Magnetization field at criticality in the Ising model Christophe - - PowerPoint PPT Presentation

magnetization field at criticality in the ising model
SMART_READER_LITE
LIVE PREVIEW

Magnetization field at criticality in the Ising model Christophe - - PowerPoint PPT Presentation

Magnetization field at criticality in the Ising model Christophe Garban ENS Lyon and CNRS Federico Camia (Vrije Universiteit Amsterdam) Joint work with Charles Newman (NYU) and Percolation and Interacting Systems, MSRI, February 2012 C. Garban


slide-1
SLIDE 1

Magnetization field at criticality in the Ising model

Christophe Garban

ENS Lyon and CNRS Joint work with

Federico Camia (Vrije Universiteit Amsterdam)

and

Charles Newman (NYU) Percolation and Interacting Systems, MSRI, February 2012

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 1 / 25

slide-2
SLIDE 2
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 2 / 25

slide-3
SLIDE 3
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 2 / 25

slide-4
SLIDE 4

What are the fluctuations

  • f

the magnetiza- tion field ?

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 2 / 25

slide-5
SLIDE 5

N N

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 3 / 25

slide-6
SLIDE 6

N N

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 3 / 25

slide-7
SLIDE 7

N N

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

  • If the spins are i.i.d

1 N

σx → N(0, 1)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 3 / 25

slide-8
SLIDE 8

N N

  • If the spins are i.i.d

1 N

σx → N(0, 1)

1 N

σxδx → W, two- dimensional white noise

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 3 / 25

slide-9
SLIDE 9

N N

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 4 / 25

slide-10
SLIDE 10

N N

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

Ising model

To each configuration σ ∈ {−1, 1}N2, one associates the Hamiltonian

H(σ) := −

i∼j σiσj

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 4 / 25

slide-11
SLIDE 11

N N

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

Ising model

To each configuration σ ∈ {−1, 1}N2, one associates the Hamiltonian

H(σ) := −

i∼j σiσj

And we define:

Pβ(σ) ∝ e−β H(σ)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 4 / 25

slide-12
SLIDE 12

N N

  • If β < βc

1 N

σx → N(0, σ2

β) 1 N

σx δx → σβW

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 5 / 25

slide-13
SLIDE 13

N N

  • If β < βc

1 N

σx → N(0, σ2

β) 1 N

σx δx → σβW

  • If β > βc

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

N 2σ0− N 2σ0+

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 5 / 25

slide-14
SLIDE 14

N N

  • If β < βc

1 N

σx → N(0, σ2

β) 1 N

σx δx → σβW

  • If β > βc

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

σx−N 2E

  • σ0
  • N

→ N(0, σ2

β)

  • As β → βc, σβ ր ∞
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 5 / 25

slide-15
SLIDE 15

N N

  • If β < βc

1 N

σx → N(0, σ2

β) 1 N

σx δx → σβW

  • If β > βc

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

σx−N 2E

  • σ0
  • N

→ N(0, σ2

β)

  • As β → βc, σβ ր ∞

And now: what about β = βc ??

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 5 / 25

slide-16
SLIDE 16

Which normalization ?

To avoid boundary issues, consider our system on the torus Z2/NZ2. We have: Var(

  • σx)

=

  • x,y

E

  • σxσy
slide-17
SLIDE 17

Which normalization ?

To avoid boundary issues, consider our system on the torus Z2/NZ2. We have: Var(

  • σx)

=

  • x,y

E

  • σxσy

N2

y

E

  • σ0σy
slide-18
SLIDE 18

Which normalization ?

To avoid boundary issues, consider our system on the torus Z2/NZ2. We have: Var(

  • σx)

=

  • x,y

E

  • σxσy

N2

y

E

  • σ0σy

N2

y

|y|−1/4 (by Onsager) ≍ N2 N2N−1/4 = N

15 4

slide-19
SLIDE 19

Which normalization ?

To avoid boundary issues, consider our system on the torus Z2/NZ2. We have: Var(

  • σx)

=

  • x,y

E

  • σxσy

N2

y

E

  • σ0σy

N2

y

|y|−1/4 (by Onsager) ≍ N2 N2N−1/4 = N

15 4

Hence it is natural to look at the random variable m(N) :=

σx N15/8 .

