Timedependent Statistics of The Ising Model: Additional Information On J. Math. Phys

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

TimeDependent Statistics of the Ising Model

Roy J. Glauber

Citation: J. Math. Phys. 4, 294 (1963); doi: 10.1063/1.1703954


View online: http://dx.doi.org/10.1063/1.1703954
View Table of Contents: http://jmp.aip.org/resource/1/JMAPAQ/v4/i2
Published by the American Institute of Physics.

Additional information on J. Math. Phys.


Journal Homepage: http://jmp.aip.org/
Journal Information: http://jmp.aip.org/about/about_the_journal
Top downloads: http://jmp.aip.org/features/most_downloaded
Information for Authors: http://jmp.aip.org/authors

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
JOURNAL OF MATIIEMATICAL PHYSICS VOLUME 4, NUMBER 2 FEBRUARY 1963

Time-Dependent Statistics of the Ising Model*


Roy J. GLAUBER
Lyman Laborawry of Physics, Harvard University, CanWridge, Massachusetts

The individual spins of the Ising model are assumed to interact with an external agency (e.g., a heat
reservoir) which causes them to change their states randomly with time. Coupling between the spins
is introduced through the assumption that the transition probabilities for anyone spin depend on
the ".a~ues of the neighboring ~I?in:" This dependence is determined, in part, by the detailed balancing
condItion obeyed by the equihbnum state of the model. The Markoff process which describes the
spin functions is analyzed in detail for the case of a closed N-member chain. The expectation values
of the individual spins and of the products of pairs of spins, each of the pair evaluated at a different
?m:' are found explicitly. The influence of a uniform, time-varying magnetic field upon the model
IS dIscussed, and the frequency-dependent magnetic susceptibility is found in the weak-field limit.
Some fluctuation-dissipation theorems are derived which relate the susceptibility to the Fourier
transform of the time-dependent correlation function of the magnetization at equilibrium.

INTRODUCTION We have attempted, therefore, to devise a form of


the Ising model whose behavior can be followed
T HE statistical study of systems of strongly
interacting particles is beset by many problems,
largely mathematical in nature. These difficulties
exactly, in statistical terms, as a function of time.
While certain of the assumptions underlying the
have motivated theorists to devote a great deal of model are to a degree arbitrary, it is surely one of
effort to devising and studying the simplest sorts the simplest ones involving N coupled particles for
of model systems which show any resemblance to which exact time-dependent solutions can be found.
those occurring in nature. The property most desired The model we shall discuss is a stochastic one.
in these models is mathematical transparency. The The spins of N fixed particles are represented as
deeper insights offered by the possibility of exact stochastic functions of time Uj(t), (j = 1, ... N),
treatment are intended to compensate for any un- which are restricted to the values ±1, and make
realistic simplifications in the formulation. The first, transitions randomly between these two values.
and most successful of these models is one intro- These transitions take place because of the inter-
duced by Ising in an attempt to expl~in the ferro- action of the spins with an external agency which
l

magnetic phase transition. While many generaliza- may be regarded as a heat reservoir. The transition
tions of this model have been studied, we may note probabilities of the individual spins, however, are
that the first true understanding of a phase transi- assumed to depend on the momentary values of the
tion in an interacting system was reached by neighboring spins as well as on the influence of
Onsager2 for the case of the two-dimensional Ising the heat bath. It is for this reason that statistical
model. correlations arise between the values of neighboring
If the mathematical problems of equilibrium sta- spins. The coupling of the spins through their transi-
tistical mechanics are great, they are at least rela- tion probabilities makes it necessary, in mathe-
tively well-defined. The situation is quite otherwise matical terms, to deal with the entire N-spin system
in dealing with systems which undergo large-scale as a unit. The spin functions form a Markoff process
changes with time. The principles of nonequilibrium of N discrete random variables with a continuous
statistical mechanics remain in largest measure un- time variable as argument. Fortunately, if the coupl-
formulated. While this lack persists, it may be useful ing of the spins is not too complicated, the dif-
to have in hand whatever precise statements can ferential equations governing the probabilities may
be made about the time-dependent hehavior of be simplified greatly, making it possible to solve for
statistical systems, however simple they may be. all of the quantities of immediate physical interest
by elementary means.
* A brief. account of this work was given at the Washington In the sections that follow, we introduce first the
D. C. meeting of the American Physical Society, 1960 [R.i individual spins interacting with the heat bath,
Glauber, ~ull. Am. Phys. Soc. 5, 296 (1960)].
1 E. ISing, Z. Physik 31, 253 (1925).
then the means by which they are coupled to one
2 L. Onsager, Phys. Rev. 65, 117 (1944). another. The description of the behavior of the model
294

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 295

is then formulated as a matter of solving for the u(t) taking on the values u = ±I we have
expectation values of the spin functions and of their
q(t) = (u(t». (4)
products. We center the subsequent discussion
largely upon explicit solutions for the single-spin The equation obeyed by the mean spin is seen
and two-spin averages, since most of the interesting from (1) to be
properties of the system may be constructed in (5)
(djdt)q(t) = -aq(t) ,
terms of these. In addition we find the time-delayed
spin correlation function, i.e. the average product so that the mean spin simply decays exponentially
of two spin variables, each evaluated at a different with a relaxation time 1/a from whatever value it
time. We then describe the model in the presence is known to have initially,
of a uniform, time-varying magnetic field. Two results q(t) = q(O)e- •
al
(6)
of this generalization are a derivation of the complex
frequency-dependent magnetic susceptibility for We may regain the individual probabilities p(±I, t)
weak fields, and a discussion of fluctuation-dissipa- from a knowledge of q(t) by means of the identities
tion relations which hold when the field-induced (2) and (3) which together yield
departures from equilibrium are small. Our efforts,
in the present paper, are confined to treating a one-
p(u, t) = HI uq(t)], (7) +
dimensional model which, as already indicated by
MANY-SPIN SYSTEM
the treatment of the Ising model at equilibrium,2
appears to be a great deal simpler thllndealing with Particles such as the one we have just discussed,
the model in two or more dimensions. each of them responding to a random spin-flipping
agency, will form the basic units of the model we
SINGLE-SPIN SYSTEM
wish to describe. We shall assume that these par-
It may be helpful in introducing our model to ticles are arranged in a regularly spaced linear array
begin by discussing the most simple of such systems: which may be closed to form an N-particle ring.
a single particle whose interaction with a heat The dynamical resemblance between this model and
reservoir of some sort causes its spin to flip between the Ising model rests on the assumption that the indi-
the values U = 1 and U = -1 randomly, but at a vidual spins of the ring are not wholly independent
known rate. We assume that no magnetic field is stochastic functions. We may, for example, introduce
present so that neither of the states U = ±1 is a tendency for a particular spin U; (j = 1 ... N) to
preferred. Then, if the rate per unit time at which correlate with its neighboring spins by assuming
the particle makes transitions from either state to that its transition probabilities between the states
the opposite one is written as a/2, the probability U; = ±I depend appropriately on the momentary
p(u, t) that the spin takes on the value U at time t spin values of the other particles. To treat any such
obeys the equation model we must consider the entire ring as a unit
and introduce a set of 2N probability functions
(djdt)p(u, t) = -!ap(u, t) +
!ap(-u, t). (1)
P(Ul' ... UNt), one for each complexion, i.e. each
This equation, or more properly, this pair of equa- set Ul, ... UN for the ring.
tions for U = ±I, preserves the normalization condi- If we let Wj(Uj) be the probability per unit time
tion that the jth spin flips from the valueuj to -Uj,
while the others remain momentarily fixed, then we
p( 1, t) + p( - 1, t) = 1. (2)
may write the time derivative of the function
The pair of equations is therefore immediately P(Ul' ... UNt) as
reducible to a single equation for a single unknown
function. A convenient choice of the latter function
is the difference of the two probabilities
q(t) = p(I, t) - p( -1, t)
+ L: w;( -U;)P(Ul'
j
... -U;, ... uNt), (8)

