Part II Tadahisa Funaki Stochastic Interface Modelsfunaki/SF/LNM1869.pdf · 2005-11-29 · Stochastic Interface Models 111 can be formulated in the framework of classical limit theorems

Part II

Tadahisa Funaki

Stochastic Interface Models

Stochastic Interface Models

Tadahisa Funaki

Graduate School of Mathematical Sciences, The University of Tokyo, Komaba,Tokyo 153-8914, JAPAN, [email protected]

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1091.2 Quick Overview of the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1111.3 Derivation of Effective Interface Models from Ising Model . . . . . . . . 1131.4 Basic Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

2 ∇ϕ Interface Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

2.1 Height Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1162.2 Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1162.3 Equilibrium States (Gibbs Measures) . . . . . . . . . . . . . . . . . . . . . . . . . . 1182.4 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1222.5 Scaling Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1242.6 Quadratic Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

3 Gaussian Equilibrium Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3.1 Gaussian Systems in a Finite Region . . . . . . . . . . . . . . . . . . . . . . . . . . . 1263.2 Gaussian Systems on Z

d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1293.3 Massive Gaussian Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1333.4 Macroscopic Scaling Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

4 Random Walk Representationand Fundamental Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.1 Helffer-Sjostrand Representation and FKG Inequality . . . . . . . . . . . . 1424.2 Brascamp-Lieb Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1464.3 Estimates of Nash-Aronson’s Type and Long Correlation . . . . . . . . . 1484.4 Thermodynamic Limit and Construction of ∇ϕ-Gibbs Measures . . 1514.5 Construction of ϕ-Gibbs Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

5 Surface Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

T. Funaki: Stochastic Interface Models, Lect. Notes Math. 1869, 105–274 (2005)www.springerlink.com c© Springer-Verlag Berlin Heidelberg 2005

106 T. Funaki

5.1 Definition of Surface Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1545.2 Quadratic Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1565.3 Fundamental Properties of Surface Tension . . . . . . . . . . . . . . . . . . . . . 1575.4 Proof of Theorems 5.3 and 5.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1585.5 Surface Tension in one Dimensional Systems . . . . . . . . . . . . . . . . . . . . 162

6 Large Deviation and Concentration Properties . . . . . . . . . . . . 164

6.1 LDP with Weak Self Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1656.2 Concentration Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1696.3 LDP with Weak Self Potentials in one Dimension . . . . . . . . . . . . . . . . 1736.4 LDP for δ-Pinning in one Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . 1806.5 Outline of the Proof of Theorem 6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 1826.6 Critical LDP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

7 Entropic Repulsion, Pinning and Wetting Transition . . . . . . 191

7.1 Entropic Repulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1927.2 Pinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1967.3 Wetting Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

8 Central Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

9 Characterization of ∇ϕ-Gibbs Measures . . . . . . . . . . . . . . . . . . 206

9.1 ϕ-Dynamics on Zd and ∇ϕ-Dynamics on (Zd)∗ . . . . . . . . . . . . . . . . . 207

9.2 Stationary Measures and ∇ϕ-Gibbs Measures . . . . . . . . . . . . . . . . . . 2089.3 Proof of Theorem 9.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2109.4 Proof of Proposition 9.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2149.5 Uniqueness of ϕ-Gibbs Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

10 Hydrodynamic Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

10.1 Space-Time Diffusive Scaling Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21710.2 The Nonlinear PDE (10.4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22010.3 Local Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22310.4 Proof of Theorem 10.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22910.5 Surface Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

11 Equilibrium Fluctuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

12 Dynamic Large Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

12.1 Dynamic LDP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23612.2 Dynamic Rate Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23712.3 Relation to the Static LDP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

13 Hydrodynamic Limit on a Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

13.1 Dynamics on a Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23913.2 Hydrodynamic Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

Stochastic Interface Models 107

14 Equilibrium Fluctuation on a Walland Entropic Repulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

14.1 The Case Attached to the Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24114.2 The Case Away from the Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24214.3 Dynamic Entropic Repulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

15 Dynamics in Two Media and Pinning Dynamicson a Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

15.1 Dynamics in Two Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24315.2 Pinning Dynamics on a Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

16 Other Dynamic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

16.1 Stochastic Lattice Gas and Free Boundary Problems . . . . . . . . . . . . . 24916.2 Interacting Brownian Particles at Zero Temperature . . . . . . . . . . . . . 25016.3 Singular Limits for Stochastic Reaction-Diffusion Equations . . . . . . . 25416.4 Limit Shape of Random Young Diagrams . . . . . . . . . . . . . . . . . . . . . . . 25816.5 Growing Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262

Abstract. In these notes we try to review developments in the last decade of thetheory on stochastic models for interfaces arising in two phase system, mostly on theso-called ∇ϕ interface model. We are, in particular, interested in the scaling limitswhich pass from the microscopic models to macroscopic level. Such limit proceduresare formulated as classical limit theorems in probability theory such as the law oflarge numbers, the central limit theorem and the large deviation principles.

Key words: Random interfaces, Effective interfaces, Phase coexistence and sepa-ration, Ginzburg-Landau model, Massless model, Random walk representation, Sur-face tension, Wulff shape, Hydrodynamic limit, Motion by mean curvature, Evolu-tionary variational inequality, Fluctuations, Large deviations, Free boundaries.

2000 Mathematics Subject Classification: 60-02 (60K35, 60H30, 60H15), 82-02(82B24, 82B31, 82B41, 82C24, 82C31, 82C41), 35J20, 35K55, 35R35


1 Introduction

1.1 Background

The water changes its state to ice or vapor together with variations in tem-perature. Each of these three states (liquid/solid/gas) is macroscopically ho-mogeneous and called a phase (or a pure phase) in physics. The water and theice can coexist at temperature 0

C. In fact, under various physical situations

especially at low temperature, more than one distinct pure phases coexist inspace and different phases are separated by fairly sharp hypersurfaces calledinterfaces. Snow crystals in the vapor or alloys consisting of two types ofmetals are typical examples. Crystals are macroscopic objects, which haveordered arrangements of atoms or molecules in microscopic scale.

Wulff [254] in 1901 proposed a variational principle, at thermodynamiclevel or from the phenomenological point of view, for determining the shapeof interfaces for crystals. Let E ⊂ R

d be a crystal shape. Its boundary ∂E isthen an interface and an energy called the total surface tension is associatedwith each interface by

W(E) =∫

∂E

σ(n(x)) dx , (1.1)

where σ = σ(n) ≥ 0 is the surface tension of flat hyperplane in Rd with unit

normal vector n ∈ Sd−1 and dx represents the volume element on ∂E. Theinterface has locally an energy σ(n(x)) depending on its tilt n = n(x) and,integrating it over the surface ∂E, the Wulff functionalW(E) is defined. Foran alloy consisting of two types of metals A and B, E is the region occupiedby A-type’s metal so that its volume is always kept invariant if the amountof each metal is fixed.

It is expected that the interface, which is in equilibrium and stable, mini-mizes its total energy and this naturally leads us to the variational problem:

minvol (E)=v

W(E) (1.2)

under the condition that the total volume of the crystal E (e.g., the region oc-cupied by A-type’s metal) is fixed to be v > 0 . The minimizer E of (1.2) andits explicit geometric expression are called the Wulff shape and the Wulffconstruction, respectively. Especially when the surface tension σ is indepen-dent of the direction n, W(E) coincides with the surface area of ∂E (exceptconstant multipliers) and (1.2) is equivalent to the well-known isoperimet-ric problem. It is one of quite general and fundamental principles in physicsthat physically realizable phenomena might be characterized by variationalprinciples. Wulff’s variational problem is one of the typical examples.

Crystals are, as we have already pointed out, macroscopic objects. It is aprincipal goal of statistical mechanics to understand such macroscopic phe-nomena in nature from microscopic level of atoms or molecules. Dobrushin,

110 T. Funaki

Kotecky and Shlosman [86] studied the Wulff’s problem from microscopicpoint of view for the first time. They have employed the ferromagnetic Isingmodel as a microscopic model and established, at sufficiently low tempera-tures, the large deviation principle for the sequence of corresponding Gibbsmeasures on finite domains when the volumes of these domains diverge toinfinity. It was shown that the large deviation rate functional is exactly theWulff functional W(E) with the surface tension σ(n) determined thermody-namically from the underlying Gibbs measures. As a consequence, under thecanonical Gibbs measures obtained by conditioning the macroscopic volumeoccupied by + spins to be constant, a law of large numbers is proved andthe Wulff shape is obtained in the limit. The results of Dobrushin et al. wereafterward generalized by Ioffe and Schonmann [152], Bodineau [20], Cerf andPisztora [52] and others; see a review paper [22].

Once an equilibrium situation is understood to some extent, the next targetis obviously the analysis of the corresponding dynamics. The situation thattwo distinct pure phases coexist and are separated by a sharp interface willpersist under the time evolution and the interface will relax slowly. The goalis to investigate the motion of interface on a properly chosen coarse space-time scale. The time evolution corresponding to the Ising model is a reversiblespin-flip dynamics, the so-called Glauber or Kawasaki dynamics which maybe the prime examples. Spin at each site randomly flips and changes its signunder the dynamics without or with conservation law. At sufficiently lowtemperatures, the interactions between spins on two neighboring sites becomestrong enough to incline them to have the common signs with high probabilityand most changes occur near the interface. The shape of interface is howeverrather complicated; for instance, it has overhangs or bubbles.

A class of effective interface models is introduced by avoiding such compli-cations and directly modeling the interface degree of freedom at microscopiclevel; see Sect. 1.3. These models are, at one side, compromises between thedescription of physical phenomena and mathematical requirements but, onthe other side, explain the phenomena in satisfactory good way. The aim ofthese notes is to try to give an overview of results mostly on the ∇ϕ interfacemodel, which is one of such effective interface models.

As we have observed, in statistical mechanics, there are at least two dif-ferent scales: macroscopic and microscopic ones. The procedures connectingmicroscopic models with the macroscopic phenomena are realized by takingthe scaling limits. The scaling parameter N ∈ N represents the ratio of themacroscopically typical length (e.g., 1 cm) to the microscopic one (e.g., 1 nm)and it is finite, but turns out to be quite large (N = 107 in this example).The physical phenomena can be mathematically understood only by takingthe limit N →∞. The dynamics involves the scalings also in time. Within amacroscopic unit length of time, the molecules collide with each other withtremendous frequency. Since the microscopic models such as the Ising modeland the ∇ϕ interface model involve randomness, the limit procedure N →∞


can be formulated in the framework of classical limit theorems in probabilitytheory.

The principal ideas behind these limit theorems are that, by the ergodicor mixing properties of the microscopic systems, the microscopic (physical)quantities are locally in macroscopic scale averaged or homogenized under thescaling limits. The macroscopic observables are obtained under such averagingeffects. However, the ∇ϕ interface model which we shall discuss in the presentnotes has only an extremely weak mixing property and this sometimes makesthe analysis of the model difficult. For instance, the thermodynamic quantitymay diverge under the usual scaling. This suggests the necessity of introducingscalings different from the usual one to obtain a nontrivial limit.

1.2 Quick Overview of the Results

In Sect. 2, the ∇ϕ interface model is precisely introduced. The basic micro-scopic objects are height variables φ of interfaces. Assigning an energy H(φ)to each height variable, its statistical ensemble in equilibrium is defined by theGibbs measures. Then, the corresponding time evolution called the Ginzburg-Landau ∇ϕ interface model is constructed in such a way that it is reversibleunder the Gibbs measures, in other words, the detailed balance is fulfilled. Thescaling limits connecting microscopic and macroscopic levels will be explained.

The ∇ϕ interface model with quadratic potentials is discussed in Sect. 3 asa warming up before studying general case with convex potentials. In Sect. 4,fundamental tools like Helffer-Sjostrand (random walk) representation, FKGinequality and Brascamp-Lieb inequality are presented.

A basic role in various limit theorems is played by the so-called surfacetension σ(u), u ∈ R

d. The function σ is a macroscopic or thermodynamicfunction and will be introduced in Sect. 5. The limit theorems under thescalings can be formulated in the terminology of probability theory as follows:

Law of large numbers (LLN): Macroscopic quantity obtained under thescaling limit from randomly fluctuating microscopic objects, i.e., heightvariables of interfaces in our model, becomes deterministic due to certainaveraging effects.

Central limit theorem (CLT): Fluctuations around the deterministiclimit are studied.

Large deviation principle (LDP): LDPs for macroscopically scaledheight variables are sometimes useful to show the LLNs.

From the physical point of view, these limit theorems are classified into twotypes: static results on the equilibrium Gibbs measures and dynamic results:

(1) Static results, Sects. 6-9.

LDP, LLN and derivation of variational principles (VP), Sect. 6:LDP was studied for Gaussian case by Ben Arous and Deuschel [12] and

112 T. Funaki

for general Gibbsian case by Deuschel, Giacomin and Ioffe [77]. For heightvariables conditioned to be positive and to have definite total volume, theshape of most probable droplet called the Wulff shape is determined asa minimizer of the total surface tension as a consequence of LDP. Addingan effect of weak self potentials to the system, Funaki and Sakagawa [123]derived the VPs of Alt and Caffarelli [5] or Alt, Caffarelli and Friedman [6].Bolthausen and Ioffe [31] discussed under additional pinning effect at awall for 2+1 dimensional system and obtained the Winterbottom shapein the limit.

Entropic repulsion (wall effect), Sect. 7.1: The entropic repulsion isthe problem to study, when a hard wall is settled at the height level0, how high the interfaces are pushed up by the randomness (i.e., theentropic effect) naturally existing in the Gibbs measures. The problem wasposed by Lebowitz and Maes [186] and then investigated by Bolthausen,Deuschel and Zeitouni [29], Deuschel [74], Deuschel and Giacomin [75] forGaussian case and by Deuschel and Giacomin [76] for general Gibbsiancase.

Pinning and wetting transition, Sects. 7.2, 7.3: The pinning is the prob-lem to study, under the effect of weak force attracting interfaces to theheight level 0, whether the field is really localized or not. The problemwas discussed by Dunlop, Magnen, Rivasseau and Roche [93], Deuscheland Velenik [81], Ioffe and Velenik [153] and Bolthausen and Velenik [32].The two effects of entropic repulsion and pinning conflict with each other,and a natural question to be addressed is which effect is dominant inthe system. In one and two dimensions, a phase transition called wettingtransition occurs. This fact was first observed by Fisher [101] in one di-mension, followed by Bolthausen, Deuschel and Zeitouni [30] and Caputoand Velenik [49].

CLT, Sect. 8: Naddaf and Spencer [202] investigated CLT for Gibbs mea-sures. The result is nontrivial since the Gibbs measures have long corre-lations.

Characterization of ∇ϕ-Gibbs measures, Definition 2.2, Sect. 9: Thefamily of all (tempered and shift invariant) ∇ϕ-Gibbs measures is charac-terized based on the coupling argument for the corresponding dynamics.This result plays a key role in the proof of the hydrodynamic limit.

(2) Dynamic results I, Sects. 10-12.

Hydrodynamic limit (LLN) and derivation of motion by mean cur-vature with anisotropy, Sect. 10: LLN is shown under the time evolu-tion. This procedure is called the hydrodynamic limit and established byFunaki and Spohn [124]. Motion by mean curvature (MMC) except forsome anisotropy is derived in the limit. The diffusion matrix of the limitequation is formally given by Hessian of the surface tension.


Equilibrium fluctuation (CLT), Sect. 11: Dynamic CLT in equilibrium isstudied and an infinite dimensional Ornstein-Uhlenbeck process is derivedin the limit by Giacomin, Olla and Spohn [135]. The identification ofthe covariance matrix with Hessian of the surface tension, however, stillremains open.

LDP, Sect. 12: Dynamic LDP was discussed by Funaki and Nishikawa [121].

(3) Dynamic results II, Sects. 13-15.The dynamics under the effects of wall or additional weak self potentials

is studied.

Hydrodynamic limit on a wall, Sect. 13: The limit is MMC with reflec-tion and described by an evolutionary variational inequality, Funaki[117].

Equilibrium fluctuation (CLT) on a wall, Sects. 14.1, 14.2: A stochas-tic PDE with reflection is obtained under the scaling limit, Funaki andOlla [122].

Dynamic entropic repulsion, Sect. 14.3: The problem of entropic repul-sion is investigated under the dynamics, Deuschel and Nishikawa [80] andothers.

Dynamics in two media, Sect. 15.1: The dynamics associated with theHamiltonian added a weak self potential is discussed.

Pinning dynamics on a wall, Sect. 15.2: Dynamics under the effects ofboth pinning and repulsion is constructed.

(4) Other dynamic models for interfaces, Sect. 16.The following five topics are discussed in the last supplementary section.

Stochastic lattice gas and free boundary problemsInteracting Brownian particles at zero temperatureSingular limits for stochastic reaction-diffusion equationsLimit shape of random Young diagramsGrowing interfaces

Funaki [116] and Giacomin [130, 132, 133] are survey papers on the ∇ϕinterface model. See also [125, 210, 224] for problems on interfaces and crystals.

1.3 Derivation of Effective Interface Models from Ising Model

Let us briefly and rather formally explain how one can derive the effectiveinterface models from the ferromagnetic Ising model at sufficiently low tem-perature. In the Ising model, the energy is associated to each ± spin configu-ration s = s(x);x ∈ Λ ∈ +1,−1Λ on a large box Λ := [−, ]d ∩ Z

d asthe sum over all bonds 〈x, y〉 in Λ (i.e., x, y ∈ Λ : |x− y| = 1)

114 T. Funaki

H(s) = −∑

〈x,y〉⊂Λ

s(x)s(y) .

The sum is usually defined under certain boundary conditions. We shall con-sider, for simplicity, only when d = 2. The function H(s) can be rewrittenas

H(s) = 2|γ| (+ constant)

in terms of the set of contours γ = γ(s) on the dual lattice corresponding tos, which separate two regions consisting of sites occupied by + and − spins,respectively, where |γ| denotes the number of bonds in γ (the total length ofγ) and an additional constant in H(s) is independent of the configurations s.Under the Gibbs measure

µ(s) =1Ze−βH(s), s ∈ +1,−1Λ ,

with the normalization constant Z, if the temperature T (β = 1/kT , k > 0is the Boltzmann constant) is sufficiently low, the configurations of spinswhich have the same values on neighboring sites overwhelm the probabil-ity, since such configurations have smaller energies. In other words, whenthere is a single large contour γ, the probability that the configurations inFig. 1.2 having bubbles arise is very little and almost negligible. We cantherefore disregard (with high probability) the configurations with bubblesand assume that the configurations like in Fig. 1.1 can only appear. Suchspin configurations s are equivalently represented by the height variablesφ = φ(x) ∈ [−, ] ∩ Z;x ∈ [−, ]d−1 ∩ Z

d−1 (in fact, we are consider-ing the case of d = 2) which measure the distances of γ from the x-axis, onefixed hyperplane. Then, the energy H(s) has another form

H(φ) = 2∑

〈x,y〉⊂[−,]d−1∩Zd−1

|φ(x)− φ(y)| (1.3)

Fig. 1.1. Possible configurations Fig. 1.2. Neglected configurations


up to an additional constant; notice that the number of horizontal bonds inγ is always fixed. The model for random interfaces φ : [−, ]d−1 ∩ Z

d−1 → Z

with the energy (1.3) is called the SOS (Solid on Solid) model. One canfurther replace the space Z for values of height variables with continuum R

and |φ(x)−φ(y)| with V (φ(x)−φ(y)), and this leads us to the ∇ϕ interfacemodel. As a generalization of the function V (η) = |η|, it is natural to supposethat the potential function V is convex and symmetric (even) so that theenergy is small when the differences of heights φ : [−, ]d−1 ∩ Z

d−1 → R onneighboring sites are small, in other words, when the interfaces are more flat.

1.4 Basic Notation

• For Λ ⊂ Zd (d dimensional square lattice),

∂+Λ = x /∈ Λ; |x− y| = 1 for some y ∈ Λ

is the outer boundary of Λ and Λ = Λ∪∂+Λ is the closure of Λ, respectively,where x /∈ Λ means x ∈ Λc = Z

d \ Λ. The inner boundary of Λ is

∂−Λ = x ∈ Λ; |x− y| = 1 for some y /∈ Λ .

• Λ Zd means that Λ is a finite subset of Z

d: |Λ|(= Λ) <∞.• O ∈ Z

d stands for the origin and, for ∈ N, Λ = [−, ]d ∩Zd denotes the

lattice cube with center O and side length 2+ 1.

• |x| = max1≤i≤d

|xi| for x = (xi)di=1 ∈ Z

d and |u| =√∑d

i=1 u2i for u = (ui)d

i=1 ∈

Rd (There will be some exceptional usages in Sect. 3). The inner product

of Rd is denoted by u · x or sometimes by (u, x) for u, x ∈ R

d.• For a bounded domain D in R

d, we denote DN = ND ∩ Zd, where

ND = Nθ; θ ∈ D ⊂ Rd and N ∈ N stands for the scaling parame-

ter. The set DN is a microscopic correspondence, which is discretized, tothe macroscopic domain D.

• The set Td = (R/Z)d ≡ (0, 1]d denotes a d dimensional unit torus (identi-

fying 0 with 1) and TdN = (Z/NZ)d ≡ 1, 2, . . . , Nd is the corresponding

microscopic lattice torus (identifying 0 with N). We also use the notationT

d = (−π, π]d.• For a topological space S, P(S) stands for the family of all Borel proba-

bility measures on S.

Acknowledgment

The results stated in Sects. 9 and 10, one of the cores in these notes, havegrown out of the visit of H. Spohn to Japan in the spring of 1995. I ammuch indebted to him, who actually got me started to work on the problemsrelated to the ∇ϕ interface model. I was stimulated by discussions with many

116 T. Funaki

people, in particular, with J.-D. Deuschel, G. Giacomin, D. Ioffe, S. Olla, G.S.Weiss and N. Yoshida. H. Sakagawa read an early version in part and gave meseveral suggestions for improvement. Professor J. Picard invited me to delivera series of lectures at the International Probability School at Saint-Flour,2003. I deeply thank all of these people.

2 ∇ϕ Interface Model

The ∇ϕ interface model has a rather simplified feature, for example, whenit is compared with the Ising model, as we have pointed out. It is, however,equipped with a sufficiently wide variety of nontrivial aspects and serves as auseful model to explain physical behavior of interfaces from microscopic pointof view. In this section we introduce the model.

2.1 Height Variables

We are concerned with a hypersurface (interface) embedded in d + 1 dimen-sional space R

d+1, which separates two distinct pure phases. Notice that, inSect. 1.3, we discussed in d dimensional space; however, here and after d isreplaced with d+1. To avoid complications, we assume that the interface hasno overhangs nor bubbles and accordingly that it is represented as a graphviewed from a certain d dimensional fixed reference hyperplane Γ located inthe space R

d+1. In other words, the location of the interface is described by theheight variables φ = φ(x) ∈ R;x ∈ Γ, which measure the vertical distancesbetween the interface and Γ . The variables φ are microscopic objects, and thespace Γ is discretized and taken as Γ = Λ( Z

d), in particular, Γ = DN witha (macroscopic) bounded domain D in R

d or lattice torus TdN or Z

d. Here Nrepresents the size of the microscopic system, and our main interest will bein analyzing the asymptotic behavior of the system under the scaling limitN →∞.

2.2 Hamiltonian

An energy is associated with each height variable φ : Γ → R by assigningpenalty according to its tilts (slopes). Namely, we define the HamiltonianH(φ) as the sum over all bonds (i.e., pairs of nearest neighbor sites) 〈x, y〉 inΓ when Γ = T

dN or Z

d, and in Γ when Γ = DN or Γ = Λ Zd in general

H(φ) ≡ HψΓ (φ) =

∑〈x,y〉⊂Γ (or Γ )

V (φ(x)− φ(y)) . (2.1)

Note that the boundary conditions ψ = ψ(x);x ∈ ∂+Γ are required todefine the sum (2.1) for Γ = DN , i.e., we assume


φ(x) = ψ(x), x ∈ ∂+Γ ,

in the sum. When Γ = Zd, (2.1) is a formal infinite sum. The (interaction) po-

tential V is smooth, symmetric and strictly convex. More precisely, through-out the present notes we require the following three conditions on the potentialV = V (η):

(V1) (smoothness) V ∈ C2(R),(V2) (symmetry) V (−η) = V (η), η ∈ R, (2.2)(V3) (strict convexity) c− ≤ V ′′(η) ≤ c+, η ∈ R, for some c−, c+ > 0.

The surface φ has low energy if the tilts |φ(x) − φ(y)| are small. The energy(2.1) of the interface φ is constructed in such a manner that it is invariantunder a uniform translation φ(x) → φ(x) + h for all x ∈ Z

d (or x ∈ Γ ) andh ∈ R. A typical example of V satisfying the conditions (2.2) is a quadraticpotential V (η) = c

2η2, c > 0.

For every Λ ⊂ Zd, Λ∗ denotes the set of all directed bonds b = 〈x, y〉 in Λ,

which are directed from y to x. We write xb = x, yb = y for b = 〈x, y〉. Foreach b ∈ (Zd)∗ and φ = φ(x);x ∈ Z

d ∈ RZ

d

, define

∇φ(b) = φ(xb)− φ(yb) .

We also define ∇iφ(x) = φ(x + ei) − φ(x), 1 ≤ i ≤ d for x ∈ Zd

where ei ∈ Zd is the i-th unit vector given by (ei)j = δij . The variables

∇φ(x) = ∇iφ(x)1≤i≤d ∈ Rd represent vector field of height differences

or sometimes called tilt (or gradients) of φ. The Hamiltonian H(φ) is thenrewritten as

H(φ) =12

∑b∈Γ∗(or Γ

∗)

V (∇φ(b)) . (2.3)

The factor 1/2 is needed because each undirected bond b = 〈x, y〉 is countedtwice in the sum. Since the energy is determined from the height differences∇φ, the model is called the ∇ϕ interface model.

Remark 2.1. (1) The sum (2.1) is meaningful only when the potential V issymmetric, while the expression (2.3) makes sense for asymmetric V . How-ever, note that the sum (2.3) is essentially invariant (except for the boundarycontributions) if V is replaced with its symmetrization 1

2V (η) + V (−η).(2) The potential V can be generalized to the bond-dependent case: Vb =Vb(η); b ∈ (Zd)∗ so that the corresponding Hamiltonian is defined by (2.3)with V replaced by Vb; see Example 5.3, Problem 10.1 below and [230]. Thisformulation truly covers the asymmetric potentials.

Remark 2.2. (1) In the quantum field theory, H is called massless Hamil-tonian and well studied in ’80s. Massive Hamiltonian is given by Hm(φ) =H(φ) + 1

2m2∑

x φ(x)2,m > 0. The Hamiltonian with weak self potentials orpinning potentials will be introduced in Sect. 6.1 or in Sect. 6.4 (see also

118 T. Funaki

Sect. 7.2), respectively. (2) In our model, height variables φ(x) themselves arenot discretized. The SOS (solid on solid) model is a model obtained discretiz-ing the height variables simultaneously: φ(x) ∈ Z+ and with V (η) = |η|, cf.Sect. 1.3 and [54], [55], [106].(3) (∆ϕ interface model) In the ∇ϕ interface model, the energy H(φ) isroughly the surface area of the microscopic interface φ. In fact, this is true forV (η) =

√1 + η2. However, if we are concerned for example with the mem-

brane as the object of our study, its surface area is preserved and alwaysconstant. Therefore the energy should be determined by taking into accountthe next order term like

∑x(∆φ(x))2, which may be regarded as the curvature

of φ, see [145].

2.3 Equilibrium States (Gibbs Measures)

Once the Hamiltonian H is specified, in the formulation of statistical me-chanics, equilibrium states called Gibbs measures can be naturally associatedtaking the effect of random fluctuations into account.

ϕ-Gibbs Measures

For a finite region Λ Zd, the Gibbs measure (more exactly, ϕ-Gibbs

measure, finite volume ϕ-Gibbs measure or local specification) for the field ofheight variables φ ∈ R

Λ over Λ is defined by

µ(dφ) ≡ µψΛ(dφ) =

1

ZψΛ

exp−Hψ

Λ (φ)dφΛ , (2.4)

with the boundary conditions ψ ∈ R∂+Λ. The term e−Hψ

Λ (φ) is the Boltzmannfactor, while

dφΛ =∏x∈Λ

dφ(x)

is the Lebesgue measure on RΛ which represents uniform fluctuations of the

interface. The constant ZψΛ is for normalization defined by

ZψΛ =

∫RΛ

exp−Hψ

Λ (φ)dφΛ . (2.5)

Note that the conditions (2.2) imply ZψΛ <∞ for every Λ Z

d and thereforeµψ

Λ ∈ P(RΛ).The reason for introducing these measures is based on a physical argument.

The uniform measure dφΛ arises from the postulate in equilibrium statisti-cal mechanics called principle of equal a priori probabilities, while theBoltzmann factor naturally appears from the Gibbs’ principle which is some-times called equivalence of ensembles: a subsystem in a very large closed


system distributed under the microcanonical ensemble (= equal probabili-ties on a system with conservation law) is described by the Gibbs measure,[72, 128, 223].

We shall often regard µψΛ ∈ P(RΛ) by considering φ(x) = ψ(x) for x ∈

∂+Λ under µψΛ. The boundary condition ψ is sometimes taken from R

Λc

, andwe regard µψ

Λ ∈ P(RZd

) in such case. When Γ = TdN , the Gibbs measure

is unnormalizable, since HψΛ (φ) is translation invariant and this makes the

normalization ZT

dN

=∞.For an infinite region Λ : |Λ| = ∞, the expression (2.4) has no meaning

since the Hamiltonian HΛ(φ) is a formal sum. Nevertheless, one can define thenotion of Gibbs measures on Z

d based on the well-known DLR formulations.For A ⊂ Z

d, we shall denote FA the σ-field of RZ

d

generated by φ(x);x ∈ A.

Definition 2.1. The probability measure µ ∈ P(RZd

) is called a Gibbs mea-sure for ϕ-field (ϕ-Gibbs measure for short), if its conditional probabilityon FΛc satisfies the DLR equation

µ( · |FΛc)(ψ) = µψΛ( · ), µ-a.e.ψ ,

for every Λ Zd.

It is known that the ϕ-Gibbs measures exist when the dimension d ≥ 3,but not for d = 1, 2. An infinite volume limit (thermodynamic limit) for µ0

Λ

as Λ Zd exists only when d ≥ 3 (cf. Sect. 4.5).

∇ϕ-Gibbs Measures

The height variables φ = φ(x);x ∈ Zd on Z

d automatically determines afield of height differences∇φ = ∇φ(b); b ∈ (Zd)∗. One can therefore considerthe distribution µ∇ of ∇ϕ-field under the ϕ-Gibbs measure µ. We shall callµ∇ the ∇ϕ-Gibbs measure. In fact, it is possible to define the ∇ϕ-Gibbsmeasures directly by means of the DLR equations and, in this sense,∇ϕ-Gibbsmeasures exist for all dimensions d ≥ 1 (cf. Sect. 4.4).

In order to describe the DLR equation for ∇ϕ-Gibbs measures, we firstclarify the structure of the state space for the ∇ϕ-field. It is obvious that theheight variable φ ∈ R

Zd

determines ∇φ ∈ R(Zd)∗ ; however, all η = η(b) ∈

R(Zd)∗ can not be the ∇ϕ-field, i.e., it may not be possible to find φ such that

η = ∇φ in general. Indeed, ∇φ always satisfies the loop condition: every sumof ∇φ along a closed loop must vanish. To state more precisely, we introducesome notion.

A sequence of bonds C = b(1), b(2), . . . , b(n) is called a chain connectingy and x (y, x ∈ Z

d) if yb(1) = y, xb(i) = yb(i+1) for 1 ≤ i ≤ n− 1 and xb(n) = x.The chain C is called a closed loop if xb(n) = yb(1) . A plaquette is a closedloop P = b(1), b(2), b(3), b(4) such that xb(i) , i = 1, .., 4 consists of fourdifferent points. The field η = η(b) ∈ R

(Zd)∗ is said to satisfy the plaquettecondition if

120 T. Funaki

(P1) η(b) = −η(−b) for all b ∈ (Zd)∗,

(P2)∑b∈P

η(b) = 0 for all plaquettes P in Zd,

where −b denotes the reversed bond of b. Note that, if φ = φ(x) ∈ RZ

d

,then ∇φ = ∇φ(b) ∈ R

(Zd)∗ automatically satisfies the plaquette condition.The plaquette condition is equivalent to the loop condition:

(L)∑b∈C

η(b) = 0 for all closed loops C in Zd.

Notice that the condition (P1) follows from (L) by taking the closed loopC = b,−b. We set

X = η ∈ R(Zd)∗ ; η satisfies the loop condition ,

then X is the state space for the ∇ϕ-field endowed with the topology inducedfrom the space R

(Zd)∗ having product topology. In fact, the height differencesηφ ∈ X are associated with the heights φ ∈ R

Zd

by

ηφ(b) := ∇φ(b), b ∈ (Zd)∗ , (2.6)

and, conversely, the heights φη,φ(O) ∈ RZ

d

can be constructed from heightdifferences η and the height variable φ(O) at x = O as

φη,φ(O)(x) :=∑

b∈CO,x

η(b) + φ(O) , (2.7)

where CO,x is an arbitrary chain connecting O and x. Note that φη,φ(O) iswell-defined if η = η(b) ∈ X .

We next define the finite volume ∇ϕ-Gibbs measures. For every ξ ∈ Xand Λ Z

d the space of all possible configurations of height differences onΛ∗ := b = 〈x, y〉 ∈ (Zd)∗; x or y ∈ Λ for given boundary condition ξ isdefined as

XΛ∗,ξ = η = (η(b))b∈Λ∗ ; η ∨ ξ ∈ X ,

where η ∨ ξ ∈ X is determined by (η ∨ ξ)(b) = η(b) for b ∈ Λ∗ and = ξ(b) forb /∈ Λ∗. The finite volume ∇ϕ-Gibbs measure in Λ (or, more precisely, in Λ∗)with boundary condition ξ is defined by

µ∇Λ,ξ(dη) =

1ZΛ,ξ

exp

−

12

∑b∈Λ∗

V (η(b))

dηΛ,ξ ∈ P(XΛ∗,ξ) ,

where dηΛ,ξ denotes a uniform measure on the affine space XΛ∗,ξ and ZΛ,ξ isthe normalization constant. We shall sometimes regard µ∇

Λ,ξ ∈ P(X ) by con-sidering η(b) = ξ(b) for b /∈ Λ∗ under µ∇

Λ,ξ as before. Note that the dimension


of the space XΛ∗,ξ is |Λ| at least if Zd \Λ is connected, since one can associate

η with φ = φΛ by

φ(x) =∑

b∈Cx0,x

(η ∨ ξ)(b), x ∈ Λ , (2.8)

where x0 /∈ Λ is fixed and Cx0,x is a chain connecting x0 and x.The finite volume ϕ-Gibbs measures and the finite volume ∇ϕ-Gibbs mea-

sures are associated with each other as we have pointed out above. Namely,given ξ ∈ X and h ∈ R, define ψ ∈ R

Zd

as ψ = φξ,h by (2.7). Then, if φ isµψ

Λ-distributed with the boundary condition ψ constructed in this way, ∇φ isµ∇

Λ,ξ-distributed. The distribution of ∇φ is certainly independent of the choiceof h.

Now, similarly to the definition of the ϕ-Gibbs measures on Zd, one can

introduce the ∇ϕ-Gibbs measures on (Zd)∗.

Definition 2.2. The probability measure µ∇ ∈ P(X ) is called a Gibbs mea-sure for the height differences (∇ϕ-Gibbs measure for short), if it satisfiesthe DLR equation

µ∇( · |F(Zd)∗\Λ∗)(ξ) = µ∇Λ,ξ( · ), µ∇-a.e. ξ ,

for every Λ Zd, where F(Zd)∗\Λ∗ stands for the σ-field of X generated by

η(b); b ∈ (Zd)∗ \ Λ∗.

Markov Property

In the Hamiltonian H(φ), the interactions among the height variables areonly counted through the neighboring sites. This structure is reflected as theMarkov property of the field of height variables φ = φ(x) under the(finite or infinite volume) ϕ-Gibbs measures µψ

Λ and µ:

Proposition 2.1. (1) Let Λ Zd and the boundary condition ψ ∈ R

∂+Λ

be given. Suppose that Λ is decomposed into three regions A1, A2, B and Bseparates A1 and A2; namely, Λ = A1∪A2∪B, A1∩A2 = A1∩B = A2∩B =∅ and |x1 − x2| > 1 holds for every x1 ∈ A1 and x2 ∈ A2. Then, underthe conditional probability µψ

Λ ( · |FB), the random variables φA1 and φA2 aremutually independent, where we denote φA1 = φ(x);x ∈ A1 etc.(2) Let µ ∈ P(RZ

d

) be a ϕ-Gibbs measure. Then, for every A Zd, the

random variables φA and φAc are mutually independent under the conditional

probability µ ( · |F∂+A).

In particular, in one dimension, φ = φ(x) is a pinned random walkunder µψ

Λ regarding x as time variables. Let η(y); y = 1, 2, . . . be an R-valuedi.i.d. defined on a certain probability space (Ω,P ) having the distributionp(a)da, where

122 T. Funaki

p(a) =1ze−V (a), a ∈ R

and z =∫

Re−V (a) da is the normalization constant. Then, we have the follow-

ing.

Proposition 2.2. Let Λ = 1, 2, . . . , N−1 ⊂ Z1 and assume that the bound-

ary conditions are given by ψ(0) = h0, ψ(N) = h1. Define the height variablesφ = φ(x);x ∈ Λ, Λ = 0, 1, 2, . . . , N by

φ(x) = h0 +x∑

y=1

η(y), x ∈ Λ ,

and consider them under the conditional probability P ( · |φ(N) = h1). Then,φΛ = φ(x);x ∈ Λ is µψ

Λ-distributed.

Shift Invariance and Ergodicity

Here, we recall the notion of shift invariance and ergodicity under the shifts forϕ-fields and ∇ϕ-fields, respectively, see, e.g., [128]. For x ∈ Z

d, we define theshift operators τx : R

Zd → R

Zd

for heights by (τxφ)(y) = φ(y − x) for y ∈ Zd

and φ ∈ RZ

d

. The shifts for height differences are also denoted by τx. Namely,τx : X → X

(or τx : R

(Zd)∗ → R(Zd)∗

)are defined by (τxη)(b) = η(b − x) for

b ∈ (Zd)∗ and η ∈ X(or η ∈ R

(Zd)∗), where b− x = 〈xb − x, yb − x〉 ∈ (Zd)∗.

Definition 2.3. A probability measure µ ∈ P(RZd

) is called shift invariantif µ τ−1

x = µ for every x ∈ Zd. A shift invariant µ ∈ P(RZ

d

) is calledergodic (under the shifts) if τx-invariant functions F = F (φ) on R

Zd

(i.e.,functions satisfying F (τxφ) = F (φ) µ-a.e. for every x ∈ Z

d) are constant (µ-a.e.). Similarly, the shift invariance and ergodicity for a probability measureµ ∈ P(X )

(or µ ∈ P(R(Zd)∗)

)are defined.

2.4 Dynamics

Corresponding to the Hamiltonian H(φ), one can naturally introduce a ran-dom time evolution of microscopic height variables φ of the interface. Indeed,we consider the stochastic differential equations (SDEs) for φt = φt(x);x ∈Γ ∈ R

Γ , t > 0

dφt(x) = − ∂H

∂φ(x)(φt)dt+

√2dwt(x), x ∈ Γ , (2.9)

where wt = wt(x);x ∈ Γ is a family of independent one dimensional stan-dard Brownian motions. The derivative of H(φ) in the variable φ(x) is givenby


∂H

∂φ(x)(φ) =

∑y∈Γ (or Γ ):|x−y|=1

V ′(φ(x)− φ(y)) , (2.10)

for x ∈ Γ . When Γ Zd, the SDEs (2.9) have the form

dφt(x) = −∑

y∈Γ :|x−y|=1

V ′(φt(x)− φt(y))dt+√

2dwt(x), x ∈ Γ , (2.11)

subject to the boundary conditions

φt(y) = ψ(y), y ∈ ∂+Γ . (2.12)

When Γ = Zd, although the Hamiltonian H is a formal sum, its derivative

(2.10) has an affirmative meaning and we can write down the SDEs for φt =φt(x);x ∈ Z

d ∈ RZ

d

, t > 0

dφt(x) = −∑

y∈Zd:|x−y|=1


2dwt(x), x ∈ Zd . (2.13)

The SDEs (2.11) with (2.12) or the SDEs (2.13) have unique solutions, sincethe coefficient V ′ in the drift term is Lipschitz continuous by our assump-tions (2.2). For (2.13), since it is an infinite system, one need to introduce aproper function space for solutions, cf. Lemmas 9.1 and 9.2. The evolution ofφt is designed in such a manner that it is stationary and, moreover, reversibleunder the Gibbs measures µψ

Λ or µ, cf. Proposition 9.4 for the associated ∇ϕ-dynamics. In physical terminology, the equation fulfills the detailed balancecondition. Such evolution or the SDEs are called Ginzburg-Landau dy-namics, distorted Brownian motion or the Langevin equation associatedwith H(φ).

The drift term in the SDEs (2.9) determines the gradient flow along whichthe energy H(φ) decreases. In fact, since the function V is symmetric andconvex, φt(x) > φt(y) implies that −V ′(φt(x) − φt(y)) < 0 so that the driftterm of (2.11) or (2.13) is negative and therefore φt(x) decreases. Converselyif φt(x) < φt(y), the drift is positive and φt(x) increases. Therefore, in bothcases, the drift has an effect to make the interface φ flat. The term

√2wt(x)

gives a random fluctuation which competes against the drift.The Dirichlet form corresponding to the SDEs (2.13) is

E(F,G) ≡ −Eµ[FLG] =∑x∈Zd

Eµ[∂F (x, φ)∂G(x, φ)] , (2.14)

for F = F (φ), G = G(φ), where Eµ[ · ] denotes the expectation under theGibbs measure µ, L is the generator of the process φt and ∂F (x, φ) :=∂F/∂φ(x), cf. Sects. 4.1 and 10.3. Indeed, at least when Γ Z

d, the genera-tor L of the process φt ∈ R

Γ determined by the SDEs (2.9) is the differentialoperator of second order

124 T. Funaki

L =∑x∈Γ

(∂

∂φ(x)

)2

−∑x∈Γ

∂H

∂φ(x)∂

∂φ(x)(2.15)

and, by integration by parts formula, we have∫

RΓ

FLG · e−H dφΓ =∫

RΓ

F∑x∈Γ

∂

∂φ(x)

∂G

∂φ(x)e−H

dφΓ

= −∑x∈Γ

∫RΓ

∂F

∂φ(x)∂G

∂φ(x)· e−H dφΓ ,

for every F = F (φ), G = G(φ) ∈ C2b (RΓ ). The Hamiltonians H may be more

general than (2.1), for instance, those with self potentials, see (6.3)

Remark 2.3. (1) The dynamics corresponding to the massive HamiltonianHm (recall Remark 2.2) can be introduced similarly. It forces the heightsφ = φ(x) to stay bounded.(2) Interface dynamics of SOS type was studied by several authors, e.g., Dun-lop [90] considered the dynamics for the corresponding gradient fields in onedimension; see also Remark 13.1 and Sect. 16.5.

2.5 Scaling Limits

Our main interest is in the analysis of the scaling limit, which passes frommicroscopic to macroscopic levels. For the microscopic height variables φ =φ(x);x ∈ Γ with Γ = DN ,T

dN or Z

d, the macroscopic height variableshN = hN (θ) are associated by

hN (θ) =1Nφ ([Nθ]) , θ ∈ D,Td or R

d , (2.16)

where [Nθ] stands for the integer part of Nθ(∈ Rd) taken componentwise.

Note that both x- and φ-axes are rescaled by a factor 1/N . This is becausethe ϕ-field represents a hypersurface embedded in d + 1 dimensional space.The functions hN are step functions. Sometimes interpolations by polilinearfunctions (or polygonal approximations) are also considered, see (6.9) and(6.21) below.

For the time evolution φt = φt(x);x ∈ Γ, t > 0 of the interface, we shallmostly work under the space-time diffusive scaling

hN (t, θ) =1NφN2t([Nθ]) . (2.17)

2.6 Quadratic Potentials

Here we take a quadratic function V (η) = 12η

2 as a typical example of thepotential satisfying our basic conditions (2.2). To rewrite the Hamiltonian


H(φ) for such V , let us introduce the discrete Laplacian ∆ ≡ ∆Λ,ψ for Λ Zd

with boundary conditions ψ ∈ R∂+Λ

∆φ(x) =∑

y∈Λ:|x−y|=1

((φ ∨ ψ)(y)− φ(x)) , x ∈ Λ , (2.18)

where φ ∨ ψ ∈ RΛ stands for the height variables which coincide with φ on Λ

and with ψ on ∂+Λ, respectively; i.e., φ ∨ ψ(x) = φ(x) for x ∈ Λ and = ψ(x)for x ∈ ∂+Λ. The summation by parts formula proves that

H0Λ(φ) = −1

2(φ,∆Λ,0φ)Λ (2.19)

where (φ1, φ2)Λ =∑

x∈Λ φ1(x)φ2(x) denotes an inner product of φ1 and φ2 ∈R

Λ. The boundary condition is taken ψ = 0 for simplicity. In particular, thefinite volume Gibbs measure µ0

Λ can be expressed as

µ0Λ(dφΛ) =

1Z0

Λ

e12 (φ,∆Λ,0φ)Λ dφΛ ,

and accordingly, φΛ forms a Gaussian field under the distribution µ0Λ with

mean 0 and covariance (−∆Λ,0)−1, the inverse operator of −∆Λ,0, see Sect.3.1 for more details.

For V (η) = 12η

2, the corresponding dynamics (2.9) is a simple discretestochastic heat equation

dφt(x) = ∆φt(x)dt+√

2dwt(x), x ∈ Γ . (2.20)

3 Gaussian Equilibrium Systems

As a warming up before studying general systems, let us consider the ∇ϕinterface model in the case where the potential is quadratic: V (η) = 1

2η2. The

corresponding system formed by the height variables φ is then Gaussian andsometimes called a free lattice field or a harmonic oscillator in physical litera-tures. For a Gaussian system, one can explicitly compute the mean, covariance(two-point correlation function) and characteristic functions. In particular, aswe shall see, the covariance of our field φ can be represented by means of thesimple random walks on the lattice, Proposition 3.2. This will be extended togeneral potentials V and called the Helffer-Sjostrand representation, see Sect.4.1 below.

We begin with systems on finite and connected regions Λ( Zd) in Sect. 3.1

and then, by taking the thermodynamic limit (i.e., Λ Zd), infinite systems

on Zd will be constructed in Sect. 3.2. We shall also discuss massive system

and see significant differences in massive and massless systems, for instance,in the speed of decay of correlation functions or the dependence of the systemon the boundary conditions, see Sect. 3.3. Sect. 3.4 deals with the macroscopicscaling limits for ϕ and ∇ϕ-fields.

126 T. Funaki

3.1 Gaussian Systems in a Finite Region

We assume that Λ Zd is connected. When V (η) = 1

2η2 and the boundary

conditions ψ ∈ R∂+Λ (or ψ ∈ R

Zd

or ψ ∈ RΛc

) are given, the correspondingHamiltonian H(φ) ≡ Hψ

Λ (φ) defined by (2.1) is a quadratic form of φ so thatthe finite volume ϕ-Gibbs measure µψ

Λ ∈ P(RΛ) (or ∈ P(RZd

)) determined by(2.4) is Gaussian.

