T. Michoel and B. Nachtergaele - The Large-Spin Asymptotics of The Ferromagnetic XXZ Chain

Markov Processes Relat.
Fields 11, 237266 (2005)

Markov
/
T
T
&
Processes
and
Related Fields
c Polymat, Moscow 2005
The Large-Spin Asymptotics of the
Ferromagnetic XXZ Chain
T. Michoel
1
and B. Nachtergaele
2
1
Instituut voor Theoretische Fysica, Katholieke Universiteit Leuven, Celestijnenlaan 200 D,
B-3001 Leuven, Belgium. E-mail: tomm@itf.fys.kuleuven.ac.be
2
Department of Mathematics, University of California, Davis, One Shields Avenue, Davis
95616-8366, USA. E-mail: bxn@math.ucdavis.edu
Abstract. We present new results and give a concise review of recent previous
results on the asymptotics for large spin of the low-lying spectrum of the fer-
romagnetic XXZ Heisenberg chain with kink boundary conditions. Our main
interest is to gain detailed information on the interface ground states of this
model and the low-lying excitations above them. The new and most detailed
results are obtained using a rigorous version of bosonization, which can be in-
terpreted as a quantum central limit theorem.
Keywords: XXZ chain, Heisenberg ferromagnet, large-spin limit, bosonization
AMS Subject Classification: 82B10, 82B24, 82D40
1. Introduction
In recent years the XXZ model has become a popular model to study prop-
erties of interface states in quantum lattice models. As an interpolation between
the Ising model and the isotropic (XXX) Heisenberg ferromagnet, the ferromag-
netic XXZ model has the interesting features of both. By considering the Ising
model and the XXX model as limiting cases of the XXZ model, intuition about
these two limits can be used to better understand the XXZ model. In this pa-
per we are interested in the large-spin asymptotics of the low-lying excitation
spectrum of the XXZ chain, in particular the excitations above the kink (or
interface) ground states of the model. In a nutshell, our main result is that the
spin-wave approximation, in the sense of Dyson [6], becomes exact in the limit
This material is based on work supported by the National Science Foundation under Grant
No. DMS0303316. T. Michoel is a Postdoctoral Fellow of the Fund for Scientic Research
Flanders (Belgium) (F.W.O.Vlaanderen).
238 T. Michoel and B. Nachtergaele
of innitely large spin. The technical statements are given in Section 2. First,
we introduce the model and the main notations and give a quick summary of
the relevant previous results.
For J = 1/2, 1, 3/2, . . ., the spin-J XXZ Hamiltonian on an interval =
[a, b] Z, with kink boundary conditions, is given by
H
J,
=
b1
x=a
H
J
x,x+1
, (1.1)
H
J
x,x+1
= J
2
(S
1
x
S
1
x+1
+S
2
x
S
2
x+1
) S
3
x
S
3
x+1
+J
_
1
2
(S
3
x
S
3
x+1
)
with S
i
x
the spin-J matrices acting on site x:
[S
i
x
, S
j
y
] = i
x,y
ijk
S
k
x
, S
x
S
x
= (S
1
x
)
2
+ (S
2
x
)
2
+ (S
3
x
)
2
= J(J + 1).
We will also use the spin raising and lowering operators: S
+
x
and S
x
, S
x
=
S
1
x
iS
2
x
.
We begin with a brief overview of the main results obtained for the Hamil-
tonians H
J,
. The spin 1/2 model, J = 1/2, is Bethe Ansatz solvable and
possesses a quantum group symmetry [19]. Consequently, there are a number of
results specic to the spin 1/2 case (e.g., see [3, 11, 18]). Since the main focus in
this paper is on large-J behavior, we will not discuss these specic results here.
The Hamiltonian (1.1) is symmetric under global rotations around the 3-axis
generated by S
3
tot,
=

x
S
3
x
, which represents the third component of the
total magnetization. Hence, H
J,
is block diagonal, and it is known that in each
sector corresponding to a given eigenvalue of S
3
tot,
there is exactly one ground
state, i.e., in each sector 0 is a simple eigenvalue [12]:
H
J,
(M)
= 0, S
3
tot,
(M)
= M
(M)
,
where M = [[J, [[J + 1, . . . , [[J. The (unnormalized) eigenvector
(M)
is given by
(M)
{mx}
P
x
mx=M
x
_
2J
J m
x
_
1/2
q
x(Jmx)
[m
x
),
where we introduced the parameter q, 0 < q < 1, by the equation 2 = q +q
1
.
These ground states have a magnetization prole that shows an interface, or
kink, with a location depending on the value of M. For a short review on the
properties of these ground states see [17].
In many instances it is important to consider the thermodynamic limit, i.e,
the limit of innitely long chains. To this end, we consider a strictly increasing
Large-spin ferromagnetic XXZ chain 239
sequence of numbers a
n
Z
+
and lim
n
a
n
= , and a sequence of volumes
n
= [a
n
+ 1, a
n
]. The set of eigenvalues of S
3
tot,n
is then
/
n
= 2a
n
J, 2a
n
J + 1, . . . , 2a
n
J
and since 2J is an integer, we have /
n
/
m
for n < m. Hence we can x
M Z, take n
0
large enough such that M /
n
for all n n
0
, and consider a
sequence of states
(M)
n
=
(M)
n
,
(M)
n
_
(M)
n
,
(M)
n
_
.
It is shown in [10, 13] that in the limit n , any such sequence converges
in norm to a unique state
(M)
on the quasi-local algebra of observables A which
is the norm completion of the algebra of local observables given by
A
loc
=
_
Z
x
M
2J+1
(C)
and each
(M)
is a ground state for the derivation
J
dened by
J
(X) = lim
Z
i[H
J,
, X],
i.e.,
(M)
_
X
(X)
_
0 for X A
loc
.
All these innite volume kink states have the same GNS Hilbert space,
namely the incomplete tensor product Hilbert space
H
J
=
_
Z
__

x
C
2J+1
_
_

yZ\
y
__
where
y
=
_
[ J), if y 0,
[J), if y > 0.
Also denote =
yZ

y
H
J
, and the GNS Hamiltonian on H
J
will be
denoted H
J
.
It was proved in [13, 21] that, for all J = 1/2, 1, 3/2, . . ., these Hamiltonians
have a gap above the ground state eigenvalue, which is 0. Let us denote the
gap by
J,M
. In the case of J = 1/2, the exact value of the gap was previously
known to be 1
1
, for all 1. In [13] a very explicit conjecture is made
about the value of the gap in the limit J . For all nite J, it is a periodic
function of M, with period 2J. The conjecture in [13] is as follows:
Conjecture 1.1. For all R, the limit lim
J
(1/J)
J,J
= (r) exists and
is given by the smallest positive eigenvalue of the Jacobi operator

