1170

Progress of Theoretical Physics, Vol. 43, No. 5, May 1970

Domain Structure in a Second-Neighbour Ising Chain

N. E. FRANKEL and D. C. RAPAPORT

School of Physics, University of Melbourne, Melbourne:> Australia

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

(Received May 16, 1969)

Two different methods of obtaining the partition function for a spin t open ended Ising
chain with interactions between nearest and second nearest neighbour spins are presented.
These are non-matrix methods, in that they do not require the diagonalization of a transfer
matrix, and they provide a useful representation for studying the analytical structure of the
partition function. The pair correlation fun13tions are also derived, using the second of these
methods. These correlation functions are utilized in a study of the ground state and the finite
temperature spin configurations. An interesting application of these results to the problem
of out-of-phase domain structure in binary alloys is discussed.

§ I. Introduction
The Ising chain is of interest in statistical mechanics because it can be
solved exactly and because it can be incorporated into some interesting problems
of statistical physics. This paper presents exact solutions for the partition func-
tion and pair correlation functions of a spin t Ising chain with both nearest and
second nearest neighbour interactions. The motivation for this work was two-
fold in nature.
Firstly, within a statistical mechanical context, it was our desire to obtain
methods of solving for the partition function and pair correlation functions of
an open ended*) spin t
chain of any length which did not rely upon the standard
matrix technique. The nearest neighbour Ising chain has so simple a solution
that in order to learn how to achieve this goal it is necessary to study a chain
with more than nearest neighbour interactions. Section 2 presents an inductive,
non-matrix method of solving for the partition function. There we present a
series representation which generates the appropriate polynomial solution for a
chain of arbitrary length. With this series representation we are able to see
the analytical structure of the partition function. Such transparency is not di-
2
rectly displayed by the matrix method which has been used in the pase), ) to
calculate the partition function.
Section 3 describes a second method. We present a difference equation for
the partition function and readily solve it to obtain a result in agreement with
that of § 2. In § 4 we calculate the pair correlation functions of Ising chain;
these results are new and are obtained using the difference equation method in-

An open ended Ising chain is one in which periodic boundary conditions are not imposed:
*)

the chain begins with site 1 and ends with site N.

 Domain Structure in a Second-Neighbour Ising Chain 1171

traduced in § 3. An interesting isomorphism between the nearest neighbour spin

t Ising chain in an external magnetic field and this second neighbour chain can
be derived and is discussed in Appendix A. Appendix B describes the exten-
sion of the techniques of § 3 to the third nearest neighbour Ising chain.
Secondly, we were interested in using our model to study some aspects of
the problem of the formation of long period superlattices 3 ) in binary alloys. It
is well known that an isomorphism exists between the Ising model and the binary

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

alloy. 4> If the problem of a three dimensional binary alloy could be reduced to
an effectively one dimensional model, then our results could be used to give
some guide to the type of behaviour to be expected. Such a model has been
proposed by Cowley and will be discussed in § 7. We have therefore used our
formulae for the pair correlation functions to study, in § 5, the ground state spin
configurations, and in § 6 the temperature dependence of the spin correlations.
Section 7 contains a discussion of these results in the light of these motivating
forces.

§ 2. Partition function-series method

Consider a one dimensional chain of N spins, where each spin si = ± 1. We

will assume that the only interactions present in the system are those between
nearest neighbour and second nearest neighbour spins, the former having the
form - J 1sisi+l and the latter the form - J 2sisi+ 2 • The Hamiltonian of the system
is then
N-1 N-2

HN= -J1 ~ sisi+l-J2 ~ sisi+2. (1)

i=l i=l

J 1 and J 2 are the interaction strengths-with our sign convention a positive in-
teraction strength corresponds to a ferromagnetic interaction between spins. The
partition function, QN, is the sum over all zN possible spin configurations of the
function exp (- HN/kT):
N-1 N-2

