• Aucun résultat trouvé

Numerical simulation of random composite dielectrics. II. Simulations including dissipation

N/A
N/A
Protected

Academic year: 2021

Partager "Numerical simulation of random composite dielectrics. II. Simulations including dissipation"

Copied!
14
0
0

Texte intégral

(1)

HAL Id: jpa-00246658

https://hal.archives-ouvertes.fr/jpa-00246658

Submitted on 1 Jan 1992

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.

Numerical simulation of random composite dielectrics.

II. Simulations including dissipation

S. Stölzle, A. Enders, G. Nimtz

To cite this version:

S. Stölzle, A. Enders, G. Nimtz. Numerical simulation of random composite dielectrics. II. Sim- ulations including dissipation. Journal de Physique I, EDP Sciences, 1992, 2 (9), pp.1765-1777.

�10.1051/jp1:1992243�. �jpa-00246658�

(2)

Classification Physics Abstracts

02.60-77.20-77.40

Numerical simulation of random composite dielectrics.

II. Simulations including dissipation

S.

St61zle,

A. Enders and G. Nimtz

II. Physikalisches Institut der Universitit zu K61n, 5000 K61n 41, Germany

(Received

21 May1992, accepted 29

May1992)

Abstract. A simulation model presented earlier for the permittivity of a composite material has now been extended to include the dissipation of energy inside the medium. It is employed to

calculate the complex dielectric function of so-called effective media, I-e- randomly distributed lossy particles in an insulating matrix. A general equation formulated before for lossless media

proves to be valid also for the dissipative mixtures investigated. However, simulations and exper-

imental data show that small deviations from random distribution can induce large deviations in the dielectric function. Experiments are reported that agree with the simulation results.

1. Introduction.

The effective

electromagnetic

parameters of

quasi-homogeneous

mixtures of different materials have been

subject

to a

large

number of

investigations

since the

beginning

of this century

([1- l8]).

In a

binary

mixture for

example,

where a volume fraction

f

of

particles (of permittivity cl

is

randomly

distributed in a matrix

(of permittivity em),

the effective dielectric function

(DF)

T is

generally

not a linear but a

complicated

sublinear function of

f,

em and e.

There have been many

attempts

for the

analytical development

of a consistent

theory,

which

describes the

physical reality

and

provides

a tool for the

interpretation

of measurements on such 'effective media'. But these theories and the

respective

formulas fail in

describing experimental

data

jig, 20], although qualitatively they

show a similar sublinear behaviour. Now as computers grow

bigger

and

faster,

a few authors have

developed

computer models for such systems and

numerical simulations of effective media have been carried out,

using

different

techniques [15- 18].

However, up to now these efforts have not led to a

knowledge

that

helps experimentalists

to

interpret

their data

unequivocally.

In [21] we have

presented

a new computer code COSME

(Complex

Solution of Maxwell's

Equations),

which was

employed

in the simulation of

binary

mixtures. We have

compared

our results to the results calculated

by

several effective medium formulas.

Although

this

procedure

did not even include lossy

dielectrics,

the

respective

formulas failed in most cases. In contrast to them we found a relation between

f,

e, em and T with a strong

similarity

to

Looyenga's

formula [7], which is valid for

f

< 0.25:

(3)

1766 JOURNAL DE PHYSIQUE I N°9

p(f)

=

f ~a(f)

+

(i fj ea(f) (ij

where

a( Ii

=

(1.60

+

0.05)

*

f

+

(0.265

+

0.005)

In this paper we report the continuation of our

investigations,

now

including lossy

dielectrics.

Apart

from

studying

the behaviour of the

complex

effective

permittivity

in the whole volume fraction range between 0 and 1, we compared

equation (ii

to the

complex

results and

finally

we carried out

experiments

to prove the

validity

of the simulation program.

2. The

algorithm.

The basis of the simulation is the

following physical

model: a volume of size 20 x 10 x 20

mm~

or 20 x 15 x 20 mm~ is surrounded

by ideally conducting walls, forrrdng

a

cavity

resonator. The solution of Maxwell's

equations

inside this volume

including

all

boundary

conditions leads to

electromagnetic

resonances characterized

by

certain distributions of the electric and

magnetic

fields

oscillating

with a

corresponding

resonance

frequency.

