• Aucun résultat trouvé

On the use of different matrix products to solve finite three-dimensional Ising systems

N/A
N/A
Protected

Academic year: 2021

Partager "On the use of different matrix products to solve finite three-dimensional Ising systems"

Copied!
9
0
0

Texte intégral

(1)

HAL Id: jpa-00210298

https://hal.archives-ouvertes.fr/jpa-00210298

Submitted on 1 Jan 1986

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.

On the use of different matrix products to solve finite three-dimensional Ising systems

Ph. Audit

To cite this version:

Ph. Audit. On the use of different matrix products to solve finite three-dimensional Ising systems.

Journal de Physique, 1986, 47 (7), pp.1119-1126. �10.1051/jphys:019860047070111900�. �jpa-00210298�

(2)

On the use of different matrix products to solve finite three-dimensional

Ising systems

Ph. Audit

Laboratoire PMTM, Université Paris-Nord, 93430 Villetaneuse, France

(Reçu le 9 dicembre 1985, accepté

sous

forme définitive le 24

mars

1986)

Résumé. 2014 On définit

un nouveau

produit matriciel, qui peut être utilisé

en

liaison avec trois autres produits

matriciels classiques, pour résoudre analytiquement des systèmes d’Ising de taille finie. A titre d’exemple,

on

déter-

mine l’expression exacte de la fonction de partition du réseau périodique anisotrope 2

x

2

x

2

en

présence de champ, qui conduit aux expressions exactes de l’aimantation et de la chaleur spécifique de

ce

système.

Abstract.

2014

A

new

matrix product is defined and used in connection with three other well-known matrix products

in order to derive analytical solutions of finite three-dimensional Ising systems. As

an

example, the exact partition

function of the periodic anisotropic 2

x

2

x

2 lattice in a field is evaluated from which analytical expressions for

the magnetization and the specific heat of this system follow.

Classification

Physics Abstracts

75.40D - 05.50

1. Introductioa

The most elegant and straightforward solutions of one-

dimensional systems of N Ising spins are supplied by the well-known transfer matrix formalism [1],

where the partition function is given by the trace of a regular product of N matrices. However, the difficulty

of calculating exact eigenvalues for the large matrices

involved in the solution of three-dimensional systems has prevented the above result from being generalized.

Notwithstanding the need of new matrix products to

build the transfer matrix has been clearly realized by Novotny [2] who proposed the row and column matrix product for this purpose. Analogically, the

basic tool of the present method is a new matrix product well suited to the three-dimensional Ising problem; and making use of this product, the partition

function is obtained as the trace of a combination of Pauli matrices, which can be calculated without any

diagonalization. We will explain the method and

apply it to the Ising 2 x 2 x 2 lattice in the presence of a field, with different couplings in the three direc- tions of space and assuming periodic boundary

conditions. In spite of the simplicity of the model

considered, the derivation of its solution is actually

non trivial, but quite instructive. Indeed, by consider- ing the derived analytical result, the different roles played by the dimensionality, the anisotropy and the

field respectively are strongly illuminated, and the difficulty encountered in seeking solutions for Ising

models taking those parameters more fully into

account can be well acknowledged and explained Furthermore, from the computer assisted derivation of the isotropic 4 x 4 x 4 Ising model partition

function [3], we have learned much of the difficulties encountered as the lattice size increases. In contrast, the considerable complexity arising from the aniso-

tropy in the presence of the field for 3d models has never, to our knowledge, been treated yet. This has guided our choice for the first model to be solved

by this new method, the solution of models with a much larger size will be given in a subsequent paper.

In the present paper, we consider the partitions

function of a system of N Ising spins Qjmn on a three- dimensional lattice, with coupling constants Ji, (i

=

1, 2, 3), along the x, y, z directions, in an external field h

Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:019860047070111900

(3)

1120

with Ki

=

JJkB T and H

=

h/kB T. In order to set

up a matrix characteristic of the three-dimensional

lattice, we start by defining the four 2 x 2 matrices Pl, P2, P3 and Q (diagonal) as

and in terms of the Pauli matrices

we have

where

and we give also for future use the matrices

Thus, the partition function (1) can be written in the form

suitable for an evaluation using the matrix formalism to be developed in the following.