QN= ~ exp (K1 ~ sisi+l + K2 ~ sisi+2),

{8i} i=l i=1
Using the identity
exp (Kss') =cosh K · (1 + ss' ·tanh K)

we can rewrite the partition function as

N-1 N-2
QN= coshN-l K1 · coshN- 2K2 ·~{II (1 + asisi+1) ·II (1 + {3sisi+2)},
{st} i=l i=l

where a= tanh K 1 and {3 =tanh K 2 •

We introduce a new set of variables ti, defined in terms of the si variables
by the ·equations
1172 N. E. Frankel and D. C. Rapaport

i=1, 2, ···N-1.
There is a one-to-one correspondence between the sets {si} and {ti} because each
si is uniquely expressible· as a product of ti:

i>1.
The partition function becomes
ON=2N coshN- 1K1·coshN- 2K2·QN,

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

(2)
where the function QN is defined in terms of the ti as
N-1 N-2
QN= 2-N L; {11 (1 + ati) ·IT (1 + (3titi+l)}. (3)
{ti} i=l i=l

We must now expand the products in (3) and sum over each of the ti. It is
obvious that, since each ti can take the values + 1 and -1, the only terms of
the expanded form of the product which contribute to (3) are those which are
either independent of ti or contain ti twice (because t/ = 1) for each value of
z. The final expression for QN will be a polynomial in a and (3 whose first term
is unity; our problem is to determine the coefficients of the remaining terms.
If N is small the problem of finding the polynomials is trivial: the first
few are

We continue this sequence until a definite pattern emerges. We are then able
to write down the general form of QN:

QN= L;
(N-l)/2N-m-1(N -i-1 ) ( i-1)
L; a2m(3i,
m=O i=m m m-1

where [x] denotes the greatest integer <x. Using the identity ( t) = (L~A.)
and relabelling, we obtain

Now smce

the term

N -f-1 ( N- .i- m -1 )(i + ~ -1) (3i

i=o N-z-2m-1 z
1s just the coefficients of 2
zN- m-l in the power series expansiOn of
 Domain Structure in a Second-Ne ighbour Ising Chain 1173

This coefficient may also be expressed in the form of a contour integral

J. (N O) - 1 f dz
m 'IJ - 2ni J o ZN-2m(1-{ 3z)m(1-z)m +1 '
where the contour C encloses only the pole at the origin. Thus

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

and if we reverse the order of summation and integration we find that

QN=J:_ f dz .1-{3z- (a 2{3z 2 /(1-z)) (a 2{3z 2/(1-{3z) (1-z))C<N - 1)/2J

2ni o zN (1- {3z) (1- z) - a 2{3z 2

The third term of the numerator makes no contributi on to the value of QN; we
can then evaluate QN by expanding the remainder of the integrand in terms of
partial fractions. Let

a={3+1,
the final result is then

As N-HXJ, {(a-b)/(a+ b)}N--70, so that in the thermodyn amic limit

QNrvl_(1 - (a-b))( a+b)N-1.

b 2 \ 2
The partition function of the infinite chain is therefore

ln Q=lim(N - 1 ln QN)
N--H:IJ

(5)
By applying the binomial theorem to (4) we obtain a more useful form of QN:
QN= (!!:__)N-1
2
{(Nff/2 ( N ) (b:) 2(1-l_) (N~/2( N -1 )(b:) r}.
2r+ 1 , a
r _
6)
(
r=O ' a 2r+ 1 ar=O

This result will be used when we later require an expression for QN where a
is replaced by its reciprocal , in which case b is imaginary.