h
(r)
, dened
below in (2.1), where r is the solution of the equation
=
+
x=
tanh((x r)) (1.2)
with determined by = cosh.
A partial result towards this conjecture was proved in [4]. Namely, there it
is shown that there are constants c
1
> 0, and c
2
> 0, independent of M and J,
such that
c
1
J
J,M
c
2
J .
In this paper we prove that the value of the gap claimed in the conjecture
is asymptotically correct. The results in this paper by themselves, however, do
not amount to a proof of the conjecture as stated. Roughly speaking, we ob-
tain the result in the grand-canonical ensemble, and with the aid of a ground
state selection mechanism that localizes the kink. The conjecture is stated in
the canonical ensemble, i.e., for xed = M/J, in which the kink is automat-
ically localized at a xed location. As is often the case the distinction between
canonical and grand-canonical results seems merely technical, but proving math-
ematical equivalence of both formulations is often highly non-trivial. In fact,
equivalence of ensembles in the usual sense does not hold in the present situa-
tion. To prove the conjecture as stated above, some additional work has to be
done. We will report on this further work in a future publication [16].
The conjecture of [13] was based on results form perturbation theory and nu-
merical calculations on small systems presented in [21], as well as on a heuristic
calculation leading to a Boson model.
The idea is to apply a rigorous version of Dysons spin wave formalism to
the XXZ chain. Mathematically speaking, the task is to control the quadratic
approximation, described by a quasi-free system of Bosons, and show that this
approximation becomes exact in the limit J .
Several authors have attempted to do this for the XXX model, with inter-
esting results [2, 22]. In these works, the authors considered the XXX model
in an external magnetic eld, and it was necessary to let the strength of the
eld diverge as J . Such a eld selects a particular ground state (out of
the innite number of them), and creates a gap in the spectrum. This allows
one to proceed, but it limits the mathematical applicability of the spin-wave
formalism. In the case of the XXZ-model, the situation is somewhat better.
First, the XXZ chain by itself (i.e., without external magnetic eld) already has
a non-vanishing spectral gap. Second, although the innite XXZ chain also has
an innite number of ground states with the degeneracy now corresponding
to the arbitrary position of the kink any eld at just one site with a non-
vanishing component in the XY-plane, a so-called pinning eld, will select a
unique ground state, for nite J [5]. Moreover, the magnitude of this eld, as
we will show, can be taken of smaller order in J. These features of the XXZ
model will allow us to prove asymptotic properties of the model itself.
2. Main results
2.1. The limiting Boson model
Our main result will be that the spectrum of the XXZ chain, in the limit of
innite spin, can be understood as the spectrum of a non-interacting system of
Bosons on the chain, with one-particle Hamiltonian,

h
(r)
, dened on
2
(Z) by
(
h
(r)
v)
x
=
x
v
x
1
(v
x1
+v
x+1
), (2.1)
where
x
=
2 cosh((x r))
2
cosh((x 1 r)) cosh((x + 1 r))
with = ln q or, equivalently, = cosh , and r R is the position of
the kink in the reference ground state (see (1.2)).

h
(r)
has the form of the
discrete Laplacian (kinetic energy) plus a diagonal term given by
x
, which is
an exponentially localized potential well centered around the interface.
We list some properties of

h
(r)
, some of which are easily proved, while other
more detailed properties about its spectrum have at this point only been veried
numerically. We will discuss these in more detail elsewhere.
(i)

h
(r)
is a positive operator;
(ii)

h
(r)
has an eigenvalue 0 with eigenvector v
0

2
(Z), up to normalization
dened by v
0,x
= 1/ cosh((x r));
(iii) the bottom of the continuous spectrum of

h
(r)
is given by 2(1
1
);
(iv) the rst excited state of

h
(r)
corresponds to an isolated eigenvalue
(r)
below the continuum;
(v) the (
1
, r)-plane is divided in a region where
(r)
is the only eigenvalue
below the continuum, and a region where there is another isolated eigen-
value between
(r)
and the bottom of the continuum.
Let

h
(r)
(x, y) be the bi-innite Jacobi matrix expressing (2.1) in the standard
Kronecker-delta basis of
2
(Z), i.e.,
h
(r)
(x, y) =
x
x,y
(
x1,y
+
x+1,y
).
The Boson Hamiltonian is then given by second quantization of

h
(r)
:
H
(r)
=
x,yZ
h
(r)
(x, y)a
x
a
y
,
where a
x
and a
y
are the creation and annihilation operators for a Boson at site
x and y, respectively. They act on the Bosonic Fock space with one-particle
space
2
(Z), T, and satisfy the canonical commutation relations (CCR)
a
y
a
x
a
x
a
y
=
x,y
1, a
y
a
x
a
x
a
y
= a
y
a
x
a
x
a
y
= 0, x, y Z.
Let

T denote the vacuum vector which, up to a scalar factor is uniquely
characterized by the property a
x
= 0, for all x Z.
We will often use the following standard orthonormal basis in T. Introduce
n = n
x
N
xZ
, ^ =
_
n
x
n
x
<
_
.
Then, the set
n
[ n ^, where
n
=
xZ
1
(n
x
!)
1/2
(a
x
)
nx
, (2.2)
is an orthonormal basis of T.
The GNS Hilbert spaces of the spin chains, H
J
, J 1/2, 1, 3/2, . . ., can
be identied with a nested sequence of subspaces of T, dened for each J, as
the linear span of all vectors
n
, with n
x
2J. We will use this identication
throughout the paper, and we will use the same symbol
n
to denote a vector
in the spin Hilbert space H
J
and the boson space T. More precisely, we will
use the projections P
n,x
on T which project onto the rst 2n boson states at
site x, i.e., on the states
n
with 0 n
x
2n, and denote P
n
=

x
P
n,x
, and
nd
H
J
= P
J
T.
More details on the Boson model are given in Section 3.3.
2.2. Ground state selection by the external eld
To prove full convergence of the low-energy spectrum, we need to add an
external eld. A physical external eld would be of the form
h
J,x

S
x
. (2.3)
For our purposes, however, the eld is a perturbation and our results will gen-
erally be more interesting if we can prove them with smaller perturbations.
Ideally, a vanishingly small eld localized at one site should suce to select a
reference ground state. It turns out that we cannot quite do this in the present
setup. We shall use a perturbation of the form
sup
x
_
h
J,x

S
x
_
, (2.4)
which is still signicantly smaller than a uniform eld. The meaning of this
operator is clear if we express states in a basis which is diagonal for each
h
J,x
S
x
.
It is important to remark that we can take
[
h
J,x
[ h
J
= h(J lnJ)
2/3
,
such that after scaling with J
1
the external eld vanishes, in contrast with
the elds employed in previous works on the XXX model [2, 22]. In fact some
results can be obtained with h
J
0, or h
J
= hln J.
Mathematically, the eld (2.4) does slightly more than a eld localized at
one site can achieve. It not only pins the interface, but also puts some control
on the local uctuations around the selected ground state, which cannot be
controlled otherwise. The fact that we can let the eld vanish as J increases,
and that we do not need a global eld like (2.3), are signs that these uctuations
are smaller in the XXZ model than in the isotropic model.
There is another mechanism of selecting a ground state, namely by restricting
the full Hilbert space to a subspace of states with xed total magnetization in
the 3-direction, since it is known that in each such sector there is exactly one
ground state, see Section 1. The ground states that are pinned by an external
eld are like grand-canonical averages of the canonical ground states with xed
magnetization. In the limit J the canonical description can be obtained
from the grand canonical one by a result analogous to the result of [21, Sections
5.11, 5.12] about equivalence of ensembles in the 2-dimensional, spin-1/2 XXZ
model. Note however that there is no equivalence of ensembles in our situation,
in the sense that correlation functions are typically dierent. But the dierence
between a canonical state and a grand-canonical state with the right average
magnetization can be expressed completely in terms of the uctuations of the
total 3-magnetization in the grand canonical state which are non-zero even in
the limit J , while they are identically zero in the canonical state. The
results about the canonical description require signicant additional work and
will be discussed elsewhere [16].
Finally we mention that because of the pinning eld (2.4), our results give
only a partial proof of Conjecture 1.1. A full proof requires in addition that
the lowest excited state can be obtained from the ground state by acting on it
with 1 spin wave operator, or more generally by a nite number of operators
independent of J. This is a problem that should be handled at the level of
the spin system, rather than in the spin wave formalism. Some results in that
direction have recently been obtained in [18].
An advantage of the grand-canonical description is that it clearly exhibits
how the Boson limit arises as the rst quantum correction to the classical limit.
2.3. Statement of the main results
For A a self-adjoint operator and a, b R, denote by P
(a,b)
(A) the spectral
projection of A onto (a, b). For A acting on Fock space, denote
the spectrum of A in T by (A);
the spectrum of P
J
AP
J
in H
J
by
J
(A);
the spectrum of P
nJ
AP
nJ
in H
nJ
by
nJ
(A), where H
nJ
is shorthand
for P
nJ
T.
Also denote s-lim the strong, or strong resolvent, operator limit for bounded,
resp. unbounded operators acting on T.
In the GNS space H
J
it is convenient to dene S
3
tot
in the renormalized
sense: S
3
tot
=
xZ
_
S
3
x
sgn(x 1/2)
, and denote
=
xZ
_
tanh((x r)) sgn
_
x
1
2
__
instead of (1.2).
Theorem 2.1. We have
s-lim
J
1
J
H
J
=

