Rencontre de Probabilit´ es 2007 - Rouen
Directed Polymers and Uniform Integrability
AlainCamanes
[email protected] Universit´e de Nantes
May 29, 2007
Outline
1 The model
The polymer and its environment The energy
2 The partition function Definition
Properties
Uniform Integrability
3 2nd order moments General method Conditional method
4 α order moments
The uniform integrability condition Examples
Outline
1 The model
The polymer and its environment The energy
2 The partition function Definition
Properties
Uniform Integrability
3 2nd order moments General method Conditional method
4 α order moments
The uniform integrability condition
Le polym` ere et son milieu
n 0
Zd
ω1
the polymer
time
Example :g ∼Bern(1/2): g= 1 for a water particle,g=−1 for an oil particle.
g(i, x)
n time 0
Zd
ω1
g(1, ω1)
the environment
Le polym` ere et son milieu
Example :g ∼Bern(1/2): g= 1 for a water particle,g=−1 for an oil g(i, x)
n time 0
Zd
ω1
g(1, ω1)
the environment
The model’s parameters
The Hamiltonian of a polymer ω of lengthn Hn(ω) =
n
X
i=1
g(i, ωi).
The more the hydrophilic polymer visits water sites, the higherHn(ω) is.
The Boltzmann’s weight with temperatureT = 1/β eβHn(ω)
Zn(β)
Probability for a polymerωto have energyHn(ω) when the temperature isT
The model’s parameters
The Hamiltonian of a polymer ω of lengthn Hn(ω) =
n
X
i=1
g(i, ωi).
The more the hydrophilic polymer visits water sites, the higherHn(ω) is.
The Boltzmann’s weight with temperatureT = 1/β eβHn(ω)
Zn(β)
Hypotheses and notations
Notations
P uniform measure on polymers paths, Q law of the environmentg,
Gn=σ g(i,x), i ≤n, x ∈Zd .
Hypotheses
The random variables (gn,x) are i.i.d., They have exponential moments
λ(β):= lnQ h
eβg i
<+∞.
Hypotheses and notations
Notations
P uniform measure on polymers paths, Q law of the environmentg,
Gn=σ g(i,x), i ≤n, x ∈Zd .
Hypotheses
The random variables (gn,x) are i.i.d., They have exponential moments
λ(β):= lnQ h
eβg i
<+∞.
Outline
1 The model
The polymer and its environment The energy
2 The partition function Definition
Properties
Uniform Integrability
3 2nd order moments General method Conditional method
4 α order moments
The uniform integrability condition Examples
The partition function
Definition
We callpartition functionof the system the quantity Wn(β) =P
h
eβHn(ω)−nλ(β) i
Questions : How doeslimnWn behave with respect to the temperatureβ? the dimension d?
The partition function
Definition
We callpartition functionof the system the quantity Wn(β) =P
h
eβHn(ω)−nλ(β) i
Questions : How doeslimnWn behave with respect to the temperatureβ? the dimension d?
Properties of the partition function
Convergence [Bolthausen 89]: There exists a random variable W∞(β)such that
Wn(β) −→
n→∞W∞(β) p.s.
0−1 law [Bolthausen 89] :
P(W∞>0)∈{0,1}. Decreasingness [Comets, Yoshida 03] :
β 7→Qhp
W∞(β)i
isdecreasing. Critical temperature : ∃βc ∈[0,+∞] such that
βc = sup{β; W∞>0} ⇔ weak disorder
Properties of the partition function
Convergence [Bolthausen 89]: There exists a random variable W∞(β)such that
Wn(β) −→
n→∞W∞(β) p.s. 0−1 law [Bolthausen 89] :
P(W∞>0)∈{0,1}.
Decreasingness [Comets, Yoshida 03] : β 7→Qhp
W∞(β)i
isdecreasing. Critical temperature : ∃βc ∈[0,+∞] such that
βc = sup{β; W∞>0} ⇔ weak disorder
Properties of the partition function
Convergence [Bolthausen 89]: There exists a random variable W∞(β)such that
Wn(β) −→
n→∞W∞(β) p.s. 0−1 law [Bolthausen 89] :
P(W∞>0)∈{0,1}.
Decreasingness [Comets, Yoshida 03] : β 7→Qhp
W∞(β)i
isdecreasing.
Critical temperature : ∃βc ∈[0,+∞] such that βc = sup{β; W∞>0} ⇔ weak disorder
Properties of the partition function
Convergence [Bolthausen 89]: There exists a random variable W∞(β)such that
Wn(β) −→
n→∞W∞(β) p.s. 0−1 law [Bolthausen 89] :
P(W∞>0)∈{0,1}.