§ 3. Partition function- differenc e equation method

In order to obtain a recurrence relation for QN we separate out the first
term of each of the products in (3):
N-1 N-2
QN = 2-N ~ { (1 + at1) (1 + {3t1t2) II (1 + at1) ·II (1 + {3titi+l)}
- {ti} i=2 i=2
1174 N. E. Frankel and D. C. Rapaport

N-1 N-2 .
+ a2-N :E {t1 II (1 + ati)
i=2
·II (1 + {3titi+l)}
i=2
{ti}

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

(7)

We introduce a function XN, defined as

The second and fourth terms of (7) are zero because they contain an unpaired
t 1 ; if the indices of the remaining two terms are reduced by one we obtain the
result
(8)

Similarly,
(9)

In (8) and (9) we have a pair of simultaneo us linear difference equations. To

solve for QN we introduce the operator E having the property that Ef(N)
= f(N + 1). XN can be eliminated between the two equations, the resulting
second order difference equation is
(10)

The solution to (10) can be written down immediate ly, it IS

QN= A;t+ N + B;t_ N,

where

and A, B are constants which depend on the boundary condition s-in this case
the expression s for Q2 and Q3 which were obtained earlier. Finally
QN= [ ({3 _ 1)2 + 4a2{3]-112 [A+ N _A-N_ {3 (A+ N-1_ A- N-1) J;

it is a simple matter to show that this reduces to ( 4), the result obtained from
the series method.

§ 4. The spin correlatio n functions

The correlation between the spins s~ and s~+r on a chain of length N is de-
fined as
 Domain Structure zn a Second-Neighbour Ising Chain 1175

(11)

m the notation of § 2. Equivalently,

l-1 l-2
= 2-N :E {[II (1 + ati) ·II (1 + (Jtiti+l)] · (1 + (Jtl-ltl)

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

{ti} i=l i=l

l+r-1 l+r-2
x [ II ti (1 + ati) · II (1 + (Jtiti+l)] · (1 + (Jtl+r-ltl+r)
i=l i=l
N-1 N-2
x [ II (1 + ati) · II (1 + (Jtiti+l)]}; (12)
i=l+r i=l+r
should the lower limit of any of the products in (12) exceed the upper limit
then that product is replaced by unity. (By definition ..78~ 82 = 0.) Since

ti (1 + ati) =a ( 1 + ~ ti)
we find that (12) can be expressed as

QN.Yt;8l+r = ar QlQr+l ( ~) QN+l-l-r [ 1 + {3.Yl 8 1 2.Y8:~1 ~) (

+ (3 (t?r+l
,J 8182 1 )Y,N+l-l-r
(~ 8182 + {32Y,l8182 g,r+l ( 1 )Y,N+l-l-r]
81828r.~r+1 a 8182 '
(13)

where we have defined the four spin correlation function .Y8:~;r 8 r• 1 as the obvi-
ous generalization of (11). (Note that .Y8~s 2 8 1 s 2 = 1 and Ys~s 2 8 2 s 8 =..7;1 8 8 . ) The cor-
relation functions and the QM are polynomials in a and {3, except where the
arguments of these functions are given explicitly as 1/a, in which case each a
in the polynomial is replaced by its reciprocal-as an example

Treating QN itself in the same way we obtain

QN= QlQr+lQN+l-l-r [1 + f3,Yl1 8 2 .Y8:~1

+ (J g,r+lY,N+l-l-r + {32Y,l
8182 8182 8182 8182srsr+1Y,N+l-l-r]
g,r+l
8182 • (14)

.Y8~l•r is then found by dividing (13) by (14).

For the special case l=r=l,

Y,N _ a+(3..78~~
1

8182-
1 +a{JY,N-1
8182
since Q2 = 1 and ..78~8 2 =a. We let N~oo and define the quantity x as
II76 N. E. Frankel and D. C. Rapaport

X = lim .?8~82 ,
N~ao

then x satisfies the equation

a+{3x
x= --"'---
I +a(3x
whose solution Is
2
x= (3 -I+ [ ((3 -IY + 4a (3]1 12

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

(I5)
2a(3
(the second solution is unphysical).
We now consider the general case and let N~oo-replacing .9:~82 by x,