H
(r)
, s-lim
J
1
J
S
3
tot
= 1.
The proof of this result is given in Section 5.2.
Corollary 2.1.
(i) If (

H
(r)
), there exists
J

J
((1/J)H
J
) such that
lim
J
J
= .
(ii) If a, b R, and a, b /
pp
(

H
(r)
), then
s-lim
J
P
(a,b)
_
1
J
H
J
_
= P
(a,b)
(

H
(r)
).
(iii) If a, b R, and (a, b), then
s-lim
J
P
(a,b)
_
1
J
S
3
tot
_
= 1.
This corollary is a direct consequence of Proposition 5.1 and [20, Theo-
rem VIII.24].
In addition we can obtain spectral concentration of (1/J)H
J
around discrete
eigenvalues of

H
(r)
(also proved in Section 5.2).
Theorem 2.2. For every isolated eigenvalue E of

H
(r)
, there exists an interval
I
J
= (E
J
, E +
J
) with lim
J
J
J
ln J
= 0,
such that for any interval I around E s.t. I (

H
(r)
) = E:
s-lim
J
P
I\IJ
_
1
J
H
J
_
= 0, s-lim
J
P
IJ
_
1
J
H
J
_
= P
{E}
_
H
(r)
_
.
Applying the same reasoning to S
3
tot
, we nd that the interval (a, b) in item
(iii) of Corollary 5.2 can be chosen as
(a, b) = (
J
, +
J
)
with again lim
J

J
J(lnJ)
1
= 0.
To prove full convergence of the spectrum, we have to add the external eld
(2.4) to H
J
, or, through the identication H
J
= P
J
T, add:
h
J
sup
x
(N
x
) (2.5)
with h
J
> 0, and N
x
= a
x
a
x
. Let us assume we add this eld to (1/J)H
J
, so
h
J
already contains the factor J
1
.
Take 0 < n
J
< J, we get that on (H
nJ
)
H
J
1
J
H
J
+h
J
sup
x
N
x
h
J
sup
x
N
x
h
J
n
J
1.
Clearly, by choosing h
J
and n
J
such that
lim
J
h
J
n
J
=
statements about the spectrum on H
J
reduce to statements about the spectrum
on H
nJ
. Or, if one chooses to make statements about the spectrum below a
certain value E, it is sucient to choose h
J
such that lim
J
h
J
n
J
> E +.
Theorem 2.3. Let n
J
= [(J(lnJ)
1
)
1/3
]. If / (

H
(r)
), then /
nJ
((1/J)
H
J
) for J large enough.
This result is proved in Section 5.3.
Hence we get convergence of the spectrum of (1/J)H
J
+ h
J
sup
x
N
x
to the
spectrum of

H
(r)
if
lim
J
h
J
= 0, lim
J
h
J
_
J
ln J
_
1/3
= .
3. Derivation of the Boson limit
3.1. The classical limit
The Boson limit can be considered as the rst quantum correction to the
classical limit. Therefore, we rst discuss the classical limit.
It is well known that for any quantum spin system, after rescaling each
spin matrix by J
1
and taking the large spin limit J , one obtains the
corresponding classical spin system [14]. For the XXZ chain in a nite volume
this is dened by the Hamiltonian
H
cl
x
_
=
b1
x=a
1
1
1
x
1
x+1
+
2
x
2
x+1
_
3
x
3
x+1
+
_
1
2
_
3
x
3
x+1
_
,
(3.1)
where
x
is a unit vector in R
3
. Minimizing this function with respect to
x
x
yields zero-energy congurations that are planar waves [21], i.e., in spherical
coordinates we nd congurations
(r)
x
() = (
(r)
x
, ), [, ] (the same at
all sites), and
(r)
x
= 2 arctan(q
xr
), where r R determines the value of the
total 3-magnetization.
Dening again = lnq or = cosh(), we have
cos
(r)
x
=
1 q
2(xr)
1 +q
2(xr)
= tanh
_
(x r)
_
, (3.2)
sin
(r)
x
=
2q
xr
1 +q
2(xr)
=
1
cosh
_
(x r)
_, (3.3)
such that the zero-energy solutions clearly describe kinks centered around r.
For the classical model, to look at the low-energy behavior amounts to mak-
ing a quadratic Taylor approximation to (3.1). At each site, the angle coordi-
nates are replaced by new coordinates
q
x
=
x
(r)
x
, p
x
= sin
(r)
x
(
x
),
and the resulting harmonic oscillator Hamiltonian is
H
cl
=
1
2
b1
x=a
_
+
x
(q
2
x
+p
2
x
)
1
_
q
x
q
x+1
+p
x
p
x+1
_
_
+
1
2
b
x=a+1
_
x
(q
2
x
+p
2
x
)
1
_
q
x
q
x1
+p
x
p
x1
_
_
, (3.4)
where
x
are given by
x
=
sin
(r)
x1
sin
(r)
x
=
cosh((x r))
cosh((x 1 r))
.
This can be derived using the identities of Lemma 5.1 below.
At sites other than the boundary sites we have a single-site potential
x
=
+
x
+
x
=
2 cosh((x r))
2
cosh((x 1 r)) cosh((x + 1 r))
which is an exponentially localized well centered around the interface.
3.2. Grand canonical states
For (
x
,
x
) a general unit vector on the sphere at site x, we can dene the
coherent spin state in (C
2J+1
)
x
(see [1, 14]):
[(
x
,
x
)) = exp
_
1
2
x
_
S
x
e
ix
S
+
x
e
ix
_
_
[J)
=
J
mx=J
_
2J
J m
x
_
1/2_
cos
1
2
x
_
J+mx
_
sin
1
2
x
_
Jmx
expi(J m
x
)
x
[m
x
).
This is particularly interesting if we choose the unit vectors at each site to
be the classical zero-energy congurations. In a nite volume , it is easy to
see that
(r)
()
_

x
e
iJ
(
(r)
x
, )
_
=
1
|
(z)
(z)
,
where
(z)
is the generating vector for the ground state vectors

(M)
, i.e., the
grand canonical ground state:
(z)
=
||J
M=||J
z
M
(M)
evaluated at z = q
r
expi = expr expi.
The fact that a classical ground state yields an exact quantum ground state
through the coherent state representation, is because H
J,
is a normal Hamil-
tonian in the sense of [14], and the classical and quantum ground state energies
are (exactly) related by the scaling factor J
2
.
Since these states are product states, their thermodynamic limit is easily
obtained. In the GNS Hilbert space H
J
, dene the embedding of [
(r)
()) as

(r)
()
_
(r)
()
_
_

xZ\
.
Lemma 3.1. For a sequence of intervals
n
= [a
n
+ 1, a
n
] tending to Z, we
have for m > n
_
_
[
(r)
m
()) [
(r)
n
())
_
_
2Jq
2an
1 q
2(aman)
1 q
2
_
q
2r
+q
22r
_
.
Proof.
_
_
[
(r)
m
()) [
(r)
n
())
_
_
2
=
_
_

xm\n
e
iJ
[(
(r)
x
, ))
xm\n
x
_
_
2
= 2
an
x=am+1
e
iJ
J [ (
(r)
x
, ))
am
x=an+1
e
iJ
J [ (
(r)
x
, ))
an
x=am+1
e
iJ
(
(r)
x
, ) [ J)
am
x=an+1
e
iJ
(
(r)
x
, ) [ J)
= 2 2
am1
x=an
1
(1 +q
2(x+r)
)
J
am
x=an+1
1
(1 +q
2(xr)
)
J
2
_
1 exp
_
J
am1
x=an
q
2(x+r)
J
am
x=an+1
q
2(xr)
__
2J
_
am1
x=an
q
2(x+r)
+
am
x=an+1
q
2(xr)
_
= 2Jq
2an
1 q
2(aman)
1 q
2
_
q
2r
+q
22r
_
,
where we used the inequalities (for u 0) 1 +u exp u and 1 expu u.
2
It follows that the sequence [
(r)
n
()) has a limit [
(r)
()) in H
J
that we can
formally write as
(r)
()
_
_

x0
exp
_
1
2
(
(r)
x
)(S
x
e
i
S
+
x
e
i
)
_
x
_
_

x>0
exp
_
1
2
(r)
x
(S
x
e
i
S
+
x
e
i
)
_
x
_
.
In general, if we write z = [z[e
i
, then
(z)
= expiS
3
tot,
(|z|)
, (3.5)
so it will be sucient to restrict our detailed analysis to the grand canonical
states
(e
r
)
. The expectation in this state will be denoted