The solution for

homogeneous

materials is

easily

found

analytically [22].

For the numerical solution for

inhomogeneous fillings

the volume inside the resonator is covered with a 3D cartesian mesh such that every mesh cell contains

only

one kind of material whereas the dielectric

properties

for different mesh cells are uncorrelated. Thus for each cell

we select

randomly

a

permittivity

e with

probability f

and a dielectric constant em with

probability

I

f.

Using

a finite difference method for the fields E and B

interdependent

via Maxwell's

equations

the

problem

is transformed into

a discretized

problem,

whose solution is

equivalent

to the solution of the

following eigenvalue equation:

Ae =

qDee

q =

A~

co POw~

(2)

Here the matrix A

represents

the discrete

rotrot-operator,

the vector e the discretized electric field vector, De is a

diagonal

matrix

containing

the different

permittivities

of all the mesh cells. q is the

eigenvalue

to be calculated

containing

A = unit

length

of the mesh cells,

co, PO "

Permittivity

and

permeability

of vacuum and w

= resonance

frequency,

at which the

field distribution e oscillates inside the resonator

(for

details see

[21]).

The above

eigenvalue problem

may be solved more

easily

whenever De is a real

matrix,

because then the

problem

is

symmetric

and real. As

reported

in [21] such a

problem,

even if it is of very

high

order

(here:

>

150000),

can be solved

by

a

spectral

shift of

eigenvalues

combined

with inverse iteration

employing

a

conjugate gradient (CG)

method for the minimization. For the simulation of

dissipative

mixtures where De is

complex,

standard CG routines fail as

they

are

only applicable

to Hermitian matrices. So a

recently published

CG

algorithm

[23] was

included in

COSME,

which works

especially

well with

complex-symrnetric

but non-Hermitian minimization

problems,

hence it is ideal for the solution of

equation (2).

The existence of

complex eigenvalues

q also causes the

frequency

w «

@

to be

complex.

Just like the

complex

value of the wave vector k of waves

propagating

inside an infinite

sample,

the

complex

value of the oscillation

frequency

w of

standing

waves inside

a resonator may be

used to calculate the

complex

DF of the

respective

material. The

corresponding

relation is

~2

0

~ ~2

(4)

14

7'

12

1o 8

70000 140000

li

5

-, ,

~ 4

I

70000 140000

N

Fig-I.

Effective permittivity vs number of nodes in the discretization mesh.

e = 50 -120, em

= 1

and f = 0.3.

where wo represents the resonance

frequency

of the empty resonator.

For the measurement of w = w' + iw" one would measure the resonance

frequency (w')

and the

quality

factor

Q,

which is

given by

w' divided

by

the FWHM

(full

width half

maximum);

w" is then calculated via the

following

relation

~,, W' W'

(~j

2Q 2Qofi$

resulting

from the time

dependence

of the electric field

(and Q

cc

vGfi:

E(t)

=

Eoe~#e~~"~e~~'~

=

Eoe~%de~~'~

Here

Q

is the measured

quality

factor of the resonator

including

a

(dissipative)

material and

Qo,

wo are the

quality

factor and resonance

frequency

of the empty resonator. In the ideal case

(infinitely conducting walls) Qo

" cc and w"

=

w'/(2 Q).

3.

Dissipative

random mixtures with different

filling

factors.

For all the simulations the size of the discretization mesh was chosen 40 x 30 x 40

respectively

40 x 20 x 40

nodes,

so the volume was divided into cells of

length

A

= 0.5 mm. This of course

corresponds

to the smallest

particle

size realized in the simulations.

For the estimation of discretization errors we carried out several simulations of identical dis- tributions of

particles

of

length

d

= 2 mm

varying only

the size of the discretization mesh. In

figure

I the

resulting

effective

permittivities

for meshes of 10 x 6 x

10,

20 x 12 x

20,

40 x 24 x 40 and 60 x 36 x 60 nodes are

displayed

vs the number of nodes. In this case e

= 50 -120 and

f

= 0.3 were

chosen,

the same behaviour was observed for e

= 15 I and

f

= 0.4. The

two smaller

grids appeared

to be too coarse to

produce

correct values.