= L:
(1-·1
up(u, t), (3) i.e., the complexion Ul, ... UN is destroyed by a
flip of any of the spins U;, but it may also be created
which is simply the expectation value of the spin by spin flip from any complexion of the form
as a function of time, i.e. if we think of the time- Ul, ... -Uj, ... UN. We shall refer to Eq. (8) as
dependent spin variable as a stochastic function the master equation since its solution would con-

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
296 ROY J. GLAUBER

tain the most complete description of the system If the spins other than U 1 are considered as fixed,
available. the stochastic model described by (8) and (9) will
approach an equilibrium in which
CORRESPONDENCE WITH THE ISING MODEL
Pie -U;) _ W;(U;)
We have already mentioned that the transition (13)
p;(U;) - W;( -U;)
probabilities Wj(u";) may be chosen to depend on
neighboring spin values as well as on u;. If we want, 1- hU;(Uj-l + Uj+1)
(14)
for example, to describe a tendency for each spin 1+ hu;(uj-I + Uj+1)
to align itself parallel to its nearest neighbors we The exponentials which occur in the ratio (12) may
may choose the probabilities Wj(u;) to be of the form
be written in the forms
(9)
exp [±(J /kT)u;(uj_1 + U;+1)]
which may be seen to take on three possible values
W;(Uj) = !a(1 - 'Y), !a(1 + 'Y)' (10) = cosh [k;' (U;_I + Ui+l) ]

[k~
The value !a corresponds to the case in which the
± UI sinh (Uj_1 + Uj+l) ] (15)
neighboring spins are antiparallel, Uj_l = -Uj+l'
When the neighboring spins are parallel to each
other the transition probability takes on the value
!a(1 - 'Y) for Uj parallel to the two of them or
= cosh (Uj_1 + Ui+I) ] [k~
!a(1 + 'Y) for Uj antiparallel. Clearly as long as 'Y
is positive the parallel configurations will be longer- ± !u;(Uj_1 + U;+I) tanh X{I(16) ~~},
lived than the antiparallel ones and we shall be
dealing with a model having ferromagnetic tenden- the latter of which is readily checked for the three
cies. Conversely negative 'Y will mean a tendency values the function can take on. The correspondence
of neighboring spins to remain aligned oppositely, between the ratios of the equilibrium probabilities
and will describe the antiferromagnetic case. We (12) and (14) may evidently be made precise by
note, incidently, that I'YI may not exceed unity. identifying the constant 'Y as
The parameter a which occurs in the transition
probabilities simply describes the time scale on 'Y = tanh (2J /kT). (17)
which all transitions take place. It has, of course,
We should mention that the particular choice we
no analog in the familiar discussions of the Ising
have made for the way in which the transition
model at equilibrium. The parameter 'Y, however,
probabilities (9) depend on neighboring spin values
describes the tendency of spins toward alignment
is motivated more by the desire for simplicity than
and thereby determines the equilibrium state of
for generality. There exist other, but less simple,
the present model much as the exchange interaction
coupling schemes which also yield the same equilib-
does in the Ising model. To indicate the quantitative
rium states as the Ising model with nearest-neighbor
correspondence between the models we write the
interactions. Some of these are discussed in the
Hamiltonian for the linear Ising model as
Appendix. There exists, furthermore, the possibility
(11) that each spin is coupled through the transition
probabilities to some or all of its more distant
When the Ising model has reached equilibrium at neighbors. We shall mention this possibility further
temperature T, the probability that the jth spin at a later point. For the present we shall continue
will take on the value U I as opposed to - U I (for a to deal with the transition probabilities (9) and
given set of values of the neighboring spins) is just discuss the mathematical treatment of the master
proportional to the Maxwell-Boltzmann factor equation based on them.
exp (-JC/kT). The ratio of the probabilities PI( -UI)
and PI(UI) corresponding to the two states for the REDUCTION OF THE PROBABILITY FUNCTION
jth spin is therefore
The functions P(Ul' ... UNt) which satisfy the
pj(-UJ _ exp [-(J/kT)u;(U;_1 + U;+l)] master equation (8) furnish, as we have noted earlier,
(12)
p;(u;) - exp [(J /kT)u;(Uj_l + U;+1)] . the fullest possible description of the system. While

