• Aucun résultat trouvé

Static properties of a random one-dimensional magnet

N/A
N/A
Protected

Academic year: 2021

Partager "Static properties of a random one-dimensional magnet"

Copied!
6
0
0

Texte intégral

(1)

HAL Id: jpa-00208363

https://hal.archives-ouvertes.fr/jpa-00208363

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.

Static properties of a random one-dimensional magnet

M. F. Thorpe

To cite this version:

M. F. Thorpe. Static properties of a random one-dimensional magnet. Journal de Physique, 1975, 36

(12), pp.1177-1181. �10.1051/jphys:0197500360120117700�. �jpa-00208363�

(2)

LE JOURNAL DE PHYSIQUE

STATIC PROPERTIES OF A RANDOM ONE-DIMENSIONAL MAGNET

M. F. THORPE

(*)

Institut

Laue-Langevin,

B.P.

156,

38042 Grenoble

Cedex,

France

(Reçu

le

24 juin 1975, accepté

le

30 juillet 1975)

Résumé. 2014 Nous donnons un calcul exact des

propriétés

thermodynamiques d’une chaine aléatoire

classique

d’Heisenberg

comportant 2 types d’atomes. La

dépendance

de la

susceptibilité

avec le

vecteur d’onde est aussi obtenue. Des résultats numériques détaillés sont donnés pour le cas parti-

culier d’un alliage magnétique-non magnétique.

Abstract. 2014 The thermodynamic

properties

of a random classical Heisenberg chain with two kinds of atoms are calculated exactly. The wave vector

dependent susceptibility

is also obtained. Detailed numerical results are given for the

special

case of a

magnetic-nonmagnetic

alloy.

Classification

Physics Abstracts

8.510

1.

Thermodynamics.

- We

study

the static pro-

perties

of a classical

Heisenberg

linear chain with two

kinds

of atom A and B

arranged randomly

with

concentration c and 1 - c,

where the nearest

neighbour exchange parameter Ji

may be

JAA, JAB

=

JBA

or

JBB depending

on the

nature of the atoms at sites i and i + 1. The unit vectors

Si

may be

thought

of as the limit of quantum mechanical

spin operators

as S -->

oo, h --->

0 and

hS ---> 1

[1], [2].

The classical

Heisenberg

linear chain with

only

one

kind of atom was first solved

by

Fisher

[1].

This solu-

tion utilizes the

expansion’

.

where

- .&.

The

Y,m(Si)

are the usual

spherical

harmonies and the

P,(x)

are

Legendre’ polynomials.

Our notation follows that of reference

[2].

The

partition

function

may be written

where we

integrate

over ail the

spin

coordinates of a

chain

with free

ends.

Using

the

expansion (2)

and the

orthogonality property

of the

spherical harmonics,

it is easy to see that :

where the

integration

over

Si picks

out

YOO(SL)

and

the

subsequent integrations

also

pick

out

only

the

YOO(Si).

The result

(5)

is true for

a finite

chain whereas in the case of a chain with

periodic boundary

condi-

tions,

it is necessary to take the

thermodynamic

limit

in order to obtain

(5).

The free energy from

(5)

is

F = - kTN In

À,o({3J) . (6)

It is almost trivial to

generalize

this result to a

random chain. The

expansion (2)

now contains a J

which

depends

on

position. However,

the crucial

spherical

harmonic

part,

which arises from the

isotropy

of the

spin interactions,

is

unchanged.

The result

(6)

becomes

which can be written more

explicitly

for our case

We can evaluate the

À,o({3J)

from eq.

(3)

Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:0197500360120117700

(3)

1178

Other

thermodynamic quantities

that can be derived

from

(8)

such as the energy and the

entropy

have simi-

lar forms.

