a

r
X
i
v
:
m
a
t
h
-
p
h
/
0
0
0
1
0
1
8
v
1


1
3

J
a
n

2
0
0
0
Propagation of Molecular Chaos by Quantum
Systems and the Dynamics of the Curie-Weiss
Model
Alexander Gottlieb
Abstract
The propagation of molecular chaos, a tool of classical kinetic the-
ory, is generalized to apply to quantum systems of distinguishable par-
ticles. We prove that the Curie-Weiss model of ferromagnetism propa-
gates molecular chaos and derive the eective dynamics of a single-spin
state in the mean-eld limit. Our treatment diers from the traditional
approach to mean-eld spin models in that it concerns the dynamics of
single-particle states instead of the dynamics of innite-particle states.
1 Introduction
The innite-particle dynamics of spin models with nite-range interactions
such as the Ising model can be dened without diculty in the norm
limit of the local (nite-particle) dynamics [21][Section 7.6]. For mean-eld
spin models such as the Curie-Weiss model, the innite-particle dynamics
can only be dened in certain representations of the innite-particle algebra,
as the limit in the strong operator topology of the local dynamics [6, 1]. The
purpose of this note is to introduce a new approach to the quantum mean-
eld dynamics via the propagation of quantum molecular chaos. The concept
of quantum molecular chaos enables us to comprehend the innite-particle
limit of the Curie-Weiss dynamics without constructing an innite-particle
dynamics.
For classical mean-eld systems, the theory of the propagation of molec-
ular chaos enables one to study the eective dynamics of nite groups of par-
ticles without dening dynamics of innite particle states. We can achieve
the same end in the quantum context by utilizing the analog for quantum
systems of theory of the propagation of molecular chaos. This device is ex-
ploited in [19], where a quantum version of propagation of molecular chaos is
1
used to derive the Vlaosov equation from the dynamics of quantum particles
in the continuum. Their approach was inspired by [3], wherein the Vlasov
equation was derived from the propagation of molecular chaos by mean-eld
systems of classical particles.
The concept of molecular chaos dates back to Boltzmann [2]. In order to
derive the fundamental equation of the kinetic theory of gases, Boltzamnn
assumed that the molecules of a nonequilibrium gas were in a state of molec-
ular disorder. Nowadays, the term molecular chaos connotes a system
of classical particles that may be regarded as having stochastically indepen-
dent and identically distributed random positions and momenta. The state
of any molecularly chaotic system is characterized by the probability law of
a single particle of the system, and so the temporal evolution of any system
that is at all times in a state of molecular chaos reduces to that of the prob-
ability law of a single particle. Both Boltzmanns equation for dilute gases
and Vlasovs equation for dilute plasmas may be interpreted as equations
that describe the dynamics of the position-velocity distribution f(x, v)dxdv
of a single particle in a molecularly chaotic gas or plasma. Because gases
and plasmas remain in a molecularly chaotic state once they have entered
one, i.e., because they propagate molecular chaos, the kinetic equations
of Boltzmann and Vlasov can be thought of as evolution equations for a
single-particle distribution f(x, v)dxdv.
The concept of propagation of molecular chaos is due to Kac [11, 12], who
called it propagation of the Boltzmann property and used it to derive the
homogenous Boltzmann equation from the innite particle asymptotics of
certain Markovian gas models. This idea was further developed in work
by [9, 16, 26, 23]. McKean [14, 15] proved the propagation of chaos by
systems of interacting diusions. See [3, 24, 20, 5, 8] for more recent work
on and some generalizations of McKeans propagation of chaos. For two
good surveys of propagation of chaos and its applications, see [25, 17].
This paper is organized as follows. Quantum molecular chaos is dened
in Section 2, and related to classical molecular chaos. Examples of quan-
tum molecular chaos are provided; it is shown that sequences of canonical
states are often molecularly chaotic. In Section 3 we dene the propagation
of quantum molecular chaos. We then prove that the Curie-Weiss model
propagates molecular chaos and solve the mean-eld dynamical equation for
the single-particle state.
2
2 Quantum Molecular Chaos
The denition of molecular chaos current in the probability literature is
equivalent to the following [25, 7]:
Denition 1 Let S be a separable metric space. For each n N, let
n
be a symmetric probability measure on S
n
, the n-fold Cartesian power of
S. (Symmetric means that the measures of rectangles are invariant under
permutations of the coordinate axes.) Let be a probability measure on S.
The sequence
n
is -chaotic if the k-dimensional marginal distribu-
tions
(k)
n
converge (weakly) to
k
as n , for each xed k N.
The quantum analog of a probability measure is a state on a C*-algebra
with identity. A state on a C*-algebra with identity / is a positive linear
functional on / that equals 1 at the identity element. The space of states on
/ endowed with the weak* topology will be denoted o(/). Molecular chaos
is an attribute of certain sequences of symmetric probability measures; we
now dene quantum molecular chaos to be an attribute of certain sequences
of symmetric states.
Denition 2 (Quantum Molecular Chaos) Let / be a C*-algebra with
identity, and denote the n-th (spatial) tensor power of / by
n
/. Let be
a state on /. For each n N, let
n
be a symmetric state on
n
/, that is,
a state on
n
/ that satises