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 297

we cannot deny that it would be desirable to know 1


these functions in their entirety we must neverthe- P(O'I' ... UN, t) = 2N {1 + ~ O'jqj(t)
less point out that, for N large, they contain vastly
more information than we usually require in practice.
+ L: j;'k
O'jO'krj,k(t) + ... }, (22)
To answer the most familiar physical questions about
which exhibits a general expansion of the probability
the system, in fact, it suffices to know just the
functions in terms of the expectation values of the
probabilities that individual spins or pairs of spins
spins and their products taken two at a time, three
occupy specified states. Alternatively, we need know
at a time, etc., i.e. the functions 1 and 0' form a
only the expectation values of spins or the average
complete orthogonal basis for the expansion of any
products of pairs of spins. Most of our attention
function of 0', and (22) is just such an expansion
in the present paper will be devoted to discussing
with N independent variables. The relation (7) for
just these functions. However before proceeding to
a single spin is a trivial example of the expansion.
the discussion, it may be helpful to indicate some
The reduced probability functions which furnish
general relations between the probability functions
the probabilities that individual spins or pairs of
and the expectation values of products of spin
spins occupy specified states, whatever may be the
variables.
states of the remaining spins, are defined by
We define the functions qj(t) to be the expectation
values of the spins oAt) regarded as stochastic func- p;(O'j' t) = L: P(O'I' .•. UN, t), (23)
(v;'vtl
tions of time:

qj(t) = (oAt» (lS)


Pjk(O'j, Uk, t) = L: P(O'I,'" UN, t), (24)
(6"''''i,''''''

= L: O'jP(O'I,
(v)
••• O'Nt). where the notation is intended to indicate summation
over all the spin variables save O'j in (23) and O'j
Here and in future work we designate by a sum and Uk in (24). If these summations are carried out
over {O'}, a sum carried out over the 2N values of upon the form (22) for P(Ul' ... UN, t) we find
the set 0'1, .•• UN' The functions r j ,k(t) are defined, Pj(O'j, t) = !{1 + O'jqj(t)}, (25)
likewise, as the expectation values of the products
Clj(t)O'k(t) : Pjk(O'j, Uk, t) = i{l + O'jqj(t)

rj .k(t) = (O'j(t)O'k(t»
(26)

= L: O'jO'kP(O'I, ••• O'Nt). (19) It should be clear that by solving for the expectation
(vi values of the spins and their products we are begin-
ning a systematic expansion of the probability
We note in particular that the "diagonal" expecta-
functions as well as finding the quantities of greatest
tion values r j , j are identically unity:
physical interest.
rj ,;(t) = 1. (20) As a preliminary step to finding the time-depen-
dent equations satisfied by the expectation values,
We next construct a general identity relating the we may write the master equation (S) in the more
probability to the expectation values as follows: compact form
Let 0' j and O'~ be two possibly different values of the
jth spin. Then the function!(1 + O';O'D equals unity d
dt P(O'I' ... UN, t)
for O'~ = O'j and zero for O'~ = -O'j. We may therefore
construct an identity expressing p(O'lj .•. O'Nt) as =- L: u'" L:
v.'
U~W",(O'~)P(Ul' .•. U~, •.• UN, t). (27)
a sum over all spins by writing
If we mUltiply both sides of this relation by Uk and
sum over all values of the 0' variables we obtain
(d/dt)qk(t) = -2 L: O'kWk(O'k)P(O'I, " . UN, t)
x L:
ler')
(1 + 0'10'0 ... (1 + O'NO'k)p(O'f, .•. O'~, t). (21)
(2S)
If we expand the product in the summand of this Similarly, if both sides of (27) are multiplied by the
relation and carry out the indicated summations, product O'jO'k (where j ~ k) and summed over the
we find 0' variables we obtain

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
298 ROY J. GLAUBER

d the Bessel functions of imaginary argunient,3


dt rj,k(t)

-2 L UjUk{Wj(Uj) + Wk(Uk)}P(Ul, , .. tJ'N, t)


exp [!x(X + X-I)] = L'" X"I,,(x) , (35)
\er)
where
(29)
(36)
If we substitute the form (9) for the transition
probabilities in (28) we obtain a recursive system Hence the time-dependent generating function is
of differential equations for the expectation v~lues given by
qk(t) :
(djdat)qk(t) = -qk(t) + h!qk-l(t) + qk+1(t)} , (30)
F(A, t) = F(A, O)e-"" L'"
k--a>
XkIk('Yat). (37)

An analogous system of equations for the expecta- We consider first the case in which all of the spin
tion values of products of pairs of spins results from expectations qk vanish initially except for one, which
the substitution of (9) in (29). For j rf k we have we may choose to be the one at the origin
(djdat)rj,k(t) = -2r j ,k(t) + hfrU-l(t) (38)
+ rj .k+l(t) + rj-l.k(t) + rj+l.k(t)}, (31) Then the initial value of the generating function is
just unity, and at later times it is
while for j = k, the functions obey the identity (20).
These equations, as we shall see, may be solved
quite readily. It is worth noting, however, that the
F(X, t) = e- a • L'"
k--co
XkIk('Yat ), (39)
assumption of forms different from (9) for the
from which we conclude, by comparing with (32),
transition probabilities leads, in many cases, to
that the spin expectations are given by
systems of equations in which the expectation values
of products of differing numbers of spins are coupled (40)
in each equation. Such systems are considerably less
An examination of the functions Ik shows that qo
tractable than the present one.
decreases steadily to zero as time increases, while
the neighboring spin expectations rise from zero to
SOLUTION FOR THE AVERAGE SPINS: positive values for a while as a form of transient
INFINITE RING
polarization induced by the positive spin at the
The coupled differential equations (30) are par- origin. The functions qk for spins neighboring the
ticularly easy to solve for the case of an infinite origin rise for times t « kha as
ring, N ~ ex>. It is convenient, for this case, to qk(t) ~ (ljlkl!)(hat)lk' e-a'. (41)
alter slightly the scheme for numbering the spins
4
by labeling a particular spin as the zeroth and They then reach a maximum at a time given, for
designating those to one side of it with positive k » 1, by at ~ k(I - "( ) - \ and, for much larger
2

integers and those to the other side with negative times, decrease as
ones. We then construct the generating function qk(t) (211j'atr~e-"(I-'Y)I . (42) f"'V

'"
F(A, t) = L Akqk(t) , (32) The most general solution for the spin expecta-
k"'-oo
tion values, corresponding to an arbitrary set of
which, according to Eq. (30), satisfies the differential initial values qle(O), may clearly be obtained from
equation (40) by linear superposition,
(J/Jat)F(A, t) = -F(A, t) + h(A + A-1)F(A, t). (33)
qle(t) = e-
a
' L'" qm(O)Ik-m('Yat), (43)
m->=CO
The solution for the generating function is evidently
where we note that the functions I" for negative
F(A, t) = F(X, 0) exp [-at + h(X + A-l)atJ, (34)
3 See, for example, G. N. Watson, Bessel Functions
which furnishes us an implicit solution for the qk(t) (Cambridge University Press, Cambridge, England, 1958),
in terms of the initial values qk(O). To make the pp. 14 and 77.
solution an explicit one we note that one of the • The locations of the maxima and various other properties
of the functions e-uI..(x) for ~ ~ 1 ~re discussed by E. W.
factors in (34) is just the generating function for Montroll, J. Math and Phys. 25, 37 (1946).