Increasing

the

grid length

from 40 to 60

obviously

does not

change

the result very much

compared

to the statis-

tical fluctuations, so for standard runs we

always

chose 40 x 30 x 40

grids

to save computer time.

(5)

1768 JOURNAL DE PHYSIQUE I N°9

, >,, i I I I i i i i i ,

, >,, i I I I I I I,, i i , ,

, >, i I, t I I II I, i I I i

, ,

, , t t i i t I ii I I i i t It i

t i i ,

, , , i i i i t I ii I I i t

i, i i i ,

,, i i i i t t I I I I I i Ii t i i t i i i ,

,

I,

, , t t i i I I I it i t t i

, ,

i , i t i t t i I i i i t t i i

i,

, , t i i i i Ii i I i i i i i I I i t,, , ,

, , t t i t t t i t, i i ,

, , t t t t I it I I I I i t t i , t , ,

i i i t I I i t I t t ,, ,,.

, i i i i it i Ii I i I i i, t ,.

, , , , i I i i, t t i i i i I I i I I t i i t ,

, , , ,, i I I I i i I I ii i i I i

, , ,

, , , ,, t t I i i I I I I Ii t t I I t , i > i i i ,

, , i i I I i i i I I I i t i i , , i ,

<, i i i t I I i i t i i1, ,

Fig.2.

Typical electric field distribution inside a resonator filled with an effective medium

(cross section).

Fields calculated by COSME.

One can

speak

of an effective medium as

long

as the

wavelength

inside every

particle

con-

siderably

exceeds the

length

d of the

particle.

If this condition

W>~

> (41

is not

fulfilled,

the wave may be scattered

by

the

inhomogeneities

and the mixture does not behave

quasi-homogeneously

any more. In this context the

problem

of the skin effect also has to be

considered,

as the

complete penetration

of the wave inside the

particle

must be ensured.

In the worst case in our simulations

,

Al

(e( was still

larger

than 7 d. The maximum

imaginary

part of e was chosen such that the

resulting

skin

depth

was still 0.75 mm, I. e.

larger

than the

length

of each

particle.

For

illustration, figure

2 shows a

typical

field distribution

(electric field)

at resonance inside

an effective medium as calculated

by

the COSME program. The

sine-shape

of the fundamental mode of the

rectangular

resonator is

clearly visible, inspite

of local field deformations

as a cause

of the

inhomogeneities (em

= I,e = 10

2i, f

=

0.I,

d

= 0.5

mrn).

Mixtures of

particles

with e

= 50

-120,

em = I and d

= 0.5 mm were

investigated

over the whole range of volume fractions between 0 and I.

Looking

at the matrix

equation (2)

it is easy to see that any

arbitrary

mixture of two materials may be transformed into a

particle-vacuum

mixture

by normalizing

all the

permittivities

to the DF of the host material.

Figure

3 shows the behaviour of the effective DF of the mixtures vs volume fraction. For small

filling

factors

(f

<

0.25)

the

typical

sublinear behaviour of

T(f)

is

observed,

below

f

= 0.12 even very close to the line which marks

Looyenga's

formula. For

f

>

0.25, however, T( ii

shows a

perfectly

linear

behaviour,

which reminds of a

simple parallel-circuit

relation.

In

figure

4 this behaviour becomes even more obvious: the effective loss tangent

f'If

= tan

changes dramatically

in the volume fraction range below

f

= 0.25

approximating

the

particles'

loss tangent

e"le'= 0A)

very

rapidly.

At

higher

volume fractions it almost does not

change

any more.

So the loss tangent and the absolute value of the DF of such an effective medium may be influenced

independently

in the different ranges of

filling

factors. Below

f

=

0.25, ?( ii

is sublinear in both the real and the

imaginary

part, hence

iii changes

very

slowly

with the volume

fraction,

in contrast to the loss

tangent.

In the

high

volume fraction range the two parameters act vice versa: almost no

change

in tan &, linear increase of

iii by adding

a few

particles

to the mixture.