<sLsL+r) =lim Ys~8L+r

N-->ao

= arQr+1 (I/ a) [I+ (3.?8:~ (I/a) . (x + ._9?8Ll82) + f3 xY8:~;r8r+l (I/a)

1 2
J.
I+f3Y's:~ • (x+YsL1 82) +{3 X.?s:~!rsr+l
1 2
Qr+l
(I6)
The four-spin correlation function may be expanded in a form similar to (I2)
and expressed in terms of the partition function and the correlation functions of
a shorter chain:

Also

hence

Qr+1g~~!r8r+l = 22 Qr+l + (a2- 22 ) Qr-1 + 2{3 (a- ~) Qr-1.?8:~1 ,

where r>3. We eliminate Ys:~ 1 by using (9) (from the definitions, XM=.?8~ 2 )
and obtain

Considering (16) we see that the spin index l appears only in the correla-
tion function Y'sz1 8 2 • It might be argued that by taking the limit z~oo and re-
placing Ysz182 by x we would obtain the correlation function for two spins both
of which are an infinite distance from the ends of the chain. This is in fact
the case, when both nearest and second nearest neighbour interactions are fer-
romagnetic (J1 and J 2 >0), but if either (or both) are antiferromagnetic there
is a tendency, as we shall soon see, for sublattices to appear for certain ranges
of J 1 and J 2, and the index l must be retained in order to indicate which par-
ticular su blattice is being considered,
 Domain Structure in a Second-Neighbour Ising Chain 1177

§ 5. Ground state configurations

The four possible zero temperature configurations are shown in Fig. 1.

States I and IV are strictly equivalent, differing only in the location of the do-
main boundaries relative to one end of the chain. Neglecting end effects, the
ground state energies are

Eio=Ewo=2J2'

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

Err 0 = - 2Jl - 2J2 ,
Em 0 = 2Jl- 2J2 .
If J 2> -tjJ1j, states II and III are stable for J 1>0 and J 1<0 respectively. If
J 2<-tiJ1j, states I and IV are stable for J 1>0 and J1<0 respectively; the choice
of state in this case depends only on the relative directions of the first two spins
of the chain which, from a further simple energy argument, can be shown to
depend only on the sign of J 1 • These results can be displayed on a phase dia-
gram, Fig. 2, the J 1-J2 plane is divided into four regions, each corresponding
to a different ground state configu,ration.
The T= 0 behaviour may also be deduced from the limiting forms of the
correlation functions. Rather than considering the general expressions for the
correlation functions we will discuss three particular examples, and for simplicity
we shall assume that J1>0 and J 2 <0. The expression for <sls2) appears in (15)
and from (16) we can obtain the expressions for the two other correlation func-
tions we propose to consider, namely

<slsa)-
f1+ax , (18)
1 +af1x
aj1+x (19)
1 +af1x
For sufficiently small values of T, K 1 and K 2 are large, so
a= tanh Kl r v 1 - 2e- 2K + 2e- 4Kt,
1

/1 =tanh K 2 -1 +2e2K
rv 2
- 2e 4K 2
,

Substituting these expansions into (15) we obtain to lowest order

J2>-tJ1'
J2< -tJl;
in both cases <s1s 2) = 1 at zero temperature. If we now use these expansions In
the evaluation of the limits of (18) and (19) we see that to lowest order the
numerators and denominators are zero. On retaining the higher order terms we
find that at T = 0,
1178 N. E. Frankel and D. C. Rapaport

These results agree with those reached earlier, and also with those of Obokata
and Oguchi 5> which were obtained by taking the zero temperature limit of the
expression for the energy of the system.

I II

---lltf!ltt--- ---tttttttt--- ][ II

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

][ N:

---11111111--- ---11111111---
Fig. 1. The allowed ground state (T=OoK) spin Fig. 2. The ground state phase diagram.
configurations.