The result

(8)

is of much wider

validity

than the

classical

Heisenberg

linear chain. A similar result is obtained for any model in which an

expansion

of the

type

(2)

can be made in which the

positive eigenvector

is

independent of the

interaction parameters. For

example,

if the classical

spin Si

is

replaced by

a

scalar ui

= + 1

to

give

a

spin 2 Ising

chain

[3],

the

positive eigenvector YOO(Si)

is

replaced by

the

positive eigenvector

of the

2 x 2 transfer matrix and the

corresponding eigen-

value becomes

and the result

(7)

would

apply

to the mixed system

[4].

The result

(7)

would not be obtained in cases where the

positive eigenvector depends

on the interaction parameters as for

example

in the

Ising

model

just

discussed with an extemal field

tenn h ui

added

;

to the Hamiltonian. In this case the

positive eigen-

vector

depends

on the ratio

Ji/h

which

prevents

closure

being exploited

in any

simple

way.

°

Notice that the

sign

of J enters

through (9)

or

(10)

which are both even functions of J. The

properties

of

these random chains in zero field are therefore inde-

pendent

of the

sign

of J. Of course this is not true for

the wave vector

dependent susceptibility

to be dis-

cussed in the next section.

2. The wave vector

dépendent susceptibility.

-

The cross-section for neutron

scattering S(g)

in the

quasielastic approximation

is

just given by

the wave

vector

dependent susceptibility (apart

from a fac-

tor

kT) [5]

For a linear

chain,

this

expression depends only

on

the

component

of q

parallel

to the chain which we

will denote

by

q. If the interatomic

spacing

is a, then :

As shown in

[2],

a

general

correlation function between

operators

oci,

oc’,,

at

sites i,

i + r may be written

In the present case we are interested in ai =

SZ

and

rx¡+r

=

Siz+r

so that

only

the 1 =

1,

m = 0 term contributes and

and

S(q)

becomes

where ] r is

necessary because the summation over r

goes over all

positive

and

negative integers.

In the

case of a pure chain with

only

one atomic

species,

this sum is

easily

evaluated to

give

where we have written u =

Â1/Â0

which may be

evaluated

directly

from

(3)

to

give

For a random

chain,

this formalism goes

through except

that there are three

separate

u, i.e. uAA, UAB =UBA’

and UBB

corresponding

to the

appropriate pair

of

atoms. The correlation function

Si. Si,,>

for a

sequence ABABAA... ABB now becomes

The

problem

is to insert this

expression

into

(12)

and

perform

the summations over the random chain.

This

procedure

is

possible

in the

general

case but

particularly simple

for the dilute

magnetic

chain.

3. Dilute

magnetic

chain. - We dilute the chain with a concentration 1 - c of

non-magnetic

atoms

so that uAA = u and UAB = UBA = UBB = 0. The free energy

(8)

becomes

where J is the

coupling

between the

magnetic

atoms

and the free energy is

proportional

to the number of

near

neighbour pairs

of

magnetic

atoms.

In order to calculate the

susceptibility,

it is easier

to do a

configurational

average rather than the sum

over i in eq.

(12).

This is an

entirely

correct

procedure

in the

thermodynamic

limit that we are interested in.

In this case it is rather

simple

and the

susceptibility

per

magnetic

atom is

This

expression

is the same as

(16)

but with u - cu.

1As T --+ 0

and q - 0,

the

susceptibility S(q) diverges

(4)

only if c

= 1 as we would expect. It is

possible

to say that there is a

phase

transition at zero temperature in the pure system. The introduction of

non-magnetic

atoms breaks up the system into

finite, non-interacting

sections and there is no

ordering

even at zero tem-

perature.

The result

(20)

takes a

particularly simple

form in

the critical

region

defined

by

which is

with the dimensionless correlation

length j given by

and

This critical

region

as defined

by (21)

is broader

than the usual one which adds

qaj >

1 to

(21),

but

is somewhat more

appropriate

in this case. At zero

temperature the correlation

length

is

given by

The results derived in this section could be

applied

to TMMC

[5]

which contains Mn++ chains with S

= 2

and which could

probably

be

doped

with

a

nonmagnetic

ion such as Zn + + . The

theory

of the

previous paragraph

could be

applied

if we correct

for the finite

spin.