Harmonic Functions and Green Functions

The mean and covariance of the height variables φ = φ(x);x ∈ Λ under µψΛ

are computable by solving the Dirichlet boundary value problem on Λ for thediscrete Laplacian ∆. Indeed, we consider the difference equation on Λ withthe boundary condition ψ

∆φ(x) :=

∑y∈Zd:|x−y|=1

(φ(y)− φ(x)) = 0, x ∈ Λ ,

φ(x) =ψ(x), x ∈ ∂+Λ,

(3.1)

which is equivalent to

∆Λ,ψφ(x) = 0, x ∈ Λ ,

where ∆Λ,ψ is the discrete Laplacian determined by (2.18). The solution φ ≡φΛ,ψ = φ(x);x ∈ Λ of (3.1) is unique and called a (discrete) harmonicfunction on Λ.

Let GΛ(x, y), x ∈ Λ, y ∈ Λ be the Green function (potential kernel) forthe discrete Laplacian ∆Λ,0 with boundary condition 0, i.e., the solution ofequations

−∆GΛ(x, y) = δ(x, y), x ∈ Λ ,

GΛ(x, y) = 0, x ∈ ∂+Λ ,(3.2)

where δ(x, y) is the Kronecker’s δ, and ∆ acts on the variable x and y isthought of as a parameter. In fact, GΛ(x, y);x, y ∈ Λ is the inverse matrixof −∆Λ(x, y);x, y ∈ Λ so that we shall denote

GΛ(x, y) = (−∆Λ)−1(x, y) ,

note that ∆Λ(x, y) is the kernel of ∆Λ ≡ ∆Λ,0: ∆Λφ(x) =∑

y∈Λ ∆Λ(x, y)φ(y).

Mean, Covariance and Characteristic Functions

The next proposition is an extension of the fact stated in Sect. 2.6 when theboundary conditions are ψ ≡ 0.


Proposition 3.1. (1) Under µψΛ, φ = φ(x);x ∈ Λ is Gaussian with mean

φΛ,ψ = φΛ,ψ(x);x ∈ Λ and covariance GΛ(x, y), i.e., µψΛ = N(φΛ,ψ, GΛ).

In particular, for x, y ∈ Λ

EµψΛ [φ(x)] = φΛ,ψ(x), (3.3)

EµψΛ [φ(x);φ(y)] = GΛ(x, y), (3.4)

where

Eµ [φ(x);φ(y)] := Eµ [φ(x)− Eµ[φ(x)] φ(y)− Eµ[φ(y)]]

stands for the covariance of φ(x) and φ(y) under µ.(2) The characteristic function of µψ

Λ is given by

EµψΛ

[e√−1(ξ,φ)Λ

]= exp

√−1(ξ, φΛ,ψ)Λ −

12(ξ, (−∆Λ)−1ξ)Λ

for ξ ∈ RΛ.

(3) If φ is µ0Λ-distributed, then φ+ φΛ,ψ is µψ

Λ-distributed.

Proof. A careful rearrangement of the sum in the HamiltonianHψΛ (φ) applying

the summation by parts formula leads us to

HψΛ (φ) = −1

2((φ− φΛ,ψ),∆Λ(φ− φΛ,ψ)

)Λ

+12

∑x∈Λ,y/∈Λ|x−y|=1

φΛ,ψ(y)∇φΛ,ψ(〈y, x〉) ,

for every φ ∈ RΛ. This is an extension of (2.19) for 0-boundary conditions

and a discrete analogue of Green-Stokes’ formula. Note that the second termin the right hand side depends only on the boundary conditions ψ and not onφ. Therefore, we have that

µψΛ(dφΛ) =

1

ZψΛ

exp

12((φ− φΛ,ψ),∆Λ(φ− φΛ,ψ))Λ

dφΛ

with a proper normalization constant ZψΛ . This immediately shows the asser-

tions (1) and (2). The third assertion (3) follows from (1) or (2).It might be useful to give another proof for (1). Actually, to show (3.3),

set its left hand side as h(x). Then, h(x) satisfies the equation (3.1). In fact,the boundary condition is obvious and, for x ∈ Λ,

∆h(x) = EµψΛ [∆φ(x)] = −Eµψ

Λ

[∂Hψ

Λ

∂φ(x)

]= 0

128 T. Funaki

by the integration by parts under µψΛ. The uniqueness of solutions of (3.1)

proves (3.3). The proof of (3.4) is similar; one may check its left hand sidesolves (3.2) in place of GΛ(x, y). This can be shown again by the integrationby parts.

It is standard to calculate the mean, covariance and other higher momentsfrom the characteristic function. Indeed, for instance, (3.4) has the third proof:We may assume ψ ≡ 0 by translating the field φ by φΛ,ψ and in this case

Eµ0Λ [(ξ, φ)2Λ] = − d2

dα2Eµ0

Λ

[e√−1α(ξ,φ)Λ

] ∣∣∣∣∣α=0

= − d2

dα2e−

α22 (ξ,(−∆Λ)−1ξ)Λ

∣∣∣∣∣α=0

= (ξ, (−∆Λ)−1ξ)Λ .

Then, the identity (3.4) follows by taking ξ = δx, δy or δx + δy in this formulaand computing their differences, where δx(·) = δ(x, ·).

In particular, for µN ≡ µ0DN

with Λ = DN taking D = (−1, 1)d and with0-boundary conditions, we have EµN [φ(O)] = 0 and the variance behaves asN →∞

EµN [φ(O)2] = (−∆DN)−1(O,O) ≈

1, d ≥ 3 ,logN, d = 2 ,N, d = 1 ,

(3.5)

where ≈ means that the ratio of the both sides stays uniformly positive andbounded. The number of the sites neighboring to each site is 2d and thereforeone can expect that, as the lattice dimension d increases, the fluctuations ofthe interfaces become smaller, in other words, they gain more stiffness. Thebehavior (3.5) of the variance agrees with this observation. When d ≥ 3, thesecond moment stays bounded as N → ∞ and accordingly ϕ-Gibbs measureis normalizable in the sense that it admits the thermodynamic limit, see Sect.3.2. For general convex potentials V , Brascamp-Lieb inequality gives at leastthe corresponding upper bound in (3.5), see Sect. 4.2. When d = 1, φ(x) isessentially the pinned Brownian motion with discrete time parameter x ∈(−N,N) ∩ Z and therefore (3.5) is standard.

Random Walk Representation

Let X = Xtt≥0 be the simple random walk on Zd with continuous time

parameter t, i.e., the generator of X is the discrete Laplacian ∆ and the jumpof X to the adjacent sites is accomplished by choosing one of them with equalprobabilities after an exponentially distributed waiting time with mean 1

2d .Let τΛ be the exit time of X from the region Λ:

τΛ := inft ≥ 0; Xt ∈ Λc .


The transition probability of the simple random walk on Λ with absorbingboundary ∂+Λ is denoted by pΛ(t, x, y) ≡ Ex[1y(Xt), t < τΛ], t ≥ 0, x, y ∈Λ, where Ex[ · ] stands for the expectation for X starting at x: X0 = x. Then,the following representations are easy.

Proposition 3.2. For every x, y ∈ Λ, we have

φΛ,ψ(x) = Ex [ψ (XτΛ)] , (3.6)

GΛ(x, y) = Ex

[∫ τΛ

0

1y (Xt) dt]

=∫ ∞

0

pΛ(t, x, y) dt. (3.7)

The middle term of (3.7) is the average of the occupation time of Xt at ybefore leaving Λ.

3.2 Gaussian Systems on Zd

Let us assume that a harmonic function ψ = ψ(x);x ∈ Zd ∈ R

Zd

is givenon the whole lattice Z

d and consider the Gaussian finite volume ϕ-Gibbsmeasures µψ

Λ ∈ P(RZd

) for all connected Λ Zd. We shall see that, if d ≥ 3,

µψΛ admits a weak limit µψ ∈ P(RZ

d

) as Λ Zd (i.e., along an increasing

sequence Λ(n)n=1,2,... satisfying ∪∞n=1Λ

(n) = Zd) and the limit µψ is a ϕ-

Gibbs measure (on Zd) corresponding to the potential V (η) = 1

2η2. A simple

but important class of the harmonic functions on Zd is given by ψ(x) = u·x+h

for u ∈ Rd and h ∈ R, where u · x denotes the inner product in R

d. The two-point correlation function of µψ decays slowly in algebraic (i.e., polynomial)order. The ∇ϕ-Gibbs measures exist for arbitrary dimension d.

Thermodynamic Limit

Since φΛ,ψ = ψ on Λ for every harmonic function ψ, from (3.3), the mean ofφ under µψ

Λ is ψ. The covariance of µψΛ is GΛ(x, y), recall (3.4). Let G(x, y) ≡

(−∆)−1(x, y) = G(x−y) be the Green function (of 0th order) of the operator∆ on Z

d, i.e.,

G(x, y) =∫ ∞

0

p(t, x, y) dt, x, y ∈ Zd ,

where p(t, x, y) denotes the transition probability of the simple random walkX on Z

d. It is well-known that G(x, y) < ∞ if and only if d ≥ 3 (i.e., if Xis transient). This can be also seen from an explicit formula for G(x) by theFourier transform:

G(x) =1

2(2π)d

∫Td

e√−1x·θ

∑dj=1(1− cos θj)

dθ , (3.8)

where Td = (−π, π]d and dθ =

∏dj=1 dθj . Since pΛ(t, x, y) ↑ p(t, x, y) as Λ

Zd, we have

130 T. Funaki

limΛZd

GΛ(x, y) = G(x, y), x, y ∈ Zd .

To study the limit of µψΛ as Λ Z

d, recalling Proposition 3.1-(3), we mayassume ψ ≡ 0. Let µ ∈ P(RZ

d

) be the distribution of a Gaussian systemφ = φ(x);x ∈ Z

d with mean 0 and covariance G(x, y), whose characteristicfunction is given by

Eµ[e√−1(ξ,φ)

]= e−

12 (ξ,(−∆)−1ξ), ξ ∈ C0(Zd) ,

where (ξ, φ) =∑

x∈Zd ξ(x)φ(x) is the inner product (on the whole lattice Zd)

and C0(Zd) denotes the family of all ξ : Zd → R satisfying ξ(x) = 0, x /∈ Λ for

some Λ Zd. The convergence of the covariances

(ξ, (−∆Λ)−1ξ) =∑

x,y∈Zd

GΛ(x, y)ξ(x)ξ(y)

−→∑

x,y∈Zd

G(x, y)ξ(x)ξ(y) = (ξ, (−∆)−1ξ)

for ξ ∈ C0(Zd) (note that both sums are finite) implies the convergence of thecharacteristic functions so that µ0

Λ weakly converges to µ as Λ Zd on the

space RZ

d

endowed with the product topology.In fact, the convergence holds under stronger topologies. To see that, let

us introduce weighted 2-spaces on Zd

2(Zd, z) := φ ∈ RZ

d

; ‖φ‖2z :=∑x∈Zd

φ(x)2z(x) <∞

for weight functions z = z(x) > 0;x ∈ Zd. We shall especially concern

with two classes of spaces (2α, ‖ · ‖α) and (2r, ‖ · ‖r) for α, r > 0 takingz(x) = (1 + |x|)−α and z(x) = e−2r|x|, respectively.

Proposition 3.3. Assume d ≥ 3. Then µ0Λ weakly converges to µ as Λ Z

d

on the spaces 2α, α > d or 2r, r > 0.

Proof. The proof is concluded once the tightness of µ0ΛΛ on these spaces is

shown. However, since

0 ≤ Eµ0Λ[φ(x)2

]= GΛ(x, x) ≤ G(x, x) = G(O) <∞ ,

we have that supΛ Eµ0

Λ [‖φ‖2α] <∞ (if α > d) and supΛ Eµ0

Λ [‖φ‖2r] <∞. Theseuniform estimates imply the tightness noting that the imbeddings 2α1

⊂ 2α2

or 2r1⊂ 2r2

are compact when 0 < α1 < α2 or 0 < r1 < r2, respectively.

Remark 3.1. Fernique’s theorem for Gaussian random variables shows that

supΛEµ0

Λ [eε‖φ‖2α ] <∞, α > d ,


for some ε > 0, e.g. [232]. In particular, this implies supΛ Eµ0

Λ [‖φ‖kα] <∞ for

α > d and k ∈ N. The same uniform estimates hold for the norm ‖φ‖r.

Finally, let us show that the limit µ of µ0Λ is actually a ϕ-Gibbs measure.

The argument below is applicable also when the potentials V are general. Wecall a function g = g(φ) on R

Zd

local if it is FΛ-measurable for some Λ Zd

and the smallest Λ is denoted by supp (g).

Proposition 3.4. µ is a ϕ-Gibbs measure.

Proof. Let Σ Zd, FΣ-measurable bounded function f and FΣc -measurable

bounded local function g be given. Then, if Λ is sufficiently large such thatΣ ∪ supp (g) ⊂ Λ, we have

Eµ0Λ [fg] = Eµ0

Λ [g(φ)EµφΣ [f ]] .

However, under the limit Λ Zd, the left and the right hand sides converge to

Eµ[fg] and Eµ[g(φ)EµφΣ [f ]], respectively. Thus we obtain the DLR equation

for µ.

In summary, we see that for every harmonic function ψ on Zd a weak limit

µψ ∈ P(RZd

) of µψΛ as Λ Z

d exists and it is a ϕ-Gibbs measure if d ≥ 3.

Remark 3.2. (1) Take ψ(x) = u · x + h for the boundary conditions. Thenthe limit µψ is not the same for different h. In this sense the massless field isquite sensitive on the boundary conditions.(2) As the results on a massive model ([15], see Remark 3.3 below) suggest,the extremal sets of the class of ϕ-Gibbs measures might be much wider thanµψ;ψ are harmonic. However, ∇ϕ-Gibbs measures are exhausted, under theassumption of shift invariance (and temperedness), by the convex hull of thegradient fields associated to these fields (see Sect. 9).

Long Correlations

As we have seen, the two-point correlation function of φ under the ϕ-Gibbsmeasure µψ coincides with the Green function G(x, y) of the simple randomwalk on Z

d, and it decays only algebraically (or in polynomial order) and notexponentially fast. In this sense the field has long dependence.

Proposition 3.5. Assume d ≥ 3. Then the two-point correlation function ofµψ is always positive and behaves like

Eµψ

[φ(x);φ(y)] ∼ k1

|x− y|d−2

as |x− y| → ∞, where |x− y| stands for the Euclidean distance and ∼ meansthat the ratio of both sides converges to 1. The constant k1 is determined by

k1 =12

∫ ∞

0

(2πt)−d2 e−

12t dt .

132 T. Funaki

Proof. The conclusion follows from the behavior G(x) ∼ k1/|x|d−2, |x| →∞ of the Green function established by Ito-McKean [155] (2.7, p.121); seealso Spitzer [238], p.308, P1 for d = 3 and Lawler [185]. Note that, in thesereferences, ∆ is normalized by dividing it by 2d.

This proposition, in particular, implies that one of the important thermo-dynamic quantities called the compressibility diverges in massless model:

∑x∈Zd

Eµψ

[φ(x);φ(y)] = ∞ .

Note that k1/|x − y|d−2, x, y ∈ Rd is the Green function (of the continuum

Laplacian) on Rd and the constant k1 has another expression

k1 = (4πd/2)−1Γ

(d

2− 1

)=

1(d− 2)Ωd

,

where Ωd is the surface area of the d− 1 dimensional unit sphere. For generalpotential V , similar asymptotics for the two-point correlation function areobtained by [202], see Sect. 4.3.

∇ϕ-Gaussian Field

We have required the assumption d ≥ 3 to construct ϕ-field on the infinitevolume lattice Z

d, but for its gradient the thermodynamic limit exists inarbitrary dimensions d including d = 1, 2. To see this, we first notice the nextlemma which is immediate from Proposition 3.1-(1). Recall that

∇iφ(x) := φ(x+ ei)− φ(x)(≡ ∇φ(x+ ei)), x ∈ Zd, 1 ≤ i ≤ d .

The bond 〈x+ei, x〉 is simply denoted by x+ei and, in particular, ei sometimesrepresents the bond 〈ei, O〉.

Lemma 3.6. Let Λ Zd and ψ ∈ R

Zd

be given. Then we have

EµψΛ [∇iφ(x)] = ∇iφΛ,ψ(x) , (3.9)

EµψΛ [∇iφ(x);∇jφ(y)] = ∇i,x∇j,yGΛ(x, y) , (3.10)

for every x, y ∈ Λ, 1 ≤ i, j ≤ d, where ∇i,x and ∇j,y indicate that theseoperators act on the variables x and y, respectively.

When d = 1, 2, although GΛ(x, y) itself is not convergent as Λ Zd, its

normalization

GΛ(x, y) :=∫ ∞

0

pΛ(t, x, y)− pΛ(t, 0, 0) dt

admits the finite limit


G(x, y) :=∫ ∞

0

p(t, x, y)− p(t, 0, 0) dt, x, y ∈ Zd ,

which is called the (normalized 0th order) Green function. One can replaceGΛ in the right hand side of (3.10) with GΛ so that the covariance of the∇ϕ-field has the limit as Λ Z

d. We therefore obtain the next proposition.

Proposition 3.7. For a harmonic function ψ on Zd, let µψ,∇ ∈ P(R(Zd)∗)

be the distribution of the Gaussian field on (Zd)∗ with mean and covariance

Eµψ,∇[∇iφ(x)] = ∇iψ(x) ,

Eµψ,∇[∇iφ(x);∇jφ(y)] = ∇i,x∇j,yG(x, y) ,

respectively. Then µψ,∇ is a ∇ϕ-Gibbs measure (see Definition 2.2 and Sect.9).

We have a family of ∇ϕ-Gibbs measures µψu,∇;u ∈ Rd by taking

ψ(x) ≡ ψu(x) := u · x. When d ≥ 3, if φ = φ(x);x ∈ Zd is µψ-

distributed, then its gradient field ∇φ = ∇φ(b); b ∈ (Zd)∗ is µψ,∇-distributed. When d = 1, the Green function is given by G(x) = − 1

2 |x|, whichproves Eµψ,∇

[∇φ(x);∇φ(y)] = δ(x− y)(≡ δ(x, y)). This, in particular, showsthat ∇φ(b); b ∈ (Z)∗ is an independent Gaussian system in one dimension.When d = 2, the Green function behaves like

G(x) = − 12π

log |x|+ c0 +O(|x|−2), |x| → ∞ ,

see Stohr [241], Spitzer [238].

3.3 Massive Gaussian Systems

In the present subsection, we study the ϕ-field associated with the massiveHamiltonian Hm(φ) introduced in Remark 2.2-(1). The mass term of Hm ac-tually has a strong influence on the field. It is localized and exhibits verydifferent features from the massless case. In particular, (1) the ϕ-Gibbs mea-sure exists for arbitrary dimensions d ≥ 1, (2) the effect of the boundaryconditions is weak (see Corollary 3.9 below) and (3) the two-point correla-tion function decays exponentially fast; in other words, the field has a strongmixing property.

Massive Gaussian ϕ-Gibbs Measures

For Λ Zd and the boundary condition ψ ∈ R

Zd

, the finite volume ϕ-Gibbsmeasure µψ

Λ;m ∈ P(RΛ) having mass m > 0 is defined by

µψΛ;m(dφΛ) :=

1

ZψΛ;m

e−HψΛ;m(φ) dφΛ

134 T. Funaki

where

HψΛ;m(φ) = Hψ

Λ (φ) +m2

2

∑x∈Λ

φ(x)2

is the massive Hamiltonian and ZψΛ;m is the normalization constant. As before,

we sometimes regard µψΛ;m ∈ P(RZ

d

). The ϕ-Gibbs measure µ ≡ µm ∈ P(RZd

)(on Z

d) having mass m is defined by means of the DLR equation with the localspecifications µψ

Λ;m in place of µψΛ in Definition 2.1. We are always concerning

the case where V (η) = 12η

2 throughout this section.

Finite Systems

Similarly to the massless case, the mean and covariance of the field φ underµψ

Λ;m can be expressed as solutions of certain difference equations and admitthe random walk representation. Indeed, consider the equations (3.1) and (3.2)with ∆ replaced by ∆−m2, respectively, i.e.,

(∆−m2)φ(x) = 0, x ∈ Λ ,

φ(x) = ψ(x), x ∈ ∂+Λ ,(3.11)

and −(∆−m2)GΛ;m(x, y) = δ(x, y), x ∈ Λ ,

GΛ;m(x, y) = 0, x ∈ ∂+Λ ,(3.12)

for y ∈ Λ. The solution of (3.11) is denoted by φ = φΛ,ψ;m, while GΛ;m(x, y)is sometimes written as

GΛ;m(x, y) = (−∆Λ +m2)−1(x, y) .

Consider the simple random walkX = Xtt≥0 on Zd as before and let σ be an

exponentially distributed random variable with mean 1m2 being independent

of X. The random walk X is killed at the time σ, in other words, it jumps toa point ∆(/∈ Z

d) at σ and stays there afterward. Every function ψ on Zd is

extended to Zd ∪∆ setting ψ(∆) = 0. The next proposition is an extension

of Propositions 3.1 and 3.2 to the massive case. The proof is similar.

Proposition 3.8. Under µψΛ;m, φ = φ(x);x ∈ Λ is Gaussian with mean

φΛ,ψ;m(x) and covariance GΛ;m(x, y). In particular, we have for x, y ∈ Λ

EµψΛ;m [φ(x)] = φΛ,ψ;m(x) = Ex [ψ (XτΛ∧σ)] , (3.13)

EµψΛ;m [φ(x);φ(y)] = GΛ;m(x, y) = Ex

[∫ τΛ∧σ

0

1y (Xt) dt]. (3.14)


Thermodynamic Limit

The random walk representation is useful to observe that the limit of µψΛ;m

as Λ Zd does not depend on the boundary condition ψ if it grows at most

in polynomial order as |x| → ∞. This property for massive field is essentiallydifferent from the massless case. If ψ grows exponentially fast, its effect mayremain in the limit of µψ

Λ;m, see Remark 3.3 below.

Corollary 3.9. If the function ψ on Zd satisfies |ψ(x)| ≤ C(1 + |x|n) for

some C, n > 0, then we have for every x ∈ Zd

limΛZd

φΛ,ψ;m(x) = 0 .

Proof. To prove the conclusion, from (3.13), it suffices to show that Px(τΛ<

σ) = o(−n) as → ∞, where Λ = [−, ]d ∩ Zd. However, since P (σ >√

) = e−m2√, this follows from the large deviation type estimate on τΛ:Px(τΛ

<√) ≤ e−C for some C > 0.

The covariance GΛ;m(x, y) of µψΛ;m converges as Λ Z

d to Gm(x, y) =Gm(x− y), where Gm(x) is defined by

Gm(x) :=∫ ∞

0

e−m2tp(t, x) dt

=1

(2π)d

∫Td

e√−1x·θ

2∑d

j=1(1− cos θj) +m2dθ ,

where p(t, x) = p(t, x,O). Note that, since m > 0, Gm(x) < ∞ for all d ≥ 1.The function Gm(x, y) is sometimes written as (−∆+m2)−1(x, y) and calledthe Green function ofm2th order of the operator∆ on Z

d. When the boundarycondition ψ satisfies the condition in Corollary 3.9, the thermodynamic limitµm ∈ P(RZ

d

) of µψΛ;m exists and it is the Gaussian measure with the mean 0,

covariance Gm(x, y) and characteristic function

Eµm

[e√−1(ξ,φ)

]= e−

12 (ξ,(−∆+m2)−1ξ) ,

for ξ ∈ C0(Zd). The limit measure is independent of the choice of ψ as longas it satisfies the condition in Corollary 3.9.

Remark 3.3. Benfatto et al. [15] characterized the structure of the class ofall massive Gaussian ϕ-Gibbs measures on R

Z when d = 1. Their result showsthat its extremal set E is given by

E = µα−,α+ ; (α−, α+) ∈ R2 ,

where µα−,α+ is the limit of the sequence of finite ϕ-Gibbs measures µψ[−,];m

with boundary condition ψ satisfying

136 T. Funaki

α± = (1− ρ2) lim→∞

ρψ(±(+ 1))

for certain ρ ∈ (0, 1). For instance, if ψ is replaced by ψ+u·x+h, the constantsα± are the same. In this respect, ϕ-Gibbs measure is not much sensitive tothe boundary conditions. The mass term m2

2

∑φ(x)2 has an effect to localize

the field. In fact, the above mentioned result implies that the shift invariantmassive Gaussian ϕ-Gibbs measure is unique in one dimension.

Short Correlations

The exponential decay of the two-point correlation function

Eµm [φ(x);φ(y)] = Eµm [φ(x)φ(y)] = Gm(x− y)

under µm is precisely stated in the next proposition. The proof of (2) is givenby a simple calculation based on the residue theorem.

Proposition 3.10. (1) When d ≥ 2, for each ω ∈ Sd−1 (i.e., |ω| = 1),determine b(ω) = bm(ω) ∈ R

d and γ ∈ R \ 0 by

12d

∑|y|=1

eb·y =m2

2d+ 1 ,

12d

∑|y|=1

yeb·y = γω .

Then, we haveGm(x) ∼ Cd|x|−

d−12 e−bm(x/|x|)·x

as |x| → ∞ for some Cd > 0.(2) When d = 1, Gm(x) has an explicit formula:

Gm(x) =12π

∫ π

−π

e√−1xθ

2(1− cos θ) +m2dθ =

e−m|x|

2 sinh m, x ∈ Z

where m > 0 is the solution of an algebraic equation cosh m = m2

2 + 1. Inparticular, m behaves such that m = m+O(m2) as m ↓ 0.

Remark 3.4. Let Cm(x− y) = Cm(x, y), x, y ∈ Rd be the Green function of

m2th order for the (continuum) Laplacian on Rd, i.e.,

Cm(x) := (−∆+m2)−1δ(x) =1

(2π)d

∫Rd

e√−1x·p

p2 +m2dp, x ∈ R

d .

The function Cm(x) has an expression by means of the modified Bessel func-tions. For example, when d = 3, we have


Cm(x) =1

4π|x|e−m|x| ,

and, for general d ≥ 1, it behaves

Cm(x) ∼ const md−32 |x|− d−1

2 e−m|x| ,

as m|x| → ∞, see [138] p.126. Note that the exponential decay rates for Gm

and Cm are different, see also [234] p.257 for d = 2.

Proposition 3.10 gives the exact exponential decay rates of the Green func-tion Gm for m > 0. However, in order just to see the exponentially decayingproperty of Gm, one can apply the Aronson’s type estimate on the tran-sition probability p(t, x, y) = p(t, x − y) of the simple random walk on Z

d:

p(t, x) ≤ min

C

td/2e−|x|2/Ct, 1

, t > 0, x ∈ Z

d , (3.15)

for some C > 0; see [202] §2, [50] for general random walks. In fact, we dividethe integral

Gm(x) =∫ ∞

0

e−m2tp(t, x) dt

into the sum of those on two intervals [0, |x|) and [|x|,∞). Then, on the firstinterval, if x = 0, we estimate the integrand as

e−m2tp(t, x) ≤ C

td/2e−|x|2/2Cte−|x|2/2Ct

≤ C

td/2e−1/2Cte−|x|2/2C|x| ≤ const e−|x|/2C ,

while on the second

e−m2tp(t, x) ≤ e−m2t

2 e−m22 |x| .

This proves that0 < Gm(x) ≤ Ce−c|x|, x ∈ Z

d ,

for some c, C > 0. See [202] Theorem B for non-Gaussian case.The Aronson’s type estimate is applicable to the massless case as well and,

though it is weaker than Proposition 3.5, we have the following:

0 < G(x) ≤ C

|x|d−2,

for some C > 0 when d ≥ 3. In fact, the change of the variables t = |x|2s inthe integral implies

G(x) =∫ ∞

0

p(t, x) dt ≤∫ ∞

0

C

td/2e−|x|2/Ct dt =

C

|x|d−2

∫ ∞

0

1sd/2

e−1/Cs ds .

Note that the last integral converges when d ≥ 3. See [202] Theorem C orTheorem 4.13 in Sect. 4.3 for non-Gaussian case.

138 T. Funaki

3.4 Macroscopic Scaling Limits

The random field φ = φ(x);x ∈ Zd is a microscopic object and our goal is

to study its macroscopic behavior. In this subsection, we discuss such problemunder the Gaussian measures µ =: µ0 (massless case, d ≥ 3) and µm,m > 0(massive case, d ≥ 1). Recall that µ and µm are ϕ-Gibbs measures on Z

d

obtained by the thermodynamic limit with boundary conditions ψ ≡ 0; seeSects. 3.2 and 3.3, respectively.

Scaling Limits

Let N be the ratio of typical lengths at macroscopic and microscopic levels.Then the point θ = (θi)d

i=1 ∈ Rd at macroscopic level corresponds to the

lattice point [Nθ] := ([Nθi])di=1 ∈ Z

d at microscopic level, recall Sect. 2.5.If x ∈ Z

d is close to [Nθ] in such sense that |x − [Nθ]| N , then x alsomacroscopically corresponds to θ. This means that observing the random fieldφ at macroscopic point θ is equivalent to taking its sample mean around themicroscopic point [Nθ]. Such averaging yields a cancellation in the fluctuationsof φ.

Motivated by these observations, let us consider the sample mean of φover the microscopic region ΛN = (−N,N ]d ∩ Z

d, which corresponds to themacroscopic box D = (−1, 1]d:

φN

:=1

(2N)d

∑x∈ΛN

φ(x) ,

note that (2N)d = |ΛN |. The field φ is distributed under µm for m ≥ 0.

Lemma 3.11. As N →∞, φN

converges to 0 in L2 under µm for all m ≥ 0.

Proof. If we denote G(x) by G0(x), we have for all m ≥ 0

Eµm

[(φ

N)2

]=

1(2N)2d

∑x,y∈ΛN

Eµm [φ(x)φ(y)] =1

(2N)2d

∑x,y∈ΛN

Gm(x− y) .

However, the Green functions admit bounds for some C, c > 0

0 < Gm(x) ≤

C

|x|d−2, m = 0, d ≥ 3 ,

Ce−c|x|, m > 0, d ≥ 1 ,

which prove the conclusion.

This lemma is the law of large numbers for ϕ-field and the next naturalquestion is to study the fluctuation of φ

Naround its limit 0 under a proper

rescaling. As we shall see, the necessary scalings will change according asm = 0 (i.e., massless case) or m > 0 (i.e., massive case) due to the differencein the mixing property of the field φ.


Fluctuations in Massive ϕ-Gaussian Field

First, let us consider the massive case: m > 0. Then the right scaling for thefluctuation of φ

Nwill be

ΦN := (2N)d2 φ

N ≡ 1

(2N)d2

∑x∈ΛN

φ(x) . (3.16)

Since (2N)d2 = |ΛN |

12 , this is the usual scaling for the central limit theorem;

recall that φ = φ(x);x ∈ Zd distributed under µm has a “nice” exponential

mixing property when m > 0.

Proposition 3.12. The fluctuation ΦN weakly converges to the Gaussiandistribution N(0,m−2) with mean 0 and variance m−2 as N →∞.

Proof. Since ΦN is Gaussian distributed with mean 0, the conclusion followsfrom the convergence of its variance:

Eµm

[(ΦN

)2]

=1

(2N)d

∑x,y∈ΛN

Gm(x− y) −→N→∞

1m2

,

note (1) in the next remark.

Remark 3.5. From Bricmont et al. [39] Proposition A1 (p. 294), we havefor µm,m > 0(1)

∑x∈Zd

Eµm [φ(O)φ(x)] = m−2,

(2)∑

x∈Zd

Eµm [φ(O)∇iφ(x)] ∼ constm−1 (m ↓ 0),

(3)∑

x∈Zd

Eµm [∇iφ(O)∇iφ(x)] is absolutely converging for each m and stays

bounded as m ↓ 0 (see Lemma 3.13 below).However, if i = j,(4)

∑x∈Zd

Eµm [∇iφ(O)∇jφ(x)] ∼ const | logm| (m ↓ 0).

Loosely speaking, as m ↓ 0, φ is expected to converge to the massless field sothat its covariances (or those of its gradients) might behave like |x− y|2−d (ormaking its gradients in x), and this may prove that

∑x∈Zd

Eµm [∇iφ(O)∇jφ(x)] ≈∫ R

r(2−d)−2 · rd−1 dr ≈ logR ,

where R ≈ m−1 is the correlation length.

140 T. Funaki

Fluctuations in Massless ϕ-Gaussian Field

Next, let us consider the massless case: m = 0 and d ≥ 3. Let φ be µ0-distributed. Since the variance m−2 of the limit distribution of ΦN under µm

diverges as m ↓ 0, the scaling (3.16) must not be correct in the masslesscase. However, if we further scale-down the value of ΦN dividing it by N andintroduce

ΦN :=1NΦN ≡ 1

(2N)d2 ·N

∑x∈ΛN

φ(x) , (3.17)

then it has the limit under µ0. In fact, the variance of ΦN behaves

Eµ0

[(ΦN

)2]

= 2−dN−d−2∑

x,y∈ΛN

G(x− y)

∼ k1N−2

∑x∈ΛN

|x|2−d ∼ k1N−2

∫|θ|≤N

|θ|2−d dθ

= k1N−2

∫ N

0

r(2−d)+(d−1) dr = O(1) .

Therefore, (3.17) is the right scaling when m = 0. This actually coincideswith the interpretation stated in Sect. 2.5: φ = φ(x);x ∈ Z

d represents theheight of an interface embedded in d + 1 dimensional space so that both x-and φ-axes should be rescaled by the factor 1/N at the same time.

If we introduce random signed measures on Rd by

ΦN (dθ) :=1

Nd2 +1

∑x∈Zd

φ(x)δx/N (dθ), θ ∈ Rd , (3.18)

then ΦN in (3.17) is represented as ΦN = 2−d/2〈ΦN (·), 1D〉, where 〈ΦN (·), f〉stands for the integral of f = f(θ) under the measure ΦN (·). In this way,studying the limit of ΦN is reduced to investigating more general problem forthe properly scaled empirical measures of φ.

Fluctuations in Massless ∇ϕ-Gaussian Field

When f = f(θ) has the form f = − ∂g∂θi

with certain g = g(θ), we can rewrite〈ΦN (·), f〉 as

〈ΦN (·), f〉 =⟨ΦN (·),− ∂g

∂θi

⟩= −N− d

2−1∑x∈Zd

φ(x)∂g

∂θi(x/N)

∼ −N− d2−1

∑x∈Zd

φ(x) ·Ng((x+ ei)/N)− g(x/N)

= −N− d2

∑x∈Zd

(φ(x− ei)− φ(x))g(x/N)

= N− d2

∑x∈Zd

∇iφ(x)g((x+ ei)/N) .


The second line is the approximation of ∂g∂θi

by its discrete derivatives. Thisrearrangement, in particular, implies that the scaling in ΦN (dθ) coincides withthe usual one of the central limit theorem, if one deals with the correspondinggradient fields ∇φ = ∇φ(x);x ∈ Z

d instead of φ.Thus it is natural to introduce the scaled empirical measures of ∇φ =

∇iφ; 1 ≤ i ≤ d:

ΨNi (dθ) ≡ ΨN

i (dθ;u) :=1

Nd2

∑x∈Zd

∇iφ(x)− uiδx/N (dθ) . (3.19)

The field ∇φ is µ∇u -distributed, where µ∇

u , u = (ui)di=1 ∈ R

d is the ∇ϕ-Gibbsmeasure µψu,∇ having boundary conditions ψ(x) = ψu(x) ≡ u · x obtainedin Proposition 3.7. Note that ui = Eµ∇

u [∇iφ(x)] and, since ∇iφ(x) − ui =∇i(φ−ψu)(x), the distribution of ΨN

i (dθ;u) under µ∇u coincides with that of

ΨNi (dθ; 0) under µ∇

0 . We may therefore assume u = 0 to study the limit. Thelimit of the variance Eµ∇

0[〈ΨN

i , g〉2]

as N → ∞ can be computed based onthe next lemma.

Lemma 3.13. ∑y∈Zd

Eµ∇0 [∇iφ(O)∇iφ(y)] =

1d.

Proof. Each term in the sum can be rewritten as

Eµ [(φ(ei)− φ(O))(φ(y + ei)− φ(y))]

=1

(2π)d

∫Td

2e√−1y·θ − e

√−1(y−ei)·θ − e

√−1(y+ei)·θ

2∑d

j=1(1− cos θj)dθ

=1

(2π)d

∫Td

e√−1y·θ 1− cos θi∑d

j=1(1− cos θj)dθ ,

which implies

d∑i=1

Eµ∇0 [∇iφ(O)∇iφ(y)] =

1(2π)d

∫Td

e√−1y·θ dθ = δ(y) .

The conclusion is shown by taking the sum in y ∈ Zd of the both sides of this

identity.

4 Random Walk Representationand Fundamental Inequalities

We are now at the position to enter into the study of the ∇ϕ interface modelfor general convex potentials V satisfying the three basic conditions (V1)-(V3)

142 T. Funaki

in (2.2). We shall first establish in this section three fundamental tools for ana-lyzing the model, i.e., Helffer-Sjostrand representation, FKG (Fortuin-Kasteleyn-Ginibre) inequality and Brascamp-Lieb inequality. Helffer-Sjostrand representation describes for the correlation functions under theGibbs measures by means of a certain random walk in random environments.Its original idea comes from [144], [237]. This representation readily impliesFKG and Brascamp-Lieb inequalities. The latter is an inequality between thevariances of non-Gaussian fields and those of Gaussian fields, which we canexplicitly compute as we have seen in Sects. 3.1 and 3.2. In particular, uniformmoment estimates on the non-Gaussian fields are obtained and these make uspossible to construct ∇ϕ-Gibbs measures on (Zd)∗ (for every d ≥ 1) and ϕ-Gibbs measures on Z

d (for d ≥ 3) by passing to the thermodynamic limit.The arguments in this section heavily rely on the convexity of the potentialV , i.e., the attractiveness of the interaction.

4.1 Helffer-Sjostrand Representation and FKG Inequality

Idea Behind the Representation

Let us shortly explain the idea behind the Helffer-Sjostrand representation.It gives the following identity for the covariance of F = F (φ) and G = G(φ)under the Gibbs measure µ:

Eµ[F ;G] =∑x∈Zd

∫ ∞

0

Eµ[∂F (x, φ0)∂G(Xt, φt)] dt . (4.1)

In the right hand side, φt = φt(x);x ∈ Zd is the ϕ-dynamics defined by the

SDEs (2.13) with µ-distributed initial data φ0, Xt is the random walk on Zd

starting at x with (temporary inhomogeneous) generator Qφt defined by

Qφf(x) =∑

y:|x−y|=1

V ′′(φ(x)− φ(y)) f(y)− f(x) ,

for f : Zd → R. Indeed, assuming Eµ[G] = 0, let H be the solution of the

Poisson equation −LH = G, where L is the generator of φt determined by(2.15) with Γ = Z

d. Then, from (2.14)

Eµ[F ;G] = Eµ[F (−LH)] =∑x∈Zd

Eµ[∂F (x, φ)∂H(x, φ)] . (4.2)

However, a simple computation (cf. (4.7) below) shows

∂(LH)(x, φ) = L∂H(x, φ) + (Q∂H(·, φ)) (x) ≡ (L+Q)∂H (x, φ)

and therefore


∂H(x, φ) = (L+Q)−1∂(LH) = E(x,φ)

[∫ ∞

0

∂G(Xt, φt) dt].

This implies the identity (4.1). The above argument is rather formal and, inparticular, one should replace the measure µ with the finite volume Gibbsmeasure [77] or with the Gibbs measure for ∇ϕ-field [135]. Note that theconvexity condition on V (i.e., V ′′ ≥ 0) is essential for the existence of therandom walk Xt.

Precise Formulation

Let the finite region Λ Zd and the boundary condition ψ ∈ R

Zd

be given. Weshall consider slightly general Hamiltonian having external field (chemicalpotential) ρ = ρ(x);x ∈ Λ ∈ R

Λ:

Hψ,ρΛ (φ) = Hψ

Λ (φ)− (ρ, φ)Λ (4.3)

and the corresponding finite volume ϕ-Gibbs measure

µψ,ρΛ (dφΛ) =

1

Zψ,ρΛ

e−Hψ,ρΛ (φ) dφΛ ∈ P(RΛ) , (4.4)

where Zψ,ρΛ is the normalization constant. This generalization will be useful

for the proof of Brascamp-Lieb inequality, cf. Lemma 4.6 and Theorem 4.9.The operator Lψ,ρ

Λ defined by

Lψ,ρΛ F (φ) :=eHψ,ρ

Λ (φ)∑x∈Λ

∂

∂φ(x)

e−Hψ,ρ

Λ (φ) ∂F

∂φ(x)

=∑x∈Λ

∂2F

∂φ(x)2− ∂Hψ,ρ

Λ

∂φ(x)∂F

∂φ(x)

for F = F (φ) ∈ C2(RΛ) is symmetric under the measure µψ,ρΛ and the associ-

ated Dirichlet form is given by

E(F,G) ≡Eψ,ρΛ (F,G) := −Eµψ,ρ

Λ

[F · Lψ,ρ

Λ G]

=Eµψ,ρΛ [(∂F, ∂G)Λ] . (4.5)

Recall that ∂F is defined by

∂xF (φ) ≡ ∂F (x, φ) :=∂F

∂φ(x)

and(∂F, ∂G)Λ ≡ (∂F, ∂G)Λ(φ) :=

∑x∈Λ

∂F (x, φ)∂G(x, φ) .

144 T. Funaki

For each φ ∈ RΛ, the operator Qφ

Λ ≡ Qφ,ψΛ,0 is introduced by

QφΛf(x) :=

∑b∈Λ∗:yb=x

V ′′(∇(φ ∨ ψ)(b))∇(f ∨ 0)(b)

=∑

y∈Λ:|x−y|=1

V ′′(φ(x)− (φ ∨ ψ)(y))(f ∨ 0)(y)− f(x) ,

for x ∈ Λ and f = f(x);x ∈ Λ ∈ RΛ under the boundary conditions φ(x) =

ψ(x) and f(x) = 0 for x ∈ ∂+Λ. In particular, when V (η) = 12cη

2, c > 0,Qφ,ψ

Λ,0 = c∆Λ,0, which is independent of φ and ψ. We further consider theoperator

L ≡ Lψ,ρΛ := Lψ,ρ

Λ +QψΛ,0

acting on the functions F = F (x, φ) on Λ × RΛ, where Qψ

Λ,0F (x, φ) :=Qφ,ψ

Λ,0F (x, φ) is the operator acting on functions with two variables. The nextlemma is simple, but explains the reason why the operator Qφ,ψ

Λ,0 is useful.

Lemma 4.1. For every x ∈ Λ and F = F (φ), we have

[∂x, Lψ,ρΛ ] ≡ ∂xL

ψ,ρΛ − Lψ,ρ

Λ ∂x = −∑y∈Λ

∂2Hψ,ρΛ

∂φ(x)∂φ(y)∂y , (4.6)

∂Lψ,ρΛ F (x, φ) = L∂F (x, φ) . (4.7)

Proof. (4.6) is obvious from the definition of Lψ,ρΛ . (4.7) follows from (4.6) by

noting the symmetry of V ′′ and

∂2Hψ,ρΛ

∂φ(x)∂φ(y)=

∑z∈Λ:|x−z|=1

V ′′(φ(x)− (φ ∨ ψ)(z)), x = y ,

−V ′′(φ(x)− φ(y)) , |x− y| = 1 ,0 , otherwise ,

for x, y ∈ Λ.

Let φt ≡ φρt = φt(x);x ∈ Λ be the process on R

Λ generated by Lψ,ρΛ ,

i.e., the solution of the SDEs (2.9) with Γ = Λ and H = Hψ,ρΛ :

dφt(x) = −∑

y∈Λ:|x−y|=1

V ′(φt(x)− φt(y))dt

+ ρ(x)dt+√

2dwt(x), x ∈ Λ ,

φt(y) = ψ(y), y ∈ ∂+Λ .

(4.8)

Let Xt, t ≥ 0 be the random walk on Λ (or, more precisely, on Λ ∪∆) with temporally inhomogeneous generator Qφt

Λ (and with killing rate


∑y∈∂+Λ:|x−y|=1 V

′′(φt(x) − ψ(y)) at x ∈ ∂−Λ). Then, L is the genera-tor of (Xt, φt). Note that the random walk Xt exists since its jump rateV ′′(∇(φt ∨ ψ)(b)) is positive from our assumption (V3).

Theorem 4.2. (Helffer-Sjostrand representation) The correlation func-tion of F = F (φ) and G = G(φ) under µψ,ρ

Λ has the representation

Eµψ,ρΛ [F ;G] =

∑x∈Λ

∫ ∞

0

Eδx⊗µψ,ρΛ [∂F (x, φ0)∂G(Xt, φt), t < τΛ] dt . (4.9)

In the right hand side, δx ⊗ µψ,ρΛ indicates the initial distribution of (Xt, φt)

and δx ∈ P(Λ) is defined by δx(z) = δ(z − x). In particular, the distributionof φ0 is µψ,ρ

Λ and the random walk Xt starts at x. τΛ = inft > 0;Xt ∈ Λc isthe exit time of Xt from Λ.

Theorem 4.2 with a special choice of F (φ) = φ(x) and G(φ) = φ(y) givesthe following extension of the formula (3.7) combined with (3.4) for quadraticpotentials to general ones; note that ∂F (z, φ) = δ(x− z) in this case.

Corollary 4.3. For every x, y ∈ Λ,

Eµψ,ρΛ [φ(x);φ(y)] = Eδx⊗µψ,ρ

Λ

[∫ τΛ

0

1y (Xt) dt].

The function F = F (φ) on RΛ is called increasing if it satisfies ∂F =

∂F (x, φ) ≥ 0 so that it is nondecreasing under the semi-order on RΛ de-

termined by “φ1 ≥ φ2, φ1, φ2 ∈ RΛ ⇐⇒ φ1(x) ≥ φ2(x) for every x ∈ Λ”.

Theorem 4.2 immediately implies the following inequality.

Corollary 4.4. (FKG inequality) If F and G are both (L2-integrable) in-creasing functions, then we have

Eµψ,ρΛ [F ;G] ≥ 0 ,

namely,Eµψ,ρ

Λ [FG] ≥ Eµψ,ρΛ [F ]Eµψ,ρ

Λ [G] .

So far, we are concerned with the representation of correlation functions.The next proposition gives the formula for the expectation of φ(x), which isan extension of (3.3) with (3.6) for quadratic potentials. See [77] for the proof.

Proposition 4.5. For x ∈ Λ, we have

EµψΛ [φ(x)] =

∫ 1

0

Eδx⊗µsψΛ [ψ (XτΛ

)] ds .

Introducing the external field ρ has an advantage in the next lemma, whichis indeed one of the tricks commonly used in statistical mechanics. We shalldenote the variance of the random variable X under µ by

var (X;µ) = Eµ[(X − Eµ[X])2

].

146 T. Funaki

Lemma 4.6. Assume ρ, ν ∈ RΛ. Then we have

d

dsEµψ,sρ

Λ [φ(x)] = Eµψ,sρΛ [φ(x); (ρ, φ)Λ] , (4.10)

Eµψ,ρΛ

[exp

(ν, φ)Λ − Eµψ,ρ

Λ [(ν, φ)Λ]]

= exp∫ 1

0

ds

∫ s

0

var ((ν, φ)Λ;µψ,ρ+s1νΛ ) ds1

. (4.11)

The left hand side of (4.11) is the generating function of (ν, φ)Λ subtractedits mean and sometimes called the free energy in physics.