(r)
:
(r)
=

(e
r
)
,
(e
r
)
_
_
_
(e
r
)
_
_
2
and its thermodynamic limit
(r)
.
In this case, the coherent states are rotations of the top state [J) through
an angle
(r)
x
around the 2-axis, [(
(r)
x
, 0)) = expi
(r)
x
S
2
x
[J). Introduce the
notation u
(r)
x
= (
(r)
x
, 0), or in Cartesian coordinates u
(r)
x
= (sin
(r)
x
, 0, cos
(r)
x
).
In the remainder, we will always keep r xed and do not make explicit the
dependence on r of various quantities. Notice that by periodicity it is sucient
to take r [0, 1).
Denote by e
1
x
, e
2
x
, e
3
x
the standard basis in R
3
(the same at every site), and
f
1
x
= cos
(r)
x
e
1
x
sin
(r)
x
e
3
x
,
f
2
x
= e
2
x
,
f
3
x
= u
(r)
x
= sin
(r)
x
e
1
x
+ cos
(r)
x
e
3
x
.
Conversely
e
1
x
= cos
(r)
x
f
1
x
+ sin
(r)
x
f
3
x
,
e
2
x
= f
2
x
,
e
3
x
= sin
(r)
x
f
1
x
+ cos
(r)
x
f
3
x
,
i.e., f
1
x
, f
2
x
, f
3
x
form an orthonormal frame for R
3
and f
1
x
, f
2
x
an orthonormal
frame for the tangent plane R
2
to the unit sphere at u
(r)
x
.
For v
x
R
3
, we denote by v
x
R
2
the projection of v
x
onto the tangent
plane at u
(r)
x
(shifted to the origin), i.e., v
x
= ( v
1
x
, v
2
x
) and
v
1
x
= v
x
f
1
x
= cos
(r)
x
v
1
x
sin
(r)
x
v
3
x
, v
2
x
= v
x
e
2
x
= v
2
x
.
Conversely, if v
x
R
2
we associate to it a vector v
x
R
3
by putting the
component along the u
(r)
x
-axis zero:
v
1
x
= cos
(r)
x
v
1
x
, v
2
x
= v
2
x
, v
3
x
= sin
(r)
x
v
1
x
.
Also denote
v
x
S
x
= v
1
x
S
1
x
+v
2
x
S
2
x
+v
3
x
S
3
x
and dene rotated spin operators [1, Eq. (3.9)]
S
1
x
= expi
(r)
x
S
2
x
S
1
x
expi
(r)
x
S
2
x
= f
1
x
S
x
= cos
(r)
x
S
1
x
sin
(r)
x
S
3
x
, (3.6)
S
2
x
= expi
(r)
x
S
2
x
S
2
x
expi
(r)
x
S
2
x
= S
2
x
, (3.7)
S
3
x
= expi
(r)
x
S
2
x
S
3
x
expi
(r)
x
S
2
x
= f
3
x
S
x
= sin
(r)
x
S
1
x
+ cos
(r)
x
S
3
x
. (3.8)
Hence we nd that [(
(r)
x
, 0)) is the top state for the rotated spin operators:
S
3
x
(
(r)
x
, 0)
_
=

S
3
x
expi
(r)
x
S
2
x
[J) = expi
(r)
x
S
2
x
S
3
x
[J) = J
(
(r)
x
, 0)
_
.
(3.9)
The rotated spin raising and lowering operators are
x
=

S
1
x
i
S
2
x
= sin
(r)
x
S
3
x
+ cos
(r)
x
S
1
x
iS
2
x
,
or
S
+
x
= sin
(r)
x
S
3
x
+ cos
2

(r)
x
2
S
+
x
sin
2

(r)
x
2
S
x
,
x
= sin
(r)
x
S
3
x
sin
2

(r)
x
2
S
+
x
+ cos
2

(r)
x
2
S
x
.
One of the main observations is that because of (3.9), it is much more con-
venient to introduce the spin wave formalism in the rotated spin basis than in
the original one. Following [22], introduce
n = n
x
N
xZ
,
^ =
_
n [
x
n
x
<
_
, ^
J
=
_
n [ x: n
x
2J,
x
n
x
<
_
, (3.10)
n
=
xZ
1
n
x
!
_
2J
n
x
_
1/2
_
x
_
nx
(r)
. (3.11)
The set
n
[ n ^
J
is an orthonormal basis for H
J
.
We conclude with a little lemma that complements (3.5).
Lemma 3.2. For < < ,
expiS
3
tot,
(r)
_
=
_
cos(/2) +i cos
(r)
x
sin(/2)
_
2J
exp
_
i
x
() sin
(r)
x

S
x
_
(r)
_
,
where
x
() =
sin(/2)
cos(/2) +i cos
(r)
x
sin(/2)
.
Proof. We write the disentanglement relation [1, Eq. (A4)]
expiS
3
x
= exp
_
i
_
1
2
sin
(r)
x
(
S
+
x
+

S
x
) + cos
(r)
x

S
3
x
__
= expiy
x
exp(lny
3
)
S
3
x
expiy
+

S
+
x
,
where
y
3
=
_
cos(/2) +i cos
(r)
x
sin(/2)
_
2
,
y
+
= y
=
sin
(r)
x
sin(/2)
cos(/2) +i cos
(r)
x
sin(/2)
=
x
() sin
(r)
x
and recall that [
(r)
) is the product state of top states for the

S-operators. 2
It is now also clear how to choose the external eld

h
J,x
in (2.3) such that
h
J,x

S
x
= h
J

S
3
x
, namely

h
J,x
= h
J
u
(r)
x
, h
J
> 0.
3.3. The Boson chain
We consider immediately the innite volume situation. Consider the Hilbert
space of wave functions
2
(Z) which we alternatively consider as the usual com-
plex Hilbert space with inner product v, w) =

xZ
v
x
w
x
or as a real linear
space with symplectic form and complex structure , i.e.,
v
2
(Z) =
_
(v
1
x
, v
2
x
) R
2
_
xZ
,
(v, w) =
xZ
v
1
x
w
2
x
v
2
x
w
1
x
,
v =
_
(v
2
x
, v
1
x
) R
2
_
xZ
.
The CCR -algebra CCR(
2
(Z), ) is generated by unitaries W(v) [ v
2
(Z)
which satisfy the commutation relations
W(v)W(w) = exp
_
i
2
(v, w)
_
W(v +w).
The Fock state is the quasi-free state on CCR(
2
(Z), ) determined by

_
W(v)
_
= exp
_
1
2
v, v)
_
.
Its GNS representation is the usual Fock representation on a Fock space T with
a vacuum vector

=
xZ
[0)
x
and creation and annihilation operators a
x
such
that
W(v) = exp
_
i
xZ
(v
x
a
x
+v
x
a
x
)
_
.
(We do not distinguish between W(v) and its representative in the Fock repre-
sentation).
In this Fock representation we can dene a quasi-free Boson Hamiltonian by
canonically quantizing the classical harmonic oscillator Hamiltonian (3.4). This
means replacing the position and momentum variables by canonical pairs q
x
,
p
x
, with commutation relations [q
x
, p
y
] = i
x,y
and
a
x
=
q
x
ip
x
2
, a
x
=
q
x
+ip
x
2
.
The result is
H
(r)
=
b1
x=a
_
+
x
_
a
x
a
x
+
1
2
_
x
a
x+1
_
+
b
x=a+1
_
x
_
a
x
a
x
+
1
2
_
x
a
x1
_
.
The corresponding innite volume derivation is denoted
(r)
() = lim
Z
i [

