The XY Model in One Dimension

Max A. Metlitski

I.

INTRODUCTION

One of the oldest problems of quantum mechanics is the one dimensional spin-1/2 antiferromagnetic Heisenberg
P ~ ~
chain with the Hamiltonian H = J i S
i Si+1 . Although the spectrum of this Hamiltonian was found exactly by H.
Bethe[1] in 1931, the solution is quite involved and does not shed much light on even such basic properties as long-range
order. A much more natural approach to the problem of interacting spin-1/2s, stems from the similarity between
spin-1/2 raising/lowering operators and fermion creation/annihilation operators. This correspondence, originally
made precise in 1928 by Jordan and Wigner[2], can be used to convert spin-1/2 systems into problems of interacting
spinless fermions. In this paper, I review the application of Jordan-Wigner transformations to a modification of the
Heisenberg chain, known as the XY model, which was originally studied in great detail in [3]. The XY model, which
can be obtained from the Heisenberg Hamiltonian by turning off the coupling between z spin components, reduces to
a free theory of spinless fermions under the Jordan-Wigner transformations. Luckily, it turns out that the low-energy
properties of the full anti-ferromagnetic Heisenberg chain, such as the presence of gapless excitations and absence of
long range order are very similar to those of the XY model (see [4] and references therein).
The first part of this paper discusses the introduction of Jordan-Wigner transformations and the spectrum of the
XY model. The second part of the paper is devoted to more advanced aspects of the XY model, such as the effects
of anisotropy and study of short, intermediate and long range order.
II.

JORDAN-WIGNER TRANSFORMATION OF THE XY MODEL

Consider the following Hamiltonian describing a chain of N spin-1/2s interacting antiferromagnetically with their
nearest neighbors:
X
H=
(S x i S x i+1 + S y i S y i+1 )
(1)
i

Here S a i are spin-1/2 operators, obeying the usual commutation relations [S a i , S b j ] = iabc ij S c i . The scale of the
coupling as well as ~ have been set to 1. We assume cyclic boundary conditions, i.e. the index i in the sum (1) runs
~N +1 = S
~1 .
over 1 . . . N with S
We can obtain a ferromagnetic counterpart of (1), by defining HF = H. Thus, once we have solved the antiferromagnetic problem exactly, we can immediately read out the solution to the ferromagnetic problem.
As usual it is convenient to introduce raising and lowering operators: ai = S x i + iS y i , ai = S x i iS y i . These
obey the algebra:
{ai , ai } = 1, {ai , ai } = {ai , ai } = 0,
[ai , aj ] = [ai , aj ] = [ai , aj ] = 0, i 6= j

(2)
(3)

The Hamiltonian (1) then takes the form:

H=

1X
(ai ai+1 + ai+1 ai )
2 i

(4)

Hence the Hamiltonian is quadratic in as. If all as obeyed canonical anticommutation relations we would be done it would be possible to diagonalize H by a linear transformation of the as! Unfortunately, as belonging to the same
site obey anticommutaion relations, while as on different sites obey commutation relations, so that unitary rotations
of the as obey neither bosonic nor fermionic algebra. The key to the solution of this problem is the Jordan-Wigner
transformation. Let,
ci = exp (i

i1
X
j=1

aj aj )ai , ci = ai exp (i

i1
X
j=1

aj aj )

(5)

2
The highly non-linear transformation (5) can be easily inverted. Observe,
ci ci = ai ai

(6)

and thus,
ai = exp (i

i1
X

cj cj )ci , ai = ci exp (i

j=1

i1
X

cj cj )

(7)

j=1

The operators ci , ci obey the canonical fermion algebra:

{ci , cj } = ij , {ci , cj } = 0, {ci , cj } = 0

(8)

We will demonstrate explicitly, {ci , cj } = ij (the other anti-commutators in (8) can be computed in a similar
fashion). Observe, [ai ai , aj aj ] = 0, (ai ai )2 = ai ai , [ai ai , aj ] = ij aj , [ai ai , aj ] = ij aj since as on different
sites commute and as on the same site behave like fermions. Therefore,
exp (i

m
X

m
Y

aj aj ) =

j=n

exp (iaj aj )

(9)

j=n

where

X
X
1
1
l
l
(i)l (ai ai ) = 1 +
(i) ai ai = 1 + (ei 1)ai ai = 1 2ai ai
l!
l!

