• Aucun résultat trouvé

Virtual phonon exchange in glasses

N/A
N/A
Protected

Academic year: 2021

Partager "Virtual phonon exchange in glasses"

Copied!
13
0
0

Texte intégral

(1)

HAL Id: jpa-00208320

https://hal.archives-ouvertes.fr/jpa-00208320

Submitted on 1 Jan 1975

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.

Virtual phonon exchange in glasses

J. Joffrin, A. Levelut

To cite this version:

J. Joffrin, A. Levelut. Virtual phonon exchange in glasses. Journal de Physique, 1975, 36 (9), pp.811-

822. �10.1051/jphys:01975003609081100�. �jpa-00208320�

(2)

VIRTUAL PHONON EXCHANGE IN GLASSES

J. JOFFRIN and A. LEVELUT

Laboratoire d’Ultrasons

(*),

Université

Pierre-et-Marie-Curie,

Tour

13, 4, place Jussieu,

75230 Paris Cedex

05,

France

(Reçu

le

5 février 1975, accepté

le 9 avril

1975)

Résumé. 2014 Cet article montre

l’importance

de l’interaction indirecte produite par l’échange de

phonons

entre les défauts des matériaux

amorphes.

Le

couplage

entre les défauts et les phonons

étant fort, il en résulte un

couplage

indirect intense; en

particulier,

le temps de relaxation transversale

produit

par cette interaction est très court (T2 ~ 10-9 s). Les mesures d’atténuation ultrasonore

qui avaient été analysées antérieurement sous

l’hypothèse

2 T1 = T2 sont réexaminées avec ce nouveau point de vue.

Abstract. 2014 In this article we demonstrate the

importance

of the indirect interaction via the

phonon field between

pairs

of defects in an amorphous material. Using only the

approximate

value

of the lattice

coupling

constant it is shown that the exchange interaction is large. The most

important

conclusion that can be drawn from this calculation is that the order of magnitude of the T2 relaxation

time of a spin

packet

is quite small

(10-9

s). It is therefore not

possible

to

interpret

the

experimental

data under the

hypothesis

2 T1 = T2 hitherto used; we re-analyse the available results from this

new

point

of view.

Classification

Physics Abstracts

7.142 - 7.270

1. Introduction. - The

phonon

field has

long

been

recognized

to be an efficient

coupling

of

particles

or excitations in

crystals.

The most famous

example

is the case of

superconductivity

where a

pair

of

electrons is bound

by

virtual

exchange

of

phonons.

The same mechanism has also been invoked in the

case of

paramagnetic impurities

in

crystals [1] ;

however because of the small value of the

coupling

constant of each

spin

with the

lattice,

even for non-

Kramers

ions,

this mechanism

yields

a

negligible

contribution to the linewidth measured in E.P.R.

[2].

On the contrary, in the case of helium 3 or 4

impurities

in helium 4 or 3

single crystals,

the

coupling

between

the

impurities

via the

phonon

field seems to be rather

strong

and probably

contributes to the anomalous value of the diffusion coefficient of

impurities,

as

observed in N.M.R.

[3].

One

might

also ask whether the anomalous value at low

temperature

of the dielectric constant of KCI

crystals containing

mole-

cular

impurities (OH-)

is not due to the appearance of a collective effect ordered

by

the

phonon

field

[4].

In view of these

examples,

the

following question

can be raised : how does the

phonon

field govern certain

properties

of

amorphous

materials

and,

in

particular, following

the model of Anderson et al.

[5]

and

Philipps [6],

the value of the

T2

relaxation time of the

spin population ?

It is

generally

admitted that an

amorphous

material

may be

thought

of as an

assembly

of defects with

randomly

distributed characteristics. Each of these defects has two conformation states

separated by

a

potential barrier,

crossed

by tunneling. Thus,

each

defect may be

represented by

an effective

spin -1

in

a local

magnetic

field whose

intensity

and orientation vary from site to site. The energy

separation

2 E of

two states of the

assembly

extends over a

large

range of energy ; it has been established

by

thermal measu-

rements at low temperature that the distribution

n(E)

of

spins

per unit energy varies very

smoothly

with E until a maximum value of the order of 1

eV, giving

a linear T variation for the

specific

heat

[7, 8].