4.2 Brascamp-Lieb Inequality

A bound on the variances under non-Gaussian ϕ-Gibbs measures by thoseunder Gaussian ϕ-Gibbs measures is called the Brascamp-Lieb inequality.More precisely, for every ν = ν(x);x ∈ Λ ∈ R

Λ, we have

var (ν, φ)Λ ≤ var ∗(ν, φ)Λ ,

where φ in the left hand side is the field distributed under µψΛ determined from

the general convex potential V , while it is distributed in the right hand sideunder the Gaussian ϕ-Gibbs measures µψ,G

Λ determined from the quadraticpotential V ∗(η) = 1

2c−η2. Note that the difference of these two potentials

V − V ∗ is still convex. Brascamp-Lieb inequality claims that the strongerconvexity of the potential has larger effect on the ϕ-field to localize it aroundits mean. This looks plausible from physical point of view. Brascamp-Liebinequality does not give any information on the mean of φ itself.

The original proof due to [33], [34] of this inequality is rather complicated,but now a simpler proof based on Helffer-Sjostrand representation is available.We prepare a lemma, in which we denote R

Λ by L2(Λ, dx) with countingmeasure dx on Λ.

Lemma 4.7. (1) For every f, g ∈ RΛ, we have

(g,−Qφ,ψΛ,0f)Λ =

14

∑b∈Λ∗

V ′′(∇φ(b))∇f(b)∇g(b)

+∑

x∈Λ,y/∈Λ|x−y|=1

V ′′(φ(x)− ψ(y))f(x)g(x) .

In particular, Qφ,ψΛ,0 is symmetric on L2(Λ, dx) for every φ ∈ R

Λ.(2) As symmetric operators on L2(Λ× R

Λ, dx× µψ,ρΛ ), we have

−Lψ,ρΛ ≥ −Qψ

Λ,0 .


(3) For every φ ∈ RΛ, as symmetric operators on L2(Λ, dx), we have

−Qφ,ψΛ,0 ≥ −c−∆Λ,0 ,

where c− is the positive constant in (V3).

Proof. (1) is a simple calculation. From (1), we see that QψΛ,0 is symmetric on

L2(Λ×RΛ, dx×µψ,ρ

Λ ). On the other hand, the identity (4.5) for the Dirichletform implies that −Lψ,ρ

Λ is nonnegative and symmetric on this space. Thus(2) is shown. (3) is obvious from (1) by taking f = g and noting V ′′ ≥ c−.

Now we state the Brascamp-Lieb inequality between the variances underµψ,ρ

Λ and µψ,GΛ , which is the Gaussian finite volume ϕ-Gibbs measure deter-

mined from the quadratic potential V ∗ without external field.

Theorem 4.8. (Brascamp-Lieb inequality) For every ν ∈ RΛ, we have

var ((ν, φ)Λ;µψ,ρΛ ) ≤ var ((ν, φ)Λ;µψ,G

Λ ) . (4.12)

Proof. Taking F = G = (ν, φ)Λ − Eµψ,ρΛ [(ν, φ)Λ] in (4.2) considered on Λ

instead of Zd, since ∂H =

(−Lψ,ρ

Λ

)−1

∂G, the left hand side of (4.12) isrewritten as ∑

x∈Λ

Eµψ,ρΛ

[ν(x)

((−Lψ,ρ

Λ

)−1

ν

)(x, φ)

],

which is bounded above by

≤∑x∈Λ

Eµψ,ρΛ

[ν(x)

(−Qφ,ψ

Λ,0

)−1

ν(x)]

≤∑x∈Λ

ν(x) (−c−∆Λ,0)−1ν(x) ,

from Lemma 4.7-(2), (3). However, the last term coincides with the right handside of (4.12), which is indeed independent of the boundary condition ψ, recall(3.4).

Theorem 4.9. (Brascamp-Lieb inequality for exponential moments)We have

Eµψ,ρΛ

[e(ν,φ−〈φ〉)Λ

]≤ e

12var ((ν,φ)Λ;µψ,G

Λ ) . (4.13)

In particular,

Eµψ,ρΛ

[e|(ν,φ−〈φ〉)Λ|

]≤ 2e

12var ((ν,φ)Λ;µψ,G

Λ ) , (4.14)

where 〈φ〉 = Eµψ,ρΛ [φ].

148 T. Funaki

Proof. (4.13) follows from (4.11) and Theorem 4.8. For (4.14), use e|x| ≤ex + e−x.

Since the right hand side of (4.13) is Eµψ,GΛ [e(ν,φ−〈φ〉)Λ ], the exponential

moments subtracted their means under ϕ-Gibbs measures are bounded aboveby those under the Gaussian measures.

Remark 4.1. (1) An estimate on

Eµψ,ρΛ

[eε(ν,φ−〈φ〉)2Λ

]

for some ε > 0 is known. This is a stronger estimate than (4.13), cf. [102],Proposition 1.1.6. See Remark 3.1 or [198] for Gaussian case.(2) See [131] for some extension of the Brascamp-Lieb inequalities.(3) For FKG or other basic inequalities used in statistical mechanics, see [138],[235]. The relation to Witten’s Laplacian is discussed in [143].

We have the Brascamp-Lieb inequality for the ∇ϕ-field∗) :

var (∇ϕ(b);µ∇Λ,ξ) ≤ c−1

− (4.15)

for every b ∈ Λ∗ and ξ ∈ X . Indeed, set HΛ(η) =∑

b∈Λ∗(η(b) + η(−b))2 +∑P∩Λ∗ =∅(

∑b∈P

η ∨ ξ(b))2 for η = η(b); b ∈ Λ∗ ∈ RΛ∗ . Then, µ∇

Λ,ξ is theweak limit of µ∇,ε

Λ,ξ as ε ↓ 0, where

µ∇,εΛ,ξ (dη) =

1Zε

Λ,ξ

exp

−

12

∑b∈Λ∗

V (η(b))− 1εHΛ(η)

∏b∈Λ∗

dη(b) ∈ P(RΛ∗) .

Since HΛ is convex, we have

var (∇ϕ(b);µ∇,εΛ,ξ ) ≤ var (∇ϕ(b);

∏b∈Λ∗

µb) ≤ var (η(b);N(0, c−1− )) = c−1

− ,

where µb = z−1e−V (η(b))/2dη(b) ∈ P(R). Taking the limit ε ↓ 0, (4.15) isobtained. Note that, even in the Gaussian case, it requires some works tohave an estimate on var (∇ϕ(b);µ∇

Λ,ξ) = ∇i,x∇i,yGΛ(x, y)|x=y for b = x+ ei.

4.3 Estimates of Nash-Aronson’s Type and Long Correlation

This section establishes the long correlation under the ϕ- and ∇ϕ-Gibbs mea-sures. The Helffer-Sjostrand representation is combined with Nash-Aronson’stype estimates on the transition probability of the random walk in randomenvironments.

∗communicated by Giacomin.


Let X = (Xt)t≥0 be the random walk on Zd with jump rates cx,±ei

(t)from x to its adjacent sites y = x± ei at time t satisfying the symmetry

cx,±ei(t) = cx±ei,∓ei

(t)

and the uniformity0 < c− ≤ cx,±ei

(t) ≤ c+ . (4.16)

Then its transition probability p(s, x; t, y) = P (Xt = y|Xs = x), t ≥ s ≥0, x, y ∈ Z

d admits the following three estimates (Propositions 4.10-4.12) ofNash-Aronson’s type. The constants C1, C2 and C > 0 depend only on d andc±.

Proposition 4.10. (Giacomin-Olla-Spohn [135], Proposition B3, B4) Forevery t ≥ s ≥ 0 and x, y ∈ Z

d, we have

p(s, x; t, y) ≤ C1

(t− s)d/2∗

exp

− |x− y|C1

√(t− s)∗

.

In addition, if |x− y| ≤√

(t− s)∗ is satisfied, then we have

p(s, x; t, y) ≥ C2

(t− s)d/2∗

,

where (t− s)∗ := (t− s) ∨ 1.

Proposition 4.11. (Delmotte-Deuschel [63]) For every t ≥ s ≥ 0 and x, y ∈Z

d, we have

p(s, x; t, y) ≤ C

(t− s)d/2e−Γ (|x−y|,t−s)/C ,

where

Γ (x, t) ≥

|x|2t, t ≥ |x| (long time regime) ,

|x|√t, t < |x| (Poisson regime) .

Especially when the jump rates are given by cx,±ei(t) = V ′′(∇φt(〈x ±

ei, x〉)), the transition probability is denoted by pφ·(s, x; t, y). The above twoestimates on p are quenched, i.e., hold uniformly in the environment φt. Forits gradient ∇pφ· , we have only annealed estimate:

Proposition 4.12. ([63]) For every t ≥ s ≥ 0, b ∈ (Zd)∗ and y ∈ Zd, we have

EµψΛ

[∣∣∇pφ·(s, ·; t, y)(b)∣∣2] 1

2 ≤ C

(t− s)(d+1)/2e−Γ (|xb−y|,t−s)/C ,

where ∇ = ∇x acts on the variable x.

150 T. Funaki

Nash-Aronson’s type estimates are applicable to derive estimates on thecorrelation functions under the Gibbs measures. The next theorem is an ex-tension of Proposition 3.5 for Gaussian fields to general convex potentials.The lower estimate implies the long correlations.

Theorem 4.13. (Naddaf and Spencer [202]) Assume d ≥ 3. Then there existC, c > 0, which depend only on d and c±, such that

0 ≤ EµψΛ [φ(x);φ(y)] ≤ C

|x− y|d−2,

for every x, y ∈ Zd. Furthermore, for the ϕ-Gibbs measure µ on Z

d which istempered, shift invariant, mean 0 and ergodic under the spatial shift (cf. Sect.4.5), we have

c

|x− y|d−2≤ Eµ [φ(x);φ(y)] ≤ C

|x− y|d−2

for every x, y ∈ Zd.

Proof. (Giacomin [130] §3.3) From Corollary 4.3 of Helffer-Sjostrand repre-sentation and Proposition 4.10, we have that

EµψΛ [φ(x);φ(y)] =

∫ ∞

0

EµψΛ

[pφ·

Λ (0, x; t, y)]dt

≤∫ ∞

0

EµψΛ

[pφ·(0, x; t, y)

]dt

≤ C1e− 1

C1|x−y| +

∫ ∞

1

C1

td/2e− |x−y|

C1t1/2 dt ,

where pφ·Λ (0, x; t, y) = P (Xt = y, t < τΛ|X0 = x) is the transition probability

of the random walk killed at the time when it goes outside of Λ. The lastintegral is, after making a simple change of variables |x− y|−2t = t′, boundedby

≤ C1

|x− y|d−2

∫ ∞

0

1td/2

e− 1

C1t1/2 dt =C ′

|x− y|d−2.

Note that this integral converges since d ≥ 3. The lower bound follows againfrom Proposition 4.10:

Eµ [φ(x);φ(y)] =∫ ∞

0

Eµ[pφ·(0, x; t, y)

]dt

≥ C2

∫ ∞

|x−y|2

dt

td/2=

2C2

(d− 2)|x− y|d−2.

For the ∇ϕ-fields, the following estimate holds.


Proposition 4.14. The dimensions d ≥ 1 are arbitrary. We have for everyx, y ∈ Z

d ∣∣∣EµψΛ [∇φ(b);∇φ(b′)]

∣∣∣ ≤ C

|xb − xb′ |d−1.

Proof. Take F (φ) = ∇φ(b) and G(φ) = ∇φ(b′) in Helffer-Sjostrand represen-tation (Theorem 4.2). Then, since ∂F (x, φ) = 1x=xb − 1x=yb, one obtains

EµψΛ [∇φ(b);∇φ(b′)] =

∫ ∞

0

EµψΛ

[∇x∇yp

φ·Λ (0, ·; t, ·)(b, b′)

]dt .

To estimate this integral, we divide the interval [0,∞) into the sum of [0, 1)and [1,∞) as in the proof of Theorem 4.13, and for the latter integral we rudelyestimate |∇x∇yp

φ·Λ (b′)| ≤ |∇xp

φ·Λ (xb′)|+|∇xp

φ·Λ (yb′)| and use Proposition 4.12.

Delmotte and Deuschel [64] have elaborated the estimate in Proposition4.14 as ∣∣∣Eµψ

Λ [∇φ(b);∇φ(b′)]∣∣∣ ≤ C

|xb − xb′ |d,

see the final remark of Giacomin [130] §3.As we have seen, the correlation functions of Gibbs measures decay slowly

and this makes the proof of CLT or LDP difficult. One can say that the loopcondition (see (L) in Sect. 2.3) for ∇ϕ-field yields the long dependence.

Remark 4.2. For the potential V (η) = η2 + λη4, λ > 0, the decay of correla-tions is discussed in [39, 40, 42, 127, 196]; see also [102] and (3.5) of [124].

4.4 Thermodynamic Limit and Constructionof ∇ϕ-Gibbs Measures

The next theorem is shown by taking thermodynamic limit for a sequenceof finite volume ∇ϕ-Gibbs measures with periodic boundary conditions, seeSect. 9.2 (and Definition 2.2) for the class (ext G∇)u of measures and recallDefinition 2.3 for shift invariance and ergodicity. The tightness of the sequenceof measures is a consequence of Brascamp-Lieb inequality. This method isapplicable only to strictly convex potentials; see Remark 4.4 for nonconvexpotentials.

Theorem 4.15. [124] (existence of ∇ϕ-Gibbs measures) For every u ∈R

d, there exists µ∇ =: µ∇u ∈ (extG∇)u, i.e., a tempered, shift invariant,

mean u and ergodic ∇ϕ-Gibbs measure µ∇u exists. Furthermore, it satisfies

Eµ∇u [eβ(η(b)−ub)

2] < ∞ for some β > 0, where ub = ui if the bond b is i-

directed.

152 T. Funaki

Proof. Let TdN = (Z/NZ)d be the lattice torus of size N and let T

d,∗N be the set

of all directed bonds in TdN . Let X

TdN

be the family of all η ∈ RT

d,∗N satisfying

the loop condition on the torus (∑

b∈Cη(b) = 0 for all closed loops C in T

dN ,

see Sect. 2.3), and define µ∇N,u ∈ P(X

TdN

) by

µ∇N,u(dη) :=

1ZN,u

exp

−

12

∑b∈T

d,∗N

V (η(b) + ub)

dηN , (4.17)

where ZN,u is the normalization constant and dηN is the uniform measure onthe affine space X

TdN

. Let µ∇N,u be the distribution of η(b) := η(b)+ub under

µ∇N,u. One can regard µ∇

N,u ∈ P(X ) by extending it periodically. Then, fromthe Brascamp-Lieb inequality (on the torus), we have for every λ > 0

supN,u∈Rd

Eµ∇N,u [eλ|η(b)−ub|] <∞ . (4.18)

This implies the tightness of the measures µ∇N,uN . Accordingly, along a

proper subsequence N ′ → ∞, µ∇N ′,u weakly converges to a certain measure

µ∇u . One can easily show that µ∇

u ∈ G∇, Eµ∇u [η(ei)] = ui, E

µ∇u [eλ|η(b)−ub|] <∞

and from Proposition 4.14∣∣∣Eµ∇

u [η(b); η(b′)]∣∣∣ ≤ C

|xb − xb′ |d−1.

We may now suppose d ≥ 2, since µ∇u is a linear combination of Bernoulli

measures when d = 1 (see Remark 4.5 below), and in this case this boundimplies the ergodicity of µ∇

u .

Remark 4.3. The periodic boundary conditions are taken for the limit mea-sure µ∇

u to be automatically shift invariant. Instead, one may consider thesequence µψu

Λ with the Dirichlet boundary conditions ψu(x) ≡ u · x and the

distributions µ∇,ψu

Λ of ∇ϕ-field under µψu

Λ. Then the tightness of µ∇,ψu

Λ

is similar, however, the shift invariance of the limit measures seems nontrivial.

Remark 4.4. The general theory in statistical mechanics seems to work forthe construction of the ∇ϕ-Gibbs measures even for nonconvex potentials V ,which diverge sufficiently rapidly as |η| → ∞, cf. [230]. Indeed, the argumentdue to Giacomin is the following. First define the specific free energy f(β) for∇ϕ-fields on T

dN at inverse temperature β > 0 in a similar manner to σ∗

(u)in (5.1) for ϕ-fields below. Then, the derivative f ′(β) (or f ′(β±)) exists forthe limit f(β) = lim→∞ f(β), from which one obtains a uniform bound. Thisimplies the tightness of µ∇

N,u.

Remark 4.5. In one dimension µ∇u is the Bernoulli measure, i.e., under µ∇

u ,∇φ(x) ≡ φ(x + 1) − φ(x);x ∈ Z is an i.i.d. sequence and the distribution


of each ∇φ(x) is given by the Cramer transform of ν(dη) = Z−1e−V (η) dη ∈P(R). In other words, if we define νλ ∈ P(R) for λ ∈ R by (5.25) and deter-mine the function λ = λ(u) of u ∈ R by the relation Eνλ [η] = u (see Sect.5.5), then we have

µ∇u (dη) =

∏x∈Z

νu(dη(x)) ∈ P(RZ), η = η(x);x ∈ Z ,

where νu = νλ(u). Indeed, [141] Theorem 3.5 shows that µ∇,ψu

Λin Remark 4.3

converges to this µ∇u as →∞. In particular, in one dimension, the potential

V needs not be convex, but the conditions (1.3)–(1.6) in [141] (with V (η) inplace of φ(x)) are sufficient.

4.5 Construction of ϕ-Gibbs Measures

The Gibbs measures for ∇ϕ-field have been constructed for all dimensionsd, but the Gibbs measures for ϕ-field are unnormalizable (i.e., finite volumeGibbs measures for ϕ-field do not converge as Λ Z

d) if d ≤ 2 and normal-izable if d ≥ 3. We have indeed the following theorem.

Theorem 4.16. If d ≥ 3, for every h ∈ R, there exists a ϕ-Gibbs measureµ ≡ µh on Z

d with mean h, i.e., Eµ[φ(x)] = h for all x ∈ Zd.

Proof. Consider the sequence of finite volume ϕ-Gibbs measures µ0Λ

∈P(RZ

d

) with 0-boundary conditions. Then, by the symmetry of V , the meanis 0: Eµ0

Λ [φ(x)] = 0. When d ≥ 3, since the variance GΛ(x, x) of the Gaussian

system is uniformly bounded in , Brascamp-Lieb inequality (Theorem 4.9)proves

supx∈Zd

sup∈N

Eµ0Λ

[eλ|φ(x)|

]<∞, λ > 0 .

Therefore, the sequence µ0Λ is tight and has a limit µ along a proper

subsequence ′ →∞. It is obvious that µ is a ϕ-Gibbs measure with mean 0.The distribution µh of φ+ h, where φ is µ-distributed, is a ϕ-Gibbs measurewith mean h.

Remark 4.6. It may be possible to construct shift invariant µh (cf. Re-mark 4.3). In fact, Giacomin suggests to consider µ0

Λ;massociated with the

massive Hamiltonian instead of µ0Λ

in the proof of Theorem 4.16. Then,µm = lim′→∞ µ0

Λ′ ;mexists and is shift invariant. Finally, if d ≥ 3, the

limit µ = limm↓0 µm exists by means of the Brascamp-Lieb inequality and wesee the shift invariance of µh.

154 T. Funaki

5 Surface Tension

The surface tension σ = σ(u), u ∈ Rd physically describes the macroscopic

energy of a surface with tilt u, i.e., a d dimensional hyperplane located inR

d+1 with normal vector (−u, 1) ∈ Rd+1. It is mathematically a fundamental

quantity, as it will appear in several limit theorems, e.g., in the rate functionalof LDP or diffusion coefficient in the hydrodynamic limit.

5.1 Definition of Surface Tension

The surface tension will be defined thermodynamically from the HamiltonianH(φ) in such a way to reflect statistical property of random interfaces withmean tilt u. Let ψu ∈ R

Zd

, u = (ui)di=1 ∈ R

d, be tilted height variablesdetermined by ψu(x) = u · x, x ∈ Z

d and set for ∈ N

σ∗ (u) =− 1

|Λ|logZψu

Λ

=− 1(2+ 1)d

log∫

RΛ

exp−Hψu

Λ(φ)

dφΛ

, (5.1)

where Λ = [−, ]d ∩ Zd is a cube with side length 2 + 1 and Zψu

Λis the

normalization constant given by (2.5) with the boundary condition ψu. Thefunction σ∗

(u) is the specific free energy of interfaces with tilt u insistedthrough the boundary condition. Note that |Λ| = (2+1)d is the order of thesurface area of the boundary of interfaces settled in d + 1 dimensional spaceR

d+1.One can show, based on the subadditivity of σ∗

in , that its limit as→∞ exists.

Theorem 5.1. [124] The limit

σ∗(u) = lim→∞

σ∗ (u) ∈ [−∞,∞)

exists.

We shall normalize the limit function σ∗(u) as σ(u) = σ∗(u) − σ∗(0) sothat σ(0) = 0; note that σ∗(u) ∈ (−∞,∞) is shown comparing with the caseof quadratic potentials, see Sect. 5.2. The function σ(u), u ∈ R

d is called the(normalized) surface tension. By Theorem 5.1, we have

σ(u) = − lim→∞

1|Λ|

logZψu

Λ

Zψ0Λ

. (5.2)

The ratio Zψu

Λ/Zψ0

Λis roughly equal to the probability to find interfaces with

mean tilt u under µψ0Λ

and therefore


µ(tilt of hN ∼ u

)

N→∞exp−Ndσ(u) , (5.3)

for the ϕ-Gibbs measure µ (with tilt 0) and macroscopically scaled heightvariables hN defined by (2.16). This broadly explains the meaning of thesurface tension.

Another Definition

Sheffield [230] gives a different but actually an equivalent definition for thesurface tension. Let µfree

Λ be the finite volume ϕ-Gibbs measure on Λ Zd

given by (2.4) and determined from the free boundary condition, i.e., thesum for the Hamiltonian (2.1) is taken for all 〈x, y〉 ⊂ Λ only. Denote thedistribution of ∇φ(b); b ∈ Λ∗ under µfree

Λ by µ∇,freeΛ∗ ∈ P(RΛ∗

). Recall thatfor two probability measures µ and ν

H(µ|ν) = Eµ

[log

dµ

dν

](5.4)

defines the relative entropy of µ with respect to ν. Then, the specific freeenergy of shift invariant measure µ∇ ∈ P(R(Zd)∗) is defined by the relativeentropy with respect to the ∇ϕ-Gibbs measure with free boundary conditionper site:

F (µ∇) = lim→∞

1|Λ|

H(µ∇Λ∗

|µ∇,free

Λ∗

) , (5.5)

where µ∇Λ∗

stands for the marginal distribution of ∇φ(b); b ∈ Λ∗

under µ∇.The surface tension has another expression

σ(u) = infµ∇:mean u

F (µ∇) , (5.6)

where the infimum is taken over all shift invariant µ∇ ∈ P(R(Zd)∗) with meantilt u: Eµ∇

[η(ei)] = ui for every 1 ≤ i ≤ d.Sheffield establishes the variational characterization for the ∇ϕ-Gibbs

measures for general tilt u: the minimizer µ∇ ≡ µ∇,(u) of (5.6) is in factthe ∇ϕ-Gibbs measure for each u. This tells that µ

∇,(u)Λ∗

= µ∇

Λ∗ ,u (= the

distribution of ∇φ(b); b ∈ Λ∗ under µψu

Λ) and for such µ∇, since one can

expect that the free boundary condition may be replaced with the 0-boundarycondition, the specific free energy F (µ∇) is the limit of

1|Λ|

H(µ∇Λ∗

,u|µ∇Λ∗

,0) ∼ −1|Λ|

logZψu

Λ

Zψ0Λ

as → ∞. This coincides with the definition (5.2) of the normalized surfacetension.

156 T. Funaki

5.2 Quadratic Potentials

The surface tension is explicitly computable for quadratic potentials. Recallthat ∆Λ ≡ ∆Λ,0 denotes the discrete Laplacian on Λ with the boundarycondition 0, see Sect. 3.1.

Proposition 5.2. Assume V (η) = c2η

2, c > 0. Then, the corresponding un-normalized surface tension σ∗(u) ≡ σ∗

c (u) is given by

σ∗c (u) =

12c|u|2 − 1

2log 2πc−1 +

12

lim→∞

1|Λ|

log det(−∆Λ) , (5.7)

where det(−∆Λ) denotes the determinant of −∆Λ

regarding it as a |Λ| ×|Λ| matrix and |u| =

√∑di=1 u

2i stands for the Euclidean norm of u. The

eigenvalues of ∆Λare known (see, e.g., [138] p.185, (9.5.12) or [108]) and

therefore det(−∆Λ) is specifically computable.

Proof. Since ψu is harmonic, we have φΛ,ψu= ψu and therefore, from the

proof of Proposition 3.1

Zψu

Λ= e−

c2 B,u

∫R

Λ

ec2 ((φ−ψu),∆Λ

(φ−ψu))Λ dφΛ

, (5.8)

where B,u is the boundary term defined by

B,u =∑

x∈Λ,y /∈Λ

|x−y|=1

ψu(y)∇ψu(〈y, x〉) .

However, the integral in the right hand side of (5.8) can be rewritten as

=∫

RΛ

e− c

2 (φ,−∆Λφ)

Λ dφΛ= (2πc−1)|Λ|/2

(det(−∆Λ

))−1/2

,

while the boundary term is equal to

B,u = 2d∑

i=1

∑|yj |≤,j =i

(ujyj + ui(+ 1))ui

= 2(+ 1)(2+ 1)d−1|u|2 .

In this way σ∗ (u) in (5.1) is explicitly calculated and the conclusion follows

by taking the limit →∞.

By the basic conditions (2.2) on V , the potential V is in general estimatedby quadratic functions both from above and below:

V (0) +12c−η

2 ≤ V (η) ≤ V (0) +12c+η

2 .


This proves the following bounds on the unnormalized surface tension σ∗(u)corresponding to V

σ∗c−(u) + d · V (0) ≤ σ∗(u) ≤ σ∗

c+(u) + d · V (0) ,

which, in particular, implies σ∗(u) ∈ (−∞,∞).

5.3 Fundamental Properties of Surface Tension

Here we summarize several properties of the surface tension. In the case ofquadratic potentials, the (normalized) surface tension is given by σc(u) =c2 |u|2 as we have seen in Proposition 5.2 and all properties listed below areobvious.

The following theorem and its corollary indicate that the function σ isstrictly convex, symmetric, σ ∈ C1 and its derivative ∇σ is Lipschitz contin-uous; in particular, a surface with gentle slope has low energy.

Theorem 5.3. ([124]; [77], [135] for (3)) The function σ = σ(u) ∈ [0,∞)enjoys the following properties.(1) (regularity) σ ∈ C1(Rd) and ∇σ = (∂σ/∂ui)d

i=1 is Lipschitz continuous,i.e., for some C > 0,

|∇σ(u)−∇σ(v)| ≤ C|u− v|, u, v ∈ Rd . (5.9)

(2) (symmetry) σ(−u) = σ(u).(3) (strict convexity) With constants c−, c+ > 0 in (V3) of (2.2), we havefor every u, v ∈ R

d that

12c−|u− v|2 ≤ σ(v)− σ(u)− (v − u) · ∇σ(u) ≤ 1

2c+|u− v|2 . (5.10)

The Lipschitz continuity of ∇σ is shown based on the coupling used for theproof of Theorem 9.3 with the help of (5.14) below. The strict convexity of σis a consequence of uniform strict convexity of σ∗

in , cf. [87] for the pyramidinequality and also [199]. See the next subsection for the proof. Theorem 5.3-(3) implies the following.

Corollary 5.4. For every u, v ∈ Rd, we have that

c−|u− v|2 ≤ (u− v) · (∇σ(u)−∇σ(v)) ≤ c+|u− v|2 . (5.11)

In particular,c−|u|2 ≤ u · ∇σ(u) ≤ c+|u|2 . (5.12)

Proof. The first estimate (5.11) is immediate by taking the sum of each sideof (5.10) with itself, but with u and v replaced by each other. The second(5.12) is from (5.11) with v = 0 noting that ∇σ(0) = 0.

158 T. Funaki

Problem 5.1. Theorem 5.3-(1) nearly establishes “σ ∈ C2(Rd)”, but this isactually not yet proved and, indeed, remains to be one of the important openproblems. Such problem on the regularity of σ is related to the CLT (Sects. 8,11). See also a recent approach by Caputo and Ioffe [48].

Remark 5.1. (Physical argument on σ) For the SOS model or the sine-Gordon model, which is its continuous-spin version in a sense, the surfacetension may have a cusp at u = 0 and therefore σ /∈ C1 in general, seeFrohlich and Spencer [107], [240]. This reflects the phenomena called rough-ening transition, cf. Remark 6.7. Moreover, if the potential V is nonconvex,it is conjectured that σ need not be in C1; i.e., the singularity of σ is relatedto the phase transition of first order exhibited by the system. See Fernandezet al. [100].

Theorem 5.5. [124] (Thermodynamic identities)

ui = Eµ∇u [η(ei)] , (5.13)

∂σ

∂ui(u) = Eµ∇

u [V ′(η(ei))] , (5.14)

u · ∇σ(u) + 1 = Eµ∇u

[d∑

i=1

η(ei)V ′(η(ei))

], (5.15)

where µ∇u ∈ P(X ) is the unique probability measure in (ext G∇)u, in other

words, ∇ϕ-pure phase, see Sect. 9.2.

The identity (5.13) is just by definition, while (5.15) is shown by the inte-gration by parts for similar integrals appearing in σ∗

, see Lemma 5.7 below.

5.4 Proof of Theorems 5.3 and 5.5

The first lemma is to replace the boundary conditions ψu in (5.2) with theperiodic boundary conditions as we have done in the proof of Theorem 4.15.The proof of the lemma can be found at Appendix II of [124].

Lemma 5.6. Let ZN,u be the normalization constant in (4.17) and set

σN (u) = − 1|Td

N |log

ZN,u

ZN,0.

Then we haveσ(u) = lim

N→∞σN (u) .

This lemma is more convenient for us than (5.2), since the ∇ϕ-Gibbsmeasure µ∇

u was constructed under the periodic boundary conditions.


Lemma 5.7. Let µ∇N,u ∈ P(X

TdN

) be the measure introduced in the proof ofTheorem 4.15. Then we have the following three identities:

ub = Eµ∇N,u [η(b)] , (5.16)

∂σN

∂ui(u) = Eµ∇

N,u [V ′(η(ei))] , (5.17)

u · ∇σN (u) + 1 = Eµ∇N,u

[d∑

i=1

η(ei)V ′(η(ei))

]. (5.18)

Proof. Since∑

b∈Ciη(b) = 0 holds for every η ∈ X

TdN

along with the closedloop Ci parallel to the ith axis, the shift invariance of µ∇

N,u (on the torus)

implies Eµ∇N,u [η(b)] = 0. This shows (5.16). To see (5.17), noting that V ′(η) =

−V ′(−η), we rewrite its left hand side as

− 1|Td

N |ZN,u

∂

∂ui

∫e− 1

2

∑b∈T

d,∗N

V (η(b)+ub)dηN

=1

|TdN |ZN,u

∫ ∑b′∈T

d,∗N :b′‖ei

V ′(η(b′) + ub′)e− 1

2

∑b∈T

d,∗N

V (η(b)+ub)dηN

and this coincides with the right hand side, where b′ ‖ ei means that the bondb′ is i-directed. Finally, the third identity (5.18) follows from

∑b∈T

d,∗N

Eµ∇N,u [η(b)V ′(η(b) + ub)] = 2|Td

N | , (5.19)

since the right hand side of (5.18) is equal to

12|Td

N |∑

b∈Td,∗N

Eµ∇N,u [(η(b) + ub)V ′(η(b) + ub)]

and if we note (5.17). However, the left hand side of (5.19) can be rewrittenas

∑b∈T

d,∗N

∫R

TdN

\O∇φ(b)V ′(∇φ(b) + ub)F (φ)

∏x∈T

dN\O

dφ(x)

= −2∑

y∈TdN

∫R

TdN

\Oφ(y)

∂F

∂φ(y)(φ)

∏x∈T

dN\O

dφ(x)

= 2∑

y∈TdN

∫R

TdN

\OF (φ)

∏x∈T

dN\O

dφ(x) = 2|TdN | ,

by the integration by parts, where φ(O) = h is arbitrarily taken and we set

160 T. Funaki

F (φ) =1

ZN,ue− 1

2

∑b∈T

d,∗N

V (∇φ(b)+ub),

for φ = φ(x);x ∈ TdN \ O. This concludes (5.19).

Proof (Theorems 5.3 and 5.5). As we have seen in the proof of Theorem 4.15,µ∇

N,u weakly converges to µ∇u as N → ∞ (we actually need not to take the

subsequence). Therefore, noting the uniform estimate (4.18), we obtain

limN→∞

Eµ∇N,u [V ′(η(ei))] = Eµ∇

u [V ′(η(ei))] . (5.20)

This convergence is uniform in u in any bounded set of Rd. Let us take the

limit N →∞ in the following trivial identity

σN (u)− σN (v) =∫ 1

0

(u− v) · ∇σN (tu+ (1− t)v) dt . (5.21)

From Lemma 5.6 the left hand side converges to σ(u)− σ(v), while the limitof the right hand side is computable by (5.17) and (5.20), and we have

σ(u)− σ(v) =∫ 1

0

d∑i=1

(ui − vi)Eµ∇tu+(1−t)v [V ′(η(ei))] dt .

However, the uniqueness of the (tempered, shift invariant) ergodic ∇ϕ-Gibbsmeasure for each tilt u (see Theorem 9.5 and Corollary 9.6 below) impliesthat, as un → u, µ∇

unweakly converges to µ∇

u . This, in particular, proves thatEµ∇

u [V ′(η(ei))] is continuous in u. Thus we have shown that σ ∈ C1(Rd) andthe identity (5.14). The symmetry of σ(u) follows from that of σ∗

(u), whichis readily seen from the symmetry of the potential V . The identity (5.15) is aconsequence of (5.18) by letting N →∞.

To prove the Lipschitz continuity (5.9) of ∇σ, we need to apply the dy-namic coupling (see Sect. 9). In fact, by noting (5.14) and

|V ′(η(ei))− V ′(η(ei))| ≤ c+ |η(ei)− η(ei)| ,we obtain (5.9) from Proposition 9.8.

The proof of the strict convexity (5.10) is only left. To this end, it sufficesto show (5.10) for σN in place of σ. Consider the Hessian of σN :

D2σN (u) =(∂2σN

∂ui∂uj

)1≤i,j≤d

.

Then, for every λ = (λi)di=1 ∈ R

d, we have that

(D2σN (u)λ, λ

)=

d∑i=1

λ2iE

µ∇N,u [V ′′(ηi(O))]

− 1|Td

N |var

d∑

i=1

λi

∑x∈T

dN

V ′(ηi(x));µ∇N,u

. (5.22)


Indeed, this identity follows by computing ∂2 logZN,u/∂ui∂uj . In the left handside ( , ) denotes the inner product of R

d (which is usually denoted by · ),while ηi(x) := η(〈x + ei, x〉) in the right hand side. Since the second termin the right hand side of (5.22) is nonpositive, we obtain from V ′′ ≤ c+ theupper bound (

D2σN (u)λ, λ)≤ c+|λ|2 . (5.23)

In order to get the lower bound, we use the Helffer-Sjostrand representationfor the ϕ-field on T

dN \ O defined by

φ(x) =∑

b∈CO,x

η(b), x = O ,

where CO,x is a chain connecting O and x. We may think of φ(O) = 0. Set

F (φ) =d∑

i=1

λi

∑x∈T

dN

V ′(φ(x+ ei)− φ(x)) .

Then, since

∂F (x, φ) =d∑

i=1

λi V ′′(φ(x)− φ(x− ei))− V ′′(φ(x+ ei)− φ(x)) ,

from Theorem 4.2 (and its proof), we have the representation

var (F ;µ∇N,u) =

∑x∈T

dN\O

⟨∂F (x, φ)(−L)−1∂F (x, φ)

⟩.

Here 〈 · 〉 denotes the expectation under µ∇N,u and L = L0,0

TdN\O +Qφ,0

TdN\O,0

.The right hand side can be further rewritten into

supf=f(x,φ)

2

∑x∈T

dN\O

〈f(x, φ)∂F (x, φ)〉 −∑

x∈TdN\O

〈f(x, φ)(−L)f(x, φ)〉

,

where the functions f satisfy the condition f(O,φ) = 0. However, each termin the above supremum can be rewritten as

∑x∈T

dN\O

〈f(x, φ)∂F (x, φ)〉 =∑

x∈TdN

d∑i=1

λi〈∇if(x, φ)V ′′(ηi(x))〉 ,

∑x∈T

dN\O

〈f(x, φ)(−L)f(x, φ)〉 =∑

x∈TdN

d∑i=1

〈V ′′(ηi(x))(∇if(x, φ))2〉

+∑

x∈TdN\O

⟨(∂f

∂φ(x)

)2⟩,

162 T. Funaki

and, therefore, we have from (5.22) that

(D2σN (u)λ, λ

)=

1|Td

N |inff

∑x∈T

dN

d∑i=1

〈V ′′(ηi(x))(λi −∇if(x, φ))2〉

+∑

x∈TdN\O

⟨(∂f

∂φ(x)

)2⟩ .

Since V ′′ ≥ c−, this identity implies

(D2σN (u)λ, λ

)≥ c−|Td

N |inff

∑x∈T

dN

d∑i=1

〈(λi −∇if)2〉 .

However, by estimating

∑x∈T

dN

d∑i=1

〈(λi −∇if)2〉 =∑

x∈TdN

d∑i=1

(λ2i + 〈(∇if)2〉) ≥ |Td

N | · |λ|2 ,

we finally get the lower bound(D2σN (u)λ, λ

)≥ c−|λ|2 . (5.24)

Now (5.23) and (5.24) establish (5.10) for σN , since we have

σN (v)− σN (u)− (v − u) · ∇σN (u)

=∫ 1

0

dt

∫ t

0

(D2σN (u+ s(v − u))(v − u), v − u

)ds

for every u, v ∈ Rd. Letting N →∞ shows (5.10) for σ.

5.5 Surface Tension in one Dimensional Systems

In one dimension (i.e., for interfaces in 1+1 dimensional space), the ∇ϕ-Gibbsmeasures are simple Bernoulli measures (cf. Remark 4.5) and the features ex-hibited by them are completely different from the higher dimensional systems.In some cases, however, the surface tension σ = σ(u), u ∈ R is explicitly com-putable and this might be useful to see.

Define νλ ∈ P(R), λ ∈ R and the normalization constant Zλ by

νλ(dη) =1Zλ

e−V (η)+λη dη , (5.25)

Zλ =∫

R

e−V (η)+λη dη ,


respectively, and introduce a function u = u(λ) as

u = Eνλ [η] ≡ d

dλlog Zλ, λ ∈ R .

Then, sinceu′(λ) = Eνλ

[(η − Eνλ [η]

)2]> 0 ,

u is strictly increasing in λ and therefore it admits an inverse function λ =λ(u). The function λ actually coincides with the differential of the surfacetension σ:

σ′(u) = λ(u) .

Indeed, from (5.14), we have an expression σ′(u) = Eνλ [V ′(η)] in one dimen-sion and, by integration by parts, one can easily find that the expectation inthe right hand side is equal to λ(u). The normalized surface tension is thusgiven by

σ(u) =∫ u

0

λ(v) dv, u ∈ R . (5.26)

Since σ′′ = λ′ > 0, one can see that σ is strictly convex and smooth.Except for the normalization, the surface tension can be expressed as the

Legendre transform of log Zλ:

σ(u) = supλ

λu− log Zλ

. (5.27)

Indeed, denoting the right hand side of (5.27) by σ(u), supλ is attained atλ = λ(u) and

λ = σ′(u) ⇐⇒ u = u(λ)

holds, see [141], (1.12) or [200], (1.2).We give three examples of V for which σ = σ(u) is explicitly computable

based on the formula (5.26).

Example 5.1. V (η) = c2η

2 with c > 0. A simple computation shows u(λ) =1cλ whose inverse function is λ(u) = cu. Therefore, we have σ(u) = 1

2cu2 and

this coincides with Proposition 5.2.

Example 5.2. V (η) = c|η| with c > 0. This potential does not satisfy theconditions (V1) and (V3) in (2.2), but σ(u) are computable. The measure νλ

is defined only for |λ| < c and, by an explicit computation, we have

u(λ) =2λ

c2 − λ2, |λ| < c .

Its inverse function is

λ(u) =√

1 + c2u2 − 1u

, u ∈ R ,

164 T. Funaki

so that the surface tension is given by

σ(u) =∫ |u|

0

√1 + c2v2 − 1

vdv, u ∈ R .

Incidentally, since λ(u) → ±c as u→ ±∞, the function σ is linearly growingas |u| → ∞.

Example 5.3. V (η) = c1η (η ≥ 0) and V (η) = −c2η (η ≤ 0) with c1, c2 > 0.This potential is even asymmetric and as the Hamiltonian we adopt the sumof positively directed bonds only: H(φ) =

∑x V (∇φ(x)), where ∇φ(x) = φ(x+

1)− φ(x), cf. Remark 2.1. Then, similarly to Example 5.2, we have

u(λ) =2λ+ c2 − c1

(c2 + λ)(c1 − λ), −c2 < λ < c1 .

Especially if c2 = +∞, we have

u(λ) =1

c1 − λ, λ(u) = c1 −

1u,

and therefore, except normalization,

σ(u) = c1u− log u, u > 0 .

The condition c2 = +∞ means that η = ∇φ(x) ≥ 0 is only realizable un-der the Gibbs measures µ∇

u . In other words, the graph of interfaces is alwaysincreasing.

6 Large Deviation and Concentration Properties

This section starts the analysis on the limit procedure under the scaling (2.16)which connects microscopic interface height variables φ = φ(x) with macro-scopic ones hN = hN (θ). We shall establish the LDP for hN as N →∞ and,as its application, obtain two types of LLNs under the (canonical) ϕ-Gibbsmeasures.

The LDP for the ∇ϕ interface model was first studied by Ben Arousand Deuschel [12] in Gaussian case. They considered the field φ(x);x ∈DN,DN = (0, N)d∩Z

d, which is distributed under the finite volume ϕ-Gibbsmeasure µ0

N ≡ µ0DN

with V (η) = 12η

2 having 0-boundary condition ψ = 0.The field is then conditioned in such a manner that φ(x) ≥ 0 and macroscopictotal volume = v (i.e. N−d−1

∑x∈DN

φ(x) = v). They proved for the condi-tioned field that on several kind of scalings the macroscopic height variableshN = hN (θ); θ ∈ D = (0, 1)d converge as N → ∞ to hD,v = hD,v(θ)which minimizes the total surface tension 1

2

∫D|∇h|2 dθ under the three con-

ditions: h = 0 at ∂D, h ≥ 0 and∫

Dh dθ = v. The function hD,v describes the

Wulff shape.


Deuschel, Giacomin and Ioffe [77] generalized the results to the non-Gaussian setting. They considered the finite volume ϕ-Gibbs measure µ0

N

with 0-boundary conditions for general macroscopic domain D and generalpotential V satisfying (2.2), and proved the LD estimates, that is, the proba-bility that hN is close to a given macroscopic surface h ∈ H1

0 (D) behaves as

µ0N

(hN ∼ h

)

N→∞exp−NdΣD(h) , (6.1)

where ΣD(h) is the (integrated) total surface tension (or sometimes calledsurface free energy) of h defined by

ΣD(h) =∫

D

σ(∇h(θ)) dθ , (6.2)

and σ = σ(u) is the (normalized) surface tension with tilt u ∈ Rd introduced in

Sect. 5. Roughly saying, the asymptotic behavior (6.1) is obtained by patchingthe relation (5.3) for localized systems. Mathematically precise formulation for(6.1) is the usual LD upper and lower bounds, see Theorem 6.1 below. ThisLDP result is an analogue of that by Dobrushin, Kotecky and Shlosman [86]for the Ising model and ΣD(h) corresponds to the Wulff functional (1.1).

These results can be further generalized for the system with weak selfpotential (one-body potential) under general Dirichlet boundary conditionsψ. We therefore state the LDP result in such settings; Sect. 6.1 for higherdimensions and Sect. 6.3 for one dimension. As an application, the LLN isproved for the finite volume ϕ-Gibbs measures (without conditioning) andthe limit profile is characterized by a variational problem which was studiedby Alt and Caffarelli [5] and others. The minimizers generate free boundariesinside the domain; Sects. 6.2 and 6.3. We also discuss the ∇ϕ interface modelfor δ-pinning with quadratic potential in one dimension; Sect. 6.4. Sect. 6.5outlines the proof of Theorem 6.1. Sect. 6.6 is devoted to the LDP for empiricalmeasures of ϕ-field distributed under the Gaussian ϕ-Gibbs measures on Z

d

when d ≥ 3. This is sometimes called the third level LDP.

6.1 LDP with Weak Self Potentials

Setting and Assumptions

A bounded domain D in Rd with piecewise Lipschitz boundary is given and

microscopic regions DN ,DN and ∂+DN , N ∈ N in Zd are defined from D,

recall Sect. 1.4. The regularity assumption on ∂D is needed to employ someresults in the theory of partial differential equations (PDEs), see Lemma 6.17.The boundary condition ψ = ψ(x);x ∈ ∂+DN for microscopic height vari-ables is given.

We assume the space, which is d + 1 dimensional, is filled by a mediachanging in the distances from the hyperplane DN . Such situation can berealized by adding self potentials (one-body potentials) U : D×R → R to the

166 T. Funaki

original Hamiltonian HψN (φ) ≡ Hψ

DN(φ) introduced in (2.1) with the boundary

condition ψ in the following manner

Hψ,UN (φ) =

∑〈x,y〉⊂DN

V (φ(x)− φ(y)) +∑

x∈DN

U( x

N, φ(x)

). (6.3)

The first term in the right hand side is HψN (φ) and the interaction potential V

is always assumed to satisfy the conditions (2.2). The statistical ensemble forthe height variables φ is then defined by the finite volume ϕ-Gibbs measureon DN

µψ,UN (dφ) =

1

Zψ,UN

exp−Hψ,U

N (φ)dφDN

, (6.4)

where Zψ,UN ≡ Zψ,U

DNis the normalization constant so that µψ,U

N ∈ P(RDN ).We shall regard µψ,U

N ∈ P(RDN ) by considering φ(x) = ψ(x) for x ∈ ∂+DN

under µψ,UN as before. When U ≡ 0, µψ,0

N coincides with µψN ≡ µψ

DNdefined

by (2.4).We consider the case that the self potential U is represented as a product

U(θ, r) = Q(θ)W (r) of two functions Q : D → [0,∞) and W : R → R andassume the following conditions on Q and W , respectively:

(Q) Q is bounded and piecewise continuous,(W) W is measurable and there exists A ≥ 0 such (6.5)

that limr→+∞

W (r) = 0, limr→−∞

W (r) = −A and

W (r) ∈ [−A, 0] for every r ∈ R.

The self potential U is called weak since it is bounded. A typical example ofW we have in mind is a function of the form

W (r) = −A · 1r<0, r ∈ R . (6.6)

This potential describes the situation that the space is filled by two differentmedia above and below the hyperplane DN . Since we assume A ≥ 0, the neg-ative values are more favorable than the positive ones for the interface heightvariables φ under the Gibbs measures. In other words the interface is weaklyattracted to the negative side, namely by the media below the hyperplaneDN . The opposite case A ≤ 0 can be easily reduced to our case A ≥ 0 byturning the interfaces upside down by the map φ → −φ and ψ → −ψ. More-over, adding a constant to W does not make any change in the Gibbs measureµψ,U

N so that, without loss of generality, we have assumed limr→+∞W (r) = 0in (6.5)-(W).

The microscopic boundary condition ψ should be scaled to have macro-scopic limits. We therefore assume that the following conditions hold forψ ∈ R

∂+DN with some C > 0, g ∈ C∞(Rd) and p0 > 2:


(ψ1) maxx∈∂+DN

|ψ(x)| ≤ CN ,

(ψ2)∑

x∈∂+DN

∣∣∣ψ(x)−Ng( x

N

)∣∣∣p0

≤ CNd .

