SAW propagating over periodic arrays of grooves, within the numerical solution of a complex approach which integrates a field theory with the theory of two-mode coupling. Note that our general
formalism takes into account the complex structure
of the evanescent modes which may exist in the
vicinity of each scatterer. A difficulty remains in the identification of the band edge with the explicit physical values of the problem under consideration.
Indeed, at
ka
A = 2 Tr(Bragg
reflectioncondition),
one can expect to be exactly at the band edge with a
zero loss. In fact, this is not
quite
true since thepresence of evanescent modes
[15, 3]
created at eachstep
(which
are also solutions of the Rayleigh propagationequation)
lead to a local stocking ofenergy at each grooves which slightly renormalizes the effective SAW velocity according to Cg =
co(l - K g 2) . Therefore, kA = ko A (1
+ K M 2)
and if
ko
A = 2 Tr, this leads to 0 = kA = 2Tr K M 2 with
Kof order unity. Reporting in equation
(14)
leads to1-
R = p (2 K / M )1/2.
Taking the numerical values of[13],
M =10-2, P = 10- 3,
K = 1, one finds thetotal loss due to leakage of the order of 1 % which is indeed very small for scatterers which individually
leak 20 times more than they reflect ! This explains
the high quality factors , 105 of SAW resonators made of two distributed reflective arrays forming an
effective « cavity » surrounding a transducer.
At the center of the stop band, 0 = - g and we
find, using equation
(13), R = 1 - p /,u
+ 0(tk 2)
indicating that the loss due to leakage are of the
order of 4. This result can be understood from a
simple « incoherent » argument valid also in a pure
dissipation case : the typical « penetration » length
of the wave in the semi-infinite periodic lattice, due
to coherent backscattering, is gloc =
dlg
sinceIs I = exp (- dl 61.,,) - e- ".
On each scatterer, the power loss is roughly p =J.L 2.
Therefore,by adding
incoherently the energies radiated by each scatterers,one finds that the total powerloss is p
gloc/ d = p/,u .
Note that asingle
scatterer may leak muchmore than it reflects
(p =
2003BC 2)
but nevertheless the total loss upon reflection of a wave incident on asemi-infinite lattice can still be small if p iz. This type of result is also obtained when equation
(10)
does not hold, i.e., when expression
(9)
holds as astrict inequality.
At the other border, cp = - 2 g and R = 1 -
(2 p /,- )112 +0 (g )
and the loss are of the order of /jL 112 > u. The incoherent argument valid at thecenter of the stop band is not valid here and contrary
to the other band edge 0 = 0 at which the leakage vanishes, the coherence enforces a stronger
leakage.
It is interesting to note that the vanishing value of
the leakage at the band edges can be checked from a
direct computation of the total energy radiated in the bulk under the form of a radiative array.
Interference at « infinity » occurs between the N
leaky bulk waves emitted from the scatterers. The non-conservation of the bulk energy when changing cp must be interpreted as a reactance. It is associated with a recirculation of the 1D-wave power source and must therefore be understood as resulting from
interference effects in the 1D-wave as described
by
the above formalism. The equivalence between
these two view points,
(which
is implicithere),
iswell established for electrical antennas and elec-
tromagnetic dipole radiation. Let us summarize the argument by computing C with the following toy model:
where Yn denotes the amplitude of the 1D-wave at the nth point of radiation and , = cp
{ 1- (Cg/ Cb)
cos9 }
is the phase delay betweentwo neighbouring scatterers, cg and cb being the phase velocity of the guided and bulk wave respect- ively. If the Yn are of unit modulus and of the form
Yn =
einY,
thesum E
ein(E + Y) is always zero exceptn
when E + Y crosses 2 7T. To demonstrate that E = 0, one has to verify that s + Y does not cross
2 7T for 0 going from 0 to 7T. Take for example 0 = 7T corresponding to an odd band edge and
1688
y = 0. Then, the condition of absence of leakage is
obtained as soon as
cg/cb
1. But, this is preciselythe condition for a guided wave to exist : at the
surface separating the « 1D-medium » from the
bulk, the-radiation of a bulk wave in a direction 0 in the absence of scatterers is realized when the projec-
tion on Ox of the bulk wavevector equals the 1D-
wavevector. This is never realized in
cgICb
1 forany 03B8: this is the general fundamental reason why guided waves, surface waves or evanescent waves
propagate more slowly than the corresponding bulk
waves. Note the loose analogy with the « inhibited spontaneous emission » discussed in
[16].