n
(A
1
A
n
) =
n
(A
(1)
/
(2)
A
(n)
)
for all permutations of 1, 2, . . . , n. For each k n, let
(k)
n
o(
k
/)
be dened by

(k)
n
(B) =
n
(B 1 1 1),
for all B
k
/, and let
k
be dened by the condition that, for all
A
1
, A
2
, . . . , A
k
/,

k
(A
1
A
2
A
k
) = (A
1
)(A
2
) (A
k
).
The sequence
n
is -chaotic if, for each k N, the states
(k)
n
con-
verge weakly* to
k
in o(
k
/) as n .
The sequence
n
is molecularly chaotic if it is -chaotic for some
state on /.
3
Suppose / is the algebra generated by the observables for a single parti-
cle of a certain species. The algebra generated by all single-particle observ-
ables in a system of n distinguishable particles of the same species is
n
/,
and states on
n
/ correspond to statistical ensembles of those n-particle
systems. Quantum molecular chaos of a sequence of n-particle states ex-
presses a condition of quasi-independence of the particles when the number
of particles is very large.
Quantum molecular chaos is related to classical molecular chaos as fol-
lows:
Theorem 1 Let / be a C* algebra with identity 1 and for each n N let

n
be a symmetric state on
n
/. The following are equivalent:
(i) The sequence
n
is -chaotic in the sense of Denition 2.
(ii) For each pair of positive elements Q
0
and Q
1
satisfying Q
0
+ Q
1
= 1,
the sequence of probability measures P
n
on 0, 1
n
dened by
P
n
(j
1
, j
2
, . . . , j
n
) =
n
(Q
j
1
Q
j
2
Q
jn
),
is P-chaotic in the classical sense of Denition 1, where P is the probability
measure on 0, 1 dened by
P(j) = (Q
j
).
Proof:
It is clear from Denition 2 that (i) = (ii). The rest of this proof is
devoted to showing that (i) = (ii).
We rst establish the following claim: Let T(o(/)) denote the space of
regular Borel probability measures on the state space of /, endowed with
the weak* topology as the dual of C(o(/)). Let be any state on /. The
measure

, a point mass at , is is the only measure T(o(/)) such

that
o(/) : [(Q) (Q)[ = 0 (1)
for every > 0 and Q / with 0 Q 1. To prove this claim, rst note
that (1) holds for every element of / if it holds for those elements Q with
0 Q 1, since every element of / is a linear combination of such positive
elements. Since (1) holds for every Q /, the Borel measure is supported
on arbitrarily small basic open neigborhoods of o(/), whence it follows
that () = 1, since is a regular measure.
4
Next we dene a couple of homeomorphisms: Let

/ denote the in-

ductive limit of the spatial tensor products
n
/. A state

/ is called
symmetric if
(A
1
A
2
A
n
1 1 )
= (A
(1)
A
(2)
A
(k)
A
k+1
A
n
1 1 )
for all elements A
1
, A
2
, . . . , A
n
of /, and all permutations of 1, 2, . . . , k,
k n. Denote the space of symmetric states on

/ by o
sym
(

/).
Strmers theorem [22][Theorem 2.8] states that there exists an ane home-
omorphism from the space T(o(/)) of regular probability measures on
o(/) to o
sym
(

/), such that (

) =

. The classical precursor

of Strmers theorem, de Finettis theorem [10], states that there exists
an ane homeomorphism : T(T(0, 1)) T
sym
(0, 1