The other

types

of

experiments

which have been

performed

on

glasses

are

transport experiments :

thermal

conductivity [7, 8]

and acoustic propaga- tion

[9].

These

experiments give

information on the

dynamical

behaviour of the

spin

system

(relaxation

of the

population,

mean free

path

of the

phonons etc...)

and all of them are related to the

change

of

properties

of each

spin

as the strain associated with the

phonon

varies.

In

fact,

any strain of the environment of the defect modifies the

parameters

of the double

well ;

conse-

quently,

an elastic wave may induce resonant transi- tions between the

spin

levels

provided

that the fre-

quency of the

phonons

is on

speaking

terms with

the energy 2 E of the

spin.

Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:01975003609081100

(3)

Following

this

simple model,

the

dynamics

of the

spin

system is

conveniently

described

by

the same

concepts

used in a

paramagnetic

system :

spin

popu-

lation, longitudinal (Tl)

and transverse

(T2)

relaxa-

tion

times,

line

saturation,

self-induced transparency for the

propagation

of an acoustic wave in a resonant

medium,

etc...

However, until now the results of

experiments

in

amorphous

materials have been

interpreted

with the

use of the a

priori

relation 2

Tl

=

T2 [9, 10] ;

such

an

approach

is attractive because

magnetic dipolar

interactions or

exchange

interactions do not occur

here for fictitious

spins.

However,

we would like to

investigate

the follow-

»ing problem :

due to the

large spin-phonon coupling

observed in most of the

glasses,

what is the influence of the

phonon

field on the static

properties

of the

spins ?

In

particular,

what is the indirect

coupling

induced between two

spins

and what are its conse-

quences ?

In

fact,

as we show

below,

the indirect interaction is

easily represented by

an

exchange

constant between

pairs

of

spins;

we shall first calculate the order of

magnitude

of this interaction.

Once it has been

estimated,

we shall calculate more

carefully

the différent relaxation

times;

a

technique developed by

Van Vleck

thirty

years ago for a mixture of

spins [11]

will be used. Because

T2

appears to be smaller than

Tl,

we are led to recalculate the beha- viour of an acoustical wave in such a resonant medium.

Finally, currently

available results whose

interpreta-

tion

depended

until now on the

hypothesis Tl T2

will be

re-interpreted

within the framework of our

theory.

In this way, some

apparently contradictory

data can be reconciled.

A few elements of the

theory developped

here are

not

entirely

new, but

they

are included in order to

present

a self-contained paper, from the

starting equations

to the numerical estimation of the para- meters introduced in the model.

The difficult

question

of the anomalous propaga- tion of a wave in a resonant medium when the sta-

tionnary

state is not reached is

beyond

the scope of this

article ;

this is the

specific problem

of self-induced

transparency.

2.

Spin-phonon coupling.

- The first

step

is an exhaustive

description

of the

coupling

between one

spin

and the

phonons.

We examine

successively

the

microscopic

mechanism of the

coupling

and the

tensorial

aspect

of the

coupling

constant.

Any

strain of the environment of a defect results in a modification of the different

parameters

of the double well. In the notation of

Phillips [12] they

are :

2 J,

the

asymmetry

of the double

well ; 1,

the distance between the two local

minima ; Vo,

the barrier

height (Fig. 1)

and

Q,

the oscillation

frequency

in an indi-

FIG. 1. - Double-well potential with asymmetry 2 J, barrier height Vo and distance 1 between the two minima.

vidual well. The last three

parameters

combine to

give

a

single parameter A,,

the

coupling

energy :

where

When there is no strain the relative

energies

of

the two

spin

states are :

Thus,

in the basis which

diagonalizes

the static

hamiltonian

Ko,

the

coupling

Hamiltonian is

simply

8 is a

component

of the strain tensor.