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 299

order are the same as those for positive order, (51)


I" = I_ n •
a result which corresponds to the known absence
AVERAGE SPINS: FINITE RING of permanent magnetization in the linear Ising model
(with interactions restricted to a finite number of
A somewhat more general means of treating the neighbors). The net effect of the spin interactions
set of equations (30) for arbitrary N may be based is to reduce the coefficient in the exponent from the
on a system of normal modes for the spin expecta- a of Eq. (6) to a(1 - "I).
tion values qk. If we seek solutions to Eqs. (30)
in the form SOLUTION FOR ONE SPIN FIXED
qk(t) = Are-PI, (44) It is interesting to investigate the behavior of
the spin system when one of the spins is assumed
where A is a constant, then we have
somehow to be fixed or frozen. We shall, for sim-
II = a{1 - hG·- 1 + r)}. (45) plicity, consider the infinite ring and let the zeroth
spin, the one at the origin, take on the fixed value
The closure of the N-spin ring requires that the
0'0 = 1. Then the differential equations derived
solution (44) be periodic in k with period N, i.e.,
earlier for the qk(t) still hold for k ~ o. In particular,
that rN = 1. Hence there are N roots for r of the
for k = 1, we have
form
(52)
r.. = exp (27r'im/N) , m = 0,1, ... N - 1, (46)
while the equations for k > 1 assume precisely the
and for these the eigenvalues 11m are
form (30). This sequence of equations for k ~ 1 is
II .. = a{1 - "I cos (27rm/N)}. (47) an inhomogeneous one because of the constant term
on the right-hand side of (52). It possesses a non-
The system of mode functions = exp (27r'imk/N) qt' vanishing equilibrium solution, which satisfies the
forms a complete orthogonal basis on the ring.
recursion relation
Hence any solution to (30) may be written in the
form k ~ 0, (53)
N-l
qk(t) = L A m e(2rimk/Nl-. m l , (48)
where qo = 1. The solution to such a linear difference
m~O equation may be written as
where the constants Am may be solved for in terms qk = 'l]lkl, (54)
of the qio(O) by using the orthogonality theorem.
These constants are where '1/ satisfies the quadratic equation
'1]2 - 2"1- 1'1] +1 = O. (55)
A rn =~ ~
N L..J qz (0)e-2rimZ/N . (49)
Z-1 It is worth noting that the same quadratic equa-
tion for 'I] holds for negative values of k as for posi-
The solution for the spin expectation values in terms
tive values of k, i.e., the equation is unchanged by
of their initial values is thus
the substitution of '1]-1 for '1]. The roots of (55),
which are always real, form a reciprocal pair. One
member of the pair, 1'-1 {I + (1 - 'Y2 ),), always
N ., has absolute value greater than unity for II'I ~ 1
= e-
ol
L i--co
1-1
L qz(O)Ik-'+;N('YOtt). (50) and therefore is of no use in solving the problem for
an infinite ring. The correct root for 'I] has absolute
The latter form of the solution is obtained from the value less than unity and is given by
former by carrying out the summation over m
(56)
explicitly. That the solutions may be expressed in
this way is obvious from the fact that the problem For this value, using the correspondence (17) with
for a finite ring may be solved by inserting periodic the static Ising model, we find
initial values in (43).
A particular consequence of the solution (50) is 'I] = tanh (J /kT). (57)
the fact that the total magnetization always de- The solution (54) exhibits clearly the tendency
creases exponentially, of any spin, in this case a fixed one, to surround itself

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
300 ROY J. GLAUBER

with a "polarization cloud." (In the antiferro- insight into the behavior of these functions by
magnetic case, "{ < 0, the signs of the induced spins simplifying the problem so that they depend, in
will alternate.) The value of 1/ given by (57) is just effect, on only one index. It often happens, in fact,
the familiar short-range order parameter of the that our knowledge of the initial state of the system
Ising model. is characterized by translational invariance, i.e.,
To complete the solution of the time-dependent our initial knowledge about all of the spins is the
equations for the qk(t), with the zeroth spin fixed, same. Then rj ,k(O) can only depend on j - k, and
we need only note that (54) constitutes a particular no other dependence on j or k can be present at
solution of the inhomogeneous system. We may add later times. In that case it becomes convenient to
to it any solution to the homogeneous system of introduce the abbreviation
equations obtained by requiring qo to vanish at all
times. Such a boundary condition may easily be r m = rk,k+m (59)
satisfied by using the method of images, since the for the spin correlation functions. We shall consider
requirement qo = 0 separates the system into two this translationally invariant situation first and then
halves which do not influence each other. (The return to the more general one presently.
infinite ring need not be imagined as closed.) If we In the uniform case the functions r m are seen to
seek a solution to the homogeneous system of equa- obey the relations
tions in which the qk assume a particular set of initial
values, say Vk for k > 0, we may reach a solution
(d/dat)rm(t) = -2rm(t) + "{{rm_l(t) + rm+l(t)} (60)
for the positive-k half of the system by using the for m .,t. 0, and
general solution (43) and imagining that the initial (61)
values of the qk at the negative sites are given by
q-k(O) = -Vk for k > 0, and that we have qo(O) = O. Aside from a trivial change of a factor of two in the
Interpreted in this way for k > 0, the solution (43) coefficients, this is precisely the sequence of equa-
may be made to fit the correct initial conditions and tions we solved in the preceding section, for the
yet, since it remains odd at k at all times, meet the single-spin averages with the zeroth spin fixed. The
boundary condition qo(t) = 0 as well. An analogous factor of two in the coefficients affects only the time
imaging procedure solves the equations for negative scale in which the functions change. In particular,
k as well. the equilibrium solution on the infinite chain is
To find the general solution to the time-dependent again given by
equations with the zeroth spin fixed we must add (62)
together the particular solution (54) for the inhomo-
geneous system and the general solution, constructed where TJ is the short-range order parameter mentioned
by the method of images, for the homogeneous earlier. The time-dependent solution for arbitrary
system, i.e., we add to the solution 1/k the solution initial correlations may be constructed immediately
to the homogeneous system which corresponds for from (58). For m > 0 we have
k > 0 to the set of initial values qk(O) - The rr rm(t) = TJ'" + e- 2a • L
'"
[rl(O) - TJI]
resulting solution for k > 0 is I-I