H
(r)
, ]
and the GNS Hamiltonian is denoted

H
(r)
,
H
(r)
=
xZ
x
a
x
a
x
1
a
x
(a
x1
+a
x+1
).
Denote a
(v) =

x
v
x
a
x
. If v is local, i.e., has only nitely many v
x
non-
zero, then
lim
Z
_
H
(r)
, a
(v)
= a
h
(r)
v),
where

h
(r)
is the bi-innite Jacobi matrix dened on
2
(Z) by
(
h
(r)
v)
x
=
x
v
x
(v
x1
+v
x+1
). (3.12)
For the nite system localized in = [a, b], we have
i
_
H
(r)
, a
(v)
= a
h
(r)
v),
where (
h
(r)
v)
x
= (
h
(r)
v)
x
for x in the bulk [a + 1, b 1], and at the boundary
sites we get
h
(r)
v
a
=
+
a
v
a
v
a+1
,

h
(r)
v
b
=
b
v
b
v
b1
.
Some important properties of

h
(r)
were given in Section 2.1. Recall the
existence of a zero-mode (property (ii)) given by:
v
0,x
= sin
(r)
x
=
1
cosh((x r))
.
That this is an eigenvector of

h
(r)
with eigenvalue zero, is easily veried:
x
v
x,0
v
x1,0
=
sin
(r)
x1
sin
(r)
x
sin
(r)
x

1
sin
(r)
x1
= 0.
The origin of the zero-mode v
0
is well understood. It arises from the rotation
symmetry of H
J,
. This can be seen from Lemma 3.2, by taking J
1/2
and
formally identifying J
1/2

S
x
with a
x
(this identication will be made more pre-
cise below). For every N N, there is a zero-energy vector
0,N
corresponding
to an N-particle occupation of v
0
:
0,N
=
1
[v
0
[
N
2
N!
a
(v
0
)
N

.
We dene P
0
the projection on v
0
and

P
0
the projection onto the zero-energy
vectors, i.e.,
P
0
=
[v
0
)v
0
[
[v
0
[
2
, (3.13)
P
0
=
NN
[
0,N
)
0,N
[. (3.14)
For completeness we recall that we will always use the standard orthonormal
basis in T, i.e., the set
n
[ n ^, where
n
=
xZ
1
(n
x
!)
1/2
(a
x
)
nx
. (3.15)
We use the same symbol
n
to denote a vector in the spin Hilbert space H
J
and
the boson space T since we will use the identication of H
J
with a subspace of
T as discussed before.
4. The large spin limit as a quantum central limit
Since the grand canonical states can be written as the all + state for rotated
spin operators, we are in the usual situation of a fully ferromagnetic state in
which we expect a boson limit after rescaling with J
1/2
, i.e., (1/
2J)
x
a
x
in some sense.
One way to make this precise is to dene uctuation operators: for v
x
R
3
,
F
J
(v
x
) =
1
J
_
v
x
S
x
(r)
(v
x
S
x
)
_
,
i.e., F
J
(v
x
) measures the deviation from the ground state expectation value of
the spin in the v
x
direction. Similar uctuation operators are used to study
uctuations of extensive observables, and their thermodynamic limit can be
taken as a noncommutative central limit [8, 9].
A connection between the spin limit J and these quantum central limits
was made in [15], with the caveat that each spin-J had to be represented as a
sum of spin-(1/2)s, instead of working with an irreducible representation. This
latter restriction is however not necessary. We have, using results from [1, 14]:
(r)
_
expiv
x
S
x
_
=
_
cos
_
1
2
[v
x
[
_
+i
v
x
u
(r)
x
[v[
sin
_
1
2
[v
x
[
__
2J
, (4.1)
(r)
(v
x
S
x
) = J(v
x
u
(r)
x
),
(r)
_
F
J
(v
x
)F
J
(w
x
)
_
=
1
2
_
v
x
w
x
(v
x
u
(r)
x
)(w
x
u
(r)
x
) +i(v
x
w
x
) u
(r)
x
_
.
The latter quantity denes a (degenerate) inner product on R
3
:
v
x
, w
x
)
x
= 2
(r)
_
F
J
(v
x
)F
J
(w
x
)
_
. (4.2)
It is not hard to use (4.1) to show that
lim
J
(r)
_
expiF
J
(v
x
)
_
= exp
_
1
2
v
x
, v
x
)
_
. (4.3)
Clearly if either v
x
or w
x
is along the u
(r)
x
-direction, then v
x
, w
x
)
x
= 0.
Hence (4.2) denes an inner product in R
2
= C, the tangent plane to the unit
sphere at u
(r)
x
. If for v
x
, w
x
R
2
, v
x
, w
x
are the corresponding vectors in R
3
(see Section 3.2),
v
x
, w
x
)
x
v
x
, w
x
)
x
= ( v
1
x
+i v
2
x
)( w
1
x
+i w
2
x
),
i.e., the standard inner product in C. We see that there are no uctuations
in the direction perpendicular to the tangent plane at the classical zero-energy
solution. Two vectors in R
3
at the same site will be called equivalent if their
projection onto the tangent plane at u
(r)
x
is the same.
For v = (v
x
R
3
)
xZ
, with only nitely many v
x
,= 0, we simply extend
this by putting
F
J
(v) =
xZ
F
J
(v
x
),
v, w) = 2
_
F
J
(v)F
J
(w)
_
=
xZ
v
x
, w
x
)
x
.
Using (4.3), and standard techniques, the quantum central limit theorem follows:
lim
J
(r)
_
expiF
J
(v
1
) . . . expiF
J
(v
n
)
_
=
_
W(v
1
) . . . W(v
n
)
_
,
where is the Fock state on the CCR-algebra CCR(
2
(Z), ) introduced in
Section 3.3, and the vectors v
1
, . . . , v
n
on the r.h.s. mean their respective equiv-
alence classes in
2
(Z). The result can be understood intuitively from
v
x
S
x
= v
1
x

S
1
x
+ v
2
x

S
2
x
= ( v
1
x
i v
2
x
)
S
+
x
+ ( v
1
x
+i v
2
x
)
x
for v
x
R
2
and corresponding v
x
R
3
.
When studying properties of the GNS Hamiltonian, it is actually easier to
make a correspondence between the GNS Hilbert spaces H
J
of
(r)
and T of .
Following [22], introduce the projection P
J,x
on T which projects onto the
rst 2J +1 Boson states at site x, i.e., on the states
n
(3.15) with 0 n
x
2J,
and denote P
J
=
x
P
J,x
, i.e., P
J
projects onto the states
n
with n ^
J
, see
(3.10). By identifying
n
(3.11) with
n
(3.15), it is clear that H
J
= P
J
T ,
where = means unitarily equivalent. Under this equivalence, we nd that the
spin operators are given by [22]:
1
2J
x
= P
J
a
x
g
J
(x)
1/2
,
1
2J
S
+
x
= g
J
(x)
1/2
a
x
P
J
, J

S
3
x
= P
J
a
x
a
x
P
J
,
(4.4)
where g
J
(x) = g
J
(a
x
a
x
) and
g
J
(n) =
_
_
_
1
1
2J
n, n 2J,
0, n > 2J.
5. The low energy spectrum
5.1. Some estimates for the Hamiltonian
We rst need the following identities:
Lemma 5.1.
cos
(r)
x1
+ cos
(r)
x+1
=
x
cos
(r)
x
, (5.1)
1
(sin
(r)
x1
+ sin
(r)
x+1
) =
x
sin
(r)
x
, (5.2)
sin
(r)
x
sin
(r)
x1
+
1
cos
(r)
x
cos
(r)
x1
=
1
, (5.3)
1
cos
(r)
x
sin
(r)
x1
sin
(r)
x
cos
(r)
x1
=
_
1
2
sin
(r)
x
, (5.4)
1
sin
(r)
x1
sin
(r)
x
+ cos
(r)
x1
cos
(r)
x