This has been done

by Hutchings

et al.

[5]

for the pure case

by inserting

a factor

S(S + 1 )

in with the

exchange

and with

S(q).

Insomuch as this

just

introduces a scale

factor,

it cannot

easily

be

checked

experimentally

and it is not clear that

S(S + 1 )

is to be

preferred

over

S2

with the

exchange. Using

the notation of

Hutchings et

al.

[4]

where the

exchange

2

Jnn

is

negative

we have to

put

in the

equations

of this

section,

so that with

we have

and

where

These results are those of

[5]

with u ---> cu. Notice that K and B are not

independent

but related at all

concentrations

by

.

The critical

scattering

occurs around

Q

=

1,

and not

at

Q

=

0,

because of the

antiferromagnetic coupling.

In

figure

1 we show a

plot

of x

against

T for various concentrations. The c = 1

theory

and

experimental points

are taken from

[5].

It can be seen that the inverse correlation

length

increases

rapidly

upon dilution.

From eq.

(23)

we can see that the

slope

of K -

1/e;

against

T is

govemed by

a factor

(1

+

c)/2 lc

which

only

increases

by 6 %

as c goes from 1 to 0.5. The

departure

from

linearity

that is

apparent

but small in

figure

1

corresponds

to

going

out of the critical

region

as we have defined it.

FIG. 1. - The inverse correlation length K plotted against tempe-

rature for various Mn concentrations in dilute Mn linear chains.

The c = 1 data points and curve are from work on TMMC [5].

(5)

1180

4. The

general

case. - For the

general

case we

consider two

magnetic species A,

B with concentra- tions c, 1 - c

leading

to three

exchange

interactions

JAA, J AB JBA,

and

JBB

and therefore to three u

functions UAA, UAB = UBA, and uBB. We must insert sequences like

(18)

into the

expression

for the sus-

ceptibility (12)

and do the summations. Instead of

summing

over all

sites i,

it is convenient to

replace

£ by

a

configuration

average. We therefore define

where zero is some

arbitrary

référence site. The super-

scripts

on C denote that a

configuration

average has been

performed

over the sites

1, 2, 3, ..., 1-

1 but

site 0 is an A atom and site 1 is also an A atom.

Clearly

we may define similar

quantities C, AB

=

CBA

and

CBB.

These

quantities

are useful because

they obey simple

recursion relations

for 1 >

1,

with the initial conditions

In

writing

down

(27)

we have

performed

the confi-

guration

average over atom 1. Similar

equations

can

be written for

CBB

and

CHA merely by interchanging

A --> B and c H 1 - c in

(27)

and

(28).

These

equations

are most

easily

solved

by

intro-

ducing

a

parameter x

to find the normal modes of

(27).

we choose

and

solving

this

equation

for x, we get two roots

x+,

so that

with

can now solve eq.

(32)

for

Ci%j

and

CAB

sepa-

rately using

the initial conditions

(28)

With these results we are now in a

position

to evaluate the

configurationally averaged susceptibility

for an A

type

atom at the

origin

The

corresponding quantity SB(q)

for a B

type

atom at the

origin

may be obtained from

(35) by letting

A H B

and c H 1 - c. The

complete susceptibility

is then

given by

The

susceptibility S(q)

will

only diverge

at zero

temperature

if either all the interactions are

ferromagnetic

or all the interactions are

antiferromagnetic.

In the first case all the u will

approach

1 as T - 0 and

S(q

=

0)

will

diverge

whilst in the second case all the u will

approach -

1 as T - 0 and

S(q

=

nla)

will

diverge.

For

mixed

ferromagnetic

and

antiferromagnetic interactions,

the

system

will still order as T -> 0 but the suscep-

tibility

that

couples

to the order

parameter

will not have a

spatial periodicity

described

by

a wavevector q.

Of course

S(q)

can still be defined and measured in these cases and is

given by (35)

and

(36).