qk(t) = 1/
k
+ e- L'"
al
(ql(O) - 1/ 1) x {lm-I(2"{at) - I m+/(2,,{at)}. (63)
I~I
As a particular example of the type of problem
x {lk-hat) - Ik+lC,,{at)}. (58) to which this result is applicable, we may suppose
An analogous solution exists for negative k values. that the spin system is suddenly subjected to a
For times t » (,,{a)-x, the solutions in all cases change of temperature; i,e" after coming to equilib-
decay exponentially to the equilibrium form. rium with a heat reservoir at temperature To, it is
suddenly placed in contact with another heat bath
at a different temperature T. In that case the initial
SOLUTION FOR THE SPIN CORRELATIONS
values of the rl are given by
We next turn our attention to the average values
of products of pairs of spin variables. The functions r/(O) = TJ~ = [tanh (J /kTo)]I , (64)
ri .• (t) which express these averages obey the two- and the way these relax into the equilibrium values
index system of Eqs. (31) for j .,t. k, and for j = k at temperature T is shown by (63).
obey the identity ri. i = 1. We can secure a rapid We return now to the general problem of solving

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 301

the two-index system of differential equations (31) Tj .j(t) = 0, is just the solution (66) antisymmetrized
without the simplifying assumption of translational in the two indices land m, i.e., for j ~ k and l ~ m
invariance. The system is an inhomogeneous one 2at {lj_,('¥at)Ik-m('¥at)
Tj ,k(t) = e-
because of the condition Tk.k(t) = 1, which plays
a role similar to that of the fixed spin in the preceding - Ij_"aC'Yat)Ik_I('¥at)} . (67)
section. The translationally invariant equilibrium The general solution to the homogeneous system is
' Tk,l = TJ Ik-ll , W h'ICh we have Just
so lutIon . d'lscussed,
obtained by superposing the solutions (67). In order
clearly satisfies the system of equations. It can be
to solve the inhomogeneous system with which we
used as a particular solution to the inhomogeneous
began, we must add the particular solution TJ i - to
k

system. To this particular solution, we must add a


the solutions we have just found. The form which
general solution to the homogeneous system ob-
satisfies the correct initial conditions for j ~ k is
tained by supplementing (31) with the conditions
j
Tk,k(t) = O. The solutions to these equations may be rj.k(t) = TJ - k + e- 2at L
[TI,m(O) - TJI-m]
I>m
obtained and the boundary conditions met by
generalizing the methods of the preceding sections X {Ij_zC'Yat)Ik-m('¥at) - Ij_m('¥at)Ik_zC'Yat)} , (68)
to deal with a two-index array Tj,k(t), i.e. a matrix,
which is the general solution for the expectation
rather than a linear sequence qj(t).
values of the spin products. When translational in-
If, for the moment, we ignore the boundary con-
variance holds, this solution may be seen to reduce
dition on Tu(t) and assume that Eqs. (31) hold
to (63) by applying the relation
even for j = k, it becomes a simple matter to solve
the equations by using a two-parameter generating '"
function analogous to (39). We then find that if I k (2x) = m~'" Ik+m(x)I m(X) , (69)
all of the initial values of Tj,k(O) vanish except one
which is a special case of the addition theorem for
which is unity, i.e., '
Bessel functions. 6
rj,k(O) = 8j1 8h " (65)
TIME-DELAYED SPIN CORRELATION FUNCTIONS
the solution for Tj ,k(t) is
The functions Ti,k(t), which we have discussed up
Tj,k(t) = e-2atlj_,('¥at)Ik_m('¥at). (66) to th~s point, describe whatever tendency the pairs
of spms Uj and Uk may have to be correlated in
Such solutions may be superposed to secure the
direction, on the average, at a particular instant
appropriate initial values and to meet the condition
of time t. Not all of the spin correlations of interest
however, have this instantaneous character. I~
Tu(t) = O. To satisfy the latter condition, we must
generalize to a two-index array the method of images
particUlar, variation of anyone spin at a given
used earlier.
instant induces polarizations among its neighbors
The matrix Tj,k(t) is, or course, symmetric. How-
which only become appreciable after finite intervals
ever, it is quite convenient to think of it as if it were
of time . .To describe correlation effects extending
antisymmetric. What we shall do is fix our attention
over an mterval of length t', we shall discuss the
for the moment, on the values of Tj ,k(t) for j > k functions (Uj(t)Uk(t + t'», i.e. the expectation values
and only attempt to deal correctly with these. We
of the products of the stochastic spin functions U·
evaluated at time t, and Uk ev~luated at time t + /
assume that these matrix elements take on their
correct initial values but that the elements Tk,j(O)
To evaluate these more general correlation func-
are given by -Tj,k(O) for j > k, and that Tj,j(O) = O.
tion~ we represent the values assumed by the spins
The matrix Tj,k which is thus assumed initially
antisymmetric, maintains its antisymmetry at later
at time t as UI, ••• UN and at the later time t t' +
as U~, ••• uJ. The probability associated with the
times and, therefore, always meets the condition
~pin values U.l' •.. UN at time t is P(Ul' ••• UN, t),
Tj,j(t) = O. In fact, it satisfies the sequence of
I.e., the solutIOn to the master equation which satis-
equations (31) including, in virtue of its antisym-
fies whatever initial conditions our physical knowl-
metry, the equation of the same form for j = k.
edge imposes. In order to carry out the averaging
We need not be embarrassed, therefore, by our in-
c?rrectly~ we must also know the probability asso-
clusion of the j = k equations in the arguments
CIated wIth the final configuration u'1, ... u'N at
leading to (66). •
tlme t + t'. The question we ask in determining that
The basic set of solutions we seek, which meets
the initial condition (65) and the boundary condition 6 Reference 3, p. 361.