Decreasingness [Comets, Yoshida 03] : β 7→Qhp
W∞(β)i
isdecreasing.
Critical temperature : ∃βc ∈[0,+∞] such that βc = sup{β; W∞>0} ⇔ weak disorder
Motivation, Results, Prospects,. . .
Why studying the partition function?
Wn=e−nλ(β) Z
enβy νn(dy), whereνn=P
δHn/n −→ Large Deviations??
In dimension 1, 2 [Carmona, Hu 02; Comets, Yoshida 03]
βc = 0
In dimension d ≥3
cf. the rest of the talk. . .
Motivation, Results, Prospects,. . .
Why studying the partition function?
Wn=e−nλ(β) Z
enβy νn(dy), whereνn=P
δHn/n −→ Large Deviations??
In dimension 1, 2 [Carmona, Hu 02; Comets, Yoshida 03]
βc = 0
In dimension d ≥3
cf. the rest of the talk. . .
Motivation, Results, Prospects,. . .
Why studying the partition function?
Wn=e−nλ(β) Z
enβy νn(dy), whereνn=P
δHn/n −→ Large Deviations??
In dimension 1, 2 [Carmona, Hu 02; Comets, Yoshida 03]
βc = 0
In dimension d ≥3
cf. the rest of the talk. . .
Uniform Integrability
Theorem (Carmona, Hu 02)
The following properties are equivalent W∞>0 a.s.
(Wn)n∈N is uniformly integrable.
Plan :
Evaluate 2nd order moments
Evaluateα order moments,α ∈(1,2]
Uniform Integrability
Theorem (Carmona, Hu 02)
The following properties are equivalent W∞>0 a.s.
(Wn)n∈N is uniformly integrable.
Plan :
Evaluate 2nd order moments
Evaluateα order moments,α ∈(1,2]
Outline
1 The model
The polymer and its environment The energy
2 The partition function Definition
Properties
Uniform Integrability
3 2nd order moments General method Conditional method
4 α order moments
The uniform integrability condition Examples
2
ndorder moments
ω1,ω2 independant random walks,
qd =P⊗2 ∀n∈N, ωn16=ω2n .
Theorem (Bolthausen 89)
As soon asλ(2β)−2λ(β)<−ln(1−qd), W∞(β)>0.
Notation :β2∈[0,+∞] the critical value (β2≤βc)
Bernoullienvironment : λ(β) = lnch(β), β2= +∞. Gaussianenvironment : λ(β) =β2/2,
β2=p
−2 ln(1−qd).
2
ndorder moments
ω1,ω2 independant random walks,
qd =P⊗2 ∀n∈N, ωn16=ω2n .
Theorem (Bolthausen 89)
As soon asλ(2β)−2λ(β)<−ln(1−qd), W∞(β)>0.
Notation :β2∈[0,+∞] the critical value (β2≤βc)
Bernoullienvironment : λ(β) = lnch(β), β2= +∞.
Gaussianenvironment : λ(β) =β2/2, β2=p
−2 ln(1−qd).
Conditional method : first steps
Definition
LetWn, we define
The sized-biasedvariableWcn : ∀ f ∈ Bb, Qh
f(cWn)i
=Q[Wnf(Wn)]. The fWn variable,
Wfn=P h
eβ(H(ω1)+H(ω2)) i
e−2nλ(β).
The following properties are equivalent
(Wn) U.I. ⇔ L(cWn) tight
⇔ L(fWn) tight
Conditional method : first steps
Definition
LetWn, we define
The sized-biasedvariableWcn : ∀ f ∈ Bb, Qh
f(cWn)i
=Q[Wnf(Wn)]. The fWn variable,
Wfn=P h
eβ(H(ω1)+H(ω2)) i
e−2nλ(β).
The following properties are equivalent
(Wn) U.I. ⇔ L(cWn) tight
⇔ L(fWn) tight
Conditional methods : problems
Identify
β∗= supn β, Qh
Wfn
i
<+∞a.s.o
Idea (Birkner 02)
Use aconditionalSanov type theorem to show λ(2β∗)−2λ(β∗) = 1 +X
n≥1
e−H(pn),
where H(pn) =−P
x∈ZdP(ωn=x) lnP(ωn=x).
Then, using Jensen’s inequality,β∗> β2.
Probleme : The large deviation principle used is not yet proven.
Conditional methods : problems
Identify
β∗= supn β, Qh
Wfn
i
<+∞a.s.o
Idea (Birkner 02)
Use aconditionalSanov type theorem to show λ(2β∗)−2λ(β∗) = 1 +X
n≥1
e−H(pn),
where H(pn) =−P
x∈ZdP(ωn=x) lnP(ωn=x).