Question

Does m(N) have a (unique) scaling limit ?

slide-20
SLIDE 20

“Subtle” issue of renormalization

Theorem (McCoy, Wu, 1967)

As N → ∞

  • σ0,0 σN,N
  • ∼ c N−1/4

where c = 21/12e3ζ′(−1)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 7 / 25

slide-21
SLIDE 21

“Subtle” issue of renormalization

Theorem (McCoy, Wu, 1967)

As N → ∞

  • σ0,0 σN,N
  • ∼ c N−1/4

where c = 21/12e3ζ′(−1)

Proposition (Rotational invariance of the two-point function)

  • σ0 σx
  • σ0 σx2

x2→∞ 1

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 7 / 25

slide-22
SLIDE 22

Notations / definitions

Rescaled lattice a Z2, a ≪ 1.

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

aZ2 1

Definition (Renormalized magnetization field)

Φa :=

  • x∈a Z2

δx σx a

15 8

ma := Φa, 1[0,1]2 ma

L := Φa, 1[0,L]2

slide-23
SLIDE 23

Notations / definitions

Rescaled lattice a Z2, a ≪ 1.

+ + + + + + + − − − − − − − − + − − − + + + − − − − − + + + + − − − − + − + + − + − − + + + − − − − − + − + + − + + − − + − − + + − − +

aZ2 1

Definition (Renormalized magnetization field)

Φa :=

  • x∈a Z2

δx σx a

15 8

ma := Φa, 1[0,1]2 ma

L := Φa, 1[0,L]2

Question

The field Φa ∈ D′. Is it the case that Φa converges as a → 0 to some random distribution Φ ?

slide-24
SLIDE 24

Road map

  • First part :

Theorem (Camia, G., Newman)

(i) The magnetization field Φa on a Z2 has a unique scaling limit as the mesh a → 0.

slide-25
SLIDE 25

Road map

  • First part :

Theorem (Camia, G., Newman)

(i) The magnetization field Φa on a Z2 has a unique scaling limit as the mesh a → 0. (ii) On a finite domain Ω with boundary conditions ξ, the magnetization field Φa

|Ω∩aZ2 has a scaling limit Φξ Ω.

slide-26
SLIDE 26

Road map

  • First part :

Theorem (Camia, G., Newman)

(i) The magnetization field Φa on a Z2 has a unique scaling limit as the mesh a → 0. (ii) On a finite domain Ω with boundary conditions ξ, the magnetization field Φa

|Ω∩aZ2 has a scaling limit Φξ Ω.

(iii) In particular, ma has a unique limiting law m = mξ which depends on the boundary conditions (usually +, −, or free).

slide-27
SLIDE 27

Road map

  • First part :

Theorem (Camia, G., Newman)

(i) The magnetization field Φa on a Z2 has a unique scaling limit as the mesh a → 0. (ii) On a finite domain Ω with boundary conditions ξ, the magnetization field Φa

|Ω∩aZ2 has a scaling limit Φξ Ω.

(iii) In particular, ma has a unique limiting law m = mξ which depends on the boundary conditions (usually +, −, or free). (iv) The scaling limit is NOT Gaussian.

slide-28
SLIDE 28

Road map

  • First part :

Theorem (Camia, G., Newman)

(i) The magnetization field Φa on a Z2 has a unique scaling limit as the mesh a → 0. (ii) On a finite domain Ω with boundary conditions ξ, the magnetization field Φa

|Ω∩aZ2 has a scaling limit Φξ Ω.

(iii) In particular, ma has a unique limiting law m = mξ which depends on the boundary conditions (usually +, −, or free). (iv) The scaling limit is NOT Gaussian.

  • Second part: properties of the limiting objects Φ and m

(i) Conformal covariance of ΦΩ (ii) Tail behavior of m etc ...

  • Applications (near-critical Ising model in the h-direction . . .)
slide-29
SLIDE 29

Behavior of the magnetization at βc for small external field

Theorem (Camia, G., Newman)

Consider Ising model on Z2 at βc with a positive external magnetic field h > 0, then σ0 ≍ h

1 15

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 10 / 25

slide-30
SLIDE 30

Behavior of the magnetization at βc for small external field

Theorem (Camia, G., Newman)

Consider Ising model on Z2 at βc with a positive external magnetic field h > 0, then σ0 ≍ h

1 15

This is the analog of (i) Onsager: σ0+

β ≍ |β − βc|1/8 as β → βc.

and (ii) Kesten - Smirnov/Werner: θ(p) = |p − pc|5/36+o(1) for site percolation on the triangular lattice.