I
T'=O~

I
I

-I

---
e=-0·48

-I
I 2 3 4 R 5 6 1 8
--4 R 5 6
-,...--
T'=o/
7 8

e=a·4a·
--- ... - - - - - - -
01-----~---~---~~~~~~---~---~
~~2~

-I

Fig. 3. The spin correlation function <stSl+R> Fig. 4. The spin correlation function <sss5+R>
for different values of the ratio p(=JdiJtD for different values of the ratio p(=J2/IJtl)
and the temperature T'(=kT/IJtD· J1>0. and the temperature T'(=kT/!JtD· Jt>O.
 Domain Structure in a Second-Neighbou r Ising Chain 1179

T'=O ------c- -
e=-0·70 ~=·0·70

I
, I
\\_---'I

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

\ I
\__J

I A A I
I\ I \ I
I I \ \ I
€=-0·48 I € =·0·48 I \ \ I
ol--~1 ~~~~~~~~~--~\~--1~
I \ I
\ I
\ I
\I

3 4 R
v
5
-d.
I 2 3
v
7

T'=O A
I
A T'=O

T'=l·~ \
I r-T'=l·2

\ I
e=o·IO

-I

Fig. 5. The spin correlation function (slsl+R) Fig. 6. The spin correlation function (s5s5+R)
for different values of the ratio p( =J2/IJ1 !) for different values of the ratio p( =Jd!J11)
and the temperature T'(=kT/!J11). J1<0. and the temperature T' ( =kT/!Jd). J1<0.

§ 6. Numerical results
Using Eqs. (6), (15), (16) and (17), · we have numerically evaluated the
spin correlation function (sLsL+R> over a range of temperatures for both positive
and negative J1 and various values of the ratio p = J 2 /l J 1j. In Figs. 3 to 6 we
display the results obtained for L = 1 and L = 5; the graphs show how the cor-
relation between spins varies with the distance, R, between them-for various
values of the temperature T'=kTjjJll·
From the numerical results the general properties of the system may be de-
duced. When J 2 is positive the behaviour is qualitatively the same as the near-
est neighbour Ising chain: the ground state is ferromagnetic if J 1 >0 and anti-
ferromagnetic if J~ <O. For T>O the correlations are all positive for J 1 >0 and
oscillate in sign for J 1<O. Should J 2 be negative, the behaviour will depend
1180 N. E. Frankel and D. C. Rapaport

on whether or not p exceeds -t. If P> -t the ground state configuration is

identical to that for J2>0; as the temperature increases from zero some of the
correlation functions change sign one or more times and, of course, eventually
approach zero. On the other hand, if P<- t, the spins group in parallel pairs
at T = 0, with successive pairs anti parallel; depending on the sign of J 1 this
configuration may correspond to either I or IV. This domain structure disap-
pear~ for T>O and the signs Of some of the correlation functions change----.:the

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

behaviour at higher temperatures is similar to that of the previous case. Even
though both positive and negative correlations exist for T>O there is no resem-
blance between them and the domain structure of the ground state; in fact when
the temperature is greater than zero all spin sites are equivalent, apart from end
effects-a result which follows from the absence of long-range order when T>O.
As this calculation considers spins close to one end of an infinite chain we
have been able to investigate the effects of the open boundary conditions. A
comparison of the results for L = 1 and L = 5 will indicate the increased impor-
tance of end effects at lower temperatures (except, of course, T = 0) and also
that where the ground state is a simple ferro- or antiferromagnetic configuration
the end effects are less pronounced than for the case in which spin pairing oc-
curs; in each case· however, the end effects decrease steadily as L increases.
The reason why, in general, L is not allowed to become infinite should by
now be apparent. We have seen that for certain values of J1 and J 2 domain
structure exists at T = 0. The correlation functions therefore oscillate between
+ 1 and -1 as L changes. Hence no unique limit of (sLsL+R> exists (for gen-
eral R) as L-'>oo and the value of L (or at least its value modulo 2) must be
retained.
Bitterlich and Jelitto 6> have made a numerical study of the two-dimensional
model with second nearest neighbour interactions. They found that the system
exhibited only a single phase transition, but there was a choice of three possible
ordered ground states, depending on the ratio of the interaction strengths. Of
the three ground states one was ferromagnetic and the other two were anti-
ferromagnetic; their phase diagram was identical in form to the one given here.
The ground state configurations of the three dimensional model are discussed by
Clapp and Moss.7)