(10)

{ai , 1 2ai ai } = ai (1 2ai ai ) + (1 2ai ai )ai = 2ai 2ai ai ai = 2ai 2(1 ai ai )ai = 0,

(11)

exp (iai ai ) =

l=0

l=1

Also, observe that

{ai , 1 2ai ai } = {ai , 1 2ai ai } = 0

(12)

which implies
[exp (i
{exp (i

m
X

aj aj ), ai ] = [exp (i

m
X

aj aj ), ai ] = 0, i 6 [n, m]

(13)

aj aj ), ai } = 0, i [n, m]

(14)

j=n
m
X

j=n
m
X

j=n

j=n

aj aj ), ai } = {exp (i

Now, we can compute the anti-commutator:

{ci , ci } = ai ai + ai ai = 1

(15)

and for j > i,

{ci , cj } = ai exp(i

i1
X

ak ak ) exp(i

= ai exp(i

ak ak )aj + aj exp(i

ak ak )aj + aj exp(i

k=i

= [ai , aj ] exp(i

j1
X

ak ak ) exp(i

i1
X

ak ak )ai

(16)

k=1

ak ak )ai = (ai aj aj ai ) exp(i

k=i
j1
X

j1
X
k=1

k=1

k=1
j1
X

j1
X

j1
X

ak ak )

(17)

k=i

ak ak ) = 0

(18)

k=i

where we have used eqs. (13),(14). Similarly, for j < i,

{ci , cj } = {cj , ci } = 0

(19)

Lets rewrite the Hamiltonian in terms of cs using eq. (7). Since cs satisfy canonical anticommutation relations,
[ci ci , cj cj ] = 0, (ci ci )2 = ci ci , [ci ci , cj ] = ij cj , [ci ci , cj ] = ij cj , we have in complete analogy to eqs. (9),

exp (i

m
X

cj cj ) =

j=n

m
Y

exp (icj cj ) =

j=n

m
Y

(1 2cj cj )

(20)

j=n

{1 2ci ci , ci } = {1 2ci ci , ci } = 0

[exp (i
{exp (i

m
X

cj cj ), ci ] = [exp (i

m
X

cj cj ), ci ] = 0, i 6 [n, m]

(22)

cj cj ), ci } = 0, i [n, m]

(23)

j=n
m
X

j=n
m
X

j=n

j=n

cj cj ), ci } = {exp(i

(21)

Now, we can compute the various terms in the Hamiltonian (4). For 1 i N 1,
ai a i+1 = exp(i

i1
X

cj cj )ci c i+1 exp(i

j=1

i
X

cj cj ) = ci exp(i

j=1

i1
X

cj cj ) exp(i

j=1

i
X

cj cj )c i+1

= ci exp(ici ci )c i+1 = ci (1 2ci ci )c i+1 = (1 2ci ci )ci c i+1 = ci c i+1 = c i+1 ci

ai+1 ai

= (ai a

i+1 )

= ci ci+1

(24)

j=1

(25)
(26)

After similarly expressing the special cyclic boundary term aN a1 + a1 aN in terms of cs, the Hamiltonian becomes,
H = Hc + Hb ,
1X
Hc =
(c i+1 ci + ci ci+1 )
2 i
N
X
1
Hb = (c1 cN + cN c1 )(exp (i
cj cj ) + 1)
2
j=1

(27)
(28)
(29)