) such that
(
p
) = p

.
Now assume that condition (ii) holds.
Let be an arbitrary but xed state on /, and for each n, extend

n

n
/ to the state

n
=
n

in o(

/) Since o(

/) is weak* compact, every subsequence of

n
has
cluster points; condition (ii) will used to prove that

is the only cluster

point of
n
. It will follow that
n
converges to

, which implies that

n
is -chaotic.
Let o(

/) be any cluster point of

n
, the limit of the subse-
quence
n
k
. Because of the increasing symmetry of the
n
k
, the state is
symmetric: o
sym
(

/). Suppose that ,=

(this assumption will

lead to a contradiction). Then
1
() ,=

and we have shown that there

must exist 0 < Q < 1 and > 0 such that

1
() : [(Q) (Q)[ > 0. (2)
Set Q
0
= Q and Q
1
= 1 Q. Dene P : o(/) T(0, 1), mapping to
P

, by
P

(j) = (Q
j
) ; j 0, 1.
Dene P

: o(

/) T(0, 1

), mapping to P

, by
P

(x
1
, x
2
, . . . ) : x
1
= j
1
, . . . , x
n
= j
n
= (Q
j
1
Q
jn
1 ).
5
Condition (ii) implies that P

(
n
k
) (P

. Since P

is continuous
and
n
k
, it follows that
P

() = (P

. (3)
The composite map
1
P

is ane and continuous, and it maps

to
P
for every o(/). The map

P : T(o(/)) T(T(0, 1))
induced by P : o(/) T(0, 1) is also ane and continuous, and also
maps

to
P
for every o(/), so

P must equal
1
P

by the
Krein-Milman theorem. That is, the following diagram commutes:
T(o(/))

o
sym
(

/)

_P

T(T(0, 1))

T
sym
(0, 1

)
By equation (3) and the commutativity of the preceding diagram,

P(
1
()) =
1
(P

()) =
P
. (4)
But equation (2) implies that

P(
1
())p T(0, 1) : [p(0) P

(0)[ > 0,
and this contradicts (4).

Corollary 1 Let / be a C* algebra with identity, and for each n N let

n
be a symmetric state on
n
/. If
(2)
n
converges to then
n
is
-chaotic.
Proof:
Let Q
0
and Q
1
be any positive elements of / such that Q
0
+ Q
1
= 1,
and let P
n
and P be as in the statement of Theorem 1. The measures P
n
are symmetric, and P
(2)
n
converges to P P since
(2)
n
converges to .
This suces to imply that P
n
is P-chaotic [25]. The -chaos of
n
now
follows from Theorem 1.

The following theorem shows that sequences of canonical states for mean-
eld systems are often molecularly chaotic:
6
Theorem 2 Let V be an operator on C
d
such that V (xy) = V (y x) for
all x, y C
d
. Let V
n
1,2
denote the operator on
n
C
d
dened by
V
n
1,2
(x
1
x
2
x
n
) = V (x
1
x
2
) x
3
x
n
,
and for each i < j n, dene V
n
ij
as acting similarly on the i
th
and j
th
factors of each simple tensor. Dene the states
n
o(
n
C
d
) by

n
(A) =
1
Z
Tr
_
e
Hn
A
_
; H
n
=
1
n

i<j
V
n
ij
; Z = Tr
_
e
Hn
_
.
The sequence
n
is -chaotic if the density operator for is the unique
minimizer of the free energy
F[D] =
1
2
Tr((D D)V ) + Tr(Dlog D).
If V is positive denite then F has a unique minimizer.
Proof Sketch: This theorem is the quantum version of Theorems 2 and 4 of
[18]. The properties of classical entropy that Messer and Spohn used to prove
those theorems are equally true for the von Neumann entropy Tr(Dlog D)
of density operators, at least when D operates on C
d
. The necessary prop-
erties of von Neumann entropy are proved in [13].