§ 7. Discussion

We have presented two different methods of solution for the partition

function in §§ 2 and 3. These methods are non-matrix methods, in that they do
not require the diagonalization of any matrix, but our. final result agrees with
that of the matrix methods, as it must. However, the se-ries representation of
the partition function in § 2 and the recurrence relation given in § 3 provide a
direction in which to search for a new analytical representation of the problem.
 Domain Structure in a Second-Neighbour Ising Chain 1181

In particular, it appears possible to use these results to establish the rela-

tionship between the solution of this and other Ising chains to the class of or-
thogonal and hi-orthogonal polynomials. With an analytical representation of these
solutions in terms of known functions it should then be possible to obtain dif-
ferential equations that the partition functions satisfy. While this is an ambitious
undertaking, we are encouraged by preliminary results presently at hand to con-
tinue this line of study as well as to attempt to extend it to higher dimensional

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

Ising models. The ultimate desire is that if the appropriate differential equa-
tions could be obtained, the analytical properties of the Ising model would then
be more amenable to study. Work in this direction is presently underway.
The pair correlation functions derived in § 4 have been used in a study of
the spin configurations as a function of temperature for all the values (both sign
and magnitude) of the nearest and second nearest neighbour interaction strengths
in §§ 5 and 6. It was pointed out in the introduction that these results could
be used in studying an effectively one dimensional binary alloy problem. It is
our intention to discuss this problem now.
The superlattice reflections in x-ray and electron diffraction patterns from
certain ordered binary alloys below their critical temperatures show splittings,
as do the diffuse maxima from certain disordered binary alloys above their criti-
cal temperatures. The splittings in the long-range ordered state of the alloy
below the critical temperature are attributed to the presence of a long period
superlattice composed of a periodic alternation of out-of-phase domains. The
splittings in the disordered state above the critical temperature are attributed to
the presence of finite but relatively large regions of the alloy where superlattice
(out-of-phase) structure exists. 3> These are small ordered regions in the disor-
dered structure which are a manifestation of the short-range order in the alloy
above its critical temperature. This is particularly true as the critical tempera-
ture is approached from a hove in which case the amount of short-range order
in the system increases. The separation of the split maxima is determined by
the period of the domain structure.
Cowley 8> has proposed an interesting model of the binary alloy in order to
interpret this phenomenon. He assumes that the alloy is composed of ordered
planes of atoms, as in the ordered structure. This is certainly correct for the
ordered state of the alloy, and the presence of a considerable amount of short-
range order in the alloy near and above the critical temperature would indicate
that it is a good approximation there as well. At an anti-phase domain boun-
dary in the alloy there is a shift of the ordered planes by an amount equal to
one half the face diagonal of a unit cell. An unshifted ordered plane will be
called an A-centered ordered plane, and a shifted plane will be called a R-
eentered ordered plane. Thus Cowley visualizes the alloy as a one dimensional
array of A- and B-centered ordered planes. There will be an interaction between
the planes arising from the interactions between the atoms of the alloy.
1182 N. E. Frankel and D. C. Rapaport

If we denote an A-centered ordered plane by the spin up direction and a

B-centered plane by the spin down direction, and assume there exists only ne-
arest and second nearest neighbour planar interactions, then we have an isomor-
phism between the Cowley model and our Ising chain.
At this point it is important to stress that in the Cowley model the order-
ing of out-of-phase domains is a separate problem, involving weak, long-range
energy terms (oscillatory potentials due to electron energies) from that of the