Then, using Jensen’s inequality,β∗> β2.
Probleme : The large deviation principle used is not yet proven.
Conditional methods : problems
Identify
β∗= supn β, Qh
Wfn
i
<+∞a.s.o
Idea (Birkner 02)
Use aconditionalSanov type theorem to show λ(2β∗)−2λ(β∗) = 1 +X
n≥1
e−H(pn),
where H(pn) =−P
x∈ZdP(ωn=x) lnP(ωn=x).
Then, using Jensen’s inequality,β∗> β2.
Probleme : The large deviation principle used is not yet proven.
Outline
1 The model
The polymer and its environment The energy
2 The partition function Definition
Properties
Uniform Integrability
3 2nd order moments General method Conditional method
4 α order moments
The uniform integrability condition Examples
Notations and results
∃α∈(1,2] supnQ[Wnα]<+∞ ⇒ (Wn)n∈N is U.I.
ω1,ω2independant random walks starting from 0 : p(t,x)=P⊗2“
0<j<t, ω1j 6=ω2j, ωt1=ωt2=x” .
Theorem (Derrida, Evans 92) Ifλ(αβ)−αλ(β)<−lnP
t,xp(t,x)α/2, then W∞>0.
Notation :βα the critical value.
Notations and results
∃α∈(1,2] supnQ[Wnα]<+∞ ⇒ (Wn)n∈N is U.I.
ω1,ω2independant random walks starting from 0 : p(t,x)=P⊗2“
0<j<t, ω1j 6=ω2j, ωt1=ωt2=x” .
Theorem (Derrida, Evans 92) Ifλ(αβ)−αλ(β)<−lnP
t,xp(t,x)α/2, then W∞>0.
Notation :βα the critical value.
Notations and results
∃α∈(1,2] supnQ[Wnα]<+∞ ⇒ (Wn)n∈N is U.I.
ω1,ω2independant random walks starting from 0 : p(t,x)=P⊗2“
0<j<t, ω1j 6=ω2j, ωt1=ωt2=x” .
Theorem (Derrida, Evans 92) Ifλ(αβ)−αλ(β)<−lnP
t,xp(t,x)α/2, then W∞>0.
How better is this method?
We define
( hQ(2) = Q h eβ2g
Q[eβ2g]ln eβ2g
Q[eβ2g]
i , hν(2) = −P
t,x p(t,x) 1−qd lnp(t,x)1−q
d.
Theorem (C., Carmona 07) There existsα such that βα > β2 if
hν(2)<hQ(2).
Remark : The optimalβα∗ satisfies
α∗βα∗λ0(α∗βα∗)−λ(α∗βα∗) =hν(α∗).
How better is this method?
We define
( hQ(2) = Q h eβ2g
Q[eβ2g]ln eβ2g
Q[eβ2g]
i , hν(2) = −P
t,x p(t,x) 1−qd lnp(t,x)1−q
d.
Theorem (C., Carmona 07) There existsα such that βα > β2 if
hν(2)<hQ(2).
Remark : The optimalβα∗ satisfies
α∗βα∗λ0(α∗βα∗)−λ(α∗βα∗) =hν(α∗).
How better is this method?
We define
( hQ(2) = Q h eβ2g
Q[eβ2g]ln eβ2g
Q[eβ2g]
i , hν(2) = −P
t,x p(t,x) 1−qd lnp(t,x)1−q
d.
Theorem (C., Carmona 07) There existsα such that βα > β2 if
hν(2)<hQ(2).
Remark : The optimalβα∗ satisfies
α∗βα∗λ0(α∗βα∗)−λ(α∗βα∗) =hν(α∗).
Some examples
Approximated values ofhν(2)
d hν(2) 3 4,6 4 3,8 5 3,6
Values ofhQ(2) for some environments
d Gaussian Binomial Poissonian
3 2,16 4,14 6,42
4 3,29 6,23 10,30
5 4,00 7,42 12,73
Inred the environments whereβ∗ > β2.
Some examples
Approximated values ofhν(2)
d hν(2) 3 4,6 4 3,8 5 3,6 Values ofhQ(2) for some environments
d Gaussian Binomial Poissonian
3 2,16 4,14 6,42
4 3,29 6,23 10,30
5 4,00 7,42 12,73
Inred the environments whereβ∗ > β2.
Conclusion
In general,β26=βc.
How to obtain a better bound on βc?
Find a link between the strong, the weak disorder phases and the large deviation rate function.
What is the link with the free energy?
pn(β) = 1
nlnWn(β)