3 The Curie-Weiss Model Propagates Chaos

A sequence of n-particle dynamics propagates chaos if molecularly chaotic
sequences of initial distributions remain molecularly chaotic for all time
under the n-particle dynamical evolutions.
In the classical context, the n-particle dynamics are Markovian. Accord-
ingly, in [7], we dened propagation of chaos in terms of Markov transition
functions:
Denition 3 For each n N, let K
n
: S
n
(S
n
) [0, ) [0, 1] be a
Markov transition function which commutes with permutations in the sense
that
K
n
(x, E, t) = K
n
( x, E, t)
for all permutations of the n coordinates of x and the points of E S
n
,
and for all t 0. (Here, (S
n
) denotes the Borel -eld of S
n
.)
7
The sequence K
n

n=1
propagates chaos if, for all t 0, the molec-
ular chaos of a sequence
n
entails the molecular chaos of the sequence
__
S
n
K
n
(x, , t)
n
(dx)
_
.
The quantum analog of a Markov transition function is a completely
positive unital map. A linear map : /
1
/
2
of C* algebras is completely
positive if, for each n N, the map from /
1
B(C
n
) to /
2
B(C
n
) that
sends AB to (A) B is positive [4]. Propagation of chaos is an attribute
of certain sequences of Markov transition functions; we now dene quantum
propagation of chaos to be an attribute of certain sequences of completely
positive unital maps.
Denition 4 (Propagation of Quantum Molecular Chaos) For each
n N, let
n
be a completely positive map from
n
/ to itself that xes the
unit 1 1
n
/ and which commutes with permutations, i.e., such
that

n
(A
(1)
A
(2)
A
(n)
) =
n
(A
1
A
2
A
n
) (5)
for all permutations of 1, 2, . . . , n, where denotes the operator on

n
/ dened by
(B
1
B
2
B
n
) = B
(1)
B
(2)
B
(n)
for all B
1
, B
2
, . . . , B
n
/.
The sequence
n
propagates chaos if the molecular chaos of a se-
quence of states
n
entails the molecular chaos of the sequence
n

n
.
Consider the case where / is the algebra generated by the observables
for a single particle of a certain species, so that
n
/ is the algebra generated
by all single-particle observables in a system of n distinguishable particles of
that species. For each n N, let the dynamics of the n-particle system be
given by a Hamiltonian operator H
n

n
/. In the Heisenberg version of
quantum dynamics, if A is the operator that corresponds to measurement
of a certain observable quantity at t = 0, the operator corresponding to the
measurement of the same quantity at time t > 0 equals e
iHnt/
Ae
iHnt/
.
The maps
n,t
:
n
/
n
/ dened by

n,t
(A) = e
iHnt/
Ae
iHnt/
(6)
are completely positive, and if they satisfy condition (5) one may ask whether
the sequence
n,t
propagates chaos.
8
We conjecture that chaos always propagates when the n-particle Hamil-
tonians H
n
are as follows: Let / = B(C
d
), the algebra of all bounded
operators on C
d
. The algebra
n
/ is isomorphic to B(
n
C
d
), and states
o(
n
/) correspond to density operators D

on B(
n
C
d
) via the equa-
tion (A) = Tr(D

A). Suppose that V is a Hermitian operator on C

d
C
d
that is symmetric in the sense that V (x y) = V (y x) for all x, y C
d
.
Let V
n
1,2
denote the operator on
n
C
d
dened by
V
n
1,2
(x
1
x
2
x
n
) = V (x
1
x
2
) x
3
x
n
,
and for each i < j n, dene V
n
ij
similarly (as acting on the i
th
and j
th
factors of each simple tensor). Dene the n-particle Hamiltonian H
n
as the
sum of the pair potentials V
n
ij
, with a 1/n scaling of the coupling constant:
H
n
=
1
n

i<j
V
n
ij
. (7)
Conjecture 1 The sequence
n,t
dened in equation (6) propagates chaos:
If
n
is -chaotic then
n

n,t
is
t
-chaotic, where the density
operator for
t
is the solution at time t of

t
D =
i

[V, D D]
(1)
D(0) = D

.
(8)
Here [V, D D]
(1)
denotes a contraction of [V, D D]: if y
i
is any or-
thonormal basis of C
d
and x C
d
,
[V, D D]
(1)
(x) =

i
< [V, D D](x y
i
), (x y
i
) > .
This conjecture will now be veried for the Curie-Weiss model of fer-
romagnetism. In this model, the ferromagnetic material is modelled by a
crystal in which the spin angular momentum of each atom is coupled to the
average spin and to an external magnetic eld. In case the applied magnetic
eld is directed along the z-axis, we may make the approximation that only
the z-components of the spins are coupled to each other and the external
eld. As in [6, 1], we consider the case of spin-
1
2
atoms, so that the space of
pure spin states of a single particle is C
2
, and the observables corresponding
9
to the measurement of the x, y and z components of spin are the Pauli spin
operators