_
1
2
cos
(r)
x
=
x
. (5.5)
Proof. These are straightforward computations using the denitions (3.2) and
(3.3) of cos
(r)
x
and sin
(r)
x
, and the addition laws for sinh and cosh. 2
With this lemma we can write the Hamiltonian in terms of the

S-operators.
Corollary 5.1.
H
J
x,x+1
= J
2
1
2
_
S
+
x

S
x+1
+

S
x

S
+
x+1
_
x,x+1

S
3
x

S
3
x+1
+J
_
1
2
_
cos
(r)
x

S
3
x
cos
(r)
x+1
S
3
x+1
_
+
_
1
2
_
sin
(r)
x

S
1
x

S
3
x+1
sin
(r)
x+1
S
3
x

S
1
x+1
_
J
_
1
2
_
sin
(r)
x

S
1
x
sin
(r)
x+1
S
1
x+1
_
,
where
x,x+1
=
+
x
+
_
1
2
cos
(r)
x
=
x+1
_
1
2
cos
(r)
x+1
.
Proof. This follows immediately from the relations (3.6)(3.8), and the previous
lemma. 2
With the Hamiltonian in terms of the

S-operators, we can apply the unitary
transformation (4.4) to write the spin Hamiltonian as an operator on T:
1
J
H
J,
= P
J
_
b1
x=a
_
+
x
g
J
(x)a
x
a
x
1
a
x+1
g
J
(x + 1)
1/2
g
J
(x)
1/2
a
x
+
b
x=a+1
_
x
g
J
(x)a
x
a
x
1
a
x1
g
J
(x 1)
1/2
g
J
(x)
1/2
a
x
1
2
2J
_
cos
(r)
b
N
2
b
cos
(r)
a
N
2
a
+
b1
x=a
x,x+1
(N
x
N
x+1
)
2
2J
+
1
2
2J
1/2
b1
x=a
_
sin
(r)
x+1
(g
J
(x + 1)
1/2
a
x+1
+a
x+1
g
J
(x + 1)
1/2
)N
x
sin
(r)
x
(g
J
(x)
1
2
a
x
+a
x
g
J
(x)
1/2
)N
x+1
_
_
P
J
, (5.6)
and likewise for the -volume GNS Hamiltonian:
1
J
H
J
= P
J
_
xZ
_
x
g
J
(x)a
x
a
x
(5.7)
1
a
x
g
J
(x)
1/2
_
g
J
(x 1)
1/2
a
x1
+g
J
(x + 1)
1/2
a
x+1
_
+
b1
x=a
x,x+1
(N
x
N
x+1
)
2
2J
+
1
2
2J
1/2
xZ
sin
(r)
x
(g
J
(x)
1/2
a
x
+a
x
g
J
(x)
1/2
)(N
x1
N
x+1
)
_
P
J
.
In the usual language of spin wave theory [6, 7, 22], the rst term (between
[. . . ]) in the Hamiltonian is called the kinematical interaction H
J,kin
, and the
second term the dynamical interaction H
J,dyn
. The last term, which describes
transitions between subspaces with constant number of particles, is not usually
present. We denote it H
J,tran
. Note that we dene these three operators with
the right scaling already included, i.e.,
1
J
H
J
= P
J
H
J,kin
+H
J,dyn
+H
J,tran
P
J
.
We will only let these operators act on vectors in H
J
, hence we may forget
about the P
J
. To further simplify some notation, introduce
the column vector A of annihilation operators:
A =
_
_
_
_
.
.
.
a
x
.
.
.
_
_
_
_
,
the diagonal matrix G
J
:
G
J
(x, y) = g
J
(x)
x,y
.
Then we can write
H
J,kin
= A
G
1/2
J
h
(r)
G
1/2
J
A
1
2J
x
a
x
a
x
,
H
(r)
= A
h
(r)
A,
where

h
(r)
is the one-particle Boson Hamiltonian, see (3.12), i.e., the matrix
with entries
h
(r)
(x, y) =
x
x,y
1
(
x1,y
+
x+1,y
).
In the following we will x for every J an n
J
N with 0 < n
J
< J and
make statements about the subspace P
nJ
H
J
of H
J
. In the end we will formulate
results on the whole of H
J
by adding an external pinning eld which will take
care of the states in (P
nJ
H
J
)
H
J
. For simplicity denote H
nJ
= P
nJ
H
J
.
Lemma 5.2. On H
nJ
we have the lower bound
H
J,kin

_

(r)
g
J
(2n
J
)
1
J
_
N
tot

(r)
A
G
1/2
P
0
G
1/2
A,
where
(r)
is the spectral gap of

h
(r)
and P
0
is dened in (3.13). An upper
bound (on the whole H
J
) is given by
H
J,kin
|
h
(r)
|N
tot
.
Proof. P
nJ
projects onto the vectors with at most 2n
J
particles per site, such
that on H
nJ
:
G
J
g
J
(2n
J
)1.
Obviously G
J
1 on H
J
. The lemma follows from the bounds on

h
(r)
:

(r)
(1 P
0
)
h
(r)
|
h
(r)
|1
and also
x
2. 2
We are going to compare the spectrum of H
J,kin
with the spectrum of

H
(r)
.
Both operators commute with N
tot
so it is sucient to compare them on eigen-
states of N
tot
. Also for H
J,dyn
it is sucient to look at eigenstates of N
tot
.
In the proof of the following lemmata we will use the following notation: for
n, m ^:
T
x
n = m i
_
_
m
x
= n
x
+ 1,
m
x1
= n
x1
1,
m
y
= n
y
, y ,= x, x + 1,
A
x
n = m i
_
m
x
= n
x
1,
m
y
= n
y
, y ,= x.
Lemma 5.3. Let
N
H
nJ
, |
N
| = 1, N
tot
N
= N
N
. Then
_
_
(

H
(r)
H
J,kin
)
N
_
_
2(1 +
1
)
n
J
N
J
,
_
_
H
J,dyn
N
_
_
4n
J
N
J
.
Proof. We can write
N
=
n
c
n
n
with
n
[c
n
[
2
= 1,
where the sum runs over n ^
J
for which

x
n
x
= N. On basis vectors, we
have
H
J,kin
n
=
x
n
x
g
J
(n
x
)
n
x
_
(n
x
+ 1)g
J
(n
x
)g
J
(n
x1
1)n
x1
1/2
x
n
x
_
(n
x
+ 1)g
J
(n
x
)g
J
(n
x+1
1)n
x+1
1/2
T
+
x
n
and for

H
(r)
the same with the g
J
1. We compare term by term. The rst
one gives:
_
_
_
n,x
c
n
x
_
n
x
n
x
g
J
(n
x
)
n
_
_
_
2
=
1
(2J)
2
n
[c
n
[
2
x
n
2
x
4N
2
n
2
J
J
2
,
where we used
x
2, n
x
2n
J
and

x
n
x
= N. For the second term we use
analogously g
J
(n
x
) g
J
(2n
J
), and nd:
_
_
_
n,x
c
n
_
[(n
x
+ 1)n
x1
]
1/2
[(n
x
+ 1)g
J
(n
x
)g
J
(n
x1
1)n
x1
]
1/2
_
x
n
_
_
_
=
_
_
_
m,x
c
T
+
x1
m
_
[m
x
(m
x1
+ 1)]
1/2
[m
x
g
J
(m
x
1)g
J
(m
x1
)(m
x1
+ 1)]
1/2
_
m
_
_
_
m
_
x
[c
T
+
x1
m
[
[m
x
(m
x1
+ 1)]
1/2
[m
x
g
J
(m
x
1)g
J
(m
x1
)(m
x1
+ 1)]
1/2
_
2
m
_
x
[c
T
+
x1
m
[[m
x
(m
x1
+ 1)]
1/2
(1 g
J
(2n
J
))
_
2
n
2
J
J
2
m
_
x
[c
T
+
x1
m
[
2
(m
x1
+ 1)
__
x
m
x
_
=
N
2
n
2
J
J
2
and the same for the third term. Summing everything together we nd
_
_
(