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
302 ROY J. GLAUBER

probability is rather different from the one answered no correlations between spins in the initial state,
by p(u), ... UN, t), since we assume that the spins as would be true, for example, for infinite tempera-
are known to have the values Ul, ••• UN at time t. ture. The remaining terms of the series describe
The values Ul, ... UN are thus to be regarded as the stabilizing effects upon the kth spin of the
initial spin values in determining the probability polarizations which exist about it in the initial state;
of finding u~, ... uk at a time tf later. We shall write For either sign of 'Y, the addition of the effects of
this conditional probability for finding ui, ... uk neighboring spins in (75) makes the correlation func-
as P(UI' '" UN lui, .,. ukt). The expectation value tion decrease in magnitude more slowly with in-
we seek for the product of two spins may then be creasing t'.
constructed by summing over all possible values In all of our work to date, we have assumed that
of the sets 0'), ..• UN and u~, .. 'uk as follows: we are in possession of some knowledge about the
system at an initial time t = 0, and have sought, in
(Uj(t)O'k(t + I'») a probabilistic sense, to answer questions about the
L p(u),'" UNt)U;P(O'l> " . UN lui, ... U/rt')0'1. behavior of the system at later times. Of course,
{ull",')
the same questions may be asked in a reversed
(70)
sense. What may we say, on the basis of knowledge
The part of this summation which is to be carried at t = 0, about the behavior of the system at
out over the variables ui, '" uk may be regarded negative times? Since the dynamical properties of
simply as the expectation value of the kth spin our model are presumably reversible, there is no
when the spins are initially U1, ... O'N' We may need to construct or solve a new master equation~
then write The probabilities are simply even functions of time.
The time t is to be construed more generally as It I
LP(UI' •.. UN lui, ... ulr, t')u~ = qk(t') , (71)
(crt} in all of the probability functions we have calcu-
lated thus far. In particular, the time-dependent
where it is understood that the initial values of the
spin correlation function (75) may be written for
qk are given by qk(O) = (flo' For the case of an infinite
t = 0 and arbitrary t' as
chain, the functions qk(t') are given in terms of these
initial values by the general solution (43) as
(U;(O)(fk(t'»r = e-att' I L'" 1JIH+lll l ha It'\). (76)
qkCt') = e- at
' L uJk-IC"at'). (72) ,--co
I

SINGLE SPIN IN A MAGNETIC FIELD


By substituting (71) and (72) into (70) we find
(U;(t)O'k(t + t'» It is not difficult to formulate the equations which
describe the behavior of our model when it is placed
= e- at ' E'" Ik_l("at') L p(U) , ••• uNt)(f;(f!. (73) in a uniform magnetic field. The influence of the
1--'" {<') magnetic field H, which we suppose is parallel to
the axis of spin quantization, is to introduce a pre-
The summation over (fl, '" O'N, however, is just
ference of the spins for either the CT = 1 or the
the instantaneous correlation Tj.k(t) defined by (19).
0' = -1 state. For the most simple case, in which
The time-delayed correlation function, therefore,
only a single spin is present, the transition proba-
reduces to
bility from U to -u may be written as
CD

(O';(t)(fk(t + t'» = e-
a
!' L
1--0>
Tj,ICt)Ik-hat') , (74) w(u) = la(l - (3u).

where the functions T;.I (t) are given, in general, If we equate the ratios of the equilibrium probabili-
by the results of the preceding section. ties calculated according to the stochastic model
For the particular case of a system in thermal and according to statistical mechanics, we find
equilibrium at temperature T, the correlation func- p( -u) _ w(u) _ 1 - (30'
tion depends only on the interval t', i.e., p(u) - w( -u) - 1 {3u +
..
(U;(t)Uk(t + t'»T = e- at
' L 'l]IH+lllhat'). (75) _ exp [-(.uH/kT)u]
1--00 - exp [(.uH/kT)uJ
The term corresponding to l = k - j is the only = 1 - u tanh (.uH /kT)
contribution which would be present if there were 1 + u tanh (.u:H/kT) , (77)

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 303

where p. is the magnetic moment associated with the state as the Ising model if we choose
spins or, more concisely, we find the correspondence
wi(u;) = w;(u;)[l - U; tanh (p.H/kT)]
{3 = tanh (p.H/kT). (78)
...~ = w;(u;)(1 - (3u;)
The equation satisfied by the expectation value
of the spin is then
= !a{ 1 - {3u; + h({3 - u;)(U;_1 +U;+l)}' (84)
The difference-differential equations satisfied by
(d/dat)q(t) = (3 - q(t). (79)
the average spins and the average products are
In the work that follows, it will be interesting to be easily constructed by means of (28) and (29). For
able to discuss the behavior of the spins in time- the average spins we find the sequence of equations
dependent magnetic fields. Since the arguments of (d/dat)qk(t) = - qk(t) + {3
statistical mechanics used in treating the Ising model
deal only with constant magnetic fields, we are free + h[qk-I(t) + qk+I(t)]
in defining the stochastic model to choose any time-
dependence of the parameter {3 which yields (78)
- !(3'Yh-I,k(t) + rk,k+I(t)], (85)
when H is constant. The simplest way of defining which differs from the sequence (30) considered
a time-dependent (3 is to retain the relation (78) earlier by the inclusion of the inhomogeneous term
when H depends on time. The solution for the {3 and, more importantly, through the inclusion of
average spins is then the pair-correlation terms rk-l,k and rk,HI' The
equations for the pair correlation are likewise found
q(t) = q(to)e-a(t-t.) + 1t e- a(t-t'){3(t')a dt', (80) to contain terms proportional to other correlation
t.
functions, i.e., the single-spin expectations and the
where to is a time at which q is known initially. expectation of the product of three spins. Such
equations appear, because of their mixed structure,
SPIN SYSTEM IN A MAGNETIC FIELD to be essentially more difficult to solve than those
treated earlier. It is not difficult, however, to solve
To construct a stochastic analog of the Ising model them in the limit of weak magnetic fields, p.H « kT,
in a magnetic field, we must first find an appropriate and by doing so we are able to discuss the time-
set of tra.nsition probabilities. To this end we note dependent magnetic susceptibility of the system.
that the Hamiltonian of the Ising model is In the weak-field limit, the parameter {3 is pro-
portional to the magnetic field, {3 = p.H /kT. The
3C = -p.H L u'" - J L U",U",+I, (81)
first-order changes of the averages qk(t) may be
'" ...
found from Eqs. (85) by using as a zeroth approxi-
so that, if the spins other than U; are considered as
mation for the functions rk-I,k and rk,k+l the solution
fixed, the ratio of equilibrium probabilities for the (68) derived for them in our earlier work. The equa-
states -U; and U; is tions for the qk(t) become in this wayan inhomo-
p;( -u;) _ exp {-(I/kT)u;[J(u;_1 + U;+I) + p.H]} geneous sequence, with the inhomogeneous terms
p;(u;) - exp {(I/kT)uj[J(u;_1 + U;+I) + p.Hll proportioned to H. The solution of these equations
is simplified considerably if we assume that .the
_ w;(u;) exp [-(p.H/kT)u;]
(82) model is in thermal equilibrium to zeroth order
- w;( -u;) exp [(p.H/kT)u;] ,
in H, i.e., that the field induces only small departures
where the identities (12) and (13) were used in from equilibrium. In that case we have
securing the latter relation. If we write the transi- (86)
tion probabilities for the model in a magnetic field
as w~(u;), the detailed balancing condition at equilib- which is independent of k, and Eqs. (85) reduce to
rium requires the sequence
d
w~(Uj) _ pj( -Uj) dat qk = - qk + h(qk-l + qk+l) + {3(1 - '1'71). (87)
w~( -Uj) - p;(u;)
w j (u;)[1 - U; tanh (p.H/kT)] We shall assume, as before, that the definition of
(83)
= w;( -u;)[1 + Uj tanh (p.H/kT)] {3 holds for time-dependent magnetic fields as well
as stationary ones. The inhomogeneous term in (87)
Hence our model will approach the same equilibrium may also be written, by using Eq. (55), as