x
=

2
_
0 1
1 0
_

y
=

2
_
0 i
i 0
_

z
=

2
_
1 0
0 1
_
.
The space of pure states of an n-spin system is
n
C
2
. For each i n, if A
is an operator on C, let A
i
denote the operator on
n
C
2
dened by
A
i
(v
1
v
2
v
n
) = v
1
A(v
i
) v
n
.
If A is Hermitian, A
i
corresponds to the measurement of the spin observable
A at the i
th
lattice site. The Hamiltonian for the n-site Curie-Weiss model
is
H
n
= J
n

i=1
_

z
i

j

z
j
n
_
H
n

i=1

z
i
=
1
n
n

i,j=1
_
J
z
i

z
j
H
z
i
_
,
where J is a positive coupling constant and H is another constant propor-
tional to the strength of the external magnetic eld.
Since
n
B(C
2
) is isomorphic to B(
n
C
2
), states
n
o(
n
B(C
2
)) cor-
respond to density operators D
n
in
n
C
2
. From denition (6) and the
fact that Tr(AB) = Tr(BA),

n

n,t
(A) = Tr(D
n
e
iHnt/
Ae
iHnt/
)
= Tr(e
iHnt/
D
n
e
iHnt/
A).
Therefore,
n

n,t
has the density operator
D
nn,t
= e
iHnt/
D
n
e
iHnt/
. (9)
For any state on B(C
2
), let D

denote the corresponding density op-

erator, and let
_
D
(t)

denote a 2 2 matrix that represents D

.
Theorem 3 The sequence of Hamiltonians H
n
propagates chaos. If
n

is a -chaotic sequence of states with [D

] =
_
a c
c d
_
, then for each t 0,

n

n,t
is (t)-chaotic, where
_
D
(t)

=
_
a ce
it(H+J(ad))
ce
it(H+J(ad))
d
_
.
10
Proof of Theorem 3:
If x 0, 1
n
for some n N, let ^(x) denote the number of 1s in x:
^(x) =
n

i=1
x
i
if x = (x
1
, x
2
, . . . , x
n
). If g : 0, 1
n
C is a symmetric function, then
g(x) depends on x only through ^(x), so that g(x) = g(y) if ^(x) = ^(y).
Lemma 1 For each n N, let f
n
: 0, 1
n
C be a symmetric function.
Suppose that
(a) there exists B < such that

s{0,1}
n [f
n
(s)[ B for all n, and
(b) there exists c C and f : 0, 1 C such that, for k = 0, 1, 2, . . . ,

z
1
,z
2
,... ,z
nk
f
n
(x
1
, . . . , x
k
, z
1
, . . . , z
nk
) cf(x
1
)f(x
2
) f(x
k
)
as n .
Then
(i) For all G C
C
([0, 1]),
lim
n

x{0,1}
n
f
n
(x)G
_
^(x)
n
_
= c G(f(1))
(ii) If c ,= 0 then 0 f(0), f(1) 1 and f(0) + f(1) = 1.
Proof of Lemma: Let
F
n
(j) =

N(x)=j
f
n
(x). (10)
Condition (b) implies that
lim
n

z
1
,z
2
,... ,z
nk
f
n
(1, 1, . . . , 1, z
1
, . . . , z
nk
) = c (f(1))
k
for all k N. Grouping the summands for which exactly j coordinates of
(1, 1, . . . , 1, z
1
, . . . , z
nk
) equal 1, we nd that
lim
n
n

j=k
F
n
(j)
_
n k
j k
_
_
n
j
_ = c (f(1))
k
. (11)
11
Now
n

j=k
F
n
(j)
_
n k
j k
_
_
n
j
_ =
n

j=k
F
n
(j)
j(j 1) (j k + 1)
n(n 1) (n k + 1)
,
while

j=k
F
n
(j)
j(j 1) (j k + 1)
n(n 1) (n k + 1)

n

j=0
F
n
(j)
_
j
n
_
k

j=0
[F
n
(j)[
_
_
k
n
_
n
+ max
jk
_

j(j 1) (j k + 1)
n(n 1) (n k + 1)