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

in-plane ordering. The Cowley model requires first of all the solving of the
ordering problem arising from short-range interactions. This gives a state of
order in the usual sense. For Cu3 Au and other similar alloys this implies (both
from approximate theory and experiment) out-of-phase boundaries parallel to 100
type planes-hence relative stability of planes of atoms perpendicular to cube
axes. Now having established this over all state of order with antiphase boun-
daries, Cowley introduces long-range, weak oscillatory potentials to get ordering
of anti-phase boundaries.*) This is the one dimensional problem which we are
interested in here.
Although there will be no phase transition in this model because the second
neighbour Ising chain has no phase transition, we expect the ground state of- the
Ising chain to correspond to that of the alloy. We would also expect that the
temperature dependence of the spin configuration above T=0°K (the effective
Tc for the Ising chain) would provide some indication of the type of ordered
plane (domain) structure to be expected in the alloy above its critical temperature.
Thus, by studying the temperature dependence of the pair correlation func-
tions of the Ising chain for different values of the interaction strengths, we can
learn something about the ordered plane configurations. A configuration of all
spins up (or equivalently, all spins down) corresponds to a configuration of all
A-centred planes (all B-centred planes) and hence no out-of-phase domain struc-
ture. A periodic configuration of up and down spins corresponds to a periodic
configuration of A- and B-centred ordered planes, and therefore represents a
periodic out-of-phase domain structure for the binary alloy.
The spin configurations of the Ising chain were given in §§ 5 and 6; the
results are displayed in Figs. 1 through 6. Depending on the magnitude and
sign of the two interaction strengths, it is possible to find the following types
of domain structure: (i) no out-of-phase domain structure at T= 0°K but the
presence of such structure at temperatures above zero; this corresponds to values
of the interaction strengths in a portion of section II of the phase diagram; (ii)
out-of-phase domain structure at both T=0°K and T>0°K-the different types
of domain structure correspond to the sectors I, III and IV of the phase diagram.
Accordingly, the presence or absence of splittings in the diffraction patterns
would be determined by the values of the interaction strengths corresponding

*> MosslO) has recently used the Friedel oscillatory potential to interpret the splitting of the
diffuse spots from Cu-Au alloys above Tc.
 Domain Structure in a Second-Neighbour Ising Chain 1183

to the appropriate sector of the phase diagram.

It is also interesting to contrast the domain structure which occurs at the
ends of the chain with that which occurs within the chain. The difference can
be clearly seen in Figs. 3 through 6. The spins (ordered planes) at the ends
of the chain are more strongly ordered than those away from the ends. This
is a manifestation of the end effects which occur in an open ended chain.
In short, we see that it is the competition between ferromagnetic and anti-

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

ferromagnetic coupling which gives rise to the different types of domain struc-
ture. These results, however, should only be taken as instructive, as it is well
known that forces of longer range play an important role in binary alloys. Nev-
ertheless, Cowley's picture of the binary alloy makes for an interesting physical
application of the analytical results presented in this paper.

Acknowledgements

We are indebted to Professor ]. M. Cowley for discussing his model with

us and for many valuable and stimulating discussions. We are grateful to Pro-
fessor S. C. Moss for his interest in this work and for many rewarding con-
versations. One of us (N.E.F.) would like to thank Professor B. W. Ninham
for stimulating the analytical motivation for this work and Professor C. J. Thomp-
son for a most useful discussion. One of us (D.C.R.) would like to acknowledge
the financial assistance of a Commonwealth Postgraduate Award.

