HAL Id: hal-01213721
https://hal.archives-ouvertes.fr/hal-01213721
Submitted on 31 Oct 2015
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
An Ergodic Description of Ground States
Eduardo Garibaldi, Philippe Thieullen
To cite this version:
Eduardo Garibaldi, Philippe Thieullen. An Ergodic Description of Ground States. Journal of Statis-
tical Physics, Springer Verlag, 2015, 158, pp.359 - 371. �10.1007/s10955-014-1139-z�. �hal-01213721�
Eduardo Garibaldi
∗Departamento de Matem´ atica Universidade Estadual de Campinas
13083-859 Campinas - SP, Brasil
garibaldi@ime.unicamp.brPhilippe Thieullen
†Institut de Math´ ematiques Universit´ e Bordeaux 1, CNRS, UMR 5251
F-33405 Talence, France
Philippe.Thieullen@math.u-bordeaux1.fr
October 5, 2010
Abstract
1 Introduction
2 Framework and Main Results
Given a finite spin set F, we introduce the configuration space Ω = F
Zd. Its elements ω ∈ Ω are written as ω = {ω
j}
j∈Zd
. Endowed with the product topology of discrete topologies, Ω is a compact metrizable space. Indeed, for j = (j
1, . . . , j
d) ∈ Z
done may consider the L
1norm kjk = |j
1| + . . . + |j
d| and define then a metric on Ω by
d(ω, ω) = 2 ¯
−inf{kjk:ωj6=¯ωj}, which is compatible with the product topology.
Notice that Z
dacts on Ω by translation. Let {e
1, . . . , e
d} be the canonical basis for the lattice Z
d. For each i ∈ {1, . . . , d}, we consider the shift transformation θ
i: Ω → Ω given by θ
i(ω) = {ω
j+ei}
j∈Zd. Given j = (j
1, . . . , j
d) ∈ Z
d, we define θ
j:= θ
1j1◦ θ
j22◦ · · · ◦ θ
djd.
Let F denote the collection of finite subsets of Z
d. The diameter and the boundary of A ∈ F are respectively diam(A) = max{kj − kk : j, k ∈ A} and
∂A := {j ∈ Z
d\ A : kj − kk = 1 for some k ∈ A}. For A ∈ F and ω ∈ Ω, we denote the restriction ω|
Asimply by ω
A.
Definition 2.1. By an invariant interaction family we mean any collection of con- tinuous maps Φ
A: Ω → R , indexed by A ∈ F , such that
1. Φ
j+A(ω) = Φ
A(θ
j(ω)), for all j ∈ Z
dand ω ∈ Ω;
∗
supported by FAPESP 2009/17075-8
†
supported by ANR BLANC07-3 187245, Hamilton-Jacobi and Weak KAM Theory
1
2 Eduardo Garibaldi and Philippe Thieullen
2. ω
A= ¯ ω
Aimplies Φ
A(ω) = Φ
A( ¯ ω).
We say that {Φ
A}
A∈Fis an invariant short range interaction family if in addi- tion there exists an integer r > 0 such that Φ
A≡ 0 whenever diam(A) > r. In this case, we say also that the invariant interaction family has range r > 0. For r = 1, one in particular says to have an interaction of nearest neighbors.
Given an invariant interaction family {Φ
A}
A∈F, we introduce the associated hamiltonian H : F × Ω → R defined by
H(Λ, ω) = H
Λ(ω) := X
A:A∩Λ6=∅
Φ
A(ω), ∀ Λ ∈ F , ω ∈ Ω. (2.1)
Notice also that the hamiltonian inherits the invariance of the interaction family, namely, H
Λ◦ θ
j= H
j+Λfor any j ∈ Z
d.
Remark 2.2. We say that {Φ
A}
A∈Fis an invariant long range interaction family if Φ
A6≡ 0 for sets A with arbitrarily large diameter. In this case, one shall assume a summability condition:
X
0∈A
kΦ
Ak
∞< ∞,
which is obviously trivial for the short range situation. Notice that the associated hamiltonian is then well defined, since
kH
Λk
∞≤ X
j∈Λ
X
j∈A
kΦ
Ak
∞= X
j∈Λ
X
0∈A
kΦ
A◦ θ
jk
∞= #Λ X
0∈A
kΦ
Ak
∞< ∞, ∀ Λ ∈ F .