(6)

60

E'

50

o

40

.

o

30

20 °

o

IO °

o

-lo

OO O-1 02 03 04 05 06 07 O-B 09 1.O

f

a)

25

E 20

o

5 *

o lo

o

5

o

o

-5

O f b)

Fig.3.

Dielectric function of the heterogeneous mixtures vs. the volume fraction (em

= 1,e

=

50 -120, d

= 0.5

mm).

Results of the numerical simulations

(.)

compared to Looyenga's formula [7]

(-). a)

Real part of I

b)

Imaginary part of I.

In

figure

5 the simulation data and the results calculated from

equation (I)

are

compared.

In the range where

(I)

is claimed to be valid for

f

< 0.25 the agreement is excellent. For further

proof

we carried out simulations with

particles

of even

higher permittivity (e

= 90

145)

and found the same agreement. So

obviously equation (I)

is also valid for

dissipative

random

mixtures with low volume fractions.

(7)

1770 JOURNAL DE PIIYSIQUE I N°9

o.5

tg 6

O.4

~ . .

O.3 O.2

O.

o-o

-o.i

O-O

f

(8)

20

6

15

1o

5

-O

f a)

'>

~

-2

-4

-6

-8

O-O

f b)

g.5.

-

equation (-) vs. volume fraction. em, e, d as in figure 3.

calculation

of the

mean polarization, can only up to the sth ntil the limits

of theomputer are

reached

[15].

So the interactions between neighboring

particles

roaches and therefore they

can only

be

valid

for very small

filling

non-random samples.

The

above

sample being higher when the

sth

multipole was included,

than

in pure dipole

(9)

1772 JOURNAL DE PHYSIQUE I N°9

~)

~@~ ~4/

)

a)

C)

Fig.6.

Non-random geometries simulated by COSME.

Obviously higher multipole

interactions have

a strong influence on the DF of a mixture.

In the simulations

presented

here the interactions between all the

particles

inside the

sample

volume are

implicitly considered,

because here Maxwell's

equations

are

directly

solved in full

generality.

This results in the decisive deviations from former well-known effective medium

relations.

4. Non-random

geometries.

The formation of connected

paths

can

easily

be enhanced or

suppressed

in the computer model.

For a

qualitative

evaluation of the influence of geometry on the

resulting

effective DF we

generated

the

following

material distributions for five different simulations:

I. Several 2D-networks on a stack but

separated by

matrix material. The networks lie

perpendicular

to the electric field inside the resonator

(Fig. 6a), f

= 0.107.

2. 2D-networks as in I. interconnected

by

columns in the direction

parallel

to the electric field

(Fig. 6b), f

= 0.15.

3. 2D-networks and columns as shown in

figure 6c, f

= 0.164.

4. As

2.,

with

f=0.23.

5. As

3.,

with

f=0.28 (due

to closer

packed 2D-networks).

Figure

7 shows the deviation of the

resulting

DF'S of such structures from those of random mixtures. From table

I,

which contains the

corresponding

relative deviations for

geometries

1.

5.,

it is obvious how

strongly

the inner

geometry

of the

inhomogeneities

influences the effective

DF,

an effect which has also been found

experimentally [24]. Especially

a

higher proportion

of

paths parallel

to the electric field increases the

resulting

effective

permittivity

whereas r results to have a lower value when the

inhomogeneities

are located

perpendicular

to the electric field.

(10)

20

lo z 4 5

. i

O f

O-O O-i O.2 O.3 O.4

a)

EI,

o

3

~ O

~

_5 .

4

-lo f

0.O O.2 O.4

b)

Fig.7.

Results from the simulations of five different non-random geometries

(.) (geometries

1.

5., see

text)

compared to the respectiv of random geometries

(-),

both displayed vs. the volume fraction.

Table I. Relative deviation of the values air calculated for non-random

geometries

from those for mixtures with

randomly

distribu ted

particles (emd ).

Geometries 1. 5. are

explained

in the text. The

corresponding

absolute values are shown in

figure

7. In this table 100 ~

corresponds

to the effective

permit tivity

for random mixtures

(for

the

respective

volume frac-

tion).