H
(r)
H
J,kin
)
N
_
_
2(1 +
1
)
Nn
J
J
.
For H
J,dyn
we use the same reasoning,
x,x+1
1, and
(n
x
n
x+1
)
2
2n
2
x
+ 2n
2
x+1
4n
J
(n
x
+n
x+1
)
to nd
_
_
H
J,dyn
N
_
_
4n
J
N
J
.
2
For H
J,tran
we have the following estimate.
Lemma 5.4. Let H
nJ
, || = 1. Then
_
_
H
J,tran
_
_
2
_
1
2
[v
0
[
1
_
(2n
J
+ 1)(4n
J
)
2
J
_
1/2
,
where [v
0
[
1
is the
1
-norm of the zero-mode.
Proof. Let
=
n
c
n
n
.
Then
_
_
_
x
sin
(r)
x
g
J
(x)
1/2
a
x
(N
x1
N
x+1
)
_
_
_
2
=
_
_
_
n,x
c
n
sin
(r)
x
g
J
(n
x
1)
1/2
n
1/2
x
(n
x1
n
x+1
)
A
x
n
_
_
_
2
=
_
_
_
n,x
c
A
+
x
n
sin
(r)
x
g
J
(n
x
)
1/2
(n
x
+ 1)
1/2
(n
x1
n
x+1
)
n
_
_
_
2
=
x
c
A
+
x
n
sin
(r)
x
g
J
(n
x
)
1/2
(n
x
+ 1)
1/2
(n
x1
n
x+1
)
n
_
x
sin
(r)
x
[c
A
+
x
n
[
2
__
x
sin
(r)
x
(n
x
+ 1)[n
x1
n
x+1
[
2
_
[v
0
[
1
(2n
J
+ 1)(4n
J
)
2
x
sin
(r)
x
[c
A
+
x
n
[
2
= [v
0
[
2
1
(2n
J
+ 1)(4n
J
)
2
and likewise for the second term. 2
5.2. Strong convergence and spectral concentration
Recall the following denitions. For A a self-adjoint operator and a, b R,
denote by P
(a,b)
(A) the spectral projection of A onto (a, b). For A acting on
Fock space, denote
the spectrum of A in T by (A);
the spectrum of P
J
AP
J
in H
J
by
J
(A);
the spectrum of P
nJ
AP
nJ
in H
nJ
by
nJ
(A).
Also denote by s-lim the strong, or strong resolvent, operator limit for bounded,
resp. unbounded operators acting on T.
In the GNS space H
J
it is convenient to dene S
3
tot
in the renormalized
sense: S
3
tot
=
xZ
_
S
3
x
sgn(x 1/2)
. Also denote
=
xZ
_
cos
(r)
x
sgn
_
x
1
2
__
,
i.e., is the 3-magnetization of a classical ground state (
(r)
x
, )
xZ
(see Sec-
tion 3.1).
Proposition 5.1. We have
s-lim
J
1
J
H
J
=

H
(r)
,
s-lim
J
1
J
S
3
tot
= 1.
Proof. Introduce the set T, the nite linear space of vectors
n
with n ^. T
is a common core for (1/J)H
J
, for all J, and

H
(r)
.
Take T arbitrary (but normalized for simplicity) and denote
=
n
c
n
n
,
N
= sup
n: c
n
=0
x
n
x
.
Note that by assumption, is a nite sum of
n
with

x
n
x
< , and hence
also N
< .
Now take J large enough such that H
J
. From the proof of Lemma 5.3
and 5.4 it is clear that we can use 2n
J
N
, as soon as 2J > N
, hence
_
_
_
_
1
J
H
J

H
(r)
_
_
_
(3 +
1
)
N
2
J
+ 2
_
1
2
[v
0
[
1
_
(N
+ 1)(2N
)
2
J
_
1/2
.
The rst result follows from [20, Theorem VIII.25 (a)].
To prove the second statement, write
S
3
x
= cos
(r)
x

S
3
x
sin
(r)
x

S
1
x
= P
J
_
cos
(r)
x
(J N
x
) sin
(r)
x
_
g
J
(x)
1/2
a
x
+a
x
g
J
(x)
1/2
__
P
J
and hence
S
3
tot
=
x
S
3
x
sgn
_
x
1
2
_
= P
J
_
J
x
cos
(r)
x
N
x
x
sin
(r)
x
_
g
J
(x)
1/2
a
x
+a
x
g
J
(x)
1/2
_
_
P
J
.
Clearly

x
cos
(r)
x
N
x
N
tot
, and for T as before
_
_
_
x
sin
(r)
x
a
x
g
J
(x)
1/2
_
n
c
n
n
_
_
_
_
2
=
_
_
_
n,x
sin
(r)
x
c
n
(n
x
+ 1)
1/2
g
J
(n
x
)
1/2
Axn
_
_
_
2
(N
+ 1)[v
0
[
2
1
.
Hence
_
_
_
_
1
J
S
3
tot
_
_
_
1
J
_
N
+ 2(N
+ 1)
1/2
[v
0
[
1
_
and (1/J)S
3
tot
1 strongly on T. 2
Corollary 5.2.
(i) If (

H
(r)
), there exists
J

J
((1/J)H
J
) such that
lim
J
J
= .
(ii) If a, b R, and a, b /
pp
(

H
(r)
), then
s-lim
J
P
(a,b)
_
1
J
H
J
_
= P
(a,b)
_
H
(r)
_
.
(iii) If a, b R, and (a, b), then
s-lim
J
P
(a,b)
_
1
J
S
3
tot
_
= 1.
Proof. The previous proposition and [20, Theorem VIII.24]. 2
In addition we can prove spectral concentration of (1/J)H
J
around discrete
eigenvalues of

H
(r)
.
Proposition 5.2. For every isolated eigenvalue E of

H
(r)
, there exists an in-
terval
I
J
= (E
J
, E +
J
) with lim
J
J
J
lnJ
= 0,
such that for any interval I around E s.t. I (

H
(r)
) = E:
s-lim
J
P
I\IJ
_
1
J
H
J
_
= 0,
s-lim
J
P
IJ
_
1
J
H
J
_
= P
{E}
_
H
(r)
_
.
Proof. Let
E
be the simultaneous eigenvector of

H
(r)
with eigenvalue E, and
of N
tot
with eigenvalue N
E
. It follows that
N
E

E

(r)
.
We have from Lemma 5.3
_
_
_
_
1
J
H
J
E
_
E
_
_
_
c(, r, E)
J
or,
lim
J
J
ln J
_
_
_
_
1
J
H
J
E
_
E
_
_
_ = 0.
Hence it follows that E is a rst order pseudo-eigenvalue with rst-order pseudo
eigenvector
E
, and the result follows from [20, Theorem XII.22]. 2
Remark 5.1. Applying the same reasoning to S
3
tot
, we nd that the interval (a, b)
in item (iii) of Corollary 5.2 can be chosen as
(a, b) = (
J
, +
J
)
with again lim
J

J
J(lnJ)
1
= 0.
5.3. Convergence of the spectrum with a pinning eld
To prove full convergence of the spectrum, we have to add the external eld
(2.4) to H
J
, or, to have a positive operator, add:
sup
x
_
[h
J,x
[J
h
J,x