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 10 / 25

slide-31
SLIDE 31

Tightness

  • Recall ma := a15/8
  • x∈[0,1]2∩a Z2σx.
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 11 / 25

slide-32
SLIDE 32

Tightness

  • Recall ma := a15/8
  • x∈[0,1]2∩a Z2σx.

Fact

lim sup

a→0

Eβc

  • (ma)2

< ∞

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 11 / 25

slide-33
SLIDE 33

Tightness

  • Recall ma := a15/8
  • x∈[0,1]2∩a Z2σx.

Fact

lim sup

a→0

Eβc

  • (ma)2

< ∞

  • Now, for the field Φa ∈ D′:

Fact

lim sup

a→0

E

  • Φa2

H−2

  • < ∞

Since BH−2(0, 1) is compact in H−3, one can take subsequential scaling limits Φak → Φ∗ in H−3.

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 11 / 25

slide-34
SLIDE 34

Tightness

  • Recall ma := a15/8
  • x∈[0,1]2∩a Z2σx.

Fact

lim sup

a→0

Eβc

  • (ma)2

< ∞

  • Now, for the field Φa ∈ D′:

Fact

lim sup

a→0

E

  • Φa2

H−2

  • < ∞

Since BH−2(0, 1) is compact in H−3, one can take subsequential scaling limits Φak → Φ∗ in H−3. It remains to characterize the subsequential scaling limits Φ∗

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 11 / 25

slide-35
SLIDE 35

Proof 1: use the breakthrough results by Smirnov

slide-36
SLIDE 36

Proof 1: use the breakthrough results by Smirnov

Theorem (Smirnov, Chelkak-Smirnov)

aZ2

≡ ≡ + a

slide-37
SLIDE 37

Proof 1: use the breakthrough results by Smirnov

Theorem (Smirnov, Chelkak-Smirnov)

aZ2 a → 0

≡ ≡ + a CLE3 ≈ SLE3

slide-38
SLIDE 38

Proof 1: use the breakthrough results by Smirnov

Theorem (Smirnov, Chelkak-Smirnov)

aZ2 a → 0

≡ ≡ + a CLE3 ≈ SLE3

Question

Can one recover the magnetization m = lima→0 ma as a functional of the CLE3 ?

slide-39
SLIDE 39

Use instead the scaling limit of FK percolation (q = 2)

Definition

Pp,q

  • ω
  • :=

1 Zp,q po(ω)(1 − p)c(ω) q#clusters

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 13 / 25

slide-40
SLIDE 40

Use instead the scaling limit of FK percolation (q = 2)

Definition

Pp,q

  • ω
  • :=

1 Zp,q po(ω)(1 − p)c(ω) q#clusters ma =

  • x∈aZ2∩[0,1]2

a15/8σx

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 13 / 25

slide-41
SLIDE 41

Use instead the scaling limit of FK percolation (q = 2)

Definition

Pp,q

  • ω
  • :=

1 Zp,q po(ω)(1 − p)c(ω) q#clusters ma =

  • x∈aZ2∩[0,1]2

a15/8σx =

  • x∈C1

a15/8σx +

  • x∈C2

a15/8σx + . . . = ξ1 Areaa(C1) + ξ2 Areaa(C2) + . . . where (ξi)i≥1 are i.i.d coin flips.

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 13 / 25

slide-42
SLIDE 42

Use instead the scaling limit of FK percolation (q = 2)

Definition

Pp,q

  • ω
  • :=

1 Zp,q po(ω)(1 − p)c(ω) q#clusters ma =

  • x∈aZ2∩[0,1]2

a15/8σx =

  • x∈C1

a15/8σx +

  • x∈C2

a15/8σx + . . . = ξ1 Areaa(C1) + ξ2 Areaa(C2) + . . . where (ξi)i≥1 are i.i.d coin flips.