4. Competition between localization and coherent
leakage.
Let us now briefly discuss the influence of disorder.
The disorder can enter in the position of the
scatterers or in their reflective or radiative power.
Consider for simplicity the case of a position disorder parametrized by 7y = k 8A and characterized by a
zero
average 17)
= 0 and a r.m.s. amplitude17 2)
=0.- 2/ g (the
notation follows from ageneralized central limit theorem
[17]).
The T-mat-rix formalism is still valid. It leads to a problem of products of random matrices which has been well studied in the literature
[1, 2].
It is known that evenin the passing bands, the waves are completely
reflected
(«
localization »regime)
by a semi-infinite system. iK is replaced by a real Lyapunov exponenty
= 03BEloc-1
inverse of the localization length which is always positive for any non-vanishing disorder a. Itremains of the order of pL near the center of the stop band whereas, in the « old » passing bands, it is no longer vanishing and reads
[1]
Equation (16)
recovers the general behaviour of the localizationlength g loc = y - 1 - (À / h )2 _ (. - 2
atsmall frequency to. The leakage loss appears to be of the same order as the localisation effect in the old
passing bands as discussed in
[6].
In concrete casessuch as the SAW
propagating
on grooves, it caneven dominate since p = 20 J.L 2. However, a non-
trivial effect occurs when one approaches an « in-
ferior » band edges at which leakage is
vanishing
inthe absence of disorder. In this region, it is known
[18, 17]
that the Lyapunov exponent has a non- analytic dependence upon the disorderBoth facts therefore suggest that Anderson locali- zation should dominate over leakage. This expec- tation can be made more precise with the following arguments.
The scaling
(17)
can be traced[17]
to the fact thatat the band edge, the problem becomes invariant under a dilatation of the lattice spacing. In our case, noting
a = (1 + b ) eiTJ = ei (TJ + JL),
the problem isinvariant under a dilatation on tk -+ 03BC /p if the
variance of the noise is changed according to
u -+ U / p 3/2.
This impliesp = 03C3 2/3
and since theLyapunov exponent y scales as A and therefore as p,
one recovers equation
(17).
Now, radiative lossesare governed by the real part 6 of b. For 6 of order
03BC 2,
, we obtain a typical loss scaling as o -4/3Another
approach
relies on the computation ofthe total energy leakage analogous to equation
(15)
for a localized mode centered at n = 0 :
with E defined in equation
(15).
We find that the total leakage loss is La- =
7T ’Y 2/ (1 - cos £ )2 = u 4/3.
This can also be reob- tained by assuming that an energy leakage of order 0’2occurs on each scatterer as suggested from the quadratic expression(12)
for L. The total loss over a03C3typical
distance ç loe = y
1 is therefore(T 2 ç loe =
413. This shows that the effect of the disorder on the
leakage scales as cr 4/3 whereas it scales with y =
u 2/3 for Anderson localization. The vicinity of the
band edges therefore offers a promising domain for observing clearly localization in experimental situ-
ations where leakage loss occurs. This result offers a non-trivial signature of the competition between two
coherent effects :
i)
Anderson localization andii)
guided to bulk mode conversion which has been studied in this article.Experiments on SAW are presently being devel- oped to test these predictions
[9].
Acknowledgments.
We are
grateful
to J. Desbois, J. Peyraud and B.Souillard for stimulating discussions and particularly
J. Desbois for useful comments on a preliminary
version of the manuscript. We acknowledge financial support from DRET under contract n° 86/177 for the research program «Propagation
acoustique
enmilieux al6atoires ».