No.

f

f T[~d deviation T"

f[d

deviation

T' from T[~d T" from

f[d

1 0.105 1.3 3.34 -61 iii 0.025 0.76 96 iii

2 0.15 8.9 5.734 + 55 iii 3.84 1.82 + l10 iii

3 0.164 7.6 6.72 + 13 iii 3.19 2.29 + 39 iii

4 0.231 14.5 13.087 + 10 iii 6.74 5.52 + 22 iii

5 0.237 II-1 13.785 19.5 iii 4.86 5.89 17.5 iii

(11)

1774 JOURNAL DE PHYSIQUE I N°9

5.

Experiments.

For the verification of the computer simulations several

inhomogeneous samples

were

prepared

and a real resonator was built for the purpose of a direct

comparison

between

physical

mea-

surements and results obtained from the numerical calculations.

The first

sample

was a block

(16

x 7 x16

mm~)

of

A1203 (96ili purity),

which was

placed

in the center

(setting A)

and then into the corner

(setting B)

of the resonator. These

geometries

are

easily reproduced by

the discretization mesh of the computer program and thus serve as very

good

test structures for the simulation results. As a second

sample

a mixture of epoxy resin

(em

= 2.98

-10.06)

and 8ili small cubes

(I

x I x I

mm~)

of a

Ti02

ceramic material

(e

= 80

I)

was

prepared

to fit

exactly

into the resonator and

according

to a random distribution

generated by

the computer

(setting C).

As a consequence of technical

problems

while

placing

the

cubes,

the desired random distribution could not be realized. So the distribution for the simulation had to be modified to represent the inner

topology

of the

sample correctly.

Inside the

rectangular

resonator

(20

x 10 x 20

mm~)

constructed for this purpose two current

loops

for H-field

coupling

were fixed

opposite

to each other in the side walls.

They

were realized

by shortening

the end of two coaxial lines

by

thin wires in the

plane

of the inner resonator

wall. Two PC-7

(7

mm

precision connector)

bulkhead connectors were mounted

directly

on the resonator side

walls,

their inner conductor

going through

holes in the resonator walls thus

realizing

the coaxial lines.

Thus the transmission

properties through

the resonator could be determined

using

a HP 85108 network

analyzer.

The network

analyzer

was calibrated with a

high quality

LRL

(two

7 mm air line and one reflect calibration

standards)

calibration

technique

in PC-7

technique [27]. Consequently

all

systematic

errors up to the resonator PC-7 connector interfaces were

removed

yielding highest possible

measurement accuracy.

First a broad

frequency

range was chosen to

identify

the transmission resonances, then the calibration was

performed

in

frequency

segments around the resonances to

give

sufficient

frequency

resolution

(down

to

frequency

steps of 200 kHz between each measurement

point)

for the determination of the resonance

frequencies

and the

Q-values. (-3

dB

points

of the

resonance

curves).

The

coupling strength

of the current

loops

was varied

by altering

the geometry of the short circuit wires thus

yielding

transmission

amplitudes

between -60 and -25 dB at the resonance

frequency

of the empty resonator. In this whole range the resonance

frequency

did not

change by

more than

10~~.

Also the

Q~value

of the empty resonator did not

change

within measure-

ment resolution. This proves that the disturbance of the

coupling

structures on the resonator

properties

is

negligible.

The resonator was

successively

loaded with the

geometrical settings A,

B and C and for each of them the lowest resonance

frequency

was determined.

In table II the results of

experiments

and simulations of the three

settings

are

directly compared.

The

imaginary

part ofw is related to the

quality

factor

Q

via

equation (3). However,

in

settings

A and B w" was below the resolution limits and has therefore been omitted.

Figure

8 shows the resonance curves of the empty resonator and of the resonator loaded with

sample

C.

The

good agreement

of cases A and B between

experiment

and simulations proves the

validity

and correctness of the COSME-results.

However,

for both these cases there is no effective DF because of the violation of relation

(4).

In case C the construction of a random distribution failed as it was

impossible

to

bring

the cubes into close

enough

contact: in the real

sample neighboring

cubes were still

separated by

a resin

layer.