Using

the

spin

operators

Sx, Sy, Sz

for a

spin -1

formula

(3)

may be written :

Also

The formulas

(4)

must not be

misinterpreted :

time reversal invariance for

example

is

required

for

4

and therefore also for the

operators Sx... Sz

which are not true

spin operators ;

on the other

hand,

X, Y,

Z refer to the axes for the

spin

space, whereas

(4)

oc,

B

= x, y, z refer to

geometrical

axes of the

material.

The

latter,

for the sake of

consistency,

must be

kept

similar

throughout

the

spin assembly,

while the

former are local

spin

axes ; there is no

simple

relation

between the two systems of axes.

Later on,

however,

it will be easier to

keep

the

same systems of axes from

spin

to

spin ;

the

price

that must be

paid

for this is a

change

in orientation of the tensor of the

coupling

constants from site i

to site

j.

As a

result,

for each

spin

there is a different

spin-phonon

tensor

G,m,

with no

simple

relation

among these tensors,

except

that

they originate

in

the same

GafJ,m given

in

(4)

so

long

as the symmetry

of the defects is the same.

Finally,

we must

emphasize

that the

spin popula-

tion defined

by

the energy difference 2 E is

composed

of

spins

with different characteristics : A and

Jo

are related

only by (3). Consequently,

the

G

i tensors

are different not

only

because local axes are tilted but also because A and

Jo

are not the same ; in this

sense the

spin packet

of energy 2 E is not

homogeneous

even if it is so within the E.P.R. standards

(inhomo-

geneous

coupling).

3.

Spin-spin coupling.

- It has

already

been known

for some time that the indirect interaction between

spins i and j

leads to a non

vanishing coupling

term

JCi’j nt only

if the

elementary

process

corresponds

to

creation and annihilation of a

phonon

of the same

mode

[2]; otherwise,

interferences reduce the pro-

babîlity

of the process to zero.

As a consequence, the energy reduction of one

spin

must be

compensated

for

by

a

corresponding

increase in energy of the other

spin.

This rule is

quite simple :

it means that the

only

resultant terms

in

3CO

are the secular tenns, those which commute with

x- 0

+

icio.

The first one is

where i

and j

refer to like

spins (Ei

=

Ej)

and also to

unlike

spins (Ei * Ej).

The second one is

Here, i and j

are

necessarily

like

spins.

The cross

coefficient

Ji,

is zero.

We shall write

At this stage the

qualitative

behaviour of

Jij

is

strongly dependent

on the model that we use for

3C,,

even if we consider the material to be

isotropic

from

the elastic

point

of view

(pure compressional

or , shear modes without

dispersion).

a)

The

simplest

case to be considered

corresponds

to

isotropic

individual

coupling :

A rather

straightforward

calculation

gives

R

= Rij

is the distance

between,

the

spins i

and

j.

qT is the modulus of the wave vector of the

phonons

with

polarization i(i

=

L, Ti

or

T2)

and energy hco = 2

E ;

with this

type

of

coupling,

i is

necessarily

L. Under the same

hypothesis,

III

The essential feature of

[8]

is

the R -1 dependence

of

JXX

as R tends to zero. At the

opposite limit,

the

interaction has an

oscillatory decreasing behaviour ;

q,i’ is a

characteristic

length

for the two

spins

i

and j

which serves as a scare for their

distance ;

is the wave

length

of the

phonons

on

speaking

terms

with each

pair

of

spins.

In the different cases that we shall

consider, ÂS

is

large

in

comparison

with the mean distance between

spins;

hence we shall not be concerned with the

oscillatory

behaviour of

(8) (R large).

b)

In the second case, it is

supposed

that the

coupling

has no

symmetry

at all. As can be

easily shown,

the smallest

anisotropy

of the

coupling

introduces a new R

dependence

in the interaction.

Even if the numerical coefficient obtained in the model is

only

a

rough approximation,

this is not

too severe a limitation for our model : the R

depen-

dence of

Xi.

is the most

important

feature.

As a first

example,

we calculate the

following

term :

and consider

only

the contribution of the shear modes

polarized perpendicularly

to the vector

R(Tl

modes

of

Fig. 2).