The Hamiltonian Hc is quadratic in anticommuting operators c, and describes a free spinless fermion on a cyclic
chain with nearest neighbor hopping. Hence Hc can be trivially diagonalized by solving the one particle Schrodinger
equation, and making a unitary transformation of the cs. The effect of the boundary term Hb can be actually taken
into account exactly (see section III A), but we choose to neglect it for now as it gives an O(1/N ) contribution to
macroscopic physical quantities.
Observe, that in the original formulation the Hamiltonian conserves total spin along the z axis: [H, S z ] = 0.
AfterP
the Jordan-Wigner transformation, this symmetry is manifested as conservation of the total fermion number
N = i ci ci , [Hc , N ] = 0. Indeed,
1
1
1
S z i = [ai , ai ] = ai ai = ci ci
2
2
2
X
X
1
N
Si z =
(ci ci ) = N
Sz =
2
2
i
i

(30)
(31)

Thus, each spinless fermion created by c carries S z = 1.

P ~ ~
It is instructive to see how the true 1D Heisenberg Hamiltonian Hh = i S
i Si+1 transforms under the JordanWigner transformation (5). Observe,
Hh = H +

Si z Si+1 z = H +

X
i

1
1
(ci ci )(c i+1 ci+1 )
2
2

(32)

Thus, unlike the XY model, the Heisenberg Hamiltonian is no longer a free theory, but a theory of spinless fermions
on a lattice with nearest neighbor hopping and nearest neighbor interactions.
Lets proceed with diagonalization of Hc . We rewrite,
Hc =

X
ij

ci Aij cj , Aij =

1
(i,j+1 + j,i+1 )
2

(33)

4
Letting {k } be a complete and orthonormal set of eigenvectors of A with eigenvalues {k }, define:
X
X
k =
ki ci , k =
ki ci
i

(34)

which imply canonical anticommutation relations of s, and conversely:

X
X
ci =
ki k , ci =
ki k
k

(35)

Expressing Hc in terms of s:
Hc =

k k k

(36)

Some of the eigenvalues of A will be negative, so it is convenient to make an additional transformation:

k = k , k < 0

k = k , k 0;

(37)

leading to
Hc =

k k k +

k,k 0

k k k =

k,k <0

where weve used, tr(A) =

satisfies,

P
k

|k |k k

|k | =

k,k <0

X
k

1
|k |(k k )
2

(38)

k = 0. The s are again canonical fermi operators. Hence the ground state |i
k |i = 0, k

(39)

and operators k generate elementary fermionic excitations with energy |k | above the ground state.
Utilizing translational invariance, we find the eigenvectors and eigenvalues of A to be:
1
kj = eikj , k = cos(k);
N

k=

2n
, N/2 n N/2 1
N

(40)

So, in thermodynamic limit N there are always gapless excitations near k = 2 (see Fig. 1 for a plot of the
dispersion relation). These have a dispersion, k = /2 + q, (k) = | sin(q)| |q| as q 0. The ground state energy
per spin becomes:
Z
Z /2
U
1 X1
1
dk 1
1
=
|k |
| cos(k)| =
4
(41)
cos(k)dk =
N
N
2
2
2
4

0
k

Lets investigate the spin properties. For simplicity we assume that N is even and not divisible by 4, so that the
ground state is non-degenerate1 (at any rate, the ground state of Hc is at most 4 times degenerate). Notice,
Sz =

ci ci

k k +

k,k 0

Hence, excitations with |k| <

ground state is,

X
X
X
N
N
N
=
k k
=
k k +
k k
2
2
2

k,k 0

(1 k k )

carry S z = +1, while excitations with |k| >

S z |i = (

X
k,k <0

X
X
N
N
=
sgn(k )k k +
1
2
2
k

k,k <0

(42)

k,k <0

N
)|i = 0
2

(43)

k,k <0

carry S z = 1. The total spin of the

(44)

Another reason for considering the case N - even, not divisible by 4, is that in this case the ground state of Hc is actually an exact
eigenstate of H with the same eigenvalue.