For ω, ω ¯ ∈ Ω and Λ ∈ F , we denote by ¯ ω
Λω
Zd\Λthe configuration of Ω that coincides with ¯ ω on Λ and with ω on Z
d\ Λ. We also introduce the following notation
Λ
n+1:= {−n, −n + 1, . . . , 0, . . . , n − 1, n}
d, ∀ n ∈ N .
Definition 2.3. We say that ω ∈ Ω is a minimizing configuration with respect to the hamiltonian H if
H
Λ(ω) ≤ H
Λ( ¯ ω
Λω
Zd\Λ), ∀ ω ¯ ∈ Ω, Λ ∈ F .
We define the minimizing ergodic value of the hamiltonian H as the constant H ¯ := inf
ω∈Ω
lim inf
n→∞
1
#Λ
nH
Λn(ω). (2.2)
Our main result concerning these minimizing objects is the following one.
Theorem 2.4. Let H be a hamiltonian defined by an invariant short range inter- action family. The set of minimizing configurations with respect to the hamiltonian H is a non-empty invariant closed set. For a minimizing configuration ω ∈ Ω, the limit lim
n→∞ 1#Λn
H
Λn(ω) always exists. Moreover, there are minimizing configu- rations ω ∈ Ω for which
n→∞
lim 1
#Λ
nH
Λn(ω) = ¯ H.
3 Proof of Theorem 2.4
The proof of Theorem 2.4 follows from a series of results regrouped in this section.
Proposition 3.1. Let H be a hamiltonian defined by an invariant short range interaction family. There exist minimizing configurations for the hamiltonian H.
The set of minimizing configurations with respect to H is invariant and closed.
Proof. For each Λ ∈ F, denote Ω
Λ:= n
ω ∈ Ω : H
Λ(ω) ≤ H
Λ( ¯ ω), ∀ ω ¯ ∈ Ω with ¯ ω
Zd\Λ= ω
Zd\Λo .
Notice that Ω
Λis non-empty: for any fixed configuration ω ∈ Ω, it clearly contains the minimum points of the map ¯ ω
Λ∈ F
Λ7→ H
Λ( ¯ ω
Λω
Zd\Λ) ∈ R . Besides, each Ω
Λis a closed set, since the map ω ∈ Ω 7→ H
Λ( ¯ ω
Λω
Zd\Λ) − H
Λ(ω) ∈ R is continuous (actually, locally constant) for all ¯ ω
Λ∈ F
Λ.
The family {Ω
Λ}
Λ∈Fis monotone: Λ ⊂ Λ
0⇒ Ω
Λ0⊂ Ω
Λ. Indeed, suppose that ω
0∈ Ω
Λ0. Let ¯ ω ∈ Ω be such that ¯ ω
Zd\Λ= ω
0Zd\Λ
. As Λ ⊂ Λ
0, we get
¯
ω
Zd\Λ0= ω
0Zd\Λ0
. Hence, H
Λ0(ω
0) ≤ H
Λ0( ¯ ω). Since H
Λ0= H
Λ+ X
A∩Λ06=∅
A∩Λ=∅
Φ
Aand Φ
A(ω
0) = Φ
A( ¯ ω) whenever ∅ 6= A ∩ Λ
0⊂ Λ
0\ Λ, we clearly obtain that H
Λ(ω
0) ≤ H
Λ( ¯ ω), which means that ω
0∈ Ω
Λ.
Therefore, the family of closed sets {Ω
Λ}
Λ∈Fhas the finite intersection property:
T
Ni=1
Ω
Λi⊃ Ω
SNi=1Λi
6= ∅. By compactness, there exists ω ∈ Ω
Λfor all Λ ∈ F.
The set of minimizing configurations with respect to H is exactly T
Λ∈F
Ω
Λ, which is clearly closed. Besides, since θ
−j(Ω
Λ) = Ω
j+Λfor all j ∈ Z
d, this set is also invariant.
Denote the Birkhoff sum by S
ΛΨ := P
j∈Λ
Ψ ◦ θ
j.
Definition 3.2. Given an invariant short range interaction family {Φ
A}
A∈F, we introduce
Ψ
0:= X
A: 0∈A
1
#A Φ
A.
Notice that Ψ
0is a real valued function since the above sum is actually finite.