We obtain :

JZZ

is obtained

from (12) by letting

E - 0

(or

qT -->

0)

Thus in the limit

qT R --> 0, JXX

as well as

JZz

increase

as R - 3 ;

this is the same

dependence

as that

of the

dipolar magnetic

interaction.

Another

example

is

given by

the term :

(5)

FIG. 2. - Diagram showing the three polarizations L, Tl and T2

relative to phonon modes with wave vector q. The vector R joining

the two spins is parallel to Oz.

We

obtain,

for the contribution of

longitudinal

modes

only :

and : o o

Here too, in the short distance range, the

leading

tenn varies as

R - 3.

In the

general

case, the contribution to

JXX

of terms

with

coupling

constant

Gll§ x Gi xl 16 npv$

is either

zero or a linear combination of

only

three functions

of qt R :

these appear in eq.

(8), (12),

and

(12bis).

Finally,

we may conclude that the total contribu- tion to

JXX

varies in the short range limit as

R - 3 ;

in the

long

range

limit,

the

leading

term is cos

(qt R )/R.

For

JZZ,

we obtain either zero or

R - 3,

for all R.

4. Calculation of the relaxation times

Tl

and

T2.

-

The

similarity

between the formal treatment of the static

properties

of the materials and those of an

assembly

of true

spins

is clear. We would like now to tum to the

dynamical properties and,

in the first

place,

estimate the

longitudinal

relaxation time

Ti

and the transverse relaxation time

T2 using

our

formalism.

4.1 LONGITUDINAL RELAXATION TIME

Ti.

-

Ti

is the relaxation time of the difference of

population

between two levels of a

spin packet separated by

an

energy 2 E. We calculate

Tl resulting

from the direct process

only,

which is the most effective one at low temperatures.

The

starting point

is formula

(4)

where we

keep only

the first term in

:Iec.

The

probability

per unit of time for a

spin i

to fall

from S.’

=

+Itosz,= -lis

In this formula q, I refer to the wave vector q and

polarization i

of the thermal

phonons,

V is the volume

of the

crystal, N(co.)

the thermal excitation of a

mode of

frequency

roq, and

F(q,,r)

and

q

are unit

vectors

parallel

to the

displacement

and wave vector.

The total contribution of the different strains 80 is :

We

separate

the

angular integration

on the direction of the wave vector and the summation over a,

fi,

y, ô.

For the

integration,

it is tedious but not difficult to calculate all the terms. A number of them

give

a

null

contribution;

the others

give typically

a coefli-

cient

1/10

that is almost

independant

of the

polariza-

tion

[13].

Now, taking

into account the coefficient

1/10,

the number of

contributing

terms and our

ignorance

of the exact form of the G tensor, we set :

The summation over i has to be

applied only

to

v«r5;

for VL ~ 1.5 vT, this

gives :

On the other hand :

With these

partial results,

one has :

A standard

procedure

then

provides

the relaxa-

(6)

tion

time Tl

of the

population

of

spins

of energy difference 2 E

1 1 / 17 B

with

In

(16), (17)

we have written a double mean value

«

(GX)2 >

instead of

(Gi)2,

because even if all the

spins

i have the same energy 2

E,

the

population

is

made up of

spins

with different d

and Jo,

and because these

quantities

are correlated with the deformation

potential ;

each of the

averagings

is

designated by

a

pair

of brackets.

To

give

a numerical value to

Ti

we retum first to

formula

(4).

The first

operation

is to describe the correlation between

84 f88, d, DA01be

and

do.

Follow-

ing Phillips [12]

in his paper on

dilatation,

we shall take all the functions of L1 and

A.

to be

independent,

and then take the correlations to be :

where b is of the order of a deformation

potential (b - 84/88)

and a is a number between 0 and 1.

a = 1 is the extreme case of

complete

correlation.

Phillips

uses a value of a close to

10-’ [12].

On the other

hand,

where

Du/De

is of the order

unity [12].

Thus

It is clear that the most

important

term is the first

one since b L-- 1 eV is much

larger

than J.