2. Matrix products.

Besides the regular product

-

denoted AB - of two

square matrices, use will also be made of the direct

product defined by

In addition the Hadamard product defined by [2, 4]

will also be needed. Useful examples of this matrix

product are given below, taking into account equa- tions (5), (6), (7), (12) and (13)

with

Interesting relations link the direct and Hadamard

products as

which follow directly from definitions (15) and (16).

However, in equation (14), the contribution of a

given spin involves a summation over its six neigh- boring spins; hence, it is necessary to define a fourth matrix product specially designed to take the three- dimensional Ising interactions into account Therefore,

we will now consider six arbitrary 2 x 2 matrices A, B, C, A’, B’, C’ and one diagonal 2 x 2 matrix

D

=

diag (d1, d2). The following expression

can be regarded as a generalization of the regular

matrix product, where, in contrast with the latter,

the sum over j in (23) is restricted to the products of

the first (respectively last) element in the A p B © C matrix row with the first (respectively last) element

in the A’ p B’ © C’ column and with the first (respec- tively last) diagonal element of the matrix D. By means

of the diagonal matrix

our special matrix product can be written in the

(4)

convenient form

In terms of the Pauli matrices (3), the matrix L1 reads

Using either of those expressions, equation (25) gives

3. The partition function of the anisotropic 2 x 2 x 2 lattice.

To illustrate the usefulness of the matrix product defined in section 2, we now, consider the 2 x 2 x 2 Ising

model endowed with the full complexity resulting from the couplings’ anisotropy and the presence of a field,

in spite of its small size. Besides, the model is assumed to satisfy periodic boundary conditions, which in this

special case, means, that each spin interacts twice with each of its three neighbours, so that the model actually displays a full three-dimensional character.

°

The eight spins of this lattice, labelled as in figure 1, can be regarded as split into two layers : (61, 0’3’ U:51 (J7)

and (Q2, (J 4’ (J 6’ 6s), and the terms of the product in equation (14) can be ordered accordingly by using equa- tion (15), so that the partition function of this model can be written as

Fig. 1.

-

Spins

on

the 2

x

2

x

2 lattice.

(5)

1122

After a summation over the second layer, or even spins, according to the rules (28) or (29), the following expression

is left over

which involves the first layer, or odd spins, only; and where for brevity, we have set

Taking into account the direct product definition (15), equation (31) now reads

or by rearranging matrix elements in the direct products we have

but further reduction results from using the Hadamard product (16) and we obtain

Then, taking equations (17), (22) and (32) into account, we introduce the following notation

(6)

and taking equation (22) into account, we can write several expressions such as

and

which yield the following relations

so that the partition function (34) is given by the trace of the Hadamard product of two matrices, which differ from one another by the order of the terms in the direct products only, it is

where the properties of the trace of a direct product, as well as the invariance of the trace under cyclic permutation

of factors have been used.

Now, application of rule (22) to expressions (35) gives

those matrices can be expressed in terms of the Pauli matrices from equations (16) and (17), and that makes the

calculation of their traces trivial; we obtain successively

(7)

1124

and

Finally, substituting equation (41) in equation (38), we obtain the desired result

We note that the derivation of this result required the use of all four different matrix products considered in section 2 ; this necessity should arise in seeking the solution of any three-dimensional Ising system, in contrast with the one-dimensional case where the use of the regular matrix product is sufficient.

It is straightforward to derive particular expressions from the general formula (42), such as for instance the zero-field partition function which is obtained by simply setting y

=

c3, g

=

S3, 6

=

0 in equation (42), according

to the definitions (18), (19) and (20) of those parameters. The result is

When one coupling Ki

=

0, that is ci

=

1 and si

=

0, there is no interaction between planes and we obtain,

from (43), the following expressions of the partition functions

Moreover, for equal couplings (ci

=

C2

=

C3

=

c, si = S 2

=

S3

=

s) in the planes we have

a result to be compared with the solution by Onsager and Kaufman [5] for an isotropic square lattice, that reduces to

for a 2 x 2 lattice, using our notation. Hence, the following relation links the two and three-dimensional partition

functions

as would be expected.