Moreover, we have the following property.
Lemma 3.3. The map Ψ
0satisfies sup
Λ∈F
sup
ωΛ= ¯ωΛ
1
#∂Λ (S
ΛΨ
0(ω) − S
ΛΨ
0( ¯ ω)) < ∞.
In order to prove this lemma, given Λ ∈ F and r > 0, we define
Int
rΛ := {j ∈ Λ : d(j, ∂Λ) ≥ r} .
4 Eduardo Garibaldi and Philippe Thieullen
Proof. Notice that
S
ΛΨ
0= X
j∈Λ
X
A: 0∈A
1
#A Φ
A◦ θ
j= X
j∈Λ
X
A:j∈A
1
#A Φ
A= X
A:A∩Λ6=∅
X
j∈A∩Λ
1
#A Φ
A= X
A:A∩Λ6=∅
#(A ∩ Λ)
#A Φ
A. (3.1) Assume the invariant interaction family {Φ
A}
A∈Fhas range r > 0. Whenever ω
Λ= ¯ ω
Λ, it is easy to see that
S
ΛΨ
0(ω) − S
ΛΨ
0( ¯ ω) = X
A∩(Λ\IntrΛ)6=∅
#(A ∩ Λ)
#A (Φ
A(ω) − Φ
A( ¯ ω))
≤ 2 X
A∩(Λ\IntrΛ)6=∅
kΦ
Ak
∞≤ 2 #(Λ \ Int
rΛ) X
0∈A
kΦ
Ak
∞,
from which the statement follows immediately.
The relation of Ψ
0with the associated hamiltonian is given below.
Proposition 3.4. Suppose {Φ
A}
A∈Fis an invariant interaction family with range r > 0. If H is the associated hamiltonian, then
kH
Λ− S
ΛΨ
0k
∞≤ #(Λ \ Int
rΛ) X
0∈A
kΦ
Ak
∞.
Proof. Using (3.1), we obtain H
Λ− S
ΛΨ
0= X
A∩IntrΛ6=∅
1 − #(A ∩ Λ)
#A
Φ
A+ X
A∩(Λ\IntrΛ)6=∅
1 − #(A ∩ Λ)
#A
Φ
A.
Notice that if A ∈ F has diameter less than r and intersects Int
rΛ, then A ⊂ Λ.
Thus, the first term in the right side of the above equation is equal to zero. Hence, we get
kH
Λ− S
ΛΨ
0k
∞≤ X
A∩(Λ\IntrΛ)6=∅
kΦ
Ak
∞≤ #(Λ \ Int
rΛ) X
0∈A
kΦ
Ak
∞.
We remark that, for Λ
n= (−n, n)
d∩ Z
d, one has #(Λ
n\ Int
rΛ
n) ≤ C
dr
dn
d−1, for some constant C
d> 0 which depends just on the dimension d. In particular, we immediately obtain the following corollary.
Corollary 3.5. Let H be a hamiltonian defined by an invariant short range inter- action family. Then
H ¯ = inf
ω∈Ω
lim inf
n→∞
1
#Λ
nS
ΛnΨ
0(ω).
We may now prove the second statement of Theorem 2.4.
Proposition 3.6. Let H be a hamiltonian defined by an invariant short range interaction family. Then, for any minimizing configuration ω with respect to H, the limit
n→∞
lim 1
#Λ
nH
Λn(ω) does exist.
Proof. Let ω be a minimizing configuration for H. By Proposition 3.4, it is equiv- alent to show that the limit lim
n→∞ 1#Λn
S
ΛnΨ
0(ω) exists. Notice also that we may assume without loss of generality that Ψ
0≥ 0. So denote
L = lim inf
n→∞
1
#Λ
nS
ΛnΨ
0(ω) and, given > 0, consider a positive integer N such that
#Λ1N
S
ΛNΨ
0(ω) < L + . Consider integers m and ` with m ≥ 1 and ` = 0, 1, . . . , N − 1. Notice that, for suitable integers j
1, . . . , j
md∈ Λ
mN, we may write
S
ΛmN+`Ψ
0=
md
X
k=1
S
ΛNΨ
0◦ θ
jk+ X
j∈ΛmN+`\ΛmN
Ψ
0◦ θ
j. Define then ω
N,k:= θ
jk(ω)
jk+ΛN
ω
Zd\(jk+ΛN)and ¯ ω
N,k:= θ
−jk(ω
N,k). As ω is a minimizing configuration, one thus gets
H
jk+ΛN(ω) ≤ H
jk+ΛN(ω
N,k) = H
ΛN( ¯ ω
N,k).