The second mean value is taken on different sites : for a

given

value of E, J and

Jo may be

different and

their

probability

distribution is

[6]

Using (20)

and

(19)

we can calculate

(GX)2 >

which is defined as

The final result is

Thus the factor

[Log

2

El Amin] -1

acts as a reduction

factor for the

coupling

between the

phonons

and the

spins.

This

explicit analytical

form

depends

of course

on the model used for

p(A, do) ;

2

E/dm;n is probably

a

large

number.

Nevertheless,

due to the

logarithm,

this factor is not very sensitive to the detailed des-

cription

of the

properties

of the

spin assembly. Phillips

takes a value of 50 for

Log AmaxI Amin

where

Using

the

following

values

we obtain

for a

spin population corresponding

to an energy difference of 700 MHz.

The calculation based on this model calls for two final comments. For true

spins,

which are all identical

(same

level

splitting

and same

spin-lattice constant),

it is

rigorously

correct that the

population

relaxes

exponentially;

the relaxation time is

unambiguously

defined. For

amorphous materials,

the

spin population

with energy

splitting

2 E is not

homogeneous

in the

sense that the

phonon coupling

constants

depend

on A and

A,

at least. The relaxation of the

population

of a

spin packet

is

probably

not

exponential.

We

have avoided the

difficulty by taking

a mean value

for the

coupling,

but this is

only

an

approximation.

On the other

hand,

with so

large

a value for the

spin-phonon coupling,

one may ask whether the Raman processes may not be

competitive

with the

direct process. A detailed calculation shows that

they

are

equally

efficient at

temperatures

of about 5 K

(lower

than for

paramagnetic salts).

Since most

of the

experiments

have been

performed

at

tempera-

tur.es less than 5

K,

it suffices to consider

only

the

direct process.

4.2 TRANSVERSE RELAXATION TIME

T2.

- The calcu-

lation of

T2

is a well known

problem

in

paramagnetic

resonance. Two concepts are useful.

First,

we distin-

guish

between like

spins

and unlike

spins :

in a

given spin assembly,

we call like

spins

those whose energy

are the same;

they

constitute a

spin packet.

Their

mutual interaction does not

change

the energy of a

pair

as a

whole,

even if we consider

flip-flop

processes

(7)

induced by terms such as (S’ S’

+

Si S+). Secondly,

in the method of moments

developped

at

length

in textbooks

[14],

the notion of thé secular part

3e! int

of the

coupling

Hamiltonian

je;nt

is

used ; by

defini-

tion, JC;nt

commutes with

Ko,

the total Zeeman

Hamiltonian ;

the different moments of the

spin

dis-

tribution are related to

JCt

and

Jeo by

formulas

given

below.

In the case of

amorphous materials, Jint is simply

For our

problem,

a

good approximation

is to , consider that the contribution to the different moments comes

essentially

from unlike

spins

since

they

are

more numerous than the like

spins.

The second moment is

given by

Its value must be calculated in some detail.

Van Vleck has

given

an

expression [13]

for

M2

for a

spin

S

= 1 :

where N is the number of

spins

of

type

i

belonging

to

the same

packet

of energy

2 E,

and

JzÙ"z

is

given by (13)

and

(4).

Using

the same

approximations

as for

Ti (with

In

(25), « (Gz)’ »

is a mean value for

Gz

over

all sites and over the total

spin population,

and

«

(G Iz)’ » is

the mean value of

G.

over all sites for the

spin population

with energy 2 E

(like spins only).

Further,

we shall use the result :

valid for a

simple

cubic lattice

[14] (a

is the minimum distance between two

spins).

We add the

following

correlation functions to those

given by (18)

and obtain

Using (27)

and the distribution

probability (20),

we

calculate « (Gz)’ »

for a

given

value of the energy 2 E. The most

important

term in

(27)

is

clearly

the

first one, since b > 2 E >

40

and

Du/De ~

1. We

obtain :

y is

kept

constant in our model and formula

(28)

no

longer depends

on E.