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
304 ROY J. GLAUBER

(88)
IN 1 11 IN 2J + (95)
x(O) = kT 1 - 11 = kT exp kT '
The sequence of Eqs. (87) differs from the sequence which is the familiar reaplt furnished by the Ising
(52), which we solved earlier, only by the inclusion model.
of this inhomogeneous term. Since the term is inde- FLUCTUATION-DISSIPATION THEOREMS
pendent of k, the particular solution required may
It is interesting to note that our result (94) for the
be chosen independent of k as well. Finding the
magnetic susceptibility is closely related to the result
particular solution is then a matter of treating the
(76) for the time-dependent correlation function. If
simplest of first-order linear differential equations.
we sum the correlation functions (76) over the indices
The general solution to the sequence (87) for an
j and k by means of the generating function (35),
infinite chain is
and multiply by l, we find the time-dependent
qk(t) = e-a(o-t o ) L: ql(to)Ik-lha(t -
I
to)] correlation function for the magnetization,

+ :T ~ +- 11:
7]
I'
to
e-a(l-OY)('-") H(t')a dt', (89)
(M(O)M(t'»T = IN 11 + 11
- 11
e-a(l-oy)IC'I. (96)

The Fourier transform of this function is


where again we have let to be the initial time. Since
the model is assumed to be in thermal equilibrium
before the magnetic field is turned on at time to,
L: (M(O)M(t'»re''''c' dt'
the initial values of the ql may be taken to vanish.
2N 1+ 11 2a(1 - 'Y)
The spin expectations therefore all have the value = p. 1 - 11 a 2(1 - 'Y)2 + w2
given by the integral term of (89).
We now introduce the stochastic magnetization 2kT
= -:;- 1m x(w) , (97)
function

(90) i.e., the imaginary, or dissipative part of the mag-


netic susceptibility is proportional to the Fourier
transform of the time-dependent magnetization cor-
whose average value is given by the sum
relation function. We thus have in hand a par-
(91) ticularly simple example of a fluctuation-dissipation
relation. Although the derivation we have given
If we let the initial time recede into the past, depends on the explicit evaluation of the functions
to -+ - co, the average magnetization obtained by involved, analogous relations are known to hold for
summing (89) becomes a wide class of mechanical systems. These relations
are derived from statistical mechanics by discussing
2N 1 2 ,
the way in which perturbations of the Liouville
(M(t» = ~T 1 ~ ~2 La> e-a(l-OY) (1-1 ') H(t')a dt'. (92)
equation affect the distribution function or density
matrix and the expectation values derived from
For the case of a magnetic field which varies har- them. Since the model we are discussing, on the
monically, H(t) = Hoe-''''', we may define a com- other hand, is a stochastic one, our equations do
plex, frequency-dependent magnetic susceptibility not follow the dynamics of the spin variables in
x(w) via the relation detail. In place of the quantum-mechanical Liouville
(MCt» = x(w)Hoe- i "". (93) equation we have the master equation, which has
altogether different properties. Our model, neverthe-
The susceptibility is then given by less, does permit the statement of a number of
simple identities analogous to the fluctuation-dis-
I/N 1 - 112 a sipation theorems of statistical mechanics, but dif-
X(w) = kT 1 + 112 a(1 - 'Y) - iw
fering from them slightly in form. Since these
relations may be of use in finding the effect of a
N 1 + 11
2
p. a(l - 'Y) (94)
= kT 1 - 11 a(1 - 'Y) - iw weak field upon the average values of quite general
functions of the spin variables, we shall derive them
In particular, in the low-frequency limit w -+ 0, we here.
find the static susceptibility We denote the change of any quantity A induced

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 305

by the presence of the weak magnetic field by the u variables, F(Ul, ... UN), we have only to mUltiply
increment symbol ~A. The change of the transition Eq. (102) through by F and sum over spins Ui'
probabilities according to (84) is then The integrand on the right-hand side may then
be recognized as an equilibrium-state average of a
~Wi(Ui) = W~(Ui) - Wi(U;) product of three stochastic functions. Expressed in
= -(p.HjkT)u;Wi(Ui)' (98) this way, the change of the average value of F
becomes
The first-order changes of the quantities involved
in the master equation (27) are related by ~(F[Ul(t), ... UN(t)))

:1 l1p(u:, ... u!., I) = ~; f", ~ (uz(t')w z[uZ(t)]F[Ul(t), ... UN(t)])r

X H(t') dt'. (103)


In particular, when the transition probabilities are
(99) given by (9) we find more simply
Now if p(u:, ... uk lUI, ... uNt) is a conditioned ~(F[Ul(t), ... UN(t)])
probability function in the sense described earlier,
i.e., it satisfies the unperturbed master equation
and reduces to IIi ~~I~I' for t = 0, then it con-
= tTaC1 - 'Y) f", ~ (uz(t')F[Ul(t), ... UN(t)]),.

stitutes a Green's function for the sequence of Eqs. X H(t') dt'. (104)
(99). If the initial time is - co, the solution to (99)
may be written as If the function F is taken to be the magnetization,
we find that it obeys the relation
l1P(Ul' ... UN, t) = - L L ur L ur'
{eT'l l tTl" ~(M(t» = (M(t»