Recall that kS
ΛNΨ
0◦ θ
j− H
j+ΛNk
∞≤ ΓN
d−1for every j ∈ Z
d, where Γ = Γ(d, r, {Φ
A}) = C
dr
dP
0∈A
kΦ
Ak
∞. Therefore, we have that S
ΛmN+`Ψ
0(ω) ≤
md
X
k=1
H
ΛN( ¯ ω
N,k) + m
dΓN
d−1+ X
j∈ΛmN+`\ΛmN
Ψ
0◦ θ
j(ω)
≤
md
X
k=1
S
ΛNΨ
0( ¯ ω
N,k) + 2m
dΓN
d−1+ 2
dh
(m + 1)
d− m
di
N
dkΨ
0k
∞. Notice that ( ¯ ω
N,k)
ΛN= ω
ΛN. We use now Lemma 3.3 to guarantee that there exists a constant κ > 0 such that S
ΛNΨ
0( ¯ ω
N,k) − S
ΛNΨ
0(ω) ≤ κ #∂Λ
N.
We obtain that S
ΛmN+`Ψ
0(ω) ≤
≤ m
d"
S
ΛNΨ
0(ω) + κ #∂Λ
N+ 2ΓN
d−1+ 2
d1 + 1 m
d− 1
!
N
dkΨ
0k
∞# ,
which yields 1
#Λ
mN+`S
ΛmN+`Ψ
0(ω) ≤
≤ 1
#Λ
NS
ΛNΨ
0(ω) + κ #∂Λ
N#Λ
N+ 2 Γ N +
1 + 1
m
d− 1
!
kΨ
0k
∞.
6 Eduardo Garibaldi and Philippe Thieullen
Hence, it follows that lim sup
n→∞
1
#Λ
nS
ΛnΨ
0(ω) ≤ L + + κ #∂Λ
N#Λ
N+ 2 Γ N .
Since the integer N can be taken arbitrarily large and > 0 can be chosen as close as one wants to zero, the proof is complete.
We will adopt an ergodic point of view. To that end, denote by M the set of Borel probability measures, equipped with the weak* topology. We consider the compact convex subset of invariant probabilities
M
θ:= n
µ ∈ M : µ ◦ θ
j= µ, ∀ j ∈ Z
do
. (3.2)
We have then the following characterization of the minimizing ergodic value ¯ H.
Proposition 3.7. Let H be a hamiltonian defined by an invariant short range interaction family. Then
H ¯ = min
µ∈Mθ
Z
Ω
Ψ
0(ω) dµ(ω).
Proof. By the ergodic decomposition theorem (see, for example, Theorem 2.3.3 in [3]), one may suppose that µ ∈ M
θis ergodic. Therefore, by Birkhoff’s ergodic theorem (see, for instance, Theorem 2.1.5 in [3]), any configuration ω belonging to the support of µ satisfies
n→∞
lim 1
#Λ
nS
ΛnΨ
0(ω) = Z
Ω
Ψ
0(ω) dµ(ω).
So thanks to Corollary 3.5, we have that ¯ H ≤ inf
µ∈MθR Ψ
0dµ.
For > 0, consider a configuration ω
∈ Ω and an arbitrarily large integer n
> 0 such that
#Λ1n
S
ΛnΨ
0(ω
) < H ¯ + and define a Borel probability measure µ := 1
#Λ
nX
j∈Λn
δ
θj(ω)∈ M.
Let µ be any weak* accumulation probability for the family {µ
}
>0when goes to zero. Clearly by construction, R
Ω
Ψ
0(ω) dµ(ω) ≤ H. So in order to obtain the ¯ opposite inequality, it is enough to argue that µ is invariant. However, notice that, for every i = 1, 2, . . . , n and for all f ∈ C
0(Ω), one has
Z
(f ◦ θ
i− f ) dµ
≤ 1
#Λ
n2 #∂Λ
nkf k
∞→ 0 as n
→ ∞, which indeed shows the invariance of µ.