Instead of v

= 5.354 +10.052 GHz a

really

random effective medium with the same

filling

factor would have had its resonance at v = 4.921 +10.047 GHz.

(12)

S [dB]

resonator

5

dB(

S21

a)

S12

ma'

10[Ge99s9932

adz v

[GHZj

S

sanipie

c

s

b)

S12

START 5.ZSOOaOOOO G4z

~~

WCP 5.4100@OOOO G4z l' Z

Fig.8.

Measured resonance curves: S21-parameter

(I.e.

transmission amplitude

S)

vs. frequency:

a)

empty resonator

b)

resonator filled with ceramic-resin mixture

So the inner

topology

of a

heterogeneous

mixture is a very

important

parameter for the

resulting

effective material parameters, as has been

pointed

out before

[14, 9].

Not even

equation (I),

which has

proved

to be correct for

completely

random

mixtures,

can be

employed

on

samples

with unknown inner geometry.

(13)

1776 JOURNAL DE PHYSIQUE I N°9

Table II. Results of the measurements on different resonator

fillings (vexp, Qexp)

in com-

parison

to the

complex

resonance

frequencies (vsim)

calculated

by

the simulation program. In

settings

A and B the

imaginary

parts of e and w were below the limit of resolution and have therefore been omit ted. The

experimental

data ofset tin g C

correspond

to a

complex frequency

of v = 5.354 + 10.052 GHz.

Setting

e I

vexp[GHz] Qexp vsim[GHz]

empty I 1. 10.595 2682 10.596

A 8.9 0.448 5.43 1357 5.448

B 8.9 0.448 5.047 1264 5.050

C 80 I 0.08 5.354 50.509 5.311 + 10.053

epoxy 2.98 10.06 1.0 6.2275 47.904

resin

6, Conclusion.

A new computer code has been

presented

which allows the simulation of

inhomogeneous lossy

materials. In this paper it was

employed

on one

species

of effective media:

randomly

distributed

dissipative particles

in an air-like matrix.

For the effective

permittivity

a sublinear

dependence

on the

filling

factor

f

was found for the range

f

< 0.25. The

empirical exponential

formula

(I),

which had been

developed

earlier for lossless

composite materials,

has

proved

to be valid also for random mixtures of

dissipative

dielectrics. For

f

< 0.25 it fits all the data very well.

Mixtures with a

higher

volume fraction showed a linear increase of T with

f.

This behaviour may be

explained by

the

gradual

formation of

interconnecting paths by

the

particles.

It

implies

the loss

tangent reaching

its

largest

value

already

at about

f

=

0.25,

whereas the absolute value of T increases much slowlier.

The

COSME-program

was tested

experimentally

with measurements carried out on three different resonator

fillings,

which were easy to model on the computer. Two of them were realized

by

a

macroscopic A1203-cube

in different

positions

and the third

sample

was a mixture of small cubes of ceramic material and a matrix of epoxy resin. The

experimental

data the

resonance

frequency

and

quality

factor of a resonator loaded with the

respective sample

were in excellent

agreement

with the results of the computer simulations.

The measurement on the ceramic-resin-mixture revealed the great

dependence

of the effective

permittivity

on the inner geometry of any

sample. Although

it was a disordered

mixture,

the distribution of the

particles

was not

completely

random but the

particles

were more

separated

from each other. For this reason the

sample

did not show the behaviour

predicted by

the

exponential

formula.

In conclusion it must be stated that the

geometrical

correlations have to be taken into ac- count, whenever

experimental

data are to be

interpreted using

effective medium theories. This

implies

the

necessity

of numerical simulations in the

general

case. For the case of a

completely

random distribution of

particles

in a

matrix, however,

we have

presented

mathematical rela- tions between the relevant parameters which describe the behaviour of such effective media in the whole range of volume fractions.

(14)

Acknowledgements.