(6.7)These conditions roughly mean that ψ(x)/N ∼ g(x/N) at x ∈ ∂+DN .

Scaling and Polilinear Interpolation

The aim is to study the macroscopic behavior of the microscopic height vari-ables φ = φ(x);x ∈ DN under the Gibbs measures µψ,U

N as N → ∞.The scaling connecting microscopic and macroscopic levels was introduced by(2.16) associating the macroscopic height variables hN = hN (θ); θ ∈ D withφ as step functions on D satisfying

hN( x

N

)=

1Nφ(x), x ∈ DN . (6.8)

However, from certain technical reasons, it turns out to be more tractable todefine hN by polilinear interpolation of the macroscopic variables on 1

NDN

determined by (6.8), i.e., for general θ ∈ D, we set

hN (θ) =∑

λ∈0,1d

[d∏

i=1

(λiNθi+ (1− λi)(1− Nθi)

)]hN

([Nθ] + λ

N

),

(6.9)where · denotes the fractional part, see (1.17) of [77]. In one dimension,(6.9) is just the usual polygonal approximation of hN (x/N), see (6.21).

LDP Result

Now we are in the position to state the LDP result. Define H1g (D) = h ∈

H1(D);h − g∣∣D∈ H1

0 (D) for g ∈ C∞(Rd), where H10 (D) stands for the

Sobolev space on D determined from the 0-boundary condition. The functiong∣∣∂D

serves for the macroscopic boundary condition as in (6.7).

Theorem 6.1. [77, 123] The family of random surfaces hN (θ); θ ∈ D dis-tributed under µψ,U

N satisfies the LDP on the space L2(D) with speed Nd andthe rate functional IU

D(h), that is, for every closed set C and open set O ofL2(D) we have that

lim supN→∞

1Nd

logµψ,UN (hN ∈ C) ≤ − inf

h∈CIUD(h) , (6.10)

lim infN→∞

1Nd

logµψ,UN (hN ∈ O) ≥ − inf

h∈OIUD(h) . (6.11)

The functional IUD(h) is given by

168 T. Funaki

IUD(h) =

ΣU

D(h)− infH1

g(D)ΣU

D if h ∈ H1g (D) ,

+∞ otherwise ,

where infH1

g(D)ΣU

D = infΣUD(h);h ∈ H1

g (D) and

ΣUD(h) =

∫D

σ(∇h(θ)) dθ −A

∫D

Q(θ)1(h(θ) ≤ 0) dθ . (6.12)

The first term in the right hand side is ΣD(h) defined by (6.2) for h ∈ H1(D).

The unnormalized rate functional ΣUD(h) is lower semicontinuous on

L2(D), which can be shown by (5.10).

Remark 6.1. Consider the case where ψ ≡ 0 (i.e., g ≡ 0) and U ≡ 0. Thenthe LDP rate functional is given by ΣD(h). Since the surface tension σ(u)attains its minimal value 0 at u = 0, the minimizer is h ≡ 0. In fact, sincePoincare’s inequality for h ∈ H1

0 (D) and then (5.10) with u = 0 imply

‖h‖2L2(D) ≤ C‖∇h‖2L2(D) ≤2Cc−

ΣD(h) ,

taking C = h ∈ L2(D); ‖h‖L2(D) ≥ a, a > 0 in (6.10), we obtain

µ0,0N (‖hN‖L2(D) ≥ a) ≤ e

−(

a2c−2C −ε

)Nd

for every ε > 0 and sufficiently large N . This means that the macroscopicinterface is flat with high probability and tilted surface appears with extremelysmall probability.

Remark 6.2. Since ∂D is piecewise Lipschitz and g∣∣D∈ C∞(D), by The-

orems 8.7 and 8.9 of [253], there exists a continuous linear trace operatorT0 : H1(D) → H

12 (∂D) such that T0u = u

∣∣∂D

for every u ∈ C∞(D) and itholds that H1

g (D) = h ∈ H1(D);T0h = g∣∣∂D.

Remark 6.3. The Gaussian case with 0-boundary condition was studied by[12] regarding hN ∈ Lp(D), 2 ≤ p < 2d/(d − 2) when D = (0, 1)d. For thegeneral Dirichlet boundary condition, the mean of the Gaussian field is shiftedby a harmonic function and therefore one can easily establish the LDP applyingthe contraction principle, see the proof of Lemma 6.6 below.

Remark 6.4. Sheffield [230] improved the topology for the LDP using Orlicz-Sobolev spaces. In addition, he discussed the LDP jointly for macroscopicheight variables and empirical measures (cf. Sect. 6.6).

Remark 6.5. If Q ≡ 1 and U is given by U(θ, r) = W (r), then it holds that

−A = − lim→∞

1|Λ|

logZ0,U

Λ

Z0Λ

. (6.13)


The right hand side represents the difference of the free energies of the inter-face in two cases with and without self potential, see (6.33) with a = b = 0in one dimension. In this sense, ΣU

D(h) above represents macroscopic totalsurface energy of the profile h.

The LDP result of Deuschel et al. [77] is a special case of Theorem 6.1:ψ ≡ 0 and U ≡ 0 so that A = 0. However, the actual proof of Theorem6.1 is given in a converse way. We reduce it to the case of U ≡ 0, sincethe potential U is weak and can be treated as a rather simple perturbation.The main effort in [123] was therefore made for the treatment of the generalboundary conditions. By a simple shift the problem can be reduced to the0-boundary case, however with bond-depending interaction potentials. Theproof of Theorem 6.1 will be outlined in Sect. 6.5. Instead, a complete proofwill be given for one dimensional system with quadratic potentials in Sect.6.3. The author takes this way, since it may be acceptable for a wide varietyof readers including nonexperts. In one dimension, one can prove the LDPunder uniform topology rather than the L2-topology.

6.2 Concentration Properties

We can deduce from Theorem 6.1 the LLNs for hN distributed under µψ,UN

(i.e., ϕ-Gibbs measure) or under its conditional probability (i.e., canonical ϕ-Gibbs measure) as N →∞ and the limits h = h(θ); θ ∈ D are characterizedby certain variational principles.

Wulff Shape

The macroscopic shape of the droplet put on a hard wall and having a definitevolume v(> 0) can be determined by the LLN for a conditioned field of hN =hN (θ); θ ∈ D. The conditions are introduced in such a manner that hN ≥ 0(wall condition; wall is put at the height level h ≡ 0) and

∫DhN (θ) dθ ≥ v (or

= v, constant volume condition).

Corollary 6.2. (Wall and constant volume conditions) For every v ≥ 0, un-der the conditional probability µ+

N,v = µ0,0N

(·∣∣hN ≥ 0,

∫DhN (θ) dθ ≥ v

)(note

that we take ψ ≡ 0, U ≡ 0), the LLN

limN→∞

µ+N,v(‖hN − hD,v‖L2(D) > δ) = 0 ,

holds for every δ > 0, where hD,v is the unique minimizer called Wulff shapeof the variational problem

minΣD(h);h ∈ H1

0 (D), h ≥ 0,∫

D

h(θ) dθ = v

. (6.14)

170 T. Funaki

Proof. We first notice that, if d ≥ 2, denoting µN ≡ µ0,0N

µN (Ω+(DN )) ≥ e−CNd−1(6.15)

for some C > 0, where Ω+(DN ) = φ ∈ RDN ;φ(x) ≥ 0 for every x ∈ DN,

see [76] and Theorem 7.2 (entropic repulsion). This bound claims that theprobability µN (Ω+(DN )) is large enough compared with the LDP estimate(at the order of e−CNd

). Setting the volume condition

Av =φ ∈ R

DN ;∫

D

hN (θ) dθ ≥ v

,

we have that

µ+N,v(‖hN − hD,v‖L2(D) > δ)

=µN (‖hN − hD,v‖L2(D) > δ,Ω+(DN ) ∩Av)

µN (Ω+(DN ) ∩Av)

≤µN (‖hN − hD,v‖L2(D) > δ,Av)

µN (Ω+(DN ))µN (Av).

The last inequality is a consequence of the FKG inequality for the denomina-tor. However, Theorem 6.1 implies

lim supN→∞

1Nd

log µN (‖hN − hD,v‖L2(D) > δ,Av) < −Σ∗v ,

lim infN→∞

1Nd

logµN (Ω+(DN ))µN (Av)

≥ −Σ∗

v ,

where

Σ∗v := inf

ΣD(h); h ∈ H1

0 (D), h ≥ 0,∫

D

h(θ) dθ = v

.

We have applied (6.15) for the second, and these two estimates prove theconclusion. Note that the value of Σ∗

v is unchanged if the conditions “h ≥0,

∫Dh(θ) dθ = v” are replaced with “

∫Dh(θ) dθ ≥ v”.

Remark 6.6. Bolthausen and Ioffe [31] proved the LLN for the Gibbs mea-sure on the wall with δ-pinning and quadratic potential under the constantvolume condition in two dimension (i.e., for interfaces in 2 + 1 dimensionalspace). The limit called Winterbottom shape is uniquely (except transla-tion) characterized by a certain variational problem, see Sect. 7.3.

The Euler equation for the minimizer hD,v of the variational problem(6.14) has the following form of the elliptic PDE:

div

(∇σ)(∇hD,v(θ))

= −cD,v, θ ∈ D ,

hD,v(θ) = 0, θ ∈ ∂D ,(6.16)


where cD,v is an appropriate constant. Indeed, the minimizer h satisfies

d

dεΣD(h+ εg)

∣∣∣∣ε=0

= 0

for all g such that∫

Dg(θ) dθ = 0. This implies that

∫D

∇σ(∇h(θ)) · ∇g(θ) dθ = −∫

D

div [∇σ(∇h)] g dθ = 0 ,

and leads us to (6.16).Dobrushin and Hryniv [85] studied the fluctuation of the Wulff shape

when d = 1. They adopted the random walk model, i.e. the SOS type model φ :0, 1, . . . , N → Z in one dimension under the condition that the macroscopicvolume of φ is always constant:

1N

N∑x=1

1Nφ(x) = v, v ∈ R .

They proved, under the one-sided Dirichlet boundary condition (i.e., φ(0) = 0)or under the two-sided conditions (i.e., φ(0) = 0, φ(N) = Nb), the LLN andthe CLT for the macroscopic height variables hN = hN (θ); 0 ≤ θ ≤ 1 definedby the polygonal approximation of φ(x)/N:(1) LLN: hN (θ) → h(θ) (N →∞), where h is the Wulff shape.(2) CLT (Fluctuation of hN around h):

√N(hN (θ)− h(θ)) =⇒ Gaussian process .

They did not impose the wall condition. See Higuchi et al. [146] for the ex-tension to the two dimensional lattice Widom-Rowlinson model.

Remark 6.7. (Wulff shape from the Ising model, cf. Sect. 1.1) For the twodimensional Ising model with nearest neighbor and ferromagnetic (attractive)interactions, it is well-known that there exists the critical temperature Tc > 0such that if T < Tc the system has the positive spontaneous magnetizationm∗ = m∗(T ) > 0. Let µN,m be the canonical Gibbs measure for such Isingmodel on [−N,N ]2 ∩ Z

2 with + boundary condition, which is obtained byconditioning the finite volume Gibbs measure in a way that the sample averageof the spins is m for some |m| < m∗. Dobrushin, Kotecky and Shlosman [86]proved under µN,m the macroscopic region occupied by − spins converges tothe Wulff shape, except translations, as N → ∞. Afterward, Ioffe [150, 151],Ioffe and Schonmann [152] extended this result for all T < Tc applying themethod of percolation.

Pisztora [213] invented the so-called L1-theory on the local sample averagesof the spins. This method is applicable to three dimension and higher, and usesthe idea based on the renormalization group called Pisztora’s coarse graining,

172 T. Funaki

see also Cerf and Pisztora [52]. It is believed that the Wulff shape has facetsif T < TR (roughening transition, ∃TR < Tc), see [233]. The review paper byBodineau, Ioffe and Velenik [22] is recommended to catch the whole picture ofthe results on the Ising model, including the derivation of the Winterbottomshape, see Sect. 7.3. See [1] for results by the middle of 1980s.

Cerf and Pisztora [53] studied the LDP under phase coexistence for Ising,Potts and random cluster models in dimensions d ≥ 3 for T < Tc. See also[3, 51].

Remark 6.8. Cohn et al. [58] considered the SOS type model on Z2 (i.e., φ :

Z2 → Z) induced from the domino tiling with equal probabilities for all possible

tilings, and proved the LLN and the LDP for the corresponding macroscopicheight variables. See also Kenyon [168].

Remark 6.9. The Wulff construction at zero temperature, but for a widerclass of Gibbs models, was studied by Descombes and Pechersky [73].

Alt-Caffarelli’s Variational Problems

The upper bound (6.10) in Theorem 6.1 implies the LLN for hN distributedunder µψ,U

N .

Corollary 6.3. If ΣUD has a unique minimizer h in H1

g (D), then the LLNholds under µψ,U

N , namely,

limN→∞

µψ,UN (‖hN − h‖L2(D) > δ) = 0 ,

for every δ > 0.

The variational problems for minimizing ΣUD were thoroughly studied by

Alt and Caffarelli [5] for nonnegative macroscopic boundary data g (one phaseproblem) with A > 0 and by Alt, Caffarelli and Friedman [6] for general g(two phases problem) especially when σ is quadratic: σ(u) = |u|2, and byWeiss [252] for more general σ. The minimizer h = h of ΣU

D generates thefree boundaries inside D. If the surface tension σ = σ(u) is smooth enough(i.e., σ ∈ C2,γ(Rd), γ > 0) and if the free boundary ∂h > 0 of the minimizerh is locally C2, then h satisfies the Euler equation

div ∇σ(∇h) = 0

in D \ ∂h > 0 and the condition

Ψ(∇h+)− Ψ(∇h−) = AQ (6.17)

on the free boundary D ∩ ∂h > 0, where Ψ(u) = u · ∇σ(u) − σ(u). TheLipschitz continuity of the minimizer h and the regularity of its free bound-ary were studied by the papers listed above and others. In our case, for theregularity of the surface tension, σ ∈ C1,1(Rd) is only known in general, recallTheorem 5.3-(1).


6.3 LDP with Weak Self Potentials in one Dimension

In this section we reformulate Theorem 6.1 in one dimension (i.e., we considerinterfaces in 1+1 dimensional space) and give a complete proof of the theorem.As we have already noticed, in one dimension, one can argue under a strongertopology determined by the uniform norm.

Reformulation of the Results

Let us take D = (0, 1) ⊂ R so that DN = 1, 2, . . . , N − 1 and ∂+DN =0, N. For simplicity, we consider the case of the quadratic potential V (η) =12η

2 with Q ≡ 1 in the self potential U(θ, r). The corresponding Gibbs measurefor the height variables φ = φ(x);x ∈ DN is then defined by

µa,b,WN (dφ) =

1

Za,b,WN

exp−Ha,b,W

N (φ)dφDN

, (6.18)

under the boundary conditions

ψ(0) = aN, ψ(N) = bN (6.19)

for some a, b ∈ R. The corresponding macroscopic boundary conditions areg(0) = a and g(1) = b at ∂D = 0, 1, recall (6.7)-(ψ2). The HamiltonianHa,b,W

N is given by

Ha,b,WN (φ) =

12

N−1∑x=0

(φ(x+ 1)− φ(x))2 +N−1∑x=1

W (φ(x)) , (6.20)

and Za,b,WN is the normalization constant. The formulas (6.18) and (6.20) cor-

respond to (6.4) and (6.3), respectively. The function W satisfies the condition(W) in (6.5).

The macroscopic height variable hN = hN (θ); θ ∈ [0, 1] is defined fromφ by the interpolation (6.9) which is, in one dimension, the polygonal approx-imation of hN (x/N) = φ(x)/N ;x ∈ DN:

hN (θ) =(θ − x

N

)φ(x+ 1) +

(x+ 1N

− θ

)φ(x),

x

N≤ θ ≤ x+ 1

N. (6.21)

Introduce two function spaces

Ca,b = h ∈ C([0, 1]);h(0) = a, h(1) = b,H1

a,b = h ∈ Ca,b;h is absolutely continuous

and its derivative h′ ∈ L2([0, 1]) .

The space Ca,b is endowed with the topology determined by the uniform-norm‖ · ‖∞. The function hN belongs to Ca,b. Since the normalized surface tension

174 T. Funaki

is σ(u) = 12u

2 for V (η) = 12η

2 (recall Proposition 5.2 and Example 5.1), thetotal surface tension of h ∈ Ca,b defined by the formula (6.2) has the form

Σ(h) ≡ Σ(0,1)(h) =12

∫ 1

0

(h′)2(θ) dθ

for h ∈ H1a,b and Σ(0,1)(h) = +∞, otherwise. We set

ΣW (h) ≡ ΣW(0,1)(h) = Σ(h)−A|h ≤ 0| , (6.22)

where | · | stands for the Lebesgue measure and h ≤ 0 = θ ∈ [0, 1];h(θ) ≤0. Note that ΣW corresponds to ΣU

D defined by (6.12). It is lower semicon-tinuous in h ∈ Ca,b and good in the sense that h ∈ Ca,b;ΣW (h) ≤ iscompact in Ca,b for each ∈ R.

Theorem 6.4. Under µa,b,WN , the family of the macroscopic height variables

hN defined by (6.21) satisfies the LDP on the space Ca,b with speed N and theunnormalized rate functional ΣW , that is, for every closed set C and open setO of Ca,b we have that

lim supN→∞

1N

logµa,b,WN (hN ∈ C) ≤ − inf

h∈CIW (h) , (6.23)

lim infN→∞

1N

logµa,b,WN (hN ∈ O) ≥ − inf

h∈OIW (h) , (6.24)

where IW (h) = ΣW (h)− infH1

a,b

ΣW is the normalized functional of ΣW .

Concentrations

Before giving the proof of Theorem 6.4, we reformulate Corollary 6.3 in onedimensional setting and study the minimizers of the functional ΣW , whichexhibit different aspects depending on the boundary conditions a and b.

Corollary 6.5. If ΣW has a unique minimizer h in H1a,b, then the LLN holds

under µa,b,WN , namely,

limN→∞

µa,b,WN

(‖hN − h‖∞ > δ

)= 0 ,

for every δ > 0.

This result is related to those obtained by Pfister and Velenik [212]. Theyconsidered the two dimensional Ising model at low temperature on a large boxwith attractive wall set at the bottom line. This line segment corresponds toour hyperplane DN , although it has an effect of hard wall at the same time,since the interfaces separating ±-phases can not go down beyond the bottomline in their setting. One of the motivations of [212] was to understand theso-called wetting or pinning/depinning transition, which will be discussed inSect. 7.3 for the ∇ϕ interface model.


Minimizers of ΣW

The functional ΣW is essentially the same as W (C) defined by (4.1) in [212],which is derived from the two dimensional Ising model; note that h(θ) ≥ 0 intheir case. The minimizer of W (C) was studied in Proposition 4.1 of [212].

Case 1. a, b > 0: The straight line h(1) connecting (0, a) and (1, b), i.e.,

h(1)(θ) = (1− θ)a+ bθ, θ ∈ [0, 1]

is a critical point of ΣW ; see Fig. 6.1. In fact, ΣW (h) = Σ(h) for h alwaysstaying in the positive side h > 0 and, for such h, the Euler equationδΣW /δh(θ) = −h′′(θ) = 0 means that h is a linear function.

Since the second term of ΣW in (6.22) makes it smaller if h stays longerin the nonpositive side h ≤ 0, there is another candidate for minimizers.Let h(2) be the curve composed of three straight line segments connectingfour points (0, a), P1(θ1, 0), P2(1 − θ2, 0) and (1, b) in this order; see Fig. 6.2.The angles at two corners P1 and P2 are both equal to α ∈ [0, π/2], which is

0 1

a

b

Fig. 6.1. The function h(1)

0 1

a

b

P1 P2


determined by the Young’s relation:

tanα =√

2A . (6.25)

Then, h(2) is a critical point of ΣW . Indeed, as we have explained above, ifthe curve is straight in the positive side h > 0, the energy is smaller. Onceh reaches the side h ≤ 0, the energy is minimal if it stays on h = 0, sinceit is a straight line and gives no contribution to Σ(h). To derive the anglerelation (6.25), set

ΣW (h(2)) ≡ F (θ1, θ2) =a2

2θ1−A(1− θ1 − θ2) +

b2

2θ2.

Then, ∂F/∂θ1 = ∂F/∂θ2 = 0 shows

176 T. Funaki

a

θ1=

b

θ2=√

2A , (6.26)

which implies (6.25). One can derive (6.25) from the free boundary condition(6.17) as well. Indeed, by Ψ(u) = |u|2− 1

2 |u|2 = 12 |u|2 and Q = 1, (6.17) reads

|∇h+|2 − |∇h−|2 = 2A and this implies ∇h+ =√

2A since ∇h− = 0. Thecondition (6.25) is the same as [49] discussed.

Since h(2)(θ) is described as

h(2)(θ) = (a−√

2Aθ)1θ≤θ1 + (b−√

2A(1− θ))1θ≥1−θ2

with θ1 and θ2 defined by (6.26), h ≡ h(2) satisfies an equation

h′′ = ν, ν =√

2A∑

θ∈∂θ;h(θ)=0

δθ , (6.27)

in the sense of generalized functions, namely,

〈h, J ′′〉 =√

2A(J(θ1) + J(1− θ2))

for every J ∈ C∞0 (0, 1). The equation (6.27) may be regarded as the Euler

equation for the functional ΣW .Case 2. a > 0, b < 0: Let h(3) be the curve composed of two straight line

segments connecting three points (0, a), P (θ1, 0) and (1, b) in this order; seeFig. 6.3. The angles at the corner P of the first and second segments to thehorizontal line are denoted by α and β ∈ [0, π/2], respectively, and obey therelation

tan2 α− tan2 β = 2A . (6.28)

These two angles depend on the boundary conditions in such a way that

a

tanα− b

tanβ= 1 .

0

1

a

b

P



Then, h(3) is a critical point of ΣW . Indeed, the critical curve has to bestraight both in the positive and nonpositive sides. The relation (6.28) isderived similarly to Case 1: Set

ΣW (h(3)) ≡ F (θ1) =a2

2θ1−A(1− θ1) +

b2

2(1− θ1)

and F ′(θ1) = 0 implies (6.28). The function h ≡ h(3) satisfies an equation

h′′ = ν, ν = (tanα− tanβ)δθ1 ,

in the sense of generalized functions.Case 3. a, b < 0: The minimizer of ΣW is the straight line connecting

(0, a) and (1, b).

Proof of Theorem 6.4

Now let us give the proof of Theorem 6.4. We prepare three lemmas, first ofwhich discusses the LDP for the finite volume Gibbs measure µa,b,0

N definedby (6.18) taking W ≡ 0 in the Hamiltonian.

Lemma 6.6. Under µa,b,0N , the family of surfaces hN satisfies the LDP on

Ca,b with speed N and the rate functional Σa,b,0(h) = Σ(h)− (b− a)2/2.

Proof. Let w = w(x);x ∈ [0, N ] be the one dimensional standard Brownianmotion starting at 0 and set hN (θ) = w(Nθ)/N, θ ∈ [0, 1]. Then, by Schilder’stheorem (see, e.g., Theorem 5.1 of [246]; Mogul’skii [200] discusses the randomwalk with general transition probabilities), the LDP holds for hN on C0 =C([0, 1]) ∩ h(0) = 0 with the rate functional Σ(h). Define φ = φ(x);x ∈[0, N ] from w as φ(x) = w(x) − xw(N)/N + (N − x)a + xb. Then, φ isthe pinned Brownian motion satisfying (6.19) and φ(x);x ∈ DN is µa,b,0

N -distributed; see Proposition 2.2. Set hN (θ) = φ(Nθ)/N, θ ∈ [0, 1], and considerthe mapping Φ : h ∈ C0 → h ∈ Ca,b defined by

Φ(h)(θ) = h(θ)− θh(1) + (1− θ)a+ θb .

Then, Φ is continuous and hN = Φ(hN ) holds. Therefore, by the contrac-tion principle (cf. [246], [79] and [72, Theorem 4.2.1]), the LDP holds for hN

with the rate functional Σ(h) = inf h∈C0:Φ(h)=h Σ(h), which coincides withΣa,b,0(h).

The proof of lemma is completed by showing a super exponential estimatefor the difference between hN and hN as in p.17 of [246]: For every δ > 0,

178 T. Funaki

P(‖hN − hN‖∞ ≥ δ

)≤

N−1∑x=0

P

(sup

θ∈[x/N,(x+1)/N ]

∣∣∣(θ − x

N

)w(x+ 1)

+(x+ 1N

− θ

)w(x)− 1

Nw(Nθ)

∣∣∣∣ ≥ δ

)

= NP

(sup

θ∈[0,1]

|w(θ)− θw(1)| ≥ Nδ

)

≤ 4NP (|w(1)| ≥ Nδ/2) = exp−N

2δ2

8+ o(N2)

,

as N →∞.

For g ∈ Ca,b and δ > 0, set B∞(g, δ) = h ∈ Ca,b; ‖h − g‖∞ < δ. Thenext lemma estimates the second term in the Hamiltonian Ha,b,W

N in (6.20).

Lemma 6.7. Let g ∈ Ca,b and 0 < δ < 1 be fixed. If hN ∈ B∞(g, δ) for Nlarge enough, then there exists some constant C > 0 such that

−AN |g ≤ 3δ| − CNδ ≤∑

x∈DN

W (φ(x)) ≤ −AN |g ≤ −3δ|+ CNδ ,

for every N large enough.

Proof. The condition (6.5)-(W) on W is needed. In fact, the upper bound isshown as ∑

x∈DN

W (φ(x)) ≤∑

x∈DN :g(x/N)≤−2δ

W (φ(x))

≤ (−A+ δ) · x ∈ DN : g(x/N) ≤ −2δ≤ −AN |g ≤ −3δ|+ CNδ ,

for every sufficiently large N . Here, the first inequality follows from W ≤ 0,the second one is because, if hN ∈ B∞(g, δ), φ(x) = NhN (x/N) ≤ −Nδ sothat W (φ(x)) → −A as N → ∞ uniformly in x ∈ DN in the sum, and thethird one is by the uniform continuity of g. The lower bound is similar:

∑x∈DN

W (φ(x)) ≥∑

x∈DN :g(x/N)>2δ

W (φ(x))−A · x ∈ DN : g(x/N) ≤ 2δ

≥ −CNδ −AN |g ≤ 3δ| ,

where the first inequality is from W ≥ −A and the second one is because, ifhN ∈ B∞(g, δ), φ(x) ≥ Nδ so that W (φ(x)) → 0 as N → ∞ uniformly inx ∈ DN : g(x/N) > 2δ.

Let ΣW− be the functional ΣW with |h ≤ 0| replaced by |h < 0|, i.e.,

ΣW− (h) = Σ(h)−A|h < 0| . (6.29)


Lemma 6.8. For every open set O of Ca,b, we have that

infh∈O

ΣW (h) = infh∈O

ΣW− (h) .

Proof. Since ΣW (h) ≤ ΣW− (h) is obvious for every h ∈ Ca,b, the conclusion

follows once we can show that

infh∈O

ΣW (h) ≥ infh∈O

ΣW− (h) . (6.30)

To this end, for every ε > 0, take h ∈ O(∩H1a,b) such that ΣW (h) ≤ infO ΣW +

ε. We approximate such h by a sequence hnn≥1 defined by hn(θ) = h(θ)−fn(θ), where fn ∈ C∞

0 ((0, 1)) are functions such that fn(θ) ≡ 1n for θ ∈

[ 1n , 1 −

1n ] and |(fn)′(θ)| ≤ 2 for every θ ∈ (0, 1

n ) ∪ (1 − 1n , 1); note that

hn ∈ H1a,b. Then, since limn→∞Σ(hn) = Σ(h) and

−|hn < 0| ≤ −|h ≤ 0|+ 2n,

we obtain that lim supn→∞ΣW− (hn) ≤ ΣW (h). However, O is an open set of

Ca,b, so that hn ∈ O for n large enough and thus (6.30) is shown.

Proof (Theorem 6.4). Let Σa,b,W and Σa,b,W− be the functionals ΣW and ΣW

−defined by (6.22) and (6.29) with Σ replaced by Σa,b,0, respectively.

Step 1 (Lower bound). Let g ∈ Ca,b and 0 < δ < 1 be given. Then, by theupper bound in Lemma 6.7 and the LD lower bound for µa,b,0

N (Lemma 6.6),we have

lim infN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ B∞(g, δ)

)]

≥ lim infN→∞

1N

log[exp (AN |g ≤ −3δ| − CNδ) · µa,b,0

N

(hN ∈ B∞(g, δ)

)]

≥ A|g ≤ −3δ| − Cδ − infh∈B∞(g,δ)

Σa,b,0(h) .

Now, suppose that an open set O of Ca,b is given. Then, for every h ∈ O andδ > 0 such that B∞(h, δ) ⊂ O, we have that

lim infN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ O

)]≥ −Σa,b,0(h) +A|h ≤ −3δ| − Cδ .

Letting δ ↓ 0, since h ∈ O is arbitrary, we have

lim infN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ O

)]≥ − inf

h∈OΣa,b,W

− (h) . (6.31)

However, by Lemma 6.8, Σa,b,W− (h) can be replaced with Σa,b,W (h) in the

right hand side of (6.31).

180 T. Funaki

Step 2 (Upper bound). Similarly, by the lower bound in Lemma 6.7 andthe LD upper bound for µa,b,0

N (Lemma 6.6), we have

lim supN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ B∞(g, δ)

)]

≤ lim supN→∞

1N

log[exp (AN |g ≤ 3δ|+ CNδ) · µa,b,0

N

(hN ∈ B∞(g, δ)

)]

≤ A|g ≤ 3δ|+ Cδ − infh∈B∞(g,δ)

Σa,b,0(h) ,

where B∞(g, δ) = h ∈ Ca,b; ‖h− g‖∞ ≤ δ is the closure of B∞(g, δ) in Ca,b.Then, by using the lower semicontinuity of Σa,b,0(h) and the right-continuityof |g ≤ 3δ| in δ, we see that for every g ∈ Ca,b and ε > 0, there exists δ > 0such that

lim supN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ B∞(g, δ)

)]≤ −Σa,b,W (g) + ε .

Therefore, the standard argument in the theory of LDP [72, 79, 246] yieldsthe upper bound

lim supN→∞

1N

log

[Za,b,W

N

Za,b,0N

µa,b,WN

(hN ∈ C

)]≤ − inf

h∈CΣa,b,W (h) (6.32)

for every compact set C of Ca,b. However, since W is bounded, the exponentialtightness for µa,b,0

N implies that for µa,b,WN : For every M > 0, there exists a

compact set K ⊂ Ca,b such that

lim supN→∞

1N

log µa,b,WN (Kc) ≤ −M .

Thus, (6.32) holds for every closed set C of Ca,b.Taking O = C = Ca,b in (6.31) and (6.32), we see that

limN→∞

1N

logZa,b,W

N

Za,b,0N

= − infh∈Ca,b

Σa,b,W (h) (6.33)

and this concludes the proof of the theorem.

6.4 LDP for δ-Pinning in one Dimension

Gibbs Measures with Pinning Potentials

Let us go back to the d dimensional setting in Sect. 6.1. The pinning is aneffect of weak force which attracts interfaces φ toward the level of height 0,


i.e., to a neighborhood of the hyperplane DN . Such effect is again realizedby adding self potentials U to the original Hamiltonian. We assume Q ≡ 1so that U(θ, r) = W (r) and denote the finite volume ϕ-Gibbs measure µψ,U

N

introduced in (6.4) by µψ,WN . Specifically, we consider the following two types

of pinning potentials.

Square-well pinning: The potential W has a form

W (r) = −b1|r|≤a, r ∈ R (6.34)

with a, b > 0. The constant s = 2a(eb−1) is called the strength of pinning.As s increases, the effect of pinning becomes stronger.δ-pinning: Under the limit a ↓ 0, b → ∞ keeping s = eJ constant forJ ∈ R, we have that

eb1|φ(x)|≤adφ(x) =⇒ eJδ0(dφ(x)) + dφ(x)

for each x ∈ DN . In this way, the finite volume Gibbs measure with δ-pinning is introduced as a weak limit of the Gibbs measure µψ,W

N withsquare-well pinning and has the following form

µψ,JN (dφ) =

1

Zψ,JN

exp−Hψ

N (φ) ∏

x∈DN

(eJδ0(dφ(x)) + dφ(x)) . (6.35)

We regard µψ,JN ∈ P(RDN ) by considering φ(x) = ψ(x) for x ∈ ∂+DN as

before. The larger J gives stronger pinning. When J = −∞, there is nopinning and µψ,−∞

N coincides with µψN = µψ

DNdefined by (2.4).

The square-well pinning potential W of (6.34) obviously does not satisfy thecondition (W) in (6.5) and the LDP is not established yet. Several proper-ties of the Gibbs measures µψ,W

N , µψ,JN with square-well or δ-pinnings will be

discussed in the subsequent section, Sect. 7.2.

LDP Result

The LDP is established for δ-pinning in one dimension with quadratic poten-tial V (η) = 1

2η2. Let us take D = (0, 1) ⊂ R so that DN = 1, 2, . . . , N − 1,

and consider the Gibbs measure µψ,JN under the boundary conditions (6.19):

ψ(0) = aN,ψ(N) = bN for a, b ∈ R. We denote µψ,JN , Zψ,J

N and ZψN (= Zψ,−∞

N )as µa,b,J

N , Za,b,JN and Za,b

N , respectively.

Theorem 6.9. (Funaki and Sakagawa [123]) Under µa,b,JN , the family of

macroscopic random surfaces hN (θ); θ ∈ [0, 1] defined by (6.21) satisfiesthe LDP on Ca,b with speed N and the rate functional IJ(h) ≡ IJ

(0,1)(h) =ΣJ(h)− inf

H1a,b

ΣJ (if h ∈ H1a,b and = +∞ otherwise), where

182 T. Funaki

ΣJ (h) ≡ ΣJ(0,1)(h) = Σ(0,1)(h) + τpin(J)|h = 0| ,

and

τpin(J) = − limN→∞

1N

logZ0,0,J

N

Z0,0N

. (6.36)

The function τpin(J) is called the pinning free energy. It is known thatthe limit exists and τpin(J) < 0 for every J ∈ R, see [123].

Minimizers of ΣJ

The minimizers of ΣJ are computable in a similar manner to what we didin Sect. 6.3 for ΣW . For instance, in the case where a > 0 and b < 0, theminimizer is either the straight line h(4) connecting (0, a) and (1, b) or thecurve h(5) composed of three straight line segments connecting four points(0, a), P1(θ1, 0), P2(1 − θ2, 0) and (1, b) in this order; see Figs. 6.4 and 6.5,respectively. If h(5) is realized, the angles at two corners P1 and P2 are bothequal to α ∈ [0, π/2] and obey the Young’s relation:

tanα =√−2τpin(J) .

6.5 Outline of the Proof of Theorem 6.1

LDP without Self Potentials

Similarly to the one dimensional case, the proof of Theorem 6.1 can be reducedto the LDP for the finite volume ϕ-Gibbs measure µψ

N (= µψ,0N ) without self

potentials. Indeed, the following proposition substitutes for Lemma 6.6. Thetopology under which the LDP holds becomes weaker when d ≥ 2.

0 1

a

b


0 1

a

P1 P2

b



Proposition 6.10. The family of random surfaces hN (θ); θ ∈ D distributedunder µψ

N satisfies the LDP on L2(D) with speed Nd and the rate functional

ID(h) =

ΣD(h)− inf

H1g(D)

ΣD if h ∈ H1g (D),

+∞ otherwise .

Once Proposition 6.10 is established, the lower bound in Theorem 6.1 isshown essentially in the same way as Theorem 6.4 with small modifications.For instance, the second estimate in Lemma 6.7 is changed as follows: If hN ∈B2(g, δ) = h ∈ L2(D); ‖h− g‖L2(D) < δ for some g ∈ L2(D), 0 < δ < 1 andfor N large enough, then there exists some constant C > 0 such that

∑x∈DN

U(x

N, φ(x)) ≤ −ANd

∫D

Q(θ)1(g(θ) ≤ −√δ) dθ + CNdδ ,

for every N large enough. Furthermore, Lemma 6.8 holds for the functionalΣU

D,−(h) defined by (6.12) with 1(h(θ) ≤ 0) replaced by 1(h(θ) < 0), i.e.,infh∈O ΣU

D(h) = infh∈O ΣUD,−(h) for every open set O of L2(D).

The proof of the upper bound, to treat the self potential term as a pertur-bation, requires slightly more careful consideration than for the lower bound,since the LDP upper bound for µψ

N (Proposition 6.10) is shown only in thespace L2(D) and not under the uniform topology. Since U is bounded, theexponential tightness for µψ,U

N can be proved from that for µψN which follows

from Lemma 6.14 below.

Treatment of Boundary Conditions

Proposition 6.10 is first established by [77] when the boundary condition ψ ≡0. By shifting the field, the problem with the general boundary condition ψcan be reduced to the 0-boundary case. Indeed, define φ as φ(x) = Ng( x

N ) forx ∈ DN (recall g ∈ C∞(Rd)) and, by shifting φ → φ− φ, introduce

HψN (φ) =

12

∑b∈DN

∗

V (∇(φ ∨ 0)(b) +∇(φ ∨ ψ)(b)) ,

and consider the associated finite volume ϕ-Gibbs measure having 0-boundarycondition:

µψN (dφ) =

1

ZψN

exp−HψN (φ)

∏x∈DN

dφ(x) .

Then the following LDP holds for µψN .

Proposition 6.11. The family of random surfaces hN (θ); θ ∈ D distributedunder µψ

N satisfies the LDP on L2(D) with speed Nd and the rate functional

184 T. Funaki

ID(h) =

ΣD(h)− inf

H10 (D)

ΣD if h ∈ H10 (D),

+∞ otherwise ,

whereΣD(h) =

∫D

σ(∇h(θ) +∇g(θ)) dθ .

Under the continuous map Φg : L2(D) → L2(D) given by Φg(h) = h + g,we have ID(h) = infID(h); h ∈ L2(D), Φg(h) = h. Hence, by the contractionprinciple, Proposition 6.10 follows from Proposition 6.11.

Proof of Proposition 6.11

(a) Convergence of Average Profiles

The proof of Proposition 6.11 is further reduced to showing the convergenceof average profiles, Lemma 6.12. We shall follow the strategy of [77]. For f ∈C∞

0 (D), set ρN = ρN (x) := f(x/N)/N ;x ∈ DN and define the HamiltonianHψ

N,f (φ) by (4.3) with Λ = DN and ρ = ρN . Another Hamiltonian HψN,f (φ)

is defined in a similar manner from HψN (φ). Then, consider the following two

finite volume ϕ-Gibbs measures:

µψN,f (dφ) =

1

ZψN,f

exp−HψN,f (φ)

∏x∈DN

dφ(x),

µψN,f (dφ) =

1

ZψN,f

exp−HψN,f (φ)

∏x∈DN

dφ(x) ,

having the different boundary conditions φ(x) = ψ(x) and φ(x) = 0 for x ∈∂+DN , respectively; recall that ψ and g satisfy the conditions (6.7). Twoprobability measures µψ

N,f and µψN,f are called Cramer transforms of µψ

N andµψ

N . We write the averages of the profile hN defined by (6.9) under µψN,f and

µψN,f as hψ

N,f (θ) = EµψN,f [hN (θ)] and hψ

N,f (θ) = EµψN,f [hN (θ)], respectively.

For f ∈ L2(D), hf denotes the unique weak solution h = h(θ) in H10 (D) of

the following elliptic PDE:

div(∇σ)(∇h(θ) +∇g(θ)) = −f(θ), θ ∈ D .

The crucial step in the proof of Proposition 6.11 is the following lemma.

Lemma 6.12. (1) hψN,f → hf in H1

0 (D) as N →∞.

(2) (LLN) limN→∞

EµψN,f [‖hN − hf‖2L2(D)] = 0.

Postponing the proof of this lemma later, we set


ΞψN,f ≡

ZψN,f

ZψN

= EµψN

[exp

1N

∑x∈DN

f( x

N

)φ(x)

].

Then, by calculating the functional derivative of ΣD(h) and using the trickto compute d

dt log ZψN,tf and integrate it in t ∈ [0, 1], Lemma 6.12-(1) yields

the following lemma.

Lemma 6.13. The limit ΛD(f) ≡ limN→∞

1Nd logΞψ

N,f exists and it holds that

ΛD(f) =∫

D

dθ

∫ 1

0

htf (θ)f(θ) dt

= suph∈H1

0 (D)

〈h, f〉 − ΣD(h)+ infH1

0 (D)ΣD ,

where 〈h, f〉 =∫

Dh(θ)f(θ) dθ. The supremum is attained at h = hf .

(b) Exponential Tightness

For the proof of the LDP upper bound in Proposition 6.11, we need thefollowing uniform exponential estimate.

Lemma 6.14. There exists ε > 0 such that

supN≥1

1Nd

logEµψN,f

exp

ε

∑x∈DN

(∣∣∣∣hN( x

N

) ∣∣∣∣2

+∣∣∣∣∇NhN

( x

N

) ∣∣∣∣2)

<∞ ,

where ∇Nu( xN ) = ∇N

i u( xN ) := N(u(x+ei

N )−u( xN ))1≤i≤d denotes the discrete

gradient of a scalar lattice field u = u( xN );x ∈ DN.

(c) Proof of Proposition 6.11

The upper bound is shown based on the exponential Chebyshev’s inequalitycombined with Lemmas 6.13 and 6.14. For the lower bound, we rely on theusual Cramer’s trick: By Lemmas 6.12-(1) and 6.13, it is easy to see that

limN→∞

1NdH(µψ

N,f |µψN ) = ID(hf ) ,

where H(µψN,f |µ

ψN ) is the relative entropy of µψ

N,f with respect to µψN ; recall

(5.4) and also see (5.4) in [77]. Combining this with Lemma 6.12-(2) andapplying the entropy inequality (cf. [79, Lemma 5.4.21]), we obtain

lim infN→∞

1Nd

log µψN (hN ∈ O) ≥ − inf

f∈C∞0 (D)

s.t. hf∈O

ID(hf ) = infh∈O

ID(h) ,

for every open set O ⊂ L2(D).

186 T. Funaki

Proof of Lemma 6.12

(a) Reduction to Lemma 6.16

We shall prove Lemma 6.12-(1). Lemma 6.12-(2) is shown from it by applyingBrascamp-Lieb inequality. The following lemma follows from (5.10).

Lemma 6.15. Let hnn≥1 be a sequence of H10 (D) and define Σf (h)

= ΣD(h) − 〈h, f〉. If limn→∞

Σf (hn) = infH1

0 (D)Σf , then hn → hf in H1

0 (D) as

n→∞.

Also by (5.10), we have

Σf (q)− Σf (hψN,f ) ≥

∫D

(∇q(θ)−∇hψN,f (θ)) · (∇σ)(∇hψ

N,f (θ) +∇g(θ)) dθ

−∫

D

(q(θ)− hψN,f (θ))f(θ) dθ ,

for every q ∈ C∞0 (D). Once we can prove that the right hand side goes to 0

as N → ∞ for every q ∈ C∞0 (D), we have lim

N→∞Σf (hψ

N,f ) = infH1

0 (D)Σf . This

combined with Lemma 6.15 completes the proof of Lemma 6.12. Hence, allwe have to prove are summarized in the following lemma.

Lemma 6.16. (1) For every q ∈ C∞0 (D),

limN→∞

∫D

∇q(θ) · (∇σ)(∇hψN,f (θ) +∇g(θ)) dθ =

∫D

q(θ)f(θ) dθ .

(2) Moreover, we have that

limN→∞

∫D

∇hψN,f (θ) · (∇σ)(∇hψ

N,f (θ) +∇g(θ)) dθ = limN→∞

∫D

hψN,f (θ)f(θ) dθ .

For the proof of Lemma 6.16, we need three lemmas.

(b) A Priori Bounds

The assumption that the domain D has a piecewise Lipschitz boundary isnecessary to show the following lemma by employing the PDE techniques likeCaccioppoli and inverse Holder inequalities.

Lemma 6.17. (1) (Lp-estimates) There exists some p ∈ (2, p0) such that

supN≥1

‖∇hψN,f‖Lp(D) <∞ and sup

N≥1‖∇hψ

N,f‖Lp(D) <∞ ,

where p0 > 2 is the constant appearing in the condition (ψ2).(2) (Fluctuation inequalities) For every e ∈ Z

d with |e| = 1, we have


limN→∞

1Nd

∑x∈DN

∣∣∣∣∇N hψN,f

(x+ e

N

)−∇N hψ

N,f

(x

N

)∣∣∣∣2

= 0 ,

limN→∞

1Nd

∑x∈DN

∣∣∣∣∇N hψN,f

(x+ e

N

)−∇N hψ

N,f

(x

N

)∣∣∣∣2

= 0 .

(c) Local Equilibria

Define QN ∈M+(D ×X ) and VN ∈M+(Rd ×X ) by

QN (dθdη) =1Nd

∑x∈DN

δ xN

(dθ)µψ,∇N,f τ−1

x (dη),

VN (dvdη) =1Nd

∑x∈DN

δ∇N hψN,f ( x

N )(dv)µψ,∇N,f τ−1

x (dη) ,

where X is the state space for the ∇ϕ-field introduced in Sect. 2.3, M+(E)stands for the class of all nonnegative measures on E , µψ,∇

N,f (dη) is the distrib-ution of η = ∇φ on X under µψ

N,f and τx : X → X denotes the shift on Zd, cf.

Definition 2.3. We regard µψN,f ∈ P(RZ

d

) by considering φ(x) = ψ(x)(= g( xN ))

for x ∈ Zd \DN as before.

Then, one can prove the following lemma, cf. Sect. 10.3-(c) for dynamics.

Lemma 6.18. For each r > 0 both the families of measures QN on D×Xr

and VN on Rd × Xr are tight, see Sect. 9.1 for the precise definition of

the space Xr. Moreover, for every limit point Q of QN, there exists νQ ∈M+(D × R

d) such that Q is represented as

Q(dθdη) =∫

Rd

νQ(dθdv)µ∇v (dη) ,

where µ∇v is the unique probability measure in (ext G∇)v, i.e., ∇ϕ-pure phase

with mean v ∈ Rd. Similarly, for each limit point V of VN, there exists

νV ∈M+(Rd × Rd) such that V is represented as

V (dvdη) =∫

Rd

νV (dvdu)µ∇u (dη) .

Now by Lemma 6.17-(1), along some subsequence, ∇hψN,f (θ)N generates

the family of Young measures ν(θ, dv) ∈ P(Rd), i.e., it holds that

limN→∞

∫D

q(θ)G(∇hψN,f (θ)) dθ =

∫D×Rd

q(θ)G(v) ν(θ, dv)dθ

for every q ∈ L∞(D) and G ∈ C0(Rd) (cf. [77, Sect. 4.3], [9]). Then, thefollowing lemma holds.

188 T. Funaki

Lemma 6.19. If the subsequence N is commonly taken, the limits νQ andνV which have appeared in Lemma 6.18 can be represented as

νQ(dθdv) = ν(θ, dv −∇g(θ)) dθ ,

andνV (dvdu) = δv(du)

∫D

ν(θ, dv −∇g(θ)) dθ .