5
So the ground state is non-degenerate and carries S z = 0 - this is the same result as in the full Heisenberg model.
Our above results can be trivially generalized to the case of ferromagnetic coupling. In that case,
HF = H Hc =

X
k

X
1
1
|k |(k k ) =
|k |(k k )
2
2

(45)

So the new ground state satisfies, k |F i = 0, k. Hence |F i can be obtained from the antiferromagnetic ground
state by turning on all the excitations, and the ferromagnetic excitations are obtained by removing the antiferromagnetic ones. In particular, the ground state energy and the excitation spectrum are the same as in the antiferromagnetic
case. Hence, the ground state is still non-degenerate and
S z |F i = (

X
k

sgn(k )k k +

k,k <0

X
X
X
N
N
N
)|F i = (
sgn(k )+
1 )|F i = (
1 )|F i = 0 (46)
2
2
2
k

k,k <0

k,k 0

This is in clear contrast to the full ferromagnetic Heisenberg model, where the ground state is greatly degenerate, and
one of the ground states carries S z = N/2.
III.

PROPERTIES OF THE XY MODEL

A.

Anisotropy Effects

It is first interesting to generalize the discussion of section II to the case when the interaction in the XY plane is
not isotropic, i.e
X
H =
((1 + )Si x Si+1 x + (1 )Si y Si+1 y )
(47)
i

where 1 1 is the anisotropy parameter. In section II, we considered the fully isotropic case = 0, which
has an additional symmetry [H, S z ] = 0. The opposite limit = 1 corresponds to the classical Ising model, in which
the ground state is Neel ordered, i.e Six |i = (1)i |i i. The methods developed in section II allow us to explicitly
study how the system properties change as we go from the totally ordered state at = 1 to the isotropic limit = 0.
First, we rewrite the Hamiltonian H in terms of operators a, a .
H =

1X
((ai a i+1 + ai+1 ai ) + (ai ai+1 + a i+1 ai ))
2 i

(48)

Now we apply the Jordan-Wigner transformation (7), to express H in terms of cs:

H = Hc + Hb
1X
((c i+1 ci + ci ci+1 ) + (ci c i+1 + ci+1 ci ))
Hc =
2 i
N
X
1
Hb = (c1 cN + cN c1 + (cN c1 + c1 cN ))(P + 1), P = exp(i
cj cj ) = exp(iN )
2
j=1

(49)
(50)
(51)

We again for now neglect the boundary term Hb as it has little effect on macroscopic physical quantities (we indicate
how to take Hb exactly into account at the end of this section). The Hamiltonian Hc is still quadratic in cs, but now
for 6= 0, [Hc , N ] 6= 0 because of the new c c, c c terms in the Hamiltonian. Recalling the correspondence, between
N and S z , the non-conservation of fermion number N simply reflects the non-conservation of S z .
Hamiltonian Hc can be diagonalized with a linear transformation of the canonical fermion operators c, c . Indeed,
write
X
1
(52)
Hc =
(ci Aij cj + (ci Bij cj ci Bij cj ))
2
i,j
Aij =

1
(i,j1 + i,j+1 ), Bij = (i,j1 i,j+1 )
2
2

(53)

6
where A is a real symmetric matrix and B is a real antisymmetric matrix. Defining,
X
k =
(gki ci + hki ci ),

(54)

k =

X
(gki ci + hki ci )

(55)

we wish to find real constants gki , hki , s.t. the s obey canonical anti-commutation relations and,
X
Hc =
k k k + const

(56)

Using the equation of motion technique, [H, k ] = k k , we obtain:

Agk + Bhk = k gk
Bgk Ahk = k hk

(57)
(58)

Letting,
ki = gki + hki , ki = gki hki

(59)

(A + B)k = k k ,
(A B)k = k k

(60)
(61)