R. Pelster is

gratefully acknowledged

for the additional low

frequency

measurements on all the materials used in the

experimental

part. The

investigations

were

supported by

the Deutsche

Forschungsgemeinschaft (SFB 341), by

the Bundesministerium fir

Forschung

und Technolc-

gie/Bonn

and

by

about 120 hours of NEC SX-3

cpu-time

of

Cologne University's

computer center.

References

[1] van Beek L.K.H., Prog. Dielectrics 7

(Heywood Books) (1967)

69-114.

[2] Bergman D., PJ1ysica A 157

(1989)

72.

[3] Landau L.D., Lifshitz E,M., Electrodynamics of Continuous Media,

(Pergamon

Press,

1960).

[4] Maxwell-Gamett J-C-, PJ1iJos. Trans. R. Soc. A 203

(1904)

385.

[5] B6ttcher C.J.F., Theory of Electric Polarization

(Amsterdam:

Elsevier,

(1952)

p. 415.

[6] Bruggeman D-A-G-, Ann. PJ1ys. 24

(1935)

636.

[7] Looyenga H., Physica

(Utrecht)

31,

(1965)

401.

[8] Lichtenecker K., Phys. Zeitschr. XXVII

(1926)

115.

[9] McLachlan D-S- et al, J. Am. Ceram. Soc. 73

(1990)

8, 2187.

[10] Nelson S-O- and You T-S-, J. PJ1ys. D Appl. PJ1ys. 23

(1990)

346.

[11] Ilgenfritz G. and Runge F., PJ1ysica A181

(1992)

69.

[12] Marton J-P- et al, Phys. Rev. B 4,

(1971)

271.

[13] Claro F. and Rojas R., Phys. Rev. B 43,

(1990)

8.

[14] Bergman D., Les Methodes de

l'homogenisation (Edition

Eyrolles,

1985).

[15] Sheu S.-Y. et aJ., Phys. Rev. B 42

(1990)

1431.

[16] Webman I. and Jortner J., Phys. Rev. B11,

(1975)

2885.

[17] Straley J-P-, Phys. Rev. B15

(1977)

5733.

[18] Cukier R-I- and Sheu S.-Y., Phys. Rev. B 42

(1990)

5342.

[19] Marquardt P. and Nimtz G., Phys. Rev. Lett. 57,

(1986)

1036.

[20] Marquardt P. and Nimtz G., Phys. Rev B 40

(1989)

7996.

[21] St61zle S. et al., J. Phys.I France 2

(1992)

401.

[22] Jackson J-D-, Classical Electrodynamics

(John

Wiey & Sons Inc., New York, 2nd Ed.,

1974).

[23] van der Vorst H-A- and Melissen J-B-M-, IEEE Trans. on Magn. 26

(1990)

706.

[24] Pelster R. et al., Phys. Rev. B 45

(1992).

[25] Lagarkov A-N-, to be published in J. Wave Propag. Applic.

(1992).

[26] Staufler D, and Aharony A., Introduction to Percolation Theory

(Taylor

and Francis, London,

1992).

[27] Eul lI.-J. and Schieck B., IEEKMTT 39

(1991)

724.

Références

Documents relatifs

On the one hand, although central limit theorems for branching random walks have been well studied and and the asymptotic expansions for lattice BRW were con- sidered recently by

Faisant l’hypoth` ese fr´ equentiste que les donn´ ees sont une r´ ealisation du mod` ele statis- tique pour une valeur θ ∗ du param` etre (inconnue), cette approche recherche

We consider random rooted maps without regard to their genus, with fixed large number of edges, and address the problem of limiting distri- butions for six different

We say that a specific pattern M occurs in a tree T if M occurs in T as an induced subtree in the sense that the node degrees for the internal (filled) nodes in the pattern match

to be calculated using the program COSME. Above a mesh size of 40 x 30 x 40 the values did not change considerably any more. To save computer time in general all our models

In these last years, several studies using the direct numerical simulation (also referred to as fully resolved or particle-resolved DNS) have been carried out in order to

For general models, methods are based on various numerical approximations of the likelihood (Picchini et al. For what concerns the exact likelihood approach or explicit

Consequently, we show that the distribution of the number of 3-cuts in a random rooted 3-connected planar triangula- tion with n + 3 vertices is asymptotically normal with mean