S
x
_
= h
J
sup
x
_
J

S
3
x
_
= h
J
sup
x
(N
x
) (5.8)
with h
J
> 0, see also the end of Section 3.2, and N
x
= a
x
a
x
. Let us assume we
add this eld to (1/J)H
J
, so h
J
already contains the factor J
1
.
Take 0 < n
J
< J as before, we get that on (H
nJ
)
H
J
1
J
H
J
+h
J
sup
x
N
x
h
J
sup
x
N
x
h
J
n
J
1.
Clearly, by choosing h
J
such that
lim
J
h
J
n
J
=
statements about the spectrum on H
J
reduce to statements about the spectrum
on H
nJ
. Or, if one chooses to make statements about the spectrum below a
certain value E, it is sucient to choose h
J
such that lim
J
h
J
n
J
> E +.
Convergence of the spectrum of H
J,kin
+ H
J,dyn
can be proved under the
weakest assumptions on n
J
. First we prove convergence of the spectrum of
H
J,kin
.
Proposition 5.3. Let n
J
= [J(lnJ)
1
], where [] denotes the integer part. If
/ (

H
(r)
), then /
nJ
(H
J,kin
Proof. Assume that
nJ
(H
J,kin
) for all J larger than some J
0
. Take > 0
arbitrary, and
J
H
nJ
a normalized approximate eigenvector:
_
_
(H
J,kin
)
J
_
_
< .
More precisely, take
J
such that
P
(,+)
(H
J,kin
)
J
=
J
.
Since [H
J,kin
, N
tot
] = 0 we can take
J
an eigenstate of N
tot
with eigenvalue
N
J
. Since
J
is orthogonal to the ground state space, it follows from Lemma 5.2
that
N
J

+

(r)
g
J
(2n
J
) J
1
.
By our choice of n
J
, we have lim
J
n
J
J
1
= 0, or lim
J
g
J
(2n
J
) = 1, and for J
large enough,
N
J

2( +)

(r)
.
Putting this into the bounds of Lemma 5.3, we nd
_
_
(

H
(r)
)
J
_
_
_
_
(

H
(r)
H
J,kin
)
J
_
_
+
_
_
(H
J,kin
)
J
_
_
4(1 +
1
)( +)

(r)
n
J
J
+
and it follows that
J
is an approximate eigenvector for

H
(r)
as well and
(

H
(r)
). 2
Now we add H
J,dyn
:
Proposition 5.4. Let again n
J
= [J(lnJ)
1
]. If / (

H
(r)
), then /
nJ
(H
J,kin
+H
J,dyn
Proof. Since H
J,dyn
0, it follows that an approximate eigenvector
J
for
H
J,kin
+H
J,dyn
must satisfy
_
_
H
J,kin
J
_
_
+
with the same notation as in the previous proposition. Hence we get the same
estimate on N
J
as before, but this implies by Lemma 5.3 that
lim
J
|H
J,dyn
J
| = 0
and
J
is an approximate eigenvector for

H
(r)
as well. 2
Alternatively, these propositions prove that the spectrum of H
J,kin
+H
J,dyn
+
h
J
sup
x
N
x
converges to the spectrum of

H
(r)
, provided
lim
J
h
J
= 0, lim
J
h
J
J
lnJ
= .
To add H
J,tran
we clearly have to relax our condition on n
J
.
Proposition 5.5. Let n
J
= [(J(lnJ)
1
)
1/3
]. If / (

H
(r)
), then for J large
enough, /
nJ
((1/J)H
J
).
Proof. By Lemma 5.4 we have for all H
nJ
|H
J,tran
| 2
_
1
2
[v
0
[
1
_
(2n
J
+ 1)(4n
J
)
2
J
_
1/2
||
and by assumption the r.h.s. goes to 0 as J . 2
Hence we get convergence of the spectrum of (1/J)H
J
+h
J
sup
x
N
x
if
lim
J
h
J
= 0, lim
J
h
J
_
J
ln J
_
1/3
= .
The results obtained in this section include the statements made in Sec-
tion 2.3.
References
[1] F.T. Arecchi, E. Courtens, R. Gilmore and H. Thomas (1972) Atomic
coherent states in quantum optics. Phys. Rev. A 6 (6), 22112237.
[2] J. Conlon and J.P. Solovej (1990) On asymptotic limits for the quantum
Heisenberg model. J. Phys. A: Math. and Gen. 23, 31993213.
[3] O. Bolina, P. Contucci and B. Nachtergaele (2000) Path integral rep-
resentation for interface states of the anisotropic Heisenberg model. Rev. Math.
Phys. 12 (10), 13251344. math-ph/9908004.
[4] P. Caputo and F. Martinelli (2003) Relaxation time of anisotropic simple
exclusion processes and quantum Heisenberg models. Ann. Appl. Probab. 13,
691721. math.PR/0202025.
[5] P. Contucci, B. Nachtergaele and W. Spitzer (2002) The Ferromagnetic
Heisenberg XXZ chain in a pinning eld. Phys. Rev. B. 66, 064429113.
[6] F. Dyson (1956) General theory of spin-wave interactions. Phys. Rev. 102 (5),
12171230.
[7] F. Dyson (1956) Thermodynamic behavior of an ideal ferromagnet. Phys. Rev.
102 (5), 12301244.
[8] D. Goderis, A. Verbeure and P. Vets (1990) Dynamics of uctuations for
quantum lattice systems. Commun. Math. Phys. 128, 533549.
[9] D. Goderis and P. Vets (1989) Central limit theorem for mixing quantum
systems and the ccr-algebra of uctuations. Commun. Math. Phys. 122, 249
265.
[10] C.-T. Gottstein and R.F. Werner (1995) Ground states of the innite q-
deformed Heisenberg ferromagnet. Preprint cond-mat/9501123.
[11] T. Koma and B. Nachtergaele (1997) The spectral gap of the ferromagnetic
XXZ chain. Lett. Math. Phys. 40, 116.
[12] T. Koma and B. Nachtergaele (1998) The complete set of ground states of
the ferromagnetic XXZ chains. Adv. Theor. Math. Phys. 2, 533558.
[13] T. Koma, B. Nachtergaele and S. Starr (2001) The spectral gap for the
ferromagnetic spin-j XXZ chain. Adv. Theor. Math. Phys. 5, 10471090.
[14] E.H. Lieb (1973) The classical limit of quantum spin systems. Commun. Math.
Phys. 31, 327340.
[15] T. Michoel and A. Verbeure (1999) Mathematical structure of magnons in
quantum ferromagnets. J. Phys. A: Math. and Gen. 32, 58755883.
[16] T. Michoel and B. Nachtergaele (2005) In preparation.
[17] B. Nachtergaele (2001) Interfaces and droplets in quantum lattice models. In:
XIII International Congress of Mathematical Physics, A. Grigoryan, A. Fokas,
T. Kibble, B. Zegarlinski (eds.), International Press, Boston, 243249. arXiv:
math-ph/0009027.
[18] B. Nachtergaele, W. Spitzer and S. Starr (2004) Ferromagnetic ordering
of energy levels. J. Stat. Phys. 116, 719738.
[19] V. Pasquier and H. Saleur (1990) Common structures between nite systems
and conformal eld theories through quantum groups. Nucl. Phys. B 330, 523
556.
[20] M. Reed and B. Simon (1972) Methods of Modern Mathematical Physics, vol. I
IV. Academic Press, New York.
[21] S. Starr (2001) Some properties for the low-lying spectrum of the ferromagnetic,
quantum XXZ spin system. Ph. D. Thesis, Univ. of California, Davis. arXiv:math-
ph/0106024.
[22] J.L. van Hemmen, A.A.S. Brito and W.F. Wreszinski (1984) Spin waves in
quantum ferromagnets. J. Stat. Phys. 37 (1/2), 187213.

T. Michoel and B. Nachtergaele - The Large-Spin Asymptotics of The Ferromagnetic XXZ Chain

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

T. Michoel and B. Nachtergaele - The Large-Spin Asymptotics of The Ferromagnetic XXZ Chain

Hochgeladen von

Copyright:

Verfügbare Formate

Markov Processes Relat.

Fields 11, 237266 (2005)

is the generating vector for the ground state vectors

. The expectation in this state will be denoted

) is the product state of top states for the

Das könnte Ihnen auch gefallen