By a minimizing probability we mean an invariant probability µ that minimizes the average value R
Ψ
0dµ, namely, such that ¯ H = R
Ψ
0dµ. Their existence is guaranteed by the previous proposition. Moreoveor, by the ergodic decomposition theorem, there always exist ergodic minimizing probabilities.
Recall now that a point at the support of an invariant probability is said to be
generic if it belongs to a subset of full measure. We may complete the proof of
Theorem 2.4 with the following result.
Theorem 3.8. Suppose {Φ
A}
A∈Fis an invariant interaction family with range r > 0. Then there are generic points ω at the support of any ergodic minimizing probability which are minimizing configurations for the associated hamiltonian H and satisfy
n→∞
lim 1
#Λ
nH
Λn(ω) = ¯ H.
Proof. For W ∈ F
ΛM, M ∈ N, consider the characteristic function χ
W: Ω → {0, 1}
which has value 1 at a point ¯ ω ∈ Ω if, and only if, ¯ ω
ΛM= W .
Let µ ∈ M
θbe an ergodic minimizing probability. Let us denote by b (φ) the subset of Ω of full measure for which Birkhoff’s ergodic theorem holds with respect to the integrable map φ : Ω → R . Obviously, any point
ω ∈ supp(µ) ∩ b (Ψ
0) ∩ \
W∈FΛM, M∈N
b (χ
W)
is generic and verify
n→∞
lim 1
#Λ
nH
Λn(ω) = lim
n→∞
1
#Λ
nS
ΛnΨ
0(ω) = Z
Ψ
0dµ = ¯ H.
Suppose on the contrary that ω is not a minimizing configuration. Hence, there shall exist ˜ ω ∈ Ω, ˜ N ∈ N and ˜ η > 0 such that
˜
ω
Zd\ΛN˜= ω
Zd\ΛN˜and H
Λ˜N
( ˜ ω) < H
Λ˜N
(ω) − η. ˜ Since in particular ω ∈ T
M∈N
b (χ
ωΛM), we have that
n→∞
lim 1
#Λ
nS
Λnχ
ωΛ ˜N+r
(ω) = µ [
¯ ω∈Ω
ω
Λ˜N+r
ω ¯
Zd\ΛN+r˜!
=: λ
N˜+r> 0.
Therefore, for n large enough, one guarantees that S
Λnχ
ωΛ ˜N+r
(ω) > #Λ
n λN+r˜2
. Denote A
n:= {j ∈ Λ
n: ω
ΛN+r˜ +j= ω
ΛN+r˜}. Let then B
n⊂ A
nbe a maximal subcollection of indices such that (Λ
N˜+r+ j) ∩ (Λ
N+r˜+ k) = ∅ whenever j, k ∈ B
n. Since for all j ∈ A
nthere must exist k ∈ B
nsuch that (Λ
N˜+r+ j) ∩ (Λ
N+r˜) 6= ∅, it follows that S
Λnχ
ωΛ ˜N+r
(ω) = #A
n≤ #B
n· #Λ
N+r˜, which yields for n large enough
1
#Λ
n#B
n> 1
#Λ
N+r˜λ
N˜+r2 .
Thus, for n sufficiently large, let us introduce the configuration ω
n∈ Ω as ω
nΛN˜+j
= ˜ ω
Λ˜N
for all j ∈ B
n, and ω
nZd\tj∈Bn(ΛN˜+j)
= ω
Zd\tj∈Bn(ΛN˜+j). From the construction, one gets that H
Λ˜N+j
(ω
n) = H
Λ˜N
( ˜ ω) for each j ∈ B
n. Notice then
8 Eduardo Garibaldi and Philippe Thieullen
that
H
Λn(ω
n) = X
j∈Bn
H
ΛN˜+j(ω
n) + X
A∩Λn6=∅, A∩tj∈Bn
(
ΛN˜+j)
=∅Φ
A(ω
n)
< #B
nH
ΛN˜(ω) − η ˜
+ X
A∩Λn6=∅, A∩tj∈Bn
(
ΛN˜+j)
=∅Φ
A(ω)
= H
Λn(ω) − η ˜ #B
n+ X
j∈Bn
H
Λ˜N
(ω) − H
Λ˜N
(θ
j(ω)) .
By the very definition of A
n, one obtains that H
Λ˜N
(ω) = H
Λ˜N