(d) Proof of Lemma 6.16

We are now in the position to prove Lemma 6.16. For every q ∈ C∞0 (D), since

(∇N )∗EµψN,f [V ′(∇φ(x))] = −f(x/N), we have by summation by parts

∫D

q(θ)f(θ) dθ = limN→∞

1Nd

∑x∈DN

∇Nq( x

N

)· Eµψ

N,f [V ′(∇φ(x))] ,

where (∇N )∗ ≡∑d

i=1∇N∗i denotes the dual operator of ∇N (cf. Sect. 10.2-

(b)) and V ′(∇φ(x)) = V ′(∇iφ(x))1≤i≤d ∈ Rd. Now by the definition of QN ,

Lemmas 6.18, 6.19 and the property (5.14) of the surface tension, we obtain∫

D

q(θ)f(θ) dθ =∫

D×X∇q(θ) · Eµ∇

v [V ′(∇φ(0))] νQ(dθdv)

=∫

D×Rd

∇q(θ) · (∇σ)(v +∇g(θ)) ν(θ, dv)dθ

= limN→∞

∫D

∇q(θ) · (∇σ)(∇hψN,f (θ) +∇g(θ)) dθ .

This shows (1). We similarly have

limN→∞

∫D

hψN,f (θ)f(θ) dθ = lim

N→∞

1Nd

∑x∈DN

∇N hψN,f

(x

N

)· Eµψ

N,f [V ′(∇φ(x))]

= limN→∞

1Nd

∑x∈DN

∇N hψN,f

(x

N

)· Eµψ

N,f [V ′(∇φ(x))]

− limN→∞

1Nd

∑x∈DN

∇(φ ∨ ψ)(x) · EµψN,f [V ′(∇φ(x))]

≡ S1 − S2 .

Now, since φ(x) = Ng( xN ), we see by the assumptions on V and ψ that

S2 = limN→∞

1Nd

∑x∈DN

∇Ng( x

N

)· Eµψ

N,f [V ′(∇φ(x))]

= limN→∞

∫D

∇g(θ) · (∇σ)(∇hψN,f (θ) +∇g(θ)) dθ .


Also, by Lemmas 6.18, 6.19 and the property of the surface tension σ, in asimilar way to the proof of (1) we can prove that

S1 = limN→∞

∫D

(∇hψN,f (θ) +∇g(θ)) · (∇σ)(∇hψ

N,f (θ) +∇g(θ)) dθ .

Therefore, the proof of (2) is also completed.

6.6 Critical LDP

Bolthausen and Deuschel [27] discussed the LDP for the empirical measures ofthe field distributed under the Gaussian ϕ-Gibbs measure on Z

d, when d ≥ 3.Here we summarize their results.

Let µ ∈ P(RZd

) be the Gaussian measure on RZ

d

with mean 0 and covari-ance (−∆)−1(x, y), x, y ∈ Z

d, recall Sect. 3.2. Define the empirical distrib-ution functional of ϕ-field by

RN (φ) :=1Nd

∑x∈DN

δτxφN∈ P(RZ

d

) , (6.37)

for φ = φ(x);x ∈ Zd ∈ R

Zd

, where DN ≡ ND ∩ Zd = [0, N − 1]d ∩ Z

d

with the choice of D = [0, 1)d. Note that Nd = |DN |. In (6.37), φN is theperiodic extension to Z

d of φ|DN= φ(x);x ∈ DN, the restriction of φ on

DN , and τx : RZ

d → RZ

d

denotes the shift. Note that RN (φ) is a shift invariantprobability measure. Since µ is ergodic under shifts, the LLN:

RN =⇒ µ (N →∞)

holds for µ-a.s. φ, where =⇒ denotes the weak convergence of measures onR

Zd

. The aim of [27] is to study the corresponding LDP.

First Result

The first result is at the order of Nd, i.e., the weak LDP of volume order:

µ (RN ∈ “neighborhood of ν”) N→∞

e−Ndh(ν|µ) (N →∞) . (6.38)

Here the rate functional is the specific entropy (specific free energy)

h(ν|µ) := limN→∞

1NdHN (ν|µ) ,

for shift invariant ν ∈ P(RZd

) and HN (ν|µ) =∫

log dνdµ |FDN

dν is the relativeentropy defined by restricting ν and µ to the σ-field FDN

= σφ(x);x ∈DN. The asymptotic property (6.38) is rudely stated, but it can be preciselyformulated as the LDP upper and lower bounds as usual. The “weak” LDP

190 T. Funaki

means that the upper bound is available only for compact sets. Similar resultsare known for lattice systems with bounded spins (e.g., Ising model) and forMarkov processes (Donsker-Varadhan theory).

Let us denote the class of all tempered (i.e., square integrable) and shiftinvariant ϕ-Gibbs measures by G and that of all ergodic µ ∈ G by ext G,respectively; recall that we are assuming V (η) = 1

2η2. From Theorem 9.10

below, we haveext G = µh;h ∈ R ,

where µh ∈ P(RZd

) is the Gaussian measure with mean h and covariance(−∆)−1(x, y). It is however known that ϕ-Gibbs measure has an entropiccharacterization (cf. Sheffield for general potentials):

h(ν|µ) = 0 ⇐⇒ ν ∈ G .

In particular, the LDP estimate (6.38) gives no useful information when ν isthe ϕ-Gibbs measure, since the rate functional is 0 for such ν.

Second Result

The order (speed) of the LDP for ν ∈ G is Nd−2, i.e., the LDP of capacityorder holds:

µ (RN ∈ “neighborhood of ν”) N→∞

e−Nd−2C(ν|µ) . (6.39)

The rate functional is given by

C(ν|µ) = inf

12ED(h); h ∈ L2(D) s.t. ν =

∫D

µh(θ) dθ

,

where

ED(h) = inf

12‖∇h‖2L2(Rd); h ∈ H1(Rd), h(θ) = h(θ) a.e. θ ∈ D

.

In particular, if ν = µh ∈ ext G, since h(θ) ≡ h, we have

C(µh|µ) = h2CapRd(D) ,

see (7.1) below for the definition of the capacity.

Third Result

If ν is the mixture of finitely many µh’s, then h(θ) in C(ν|µ) is a stepfunction so that the condition h(θ) ∈ H1(Rd) is never fulfilled and thereforeC(ν|µ) = +∞. This means that (6.39) is not at the correct order for such ν.Indeed, for such ν, the LDP with speed Nd−1, i.e., the LDP of surface orderholds and we have that


limN→∞

1Nd−1

HN (µhN|µ) <∞ ,

for the sequence of Gaussian fields µhN∈ P(RDN ) on DN with mean h(x/N)

and covariance (−∆)−1(x, y).

Remark 6.10. The LDP for the Gaussian ϕ-field on Zd, whose covariance

is given by the Green function of a long-range random walk, is studied by [46],[47].

7 Entropic Repulsion, Pinning and Wetting Transition

In this section we discuss the subjects of entropic repulsion and pinning. Theentropic repulsion is the problem to study, when a hard wall is settled atthe height level 0, how high the interfaces are pushed up by the random fluc-tuations naturally caused by the Lebesgue measure dφΛ in the Gibbs measure(2.4), in other words, by the entropic effects of the measure. The pinningis, on the other hand, the problem to study, under the effect of weak forceattracting interfaces to the height level 0, whether the field is really localizedor not. This is the energy effect. These two effects conflict with each otherand therefore, it is an interesting question to know which effect is dominantin the system. This leads us to the problem of the wetting transition.

We first briefly summarize the results. Let µN = µ0DN

be the finite volumeϕ-Gibbs measure onDN with 0-boundary conditions. The goal is to investigatethe asymptotic behavior as N →∞ of height variables φ under µN and underµN with wall or/and pinning effects.

(a) No wall nor pinning (cf. (3.5), Theorem 4.13):d = 2 =⇒ delocalized (|φ(x)| ≈

√logN)

d ≥ 3 =⇒ localized (φ(x) = O(1))massless (algebraic decay of two-point correlations)

(b) Wall effect only (Entropic repulsion, Theorem 7.3):d = 2 =⇒ delocalized (φ(x) ≈ logN)d ≥ 3 =⇒ delocalized (φ(x) ≈

√logN)

(c) Pinning effect only (Theorems 7.4, 7.5):d = 2 =⇒ localized and mass generation

(exponential decay of two-point correlations)d ≥ 3 =⇒ localized, mass generation

(d) Both wall and pinning effects (Theorem 7.7):d = 1, 2 =⇒ wetting transition occurs, i.e., if the strength of

pinning is strong, the field is localized; while, ifit is weak, the field is delocalized.

d ≥ 3 =⇒ no wetting transition and the field is always localized.

192 T. Funaki

7.1 Entropic Repulsion

First let us remind some notation. For Λ Zd, the finite volume ϕ-Gibbs

measure µ0Λ ∈ P(RZ

d

) is defined by (2.4) with 0-boundary condition. We shalldenote by D the class of all connected and bounded domains D in R

d withpiecewise smooth boundaries ∂D. We take D ∈ D and fix it, and simply writeµN for µ0

DNwith the choice of Λ = DN . When V (η) = 1

2c−η2, µ0

Λ and µN

are denoted by µ0,GΛ and µG

N , respectively. For Λ Zd, consider the entropic

repulsion event defined by

Ω+(Λ) = φ;φ(x) ≥ 0 for every x ∈ Λ .

This event is realized by putting a wall at height level 0 on the region Λ.We explain the results on the entropic repulsion following Deuschel and

Giacomin [76]. The conditions (V1)-(V3) in (2.2) are always assumed on thepotential V . The first result is for the probability estimate on the entropicrepulsion event when the wall is put at strictly inside of D: A D meansthat A ⊂ D and dist(A,Dc) > 0. We shall write

logd(N) =

logN, d ≥ 3 ,(logN)2, d = 2 .

See Remark 7.2 below for d = 1.

Theorem 7.1. [76] Assume A,D ∈ D and A D. Then there exist 0 <C1 ≤ C2 <∞ such that

−C2 ≤ lim infN→∞

1Nd−2 logd(N)

log µN (Ω+(AN ))

≤ lim supN→∞

1Nd−2 logd(N)

logµN (Ω+(AN )) ≤ −C1 .

Proof (Partially). We only outline the proof of the lower bound for theGaussian case: µN = µG

N (with c− = 1) when d ≥ 3. The bound can beshown with C2 = 2dG(O,O)CapD(A), where G denotes the Green functionof ∆ on Z

d and

Cap D(A) = inf

12‖∇h‖2L2(D); h ∈ C1

0 (D), h ≥ 1A

(7.1)

is the capacity; note that 12‖∇h‖2L2(D) = ΣD(h) for the Gaussian case (the

proof of the lower bound goes essentially in a similar way also for d = 2 ifG(O,O) is suitably modified). In fact, transforming the measure µG

N into µG,φN

under the map φ → φ+ φ where φ(x) =√a logNh(x/N) with h(θ) ≥ 1A(θ),

one can show the LLN for the transformed measure µG,φN :

limN→∞

µG,φN (Ω+(AN )) = 1 , (7.2)


if a > 2dG(O,O); i.e., the probability that the interfaces touch the wall be-comes negligible if they are pushed up to the sufficiently high level by addingφ. The price to adding φ should be paid by the relative entropy which behavesas

limN→∞

1Nd−2 logN

H(µG,φN |µG

N ) =a

2‖∇h‖2L2(D) . (7.3)

Applying the entropy inequality, (7.2) combined with (7.3) shows

1Nd−2 logN

logµG

N (Ω+(AN ))

µG,φN (Ω+(AN ))

≥ − 1Nd−2 logN

1

µG,φN (Ω+(AN ))

H(µG,φ

N |µGN ) +

1e

−→N→∞

−a2‖∇h‖2L2(D) .

Thus the lower bound is obtained with C2 = 2dG(O,O)CapD(A).

Remark 7.1. The problem of the entropic repulsion was posed by Lebowitzand Maes [186]. They notify that the probability of Ω+,δ(AN ) = φ;φ(x) ≥δ for every x ∈ AN, δ > 0 instead of Ω+(AN ) may be estimated as follows:

µN (Ω+,δ(AN )) ≤ µN (X ≥ δ)

≈ exp−δ2/2var (X;µN )

≤ exp

−δ2/2var (X;µG

N )≤ exp

−Cδ2Nd−2

,

where X = 1|AN |

∑x∈AN

φ(x). The second line is true at least if the ϕ-fieldhas Gaussian tail and the Brascamp-Lieb inequality proves the third line.The above calculation roughly explains the capacity order appearing in (7.4),though the logarithmic correction does not come out because of the differenceof Ω+,δ(AN ) from Ω+(AN ).

The second result is for the case where the wall is put over the wholedomain D, i.e., we take A = D, cf. (6.15).

Theorem 7.2. [76] Assume D ∈ D. Then there exist 0 < C1 ≤ C2 <∞ suchthat

−C2 ≤ lim infN→∞

1Nd−1

log µN (Ω+(DN ))

≤ lim supN→∞

1Nd−1

log µN (Ω+(DN )) ≤ −C1 .

Moreover, we have that

limL→∞

lim supN→∞

1Nd−1

∣∣logµN (Ω+(DN ))− log µN (Ω+(∂LDN ))∣∣ = 0 ,

where ∂LDN = x ∈ DN ; dist (x,DcN ) ≤ L.

194 T. Funaki

Proof (Partially). In the proof of Theorem 7.1, one can take h ≡ 1 so thatφ(x) ≡

√a logN and

limN→∞

µG,φN (Ω+(DN )) = 1

holds for a > 2dG(O,O). Then, the lower bound is shown with the speedNd−1 logN instead of Nd−1. To remove the log factor, further delicate ar-guments are required. For the non-Gaussian case, the application of theBrascamp-Lieb inequality gives the same result for a > 2dG(O,O)/c−.

Theorem 7.1 claims that the probability behaves as

µN (Ω+(AN )) e−CNd−2 logd(N) , (7.4)

i.e., the decay is essentially of capacity order except for the logarithmiccorrection, while Theorem 7.2 indicates that

µN (Ω+(DN )) e−CNd−1, (7.5)

i.e., the decay is much faster than (7.4) and it is of surface order. The behav-ior of φ(x) near the boundary of DN substantially contributes to the decayof the probability in (7.5). Indeed, since φ satisfies the boundary conditionφ(x) = 0 at x ∈ ∂+DN , φ can be negative near the boundary with highprobability and this makes µN (Ω+(DN )) smaller. If one can assume that theprobability for φ(x) getting to the positive side at each x near ∂+DN behavesas e−c and is nearly independent when x’s are apart from each other, thenwe get the order in (7.5). Once the interface comes to the positive side, theprobability to stay there is governed by (7.4) at the inside of DN , which isnegligible compared with the boundary effect (7.5).

The third result is for the estimate giving how high the interfaces arepushed up by the effect of the wall put at A = D, i.e., our object is theconditional probability

µ+N = µN (·|Ω+(DN )) .

This measure was introduced in Corollary 6.2 imposing the constant volumecondition at the same time.

Theorem 7.3. [76] (1) (Upper bound) For every C > 2dc−G∗

d ,

limN→∞

infx∈DN

µ+N (φ(x) <

√C logd(N)) = 1 ,

where G∗d = G(O,O) (when d ≥ 3), G∗

2 = limN→∞GDN(O,O)/ logN (when

d = 2), and G and GDNare the Green functions of ∆ on Z

d and DN , respec-tively.(2) (Lower bound) There exists K > 0 such that

limN→∞

infx∈D

(δ)N

µ+N (φ(x) >

√K logd(N)) = 1 ,

for every δ ∈ (0, 1), where D(δ)N = DN \ ∂δNDN .


Proof (Outline). We assume d ≥ 3. Note that the Brascamp-Lieb inequalityholds for µ+

N since the conditioning under Ω+(DN ) is equivalent to addingto the Hamiltonian the self potential term

∑x W (φ(x)), where W (r) = 0

for r ≥ 0 and +∞ for r < 0, and such W can be regarded convex (byapproximating it by a sequence of convex potentials). Therefore, we have

µ+N (φ(x) >

√C logN)

≤ exp

−

c−((√C logN − Eµ+

N [φ(x)]) ∨ 0)2

2GDN(x, x)

.

This proves the upper bound, if one can show

lim supN→∞

supx∈DN

Eµ+N [φ(x)]√C logN

< 1 . (7.6)

To see (7.6), the method of changing the measure (as in the proof of Theorem7.1), the FKG and Brascamp-Lieb inequalities are applied. The details areomitted. The lower bound is much more delicate. The lattice is divided intoeven and odd sites, and then the Markov property of the ϕ-field is effectivelyused.

As Theorem 7.3 suggests, the expectation of the height variables behavesas follows at the inside (in the macroscopic sense) of DN :

Eµ+N [φ(O)] ≈

N→∞

√logd(N) , (7.7)

assuming O ∈ D (the interior of D) for simplicity. This means that, oncethe wall is put on DN , the interfaces are pushed up to the level of orderO(

√logd(N)) inside of DN . The behavior of φ(x) was given by (3.5) when

there was no wall including the non-Gaussian case, see also [21]. Comparedwith this, we see that the wall pushes up the interfaces further at the order of√

logN for every d ≥ 2. Bolthausen et al. [28] and Daviaud [60] studied thefine behavior of maxx φ(x) for d = 2 for Gaussian case.

Remark 7.2. When d = 1, the height variables behave as |φ(x)| ≈√N under

µN , recall (3.5). As we shall see in Sect. 14, they behave as φ(x) ≈√N under

µ+N too. Namely, the wall does not change the order of the heights of the ϕ-field

in one dimension.

Much more precise results are known especially for the Gaussian ϕ-field indimensions d ≥ 3. Deuschel and Giacomin [75] proved that the distribution ofφ(x)− aN ;x ∈ Z

d under µ+N (regarding φ(x) = 0 on Dc

N ) weakly convergesas N → ∞ to the Gaussian ϕ-Gibbs measure µ ≡ µZd = N(0, G(x, y)) onZ

d, where aN := Eµ+N [φ(O)] ∼

√4G(O,O) logN . In other words, the wall

has an effect on the ϕ-field simply pushing it up by aN . In addition, the

196 T. Funaki

following precise asymptotic estimates on the probabilities µZd(Ω+(DN )) andµN (Ω+(DN )) were established by Bolthausen et al. [29] and Deuschel [74],respectively:

limN→∞

1Nd−2 logN

log µZd(Ω+(DN )) = −2G(O,O)Cap Rd(D) , (7.8)

limN→∞

1Nd−1

logµN (Ω+(DN )) = −c , (7.9)

where c is a certain constant determined from the surface tension.The entropic repulsion for the Gaussian ϕ-field on a (quenched) random

hard wall was discussed by Bertacchi and Giacomin [18], when d ≥ 3. Theentropic repulsion for two interfaces (over the wall) to lie one above the otheris discussed by Bertacchi and Giacomin [19] and Sakagawa [227]. The entropicrepulsion for interfaces between two walls is studied by Sakagawa [228]. Thisproblem was first discussed by Bricmont et al. [38].

Remark 7.3. (1) The entropic repulsion for the Gaussian field with covari-ance P (−∆)−1(x, y) is discussed by Sakagawa [226] for polynomials P inhigher dimensions (i.e., at the transient regime: d ≥ 3 when P (a) = a). Thephysical motivation comes from [145].(2) The probability estimate on the entropic repulsion event is, in general,more delicate in lower dimensions (i.e., at the recurrent regime). Sinai [236]considered the field φ = φ(x) ∈ Z;x ∈ Z+ with mean 0, covariance(−∆)−2(x, y) satisfying φ(0) = 0 in one dimension and proved that

C1N−1/4 ≤ P (φ(x) ≥ 0 for every 0 ≤ x ≤ N) ≤ C2N

−1/4 , (7.10)

where ∆ is the discrete Laplacian on Z+ determined from the Dirichlet 0-boundary condition at x = 0. Theorem 7.1 and (7.8) show the exponentialdecay of the probability when d ≥ 3, while (7.10) exhibits the decay in powerlaw which is much delicate. Note that the field φ can be constructed as φ(x) =∑x

y=0 η(y) from the simple and symmetric random walk η(y); y ∈ Z+ on Z

with time parameter y.

7.2 Pinning

Recall that µψ,WN is the finite volume ϕ-Gibbs measure (6.4) with U(θ, r) =

W (r). The square-well pinning and δ-pinning were introduced in Sect. 6.4.We first shortly summarize the known results on the pinning problem.

Dunlop et al. [93] first proved the localization of the ϕ-field underthe square-well pinning, namely the uniform boundedness in N of the ex-pected height variables Eµ0,W

N [|φ(x)|] under the ϕ-Gibbs measures µ0,WN with

0-boundary conditions or the existence of infinite volume limit of µ0,WN as

N → ∞, if the Hamiltonian contains arbitrarily weak pinning potentials Wwhen d = 2 for quadratic V . This should be compared with the case without


pinning (i.e., W ≡ 0) in which the localization occurs only when d ≥ 3 andalso compared with the case of strong pinning (or massive) potentials satisfy-ing lim|r|→∞W (r) = +∞ for which the localization occurs for all dimensions.The result of [93] is extended for general convex potential V by Deuscheland Velenik [81] later. In addition to the localization, the mass generation,namely the exponential decay of the correlations of the ϕ-field is shown byIoffe and Velenik [153] for d = 2 with δ-pinning, see also [92]. Further preciseestimates on the asymptotic behaviors of the mass and the degree of local-ization by means of the variances of the field as the pinning effect becomessmaller were established by Bolthausen and Velenik [32].

Let us state the results in more details. The first result is on the localiza-tion due to Deuschel and Velenik [81]. The ϕ-field is localized even withoutthe pinning if d ≥ 3. Therefore, the interesting case is in the two dimensionso that we take D = [−1, 1]2 and consider the ϕ-field on DN = [−N,N ]2 ∩Z

2

adding square-well or δ-pinnings. The next theorem gives an estimate on thedecay of the tail distribution of the height variables. This implies the localiza-tion as we shall see later. We write µa,b

N for µ0,WN with the square-well pinning

potential W given in (6.34) and µJN for µ0,J

N with δ-pinning, respectively. The0-boundary conditions are imposed in both measures.

Theorem 7.4. ([81], Non-Gaussian tail) (1) For every a, b > 0, there existconstants C1 = C1(s), C2 = C2(s) > 0 such that

e−C1T 2/ log T ≤ µa,bN (φ(x) ≥ T ) ≤ e−C2T 2/ log T , (7.11)

for every sufficiently large T > 0 and every N ∈ N, where the upper boundholds for every x ∈ DN while the lower bound holds only for x ∈ DN satisfyingdist (x, ∂DN ) ≥ d0T/ log T with some d0 > 0. Recall that s = 2a(eb−1) standsfor the strength of the pinning.(2) Letting a ↓ 0 and b→∞, a similar estimate holds for µJ

N with δ-pinning.

The upper bound implies, in particular, that the exponential moments of|φ(x)| are uniformly bounded in N under µa,b

N or µJN so that these measures

are tight and admit the limits µa,b and µJ , respectively, along a proper sub-sequence N ′ → ∞. It is obvious that the limits µa,b or µJ are the ϕ-Gibbsmeasures with square-well or δ-pinnings on Z

2. The fields are thus localizedin two dimension.

Remark 7.4. (1) The lower bound in (7.11) indicates that the decay of thetail distribution of φ(x) is slower than that of the Gaussian distribution; i.e.,φ(x) can take rather large values. Without pinning, if d ≥ 3, the ϕ-field existson Z

d and the Brascamp-Lieb inequality shows its Gaussian tail: µ(φ(x) ≥T ) e−CT 2

. Therefore, with pinning, the tail distributions exhibit completelydifferent behaviors for d = 2 and d ≥ 3.(2) The estimate (7.11) holds for the Gibbs measures µa,b and µJ on Z

2 bytaking the limit N →∞.

198 T. Funaki

The second result is on the mass generation due to Ioffe and Velenik[153] for d = 2 and δ-pinning. The case of d ≥ 3 is discussed by [41].

Theorem 7.5. [153] For every J ∈ R, there exist constants m = m(J), C =C(J) > 0 such that

EµJN [φ(x);φ(y)] ≤ Ce−m|x−y|, x, y ∈ Z

2 ,

i.e., the covariances under µJN have the exponential decay estimates uniformly

in N .

In the proof of Theorems 7.4 and 7.5, the following expansions of themeasures µJ

N and µa,bN play the key role.

Lemma 7.6. (1) For each φ ∈ RDN (or ∈ R

Zd

regarding φ ≡ 0 on DcN ) define

the pinned region (random region) by

A ≡ AN (φ) = x ∈ DN ;φ(x) = 0 . (7.12)

Then the measure µJN admits the expansion

µJN (·) =

∑A⊂DN

νJN (A)µ0

DN\A(·) ,

where νJN (A) = µJ

N (A = A) is the probability that the pinned region is A andµ0

DN\A(·) is the ϕ-Gibbs measure on DN \A with 0-boundary condition.

(2) The measure µa,bN admits the similar expansion:

µa,bN (·) =

∑A⊂DN

νa,bN (A)µa,b

N,Ac(·) ,

where A = x ∈ DN ; |φ(x)| ≤ a and νa,bN (A) = µa,b

N (A = A), µa,bN,Ac(·) =

µa,bN (·|A = A).

For the proof of two theorems, using Lemma 7.6, Helffer-Sjostrand rep-resentation and Brascamp-Lieb inequality are applied, but the details areomitted.

Remark 7.5. The constant m = m(J) arising in Theorem 7.5 behaves aslimJ↓−∞m(J) = 0, since the decay of the correlation is algebraic at J = −∞.[32] studied the detailed behavior of m(J) when V (η) = c

2η2. In particular, if

d = 2, limJ↓−∞EµJN [φ(x)2] = ∞ since the field is delocalized at J = −∞.

They investigated the fine behavior of the variances as well.

7.3 Wetting Transition

The problem of the wetting transition, which is studied by Pfister and Velenik[212] for the Ising model, is recently discussed for the ∇ϕ interface model as


well by several authors. The effects of the hard wall and the pinning near 0-level are introduced at the same time. The ϕ-field can take only nonnegativevalues. Recall that the field on a hard wall is delocalized for all dimensionsd (Theorem 7.3-(2)) while the pinning localizes the field (Theorem 7.4). Theformer is caused by the entropy effect and the latter is by the energy effect.

Fisher [101] proved the existence of the wetting transition, namely thequalitative change in the localization/delocalization of the field depending onwhich of these two competitive effects dominate the other, when d = 1 for theSOS type discrete model. This result is extended by Caputo and Velenik [49]for d = 2. The precise path level behavior is discussed by Isozaki and Yoshida[154] when d = 1. Bolthausen et al. [30] showed that, contrarily when d ≥ 3, notransition occurs and the field is always localized, i.e., only the phase of partialwetting appears. Bolthausen and Ioffe [31] proved the LLN in the partialwetting phase in two dimension (i.e., d = 2) under the Gibbs measures with0-boundary conditions, hard wall, δ-pinning and quadratic V conditioned thatthe macroscopic total volume of the interfaces is kept constant. They derivedthe so-called Winterbottom shape in the limit and the variational problemcharacterizing it, cf. Remark 6.6. The one dimensional case with general Vwas discussed by De Coninck et al. [61].

Let us state the results more precisely. We deal with the δ-pinning only,but the square-well pinning can be treated essentially in the same way. Thefinite volume ϕ-Gibbs measure µJ,+

N ∈ P(RZd

+ ) with hard wall and δ-pinningis defined by

µJ,+N (·) = µJ

N (·|Ω+(DN )) , (7.13)

for J ∈ [−∞,∞), where R+ = [0,∞). We take V (η) = 12η

2 and consideron the region DN = [−N,N ]d ∩ Z

d determined from D = [−1, 1]d. Recallthat µJ

N ∈ P(RZd

) is the finite volume ϕ-Gibbs measure with δ-pinning and0-boundary condition, and Ω+(DN ) = φ;φ(x) ≥ 0 for every x ∈ DN. IfJ = −∞, there is no pinning effect so that µ−∞,+

N = µ+N .

The random region A ≡ AN (φ) defined by (7.12) is regarded as the dryregion, because the heights are 0 on A so A is the region not covered bythe matter under our consideration. One can expect that A is wide if theinterface φ is localized, while it is narrow if φ is delocalized. In this sense,the localized or delocalized states are called partial wetting or completewetting, respectively.

For instance, the materials such as gasoline or oil dropped on a flat planespread over like a film. This is the state of complete wetting. On the otherhand, a small droplet of the water does not spread over on the desk. This isthe partial wetting. The place not covered by the water is the dry region.

Existence and Nonexistence of the Wetting Transition

Recent researches by Caputo and Velenik [49] and Bolthausen et al. [30] showthat when d = 1 and 2 the competition between entropy effect and energy ef-fect brings about the phase transition called wetting transition from complete

200 T. Funaki

wetting to partial wetting when the parameter J increases, while when d ≥ 3the pinning effect is always dominant and only the state of partial wettingappears for every J ∈ R.

As a natural index to judge that the system lies in the states of completeor partial wettings, let us introduce the mean density of the dry region

ρN (J) =1

|DN |EµJ,+

N [|AN |] ≥ 0 .

Observing that the limit of ρN (J) is 0 or positive, one can probe whether themost region is wet or the dry region substantially survives. Namely, we callthe complete wetting if

ρ(J) := limN→∞

ρN (J) = 0 , (7.14)

and the partial wetting if

ρ(J) := lim infN→∞

ρN (J) > 0 . (7.15)

Lemma 7.8 below shows that ρ(J) is nondecreasing in J .The result is summarized in the next theorem. Recall that the potential is

taken as V (η) = 12η

2.

Theorem 7.7. (1) [49] When d = 2, the critical value Jc ∈ R exists suchthat ρ(J) = 0 (i.e., complete wetting) for J < Jc and ρ(J) > 0 (i.e., partialwetting) for J > Jc.(2) [30] When d ≥ 3, ρ(J) > 0 (i.e., partial wetting) for every J ∈ R.

Remark 7.6. [49] The wetting transition exists for the SOS model with thepotential V (η) = |η| for all dimensions d ≥ 1.

The proof of Theorem 7.7 is omitted, instead we state two simple lemmaswhich are needed for the proof of the theorem. Let ZJ,+

N be the normalizationconstant for µJ,+

N :

ZJ,+N =

∫R

DN+

e−H0N (φ)

∏x∈DN

dφ(x) + eJδ0(dφ(x))

.

In particular, for J = −∞

Z−∞,+N ≡ Z+

N =∫

RDN+

e−H0N (φ)

∏x∈DN

dφ(x)

is the normalization constant when there is no pinning. Then we have thefollowing lemma.

Lemma 7.8. The dimension d ≥ 1 is arbitrary. We have that


d

dJlogZJ,+

N = EµJ,+N [|AN |] , (7.16)

d2

dJ2logZJ,+

N = EµJ,+N [|AN |; |AN |] . (7.17)

Moreover, the limit

τwall(J) = − limN→∞

1|DN |

logZJ,+

N

Z+N

∈ (−∞, 0] (7.18)

exists.

Proof (Outline). Use the expansion in Lemma 7.6-(1) to show (7.16) and(7.17). The existence of the limit τwall(J) is similar to that for the surfacetension, Theorem 5.1.

The constant τwall(J) is called the wall free energy (or, more precisely,the wall+pinning free energy) which is expressed as the difference

τwall(J) = − limN→∞

1|DN |

logZJ,+N + lim

N→∞

1|DN |

logZ+N ,

between the surface tensions under the hard wall with and without the pin-ning.

Lemma 7.9. The dimension d ≥ 1 is arbitrary. For every large enough J ∈ R,ρ(J) > 0 (i.e., partial wetting) holds.

Proof. The identity (7.16) shows∫ J

−∞ρN (J ′) dJ ′ =

1|DN |

logZJ,+

N

Z+N

. (7.19)

Here, the replacement of the region of the integration from RDN+ into R

DN

shows Z+N ≤ Z0

DNand this implies that

lim supN→∞

1|DN |

logZ+N ≤ lim

N→∞

1|DN |

logZ0DN

= −σ∗(0) ,

where σ∗(0) is the unnormalized surface tension at u = 0. On the otherhand, ZJ,+

N can be bounded from below by the integral with respect to∏x∈DN

eJδ0(dφ(x)) so that we have

ZJ,+N ≥ eJ|DN | .

These two bounds combined with (7.19) show that

lim infN→∞

∫ J

−∞ρN (J ′) dJ ′ ≥ J + σ∗(0) ,

for every J ∈ R. The right hand side is positive if J is sufficiently large. Thisproves the conclusion noting that ρN (J) is nondecreasing in J .

202 T. Funaki

When d ≥ 3, one can prove that τwall(J) < 0 for all J ∈ R; the proofis omitted. This physically means that the interfaces always feel the pinningeffect at macroscopic level. Then, similarly to the proof of Lemma 7.9, one candeduce ρ(J) > 0 from τwall(J) < 0 and this completes the proof of Theorem7.7-(2). Velenik [249] discussed the delocalization of interfaces above a wall ina complete wetting regime, and in an external field.

Remark 7.7. When d = 1, Isozaki and Yoshida [154] established the limittheorem at path level for the SOS type model: φ(x) ∈ Z+, which is extendedby Deuschel, Giacomin and Zambotti [78] for R+-valued case including thecritical regime; see also [30]. When d = 1, the wetting transition for pluralinterfaces of SOS type was discussed by Tanemura and Yoshida [245].

Winterbottom Shape

Bolthausen and Ioffe [26, 31] add the pinning effect to the argument for thederivation of the Wulff shape in Sect. 6.2. Assuming d = 2 and V (η) = 1

2η2,

the measure µJ,+N,v is introduced from µJ,+

N by conditioning on the macroscopicvolume of droplets as before:

µJ,+N,v = µJ,+

N

(·∣∣∣∫

D

hN (θ) dθ ≥ v

).

[31] proved that, if τwall(J) < 0, the LLN holds under µJ,+N,v and the limits are

the minimizers of the functional

ΣJD(h) = ΣD(h) + τwall(J)|θ ∈ D; h(θ) = 0|

under the conditions that h ≥ 0 and∫

Dh dθ = v. The second term represents

that the interfacial energy is smaller if the dry region (i.e., the region on whichh(θ) = 0) is wider; i.e., if τwall(J) < 0, it is more favorable for the interfaceto stay on the wall because of the pinning effect. The minimizer of ΣJ

D(h) iscalled the Winterbottom shape, which is unique except the translation.

Remark 7.8. (1) De Coninck et al. [61] discussed the above problem forgeneral potential V when d = 1, i.e., conditioning µ+

N or µJ,+N as

∑N−1x=1 φ(x) =

N2v (constant macroscopic volume condition for the interfaces), they derivedthe Wulff shape or the Winterbottom shape, respectively. They further obtainedthe Young’s relation for the angle of the interface to the wall at the point ittouches the wall.(2) De Coninck et al. [62] discussed the generalization of Young’s law for theSOS type interfaces on a substrate which is heterogeneous, rough and realizedas another SOS interface.

Remark 7.9. (Ising model) The wetting transition and the derivation of theWinterbottom shape from the two dimensional Ising model were studied byPfister and Velenik [212]. They impose the + spins on the upper half of the


boundary of the cube with size N and − spins on the lower half. Moreover,at the lower segment of the boundary of the cube, magnetic field is added andthis gives the pinning effect. Changing the strength of the magnetic field, thephase separation curve prefers to stay on the lower boundary of the cube. Threeand higher dimensional Ising model was investigated by Bodineau, Ioffe andVelenik [23].

Remark 7.10. We refer to [187, 188, 192, 193] by Lipowsky et al. for physicalmotivations to the problem of the wetting transition.

8 Central Limit Theorem

The long correlations of the ϕ-field (cf. Theorem 4.13) make the proof of limittheorems like the LDP, the CLT and others nontrivial. This section discussesthe central limit theorem (CLT) for an infinite system.

We first assume d ≥ 3 and consider the ϕ-Gibbs measure µ on Zd con-

structed by Theorem 4.16. If µ has a very nice mixing property, it is easy toshow the CLT under the usual scaling for the fluctuation of the ϕ-field:

ΦN = N−d/2∑

x∈DN

φ(x)−Eµ[φ(x)] =⇒N→∞

N(0,m) ,

for some m > 0, where D is a bounded domain of Rd and note that Nd/2 ≈√

|DN |. We have seen this in Proposition 3.12 for massive Gaussian systems.However, Theorem 4.13 actually implies that

Eµ[ΦN2] ≈∑

|x|≤N

|x|2−d ≈ N2 −→N→∞

∞ .

Therefore, ΦN does not give the right scaling and, as we did in (3.17) or in(3.18), we should scale-down it and consider

ΦN = N−1ΦN = N−d/2∑

x∈DN

N−1 φ(x)− Eµ[φ(x)] (8.1)

or the random signed measures

ΦN (dθ) = N−d/2∑x∈Zd

N−1 φ(x)− Eµ[φ(x)] δx/N (dθ) , (8.2)

for θ ∈ Rd. Since (8.2) is the usual CLT scaling for the ∇ϕ-field (recall Sect.

3.4), it is natural to introduce the fluctuation fields ΨNi , 1 ≤ i ≤ d for ∇φ =

∇φ(b); b ∈ (Zd)∗ as

ΨNi (dθ) = N−d/2

∑x∈Zd

∇iφ(x)− ui δx/N (dθ) , (8.3)

204 T. Funaki

for θ ∈ Rd, where ui = E[∇iφ(x)]; recall (3.19) for the Gaussian case. In fact,

Naddaf and Spencer [202] studied ΨNi under the ergodic ∇ϕ-Gibbs measure

µ∇u with mean u = (ui)d

i=1 ∈ Rd, i.e. µ∇

u ∈ (ext G∇)u, and established theCLT. The lattice dimension d ≥ 1 is arbitrary, since the ∇ϕ-field is dealt with.The result is later extended to the dynamic level by Giacomin et al. [135],which actually concludes the static result of Naddaf and Spencer, see Sect.11.

In (8.1) or in (8.2), φ(x) is divided by N . This may be explained in thefollowing manner: Our real object is the ∇ϕ-field and, from this point of view,φ(x) is expressed as the sum of∇φ(b)’s along a path connecting O and x. Sincethe typical length of the path is N , it is natural for the ϕ-field to be dividedby N .

We now state the CLT result. We write 〈Ψ, f〉 =∫

Rd f dΨ or 〈f, g〉 =∫Rd fg dθ.

Theorem 8.1. [202] There exists a positive definite d × d matrix q =(qij(u))1≤i,j≤d such that

limN→∞

Eµ∇u

[e√−1〈ΨN

i ,f〉]

= exp−1

2

⟨∂f

∂θi, A

∂f

∂θi

⟩

= exp

12

∫Rd

k2i

k · qk |f(k)|2 dk

holds for every f = f(θ) ∈ C∞0 (Rd). Here, A is a positive definite integral op-

erator determined by A−1 = −∑d

i,j=1 qij∂2

∂θi∂θj, f(k) is the Fourier transform

of f and k · qk =∑

ij qijkikj. The concrete form of the matrix q will be givenin Theorem 11.1, (11.1).

We outline the proof of Theorem 8.1. The potential V is always supposedto satisfy the conditions (V1)–(V3) in (2.2). The basic idea is the usage of theHelffer-Sjostrand representation on Z

d.Consider the differential operators ∂x, ∂

∗x and L defined by

∂x =∂

∂φ(x), ∂∗x = −∂x +

∑y∈Zd:|x−y|=1

V ′(φ(x)− φ(y)) ,

L = −∑x∈Zd

∂∗x∂x ,

acting on the functions F = F (φ) of φ = φ(x);x ∈ Zd, recall Sect. 4.1. ∂∗x is

the dual operator of ∂x (with respect to the ϕ-Gibbs measure µ at least whend ≥ 3) and L is the generator of the SDEs (2.13) on Z

d, i.e., the operator(2.15) with Γ = Z

d. We further introduce the operator

L = L+Q

acting on F = F (x, φ), where Q is defined by


QF (x, φ) = −d∑

i=1

(∇∗i V

′′(∇iφ(·))∇i)F (x, φ) .

Recall that

∇ig(x) = g(x+ ei)− g(x), ∇∗i g(x) = g(x− ei)− g(x) .

In Sect. 4.1, Q is denoted by QφZd . Assuming u = 0 for simplicity, set

GN (t) = Eµ∇0

[et〈ΨN

i ,f〉].

Then, taking

vN (x, φ) = (−L)−1∇∗i f

N (x), fN (x) = N1− d2 f(x/N) ,

we have for every x ∈ Zd

∂x

∑

y∈Zd

∂∗yvN

(x, φ) = −LvN (x, φ) = ∇∗

i fN (x) = N∂x〈ΦN ,∇N∗

i f〉 ,

where ∇N∗i f(θ) = N

(f(θ − ei/N) − f(θ)

). We have used [∂x, ∂

∗y ] = ∂x∂yH

(H is a formal Hamiltonian on Zd) and ∂xv

N (y, φ) = ∂yvN (x, φ) for the first

equality. These identities imply

N−1∑y∈Zd

∂∗yvN = 〈ΦN ,∇N∗

i f〉(

= 〈ΨNi , f〉

).

We accordingly obtain

d

dtGN (t) = Eµ∇

0

[〈ΦN ,∇N∗

i f〉et〈ΨNi ,f〉

]

= Eµ∇0

∑

x∈Zd

N−1∂∗xvNet〈ΨN

i ,f〉

=∑x∈Zd

Eµ∇0

[N−1vN∂xe

t〈ΨNi ,f〉

]

=∑x∈Zd

Eµ∇0

[N−2vN t∇∗

i fN (x)et〈ΨN

i ,f〉]

= tEµ∇0

∑

x∈Zd

N−2vN∇∗i f

N (x)−Af

et〈ΨN

i ,f〉

+ tAfG

N (t) ,

where Af is an arbitrary constant. The next proposition is essentially a ho-mogenization result for the random walk in random environment:

206 T. Funaki

Proposition 8.2. Take Af = 〈∂f/∂θi, A∂f/∂θi〉. Then we have

limN→∞

Eµ∇0

[(∇∗

i fN , (−N2L)−1(∇∗

i fN ))−Af

2]

= 0 ,

where ( , ) means the inner product on Zd.

Once this is shown, the first term in the last line of the above equalitiesvanishes as N →∞. Thus we have

limN→∞

d

dtlogGN (t) = tAf ,

which concludes the proof of Theorem 8.1. Remark 8.1. In one dimension, ∇φ(x);x ∈ Z form i.i.d. for general(nonconvex) potential. Therefore, the CLT is obvious and q(u) coincides withthe variance of νu,see Remark 4.5.

9 Characterization of ∇ϕ-Gibbs Measures

In Sect. 4.4, for each average tilt u ∈ Rd, an infinite volume∇ϕ-Gibbs measure

µ∇u on (Zd)∗, which is tempered (i.e., square integrable), shift invariant and

ergodic under shifts, in other words, ∇ϕ-pure phase was constructed, seeTheorem 4.15 and recall Definitions 2.2 and 2.3 for the notion of the∇ϕ-Gibbsmeasures, shift invariance and ergodicity. The tightness argument based onthe Brascamp-Lieb inequality was applied.

This section addresses the uniqueness problem for the ∇ϕ-pure phase µ∇u

for each u ∈ Rd. The well-known Dobrushin’s uniqueness argument [83, 84]

does not work here, since if it works the correlation functions of the fieldsmust decay exponentially fast, [179]. But this can not happen as we have seenalready. We shall solve the problem based on the dynamic coupling, i.e.,by characterizing all equilibrium (stationary) measures for the dynamics ofgradient fields ∇φt associated with those of heights fields φt determined bythe SDEs (2.13) by means of the relation (2.6):

ηt(b) ≡ ∇φt(b) = φt(xb)− φt(yb) ,

where b = 〈xb, yb〉 ∈ (Zd)∗ are directed bonds in Zd. Considering ∇φt is

natural in the sense that the dynamics (2.13) for the ϕ-field on Zd is invariant

under the uniform translation φ(x) → φ(x) + h.Since ∇ϕ-Gibbs measures are reversible and therefore stationary for the

stochastic processes ηt = ηt(b); b ∈ (Zd)∗ under the subsidiary assumptionsof shift invariance and temperedness (see Proposition 9.4), the study of sta-tionary measures for ηt yields an information for the ∇ϕ-Gibbs measures, seeDefinition 9.1 below for stationarity and reversibility. Our result can roughlybe stated as follows: Under the conditions (V1)-(V3) in (2.2) on the potential


V , for each u ∈ Rd there exists a unique tempered, shift invariant, ergodic

under shifts and stationary probability measure µ∇u for ηt with mean u (aver-

age tilt), cf. Theorem 9.3. Especially, there exists a unique ∇ϕ-Gibbs measurewhich is tempered, shift invariant, ergodic under shifts and has mean u, seeCorollary 9.6. This will play an important role in establishing the hydrody-namic limit later, and has been already applied to show several properties ofthe surface tension σ = σ(u), see Sect. 5.4, and also to prove the LDP, seeSect.6.5.

9.1 ϕ-Dynamics on Zd and ∇ϕ-Dynamics on (Zd)∗

According to (2.13) the dynamics of the height variables φt = φt(x) ∈ RZ

d

is governed by the SDEs

dφt(x) = −∑

b:xb=x

V ′(∇φt(b)) dt+√

2dwt(x), x ∈ Zd , (9.1)

where wt(x);x ∈ Zd is a family of independent one dimensional standard

Brownian motions. The potential V is always assumed to satisfy the conditions(V1)–(V3). The dynamics for height differences ηt = ηt(b) ∈ R

(Zd)∗ is thendetermined by the SDEs

dηt(b) = −

∑b:xb=xb

V ′(ηt(b))−∑

b:xb=yb

V ′(ηt(b))

dt+

√2dwt(b) , (9.2)

for b ∈ (Zd)∗, where wt(b) = wt(xb)−wt(yb). Indeed, writing down the SDEs(9.1) for φt(xb) and φt(yb) and then taking their difference, (9.2) are readilyobtained. Since ηt fulfills the loop condition, the state space of the process ηt

is X which has been introduced in Sect. 2.3.The relationship between the solutions of (9.1) and (9.2) is summarized

in the next lemma. Recall that the height differences ηφ are associated withthe heights φ by (2.6) and, conversely, the heights φη,φ(O) can be constructedfrom height differences η and the height variable φ(O) at x = O by (2.7). Wealways assume η0 ∈ X for the initial data of (9.2).

Lemma 9.1. (1) The solution of (9.2) satisfies ηt ∈ X for all t > 0.(2) If φt is a solution of (9.1), then ηt := ηφt is a solution of (9.2).(3) Conversely, let ηt be a solution of (9.2) and define φt(O) through (9.1) forx = O and ∇φt(b) replaced by ηt(b) with arbitrary initial condition φ0(O) ∈ R.Then φt := φηt,φt(O) is a solution of (9.1).

To discuss the existence and uniqueness of solutions to (9.2), the space Xis rather big and therefore we introduce weighted 2-spaces on (Zd)∗ as wehave done for ϕ-field in Sect. 3.2:

208 T. Funaki

2r,∗ ≡ 2r((Zd)∗) :=

η ∈ R

(Zd)∗ ; |η|2r :=∑

b∈(Zd)∗

η(b)2e−2r|xb| <∞

,

for r > 0. The increasing order of η is controlled exponentially in the space2r,∗. We denote Xr = X∩2r,∗ equipped with the norm | · |r. Then, the condition(V.3) on V implies global Lipschitz continuity in Xr, r > 0, of the drift termof the SDEs (9.2). Therefore, a standard method of successive approximationsyields the following lemma.

Lemma 9.2. For each η ∈ Xr, r > 0, the SDEs (9.2) have a unique Xr-valuedcontinuous solution ηt starting at η0 = η.

We are now in the position to precisely define the stationarity and the re-versibility of probability measures under the dynamics ηt. Let C∞

loc,b(X ) denotethe family of all (tame) functions F on X of the form F (η) = F (η(b); b ∈ Λ∗)for some Λ Z

d and F ∈ C∞b (RΛ∗

).