(A B)(A + B)k = 2k k ,
(A + B)(A B)k = 2k k

(62)
(63)

which implies,

Assuming det(A B) 6= 02 , all eigenvalues 2k of the real symmetric matrix R = (A B)(A + B) are strictly positive,
and hence N real orthonormal eigenvectors k of R, generate N real solutions of eqs. (57), (58) with k > 0, if we
let k = k 1 (A + B)k .
After transformations (54), (55), the Hamiltonian becomes,
Hc =

X
k

1
k (k k )
2

(64)

The ground state is again defined as k |i = 0, k, and the spectrum is that of free fermions with excitation energies
k .
2
1+ 2
It remains to diagonalize the matrix ((A B)(A + B))ij = 1
4 (i+1,j1 + i1,j+1 ) + 2 ij . Due to translational
invariance, we find3 ,
r
kj =

2
sin(kj), k > 0,
N

k = (1 (1 2 ) sin2 k)1/2
r
2
2n
N
N
kj =
cos(kj), k 0, k =
, n
1
N
N
2
2

(65)
(66)

The corresponding k can be calculated from eq. (60):

1
1
(A + B)jn kn =
((k,j+1 + k,j1 ) + (k,j+1 k,j1 ))
k
2k
1
=
(cos(k)kj + sin(k)k,j )
k

kj =

2
3

(67)
(68)

if det(A B) = 0 we can still represent the Hamiltonian in the form

q (56)
1
Actually, k for k = 0, and k = have a normalization factor
. The effect of these two modes, is however, negligible in the limit
N
N

7
We observe that for any asymmetry 6= 0, the excitation spectrum develops a gap = ( 2 ) = || (see Fig. (1)
for a plot of the dispersion relation). The ground state energy per spin becomes,
Z
U
1 X1
1 /2
1 p
(69)
=
k
dk(1 (1 2 ) sin2 (k))1/2 = E( 1 2 )
N
N
2
0

where E is an elliptic E function.

Finally, lets discuss how to take the boundary term Hb into account. We observe, that even though for arbitrary
, [N , Hc ] 6= 0, we have [P, Hc ] = [P, H] = 0, where P = exp (iN ). This corresponds to the symmetry with respect
to rotations around z axis by . Thus, we can diagonalize Hc and P simultaneously. Notice, that P has eigenvalues
1, and {P, k } = 0. So if the ground state |i of Hc with energy E0 carries P = 1, then Hb |i = 0 and hence
|i is an eigenstate of full H with the same energy E0 . In addition, all eigenstates of Hc with an even number of
excitations also have P = 1 and hence are exact eigenstates of H. As shown in section II, for = 0, N even, and
not divisible by 4, the ground state of Hc carries S z = 0, and hence N = N2 is odd, and P = 1. By continuity, this
is true for all . Therefore, in this case, the ground state of Hc is guaranteed to be a true eigenstate of H. Notice,
that these complications could be avoided if we considered a chain with free ends. In this case, no extra boundary
term arises in transformation from as to cs. Yet, this case has no translational invariance, making k s somewhat
ugly.
B.

Magnetic Order

Lets investigate the magnetic order properties of the XY model. We know that for = 1 the ground state
has long range Neel order, i.e. Si x |i = (1)i |i (and in the ferromagnetic case, Si x |i = |i). It is interesting
to understand what happens as decreases from 1 to 0. It is clear that states with spin-x components ordered and
spin-y components ordered start to compete, yet it is not obvious, whether the ground state has any long and/or
short range order at = 0.
It is best to define the magnetic order in terms of correlators
X
X
~i S
~j i =
a ij = hSi a Sj a i, ij =
hS
a ij
(70)
a

where the expectation values are taken in the ground state of H (or in the canonical ensemble if we are working at
finite temperature). This way, our definition is not sensitive to the degeneracy of the ground state, which can affect
the single spin expectations hSi a i. The correlators can be easily expressed in terms of fermionic operators c, c . For
j > i, we obtain,
j1
X
1
1
cn cn )(cj + cj )i
h(ai + ai )(aj + aj )i = h(ci + ci ) exp(i
4
4
n=i