Most

experiments

do not measure

spin-spin

correlations but rather

magnetisation-magnetisation

corre-

lations. The

theory

that we have worked out can

easily

be modified to include this if we

put Mi

= gA

Si

or

gB

Si depending

on whether the site i is

occupied by

an A or B atom, then eq.

(12), (35)

and

(36)

become

(6)

Rather than write out more

explicitly

the

lengthy expression

for

M(q),

we will

only

do so for a few

special

cases.

When q

=

0,

we have

This

clearly diverges

at T = 0 for all

ferromagnetic couplings,

i.e. UAA = UAB = UBA = UBB = 1. At

high

tem-

peratures

this becomes

that is the average

squared

moment.

(Combining

this with the

1/kT

factor that we have not been

carrying,

we get the

expected

Curie

law.)

In the critical

region

defined

by (21)

we may write

M(q)

in a form similar to

(22)

where

and

The appearance of the B tenn is a néw

feature;

it does not occur in the pure or dilute

magnetic

systems. It reflects the disorder

directly

and could

be

quite sizeable,

as of course in

applying

this work

to real

s stems with

finite

spins, appropriate spin

factors

S(S + 1)

must be

incorporated

into

the g

factors.

5. Conclusions. - We havé shown that the static

properties

of random classical

Heisenberg

linear

chains can be worked out

by exploiting

the

indepen-

dence of the

positive eigenvector

from the

exchange

interactions. It should be

possible

to

apply

these

results to real systems if

appropriate spin

factors are

incorporated

into the

exchange

interactors and

the g

factors.

However, although

quantum corrections do not seem to be

important

for Mn++

(S

=

2),

this

may not be the case for

Ni (S

=

1). Finally

we

note that in many mixed systems, the relation

is

obeyed [6].

If this is also the case in linear

chains,

then from eq.

(23a)

the square root of the inverse correlation

length

should be linear in the concen-

tration.

Acknowledgments.

- 1 should like to thank S. W.

Lovesey

for a useful conversation and the ILL for their

hospitality

in the summer of 1975.

Note added in proof This problem has also been considered by T. Tonegawa, H. Shiba and P. Pincus, Phys. Rev.11 (1975) 4683. The

work here corresponds here to their quenched limit. They gave not considered q dependent quantities but our expressions for the pair corre-

lations are in agreement.

References [1] FISHER, M. E., Am. J. Phys. 32 (1964) 343.

[2] THORPE, M. F. and BLUME, M., Phys. Rev. B 5 (1972) 1961.

[3] See for example MATTIS, D. in The Theory of Magnetism (Harper

and Row, London) 1965, p. 236.

[4] WORTIS, M., Phys. Rev. B 10 (1974) 4665.

[5] HUTCHINGS, M. T., SHIRANE, G., BIRGENEAU, R. J. and HOLT, S. L., Phys. Rev. B 5 (1972) 1999.

[6] COWLEY, R. A. and BUYERS, W. J. L., Rev. Mod. Phys. 44 (1972)

406.

Références

Documents relatifs

After finding that the static properties of dense polymers very well compare to ideal chains, it is especially interesting to check the chain dynamics. Since the chain

Accurate values for the ground state energy, radial distribution function, average exchange energy and momentum distri- bution are obtained at low atomic densities.. Then

we expect large quantum fluctuation and also relevant cooperative phenomena. Importance of quantum fluctuation is also enhanced for frustrated systems and the absence of

In view of the fact that the transport of spin fluctuations is slow at high temperatures, the most ef- ficient and accurate method is first to reduce the statis-

In the former phase, which is called the spin-fluid phase by Haldane [9], any type of long-range order does not exist, as ex- pected; while the latter phase, called the

temperature Tc is confirmed; quantitative evaluations of the spin pair correlation function below Tc,.. and of the transition temperature Tc are given; the results are

In this temperature region there- fore, our results can be considered to be representative for the ideal square Heisenberg ferromagnet.. [2] the isomor- phous

In this section we construct an estimator that works well whenever the distribution of X is sufficiently spherical in the sense that a positive fraction of the eigenvalues of