Definition 9.1. We say that µ∇ ∈ P(X ) is stationary under ηt ifEµ∇

[F (η0)] = Eµ∇[F (ηt)] for all t ≥ 0 and F ∈ C∞

loc,b(X ), where Eµ∇[ · ]

means the expectation for ηt with initial distribution µ∇. We say that µ∇

is reversible if Eµ∇[F (η0)G(ηt)] = Eµ∇

[F (ηt)G(η0)] for all t ≥ 0 andF,G ∈ C∞

loc,b(X ).

The reversibility implies the stationarity.

9.2 Stationary Measures and ∇ϕ-Gibbs Measures

Let us formulate the results precisely. We shall consider the following classesof probability measures on X :

P2(X ) = µ∇ ∈ P(X ); Eµ∇[η(b)2] <∞ for every b ∈ (Zd)∗,

S = µ∇ ∈ P2(X ); shift invariant and stationary under ηt,ext S = µ∇ ∈ S; ergodic under shifts,(ext S)u = µ∇ ∈ ext S; µ∇ has mean u, u = (ui)d

i=1 ∈ Rd .

Recall Definition 2.3 for the shift invariance and the ergodicity. The last con-dition “µ∇ has mean u” means Eµ∇

[η(ei)] = ui for every 1 ≤ i ≤ d, whereei ∈ Z

d denotes the i-th unit vector and the bond 〈ei, O〉 is also denoted byei. The measures µ∇ ∈ P2(X ) are called tempered. The set P(Xr), r > 0, isdefined correspondingly and P2(Xr) stands for the set of all µ ∈ P(Xr) suchthat Eµ[|η|2r] <∞. Note that S ⊂ P2(Xr) for every r > 0.

With these notation, the uniqueness of stationary measures under the ∇ϕ-dynamics is formulated in the next theorem.

Theorem 9.3. [124] (uniqueness of stationary measures for ηt) For everyu ∈ R

d there exists at most one µ ∈ (extS)u.


The proof of this theorem is given based on a coupling argument; namely,assuming that there exist two different measures µ∇, µ∇ ∈ (ext S)u, we con-struct two solutions φt and φt of the SDEs (2.13) with common Brownianmotions wt = wt(x);x ∈ Z

d in such a way that the gradients ∇φ0 and ∇φ0

of their initial data are distributed under µ∇ and µ∇, respectively. Then, com-puting the time derivative of

∑x∈Λ E[(φt(x)− φt(x))2] for each Λ Z

d, onecan finally conclude µ∇ = µ∇ by letting Λ Z

d and noting the ergodicity ofµ∇ and µ∇. The ergodicity helps to deal with the boundary terms. The strictconvexity of V plays an essential role. The details will be discussed in Sect.9.3.

We now consider the family of the ∇ϕ-Gibbs measures on (Zd)∗, recallDefinition 2.2:

G∇ = µ∇ ∈ P2(X ); shift invariant ∇ϕ-Gibbs measures .

The classes ext G∇ and (ext G∇)u are similarly defined, so that µ∇ ∈ ext G∇ isergodic under shifts and µ∇ ∈ (ext G∇)u has mean u. Note that, if µ ∈ P2(X )is shift invariant, then µ ∈ P2(Xr) for all r > 0. Since the finite volume ∇ϕ-Gibbs measure µ∇

Λ,ξ is reversible under the evolution governed by the finitedimensional SDEs for ∇ϕ-field on Λ∗ with boundary condition ξ, one canshow the following proposition by letting Λ Z

d, see Sect. 9.4

Proposition 9.4. Every µ∇ ∈ G∇ is reversible under the dynamics ηt definedby the SDEs (9.2). In particular, we have G∇ ⊂ S.

Theorem 9.3 and Proposition 9.4 imply the uniqueness of the tempered,shift invariant and ergodic ∇ϕ-Gibbs measures for each mean tilt u:

Theorem 9.5. [124] (uniqueness of ∇ϕ-Gibbs measures) For every u ∈ Rd

there exists at most one µ∇ ∈ (extG∇)u.

Proof. By Proposition 9.4, µ∇ ∈ extG∇ implies µ∇ ∈ extS. Consequentlythe conclusion follows from Theorem 9.3.

Combining this with Theorem 4.15 (existence of ∇ϕ-Gibbs measures) wehave the following corollary.

Corollary 9.6. [124] (characterization of all tempered and shift invariant∇ϕ-Gibbs measures) For every u ∈ R

d, (extG∇)u = µ∇u . In particular, the

family G∇ is the convex hull of µ∇u ; u ∈ R

d.Remark 9.1. As we saw in Sect. 3.2, the Gaussian measure µψ = N(ψ,G)is a ϕ-Gibbs measure for quadratic potentials if d ≥ 3 and ψ is harmonic,where G is the Green function. In particular, if φ = φ(x);x ∈ Z

d is µψ-distributed and ∇ψ is not shift invariant, the distribution of ∇φ is a ∇ϕ-Gibbsmeasure which is not shift invariant. The characterization of unshift invariant∇ϕ-Gibbs measures for general convex potentials V is not known.

210 T. Funaki

9.3 Proof of Theorem 9.3

Energy Inequality

We first prepare an energy inequality for φt. After computing the time deriv-ative, the proof is essentially due to the rearrangement of the sum, which is adiscrete analogue of Green-Stokes’ formula.

Lemma 9.7. Let φt and φt be two solutions of (9.1) and set φt(x) := φt(x)−φt(x). Then, for every Λ Z

d, we have

∂

∂t

∑x∈Λ

(φt(x)

)2

= IΛt +BΛ

t , (9.3)

where

IΛt = −

∑b∈Λ∗

∇φt(b)V ′(∇φt(b))− V ′(∇φt(b))

,

BΛt = 2

∑b∈∂Λ∗

φt(yb)V ′(∇φt(b))− V ′(∇φt(b))

,

and ∂Λ∗ = b ∈ (Zd)∗;xb /∈ Λ, yb ∈ Λ. The interior term IΛt and the boundary

term BΛt admit the following bounds, respectively,

IΛt ≤ −c−

∑b∈Λ∗

(∇φt(b)

)2

, (9.4)

BΛt ≤ 2c+

∑b∈∂Λ∗

|φt(yb)| |∇φt(b)| . (9.5)

Proof. From the equation (9.1),

∂

∂t

(φt(x)

)2

= −2∑

b:xb=x

Φt(b) φt(x) = −

∑b:xb=x

Φt(b)−∑

b:yb=x

Φt(b)

φt(x) ,

whereΦt(b) := V ′(∇φt(b))− V ′(∇φt(b)) .

The second equality uses the symmetry of V which implies V ′(∇φ(b)) =−V ′(∇φ(−b)). The right hand side summed over x ∈ Λ becomes

−∑x∈Λ

φt(x)∑

b:xb=x

Φt(b) +∑x∈Λ

φt(x)∑

b:yb=x

Φt(b)

= −∑b∈Λ∗

∇φt(b)Φt(b)−∑

b:xb∈Λ,yb /∈Λ

φt(xb)Φt(b) +∑

b:yb∈Λ,xb /∈Λ

φt(yb)Φt(b)

= IΛt +BΛ

t ,

which proves (9.3). To obtain the term BΛt we again used the symmetry of V .

The two bounds (9.4) and (9.5) follow from the condition (V.3) on V .


Dynamic Coupling

The proof of Theorem 9.3 is reduced to a proposition which also implies theLipschitz continuity of the derivative of the surface tension σ(u), see Theorem5.3 above. Suppose that there exist µ∇ ∈ (extS)u and µ∇ ∈ (extS)v for u, v ∈R

d. Let us construct two independent Xr-valued random variables η = η(b)and η = η(b) on a common probability space (Ω,F , P ) in such a mannerthat η and η are distributed by µ∇ and µ∇ under P , respectively. We defineφ0 = φη,0 and φ0 = φη,0 using the notation in (2.7). Let φt and φt be the twosolutions of the SDEs (9.1) with common Brownian motions having initialdata φ0 and φ0. In view of Lemmas 9.1 and 9.2 such solutions certainly exist.Since µ∇, µ∇ ∈ S, we conclude that ηt := ηφt and ηt := ηφt are distributedby µ∇ and µ∇, respectively, for all t ≥ 0. Our claim is then the following.

Proposition 9.8. There exists a constant C > 0 independent of u, v ∈ Rd

such that

lim supT→∞

1T

∫ T

0

d∑i=1

EP [(ηt(ei)− ηt(ei))2] dt ≤ C|u− v|2 . (9.6)

Once this proposition is proved, Theorem 9.3 immediately follows. Indeed,suppose that there exist two measures µ∇, µ∇ ∈ (extS)u. Then Proposition9.8 with u = v implies

limT→∞

∫|η − η|2rPT (dηdη) = 0 , (9.7)

where PT is a shift invariant probability measure on Xr × Xr, r > 0, definedby

PT (dηdη) :=1T

∫ T

0

P(ηt(b), ηt(b); b ∈ (Zd)∗ ∈ dηdη

)dt .

The first marginal of PT is µ∇ and the second one is µ∇. Thus (9.7) impliesthat the Vaserstein distance between µ∇ and µ∇ vanishes and hence µ∇ = µ∇,see, e.g., [109], p.482 for the Vaserstein metric on the space P2(Xr). Thisconcludes the proof of Theorem 9.3.

Proof of Proposition 9.8

Step 1. We apply Lemma 9.7 to the differences φt(x) := φt(x)− φt(x) andobtain, with the choice Λ = Λ,

EP

[ ∑x∈Λ

(φT (x)

)2]

+ c−

∫ T

0

EP

∑

b∈Λ∗

(∇φt(b)

)2

dt

≤ EP

[ ∑x∈Λ

(φ0(x)

)2]

+ 2c+∫ T

0

EP

∑

b∈∂Λ∗

|φt(yb)| |∇φt(b)|

dt (9.8)

212 T. Funaki

for every T > 0 and ∈ N. Set

g(t) =d∑

i=1

EP

[(∇φt(ei)

)2].

Then, noting that the distribution of (ηt, ηt) = (∇φt,∇φt) on Xr × Xr isshift invariant, the second term on the left hand side of (9.8) coincideswith c−d

−1|Λ∗ |

∫ T

0g(t) dt. On the other hand, estimating |φt(yb)| |∇φt(b)| ≤

γ|∇φt(b)|2 +−1γ−1|φt(yb)|2/2 for arbitrary γ > 0, the second term on theright hand side is bounded by

c+γd−1|∂Λ∗

|∫ T

0

g(t) dt+ c+−1γ−1|∂Λ∗

|∫ T

0

supy∈∂Λ

‖φt(y)‖2L2(P ) dt .

Then, choosing γ = c−/2c+c0 with c0 := sup≥1|∂Λ∗ |/|Λ∗

| <∞, we obtainfrom (9.8)

∫ T

0

g(t) dt ≤ 2dc−|Λ∗

|EP

[ ∑x∈Λ

(φ0(x)

)2]

+(2c+c0)2d

(c−)2

∫ T

0

supy∈∂Λ

‖φt(y)‖2L2(P ) dt , (9.9)

where we have dropped the nonnegative first term on the left hand side of(9.8).Step 2. Here we derive the following bound on the boundary term: For eachε > 0 there exists an 0 ∈ N such that

supy∈∂Λ

‖φt(y)‖2L2(P ) ≤ C1

(ε22 + 2|u− v|2 + −2t

∫ t

0

g(s) ds)

(9.10)

for every t > 0 and ≥ 0, where C1 > 0 is a constant independent of ε, , andt. To this end, as an immediate consequence of the mean ergodic theorem (cf.[24, 175]) applied to µ∇ ∈ (ext S)u, we have

lim|x|→∞

1|x| ‖φ

η,0(x)− x · u‖L2(µ∇) = 0 (9.11)

and correspondingly for µ∇ with v in place of u. Taking Λ′ = Λ[/2] one obtains

‖φt(y)‖L2(P ) ≤‖φt(y)−1|Λ′|

∑x∈Λ′

φt(x)− y · u‖L2(P )

+ ‖φt(y)−1|Λ′|

∑x∈Λ′

φt(x)− y · v‖L2(P )

+ ‖ 1|Λ′|

∑x∈Λ′

φt(x)‖L2(P ) +√d|u− v|


=:I1 + I2 + I3 + I4 ,

for y ∈ ∂Λ. However, since∑

x∈Λ′ x = 0 and using (9.11),

I1 ≤1|Λ′|

∑x∈Λ′

‖φt(y)− φt(x)− (y − x) · u‖L2(P )

=1|Λ′|

∑x∈Λ′

‖φη,0(y − x)− (y − x) · u‖L2(µ∇) ≤ ε

provided is sufficiently large; recall that ∇φt is distributed by µ∇ for allt ≥ 0. Similarly, I2 ≤ ε for sufficiently large . Finally, since as in the proofof Lemma 9.7

∂

∂t

∑x∈Λ′

φt(x)

= −

∑x∈Λ′

∑b:xb=x

Φt(b) =∑

b∈(∂Λ′)∗

Φt(b) ,

I3 is bounded as

I3 ≤∥∥∥∥∥

1|Λ′|

∑x∈Λ′

φ0(x)

∥∥∥∥∥L2(P )

+∫ t

0

1|Λ′|

∑b∈(∂Λ′)∗

‖Φs(b)‖L2(P ) ds .

The right hand side can be further estimated as

∑b∈(∂Λ′)∗

‖Φs(b)‖L2(P ) ≤ c+d−1|(∂Λ′)∗|

d∑i=1

‖∇φs(ei)‖L2(P )

and, using again (9.11),∥∥∥∥∥

1|Λ′|

∑x∈Λ′

φ0(x)

∥∥∥∥∥L2(P )

≤ 1|Λ′|

∑x∈Λ′

‖φη,0(x)− x · u‖L2(µ∇)

+ ‖φη,0(x)− x · v‖L2(µ∇) + |x| · |u− v|

≤ ε+√d|u− v| ,

for sufficiently large . Therefore,

I3 ≤ ε+√d|u− v|+ c+d

−1|Λ′|−1|(∂Λ′)∗|∫ t

0

d∑i=1

‖∇φs(ei)‖L2(P ) ds

for sufficiently large . This completes the proof of (9.10).Step 3. Using (9.11), one can choose 1 ∈ N such that

1|Λ|

∑x∈Λ

EP

[(φ0(x)

)2]

214 T. Funaki

≤ 3|Λ|

∑x∈Λ

‖φη,0(x)− x · u‖2L2(µ∇)

+|x · u− x · v|2 + ‖φη,0(x)− x · v‖2L2(µ∇)

≤ ε22 + 3d2|u− v|2, (9.12)

for every ≥ 1. Inserting the estimates (9.10) and (9.12) into (9.9), we have∫ T

0

g(t) dt ≤ C2(ε22 + 2|u− v|2)

+ C2−2

∫ T

0

(ε22 + 2|u− v|2 + −2t

∫ t

0

g(s) ds)dt

≤ C2(ε2 + |u− v|2)(2 + T ) + C2−4T 2

∫ T

0

g(t) dt

for every T > 0 and ≥ 2 := max0, 1, which may depend on u, v andε > 0. C2 is a constant independent of u, v and ε. Choosing = (2C2T

2)1/4

and letting T →∞, we obtain

lim supT→∞

1T

∫ T

0

g(t) dt ≤ 2C2(√

2C2 + 1)(ε2 + |u− v|2)

for every ε > 0. Finally, letting ε→ 0, the desired estimate (9.6) is shown.

9.4 Proof of Proposition 9.4

We establish the reversibility of the ∇ϕ-Gibbs measures under the dynamics(9.2). To this end, we need the approximation of the solutions of (9.2) by thecorresponding finite volume equations, cf. [89, 231, 257] for related results.For every ξ ∈ X and Λ Z

d, let us consider the SDEs

dηt(b) = −

∑b:xb=xb∈Λ

V ′(ηt(b))−∑

b:xb=yb∈Λ

V ′(ηt(b))

dt

+√

2dwΛt (b), b ∈ Λ∗,

ηt(b) = ξ(b), b /∈ Λ∗ ,

η0(b) = ξ(b), b ∈ (Zd)∗,

(9.13)

where wΛt (b) = 1xb∈Λwt(xb)− 1yb∈Λwt(yb). The distribution on the space

C([0, T ],X ) of the solution ηt ≡ ηΛt is denoted by PΛ

ξ . The distribution of thesolution of the SDEs (9.2) starting at ξ is denoted by Pξ. Then the next lemmais standard by showing the tightness of PΛ

ξ Λ and the unique characterizationof Pξ in terms of the martingale problem, see Proposition 2.2 of [124]. RecallSect. 9.1 for the space C∞

loc,b(X ).


Lemma 9.9. For every ξ ∈ Xr and F ∈ C∞loc,b(X ),

limΛZd

EP Λξ [F (ηt)] = EPξ [F (ηt)] .

We are now in the position to complete the proof of Proposition 9.4. Tothis end, it suffices to show that every µ∇ ∈ G∇ satisfies∫

Xr

F (ξ)EPξ [G(ηt)]µ∇(dξ) =∫Xr

EPξ [F (ηt)]G(ξ)µ∇(dξ) (9.14)

for every t ≥ 0 and F,G ∈ C∞loc,b(X ), cf. Definition 9.1. However, for every

ψ ∈ RZ

d

and Λ Zd, if we consider the SDEs for φt ∈ R

Zd

:

dφt(x) = −∑

b:xb=x

V ′(∇φt(b)) dt+√

2dwt(x), x ∈ Λ ,

φt(x) = ψ(x), x /∈ Λ ,

φ0(x) = ψ(x), x ∈ Zd ,

(9.15)

then the finite volume ϕ-Gibbs measure µψΛ is clearly reversible under (9.15);

recall (4.5) taking ρ = 0. Therefore, since ηt = ∇φt satisfies (9.13) providedψ = φξ,0, µ∇

Λ,ξ is reversible under (9.13), i.e.,∫X

Λ∗,ξ

F (η)EP Λη∨ξ [G(ηt)]µ∇

Λ,ξ(dη) =∫X

Λ∗,ξ

EP Λη∨ξ [F (ηt)]G(η)µ∇

Λ,ξ(dη)

(9.16)for all ξ ∈ X if both F and G are supported in Λ. For given µ∇ ∈ G∇,integrating both sides of (9.16) with respect to µ∇(dξ) we have by the DLRequation∫

Xr

F (ξ)EP Λξ [G(ηt)]µ∇(dξ) =

∫Xr

EP Λξ [F (ηt)]G(ξ)µ∇(dξ) .

Hence, (9.14) follows from Lemma 9.9 by letting Λ Zd.

Remark 9.2. (1) Results similar to Proposition 9.4 together with its conversewere obtained for lattice systems by [89, 231, 258, 225] and for continuum sys-tems by [156, 109].(2) The dynamic approach might work also to construct the ∇ϕ-Gibbs mea-sures, see [109], Proposition 6.2 for the massive continuum field.

Remark 9.3. Sheffield [230] gives a different proof for Theorem 9.5 in moregeneral setting based on the argument called “cluster swapping”.

Remark 9.4. Gawedzki and Kupiainen [127] considered the ∇ϕ interfacemodel with V (η) = 1

2η2+λη4 and proved that, applying the Wilson-Kadanoff’s

renormalization group repeatedly, the limit becomes the massless Gaussianfield with the potential 1

2c(λ)η2 with a proper positive constant c(λ) determineddepending on λ.

216 T. Funaki

Remark 9.5. The nonuniqueness of µ∇u for nonconvex potential V is un-

known. The ground states corresponding to such potential are analyzed by[201] for continuum field.

9.5 Uniqueness of ϕ-Gibbs Measures

The existence of shift invariant and ergodic ϕ-Gibbs measures is not fullyestablished in general (cf. Theorem 4.16 and Remark 4.6) except the Gaussiancase (cf. Sect. 6.6). However, the uniqueness can be shown from Theorem 9.5.

Theorem 9.10. For every h ∈ R, the square integrable, shift invariant andergodic (under the shifts) ϕ-Gibbs measure µ with mean h (i.e., Eµ[φ(x)] = hfor every x ∈ Z

d) is unique; recall Definition 2.3 for shift invariance andergodicity of ϕ-fields.

Proof. Let φ = φ(x);x ∈ Zd be µ-distributed. For each x ∈ Λ, φ(x) is

represented as

φ(x) =1

|∂Λ|∑

y∈∂Λ

φ(y) +1

|∂Λ|∑

y∈∂Λ

∑b∈Cy,x

∇φ(b) ,

where Cy,x are chains connecting y and x. However, letting → ∞, the firstterm in the right hand side converges to h in L1(µ) by the ergodicity of µ.Therefore, from Theorem 9.5, we see that every finite dimensional distributionof φ under µ is uniquely determined.

For the infinite volume dynamics φt = φt(x);x ∈ Zd and ηt = ηt(b); b ∈

(Zd)∗, the convergence rate to the equilibrium or the algebraic decay ofcorrelations:

cov (F (φ0), G(φt)) ∼ ct−(d−2)/2, cov (F (η0), G(ηt)) ∼ ct−d/2

as t → ∞ under equilibrium are proved by [64] for d ≥ 3 and d ≥ 2, respec-tively; compare this with the static results in Sect. 4.3.

10 Hydrodynamic Limit

We now entirely move toward the investigation on the dynamics for theheights. The random time evolution of microscopic height variables φt =φt(x), t ≥ 0 was naturally introduced in Sect. 2.4 from the HamiltonianH(φ) by means of the Langevin equations. This section analyzes its macro-scopic behavior under the space-time diffusive scaling defined by (2.17). Weshall establish the LLN, called hydrodynamic limit, for ϕ-dynamics on T

dN

or on DN . It is shown that the evolutional law of the macroscopic interfacesis governed by the motion by mean curvature with anisotropy in the limit,


and described by the nonlinear PDE with diffusion coefficient formally givenby the Hessian of the surface tension σ = σ(u). The corresponding CLT for∇ϕ-dynamics on (Zd)∗ which is in equilibrium and the LDP for ϕ-dynamicson T

dN will be studied in Sects. 11 and 12, respectively. The dynamics with

the wall effect or those in two media realized by adding weak self potentialwill be discussed in Sects. 13, 14 and 15.

10.1 Space-Time Diffusive Scaling Limit

Let us consider the SDEs (2.9) on a big but finite lattice domain Γ . Weshall take Γ = T

dN (i.e., we discuss under periodic boundary conditions) or

Γ = DN for a bounded domain D in Rd having piecewise Lipschitz boundary

with properly scaled boundary conditions ψ ∈ R∂+DN . More exactly saying,

when Γ = TdN , the SDEs have the form

dφt(x) = −∑

y∈TdN :|x−y|=1


2dwt(x) , (10.1)

for x ∈ TdN , while, when Γ = DN , they have the form

dφt(x) = −∑

y∈DN :|x−y|=1


2dwt(x) , ()′

for x ∈ DN with the boundary conditions

φt(y) = ψ(y), y ∈ ∂+DN . ()′

(a) Main Theorem

Under the space-time diffusive scaling for the evolution of microscopic heightvariables φt = φt(x) of the interface, macroscopic height variables hN (t) =hN (t, θ) are defined as step functions on the torus T

d or on the domain Dby the formula (2.17):

hN (t, θ) =1NφN2t([Nθ]), θ ∈ T

d or D .

Or, we adopt the definition

hN (t, θ) =∑

x∈TdN (or DN )

1NφN2t(x)1B(x/N,1/N)(θ) , (10.2)

where B(θ, a) =∏d

i=1[θi−a/2, θi +a/2) denotes the d dimensional cube (box)with center θ = (θi)d

i=1 and side length a > 0.The goal is to study the behavior of hN (t) as N → ∞. Two definitions

(2.17) and (10.2) coincide if B(θ, a) is taken as∏d

i=1[θi, θi +a). The differenceis therefore only the componentwise shift in the variable θ by 1/2N , but this isnegligible in the limit. The conditions (V1)-(V3) in (2.2) are always assumedon the potential V .

218 T. Funaki

Theorem 10.1. (Hydrodynamic Limit, Funaki and Spohn [124] on thetorus T

d, Nishikawa [204] on D with boundary conditions) Assume that initialrandom configuration φ0 = φ0(x);x ∈ T

dN of the SDEs (10.1) converges to

some nonrandom h0 ∈ L2(Td) in the sense that

limN→∞

E[‖hN (0)− h0‖2] = 0 , (10.3)

where ‖ · ‖ denotes the usual L2-norm of the space L2(Td). Then, for everyt > 0,

limN→∞

E[‖hN (t)− h(t)‖2] = 0

holds and the limit h(t) = h(t, θ) is a unique weak solution of the nonlinearPDE

∂h

∂t(t, θ) = div ∇σ(∇h(t, θ))

≡d∑

i=1

∂

∂θi

∂σ

∂ui(∇h(t, θ))

, θ ∈ T

d, (10.4)

having initial data h0, where σ = σ(u) is the normalized surface tension de-fined in Sect. 5.1, (5.2).

The theorem is only stated for the torus Td, but a similar result holds on

D and in the space L2(D). In this case, the PDE (10.4) requires a macroscopicboundary condition g at ∂D. The PDE (10.4) describes the motion by meancurvature (MMC) with anisotropy, see the next paragraph (b). The limit h(t)is nonrandom and therefore Theorem 10.1 is at the level of the LLN.

Remark 10.1. (1) If σ ∈ C2(Rd) which is not yet shown (see Problem5.1), the diffusion coefficient of the PDE (10.4) is given by the Hessian(∂2σ/∂ui∂uj)ij of σ.(2) In the Gaussian case (i.e., V (η) = 1

2η2), σ(u) = 1

2 |u|2 and the limit equa-tion (10.4) is linear heat equation. In fact, this can be directly seen, since thedrift term of the SDEs (10.1) is ∆φt(x) for such potential V ; recall the SDEs(2.20) and that ∆ denotes the discrete Laplacian. The space-time diffusivescaling leads the discrete Laplacian to the continuum one.

(b) Physical Meaning of the PDE (10.4)

The total surface tensions ΣTd(h) on the torus Td or ΣD(h) onD of the macro-

scopic surface h = h(θ) were introduced in (6.2). The Frechet derivatives ofΣ = ΣTd or ΣD are given by

δΣ

δh(θ)(h) = −div (∇σ)(∇h(θ)) .

Therefore, the hydrodynamic equation (10.4) can be regarded as a gradientflow for Σ


∂h

∂t(t) = −δΣ

δh(h(t)) , (10.5)

namely the surface moves relaxing its total surface energy. For isotropic motionby mean curvature one would have σ(u) =

√1 + |u|2; note that Σ(h) is the

surface area of h and δΣ/δh is the mean curvature in such case. In our casethis is likely to hold for small |u|, however σ(u) |u|2 for large |u|, whichreflects the constraints due to the underlying microscopic lattice structure.

(c) Formal Derivation of the PDE (10.4)

We work on Td just for fixing the notation. For every test function f = f(θ) ∈

C∞(Td), we have that

〈hN (t), f〉 :=∫

Td

hN (t, θ)f(θ) dθ

=1

Nd+1

∑x∈T

dN

φN2t(x)[f ]N (x/N) ,

where[f ]N (x/N) := Nd

∫B(x/N,1/N)

f(θ) dθ .

Then, applying Ito’s formula and recalling the symmetry of V in our basicconditions (2.2), we have by summation by parts that

〈hN (t), f〉 − 〈hN (0), f〉

= −∫ t

0

1Nd

d∑i=1

∑x∈T

dN

V ′(∇iφN2s(x))∂f

∂θi

( x

N

)ds+ o(1). (10.6)

The last error term o(1) involves those for the replacement of ∇Ni [f ]N (macro-

scopically normalized discrete differential, see Sect. 10.2-(b) or Lemma 6.14)with ∂f/∂θi and the martingale term:

mN (t, f) =√

2Nd+1

∑x∈T

dN

wN2t(x)[f ]N( x

N

),

which goes to 0 since E[mN (t, f)2] = O(N−d). Note that the divergent factorN2 carried under the time change has disappeared, see Problem 10.1 below.

The left hand side of (10.6) would converge to 〈h(t), f〉 − 〈h0, f〉. On theother hand, the right hand side consists of a large scale sum of complex vari-ables. However, one would expect the so-called local equilibrium stateswere realized in the system, i.e., around each macroscopic space-time point(s, θ) the distribution of the gradient field corresponding to the microscopicheight variables φN2s(x);x ∼ Nθ would reach the equilibrium state µ∇

∇h(s,θ),

220 T. Funaki

which is the∇ϕ-Gibbs measure with mean tilt∇h(s, θ). Thus, by means of thelocal ergodicity, i.e., under large sum, V ′(∇iφN2s(x)) in the right hand sidecould be replaced with its ensemble average Eµ∇

∇h(s,x/N) [V ′(∇iφ(0))] whichcoincides with ∂σ/∂ui(∇h(s, x/N)) from (5.14); note that ∇iφ(0) = ∇φ(ei)under different two notation. Therefore, one would obtain in the limit

d

dt〈h(t), f〉 = −〈∇σ(∇h(t)),∇f〉 ,

for every f ∈ C∞(Td), which is a weak form of the nonlinear PDE (10.4).The actual proof given in [124] is slightly different. It is based on the methodof entropy production initiated by [141] (see also [170, 239]) and its variant,the H−1-method, by which one can avoid the so-called two blocks’ estimatenecessary in the standard route for establishing the hydrodynamic limit. Theresult in Sect. 9 (Corollary 9.6) is substantial to complete the proof.

Remark 10.2. (1) In one dimension, Theorem 10.1 gives essentially the sameresult that [141] obtained (without convexity condition on V ), since ηt ≡ ∇φt

satisfies the same SDEs that [141] considered. However, in higher dimensions,our ∇ϕ-Gibbs measures have long correlations and the situation is very dif-ferent from [141].(2) The martingale term mN (t, f) looks simply disappearing in the limit aswe have mentioned, but, in fact, this is not really true. For a > 0, considerthe SDEs (10.1) with the Brownian motions

√2wt(x) replaced by

√2awt(x).

Then, starting from such SDEs, we have the limit equation for the macroscopicheights

∂h

∂t(t, θ) = adiv (∇σa)(∇h(t, θ)) ,

where σa is the surface tension determined by the potential a−1V . Indeed,under the time change φt := φa−1t, one can apply Theorem 10.1 for φt. WhenV is quadratic, a∇σa(u) does not depend on a, but this may not be true ingeneral.

Problem 10.1. In (10.6), we have used the summation by parts formulanoting that V ′(η) = −V ′(−η) and this makes the right hand side of orderO(1) as N → ∞. If the potential V is asymmetric (cf. Remark 2.1), suchcancellation does not occur and the right hand side remains to be O(N) atleast at first look. Models involving such divergent quantities are called ofnongradient. The hydrodynamic limit for the Ginzburg-Landau dynamicsof nongradient type might be established based on Varadhan’s argument [247].

10.2 The Nonlinear PDE (10.4)

We start now the proof of Theorem 10.1 on Td. This subsection summarizes

results on the nonlinear PDE (10.4). We recall that the surface tension σsatisfies the properties stated in Sect. 5.3.


(a) Existence and Uniqueness of Solutions

Let us introduce a triple of real separable Hilbert spaces V ⊂ H = H∗ ⊂ V ∗ byH = L2(Td), V = H1(Td) := h ∈ H; |∇h| ∈ H and V ∗ = H−1(Td). We alsodenote by Hd the d-fold direct product of H. These three spaces are equippedwith their standard norms denoted by ‖ · ‖, ‖ · ‖V and ‖ · ‖V ∗ , respectively.The duality relation V 〈·, ·〉V ∗ between V and V ∗ satisfies V 〈v, h〉V ∗ = 〈v, h〉if v ∈ V and h ∈ H, where 〈·, ·〉 is the scalar product of H. We consider thenonlinear differential operator

A(h) =d∑

i=1

∂

∂θiσ′

i(∇h), h ∈ V ,

where σ′i(u) := ∂σ/∂ui, u ∈ R

d. The next lemma follows from Theorem 5.3and Corollary 5.4:

Lemma 10.2. The operator A : V → V ∗ has the following properties forall h, h1, h2 ∈ V . The constants c− and C are those appeared in (V3) andTheorem 5.3-(1), respectively.(A1) (semicontinuity) V 〈h,A(h1 + λh2)〉V ∗ is continuous in λ ∈ R,(A2) (monotonicity) V 〈h1−h2, A(h1)−A(h2)〉V ∗ ≤ −c−‖∇h1−∇h2‖2,(A3) (coercivity) V 〈h,A(h)〉V ∗ + c−‖h‖2V ≤ c−‖h‖2,(A4) (growth condition) ‖A(h)‖V ∗ ≤ C‖h‖V .

We call h(t) a solution (or anH-solution) of (10.4) with initial data h0 ∈ Hif h(t) ∈ C([0, T ],H) ∩ L2([0, T ], V ) and

h(t) = h0 +∫ t

0

A(h(s)) ds

holds in V ∗ for a.e. t ∈ [0, T ]. The general theory on nonlinear PDEs (e.g.,[10, 37, 178, 255]) proves the existence and uniqueness of solutions to (10.4)under the conditions (A1)–(A4) of Lemma 10.2.

Proposition 10.3. For every initial data h0 ∈ H the PDE (10.4) has aunique solution h(t). In addition, it admits the uniform bound

sup0≤t≤T

‖h(t)‖2 +∫ T

0

‖h(t)‖2V dt ≤ K(‖h0‖2 + 1) , (10.7)

where K is a constant depending only on c− and T .

(b) Discretization Scheme and Its Convergence

In order to prove Theorem 10.1, one needs to compare the discrete variablehN (t) with the continuum one h(t). It is therefore convenient to introduceh,N (t), a solution of lattice approximated version of the PDE (10.4), and

222 T. Funaki

compare hN (t) with h,N (t). For this purpose we define the finite differenceoperators

∇Ni f(θ) = N(f(θ + ei/N)− f(θ)) ,

∇N∗i f(θ) = −N(f(θ)− f(θ − ei/N)), θ ∈ T

d, 1 ≤ i ≤ d,

∇N = (∇N1 , ...,∇N

d ) .

With these notations the discretized PDE of (10.4) reads

∂

∂th,N (t, θ) = AN (h,N (t))(θ) := −

d∑i=1

∇N∗i σ′

i(∇Nh,N (t, θ)) , (10.8)

for θ ∈ 1N T

dN ≡ θ ∈ T

d;Nθ ∈ TdN ⊂ T

d. It has to be solved with the initialdata

h,N0 (θ) = [h0]N (θ) := Nd

∫[[θ]]N

h0(θ′) dθ′ (10.9)

where [[θ]]N stands for the box with center in 1N T

dN of side length 1

N containingθ ∈ T

d. Denoting by [θ]N the center of the box [[θ]]N , we extend h,N (t, θ) toT

d as a step function,

h,N (t, θ) := h,N (t, [θ]N ), for θ ∈ Td . (10.10)

We mention the convergence of the solution h,N (t) of the discretized PDE(10.8) to h(t) as N → ∞. The monotonicity of the operator is essential forthe proof, see Proposition I.2 in [124] for details.

Lemma 10.4. (1) For every t > 0, h,N (t) converges to h(t) weakly in H asN →∞, where h(t) is the unique solution of (10.4) with initial data h0 ∈ H.(2) Assume supN∈N

‖∇Nh,N0 ‖ <∞ in addition. Then the above convergence

holds strongly in H.

(c) Uniform Lp-Bound on ∇Nh,N (t);N ∈ N

In the proof of Theorem 10.1, a certain function of ∇φ-variables divergingquadratically in ∇φ arises. Such function can be controlled in the limit, if auniform Lp-bound on the ∇ϕ-dynamics is available for some p > 2, since itimplies the uniform L2-integrability of the function. However, unfortunately,we can only derive a uniform L2-bound for the ∇ϕ-dynamics, see Sect. 10.3-(a). We shall introduce the notion of coupled local equilibria and show that auniform Lp-bound for the discretized PDE, which is derived here, compensateswith the missing estimate on the ∇ϕ-dynamics.

Let h,N (t) be the solution of (10.8) with initial data h,N0 satisfying

supN∈N‖∇Nh,N

0 ‖ < ∞. We shall derive a uniform Lp-bound on ∇Nh,N (t)in N . The norm of the space Lp(Td) is denoted by ‖ · ‖p, 1 ≤ p ≤ ∞; recallthat ‖ · ‖2 is simply denoted by ‖ · ‖.


Lemma 10.5. We have that

supN∈N

supt≥0

‖∇Nh,N (t)‖ <∞ , (10.11)

supN∈N

∫ T

0

‖∇N∇Nh,N (t)‖2 dt <∞, T > 0 , (10.12)

where ‖∇N∇Nh‖2 =∑d

i,j=1 ‖∇Ni ∇N

j h‖2 , and for some p > 2,

supN∈N

∫ T

0

‖∇Nh,N (t)‖pp dt <∞ . (10.13)

Proof. The proof is due to an idea quite common in the theory of PDE, e.g.,see [180], p.433. Denoting h = h,N for simplicity, we have from (10.8),

d

dt‖∇Nh(t)‖2 = −2Nd−2

∑θ∈ 1

N TdN

d∑i=1

∇Nh(t, θ + ei/N)−∇Nh(t, θ)

·∇σ(∇Nh(t, θ + ei/N))−∇σ(∇Nh(t, θ))

≤ −2c−‖∇N∇Nh(t)‖2 . (10.14)

We have used ∇Ni ∇N∗

j = ∇N∗j ∇N

i and ∇Nj ∇N

i = ∇Ni ∇N

j , and subsequentlyCorollary 5.4. Hence,

‖∇Nh(t)‖2 + 2c−∫ t

0

‖∇N∇Nh(s)‖2 ds ≤ ‖∇Nh0‖2 ,

which shows (10.11) and (10.12). To show (10.13), we need Sobolev’s lemmafor lattice functions,

‖f‖22∗ ≤ C(‖∇Nf‖2 + ‖f‖2), f = f(θ), θ ∈ TdN, (10.15)

for some C > 0 independent of the lattice spacing N . Here 2∗ is the Sobolevconjugate of 2 defined by 2∗ = 2d/(d− 2) if d ≥ 3, 2∗ is an arbitrary numberlarger than 1 if d = 2 and 2∗ = ∞ if d = 1. Given (10.15), the proof of (10.13)can be completed from (10.11) and (10.12) using Holder’s inequality

‖f‖p ≤ ‖f‖1−τ‖f‖τq

with the choice of q = 2∗, p = 4− 4/2∗(> 2) and τ = p/2.

10.3 Local Equilibria

(a) Uniform Bound on Second Moments

As we have mentioned in Sect. 10.1-(c), one would expect that at positive(macroscopic) times the interface has locally a definite tilt u and a statistics

224 T. Funaki

as specified by the ∇ϕ-Gibbs measure µ∇u . Such a strong property will come

out only indirectly. However for the space-time averaged measure we willestablish that it is some mixture of ∇ϕ-Gibbs measures. In fact such propertywill be established for the measure coupled to the solution of a discretizedversion of the PDE (10.4), see Proposition 10.8 for a precise statement.

Let µ∇,Nt ∈ P(X

TdN

) be the distribution of ∇φt on XT

dN

, the state spacefor the ∇ϕ-field on the torus defined in the proof of Theorem 4.15, and letAvT (µ∇,N ) be its space-time average over [0, N2T ]× T

dN :

AvT (µ∇,N ) =1Nd

∑x∈T

dN

1N2T

∫ N2T

0

µ∇,Nt τ−1

x dt ,

for T > 0. Here τx : XT

dN→ X

TdN

denotes the shift by x on TdN (cf. Definition

2.3 on Zd) and note that Nd = |Td

N |. ν∇ ∈ P(XT

dN

) is always regarded asν∇ ∈ P(X ) by extending it periodically. We shall simply denote by µ∇

N =µ∇

N,0 ∈ P(XT

dN

) the finite volume ∇ϕ-Gibbs measure with periodic boundaryconditions and tilt u = 0 (see the proof of Theorem 4.15).

To obtain uniform L2-bounds, we again use a coupling argument for theSDEs (10.1) on T

dN . Assume that two initial data (RT

dN -valued random vari-

ables) φ0 = φ0(x);x ∈ TdN and φ0 = φ0(x);x ∈ T

dN are given and let φt

and φt be the corresponding two solutions of the SDEs (10.1) with commonBrownian motions. The macroscopic ϕ-fields obtained from φt and φt by scal-ing in space, time and magnitude as in (10.2) are denoted by hN (t, θ) andhN (t, θ), θ ∈ T

d, respectively. Recall that ‖ · ‖ denotes the norm of the spaceL2(Td).

Lemma 10.6. (1) We have for every t > 0

E[‖hN (t)− hN (t)‖2] ≤ E[‖hN (0)− hN (0)‖2] .

(2) Assume the condition (10.3) on the distribution µN0 of φ0. Then,

supN∈N

EAvT (µ∇,N )[η(b)2] <∞, b ∈ (Zd)∗ .

Proof. As in Lemma 9.7 we have

∂

∂t

∑x∈T

dN

(φt(x)

)2

≤ −c−∑

b∈(TdN )∗

(∇φt(b)

)2

with φt := φt − φt. On the torus TdN there is no boundary term. Integrating

both sides in t and dividing by Nd+2, we obtain

E[‖hN (t)− hN (t)‖2] + c−

∫ t

0

E

1Nd+2

∑b∈(Td

N )∗

(∇φs(b)

)2

ds

≤ E[‖hN (0)− hN (0)‖2] . (10.16)


This shows (1).We now take a special φ0: φ0(x) =

∑b∈CO,x

η(b) with the chain CO,x

connecting O and x and with XT

dN

-valued random variable η distributed underµ∇

N . Then,

d∑i=1

EAvT (µ∇,N )[η(ei)2] =1Nd

∑b∈(Td

N )∗

1N2T

∫ N2T

0

E[(∇φt(b))

2]dt

≤ 2Nd

∑b∈(Td

N )∗

1N2T

∫ N2T

0

E

[(∇φt(b)

)2]dt+ 2

d∑i=1

Eµ∇N [η(ei)2]

≤ 4Tc−

E[‖hN (0)‖2] + E[‖hN (0)‖2]

+ 2

d∑i=1

Eµ∇N [η(ei)2] .

We used the stationarity of µ∇N under the SDEs (9.2) on (Td

N )∗ for the sec-ond line and then (10.16) in the third. The last term in the right handside is bounded in N because of the uniform bound (4.18), take u = 0.Therefore, since µN

0 satisfies (10.3), the assertion (2) follows if one can showsupN∈N

E[‖hN (0)‖2] <∞. To this end we choose the chain CO,x connecting Oand x as follows: First we connect O and (x1, 0, . . . , 0) through changing onlythe first coordinate one by one. Then (x1, 0, 0, . . . , 0) and (x1, x2, 0, . . . , 0) areconnected through changing the second coordinate, etc.. With this choice,

E[‖hN (0)‖2] =1

Nd+2

∑x∈T

dN

E

∑

b∈C0,x

η(b)

2

≤ 1Nd+2

∑x∈T

dN

dNE

∑

b∈C0,x

η(b)2

≤ C

d∑i=1

Eµ∇N [η(ei)2] ,

which is bounded in N .

(b) Method of Entropy Production

To establish the local equilibria we will essentially follow the route of [141].We first note that, as pointed out in Sect. 4.1 (with ρ = 0 and Λ replaced byT

dN ), the generator of the process φt = φt(x);x ∈ T

dN is given by

LN =∑

x∈TdN

Lx ,

in which Lx are differential operators

Lx := eHN∂

∂φ(x)e−HN

∂

∂φ(x)= ∂2

x − ∂xHN · ∂x ,

226 T. Funaki

where HN = HT

dN

(φ) is the Hamiltonian defined by (2.1) on TdN and ∂x =

∂/∂φ(x). To write down the generator for the corresponding ∇ϕ-dynamicsηt ≡ ∇φt, which is the solution of the SDEs (9.2) on (Td

N )∗, we further notethat

∂x = 2∑

b:xb=x

∂

∂η(b)(10.17)

as operators acting on the functions F = F (η) ∈ C2b (X

TdN

) of variables ∇φ.Replacing ∂x in the definition of Lx with this formula, we obtain the differ-ential operators

L∇N =

∑x∈T

dN

L∇x ,

L∇x =

∑b,b∈(Td

N )∗:xb=xb=x

4

∂2

∂η(b)∂η(b)− 2V ′(η(b))

∂

∂η(b)

,

for x ∈ TdN . Two operators Lx and L∇

x (and therefore LN and L∇N ) coincide

when they act on C2b (X

TdN

). Thus L∇N is the generator corresponding to ηt.

Through integrating by parts its Dirichlet form is given by

−∫X

TdN

FL∇NGdµ∇

N

=∑

x∈TdN

∫X

TdN

∂xF ∂xGdµ∇N ,

= 4∑

x∈TdN

∫X

TdN

( ∑b:xb=x

∂F

∂η(b)

)( ∑b:xb=x

∂G

∂η(b)

)dµ∇

N , (10.18)

for F,G ∈ C2b (X

TdN

). For ν∇ ∈ P(XT

dN

) let IN (ν∇) be the entropy produc-tion defined by

IN (ν∇) = −4∫X

TdN

√FNL

∇N

√FN dµ∇

N ,

where FN (η) = dν∇/dµ∇N .

In order to apply the argument of [141], it is convenient to extend thedifferential operators on the whole lattice Z

d or on Λ Zd in the following

manner. Let C2loc,b(X ) be the class of all tame functions F on X of the form

F (η) = F (η(b); b ∈ Λ∗) for some Λ Zd and F ∈ C2

b (RΛ∗). We regard

L∇x , x ∈ Z

d, the differential operators acting on C2loc,b(X ); the sum in the

right hand side should be taken for b, b ∈ (Zd)∗ : xb = xb = x. We furtherdefine the differential operator L∇

Λ , Λ Zd, acting on C2

loc,b(X ) by

L∇Λ =

∑x∈Λ

L∇x .


The next lemma claims that, if the entropy production per unit volume con-verges to 0, the limit measure must be a superposition of µ∇

u ;u ∈ Rd.

Lemma 10.7. Let a sequence µ∇,N ∈ P(XT

dN

);N ∈ N be given, which istight in P(X ) and satisfies

limN→∞

N−dIN (µ∇,N ) = 0 . (10.19)

Then, every limit point ν∇ ∈ P(X ) of µ∇,N is a ∇ϕ-Gibbs measure.

Proof. On the infinite lattice we define the entropy production as follows: Forν∇ ∈ P(X ) and Λ Z

d,

IΛ(ν∇) = −4∫X

√FΛL

∇Λ

√FΛ dµ

∇ ,

where FΛ = dν∇/dµ∇|FΛ∗ with the σ-field FΛ∗ of X generated by η(b); b ∈

Λ∗ and µ∇ = µ∇0 ∈ P(X ) is the ∇ϕ-Gibbs measure with tilt u = 0. Consid-

ering µ∇,N ∈ P(X ), we have

IΛ(µ∇,N ) = sup−

∫X

L∇ΛG

Gdµ∇,N ; G is positive and FΛ∗ - measurable

≤ sup

−

∫X

TdN

L∇ΛG

Gdµ∇,N ; G is positive function on X

TdN

=|Λ|Nd

IN (µ∇,N ) = |Λ| × o(1)

as N →∞ by assumption, see [141] for the first and the third equalities. SinceIΛ is lower semicontinuous, the above bound implies IΛ(ν∇) = 0 for all weaklimits ν∇ in P(X ) of µ∇,N as N → ∞. To show that ν∇ is a ∇ϕ-Gibbsmeasure, we choose some FΛ∗ -measurable G ∈ C2

loc,b(X ). Then∣∣∣∣∫XL∇

ΛGdν∇∣∣∣∣ =

∣∣∣∣∫XL∇

ΛG · FΛ dµ∇∣∣∣∣

= 4

∣∣∣∣∣∑x∈Λ

∫X

( ∑b:xb=x

∂G

∂η(b)

)( ∑b:xb=x

∂FΛ

∂η(b)

)dµ∇

∣∣∣∣∣

≤ 2

√√√√∑x∈Λ

∫X

( ∑b:xb=x

∂G

∂η(b)

)2

dν∇ ×√IΛ(ν∇) = 0 .