_
j
n
_
k

__
B
_
_
k
n
_
n
+ max
jk
_
_
j
n
_
k

_
j k + 1
n
_
k
__
B
__
k
n
_
n
+
1
n
k
max
jk
__
k
1
_
j
k1
(k 1) +
_
k
2
_
j
k2
(k 1)
2
+ + (k 1)
k
__
B
__
k
n
_
n
+
1
n
__
k
1
_
k +
_
k
2
_
k
2
+ + k
k
__
B
__
k
n
_
n
+
1
n
(1 + k)
k
_
,
and therefore
lim
n
n

j=k
F
n
(j)
_
n k
j k
_
_
n
j
_ = lim
n
n

j=0
F
n
(j)
_
j
n
_
k
. (12)
Condition (a) implies that

j
[F
n
(j)[ B for all n. This bound, equa-
tions (11) and (12), and the fact that any continuous function on [0, 1] can
be approximated uniformly by polynomials, imply that
lim
n
n

j=0
F
n
(j)G
_
j
n
_
= c G(f(1))
for all G C
C
([0, 1]). This establishes (i), in view of the denition (10) of
F
n
.
12
If c ,= 0, conclusion (i) implies that f(1) [0, 1]. Condition (b) for k = 0
and k = 1 states that
lim
n

z
1
,z
2
,... ,zn
f
n
(z
1
, z
2
, . . . , z
n
) = c
lim
n

z
1
,z
2
,... ,z
n1
f
n
(1, z
1
, . . . , z
n1
) = cf(1)
lim
n

z
1
,z
2
,... ,z
n1
f
n
(0, z
1
, . . . , z
n1
) = cf(0),
so it follows that f(0) + f(1) = 1 if c ,= 0. This establishes (ii), concluding
the proof of the lemma.

Let e
1
= (1, 0) and e
2
= (0, 1). For each n N, let
c
n
=
_
e
j
1
e
j
2
e
jn

n
C
2
[ j
1
, j
2
, . . . , j
n
1, 2
_
.
Denote by [A]
k
1
,... ,kn
j
1
,... ,jn
the matrix elements for an operator A on
n
C
2
relative
to the basis c
n
, that is, for j
1
, k
1
, . . . , j
n
, k
n
1, 2, let
[A]
k
1
,... ,kn
j
1
,... ,jn
= A(e
j
1
e
j
2
e
jn
), e
k
1
e
k
2
e
kn
) .
The sequence
n
is -chaotic. This implies that, for each k N,

z
1
,z
2
,... ,z
nk
{1,2}
[D
n
]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
[D

]
y
1
x
1
[D

]
y
2
x
2
[D

]
y
k
x
k
,
as n . To show that
n

n,t
is (t)-chaotic, it suces to show that
for all k N,

z
1
,z
2
,... ,z
nk
{1,2}
_
D
nn,t

y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk

_
D
(t)

y
1
x
1
_
D
(t)

y
2
x
2

_
D
(t)

y
k
x
k
(13)
as n . We proceed to verify (13).
The operator H
n
is diagonalized by the basis c
n
, and its diagonal entries
are
[H
n
]
k
1
,... ,kn
k
1
,... ,kn
=
1
n
n

r,s=1
(J(e
kr
)(e
ks
) H(e
kr
)) , (14)
13
where
(e
1
) = +

2
(e
2
) =

2
.
Abbreviating J(x)(y) H(x) by w(x, y), equations (9) and (14) imply
that
[D
nn,t
]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
=
exp
_
_
it
n
k

r,s=1
(w(y
r
, y
s
) w(x
r
, x
s
))
_
_
exp
_
H
it

n k
n
k

r=1
((x
r
) (y
r
))
_

[D]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
exp
_
J
2it
n
k

r=1
n

s=k+1
(z
s
)((x
r
) (y
r
))
_
.
Therefore,
lim
n

z
1
,... ,z
nk
{1,2}
_
D
nn,t

y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
= exp
_
H
it

r=1
((x
r
) (y
r
))
_

lim
n

z
1
,... ,z
nk
{0,1}
[D]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
exp
_
J
2it
n
k

r=1
n

s=k+1
(z
s
)((x
r
) (y
r
))
_
= exp
_
H
it

r=1
((x
r
) (y
r
))
_

lim
n

z
1
,... ,z
nk
{1,2}
[D]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
exp
_
J
2it