Claim

Uniformly in a → 0, E

  • diam(Ci)≤ǫ

ξi Areaa(Ci) 2 ≤ C ǫ7/4

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 13 / 25

slide-43
SLIDE 43
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 14 / 25

slide-44
SLIDE 44

a Z2

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 14 / 25

slide-45
SLIDE 45

a Z2 ǫ

a ≪ ǫ ≪ 1

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 14 / 25

slide-46
SLIDE 46

a Z2 ǫ #points =

  • i

Xi ∼

  • i

Yi K ǫ a 15/

a ≪ ǫ ≪ 1

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 14 / 25

slide-47
SLIDE 47

a Z2 #points =

  • i

Xi ∼

  • i

Yi K ǫ a 15/

Bi Bj

E

  • X − K( ǫ

a)15/8Y 2 =

  • i,j

E

  • Xi − K( ǫ

a)15/8Yi Xj − K( ǫ a)15/8Yj

  • =
  • (E
  • X2

)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 14 / 25

slide-48
SLIDE 48

Proof 2: use the recent results by Dubédat or Chelkak, Hongler, Izyurov

slide-49
SLIDE 49

Proof 2: use the recent results by Dubédat or Chelkak, Hongler, Izyurov

Theorem (Dubédat or Chelkak, Hongler, Izyurov)

Ω z1 z2 z3

a Z2

slide-50
SLIDE 50

Proof 2: use the recent results by Dubédat or Chelkak, Hongler, Izyurov

Theorem (Dubédat or Chelkak, Hongler, Izyurov)

Ω z1 z2 z3

a Z2

There exist n-point correlation functions z1, . . . , zn → z1, . . . , zn+

Ω ,

slide-51
SLIDE 51

Proof 2: use the recent results by Dubédat or Chelkak, Hongler, Izyurov

Theorem (Dubédat or Chelkak, Hongler, Izyurov)

Ω z1 z2 z3

a Z2

There exist n-point correlation functions z1, . . . , zn → z1, . . . , zn+

Ω ,

so that as the mesh a → 0, and if the points z1, . . . , zn remain “macroscopically far apart”, a− n

8 E +

  • σza

1 . . . σza n

  • ∼ z1, . . . , zn+

slide-52
SLIDE 52

Proof 2: use the recent results by Dubédat or Chelkak, Hongler, Izyurov

Theorem (Dubédat or Chelkak, Hongler, Izyurov)

Ω z1 z2 z3

a Z2

There exist n-point correlation functions z1, . . . , zn → z1, . . . , zn+

Ω ,

so that as the mesh a → 0, and if the points z1, . . . , zn remain “macroscopically far apart”, a− n

8 E +

  • σza

1 . . . σza n

  • ∼ z1, . . . , zn+

This suggests that one should have for each n ≥ 1: E +

  • mn

a

a→0

  • z1, . . . , zn+

Ω dz1 . . . dzn

slide-53
SLIDE 53

In order to show that Φa converges to a limiting random distribution, two things thus remain to be proved:

slide-54
SLIDE 54

In order to show that Φa converges to a limiting random distribution, two things thus remain to be proved:

  • 1. Show that the nth moment E +

  • mn

a

  • indeed converges towards

µn :=

  • z1, . . . , zn+

Ω dz1 . . . dzn

as a → 0.

slide-55
SLIDE 55

In order to show that Φa converges to a limiting random distribution, two things thus remain to be proved:

  • 1. Show that the nth moment E +

  • mn

a

  • indeed converges towards

µn :=

  • z1, . . . , zn+

Ω dz1 . . . dzn

as a → 0.

  • 2. Show that the sequence of moments {µn}n≥1 uniquely characterizes a

probability law m (moment problem).

slide-56
SLIDE 56

In order to show that Φa converges to a limiting random distribution, two things thus remain to be proved:

  • 1. Show that the nth moment E +

  • mn

a

  • indeed converges towards

µn :=

  • z1, . . . , zn+

Ω dz1 . . . dzn

as a → 0.

  • 2. Show that the sequence of moments {µn}n≥1 uniquely characterizes a

probability law m (moment problem). The second property will follow from the following proposition:

Proposition (Camia, G., Newman)

For all fixed t > 0, one has sup

a>0

E +

  • et ma

< ∞

slide-57
SLIDE 57

Second part: properties of Φ and m

  • Conformal covariance properties
  • Tail estimates, i.e. P
  • m > x
  • for x ≫ 1.
  • Analyticity of the probability density function of m
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 17 / 25

slide-58
SLIDE 58

Conformal covariance

Theorem

(i) m[0,λ]2

(d)

≡ λ15/8 m[0,1]2

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 18 / 25

slide-59
SLIDE 59

Conformal covariance

Theorem

(i) m[0,λ]2

(d)