We then have

If we introduce these different results in

(25)

we

finally

obtain

At this stage it is difficult to deduce the linewidth of the

spin-packet

if we have no information on the

. lineshape ;

at the very

least,

the fourth moment

M4

must be estimated in order to

distinguish

between

Gaussian or Lorentzian

shapes.

The

procedure

for

M4

follows the same lines as

for

M2.

The estimation is based on the standard for- mula of Van Vleck and on the

following

observations.

There are two

important

terms in

M4 :

The first one is the sum

rhe second one is

The other terms result from a summation on

i

j k, i j,

i

j, p

or

i,

p q with the restric- tion

Ep

=

Eq ;

the conditions

E,

=

Ej

=

Ele

and

Ep = Eq greatly

reduce the number of

spins entering

into the sum. If

they

are distributed at

random,

their contribution is small. We discard the

possibility

that

they

are clustered.

The two first contributions to

M4

are

nearly equal

to

.__.

The result

(33)

shows that the line

shape

is

typically

(8)

Gaussian ; consequently, T2

is related in a

simple

way to

M2 by

Its order of

magnitude

can now be estimated.

Using

y ~

10-2 [12], a

=

3 A,

b

= 10-12

erg,

p = 2

g/cm’,

and VL = 5 x

105

cm

s-1;

; we have

This estimation calls for a few remarks.

As a result of the Gaussian line

shape,

the relaxa- tion of the transverse

magnetization

is not exponen- tial.

The numerical value

(35)

can be considered

good

within an order of

magnitude only ;

but this

shows,

in any case, that the

equality

2

Tl

=

T2

cannot be

used for the

interpretation

of the

experimental

results.

The result

(34)

is almost

independent

of E.

However,

it is clear that when

2 ET21h

is smaller than one, the method of moments is no

longer

convenient : its becomes

impossible

to

distinguish

between secular and non-secular terms with

respect

to

Ko.

In the

following,

we will consider that we are

beyond

that

limit,

in

particular,

when

considering

the

propagation

of an acoustic wave, we shall take its

frequency

to

be

sufficiently large

so that

wT2

> 1 : this

implies

a

frequency larger

than 150 MHz.

In the calculation of the moments, there is a pro- blem with the

long

range behaviour of the interaction between like

spins

which varies as

q2 R -1

cos

(qR)

for

qR »

1. For

example,

it is not evident that

M2

is convergent.

Two facts

support

the idea that this contribution is

negligible.

i)

The number of like

spins

is a small fraction of the total number : the

spins

are

uniformly

distributed

over about 1 eV or

10-12

erg; the

spin packet

under

consideration

occupies

a width

1iT21 ’" 10-18

erg.

The ratio is theri

10-6.

The same reason was

already

invoked in

discarding

some terms in

M4.

ii)

The

phonon attenuation,

which occurs when R >

À,

has to be taken into account.

Simply adding

an

imaginary part

to the wavevector is mathemati-

cally

incorrect

[2],

but it introduces a useful conver- gence factor and even for a weak attenuation the contribution to

T2

from like

spins

becomes very small.

For our final

remark,

we return to the

expression

for

MZ ;

formula

(24)

is the

high temperature

appro- ximation to

M2,

and is

justified

as

long

as 2 E

kT,

which is the case for our numerical values. Since most of the

experimental

results have been obtained in the

temperature

range 0.3 K-1

K,

the

approxima-

tion remains valid for 2 E

10-16

erg. But even if it were necessary to extend the calculation to much lower temperatures

(or

to much

higher energies),

this could be done

nearly exactly

because the

spin

°

is t (1).

5. Ultrasonic atténuation. - We now tum to the

problem

of the

propagation

of an elastic wave in

amorphous materials,

and in

particular

to the calcula-

tion of the attenuation of the wave due to

coupling

with the

pseudo-spin

system. The

problem

will be

limited to the case of a

longitudinal

ultrasonic wave

in which x is the direction of

propagation

and vibra-

tion. This

problem

is to be solved for any acoustic power, in order to treat at the same time the saturation and the non-saturation cases of the

spin population.

Theoretically

this is easy : the

coupled

evolution

equations

of the elastic variables and of the

spin

variables are solved

simultaneously.