r=1
((x
r
) (y
r
))
1
n
n

s=k+1
(z
s
)
_
.
The last limit in the preceding equation may be calculated thanks to
Lemma 1. To apply the lemma, x x
1
, y
1
, . . . , x
k
, y
k
1, 2 and dene
f
n
: 1, 2 C; f
n
(z
1
, z
2
, . . . , z
n
) = [D]
y
1
,... ,y
k
,z
1
,... ,zn
x
1
,... ,x
k
,z
1
,... ,zn
G : [0, 1] C; G(s) = exp
_
Jit
k

r=1
((x
r
) (y
r
))(s (1 s))
_
.
14
The functions f
n
are symmetric and satisfy condition (b) of the lemma, with
f(1) = [D

]
e
1
e
1
f(2) = [D

]
e
2
e
2
c = [D

]
y
1
x
1
[D

]
y
2
x
2
[D

]
y
k
x
k
.
The f
n
also satisfy condition (a) of the lemma, for

[D]
y
1
,... ,y
k
,z
1
,... ,zn
x
1
,... ,x
k
,z
1
,... ,zn

1
2
_
[D]
x
1
,... ,x
k
,z
1
,... ,zn
x
1
,... ,x
k
,z
1
,... ,zn
+ [D]
y
1
,... ,y
k
,z
1
,... ,zn
y
1
,... ,y
k
,z
1
,... ,zn
_
by the positivity of D
n
, so that

z
1
,... ,z
nk
{1,2}

[D]
y
1
,... ,y
k
,z
1
,... ,zn
x
1
,... ,x
k
,z
1
,... ,zn

tr(D
n
) = 1.
Conclusion (i) of Lemma 1 now reveals that

z
1
,... ,z
nk
{1,2}
[D]
y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
exp
_
J
2it

r=1
((x
r
) (y
r
))
1
n
n

s=k+1
(z
s
)
_
converges to
[D

]
y
1
x
1
[D

]
y
2
x
2
[D

]
y
k
x
k
exp
_
Jit
_
[D

]
e
1
e
1
[D

]
e
2
e
2
_
k

r=1
((x
r
) (y
r
))
_
,
whence
lim
n

z
1
,... ,z
nk
{1,2}
_
D
nn,t

y
1
,... ,y
k
,z
1
,... ,z
nk
x
1
,... ,x
k
,z
1
,... ,z
nk
= exp
_
H
it

r=1
((x
r
) (y
r
))
_

[D

]
y
1
x
1
[D

]
y
2
x
2
[D

]
y
k
x
k
exp
_
Jit
_
[D

]
e
1
e
1
[D

]
e
2
e
2
_
k

r=1
((x
r
) (y
r
))
_
=
k

r=1
[D

]
yr
xr
exp
_
it ((x
r
) (y
r
))
_
H/ + J
_
[D

]
e
1
e
1
[D

]
e
2
e
2
___
.
15
This shows that
n

n,t
is (t)-chaotic, where
_
D
(t)

y
x
= [D

]
y
x
exp
_
it ((x) (y))
_
H/ + J
_
[D

]
e
1
e
1
[D

]
e
2
e
2
___
.
Finally, it may be veried that the density operators D
(t)
satisfy equa-
tion (8) of Conjecture 1.