(8)

4. Thermodynamical functions of the 2 x 2 x 2 isotropic Ising system.

Having performed this simple check of the general expression (42), we now proceed to the isotropic case : K1 = K2

=

K3

=

K, and readily obtain from (42), first the partition function with field

and second the partition function without field

In order to be fully aware of the difficulties induced by the presence of a field in three-dimensional Ising systems, it is certainly instructive to compare equations (42) to (43), or equations (52) to (51).

The specific heat per spin is obtained from the expression

where Z8 and Zg are the first and second derivative with respect to c of expression (52). A numerical calculation of C as a function of T shows that the specific heat is a slowly varying function of T with a maximum. This behaviour is qualitatively analogous to the specific heat variation of the 2 x 2 lattice [6]. In the 2 x 2 x 2 case,

the maximum C/8 kB

=

0.61119 was found for kB TjJ

=

3.57066 or K

=

0.28006, that is not so far from the most accepted value [7] for the critical coupling of the infinite three-dimensional Ising model in zero-field, K,

=

0.22166.

Concerning the magnetization, it is obtained by differentiating equation (51) with respect to h as

We note that

dc’ldh

=

(kB T)-1 s’ and ds’/dh

=

(kB T)-1 c’ and upon setting C3

=

c in equations (18) (20), we have

Using those relations we find

and note that M(O)

=

0; the system being finite does not exhibit spontaneous magnetization.

In conclusion, through the full solution of a simple model, we have shown that exact analytic solutions

of finite-size three-dimensional Ising models could benefit from the above definition of a new matrix product

used in connection with three other classical products. It will be shown in a forthcoming paper that larger N x N x

N lattices can benefit from the present scheme which summarily consists in three steps : first a summation over

(N x N x N)/2 spins by using the special matrix product, second a reduction to N/2 terms by means of the Hada-

mard product, third the determination of the trace of the transfer matrix by taking advantage of the properties

of the Pauli matrices.

(9)

1126

References

[1] KRAMERS, H. A., WANNIER, G. H., Phys. Rev. 60 (1941)

252.

[2] NOVOTNY, M. A., J. Math. Phys. 20 (1979) 1146.

[3] PEARSON, R. B., Phys. Rev. 26 (1982) 6285 and

ITZYKSON, C., PEARSON, R. B., ZUBER, J. B., Nucl.

Phys. B 220 (1983) 415.

[4] HALMOS, P. R., Finite-dimensional vector spaces (Prince- ton, N. J. : Van Nostrand) 1958.

[5] KAUFMANN, B., Phys. Rev. 76 (1946) 1232.

[6] FERDINAND, A. E., FISHER, M. E., Phys. Rev. 185 (1969)

832.

[7] PAWLEY, G. S., SWENDEN, R. H., WALLACE, D. J.,

WILSON, K. G., Phys. Rev. B 29 (1984) 4030.

Références

Documents relatifs

In this article we consider the production of 100 individuals with different profiles of aphasia and 100 individuals without brain damage, from the corpus

Aware of the grave consequences of substance abuse, the United Nations system, including the World Health Organization, has been deeply involved in many aspects of prevention,

Banica introduced the notions of dephased and undephased de- fects for matrices F G , and showed that they give upper bounds for the tangent spaces at F G to the real algebraic

Neural networks provide multi-class classification regardless of the linear time-separability of classes. In addition, neural networks are able to determine the presence

In order to do so, compare the free energy of a perfectly ordered system with the free energy of a partially disordered case that introduces the less energetic frustrations

In the first part, by combining Bochner’s formula and a smooth maximum principle argument, Kr¨ oger in [15] obtained a comparison theorem for the gradient of the eigenfunctions,

As the number of isomorphism classes of such curves is infinite, we obtain infinitely many conjugacy classes of de Jonquières involutions in the Cremona group.. (In fact, there

Using Howlett's result, we can give explicit bounds for two cases: when G is metabelian, or nilpotent-by-abelian, and when one of the factors is abelian.. (*) Indirizzo