A macroscopical description tor the spm system is

adopted,

and we shall

speak

in terms of its

magnetiza-

tion

or, equivalently,

in terms of mean

spin operators.

We start with a

Lagrangian density

which includes both a

purely

elastic and a

magnetoelastic

term :

u is the elastic

displacement (along

the x

direction)

and

X is the

spin density.

In this

paragraph

we will

always

write

Gx

for

{

«

(G’x )2 » 11/2.

In the model

developed above,

the contributions to

JU

from shear and

compressive

waves were not

separated ; however,

when we consider the

propagation

of an elastic wave, this is necessary ; this remark will be useful in

correlating

our results

with

expérimental

data when it is

possible

to

distinguish

between the

coupling

constants.

The effective

coupling

constant

Gx

and the mean

spin operator

s as defined

by :

have been introduced in the

Lagrangian density

in the

following

manner.

The total

magnetization

is the sum of

elementary magnetizations of spin packets

with Larmor

frequency

roo

(where hmo

=

2 E).

Each of these

packets

is

coupled

to the

phonons by

a mean

coupling

constant

depending

on wo and the total

magnetoelastic

energy is

proportional

to :

s(mo)

is the

elementary magnetization ;

f (coo)

is a normalized function :

(9)

which is related to the

spectral density n(E)

introduced

by Phillipps [6] :

In

fact,

the

coupling

constant

Gx

is

only weakly dependent (logarithmically)

on roo

(eq. (21)); therefore, Gx(coo) ~-- Gx

can be removed from the

integral,

and we have

More

generally,

the mean

spin operator,

which is

proportional

to the total

magnetization,

is defined

by

At thermal

equilibrium,

one has :

Such a

description

in terms of mean

operators (total

or

elementary)

is valid because there are a

large

number

of

spins

in a volume whose dimensions are of the order of an ultrasonic

wavelength ;

more

explicitly :

Applying

the

Euler-Lagrange equation

to the

Lagrangian density,

we obtain :

which is the

propagation equation

of the elastic wave

coupled

to the

spins.

vL =

(Clp)112

is the sound

velocity

without

coupling.

In this

equation

we have

deliberatly neglected

the

term

since it does not oscillate at a

frequency

close to the

acoustic

frequency.

Despite

the results obtained in the

previous

para-

graph (Gaussian

line

shape),

we shall take the beha-

viour of the

magnetization

to be

governed by

Bloch

equations

with relaxation times

Ti

and

T2 given by

eq.

(17)

and

(34). Indeed,

we shall set

T 1

=

Tl 1

and

T 2

=

Ti 1

and write :

where a is the strain.

The formulas

(39)

are the

equations

of evolution of the

magnetization s(mo) (mo is omitted)

in the labo-

ratory

frame.

As is

customary

in

spin

resonance, we transform to a

rotating

frame where

spin

variables are indicated

by

a

tilde.

However,

because of the

propagative

character

of the

perturbation,

we have to choose a

rotating

frame which varies from

point

to

point,

in order to

be in

phase

with the wave

[15].

We write the

displacement

associated with the wave

The attenuation and the

change

of

phase velocity

are taken into account since the

amplitude

A is a

function of x

and t,

and the

phase

lp is a function of x.

The desired transformation is :

Neglecting

all

rapidly oscillating

terms, we obtain :

with the

approximation

of weak attenuation and small

change

of

velocity :

In the

following,

we

put :

(10)

We now retum to the

propagation

eq.

(38)

and we

obtain :

We then have to solve the

following

five

coupled equations :

Such a

problem

is

beyond

the scope of this

article,

and we limit ourselves to the search for

stationary solutions,

i.e. solutions in which all time derivatives

are zero

This

type

of solution describes the case where the duration of a wave train is

long compared

to the

characteristic times

Ti

and

T2 ;

this may or may not be the case

experimentally.

The

system

to be resolved is then :

The last three

equations

are first solved as functions of so. We find :

These results must be substituted into the first two

equations of (41 ).