(71)

j1
Y
1
(1 2cn cn )(cj + cj )i
h(ci + ci )
4
n=i

(72)

j1
Y
1
((cn + cn )(cn cn ))(cj + cj )i
h(ci + ci )(1 2ci ci )
4
n=i+1

(73)

j1
Y
1
h(ci ci )
((cn + cn )(cn cn ))(cj + cj )i
4
n=i+1

(74)

xij =

We introduce the operators,

Ai = ci + ci , Bi = ci ci

(75)

{Ai , Aj } = ij , {Bi , Bj } = 2ij , {Ai , Bj } = 0

(76)

These obey the algebra,

In terms of A, B, eq. (71) becomes,

xij =

1
hBi Ai+1 Bi+1 Ai+2 . . . Aj1 Bj1 Aj i
4

(77)

8
Similarly, for y and z correlators,
1
yij = (1)ji hAi Bi+1 Ai+1 Bi+2 . . . Bj1 Aj1 Bj i
4
1
z
ij = hAi Bi Aj Bj i
4

(78)
(79)

Now since As and Bs are anticommuting variables, their expectation values can be evaluated with Wicks theorem,
in terms of sums of all possible contractions of pairs of operators. Observe,
X
X
X
ki kj = ij
(80)
hAi Aj i =
ki k0 j h(k + k )(k0 + k0 )i =
ki k0 j hk k0 i =
k,k0

hBi Bj i =

k,k0

ki k0 j h(k k )(k0 k0 )i =

k,k0

hBi Aj i =

ki kj = ij

(81)

ki kj = ( T )ij = Gij

(82)

ki k0 j h(k k )(k0 + k0 )i =

k,k0

X
k

The matrix G acts as a Greens function in our calculations. Notice that in expressions for correlators (77), (78),
(79), all As live on different sites, and hence contractions of As always give 0 by (80). Similarly, contractions of Bs
are always 0. Hence, only contractions of As with Bs appear when we evaluate our correlators with Wicks theorem.
Summing all these contractions we obtain:
1 X
1
xij =
sgn(p)(Gi,p(i+1) Gi+1,p(i+2) . . . Gj1,p(j) ) = det Gij
(83)
x
4
4
pS[i+1,j]

yij =
zij

1
4

sgn(p)(Gi+1,p(i) Gi+2,p(i+1) . . . Gj,p(j1) ) =

pS[i,j1]

1
det Gij
y
4

1
= (Gii Gjj Gij Gji )
4

(84)
(85)

ij
where S[a, b] is the group of permutations of integers {n : a n b} and matrices Gij
x , Gy are given by
ij
(Gij
x )nm = Gi+n1,i+m , (Gy )nm = Gi+n,i+m1 , 1 n, m j i

(86)

So the various correlators are just subdeterminants of the Greens function G. It now remains to evaluate these
subdeterminants. Before we perform this calculation, lets indicate how to generalize our results to the case of
ferromagnetic coupling, HF = H. In that case, we have as explained in section II, , and hence gki hki ,
which implies ki ki , ki ki . It follows from eq. (82), that G G, and hence by eqs. (83), (84)
ji x,y
x,y
ij , zij zij .
ij (1)
We cant make any further progress without calculating the Greens function G defined in eq. (82). Recall,
X
X 1
Gij =
ki kj =
(cos(k)ki + sin(k)k,i )kj
(87)
k
k
k

Z /2
1 + (1)ij+1
2
1
=

dk (cos(k) cos(k(i j)) sin(k) sin(k(i j)))

(88)
2

k
0
where weve obtained the last line by some trivial algebra, from eqs. (66), (67). Notice that Gij depends only on the
difference i j as should be expected for a cyclic chain due to translational invariance. Hence, we write Gij = Gij .
Also, Gij vanishes when i j is even.
We can evaluate Gr analytically in the isotropic case = 0 and in the Ising limit = 1. For = 0, we have
Z
2 /2
2
Gr =
cos(kr) = (1)(r+1)/2 , r-odd
(89)
0
r
while for = 1,
2
Gr =

/2

dk cos(k(r + 1)) = r,1

0

Now we are ready to investigate order in the ground state of the XY chain.