Appendix A

It is a simple matter to demonstrate that the solution to the nearest neighbour

Ising problem in a magnetic field can be obtained from our results. The Hamil-
tonian of a chain of N -1 sites in a field B is
N-2 N-1
Hfv -1 = - J ~ SiSi+l- fJ.oB ~ Si •
i=1 i=1

The partition function is then

If we replace /3J by our K 2 and /3fJ. 0B by our K 1 we see that

0 1 -1Q N,
-N-1-2

On being the partition function of a chain of N sites with second nearest neigh-
bour interactions. The spin correlation functions for the two systems are clearly
identical in form if the same change of variable is made.
1184 N. E. Frankel and D. C. Rapaport

Appendix B
We can extend the difference equation method to the third neighbour pro-
blem. If - J3sisi+S is the interaction between spins si and si+S we modify the
Hamiltonian in (1) by the addition of a term
N-3
~Js ~ SiSi+S.
i=l

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

The polynomial part of the partition function is then
N-1
QN= z-N ~ { (1 + asls2) (1 +{3slss) (1 + rsls4) . II (1 + asisi+l)
{Si} i=2

N-2 N-3

X II (1 + {3sisi+2) . II (1 + rsisi+S)}'
i=2 i=2

r
where =tanh (f3Js) (no advantage is gained by expressing QN In terms of the
ti variables). We introduce the functions

where Y/'{s 1 is the spin correlation function-

N-l N-2 N-3
II (1 + rsisi+s)}'
YsZ.:Sj = z-N ~ {sisj. II (1 + asisi+l) . II (1 + {3sisi+2) . i=l
i=l
{si} i=l
and obtain a set of linear difference equations-

(1-E)QN+a{3XN+arYN+f3rZN=0,
aQN+ ((3-E)XN+rYN+a(3rZN=0,
f3QN+ aXN+ (af3r- E) YN+ rZN= o ,
a{3QN+XN+f3rYN+ (ar-E)ZN=O.
These reduce to a single fourth order difference equation

{E 4 - ((3 + 1) (ar + 1)E 3 + (3 (1- a 2) (1- r 2)E 2

+ r (a- r) (1- (3) (1- a 2) (1- /3 2) E + r 2(1- a 2) (1- /3 2)} QN= o.
We shall not proceed any further with the third nearest neighbour problem as
it requires solving a quartic equation. In any case, the second nearest neighbour
chain displays all the features in which we are interested; nothing essentially
new is provided by the solution to the more difficult problem.
In conclusion we should point out that Fisher and T emperley 9) have discussed
the second and third nearest neighbour problems. Their results are expressed
in the form of algebraic equations but, whereas our equations are of second and
fourth order for the second and third neighbour problems respectively, their equ-
ations are of fourth and eighth order. A similar observation has been made 2 )
in connection with the transfer matrix approach to this problem, namely that the
introduction of the ti variables reduces the matrices to half their original size.
 Domain Structure in a Second-Neighbour }sing Chain 1185
Reference-s

1) E. W. Montroll, ]. Chern. Phys. 10 (1942), 61.

2) J. F. Dobson, ]. Math. Phys. 10 (1969), 40.
3) H. Sato and R. S. Toth, Long Period Superlattices in Alloying Behaviour and Effects in
. Concentrated Solid Solutions (Gordon and Breach Science Publishers, Inc., New York,
1965).
]. M. Cowley, ]. Aust. Inst. Metals 11 (1966), 258.
4) P. C. Clapp and S. C. Moss, Phys. Rev. 142 (1966), 418.

Downloaded from https://academic.oup.com/ptp/article/43/5/1170/1921747 by guest on 01 February 2024

5) The second nearest neighbour problem has been treated via a generalization of the Bethe
approximation by T. Obokata and T. Oguchi, J. Phys. Soc. Japan 25 (1968), 322.
6) W. Bitterlich and R. ]. Jelitto, Phys. Stat. Sol. 28 (1968), 365.
7) P. C. Clapp and S. C. Moss, Phys. Rev. 171 (1968), 754.
8) J. M. Cowley, private communication and to be published.
9) M. E. Fisher and H. N. V. Tern perley, Rev. Mod. Phys. 32 (1960), 1029.
10) S. C. Moss, Phys. Rev. Letters 22 (1969), 1108.