≡ λ15/8 m[0,1]2

Ω ˜ Ω f ω ˜ ω = f(ω)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 18 / 25

slide-60
SLIDE 60

Conformal covariance

Theorem

(i) m[0,λ]2

(d)

≡ λ15/8 m[0,1]2

Ω ˜ Ω f ω ˜ ω = f(ω)

(ii) f∗Φ(ω) and Φ(˜ ω) are a.c. Furthermore, ∀ φ ∈ C ∞

c (Ω)

˜ φ = φ ◦ f −1 ∈ C ∞

c (˜

Ω) , one has

  • Φ(˜

ω), ˜ φ

  • =
  • Φ(ω), φ(z) |f ′(z)|15/8
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 18 / 25

slide-61
SLIDE 61

Tail behavior: heuristics

Let’s try to understand P

  • m > x
  • . . .
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 19 / 25

slide-62
SLIDE 62

Tail behavior: heuristics

Let’s try to understand P

  • m > x
  • . . .

Fact

There is a universal constant C > 0 s.t. P

  • mξ > x
  • ≥ e−C x16
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 19 / 25

slide-63
SLIDE 63

Tail behavior: heuristics

Let’s try to understand P

  • m > x
  • . . .

Fact

There is a universal constant C > 0 s.t. P

  • mξ > x
  • ≥ e−C x16

This suggests

P

  • mξ > x
  • ≈ e−O(1)x16
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 19 / 25

slide-64
SLIDE 64

Tail behavior: main theorem

Theorem

There exists a universal constant c > 0 such that for any boundary conditions ξ around [0, 1]2, as x → ∞: log P

  • mξ > x
  • ∼ −c x16
slide-65
SLIDE 65

Tail behavior: main theorem

Theorem

There exists a universal constant c > 0 such that for any boundary conditions ξ around [0, 1]2, as x → ∞: log P

  • mξ > x
  • ∼ −c x16

Proof: Goes through the study of the moment generating function of mξ:

slide-66
SLIDE 66

Tail behavior: main theorem

Theorem

There exists a universal constant c > 0 such that for any boundary conditions ξ around [0, 1]2, as x → ∞: log P

  • mξ > x
  • ∼ −c x16

Proof: Goes through the study of the moment generating function of mξ:

Proposition

There exists a universal constant b > 0 such that for any boundary conditions ξ around [0, 1]2, one has as t → ∞: log E

  • et mξ

∼ −b t

16 15

slide-67
SLIDE 67

Tail behavior: main theorem

Theorem

There exists a universal constant c > 0 such that for any boundary conditions ξ around [0, 1]2, as x → ∞: log P

  • mξ > x
  • ∼ −c x16

(by a Tauberian Theorem) Proof: Goes through the study of the moment generating function of mξ:

Proposition

There exists a universal constant b > 0 such that for any boundary conditions ξ around [0, 1]2, one has as t → ∞: log E

  • et mξ

∼ −b t

16 15

COROLLARY: mξ is indeed not Gaussian.

slide-68
SLIDE 68

Why would m = lim ma have exponential moments ??

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-69
SLIDE 69

Why would m = lim ma have exponential moments ??

Theorem (GHS inequality, Griffiths, Hurst, Sherman, 1970)

Zβ,h :=

σ e−βH(σ)+h P σx is such that

∂3

h

  • log Zβ,h
  • ≤ 0
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-70
SLIDE 70

Why would m = lim ma have exponential moments ??

Theorem (GHS inequality, Griffiths, Hurst, Sherman, 1970)

Zβ,h :=

σ e−βH(σ)+h P σx is such that

∂3

h

  • log Zβ,h
  • ≤ 0

∂3

h

  • log(
  • e−βcH+h P σx)
  • ≤ 0
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-71
SLIDE 71

Why would m = lim ma have exponential moments ??

Theorem (GHS inequality, Griffiths, Hurst, Sherman, 1970)

Zβ,h :=

σ e−βH(σ)+h P σx is such that

∂3

h

  • log Zβ,h
  • ≤ 0

∂3

h

  • log(
  • e−βcH+h P σx)
  • ≤ 0

⇔ ∂3

h

  • log(

e−βcH+h P σx e−βcH )

  • ≤ 0
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-72
SLIDE 72

Why would m = lim ma have exponential moments ??