(90)

9
1.

Short Range Order

Lets find out what are the correlations in the XY model between nearest neighbors at finite anisotropy. This
requires calculating,
1
1
Gi,i+1 = G1
4
4
1
1
= Gi+1,i = G1
4
4
1
1
= (Gii Gi+1,i+1 Gi,i+1 Gi+1,i ) = G1 G1
4
4

x1 = xi,i+1 =

(91)

y1 = yi,i+1

(92)

z1 = zi,i+1

(93)

and involves only the nearest neighbor Greenss functions G1 , which can be calculated in terms of elliptic E and K
functions:
p
p
2 1
G1 =
(E( 1 2 ) K( 1 2 ))
(94)
1
The correlators, x1 , y1 , z1 are plotted in Fig. 2. Observe, that for any all of these are negative, and hence
the nearest neighbors clearly display anti-ferromagnetic correlations. When = 1, we have as expected from Ising
model, x = 41 , y = 0, z = 0. As decreases, we see that anti-ferromagnetic correlations between y components
of neighboring spins start to develop, competing with the correlations of S x . The appearance of correlations between
y components also seems to induce an anti-ferromagnetic correlation between z components. Finally, in the isotropic
case = 0, the correlations between x and y components of neighboring spins become equal (as expected from
1
symmetry reasons), and x1 = y1 = 2
, z1 = 12 . So anti-ferromagnetic short-range order persists in the isotropic
limit = 0. Finally, if the coupling is ferromagnetic, then as indicated above, x,y
x,y
and z1 z1 . Hence,
1
1
for the ferromagnetic XY model, the x and y components of spins display ferromagnetic correlations, while the z
components of spins display anti-ferromagnetic correlations.
2.

Intermediate Range Order

We can numerically evaluate the determinants (83), (84) using the Green function (87). We display the resulting
correlators xr = xi,i+r , yr = yi,i+r , zr = zi,i+r in Figs. 3, 4, 5 for several values of > 0. The correlations of x
r
and y spin components are clearly anti-ferromagnetic as x,y
(1) . The z spin components are also correlated
r
anti-ferromagnetically, but only for sites i, i + r, where r is odd. It seems that for = 0, xr tends to 0 as r ,
and for 6= 0, xr tends to a non-zero constant as r . On the other hand both y and z tend to 0 as r . In
the next two sections we indicate how to prove these claims analytically.
3.

Absence of Long Range Order in Isotropic XY model

It can be shown analytically that in the isotropic ( = 0) XY model limr ar = 0, i.e. there is no long range
order. For z this statement is trivial, as
zr =

1
, r-odd, zr = 0, r-even lim zr = 0
r
2 r2

(95)

However, for x, y spin components there is some work to be done estimating the subdeterminants of matrices Grx,y =
T
Gi,i+r
given in eqs. (83), (84). It is easy to see, using Gr = Gr for = 0, that Gry = (Grx ) , and hence
x,y
xr = yr = 14 det(Grx ) as expected for symmetry reasons. One can use Hadamards Theorem to bound the determinant
of Grx by the product of norms of its rows:
det(Grx )2

r X
r
Y

(Grx )2ij

(96)

i=1 j=1
2
One can further use the orthogonality of matrix G and the explicit form of Gr = (1)(r+1)/2 r
for r-odd, to bound
r
the norm of each row of Gx , obtaining at the end,

x,y
=
r

2
1
det(Grx ) r2/ const 0, as r
4

(97)

10
Finally, let us remark that a similar argument using Hadamards Theorem can be used to show that at any finite
temperature, the correlators ar fall off exponentially as r , and no long range order exists for any anisotropy .
For further details see the original paper [3].
4.