4 Future Work
In future work, we hope to prove Conjecture 1, or at least to prove that the
mean-eld Heisenberg model propagates chaos. We shall also investigate the
propagation of chaos by open systems (coupled to thermal baths) and prove
the H-theorem for those processes.
5 Acknowledgements
I am indebted to William Arveson, Sante Gnerre, Lucien Le Cam, Marc
Rieel, and Georey Sewell for their advice and for their interest in this
research. The formulation and proof of Theorem 1 was especially inspired
by Elizabeth Kallman. I thank Alexandre Chorin for introducing me to
statistical physics and other topics in applied mathematics. This work was
supported in part by the Oce of Science, Oce of Advanced Scientic
Computing Research, Mathematical, Information, and Computational Sci-
ences Division, Applied Mathematical Sciences Subprogram, of the U.S.
Department of Energy, under Contract No. DE-AC03-76SF00098.
References
[1] F. Bagarello and G. Morchio. Dynamics of mean-eld spin models from
basic results in abstract dierential equations. Journal of Statistical
Physics 66: 849 - 866, 1992.
[2] L. Boltzmann. Lectures on Gas Theory. Dover Publications, New York,
1995.
[3] W. Braun and K. Hepp. The Vlasov dynamics and its uctuations in
the
1
n
limit of interacting classical particles. Communications in Math-
ematical Physics 56: 101-113, 1977.
16
[4] E. B. Davies. Quantum Theory of Open Systems. Academic Press, Lon-
don, 1976.
[5] D. Dawson and J. G artner. Large deviations from the McKean-Vlasov
limit for weakly interacting diusions. Stochastics 20: 247-308, 1987.
[6] G. G. Emch and H. J. F. Knops. Pure thermodynamical phases as
extremal KMS states. Journal of Mathematical Physics 11 (no. 10):
3008 - 3018, 1970.
[7] A. Gottlieb. Markov transitions and the propagation of chaos (PhD
Thesis). Lawrence Berkeley National Laboratory Report, LBNL-42839,
1998.
[8] C. Graham. McKean-Vlasov Ito-Skorohod equations, and nonlinear dif-
fusions with discrete jump sets. Stochastic Processes and their Applica-
tions 40: 69-82, 1992.
[9] F. A. Gr unbaum. Propagation of chaos for the Boltzmann equation.
Archive for Rational Mechanics and Analysis 42: 323-345, 1971.
[10] E. Hewitt and L. Savage. Symmetric measures on Cartesian products.
Transactions of the American Mathematical Society 80: 470-501, 1955.
[11] M. Kac. Foundations of kinetic theory. Proceedings of the Third Berke-
ley Symposium on Mathematical Statistics and Probability, Vol III. Uni-
versity of California Press, Berkeley, California, 1956.
[12] M. Kac. Probability and Related Topics in Physical Sciences. American
Mathematical Society, Providence, Rhode Island, 1976.
[13] O. E. Lanford III and D. W. Robinson. Mean entropy of states in quan-
tum statistical mechanics. Journal of Mathematical Physics 9 (7): 1120-
1125, 1968.
[14] H. P. McKean, Jr. A class of Markov processes associated with nonlinear
parabolic equations. Proceedings of the National Academy of Science 56:
1907-1911, 1966.
[15] H. P. McKean, Jr. Propagation of chaos for a class of nonlinear parabolic
equations. Lecture Series in Dierential Equations 7: 41-57. Catholic
University, Washington, D.C., 1967.
[16] H. P. McKean, Jr. Fluctuations in the kinetic theory of gases. Commu-
nications in Pure and Applied Mathematics 28: 435-455, 1975.
17
[17] S. Meleard. Asymptotic behavior of some interacting particle systems;
McKean-Vlasov and Boltzmann models. Lecture Notes in Mathematics,
1627. Springer-Verlag, Berlin, 1995.
[18] J. Messer and H. Spohn. Statistical mechanics of the isothermal Lane-
Emden equation. Journal of Statistical Physics 29 (3): 561-578, 1982.
[19] H. Narnhofer and G. Sewell. Vlasov hydrodynamics of a quantum me-
chanical model. Communications in Mathematical Physics 79: 9-24,
1981.
[20] K. Oelschl ager. A law of large numbers for moderately interacting diu-
sion processes. Zeitschrift f ur Wahrscheinlichkeitstheorie und verwandte
Gebiete 69: 279-322, 1985.
[21] D. Ruelle. Statistical Mechanics: Rigorous Results. Addison Wesley,
Redwood City, California, 1969.
[22] E. Strmer. Symmetric states of innite tensor products of C*-algebras.
Journal of Functional Analysis 3: 48-68, 1969.
[23] A. Sznitman.

Equations de type de Boltzmann, spatialement ho-
mog`enes. Zeitschrift f ur Wahrscheinlichkeitstheorie und verwandte Ge-
biete 66: 559-592, 1984.
[24] A. Sznitman. Nonlinear reecting diusion process, and the propaga-
tion of chaos and uctuations associated. Journal of Functional Analy-
sis 56: 311-336, 1984.
[25] A. Sznitman. Topics in propagation of chaos. Lecture Notes in Mathe-
matics, 1464. Springer-Verlag, Berlin, 1991.
[26] K. Uchiyama. Fluctuations of Markovian systems in Kacs caricature
of a Maxwellian gas. Journal of the Mathematical Society of Japan 35
(3): 477-499, 1983.
18