Theorem (GHS inequality, Griffiths, Hurst, Sherman, 1970)

Zβ,h :=

σ e−βH(σ)+h P σx is such that

∂3

h

  • log Zβ,h
  • ≤ 0

∂3

h

  • log(
  • e−βcH+h P σx)
  • ≤ 0

⇔ ∂3

h

  • log(

e−βcH+h P σx e−βcH )

  • ≤ 0

⇔ ∂3

h

  • log E
  • eh P σx

≤ 0

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-73
SLIDE 73

Why would m = lim ma have exponential moments ??

Theorem (GHS inequality, Griffiths, Hurst, Sherman, 1970)

Zβ,h :=

σ e−βH(σ)+h P σx is such that

∂3

h

  • log Zβ,h
  • ≤ 0

∂3

h

  • log(
  • e−βcH+h P σx)
  • ≤ 0

⇔ ∂3

h

  • log(

e−βcH+h P σx e−βcH )

  • ≤ 0

⇔ ∂3

h

  • log E
  • eh P σx

≤ 0 log E

  • eh P σx

≤ h

  • σx + h2

2 Var[

  • σx]
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 21 / 25

slide-74
SLIDE 74

log E

  • eh P σx

≤ h

  • σx + h2

2 Var[

  • σx]

With h := ta15/8, this gives sup

a>0

  • log E
  • etma

≤ sup

a>0

  • tE
  • ma

+ t2 2 Var[ma]

  • < ∞
slide-75
SLIDE 75

log E

  • eh P σx

≤ h

  • σx + h2

2 Var[

  • σx]

With h := ta15/8, this gives sup

a>0

  • log E
  • etma

≤ sup

a>0

  • tE
  • ma

+ t2 2 Var[ma]

  • < ∞

Using the estimate with ˜ t > t > 0, we get

Fact

E

  • etm

= lim

a→0 E

  • etma

< ∞

slide-76
SLIDE 76

Get a hand on the exponential moments

Main tool: use scaling ! log E

  • etm

= log E

  • et/λ15/8mλ
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 23 / 25

slide-77
SLIDE 77

Get a hand on the exponential moments

Main tool: use scaling ! log E

  • etm

= log E

  • et/λ15/8mλ

= log E

  • emλt

( with λt := t8/15)

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 23 / 25

slide-78
SLIDE 78

Get a hand on the exponential moments

Main tool: use scaling ! log E

  • etm

= log E

  • et/λ15/8mλ

= log E

  • emλt

( with λt := t8/15) = λ2

t

1 λ2

t

log E

  • emλt
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 23 / 25

slide-79
SLIDE 79

Get a hand on the exponential moments

Main tool: use scaling ! log E

  • etm

= log E

  • et/λ15/8mλ

= log E

  • emλt

( with λt := t8/15) = λ2

t

1 λ2

t

log E

  • emλt

∼ λ2

t

lim

L→∞

1 L2 log E

  • emL
  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 23 / 25

slide-80
SLIDE 80

Get a hand on the exponential moments

Main tool: use scaling ! log E

  • etm

= log E

  • et/λ15/8mλ

= log E

  • emλt

( with λt := t8/15) = λ2

t

1 λ2

t

log E

  • emλt

∼ λ2

t

lim

L→∞

1 L2 log E

  • emL

∼ t16/15C

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 23 / 25

slide-81
SLIDE 81

Analyticity of the density function of m

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 24 / 25

slide-82
SLIDE 82

Analyticity of the density function of m

Theorem

For all t ∈ R and any boundary condition ξ, one has |E ξ ei t m | ≤ e−c |t|

16 15 .

In particular, the density function f = fΩ is an entire function on the whole plane C !

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 24 / 25

slide-83
SLIDE 83

Further applications

A first one:

Theorem

Consider Ising model on Z2 at βc with a positive external magnetic field h > 0, then σ0 ≍ h

1 15

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 25 / 25

slide-84
SLIDE 84

Further applications

A first one:

Theorem

Consider Ising model on Z2 at βc with a positive external magnetic field h > 0, then σ0 ≍ h

1 15

A second one:

Theorem

Consider Ising model on aZ2 with external magnetic field h a15/8, then there is a scaling limit as a → 0 towards a massive near-critical model.

  • C. Garban (ENS Lyon and CNRS)

Magnetization field of the Ising model 25 / 25