End-to-End Order

Although the subdeterminants det(Gij

x,y ) are difficult to analyze for arbitrary and i, j, they greatly simplify for
i = 1, j = N , i.e. when we are computing the correlations between the first and the last spin in the chain. Indeed,
from eq. (71), we see that
x1N =

X
1
1
h(c1 + c1 )(cN cN ) exp(i
cn cn )i = hA1 BN P i
4
4
n

(98)

where P = exp(iN ). As was noted in section III A, P |i = |i, so we obtain,

1
x1N = GN 1 P
4

(99)

1
y1N = G1N P
4

(100)

and similarly,

So the complicated subdeterminants reduce to just a single Greens function. However, for the cyclic chain this is
not telling us much about the long range order, since sites 1 and N are nearest neighbors. For the chain with free
ends, however, the end-to-end order is a meaningful indication of long range order (although the two need not agree
numerically exactly, due to end effects). Computing the Greens functions for a chain with free ends (by solving eqs.
(60), (61) to find k , k , and then using the definition of Greens function (82)), one obtains [3] in the limit N
y1N = 0, x1N =

, >0
(1 + )2

(101)

Hence, there is no end-to-end order of the y spin components for any > 0, while there is end-to-end order of the x
spin components for any positive anisotropy . Also, the end-to-end order vanishes in the isotropic case = 0. These
results support our numerical computations of section III B.2, and the analytical results of section III B.3.

[1] H. Bethe, Z. Phys., 71, 205 (1931).

[2] P. Jordan and E. Wigner, Z. Physik 47, 631 (1928).
[3] E. Lieb, T. Schultz, D. Mattis, Ann. of Phys., 16, 407 (1961);
E. Lieb, D. Mattis, Mathematical Physics in One Dimension. Academic Press, New York and London, (1966).
[4] A. Tsvelik, Quantum Field Theory in Condensed Matter Physics. Cambridge University Press, Cambridge, (1995).

11

Excitation Spectrum of the XY Model

1

0.8

0.6
E
0.4

0.2

k
FIG. 1: In this figure we show the spectrum of the excitations of the XY model for different values of the anisotropy . The
red curve corresponds to = 0, while the blue curve corresponds to = 0.2. The spectrum is gapless in the isotropic case, and
acquires a gap in the anisotropic case.

12

Short Range Order in the XY Model

0.25

0.2

0.15
-rho
0.1

0.05

0.2

0.4

0.6

0.8

gamma
FIG. 2: In this figure we display the nearest neighbor spin correlations in the XY model as a function of anisotropy . The
red curve corresponds to x1 , the blue curve corresponds to y1 and the green curve to z1 . We clearly see that the competition
between x and y spin correlations as 0.

13

Intermediate Range Sx Order in XY Model

0.25

0.2

0.15
|rho|
0.1

0.05

10

12

r
FIG. 3: In this figure we display correlations between x components of spins (1)r xr as a function of their separation r for
several values of anisotropy . The red curve corresponds to = 0, the blue curve corresponds to = 0.1 and the green curve
to = 1/3. We see that for the isotropic case = 0, the correlator slowly tends to 0, while for non-zero anisotropy , the
correlator tends to a non-zero limit.

14

Intermediate Range Sy Order in XY Model

0.14
0.12
0.1
|rho| 0.08
0.06
0.04
0.02
0

10

12

r
FIG. 4: In this figure we display correlations between y components of spins (1)r yr as a function of their separation r for
several values of anisotropy . The red curve corresponds to = 0, the blue curve corresponds to = 0.1 and the green curve
to = 1/3. We see that for all the correlator tends to 0.

15

Intermediate Range Sz Order in XY Model

0.1

0.08

0.06
|rho|
0.04

0.02

10

12

r
FIG. 5: In this figure we display correlations between z components of spins zr as a function of their separation r for several
values of anisotropy . The red curve corresponds to = 0, the blue curve corresponds to = 0.1 and the green curve to
= 1/3. We see that for all the correlator tends to 0.