Since our main concem in this paper is with the ultrasonic

attenuation,

we concentrate on the first

equation only

and obtain :

The

quantities f (coo),

so,

Tl, T2 in

this

integral depend only weakly

on roo in

comparison

with the

resonant

denominator,

so that

they

may be taken out of the

integral.

We thus obtain :

Setting

with

we

finally

have :

The solution of this differential

equation,

with the

condition A =

Ao

for x =

0,

is

given by

(11)

Two

limiting

cases are of interest :

The solution then become

or

This is the non-saturation limit and

ln.s

is the usual attenuation

length.

The solution is :

where E is the ultrasonic

intensity.

One can define an attenuation

length

for the acoustic power

by :

It is

easily

seen from the differential

equation

that :

Defining

a critical ultrasonic

intensity

or

amplitude Ac by :

we have

In order to facilitate the

comparison

with

experi-

mental

data,

it is useful to write the critical condi- tion

(46)

as a function of the ultrasonic flux

It then can be rewritten as

and in the saturated

regime,

we have :

or

This result indicates

that ls 1

is

weakly temperature dependent (ls 1 N T- 1/2) further,

in the

region

of

high temperatures (1iro kT),

we

predict

that

1.-..l

varies

aSW2 (which

is a well-known

result),

whereas in

the saturated

regime ls 1

must increase as

ro3,

instead

of

w4.

These are two differences with the standard

theory [9].

, 6.

Comparison

with ultrasonic

experiments.

- An

ultrasonic wave

propagating through

an

amorphous

material suffers two

types

of attenuation.

First,

the elastic wave modulates the

spin popula-

tion and a non-resonant attenuation occurs

[16] ;

it

originates

in the non reversible

part

of

the modulation

and is related to

Tl, essentially.

The same

phenomenon

has been described and observed for

paramagnetic

systems

[17, 18].

Secondly,

there is a resonant attenuation due to transitions between the two

levels;

its order of

magnitude

has been calculated in the

previous

para-

graph (eq. (43)).

Experimentally,

these two contributions are

easily distinguished :

the first one is

independent

of the

acoustical

intensity,

whereas the second is

subject

to

saturation,

as has been shown.

Several observations have been

reported

of the

non-linear character of ultrasonic attenuation on

different

amorphous

materials

[9, 10], [19, 20, 21, 22].

We shall compare our

theory

with the paper

by

Arnold et al.

[9]

which

provides

many

quantitative

measurements; up to now

only experiments

with

longitudinal

ultrasonic waves have been

reported ;

and information can therefore be deduced

only

for

the

longitudinal coupling

constants.

6.1 EVALUATION oF

Gx

AND

Tl.

- The effective

coupling

constant

Gx

can be deduced

independently

from the value of the non-saturated attenuation :

1 __1 ... B n 2 ...2

Using

the

following

numerical values taken from

specific

heat measurements and observations of the ultrasonic attenuation :

one obtains

With this value the relaxation times can be

easily

calculated :

6.2 EVALUATION or

T2

AND

Gz.

- In order to

evaluate

T2,

the formula

giving

the critical acoustic

Références

Documents relatifs

Key words: Mulholland inequality, Hilbert-type inequality, weight function, equiv- alent form, the best possible constant factor.. The above inequality includes the best

To achieve our objectives of assessing the level of readiness of Nigeria, in respect of provisioning of required infrastructures, when the implementation of the

This paper presents three presumably typical use cases and associated interaction design proposals, which illustrate (a) how untrained users can benefit from the organization of

In these applications, the probability measures that are involved are replaced by uniform probability measures on discrete point sets, and we use our algorithm to solve the

Bounded cohomology, classifying spaces, flat bundles, Lie groups, subgroup distortion, stable commutator length, word metrics.. The authors are grateful to the ETH Z¨ urich, the MSRI

[r]

The purpose of this research is to highlight, through an empirical analysis, the impact of variables macro-economic (monetary mass, budget deficit, exchange term,

Up to now, while highly developed in silicate glasses, the ion exchange process is still very little studied in chalcogenide glass family because of the lack of composition with