This implies that ν∇|FΛ∗ is stationary under L∇

Λ , the generator for the SDEs(9.13) when the boundary condition ξ is fixed. The dynamics defined by (9.13)is ergodic. This can be seen through the diffeomorphism J : XΛ∗,ξ ! η → φ =φ(x);x ∈ Λ ∈ R

Λ defined by (2.8) and from the fact that the dynamics

228 T. Funaki

for φt = φt(x);x ∈ Λ defined by the SDEs (9.15) is ergodic. Its uniquestationary measure is the finite volume ∇ϕ-Gibbs measure µ∇

Λ,ξ ∈ P(XΛ∗,ξ),cf. Sect. 9.4, which implies the DLR equations for ν∇,

ν∇(·|F(Zd)∗\Λ∗)(ξ) = µ∇Λ,ξ(·), ν∇-a.e. ξ .

This proves that ν∇ is a ∇ϕ-Gibbs measure.

(c) Coupled Local Equilibria

Setu,N (t, x) ≡ (u,N

1 (t, x), ..., u,Nd (t, x)) = ∇Nh,N (t, x/N)

for x ∈ TdN and consider the probability measures

pN (dηdu) =1t

∫ t

0

1Nd

∑x∈T

dN

1u ,N (s,x)∈duµ∇,NN2s τ

−1x (dη) ds

on XT

dN×R

d (and therefore on X×Rd by periodic extension). This means that

we have coupled the distribution of the stochastic dynamics and the solutionof the discrete PDE (10.8). Lemmas 10.5 and 10.6-(2) prove

supN∈N

supb∈(Zd)∗

∫η(b)2 + |u|p pN (dηdu) <∞ (10.20)

for some p > 2. In particular, pN ;N ∈ N is tight in P(X × Rd) and, con-

sequently, one can choose from an arbitrary sequence N ′ →∞ a subsequenceN ′′ →∞ such that pN ′′

(dηdu) converges weakly on X ×Rd to some p(dηdu)

as N ′′ →∞.To characterize p, the following entropy bound is imposed on the initial

distributions µ∇,N0 ,

limN→∞

N−(d+2)HN (µ∇,N0 ) = 0 . (10.21)

This condition will be removed later. Here HN (ν∇) ≡ H(ν∇|µ∇N ) denotes the

relative entropy of ν∇ ∈ P(XT

dN

) with respect to µ∇N , recall (5.4).

Proposition 10.8. Under the condition (10.21), there exists λ ∈ P(Rd×Rd)

such that p can be represented in the form

p(dηdu) =∫

v∈Rd

µ∇v (dη) λ(dvdu) .

Proof. For G = G(u) ∈ Cb(Rd) and p(dηdu) ∈ P(X ×Rd) we shall denote the

integration of G with respect to p(dηdu) in u by p(dη,G) ∈ M(X ); the classof all signed measures on S having finite total variations is denoted by M(S).The subsequence N ′′ is simply denoted by N .


First we note that p(dη,G) is shift invariant for every G ∈ Cb(Rd). In fact,this can be shown by Lemma 10.5 first for G ∈ C1

b (Rd) and then for generalG by approximation.

We next show that, for every G ∈ Cb(Rd), p(dη,G) has a representation

p(dη,G) =∫

Rd

µ∇v (dη)λ(dv,G) (10.22)

with some λ(dv,G) ∈ M(Rd). Set pN (dη,G) := pN (dη,G)/∫X pN (dη,G) ∈

P(X ) for G ≥ c > 0. Then, since G > 0 and the entropy production IN (ν∇)is convex in ν∇, we have

IN (pN ) ≤ ‖G‖∞ct

∫ t

0

IN (µ∇,NN2s ) ds

=‖G‖∞ctN2

HN (µ∇,N

0 )−HN (µ∇,NN2t )

.

The second line is shown by noting that FN (t) = dµ∇,Nt /dµ∇

N is the solutionof the forward equation ∂FN (t)/∂t = L∇

NFN (t) and then using (10.18). SinceHN (µ∇,N

N2t ) ≥ 0, we conclude from the assumption (10.21) that pN ;N ∈ Nsatisfies the condition (10.19) and therefore Lemma 10.7 shows that its weaklimit p(·, G) = p(·, G)/

∫X p(dη,G) is a ∇ϕ-Gibbs measure. However, p(·, G)

is shift invariant as we have seen above and p(·, G) ∈ P2(X ) by using (10.20).Hence p(·, G) ∈ G∇ and consequently we see from Corollary 9.6

p(·, G) =∫

Rd

µ∇v (·) λ(dv,G)

for some λ(·, G) ∈ P(Rd). Thus we have obtained (10.22) for uniformly pos-itive G ∈ Cb(Rd) by taking λ(dv,G) =

∫X p(dη,G) × λ(dv,G). It also holds

for general G.The final task is to show that λ(dv,G) in (10.22) is represented as

λ(dv,G) =∫

Rd

G(u) λ(dvdu)

for every G ∈ Cb(Rd) with some λ ∈ P(Rd × Rd). To this end, one can apply

Birkhoff’s individual ergodic theorem for µ∇v , and then Stone-Weierstrass’s

theorem ([218], p.121) and Riesz-Markov’s theorem ([218], p.111). The detailsare omitted.

10.4 Proof of Theorem 10.1

We compare the solution of the SDEs with that of a discretized version of thePDE. Recalling the definition of h,N (t) from Sect. 10.2-(b), we have

E[‖h(t)− hN (t)‖2

]≤ 2‖h(t)−h,N (t)‖2+2E

[‖h,N (t)− hN (t)‖2

]. (10.23)

230 T. Funaki

The first term refers to the PDE only. In Lemma 10.4-(2) we proved that itconverges to zero in the limit N → ∞ under additional uniform bound on∇Nh,N

0 . But this assumption can be easily removed by approximating theinitial data h0 ∈ L2(Td) with smooth functions. In this subsection only thesecond term is handled. We first assume the entropy bound (10.21), which isactually removable.

By a straightforward computation of the L2-norm using Ito’s formula

E[‖h,N (t)− hN (t)‖2]

= E

1Nd

∑x∈T

dN

(h,N (t, x/N)−N−1φN2t(x))2

(10.24)

= E[‖h,N (0)− hN (0)‖2]− 2∫ t

0

(IN1 (s)− IN

2 (s)− IN3 (s) + IN

4 (s)) ds ,

where

IN1 (s) =

1Nd

∑x∈T

dN

d∑i=1

σ′i(u

,N (s, x))u,Ni (s, x) ,

IN2 (s) =

1Nd

∑x∈T

dN

d∑i=1

σ′i(u

,N (s, x))E[∇iφN2s(x)] ,

IN3 (s) =

1Nd

∑x∈T

dN

d∑i=1

u,Ni (s, x)E[V ′(∇iφN2s(x))] ,

IN4 (s) =

1Nd

∑x∈T

dN

d∑i=1

E[∇iφN2s(x)V ′(∇iφN2s(x))]− 1 .

Recall u,N (s, x) = ∇Nh,N (s, x/N). With the notation pN (dηdu), these termscan be rewritten as

∫ t

0

INk (s) ds = t

∫X×Rd

fk(η, u) pN (dηdu) ,

for k = 1, 2, 3, 4, where

f1(η, u) =d∑

i=1

uiσ′i(u), f2(η, u) =

d∑i=1

η(ei)σ′i(u) ,

f3(η, u) =d∑

i=1

V ′(η(ei))ui, f4(η, u) =d∑

i=1

η(ei)V ′(η(ei))− 1 .

One can pass to the limit for the first three terms, where the limit N ′′ → ∞should be taken along the subsequence N ′′ chosen in Sect. 10.3-(c),


limN ′′→∞

∫ t

0

IN ′′

k (s) ds = t

∫X×Rd

fk(η, u) p(dηdu) , (10.25)

for k = 1, 2, 3. Indeed, noting that |∇σ(u)| ≤ C(1 + |u|) and |V ′(η(ei))| ≤C(1 + |η(ei)|), we see

|f1(η, u)|p/2 ≤ C(1 + |u|p), p/2 > 1 ,

|f2(η, u)|q + |f3(η, u)|q ≤ C

(1 + |u|p +

d∑i=1

|η(ei)|2),

for q = 2p/(2+p) > 1. Therefore fk(η, u), k = 1, 2, 3, are uniformly integrablewith respect to the probability measures pN ;N ∈ N because of the uniformbound (10.20). Since pN ′′

converges weakly to p, we obtain (10.25). For thefourth term, since ηV ′(η) ≥ c−η

2 ≥ 0 (η ∈ R), one can apply Fatou’s lemmato obtain

lim supN ′′→∞

(−

∫ t

0

IN ′′

4 (s) ds)≤ −t

∫X×Rd

f4(η, u) p(dηdu). (10.26)

Summarizing (10.25), (10.26) and together with Proposition 10.8, Theorem5.5, we have proved that

lim supN ′′→∞

[−

∫ t

0

(IN ′′

1 (s)− IN ′′

2 (s)− IN ′′

3 (s) + IN ′′

4 (s)) ds]

≤ t

∫R2d

−u · ∇σ(u) + v · ∇σ(u) + u · ∇σ(v)− v · ∇σ(v) λ(dvdu)

= −t∫

R2d

(u− v) · (∇σ(u)−∇σ(v)) λ(dvdu) ≤ 0 . (10.27)

The convexity of σ (cf. Corollary 5.4) implies the nonpositivity of the lastintegral. This holds for some subsequence N ′′ → ∞ of arbitrarily takensequence N ′ → ∞. Hence, going back to (10.24), we have without takingsubsequence

limN→∞

E[‖h,N (t)− hN (t)‖2

]= 0 . (10.28)

Finally, from (10.23) we conclude Theorem 10.1 under the auxiliary entropyassumption (10.21).

However, the entropy assumption can be removed. The idea originally ob-served by Lu [194] is, roughly saying, that after a short time (macroscopicallyof order O(N−2)) the system gains the entropy bound. The details are omit-ted. Remark 10.3. (1) The technique employed for the proof of Theorem 10.1is called H−1-method, since the L2-norms for the height variables can be re-garded as the H−1-norms for the height differences. This method was proposedby Chang and Yau [56] to establish the nonequilibrium fluctuation for the first

232 T. Funaki

time. It has an advantage to skip the so-called 2-blocks’ estimate. As we haveseen, the 1-block estimate follows from the three steps: (a) showing that limitmeasures have entropy production 0, (b) 0-entropy production implies the ∇ϕ-Gibbs property, and (c) characterization of all ∇ϕ-Gibbs measures. Indeed, inSect. 9, we have characterized all stationary measures for the infinite volume∇ϕ-dynamics (Theorem 9.3), which is stronger result than (c). Under suchsituation, the proof of 1-block estimate can be simplified, in particular, oneneed not rely on the entropy production, see [114, 220, 240].(2) Under different setting, Abraham et al. [2] discussed the dynamics withboundary conditions related to the wetting transition. They observed the in-terfaces sideways and proved that the wetting transition occurs depending onthe strength of the potential V . See also [59]. For lattice gasses, the boundaryconditions were discussed by [96, 171].(3) From the view point of the nonlinear PDE theory, Giga and Giga [136, 137]studied the case where the surface tension has anisotropy and singularity. Inparticular, they treated the evolution of facets.(4) The hydrodynamic limit involving phase transitions is not well establishedexcept [219].

10.5 Surface Diffusion

So far, we have been considering the dynamics (2.9) (or, equivalently (2.11)–(2.13) or (10.1)) associated with the HamiltonianH in the sense that they havethe ϕ-Gibbs measures as their equilibrium states. Indeed, one can introducevarious types of dynamics having such properties. For instance, let A be a(nonnegative) operator acting on the spatial variable x and consider the SDEs

dφt(x) = −A ∂H

∂φ(x)(φt)dt+

√2Adwt(x) , (10.29)

for x ∈ Γ . Then, at least formally, φt is reversible under the ϕ-Gibbs measureson Γ . This can be seen by writing down the corresponding generator andchecking that it is symmetric under the ϕ-Gibbs measures. See Hohenbergand Halperin [147] for the physical background for the SDEs (10.29). Thedynamics (2.9) is the special case that A = I (identity map).

Let us, in particular, take A = −∆. Then, (10.29) can be rewritten as

dφt(x) =∑

y:|x−y|=1

∑z:|z−y|=1

V ′(φt(y)− φt(z))

−∑

z:|z−x|=1

V ′(φt(x)− φt(z))

dt

+√

2dwt(x) , (10.30)


for x ∈ Γ , where wt(x);x ∈ Γ are Gaussian processes with mean 0 andcovariance

E[wt(x)ws(y)] = −∆(x, y) · t ∧ s ,which is a precise realization of

√−∆wt(x).∆(x, y) is the kernel of the discrete

Laplacian ∆.The SDEs (10.30) on the lattice torus Γ = T

dN have, contrarily to the

dynamics (2.9), the conservation law (at microscopic level):∑x∈T

dN

φt(x) =∑

x∈TdN

φ0(x) ,

for t > 0. Indeed, taking the sum in x, the drift term in (10.30) cancels outand, moreover,

∑x∈T

dNwt(x) = 0 holds. The sum

∑x∈T

dNφt(x) represents the

total volume of the phase below the interface. The dynamics determined by(2.9) does not have such property. In this sense, the time evolutions definedby (2.9) and (10.30) may correspond to the Glauber and Kawasaki dynamics,respectively, in particles’ systems. The SDEs (10.30) are sometimes adopted asa model for alloys, since the total numbers of atoms of two kinds of metals arepreserved, respectively, under the time evolution. Because of such conservationlaw, one may think of that the atoms move around only over the surfaceseparating two phases. Thus the model is called the surface diffusion, see[240] for details.

Nishikawa [203] introduced the macroscopic scaling

hN (t, θ) =∑

x∈TdN

1NφN4t(x)1B(x/N,1/N)(θ) , (10.31)

for θ ∈ Td and the solution φt = φt(x);x ∈ T

dN of (10.30) on the lattice

torus and proved the following theorem. The potential V satisfies the basicconditions (V1)-(V3) in (2.2).

Theorem 10.9. Assume that hN (0) satisfies two conditions

limN→∞

E[‖hN (0)− h0‖2H−1(Td)] = 0 ,

supN

E[‖hN (0)‖2L2(Td)] <∞ ,

for some h0 ∈ L2(Td). Then, for every t > 0,

limN→∞

E[‖hN (t)− h(t)‖2H−1(Td)] = 0

holds and the limit h(t) = h(t, θ) is a unique weak solution of the nonlinearPDE

∂h

∂t(t, θ) = −∆ [div (∇σ)(∇h(t, θ))]

≡ −d∑

i,j=1

∂2

∂θ2j

∂

∂θi

∂σ

∂ui(∇h(t, θ))

, θ ∈ T

d , (10.32)

234 T. Funaki

having initial data h0, where H−1(Td) stands for the Sobolev space over Td

equipped with the standard norm ‖ · ‖H−1(Td) and σ = σ(u) is the normalizedsurface tension as before.

Compared with the scaling (10.2) for the case without conservation law,the space-time scaling ratio is N : N4 in (10.31). The limit equation (10.32)is of fourth order. For instance, when the potential is quadratic V (η) = 1

2η2,

we easily see that the scaling is proper and the limit equation is of the form

∂h

∂t= −∆2h .

The space-time scaling ratio is closely related to the spectral gap of the gener-ator corresponding to the process φt = φt(x);x ∈ T

dN. The gap behaves as

O(1/N2) as N →∞ for the dynamics (10.1), while it behaves as O(1/N4) for(10.30). In other words, the system with conservation law requires longer timeto relax to the equilibrium state. These gaps are seen from the logarithmicSobolev inequalities, which are obtained based on the theory of Bakry andEmery [8] noting that our potential V is convex, cf. [111]. Compare with theresults [126, 183, 195] for particles’ systems.

The proof of Theorem 10.9 is similar to that of Theorem 10.1, but basedon the H−2-method rather than H−1-method since the basic norms change.The main task is again the characterization of all ∇ϕ-Gibbs measures corre-sponding to the dynamics (10.30) on Γ = Z

d. Under the conservation law,such Gibbs measures should be called the canonical ∇ϕ-Gibbs measures. Un-fortunately, the method of energy inequality developed in Sect. 9.3 does notwork well for conservative system. Instead, Nishikawa proved that the classof canonical ∇ϕ-Gibbs measures and that of ∇ϕ-Gibbs measures for non-conservative system coincide under the shift invariance. Thus one can applyCorollary 9.6 to characterize the canonical ∇ϕ-Gibbs measures.

In Sect. 10.1-(d), we notified that the macroscopic interface equation (10.4)is nothing but the gradient flow for the total surface tensionΣ. The basic spacewas L2(Td) there and the Frechet derivatives were computed on this space.For the conservative system, the basic space should be replaced with H−1(Td)and the Frechet derivative δΣ/δh(θ) of Σ must be computed on this space,i.e.,

d

dεΣ(h+ εg)

∣∣∣∣ε=0

=

(δΣ

δh, g

)

H−1

,

for every g ∈ C∞(Td), where the inner product is defined by (f, g)H−1 =((−∆)−1f, g)L2 . Thus we have

δΣ

δh(θ)= ∆ [div (∇σ)(∇h(θ))]

and therefore the limit equation (10.32) is again the gradient flow for Σ andhas the form


∂h

∂t(t) = − δΣ

δh(h(t)) .

The derivation of the motion by mean curvature from bistable reaction-diffusion equations (sometimes called Allen-Cahn equation) via singular limitis extensively studied in recent years in nonlinear PDE theory, see [205] andalso Sect. 16.3 (for the equations with noises). The conservative system is de-scribed by the fourth order PDE called Cahn-Hilliard equation. The interfacialequation (10.32) derived here might coincide with the equation derived fromCahn-Hilliard type equation via singular limit. See Visintin [251] for variousapproaches to the problems related to the phase transitions from the viewpoint of PDE theory.

Bertini et al. [17] derived fourth order PDE via hydrodynamic limit. Thelimit equation is the same as (10.32) in Theorem 10.9 in one dimension. How-ever, in higher dimensions, the equilibrium measures of the model treated by[17] are Bernoulli product measures, while they have long correlations for themodel discussed here. Therefore, these two results are essentially different.

11 Equilibrium Fluctuation

Let φt(x);x ∈ Zd be the solution of the SDEs (2.13) with initial data φ0

whose gradients ∇φ0 are distributed according to the ∇ϕ-pure phase µ∇u for

some u ∈ Rd. Note that the process ηt ≡ ∇φt is in equilibrium. Consider an

S ′(Rd)-valued process

ΨNi (t, dθ) = N−d/2

∑x∈Zd

(∇iφN2t(x)− ui) δx/N (dθ), θ ∈ Rd ,

for each 1 ≤ i ≤ d, where S ′(Rd) stands for the class of Schwartz distributionson R

d. The potential V satisfies the conditions (V1)-(V3) in (2.2).

Theorem 11.1. (Equilibrium fluctuation, [135]) The process ΨNi (t) weakly

converges as N →∞ in the space C([0,∞),S ′(Rd)) to an equilibrium solutionΨi(t) of the stochastic PDE (stochastic partial differential equation)

∂Ψi

∂t(t, θ) = −AΨi(t, θ) +

√2∂B

∂θi(t, θ) ,

where B(t, θ) is the space-time white noise and

A = −d∑

i,j=1

qij∂2

∂θi∂θj.

The positive definite d × d matrix (qij) ≡ (qij(u))1≤i,j≤d is characterized bythe variational formula:

236 T. Funaki

v · qv =2 infF

d∑i=1

Eµ∇u

[(ui −DiF (η))2V ′′(η(ei))

]

+∑x∈Zd

Eµ∇u [(∂xF )2]

, (11.1)

where v ∈ Rd, the infimum is taken over all F = F (η) ∈ C∞

loc,b(X ), DiF (η) =F (τei

η) − F (η), τei: X → X is the shift (cf. Definition 2.3) and ∂xF (η) is

defined by (10.17).

The proof relies on the Helffer-Sjostrand type representation for correla-tion functions of the process ηt ≡ ∇φt in terms of random walk in movingrandom environment. Such representation is also used in [202] (see Sect. 8)for static correlation functions. Then the problem is reduced to establishingthe homogenization for this random walk, cf. [173].

In the stochastic PDE appearing in Theorem 11.1, the strength of thenoise (diffusion coefficient), the drift term (indicating frictional drag ordissipation of energy) and the variance of the equilibrium measure (calledfluctuation) automatically satisfy a certain relation, which is called thefluctuation-dissipation theorem. Einstein relation for the Langevin equa-tion is the typical and simplest example.

Remark 11.1. Theorem 11.1 covers the static CLT discussed in Sect. 8.

Problem 11.1. [135] If σ ∈ C2(Rd) (see Problem 5.1), does the covariancematrix q(u) = (qij(u)) appearing in the CLT coincide with the Hessian ofσ(u)? Recall that σ(u) arises in the LLN and the LDP in various manner.Such identity is ordinary in statistical mechanics, for instance if the Gibbsmeasures have exponentially fast mixing property.

12 Dynamic Large Deviation

12.1 Dynamic LDP

The hydrodynamic limit is the LLN for the macroscopic height variables hN (t)defined in (2.17). In this section we study the corresponding LDP on thelattice torus T

dN . We always assume the conditions (V1)-(V3) in (2.2) on

the potential V . The result is the following: Assume that the initial dataφ0(x) ≡ φN

0 (x);x ∈ TdN of the SDEs (10.1) are deterministic and satisfy

supN

|φN

0 (O)|+ 1Nd

∑b∈(Td

N )∗

(∇φN0 (b))2

<∞ . (12.1)

We also assume the condition (10.3) without taking expectations for the cor-responding hN (0) and some h0 ∈ L2(Td).


Theorem 12.1. (Dynamic LDP, Funaki and Nishikawa [121] on Td) The

LDP holds for hN (t); t ∈ [0, T ] with speed Nd and rate functional IT (h):

P(hN (t) ∼ h(t), t ≤ T

)

N→∞exp−NdIT (h) ,

where h(t) = h(t, θ) is a given motion of surface. More precisely, for everyclosed set C and open set O of C([0, T ], L2

w(Td)), we have that

lim supN→∞

1Nd

logP (hN (·) ∈ C) ≤ − infh∈C

IT (h) , (12.2)

lim infN→∞

1Nd

logP (hN (·) ∈ O) ≥ − infh∈O

IT (h) , (12.3)

where L2w(Td) is the space L2(Td) equipped with the weak topology and

C([0, T ], L2w(Td)) stands for the class of all continuous functions h : [0, T ] →

L2w(Td). The precise form of the rate functional IT (h) ≡ IT (h(·)) is stated in

the subsequent section.

12.2 Dynamic Rate Functional

For each h = h(t, θ) which is differentiable,

IT (h) =14

∫ T

0

dt

∫Td

∂h

∂t(t, θ)− div[(∇σ)(∇h(t, θ))]

2

dθ , (12.4)

if h(0) = h0 and IT (h) = +∞ if h(0) = h0, where σ is the normalizedsurface tension. More precisely saying, for h ∈ C([0, T ], L2

w(Td)) satisfyingh(t) ∈ H1(Td) for a.e. t ∈ [0, T ]

IT (h) = supJ=J(t,θ)∈C1([0,T ]×Td)

IT (h;J) ,

where

IT (h;J) =∫

Td

J(T, θ)h(T, θ) dθ −∫

Td

J(0, θ)h0(θ) dθ

−∫ T

0

dt

∫Td

∂J

∂t(t, θ)h(t, θ) dθ

+∫ T

0

dt

∫Td

∇J(t, θ) · ∇σ(∇h(t, θ)) dθ −∫ T

0

dt

∫Td

J2(t, θ) dθ ,

and H1(Td) denotes the Sobolev space on Td.

The upper bound in Theorem 12.1 is shown based on the exponentialChebyshev’s inequality, while for the lower bound the hydrodynamic limitfor a weakly perturbed system is established and then Girsanov’s formulais applied. These ideas are rather standard. Essential role is played by the

238 T. Funaki

superexponential estimate, namely the probability of replacing sample meanover box of side length Nε with ensemble mean is superexponentially smallas N →∞ and then ε ↓ 0. The H−1-method is effectively used to prove suchestimate.

Remark 12.1. In one dimension, Theorem 12.1 was proved by Donsker andVaradhan [88]. See also [172, 184] for the LDPs corresponding to the hydro-dynamic limit.

12.3 Relation to the Static LDP

Let µψN ≡ µψ

DNbe the finite volume ϕ-Gibbs measure (2.4) on DN with

boundary condition ψ satisfying the conditions (6.7). Then, Theorem 6.1 withU ≡ 0 shows the LDP for macroscopic height variables hN = hN (θ); θ ∈ Ddistributed under µψ

N and the (unnormalized) rate functional is given by thetotal surface tension ΣD(h) in (6.2).

Going back to the torus, since the distribution of ∇NhN (T ) weakly con-verges as T →∞ to the macroscopically scaled ∇ϕ-field under the finite vol-ume ∇ϕ-Gibbs measure on T

dN , one would expect that the static LDP could

be recovered from the dynamic LDP. An affirmative answer is not known atpresent, however one can at least recover the static rate functional from thedynamic one as T →∞. In fact, denoting the distribution of hN (T ) by µN (T ),the contraction principle implies the LDP for µN (T )N with rate functional

ST (h) = infIT (h); h = h(t, θ) s.t. h(T, θ) = h(θ)

for h = h(θ) ∈ H1(Td). The relationship between ST (h) and the total surfacetension ΣTd(h) on the torus (defined by (6.2) with D = T

d) is stated in thefollowing proposition.

Proposition 12.2.lim

T→∞ST (h) = ΣTd(h) .

The limit in the left hand side is called a quasi-potential and the corre-sponding classical flow is a gradient flow for the potential Σ as pointed outin Sect. 10.1. In such case the quasi-potential coincides with the potential Σitself. The infimum is attained by the reversed trajectory of the gradient flow.

13 Hydrodynamic Limit on a Wall

Under the static situation, several modifications were made to the Hamiltonianand the corresponding ϕ-Gibbs measures. We have considered, for instance,wall effect by conditioning φ ≥ 0, two media system by introducing weak selfpotentials and pinning effect near the height level 0. The associated dynamicscan be constructed in such a manner that it is reversible under the modified


ϕ-Gibbs measures. In the following sections, we discuss the problems of hy-drodynamic limit and fluctuations for such dynamics. Entropic repulsion isalso studied.

13.1 Dynamics on a Wall

The dynamics for the microscopic interfaces φt = φt(x);x ∈ Γ on a wall isintroduced by SDEs of Skorokhod type:

dφt(x) = − ∂H

∂φ(x)(φt) dt+

√2dwt(x)

+1Nf

(t

N2,x

N,

1Nφt(x)

)dt+ dt(x), x ∈ Γ , (13.1)

subject to the conditions

φt(x) ≥ 0, t(x) , 0(x) = 0 and∫ ∞

0

φt(x) dt(x) = 0 , (13.2)

for each x ∈ Γ , where ∂H/∂φ(x) is defined as in (2.10), wt = wt(x);x ∈ Γis a family of independent one dimensional standard Brownian motions andf = f(t, θ, h) is a given macroscopic external force, for instance, a mild pinningeffect on the interfaces from the wall. The interfaces can move over the wallsettled at height level 0 so that the height variables always satisfy φt(x) ≥ 0.The condition “t(x) ” means that t(x) called local time of φt(x) at 0 isnondecreasing in t and the last condition in (13.2) implies that t(x) increasesonly when φt(x) = 0. In particular, dt(x) = 0 if φt(x) > 0 and dt(x)represents a strong repelling force from the wall when the interfaces touchit. The external force f is microscopically scaled in the equation (13.1) tohave nontrivial macroscopic limit, see Theorem 13.1. If f = f(t, θ, h) is jointlycontinuous in these three variables and Lipschitz continuous in h, then theSDEs (13.1) subject to (13.2) (and with boundary conditions (2.12)) have aunique solution, see [191, 244].

The unique stationary measure of the dynamics determined by (13.1)–(13.2) when f = 0 and Γ = DN with 0-boundary conditions is given byµ+

N ≡ µ0DN

( · |φ ≥ 0), the conditional probability of the finite volume ϕ-Gibbsmeasure µ0

DN. This measure is reversible under the dynamics.

13.2 Hydrodynamic Limit

In the next theorem, we work on the lattice torus Γ = TdN and assume the

following conditions on f

240 T. Funaki

(E1) f ∈ C1([0,∞)× Td × [0,∞)) ,

(E2) there exist constants C > 0 and κ ∈ (0, 1) such that

|f(t, θ, h)|+∣∣∣∣ ∂f∂θi

(t, θ, h)∣∣∣∣ +

∣∣∣∣∂f∂t (t, θ, h)∣∣∣∣ ≤ C(1 + |h|κ) ,

for 1 ≤ i ≤ d and

− C ≤ ∂f

∂h(t, θ, h) ≤ 0,

for every (t, θ, h) ∈ [0,∞)× Td × [0,∞) .

The condition (E2) means the sublinear growth and nonincreasing propertyof f in h.

Theorem 13.1. (Hydrodynamic limit, Funaki [117] on Td) As N → ∞,

the macroscopic height variables hN (t, θ) defined by (2.17) or (10.2) convergeto h(t, θ) in probability, i.e., for every ϕ ∈ H = L2(Td) and t, δ > 0 ,

limN→∞

P(|〈hN (t), ϕ〉 − 〈h(t), ϕ〉| > δ

)= 0 ,

if this condition holds at t = 0 and if supNE[‖hN (0)‖2H ] <∞. The limit h(t, θ)is a unique solution of the evolutionary variational inequality(MMC withreflection (obstacle)):

h ∈ L2([0, T ], V ),∂h

∂t∈ L2([0, T ], V ′), ∀T > 0 , (a)

⟨∂h

∂t(t), h(t)− v

⟩+ 〈∇σ(∇h(t)),∇h(t)−∇v〉

≤ 〈f(t, h(t)), h(t)− v〉 , a.e. t, ∀v ∈ V : v ≥ 0 , (b)h(t, θ) ≥ 0, a.e. , (c)h(0, θ) = h0(θ) , (d)

where V = H1(Td), V ′ = H−1(Td) and 〈·, ·〉 denotes the inner product of H(or Hd) or the duality between V ′ and V .

Note that, if h(t, θ) > 0, the condition (b) implies that h(t) satisfies thePDE (10.4) with external force f (roughly saying, by taking v = h(t) + εv forsufficiently small ε and v ∈ V ):

∂h

∂t(t, θ) = div (∇σ)(∇h(t, θ))+ f(t, θ, h(t, θ)) .

The evolutionary variational inequality (EVI) describes the strong repellingeffect from the wall when the macroscopic interfaces touch it, i.e., h(t, θ) = 0.

The proof of Theorem 13.1 is completed based on the penalty method,comparison theorem on SDEs, superexponential 1-block and 2-blocks’ esti-mates, tightness argument from energy inequality and results on the EVI dueto Bensoussan and Lions [16].


Remark 13.1. Rezakhanlou [221, 222] derived a Hamilton-Jacobi equationunder hyperbolic scaling from growing SOS dynamics (φ(x) ∈ Z) with con-straints on the gradients (e.g., ∇φ(x) ≤ v). Related results were obtained byEvans and Rezakhanlou [95] and Seppalainen [229].

14 Equilibrium Fluctuation on a Walland Entropic Repulsion

The first two subsections discuss the fluctuation problem for equilibrium ϕ-dynamics on a wall in one dimension with boundary conditions. The thirdsubsection briefly summarizes recent results on dynamic entropic repulsion.

14.1 The Case Attached to the Wall

Let us consider the equilibrium dynamics φt on the wall in one dimension, i.e.,φt is a solution of the SDEs (13.1)–(13.2) with d = 1, Γ = DN ≡ 1, 2, . . . , N−1,D = (0, 1), f = 0 under the 0-boundary conditions:

φt(0) = φt(N) = 0 (14.1)

and with an initial distribution µ+N ≡ µ0

DN( · |φ ≥ 0). Macroscopic fluctua-

tion field around the hydrodynamic limit h(t, θ) = 0 is defined by

ΦN (t, θ) =√NhN (t, θ)

=∑

x∈DN

1√NφN2t(x)1[ x

N − 12N , x

N + 12N )(θ), θ ∈ D = [0, 1] .

Since ΦN (t, θ) ≥ 0, the limit is certainly non-Gaussian if it exists and theresult must be different from the usual CLT. In fact, we have the followingtheorem.

Theorem 14.1. (Equilibrium fluctuation, Funaki and Olla [122]) As N →∞, ΦN (t, θ) weakly converges to Φ(t, θ) on the space C([0, T ],H−α([0, 1])) ∩L2

w([0, T ] × [0, 1]) for every T > 0 and α > 1/2. The limit Φ(t, θ) is aunique weak stationary solution of the stochastic PDE with reflectionof Nualart-Pardoux type (cf. [208]):

∂Φ

∂t(t, θ) = q

∂2Φ

∂θ2(t, θ) +

√2B(t, θ) + ξ(t, θ), θ ∈ (0, 1) ,

Φ(t, θ) ≥ 0,∫ ∞

0

∫ 1

0

Φ(t, θ) ξ(dtdθ) = 0 ,

Φ(t, 0) = Φ(t, 1) = 0, ξ: random measure ,

where H−α([0, 1]) is the Sobolev space on [0, 1] determined from Dirichlet 0-boundary conditions, B(t, θ) is the space-time white noise, q = 1/〈η2〉ν0(=1/u′(0)) and ν0 ∈ P(R) is defined by (5.25) with λ = 0.

242 T. Funaki

The proof is based on the penalization, i.e., we replace the terms of thelocal times t(x) with strong positive drifts when the interfaces try to movetoward the negative side. This replacement gives the lower bound for theSDEs (13.1)–(13.2) by comparison theorems. Therefore the equilibrium fluc-tuation result for the SDEs with penalization, which is established throughthe so-called Boltzmann-Gibbs principle, gives the lower estimate for the limitΦ(t, θ). On the other hand, to obtain the upper bound for Φ(t, θ), since we areconcerned with the equilibrium situation, once the dynamic lower bound isestablished, one only needs to show the static upper bound. Indeed, it is notdifficult to prove that the stationary measure µ+

N weakly converges under thescaling of our interest to the distribution of (properly time changed) three di-mensional pinned Bessel process, which is the stationary measure of thestochastic PDE of Nualart-Pardoux type. Zambotti [256] has some extension.

14.2 The Case Away from the Wall

In the last subsection, we have discussed under the 0-boundary conditions(14.1). Then the macroscopic heights hN (t, θ) converges to 0 and, in thissense, the interfaces are attached to the wall at the macroscopic level.

Here, taking a, b > 0, let us consider the SDEs (13.1)–(13.2) under thepositive boundary conditions

φt(0) = aN, φt(N) = bN . (14.2)

The initial distribution is taken as µ+,aN,bNDN

≡ µaN,bNDN

(·|φ ≥ 0), the finitevolume ϕ-Gibbs measure on DN = 1, 2, . . . , N−1 with boundary conditionsaN and bN at x = 0 and N , respectively, conditioned to be φ ≥ 0. Then thesystem is stationary and reversible. Since the limit of hN (t, θ) as N → ∞ isgiven by

h(θ) = a+ (b− a)θ, θ ∈ [0, 1] ,

the fluctuation field is defined by

ΦN (t, θ) =√N

(hN (t, θ)− h(θ)

).

Theorem 14.2. ([122]) As N → ∞, ΦN (t, θ) weakly converges to Φ(t, θ)on the space C([0, T ],H−α([0, 1])) ∩ L2

w([0, T ] × [0, 1]) for every T > 0 andα > 1/2. The limit Φ(t, θ) is a unique weak stationary solution of the stochasticPDE:

∂Φ

∂t(t, θ) = qb−a

∂2Φ

∂θ2(t, θ) +

√2B(t, θ), θ ∈ (0, 1) ,

Φ(t, 0) = Φ(t, 1) = 0 ,

where qu = 1/u′(λ) with λ = λ(u), see Sect. 5.5.


Intuitively saying, since the fluctuation of the interfaces is of O(√N), they

do not feel the effect from the wall under the boundary conditions (14.2) sothat the limit of the fluctuation fields becomes Gaussian contrary to the casestudied in Theorem 14.1.

Problem 14.1. In higher dimensions, the expected scaling for the fluctuationfield might be

Nd/2(hN (t, θ)− E[hN (t, θ)]

),

i.e., the order of the fluctuation for the ϕ-field is expected to be of O(N−d/2+1).Compare this with the results in entropic repulsion and then you see that thefluctuation is much smaller than the order of the mean E[hN (θ)] except whend = 1. The fluctuation field accordingly may not feel the wall and the limitmight be Gaussian even under the 0-boundary conditions when d ≥ 2.

14.3 Dynamic Entropic Repulsion

The problem of entropic repulsion (see Sect. 7.1 for static results) can beinvestigated under the time evolutions.

In fact, Deuschel and Nishikawa [80] discussed the ϕ-dynamics on a wall onZ

d, i.e. the dynamics governed by the SDEs of Skorokhod type (13.1)–(13.2)with f ≡ 0 and Γ = Z

d. They proved that, starting from an i.i.d. initialdistribution with finite variance, the solution behaves as φt(x) = O(

√logd(t))

as t→∞ for d ≥ 2.Dunlop et al. [91] studied the SOS type dynamics on a wall on Z (under

the constraint |∇φt(x)| = 1) and proved that c1t1/4 ≤ E[φt(x)] ≤ c2t1/4 log t.

Ferrari et al. [98] and Fontes et al. [103] discussed the serial harness process(introduced by Hammersley). In particular, the latter was concerned withthe dynamic entropic repulsion on a random wall, cf. [18] under the staticsituation.

15 Dynamics in Two Media and Pinning Dynamicson a Wall

15.1 Dynamics in Two Media

Let us consider the microscopic dynamics associated with the Hamiltonian(6.3) having a weak self potential in one dimension, i.e., let φt = φt(x); 0 ≤x ≤ N be the solution of the SDEs

dφt(x) = − ∂H

∂φ(x)(φt) dt+

√2dwt(x)− f(φt(x))dt ,

for 1 ≤ x ≤ N − 1 with the boundary conditions

φt(0) = aN, φt(N) = bN ,

244 T. Funaki

where f(r) = W ′(r); we assume Q ≡ 1 so that U(θ, r) = W (r) for simplicity.Then the following result is expected on its hydrodynamic behavior.

Assume a ≥ 0 ≥ b and A =∫

Rf(r) dr ≥ 0. Then, as N →∞, the macro-

scopically scaled height variable hN (t, θ) converges to h(t, θ) in probability. Thelimit h(t, θ) is a solution of the free boundary problem for the nonlinearPDE (10.4)

∂h

∂t(t, θ) =

∂

∂θ

σ′

(∂h

∂θ(t, θ)

)on (t, θ);h(t, θ) = 0 ,

Ψ(h′+(t, θ))− Ψ(h′−(t, θ)) = A on (t, θ);h(t, θ) = 0 ,h(t, 0) = a, h(t, 1) = b ,

where Ψ(u) = σ′(u)u − σ(u) and, h′+(θ) and h′−(θ) are derivatives of h at θfrom the positive and negative sides of h, respectively.

The above mentioned free boundary problem was studied by Caffarelliet al. [43, 44, 45].

15.2 Pinning Dynamics on a Wall

This subsection is taken from unpublished notes based on a discussion withJ.-D. Deuschel. We construct the dynamics of microscopic interfaces underthe effects of pinning and repulsion, and discuss its reversibility.

Dynamics Without Volume Conservation Law

Let Λ Zd be given. For nonnegative height variables φ = φ(x);x ∈ Λ ∈ R

Λ+

on Λ, the Hamiltonian H(φ) ≡ HψΛ (φ) with boundary conditions φ(x) =

ψ(x) ≥ 0, x ∈ ∂+Λ was introduced in (2.1). We consider the SDEs for φt =φt(x);x ∈ Λ

dφt(x) = −1(0,∞)(φt(x))∂H

∂φ(x)(φt)dt

+ 1(0,∞)(φt(x)) ·√

2dwt(x) + dt(x), x ∈ Λ , (15.1)

subject to the conditions:

(a) φt(x) ≥ 0, t(x) , 0(x) = 0,

(b)∫ ∞

0

φt(x) dt(x) = 0, (15.2)

(c) ct(x) =∫ t

0

10(φs(x)) ds,


for every x ∈ Λ. We shall choose c = eJ (≥ 0) for J ∈ [−∞,∞). The boundaryconditions (2.12) at y ∈ ∂+Λ is automatically imposed through the Hamil-tonian Hψ

Λ .The first basic problems we should address are (1) construction and

uniqueness of dynamics and (2) identification of invariant or reversible mea-sures. For the problem (1) we refer to [149, 242, 243]. The case of Λ = Z

d

should also be considered.

Reversibility

Set Ω+(Λ) = RΛ+ and let µψ,J,+

Λ ∈ P(Ω+(Λ)) be the finite volume ϕ-Gibbsmeasure with hard wall and δ-pinning defined by (7.13) (with DN replacedby Λ and with boundary condition ψ), i.e.,

µψ,J,+Λ (dφ) =

1

Zψ,J,+Λ

e−HψΛ (φ)

∏x∈Λ

(cδ0(dφ(x)) + dφ+(x)

),

where c = eJ and dφ+(x) stands for the Lebesgue measure on R+. We denote

A(φ) = x ∈ Λ;φ(x) = 0,

and

B(φ) = x ∈ Λ;φ(x) > 0 ,

for φ ∈ Ω+(Λ). The sets A(φ) and B(φ) represent dry and wet regions associ-ated with the height variables φ, respectively. Then the space Ω+(Λ) can bedecomposed in two ways as

Ω+(Λ) =⋃

A⊂Λ

Ω0A =

⋃B⊂Λ

Ω+B

where Ω0A = φ ∈ Ω+(Λ);A(φ) = A and Ω+

B = φ ∈ Ω+(Λ);B(φ) = B,respectively.

Let us return to the SDEs (15.1)–(15.2). The corresponding generatorswhen φt moves on the region Ω+

B are given by

LB =∑x∈B

Lx

where

LxF (φ) = eH(φ) ∂

∂φ(x)

(e−H(φ) ∂F

∂φ(x)

), φ ∈ Ω+(Λ) ,

for F = F (φ) ∈ C2b (Ω+(Λ)). We simply denote H(φ) for Hψ

Λ (φ). We set

L = LΛ ,

246 T. Funaki

which is the free generator without pinning nor repulsion. In order to glue Ω+B

to Ω+B∪x at φ(x) = 0, x /∈ B, we need to introduce the boundary operator L

by

LF (φ, x) =1c

∂F

∂φ(x)− LxF (φ), φ ∈ Ω+(Λ), x ∈ A(φ) .

Note that LF (φ, x) can be rewritten as

LF (φ, x) = LB(φ)F (φ) +1c

∂F

∂φ(x)− LB(φ)∪xF (φ) ,

and compare this expression with the boundary operator Lf(x) defined bythe formula (7.2) of [149], p.204. The gluing operators for Ω+

B\C with Ω+B for

C ⊂ B is unnecessary if |C| ≥ 2, since the direct transitions between such twosets never occur (more precisely, occur with probability 0) for φt.

Lemma 15.1. For F ∈ C2b (Ω+(Λ)),

F (φt)− F (φ0)−∫ t

0

LF (φt) dt−∑x∈Λ

∫ t

0

cLF (φt, x) dt(x)

is a martingale.

Proof. Applying Ito’s formula, we have

dF (φt) =∑x∈Λ

∂F

∂φ(x)(φt) dφt(x) +

∑x∈Λ

∂2F

∂φ(x)2(φt)1(0,∞)(φt(x)) dt

=∑x∈Λ

1(0,∞)(φt(x))LxF (φt) dt+∑x∈Λ

∂F

∂φ(x)(φt) dt(x) + dmt

= LF (φt) dt+∑x∈Λ

cLF (φt, x) dt(x) + dmt ,

where

mt =∑x∈Λ

∫ t

0

1(0,∞)(φs(x)) ·√

2∂F

∂φ(x)(φs) dws(x)

is a martingale. Note that dt(x) = 0 if x ∈ B(φt) and dt = cdt(x) if x ∈A(φt), which follow from the conditions (15.2)-(b) and (c), respectively.

Lemma 15.2. Let F,G ∈ C2b,0(Ω

+(Λ)) and assume that F satisfies the“boundary conditions” LF (φ, x) = 0 for every φ ∈ Ω+(Λ) and x ∈ A(φ).Then, we have

∫Ω+(Λ)

GLF dµψ,J,+Λ = −

∫Ω+(Λ)

∑x∈B(φ)

∂F

∂φ(x)∂G

∂φ(x)dµψ,J,+

Λ . (15.3)

In particular, µψ,J,+Λ is reversible for (L, L)-diffusion, cf. [149], p.204.


Proof. The probability measure µψ,J,+Λ admits a decomposition (recall Lemma

7.6):µψ,J,+

Λ (·) =∑A⊂Λ

ν(A)µ0A(·) ,

where

µ0A(dφ) ≡ µ+

Ac(dφ) =1Z0

A

e−H(φ)∏

x∈Ac

dφ+(x)∏x∈A

δ0(dφ(x)) ,

ν(A) =c|A|Z0

A

Zψ,J,+Λ

= µψ,J,+Λ (A(φ) = A) .

If x ∈ Ac(≡ Λ \A), by the integration by parts,∫

Ω+(Λ)

GLxF µ0A(dφ) = −

∫Ω+(Λ)

∂G

∂φ(x)∂F

∂φ(x)µ0

A(dφ)

−∫

Ω+(Λ)

G · e−H ∂F

∂φ(x)

∣∣∣∣φ(x)=0

1Z0

A

∏y∈Ac

dφ+(y)∏y∈A

δ0(dφ(y))

and the second term can be further rewritten as

−∫

Ω+(Λ)

G∂F

∂φ(x)

Z0A∪xZ0

A

µ0A∪x(dφ) .

Therefore,∫

Ω+(Λ)

GLF µψ,J,+Λ (dφ) =

∑x∈Λ

∑A⊂Λ

ν(A)∫

Ω+(Λ)

GLxF µ0A(dφ)

=−∑x∈Λ

∑A:x∈Ac

ν(A)∫

Ω+(Λ)

∂G

∂φ(x)∂F

∂φ(x)µ0

A(dφ)

−∑x∈Λ

∑A:x∈Ac

ν(A)∫

Ω+(Λ)

G∂F

∂φ(x)

Z0A∪xZ0

A

µ0A∪x(dφ)

+∑x∈Λ

∑A:x∈A

ν(A)∫

Ω+(Λ)

GLxF µ0A(dφ) .

The first term coincides with the right hand side of (15.3). Setting A′ :=A∪x first and then writing A′ by A again, the second term can be rewrittenas

−∑x∈Λ

∑A:x∈A

ν(A \ x)∫

Ω+(Λ)

G∂F

∂φ(x)Z0

A

Z0A\x

µ0A(dφ)

= −∑x∈Λ

∑A:x∈A

c−1ν(A)∫

Ω+(Λ)

G∂F

∂φ(x)µ0

A(dφ) .

248 T. Funaki

Therefore, the sum of the second and the third terms becomes

∑x∈Λ

∑A:x∈A

ν(A)∫

Ω+(Λ)

G

LxF − c−1 ∂F

∂φ(x)

µ0

A(dφ)

= −∫

Ω+(Λ)

G∑

x∈A(φ)

LF (φ, x)µψ,J,+Λ (dφ) = 0

by the boundary conditions.