= k~ a(l - 'Y) f", (M(t')M(t»TH(t') dt'. (105)

x p(u:' ... uk I Ul, ... UN, t - t') dt'. (100) Since the equilibrium state is stationary, the thermal
average in the integrand can only depend on t - t'.
We next substitute the expression (98) for the
Hence for the case of a harmonic field H(t)
increment of the transition probabilities into (100)
Hoe- i .", we find
and sum explicitly over the values of u", finding

l1P(Ul' '" UN, t) = J!.... L L ur x(w) = k~ a(l - 'Y) lo'" (M(O)M(t»~i'" dt. (106)
kT I~'I z

X f", H(/') {Wz(ur, t')p(u:, ..•


The foregoing relations are rather similar in
ur, ... u!., I')
structure to the complex forms of the fiuctuation-
dissipation theorems of statistical mechanics, and
+ Wz(-ur, t')p(uL ... -ur, ... UN, t')} furnish us with similar information. They differ
from those relations, however, in two respects
X p(uL ... u!. I Ul, '" UN, t - t') dl'. (101)
illustrated by comparing (97) and (106). The former
The detailed balancing relation (83) assures us equation relates the imaginary part of the suscepti-
that the two products within the curly brackets of bility to the transform of the correlation function;
(101) are equal, i.e. that the probability increment while the latter relates the real part to it with a
may be simplified to the form different proportionality constant. Although both
types of relation hold true for the model at hand,
l1P(Ul' '" UN, t) it is interesting to see how the difference between
them arises. For this purpose let us consider the
2
= k TP. L L
I (cr' J
f'
H(/')p(uf,··· u/r, 1')urwz(U', I') stochastic function
-co

X p(U:, ... u!. lUI, ... UN, t - t') dt'. (102) L(t) = -2p. L".
u",(t)w... (u"., t). (107)

To evaluate the change induced by the magnetic The expectation value of L is the time derivative
field in the expectation value of any function of the of the average magnetization. To see this, we use

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
306 ROY J. GLAUBER

(28) to write to which they correspond are fairly simple, however.


The eigenvectors which are formed from the products
d d of rth degree polynomials with p(O) have eigen-
{L(t» = JJ. ~ dt qm(t) = dt {M(t». (108)
values
The function L(t) itself, however, is not the time 11 = I'm, + I'm, + .,. + I'm., (113)
derivative of M(t). If it were, the substitution of
M(t) for it in (103) would lead to precisely the where the I'mj are given by (47), and the set of inte-
relations furnished by discussions based on the gers m I , • • • m is selected from 0, 1 .,. N - 1
T

Liouville equation. The relations we find instead with no repetitions. The number of such eigenvalues
are evidently quite similar in content.
is given by the binomial coefficient (~) . The full
ALTERNATIVE METHOD AND GENERALIZATION set of 2N eigenvalues is obtained by allowing r to
range from 0 to N.
It may be of interest to mention briefly another
In particular, the largest eigenvalue is obtained
way of studying Markoff processes, one rather dif-
for r = N and is v = N. The eigenvector for this
ferent from the preceding discussion. The 2N values
case is simply proportional to
of the probability function P«(J'I, ••• (J'N, t) may be
regarded as the components of a vector p. Then by N

suitably defining the elements of a matrix M, we p(N)«(J'I, ••• (J'N, t) = II (J'je- Na ,. (114)
i-I
may write the master equation (27) in the form
All of the foregoing discussion has been restricted
(djdt)p = Mp, (109) to the case of nearest-neighbor coupling among spins
in order to make contact with the familiar studies
which suggests that p is a superposition of eigen-
vectors p (a) which satisfy of the Ising model. The coupling may be extended
to include the first n nearest neighbors by intro-
(110) ducing the transition probability

One eigenvector, at least, is quite well-known to us.


The probability distribution for the Ising model at
wj«(J';) = ! J1 -
~
!(J'j t
1-1
'YI«(J'j-l + (J'j+I)}, (115)
equilibrium, the normalized Maxwell-Boltzmann dis-
tribution, corresponds to the eigenvalue v = O. It is where LI 11'11 ::; 1. The methods of the preceding
sections deal equally with the equations which follow
P(O)«(J'I, ••• (J'N) = Z-l exp [(J jkT) E (J'I(J'I+I], (111) from this more general type of coupling. The only
I
significant change is that the quadratic equation,
where Z, the normalizing factor, is the partition (55), for the short-range order is replaced by an
function. equation of 2nth degree which has n roots 711, ••• 71"
Other eigenvectors may be sought by multi- with absolute value less than unity. The equilibrium
plying p (0) by sums of products of spin variables solution for the average spins, when the zeroth spin
with undetermined coefficients. For example, if we is fixed, is then an expression of the form
write ..
qk = E C;71~,
i-I
(116)

where the coefficients Cj must be determined from the


we find that the condition that this form satisfy
condition qo = 1 and the equations for qI, '" qn-l.
(109) is that the functions ai(t) satisfy the same
These spin averages then determine the equilibrium
sequence of equations (30) as we discussed earlier
spin correlations rj.k in precisely the way described
in connection with qi(t). The mode functions r::, earlier.
where rm is given by (46) therefore furnish us with
N different eigenvectors corresponding to roots I'm ACKNOWLEDGMENT
given by (47).
The eigenvectors which are constructed by multi- The author would like to thank the Bell Telephone
plying p (0) by higher-order polynomials in (J'l, ••• (J'N, Laboratories, Murray Hill, New Jersey, for their
are somewhat more complicated in form, and will hospitality during a period of several summer weeks
be discussed in a later publication. The eigenvalues in which this work was carried out.

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
TIME-DEPENDENT STATISTICS OF THE ISING MODEL 307
APPENDIX general solution is given by the form
We have already noted that other forms of the Wj(CTj) = !a {I + ~CTj-1CT;+1
transition probability than (9) are capable of bring-
ing the stochastic model to the same equilibrium
- h(1 + ~)CTj(CTj_l + CTj+1)}, (117)
state as the Ising model. The condition that such a in the absence of any magnetic field. In this form
transition probability Wj(CTj) must satisfy is that the the parameter 'Y must still be identified with the
ratio Wj(CT;)/Wj( -CTj) be equal to the equilibrium constant (17), but the parameter ~ has no analog
probability ratio (12). If we assume that Wj(CTj) in the discussions of the Ising model at equilibrium,
depends symmetrically on the two neighboring spins and may evidently be chosen arbitrarily. It was
CTj_l and CTj+1 as well as on CTj, then the condition assumed to vanish in our discussions of the time-
just mentioned may be regarded as a functional dependent model since its presence materially com-
equation for the transition probability. Its most plicates the equations for the spin expectation values.

Downloaded 28 Jul 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions

You might also like