Dynamics With Volume Conservation Law

Mixing the ideas behind the SDEs (10.30) and (15.1), one can introduce dy-namics with conservation law by means of other SDEs

dφt(x) =1(0,∞)(φt(x))∆∂H

∂φ(x)(φt)dt

+ 1(0,∞)(φt(x)) ·√

2dw∆t (x)−∆d∆t (x), x ∈ Λ , (15.4)

subject to the conditions (15.2) with t(x) replaced by ∆t (x). The operator ∆in the first and the last terms of the right hand side is the discrete Laplacianacting on the variable x and the Brownian motions w∆

t (x);x ∈ Λ has acovariance structure

E[w∆t (x)w∆

s (y)] = −∆Λ(x, y) · t ∧ s ,

where ∆Λ(x, y), x, y ∈ Λ is the kernel of the discrete Laplacian.The first fundamental questions are the same as before, i.e., construction,

uniqueness of the dynamics and the identification of all reversible measures.Then, an interesting question for the dynamics is the derivation of the motionof the Winterbottom shape, cf. Sect. 7.3.

Let us consider the SDEs (15.4) taking Λ = TdN . The corresponding macro-

scopic height variables are defined by

hN (t, θ) =1N

∑x∈T

dN

φNαt(x)1B( xN , 1

N )(θ), θ ∈ Td .

To pick up the correct scaling Nα in time, we take a test function f = f(θ) ∈C∞(Td) and consider 〈hN (t), f〉 as in Sect. 10.1-(c). Then, its martingale termis given (with small error) by

√2

Nd+1

∑x∈T

dN

f( x

N

)1(0,∞)(φNαt(x))w∆

Nαt(x)

whose quadratic variational process is


= 2Nα−2d−2t∑

x,y∈TdN

f( x

N

)f( y

N

)

× 1(0,∞)(φNαt(x)) (−∆(x, y)) 1(0,∞)(φNαt(y))

∼ 2Nα−d−4t〈f1Dt, (−∆)f1Dt

〉

where Dt := the support of the Winterbottom shape (arising in the limit) attime t and the last ∆ is the continuum Laplacian on T

d. Therefore, one canexpect that the correct time scaling should be α = d+ 4.

16 Other Dynamic Models

16.1 Stochastic Lattice Gas and Free Boundary Problems

Particle Systems on Zd

At sufficiently low temperature, physical systems exhibit phase transition phe-nomena. Suppose that more than one phase coexist in the initial state of thesystem. Then a boundary will separate the phases, and this phase bound-ary would move according to a proper evolutional rule. Determination of themotion of the phase boundary is called the problem of phase separation, dy-namic phase transition, pattern formation, etc., and is analyzed in variouskinds of situations. So far, we have been mostly concerned with the ∇ϕ inter-face model. This subsection briefly summarizes approaches from the particlesystems.

Infinitely many particles are scattered over Zd and each of them performs a

random walk with jump rate determined from the surrounding configurationunder the exclusion rule that at most one particle can occupy each site ateach time. Such particle system is called the (stochastic) lattice gas or theKawasaki dynamics. The model for generation and extinction of particles ateach site is called the Glauber dynamics. The system taking all these effects(i.e., jumps, generations and extinctions) into account is called the Glauber-Kawasaki dynamics. Liggett [189, 190] are basic references for the interactingparticle systems.

Starting from the Glauber dynamics and others, Spohn [240] studied apattern formed after proper scaling and derived the motion by mean curva-ture. The argument there is rather heuristic, but contains several suggestiveconjectures. Presutti and others [69, 129, 163], derived (isotropic) motion bymean curvature for the interfaces from the Glauber-Kawasaki dynamics. Sinceone can derive the reaction-diffusion equation from the Glauber-Kawasaki dy-namics under proper hydrodynamic scaling limit (cf. [66, 70]) and the motionby mean curvature can be obtained from the reaction-diffusion equation undersingular limit (cf. Sect. 16.3), these results are thought of as the two scalingsare accomplished at once. In [67, 68, 164], the motion by mean curvature was

250 T. Funaki

derived from the Glauber dynamics corresponding to the Kac’s type poten-tial with long range interaction. The arising limits are nonrandom, while [69]treated the case that several random phase separation points appear. See areview paper by Giacomin et al. [134].

Free Boundary Problems

One phase or two phases Stefan free boundary problems are derived fromthe systems with two types of particles (e.g., A/B types) in which each typeperforms the Kawasaki dynamics possibly with different jump rates dependingon the types and, if different type of particles meet, both of them disappear.One dimensional case was discussed by Chayes and Swindle [57], and Funaki[114] extended their results to higher dimensions taking the effect of latentheat into account. See Quastel [217], Landim et al. [181], Ben Arous andRamırez [14], Gravner and Quastel [139] for investigations of systems withtwo types of particles. Komoriya [174] dealt with the case that, if differenttype of particles meet, they reflect. The result in [181] is applicable to thesystem of three types of particles called Potts model. See Ben Arous et al.[13] for internal DLA in a random environment.

16.2 Interacting Brownian Particles at Zero Temperature

The Wulff shape of interfaces for crystals is derived from the ferromagneticIsing model as we have mentioned in Sect. 1.1; see also Remark 6.7. This isa static result so that a natural question arises: Can one derive the motionof the Wulff shape from the corresponding dynamic system? The Kawasakidynamics, a system of interacting random walks on the lattice Z

d, is indeedthe stochastic evolutions which have the canonical Gibbs measure for theIsing model as an equilibrium state. Unfortunately, the Kawasaki dynamicshas a certain technical difficulty because of their discrete nature at this mo-ment. Instead, one can consider its continuum version, a system of interactingBrownian particles in R

d. As we have seen (in Sect. 10), for analyzing thesystem, the structure of the Gibbs measures corresponding to such dynamics(with infinitely many particles) needs to be clarified first. However, if d ≥ 2,this (and therefore the Wulff shape for this system) is not known for longexcept the simplest situation that the temperature of the system is zero. Weshall therefore study the system under the zero temperature limit hoping thatthis would serve as the first stage toward a deeper analysis on the motion ofthe Wulff shape. This subsection reviews the results of [118, 119]. The modelwas suggested by D. Ioffe.

Model and Problems

The time evolution of the positions of interacting Brownian N particles in Rd,

denoted by x(t) = (xi(t))Ni=1 ∈ (Rd)N , is prescribed by the SDEs


dxi(t) = −β2∇xi

H(x(t)) dt+ dwi(t), 1 ≤ i ≤ N , (16.1)

where β > 0 represents the inverse temperature of the system and (wi(t))Ni=1

is a family of independent d dimensional Brownian motions. The Hamiltonianis the sum of pairwise interactions between particles

H(x) =∑

1≤i<j≤N

U(xi − xj)

and the gradient ∇xiH(x) ≡

∑j =i∇U(xi − xj) is taken in the variable xi.

The potential U(x) = U(|x|), x ∈ Rd is radially symmetric and satisfies the

following three conditions:

(U1) U ∈ C30 (R),where U(−r) := U(r) ,

(U2) U attains a unique minimum at r = a > 0 such that U(a) = minr>0

U(r),

(U3) c = U ′′(a) > 0 .

The range of U is defined by b = infr > 0;U(s) = 0 for every s > r.The problem is to study the zero temperature limit for the system (16.1).

One can expect that the configurations x are crystallized (i.e., frozen) in anequal distance a as β → ∞. More precisely, we shall study the followingproperties.

• Microscopic behavior: The structure of crystallization is kept under thetime evolution except the isometric movement.

• Macroscopic behavior: The limits of translational and rotational motionsare characterized.

• Coagulation of several crystals for one dimensional system.

Rigid Crystals

A configuration z = (zi)Ni=1 ∈ (Rd)N is called a crystal if

|zi − zj | = a or |zi − zj | > b

for every i = j. (A certain condition is required on b for such z to exist.) Forθ ∈ SO(d) and η ∈ R

d, let ϕθ,η be an isometry on Rd or on (Rd)N defined

by ϕθ,η(y) = θy + η for y ∈ Rd and ϕθ,η(x) = (ϕθ,η(xi))N

i=1 for x = (xi)Ni=1.

A crystal z is called rigid if the energy H increases under any perturbativetransformations except isometries, i.e., if there exists δ > 0 such that

H(x) > H(z) for every x ∈M(δ) \M ,

where M ≡ Mz = ϕθ,η(z); θ ∈ SO(d), η ∈ Rd and M(δ) is a δ-

neighborhood of M in (Rd)N . The rigidity means that z has no internal

252 T. Funaki

degree of freedom except for the isometry. For example in 2 dimension, thethree vertices of equilateral triangle form a rigid crystal, but the four verticesof square do not. The rigid crystal is a local minimum of H by definition, butnot necessarily a global one.

We further introduce the notion of infinitesimal rigidity. Tangent spaceto M at z is defined by

Hz =Xz + h;X ∈ so(d), h ∈ R

d⊂ (Rd)N ,

whereXz+h = (Xzi+h)Ni=1 and so(d) = X : d×d real matrices;X+tX = 0

is the Lie algebra of SO(d). Let H⊥z be the orthogonal subspace to Hz in

(Rd)N . Note that, if δ > 0 is sufficiently small, x ∈ M(δ) admits a uniquedecomposition x = z(x) + h(x) such that z(x) ∈ M and h(x) ∈ H⊥

z(x). TheHessian of H on M is given by

E(h) ≡ Ez(h) =c

a2

∑〈i,j〉

(hi − hj , zi − zj)2 ,

for h = (hi)Ni=1 ∈ (Rd)N . The sum 〈i, j〉 is taken over all pairs i, j such that

|zi − zj | = a. A crystal z is called infinitesimally rigid if

E(h) = 0 ⇐⇒ h ∈ Hz

i.e., the Hessian is nondegenerate to the orthogonal direction.Note that “h ∈ Hz ⇒ E(h) = 0” is obvious, in fact, the translational

invariance of H implies E(h) = 0, while its rotational invariance impliesE(Xz) = 0. The infinitesimal rigidity implies the rigidity.

Microscopic Shape Theorem

The basic scaling parameter is the ratio ε = (microscopic spatial unitlength)/(macroscopic spatial unit length) so that (εxi)N

i=1 is the macroscopiccorrespondence to the microscopic configuration x = (xi)N

i=1. The number ofparticles and the inverse temperature will be rescaled in ε: N = N(ε), β =β(ε). Let a sequence of infinitesimally rigid crystals z(ε) = (z(ε)

i )Ni=1; 0 < ε <

1 be given. For sufficiently small c > 0, the c-neighborhood of M = Mz(ε) isdefined by M∇(c) ≡M∇,N (c) = x ∈ (Rd)N ; ‖∇h(x)‖∞ ≤ c, where h(x) ∈H⊥

z(x) was determined by decomposing x and ‖∇h‖∞ = sup〈i,j〉 |hi − hj |. Weintroduce the macroscopic time change for the solution x(t) of the SDEs (16.1)

x(ε)(t) = x(ε−κt), κ = d+ 2 .

Theorem 16.1. [118] Let c(ε) ↓ 0 (as ε ↓ 0) be given and assume thatx(ε)(0) ∈ M∇(c′(ε)) for some c′(ε) c(ε) (i.e., the initial configuration isnearly an infinitesimally rigid crystal) and β = β(ε) →∞ sufficiently fast asε ↓ 0. Then, we have that


limε↓0

P (σ(ε) ≥ t) = 1 ,

for every t > 0, where σ(ε) = inft ≥ 0;x(ε)(t) /∈ M∇(c(ε)). In other words,asymptotically with probability one x(ε)(t) keeps its rigidly crystallized shapewithin fluctuations c(ε).

For the proof of the theorem, Lyapunov type argument is applied combin-ing with a spectral gap estimate for E(h).

Motion of a Macroscopic Body

We say that a sequence x(ε) = (x(ε)i )N

i=1, N = N(ε) of configurations has amacroscopic density function ρ(y), y ∈ R

d if

εdN∑

i=1

δεx

(ε)i

(dy) ⇒ ρ(y) dy

weakly as ε ↓ 0. The initial configurations x(ε)(0) = z(ε)(= (z(ε)i )N

i=1) areassumed to be infinitesimally rigid crystals with macroscopic density functionρ(y) and |z(ε)

i | ≤ Rε−1 for all i and ε and for some R > 0. The particles’number behaves as N = N(ε) ∼ ρε−d, where ρ =

∫Rd ρ(y) dy. We may assume,

by shifting the system if necessary, that the body is centered:∫

Rd yρ(y) dy = 0.Let Q = (qαβ)1≤α,β≤d be the matrix defined by

qαβ =∫

Rd

yαyβρ(y) dy ,

where yα denotes the αth component of y. We may assume by rotating thesystem that Q is diagonal. The sum

vαβ = qαα + qββ

is called moments of inertia.

Theorem 16.2. [118] Assume that β = β(ε) → ∞ sufficiently fast as ε ↓ 0.Then, x(ε)(t) has a macroscopic density function ρt(y), which is congruent toρ(y), i.e., ρt(y) = ρ

(ϕ−1

θ(t),η(t)(y)). The translational and rotational motions

(η(t), θ(t)) of the limit body are random and characterized as follows:

(1) η(t) and θ(t) are mutually independent.(2) η(t) = (d dimensional Brownian motion)/

√ρ

(3) θ(t) is a Brownian motion on SO(d) which is a solution of the SDE ofStratonovich’s type

dθ(t) = θ(t) dm(t), θ(0) = I

where m(t) = (mαβ(t)) is an so(d)-valued Brownian motion such thatthe components mαβ(t);α < β in the upper half of the matrix m(t)are mutually independent and mαβ(t) = (one dimensional Brownianmotion)/

√vαβ.

254 T. Funaki

Coagulation in One Dimension

Consider the SDEs (16.1) in R taking β = ε−γ , γ > 0:

dxi(t) = −ε−γ

2

∑j =i

U ′(xi(t)− xj(t)) dt+ dwi(t), 1 ≤ i ≤ N .

Theorems 16.1 and 16.2 have dealt with motion of single macroscopic body.In one dimension one can establish coagulation of several bodies (bodies arecalled rods in one dimension). We need additional assumptions on U such thatthe well at a is deep and located away from 0.

Since the rods evolve independently until they meet (i.e., until the timewhen the microscopic distance between two rods becomes b) and since theanalysis of multiple rods can be essentially reduced to the two rods case, weassume the following two conditions on the initial configuration x(0):

(1) x(0) = x(1)(0)∪x(2)(0) consists of two nearly rigid crystals (called chainsin one dimension) with particles’ numbers N1 = [ρ1ε

−1], N2 = [ρ2ε−1]

and fluctuation εν , ν > 1/2, i.e., x()(0) ∈ M∇,N(εν), = 1, 2, whereρ1, ρ2 > 0.

(2) The distance of these two chains (that between the right most particle ofx(1)(0) and the left most one of x(2)(0)) is b.

Theorem 16.3. [119] Let x(ε)(t) = x(ε−3t) be the macroscopically timechanged process and assume the above condition on x(0). Take another ν′ > 0and suppose that γ > max4, 2ν′ + 3. Then, for every δ > 0, we have

limε↓0

Px(ε)(t) ∈M∇,N (εν

′) for some t ≤ ε1−δ

= 1 ,

where N = N1 +N2.

This theorem claims that two rods in x(ε)(t) coagulate and form a singlerod within a very short time ε1−δ. After the coagulation x(ε)(t) moves as asingle rod as we have seen in Theorem 16.1.

16.3 Singular Limits for Stochastic Reaction-Diffusion Equations

In this subsection we refer to the results concerning a reaction-diffusion equa-tion with additive noise, i.e., (16.2) below. See the survey paper [113] for moredetails. The equation (16.2) involves a small parameter ε > 0 representing thetemperature of the system; ε = β−1 in the last subsection. The spatial variablex is already macroscopic. Taking the zero temperature limit ε ↓ 0 again, ora singular limit in mathematical terminology, we expect that the solution ofthe equation converges under an appropriate time change to a point at whichthe potential appearing in the nonlinear reaction term is minimized. Pointsat which the potential attains its minimum, simply called minimal points or


bottoms, correspond to phases in the physical context. If we have more thanone phase, phase separation will occur.

Reaction-diffusion equations have been used to describe various kinds ofphenomena, including dynamical phase transitions. In connection with mi-croscopic particle systems, we note that the equation treated here is consid-ered to describe an intermediate (mesoscopic) level, between microscopic andmacroscopic. Indeed, the reaction-diffusion equation can be derived from theGlauber-Kawasaki dynamics by means of the hydrodynamic limit (as we havereferred in Sect. 16.1), and the noise term naturally appears in the fluctuationproblem.

Stochastic Reaction-Diffusion Equations

The following reaction-diffusion equation with noise is considered in this sub-section:

∂u

∂t= ∆u+

1εf(u) + wε(t, x), t > 0, x ∈ D , (16.2)

where wε(t, x) is a noise depending on ε > 0 and D is a domain in Rd. We

assume that the reaction term f ∈ C∞(R) is bistable:

∃u∗ ∈ (−1, 1) such that f(±1) = f(u∗) = 0, f ′(±1) < 0, f ′(u∗) > 0 ,

and fulfills a technical condition:

∃C, p > 0 such that |f(u)| ≤ C(1 + |u|p) and supuf ′(u) <∞ ,

which ensures the existence and uniqueness of the solution of (16.2) for noiseswe shall treat. Two values ±1 are stable points of f and u∗ is an unstablepoint. A function F satisfying f = −F ′ is the associated potential. Our goalis to study the behavior of the solution u = uε(t, x) of (16.2) as ε ↓ 0. Whenthe reaction term dominates the noise term, multiplying both sides by ε yields

limε↓0

f(uε) = 0

formally, while the unstable solution is not considered to appear in the limit.So we guess that

limε↓0

uε(t, x) = +1 or − 1

depending on (t, x). In other words, a random boundary separating +1 and−1 might appear. To find the motion of the boundary is the main problemwhich we discuss here. The results in the absence of noise (i.e., in the casewε = 0) are summarized in [113].

256 T. Funaki

Singular Limits

(a) The Case where d = 1,D = R and wε(t, x) = εγa(x)wh(t, x)

Here γ > 0, and a ∈ C20 (R) is a function representing the magnitude of the

noise at each point. The condition that a has a compact support is introducedso that the problem can be localized and the boundary conditions uε(t,±∞) =±1 of the equation (16.2) at x = ±∞ hold. wh, 1/2 ≤ h ≤ 1, is a self-similarGaussian noise, i.e., the covariance structure of the noise is formally given by

〈wh(t, x)wh(s, y)〉 = δ0(t− s)Qh(x− y) ,

where δ0 is the δ-function at 0 and Qh is the Riesz potential of order 2h− 1:

Qh(x) =

h(2h− 1)|x|2h−2

δ0(x)

, 1/2 < h ≤ 1 ,, h = 1/2 .

In particular, w1/2(t, x) is the space-time white noise and w1(t, x) = w(t) isthe one parameter white noise independent of the spatial variable. Under theabove conditions, the stochastic PDE (16.2) has a unique solution uε(t, x) inthe sense of mild solutions or generalized functions. Although the solution isnot differentiable, it is Holder continuous:

uε(t, x) ∈ ∩δ>0Ch2 −δ,h−δ((0,∞)× R), a.s.

Under suitable assumptions on the initial condition and an assumption ofsymmetry of the reaction term, i.e., f is odd: f(u) = −f(−u), we can prove thefollowing theorem. The function m = m(y) is a standing wave solution to thereaction-diffusion equation, i.e. a solution of the ODE (ordinary differentialequation) on R

m′′ + f(m) = 0, y ∈ R ,

satisfying m(±∞) = ±1. Since m is determined uniquely up to parallel dis-placement, we normalize it as m(0) = 0.

Theorem 16.4. [110, 112] There exists γ(h) > 0 such that for all γ ≥ γ(h)

uε(t, x) := uε(ε−2γ−ht, x) =⇒ε↓0

χξt(x) ,

where χξ(x) = 1x>ξ − 1x<ξ. The motion of the phase separating point ξtthat appeared in the limit is governed by the SDE

dξt = α1a(ξt)dBt + α2a(ξt)a′(ξt)dt , (16.3)

where Bt is a one dimensional Brownian motion and the constants α1 = α1(h)and α2 = α2(h) are given by


α21 =

1‖m′‖4L2(R)

∫R2m′(x)m′(y)Qh(x− y) dxdy ,

α2 = − 1‖m′‖2L2(R)

∫ ∞

0

dt

∫R3xp(t, x, z)p(t, y, z)

× f ′′(m(z))m′(z)Qh(x− y) dxdydz ,

respectively. The function p(t, x, y) is the fundamental solution of the lin-earized operator ∂/∂t− ∂2/∂y2 + f ′(m(y)).

A related result was obtained in the microscopic situation by Presutti et al.[36]; see also [35]. In physics, the case h = 1/2 (that is, the space-time whitenoise) is important, and it is of particular interest to identify the diffusioncoefficient governing the motion of the random boundary in the limit. We seefrom Theorem 16.4 that the diffusion coefficient (also called mobility) of thelimit SDE is

α1(1/2)2 = ‖m′‖−2L2(R)

if h = 1/2. Here ‖m′‖2L2(R) is called the surface tension in this model. Thisresult is consistent with conjectures of Kawasaki and Ohta [166] and Spohn[240].

(b) The Case where d = 2 and D is a Bounded Domain in R2 with Smooth

Boundary

In (a) we considered the one dimensional case, while results for higher dimen-sional case, especially the two dimensional case, are described here. Consider(16.2) with Neumann boundary condition: ∂u/∂n = 0 (x ∈ ∂D). Here the as-sumption that f is an odd function is not necessary; instead, we merely assumeA(f) ≡

∫ 1

−1f(u) du = 0. For simplicity the noise is assumed to be independent

of the spatial variable, depending on only the time variable: wε(t, x) = ξεt/√ε,

where ξεt = ε−γξ(ε−2γt), 0 < γ < 1/3, and ξ(t) ∈ C1(R+), a.s. is a mean 0 sta-

tionary process with the strong mixing property. Roughly speaking, ξεt ⇒ αwt

as ε ↓ 0, where wt is a one dimensional Brownian motion and α is a constantgiven by the Green-Kubo formula

α =

√2∫ ∞

0

E[ξ(0)ξ(t)] dt .

Then, under suitable initial conditions, the following holds for the solutionuε(t, x) of (16.2).

Theorem 16.5. [115] As long as the phase separation curve Γt in the limitis strictly convex and does not touch the boundary ∂D, we have

uε(t, x) =⇒ε↓0

χΓt(x) ,

258 T. Funaki

where the motion of the curve Γt is given by a random perturbation of thecurvature flow:

V = κ+ (c0α)wt . (16.4)

Here, V is the inner normal speed of Γt, κ denotes the curvature of Γt, andc0 is given by

c0 =√

2/∫ 1

−1

du

√∫ 1

u

f(v) dv .

Remark 16.1. In the case where wε(t, x) is the space-time white noise andD = R

3, Kawasaki and Ohta [166, 209] considered (16.2) and derived a ran-dom mean curvature flow describing the motion of the phase boundary, callingthe limit the drumhead model. However, equation (16.2) does not have amathematical meaning in dimensions greater than one, and the proof of exis-tence of the solution is impossible. Other physical papers on the derivation ofkinetic equations of interfaces or analysis of their equilibrium states include[4, 82, 211]. Among the literatures concerning interface curves we mention[142], and its probabilistic version is discussed in [162].

Remark 16.2. (1) Problems discussed in Sects. 16.2 and 16.3 can be viewedas convergence of solutions of SDEs in an infinite (or large) dimensional spacetoward a submanifold where the minimum energy is attained. Such problemsare treated in [120, 165] in finite dimensional spaces.(2) A related problem of a large deviation principle is studied in [25], etc.,especially for the case where minimal points of the rate functional are notunique, and [97] discussed the case of stochastic PDEs for which exactly twominimal points exist.

Remark 16.3. The zero temperature limit and the metastable behavior of theGlauber dynamics (under the periodic boundary conditions) or the Kawasakidynamics (in an infinite gas reservoir) were investigated by Ben Arous andCerf [11] and by den Hollander et al. [65], respectively.

16.4 Limit Shape of Random Young Diagrams

An asymptotic shape of typical Young diagrams is studied by Vershik [250],cf. Kerov [169]. The similarity of this approach to the Wulff problem in theIsing model is discussed by Shlosman [233]. The related stochastic dynamicsis analyzed by several authors; especially, a randomly growing Young diagramby Johansson.

Model

For each N ∈ N, an array p = (n1, n2, . . . , nk) of positive integers is called apartition of N if n1 ≥ n2 ≥ · · · ≥ nk and

∑ki=1 ni = N for some 1 ≤ k ≤ N .


The number of groups of size i in the partition p is defined by ri = j;nj = i,1 ≤ i ≤ N . Then a Young diagram is associated with p by

ϕp(x) =∑i>x

ri, x ≥ 0 ,

which is the number of groups with size larger than x. The function ϕp is anonincreasing step function satisfying ϕp(0) = k, ϕp(x) = 0 for x ≥ n1 and∫ ∞0ϕp(x) dx = N . More precisely, the region A = Ap surrounded by the step

function ϕp and the boundary ∂R2+ of quadrant is called the Young diagram;

note that each A is built by piling up unit cubes in R2+.

Let PN be the set of all partitions p of N . Vershik [250] introduced sev-eral kinds of statistics (like finite volume Gibbs measures) on the set PN ,which leads to random Young diagrams. The most typical and natural oneis the uniform probability on it, sometimes called Bose statistics, defined byµN

B (p) = 1/PN for every p ∈ PN . Fermi statistics is the uniform probabilityµN

F on the set of strict partitions p such that n1 > n2 > · · · > nk.

Scaling and Law of Large Numbers

Introduce the scaling for Young diagrams in two dimensions as

ϕNp (x) =

1√Nϕp(√Nx), x ≥ 0 ,

so that∫ ∞0ϕN

p (x) dx = 1. Then, as N → ∞, the LLN holds and the limitshapes of ϕN

p (x) are given by curves in R2+:

ΦB(x) = − 1α

log(1− e−αx

), α =

π√6,

ΦF (x) =1β

log(1 + e−βx

), β =

π√12

,

under µNB and µN

F , respectively; see [250]. Three dimensional case called sky-scraper problem is discussed by Cerf and Kenyon [51]; see also [167]. The limitshape is a surface in R

3+, which has flat pieces (called facets) embedded on

∂R3+ (being consisted of three quadrants R

2+).

Large Deviation

The corresponding LDP is established by Dembo, Vershik and Zeitouni [71]for µN

B and µNF in two dimension. The speed of the LDP is

√N and its rate

functionals IB and IF are explicitly specified. In particular, the limit shapesΦB and ΦF are the solutions of the variational problems for IB and IF , re-spectively. Shlosman [233] discussed the analogy between the Wulff problemarising from the Ising model and the variational problem for IB including

260 T. Funaki

higher dimensional case. Especially, he observes that the latter problem hasa similar solution to the Wulff construction.

For the proof, the conditioning approach of Fristedt is effectively used.For instance, for Bose statistics, we have the following representation. LetZjj=1,2,...,N be a sequence of independent Z+-valued random variables withgeometric distributions of parameter qj , respectively. Then, under condition-ing that

∑Nj=1 jZj = N , Zjj has the same distribution as rjj distributed

under µNB for every q ∈ (0, 1). This corresponds to the equivalence of ensemble

in statistical mechanics.

Central Limit Theorem

The corresponding CLT is established by Pittel [214]. In fact, under µNB ,

VN (x) = N−1/4xϕp

(√Nα−1 log 1/(1− x)

)−√Nα−1 log 1/x

, x ∈ [0, 1]

weakly converges to a certain Gaussian process V (x).

Dynamic Approach

Let ΩN be the set of all r = (r1, r2, . . . , rN ) determined from partitions p ∈PN . Durrett, Granovsky and Gueron [94] introduced a stochastic evolutionof rt, t ≥ 0 called the coagulation-fragmentation process for each N . It is aMarkov chain on ΩN determined by the rates of coagulation of two groups ofsizes i and j into a single group of size i+ j and those of fragmentation of agroup into two groups. The reversibility of the process rt is characterized undera proper assumption on the rates (see also [140]) and an explicit formula of theinvariant measure is found. It has a conditioning representation of Fristedt’stype but the distributions of random variables Zjj should be replaced fromgeometric to Poissonian.

Norris [206, 207] proved the LLN (a kind of mean field limit) for coagu-lation process (without fragmentation) under a suitable scaling as N → ∞.The limit is described by the Smoluchowski’s coagulation equation, which isa nonlinear equation associated with an infinite system. From the view pointof the Young diagrams, his scaling is in a different regime, i.e., the verticalsize behaves as O(N), while the horizontal one is kept at O(1) as N →∞.

Randomly Growing Young Diagram

Let Y = ∪∞N=1ϕp; p ∈ PN be the set of all Young diagrams. A randomly

growing Young diagram (sometimes called the corner growth model) is aMarkov chain ϕ(t), t = 1, 2, . . . on Y defined by the following transition rule.We call a unit cube Q ⊂ R

2+ a corner outside of ϕ ∈ Y if the left and lower

sides of Q are both on ϕ or on ∂R2+. Then, ϕ(t+1) is obtained by picking each


corner outside of ϕ(t) independently with probability p ∈ (0, 1) and addingthose corners to ϕ(t). The explicit form of the asymptotic shape ϕ0 of 1

tϕ(t)as t→∞ is known. This is a result of LLN type. Johansson [158] (see also [7]and a review paper [159]) further studied the fluctuation of ϕ(t) around tϕ0

and found that the limit distribution is described by the Tracy-Widom largesteigenvalue distribution arising in the random matrix theory. The model withcontinuous time parameter ϕ(t); t ≥ 0 can be discussed in a parallel way.

The randomly growing Young diagram is equivalent to the totally asym-metric one dimensional exclusion process (first observed by Rost, see also[90, 215]), the last-passage directed percolation model and it is related to theHammersley’s model [157], domino tilings and dimer model [160]. The distrib-ution of the last-passage time in the directed percolation model, in particular,behaves asymptotically as N →∞ as

c1N + c2N1/3X ,

where c1 and c2 are certain positive constants, N is the size of the system andX is a random variable with Tracy-Widom distribution; [99, 158]. In recentyears, it is realized that this type of asymptotic behavior arises universally inseveral related models. Prahofer and Spohn [216] found a connection in theso-called polynuclear growth (PNG) model studying the fluctuations from itslimit semi-circular shape.

Results for the (undirected) first-passage percolation, including an asymp-totic shape and fluctuations of percolated region, are reviewed by [148]. See[197] for the directed first-passage and last-passage percolation models.

16.5 Growing Interfaces

Krug and Spohn [177] give a nice review on growing interfaces, includingEden models, SOS models, ballistic deposition models, KPZ equation, directedpolymer, percolations, DLA (nonlocal model) and others.

SOS Dynamics

The SOS type dynamics is the ϕ-dynamics φ = φt(x) with values in Z.If φt(x) is nondecreasing in t for each x, it defines the growing interfaces orsometimes called marching soldiers. In one dimension, under the constraintthat ∇φ(x) = 0 or −1 on the height differences, the associated ∇ϕ-dynamicsis the totally asymmetric exclusion processes as we have pointed out in thelast subsection. For more general case of height differences ∇φ(x) ∈ Z, ∇ϕ-dynamics is the zero-range processes. The hyperbolic space-time scaling limithas been studied by Seppalainen and Rezakhanlou; cf. Remark 13.1. Fritz[104, 105] extended the Tartar-Murat theory of compensated compactness toprove the hydrodynamic limit (LLN); cf. [182]. Varadhan [248] established theLDP for the asymmetric exclusion process.

262 T. Funaki

KPZ Equation

A model for interfaces growing by depositions of particles, which randomly fallfrom the ambient atmosphere on the surface, was introduced by Kardar, Parisiand Zhang [161]. The dynamics of the surface is described by a stochastic PDEcalled KPZ equation:

∂

∂th(t, x) = ∆h(t, x) + |∇h(t, x)|2 + w(t, x) ,

where h(t, x) is the height of the surface and w(t, x) is the space-time whitenoise; see also [176].

References

1. D. Abraham, Surface structures and phase transitions – exact results, In:Phase Transitions and Critical Phenomena, 10 (1986), pp. 1–74, AcademicPress. 172

2. D. Abraham, P. Collet, J. De Coninck and F. Dunlop, Langevin dy-namics of an interface near a wall, J. Statis. Phys., 61 (1990), pp. 509–532. 232

3. K.S. Alexander, Cube-root boundary fluctuations for droplets in random clus-ter models, Commun. Math. Phys., 224 (2001), pp. 733–781. 172

4. S.M. Allen and J.W. Cahn, A microscopic theory for antiphase boundarymotion and its application to antiphase domain coarsening, Acta Metall., 27(1979), pp. 1085–1095. 258

5. H.W. Alt and L.A. Caffarelli, Existence and regularity for a minimumproblem with free boundary, J. Reine Angew. Math., 325 (1981), pp. 105–144. 112, 165, 172

6. H.W. Alt, L.A. Caffarelli and A. Friedman, Variational problems withtwo phases and their free boundaries, Trans. Amer. Math. Soc., 282 (1984),pp. 431–461. 112, 172

7. J. Baik, P. Deift and K. Johansson, On the distribution of the length of thelongest increasing subsequence of random permutations, J. Amer. Math. Soc.,12 (1999), pp. 1119–1178. 261

8. B. Bakry and M. Emery, Diffusions hypercontractives, In: Seminaire deProbabilites XIX, Lect. Notes Math., 1123 (1985), Springer, pp. 179–206. 234

9. J.M. Ball, A version of the fundamental theorem for Young measures, in PDEsand continuum models of phase transitions (Nice, 1988), Springer, LectureNotes in Physics, 344 (1989), pp. 207–215. 187

10. V. Barbu, Nonlinear Semigroups and Differential Equations in Banach Spaces,Noordhoff, 1976. 221

11. G. Ben Arous and R. Cerf, Metastability of the three-dimensional Isingmodel on a torus at very low temperatures, Electron. J. Probab., 1 (1996),approx. 55 pp. (electronic). 258

12. G. Ben Arous and J.-D. Deuschel, The construction of the d + 1-dimensional Gaussian droplet, Commun. Math. Phys., 179 (1996), pp. 467–488. 111, 164, 168

13. G. Ben Arous, J. Quastel and A.F. Ramırez, Internal DLA in a randomenvironment, Ann. Inst. Henri Poincare, 39 (2003), pp. 301–324. 250


14. G. Ben Arous and A.F. Ramırez, Asymptotic survival probabilities in therandom saturation process, Ann. Probab., 28 (2000), pp. 1470–1527. 250

15. G. Benfatto, E. Presutti and M. Pulvirenti, DLR measures for one-dimensional harmonic systems, Z. Wahr. verw. Gebiete, 41 (1978), pp. 305–312. 131, 135

16. A. Bensoussan and J.L. Lions, Applications of Variational Inequalities inStochastic Control, North-Holland, 1982. 240

17. L. Bertini, C. Landim and S. Olla, Derivation of Cahn-Hilliard equationsfrom Ginzburg-Landau models, J. Statis. Phys., 88 (1997), pp. 365–381. 235

18. D. Bertacchi and G. Giacomin, Enhanced interface repulsion from quenchedhard-wall randomness, Probab. Theory Relat. Fields, 124 (2002), pp. 487–516. 196, 243

19. D. Bertacchi and G. Giacomin, Wall repulsion and mutual interface repul-sion: an harmonic crystal model in high dimensions, Stoch. Proc. Appl., 110(2004), pp. 45–66. 196

20. T. Bodineau, The Wulff construction in three and more dimensions, Commun.Math. Phys., 207 (1999), pp. 197–229. 110

21. T. Bodineau, G. Giacomin and Y. Velenik, On entropic reduction of fluc-tuations, J. Statist. Phys., 102 (2001), pp. 1439–1445. 195

22. T. Bodineau, D. Ioffe and Y. Velenik, Rigorous probabilistic analysis ofequilibrium crystal shapes, J. Math. Phys., 41 (2000), pp. 1033–1098. 110, 172

23. T. Bodineau, D. Ioffe and Y. Velenik, Winterbottom construction for fi-nite range ferromagnetic models: an L1-approach, J. Statist. Phys., 105 (2001),pp. 93–131. 203

24. D. Boivin and Y. Derriennic, The ergodic theorem for additive cocycles ofZ

d or Rd, Ergod. Th. Dynam. Sys., 11 (1991), pp. 19–39. 212

25. E. Bolthausen, Laplace approximations for sums of independent random vec-tors, Part II. Degenerate maxima and manifolds of maxima, Probab. TheoryRelat. Fields, 76 (1987), pp. 167–206. 258

26. E. Bolthausen, Large deviations and perturbations of random walks and ran-dom surfaces, European Congress of Mathematics, Vol. I (Budapest, 1996),pp. 108–120, Progr. Math., 168, Birkhauser, 1998. 202

27. E. Bolthausen and J.-D. Deuschel, Critical large deviations for Gaussianfields in the phase transition regime I, Ann. Probab., 21 (1993), pp. 1876–1920. 189

28. E. Bolthausen, J.-D. Deuschel and G. Giacomin, Entropic repulsion andthe maximum of two dimensional harmonic crystal, Ann. Probab., 29 (2001),pp. 1670–1692. 195

29. E. Bolthausen, J.-D. Deuschel and O. Zeitouni, Entropic repulsion of thelattice free field, Commun. Math. Phys., 170 (1995), pp. 417–443 (Erratum:ibid, 209 (2000), pp. 547–548). 112, 196

30. E. Bolthausen, J.-D. Deuschel and O. Zeitouni, Absence of a wet-ting transition for a pinned harmonic crystal in dimensions three and larger,J. Math. Phys., 41 (2000), pp. 1211–1223. 112, 199, 200, 202

31. E. Bolthausen and D. Ioffe, Harmonic crystal on the wall: a microscopicapproach, Commun. Math. Phys., 187 (1997), pp. 523–566. 112, 170, 199, 202

32. E. Bolthausen and Y. Velenik, Critical behavior of the massless free fieldat the depinning transition, Commun. Math. Phys., 223 (2001), pp. 161–203. 112, 197, 198

33. H.J. Brascamp and E.H. Lieb, On extensions of the Brunn-Minkowski andPrekopa-Leindler theorems, including inequalities for log concave functions,and with an application to the diffusion equation, J. Funct. Anal., 22 (1976),pp. 366–389. 146

264 T. Funaki

34. H.J. Brascamp, E.H. Lieb and J.L. Lebowitz, The statistical mechanics ofanharmonic lattices, Bull. Int. Statis. Inst., 1975, pp. 393–404. 146

35. S. Brassesco and P. Butta, Interface fluctuations for the D = 1 stochasticGinzburg-Landau equation with nonsymmetric reaction term, J. Statist. Phys.,93 (1998), pp. 1111–1142. 257

36. S. Brassesco, A. De Masi and E. Presutti, Brownian fluctuations of theinstanton in the d = 1 Ginzburg-Landau equation with noise, Ann. Inst. HenriPoincare, 31 (1995), pp. 81–118. 257

37. H. Brezis, Operateurs Maximaux Monotones et Semi-Groupes de Contractionsdans les Espaces de Hilbert, North-Holland, Amsterdam, 1973. 221

38. J. Bricmont, A. El Mellouki and J. Frohlich, Random surfaces in statis-tical mechanics: roughening, rounding, wetting,. . . , J. Statist. Phys., 42 (1986),pp. 743–798. 196

39. J. Bricmont, J.-R. Fontaine, J.L. Lebowitz and T. Spencer, Latticesystems with a continuous symmetry, I. Perturbation theory for unboundedspins, Commun. Math. Phys., 78 (1980), pp. 281–302. 139, 151

40. J. Bricmont, J.-R. Fontaine, J.L. Lebowitz and T. Spencer, Lattice sys-tems with a continuous symmetry, II. Decay of correlations, Commun. Math.Phys., 78 (1981), pp. 363–371. 151

41. D. Brydges, J. Frohlich and T. Spencer, The random walk representationof classical spin systems and correlation inequalities, Commun. Math. Phys.,83 (1982), pp. 123–150. 198

42. D. Brydges and H.-T. Yau, Grad ϕ perturbations of massless Gaussian fields,Commun. Math. Phys., 129 (1990), pp. 351–392. 151

43. L.A. Caffarelli and J.L. Vazquez, A free-boundary problem for the heatequation arising in flame propagation, Trans. Amer. Math. Soc., 347 (1995),pp. 411–441. 244

44. L.A. Caffarelli, C. Lederman and N. Wolanski, Uniform estimates andlimits for a two phase parabolic singular perturbation problem, Indiana Univ.Math. J., 46 (1997), pp. 453–489. 244

45. L.A. Caffarelli, C. Lederman and N. Wolanski, Pointwise and viscositysolutions for the limit of a two phase parabolic singular perturbation problem,Indiana Univ. Math. J., 46 (1997), pp. 719–740. 244

46. P. Caputo and J.-D. Deuschel, Large deviations and variational principlefor harmonic crystals, Commun. Math. Phys., 209 (2000), pp. 595–632. 191

47. P. Caputo and J.-D. Deuschel, Critical large deviations in harmonic crys-tals with long-range interactions, Ann. Probab., 29 (2001), pp. 242–287. 191

48. P. Caputo and D. Ioffe, Finite volume approximation of the effective diffu-sion matrix: the case of independent bond disorder, Ann. Inst. H. Poincare, 39(2003), pp. 505–525. 158

49. P. Caputo and Y. Velenik, A note on wetting transition for gradient fields,Stoch. Proc. Appl., 87 (2000), pp. 107–113. 112, 176, 199, 200

50. E.A. Carlen, S. Kusuoka and D.W. Stroock, Upper bounds for symmetricMarkov transition functions, Ann. Inst. H. Poincare, 23 (1987), pp. 245–287. 137

51. R. Cerf and R. Kenyon, The low-temperature expansion of the Wulff crystalin the 3D Ising model, Commun. Math. Phys., 222 (2001), pp. 147–179. 172, 259

52. R. Cerf and A. Pisztora, On the Wulff crystal in the Ising model, Ann.Probab., 28 (2000), pp. 947–1017. 110, 172

53. R. Cerf and A. Pisztora, Phase coexistence in Ising, Potts and percolationmodels, Ann. Inst. H. Poincare, 37 (2001), pp. 643–724. 172


54. F. Cesi and F. Martinelli, On the layering transition of an SOS surface in-teracting with a wall, I. equilibrium results, J. Statis. Phys., 82 (1996), pp. 823–913. 118

55. F. Cesi and F. Martinelli, On the layering transition of an SOS surfaceinteracting with a wall, II. the Glauber dynamics, Commun. Math. Phys., 177(1996), pp. 173–201. 118

56. C.C. Chang and H.-T. Yau, Fluctuations of one dimensional Ginzburg-Landau models in nonequilibrium, Commun. Math. Phys., 145 (1992), pp. 209–239. 231

57. L. Chayes and G. Swindle, Hydrodynamic limits for one-dimensional parti-cle systems with moving boundaries, Ann. Probab., 24 (1996), pp. 559–598. 250

58. H. Cohn, R. Kenyon and J. Propp, A variational principle for dominotilings, J. Amer. Math. Soc., 14 (2001), pp. 297–346. 172

59. P. Collet, F. Dunlop, J. Fritz and T. Gobron, Langevin dynamics of asemi-infinite interface, Markov Processes Rel. Fields, 3 (1997), pp. 261–274. 232

60. O. Daviaud, Extremes of the discrete two-dimensional Gaussian free field,math.PR/0406609, preprint, 2004. 195

61. J. De Coninck, F. Dunlop and V. Rivasseau, On the microscopic validityof the Wulff construction and of the generalized Young equation, Commun.Math. Phys., 121 (1989), pp. 401–419. 199, 202

62. J. De Coninck, S. Miracle-Sole and J. Ruiz, Rigorous generalizationof Young’s law for heterogeneous and rough substrates, J. Statis. Phys., 111(2003), pp. 107–127. 202

63. T. Delmotte and J.-D. Deuschel, On estimating the derivatives of sym-metric diffusions in stationary random environment, preprint, 2002. 149

64. T. Delmotte and J.-D. Deuschel, Algebraic L2 convergence rates for ∇φinterface models, preprint, 2003. 151, 216

65. F. den Hollander, F.R. Nardi, E. Olivieri and E. Scoppola, Dropletgrowth for three-dimensional Kawasaki dynamics, Probab. Theory Relat.Fields, 125 (2003), pp. 153–194. 258

66. A. De Masi, P.A. Ferrari and J.L. Lebowitz, Reaction diffusion equationsfor interacting particle systems, J. Statis. Phys., 44 (1986), pp. 589–644. 249

67. A. De Masi, E. Orlandi, E. Presutti and L. Triolo, Motion by curvatureby scaling nonlocal evolution equations, J. Statis. Phys., 73 (1993), pp. 543–570. 249

68. A. De Masi, E. Orlandi, E. Presutti and L. Triolo, Glauber evolutionwith Kac potentials, I. Mesoscopic and macroscopic limits, interface dynamics,Nonlinearity, 7 (1994), pp. 633–696. 249

69. A. De Masi, A. Pellegrinotti, E. Presutti and M.E. Vares, Spatialpatterns when phases separate in an interacting particle system, Ann. Probab.,22 (1994), pp. 334–371. 249, 250

70. A. De Masi and E. Presutti, Mathematical Methods for Hydrodynamic Lim-its, Lect. Notes Math., 1501 (1991), Springer. 249

71. A. Dembo, A. Vershik and O. Zeitouni, Large deviations for integer par-titions, Markov Processes Relat. Fields, 6 (2000), pp. 147–179. 259

72. A. Dembo and O. Zeitouni, Large Deviations Techniques and Applications,2nd edition, Applications of Mathematics 38, Springer, 1998. 119, 177, 180

73. X. Descombes and E. Pechersky, Droplet shapes for a class of models inZ

2 at zero temperature, J. Statis. Phys., 111 (2003), pp. 129–169. 17274. J.-D. Deuschel, Entropic repulsion of the lattice free field, II. The 0-boundary

case, Commun. Math. Phys., 181 (1996), pp. 647–665. 112, 196

266 T. Funaki

75. J.-D. Deuschel and G. Giacomin, Entropic repulsion for the free field: path-wise characterization in d ≥ 3, Commun. Math. Phys., 206 (1999), pp. 447–462. 112, 195

76. J.-D. Deuschel and G. Giacomin, Entropic repulsion for massless fields,Stoch. Proc. Appl., 89 (2000), pp. 333–354. 112, 170, 192, 193, 194

77. J.-D. Deuschel, G. Giacomin and D. Ioffe, Large deviations and concen-tration properties for ∇ϕ interface models, Probab. Theory Relat. Fields, 117(2000), pp. 49–111. 112, 143, 145, 157, 165, 167, 169, 183, 184, 185, 187

78. J.-D. Deuschel, G. Giacomin and L. Zambotti, Scaling limits of equilib-rium wetting models in (1+1)-dimension, preprint, 2004. 202

79. J.-D. Deuschel and D.W. Stroock, Large Deviations, Academic Press,1989. 177, 180, 185

80. J.-D. Deuschel and T. Nishikawa, The dynamic of entropic repulsion, inpreparation, 2004. 113, 243

81. J.-D. Deuschel and Y. Velenik, Non-Gaussian surface pinned by a weakpotential, Probab. Theory Relat. Fields, 116 (2000), pp. 359–377. 112, 197

82. H.W. Diehl, D.M. Kroll and H. Wagner, The interface in a Ginsburg-Landau-Wilson model: Derivation of the drumhead model in the low-temperature limit, Z. Physik B, 36 (1980), pp. 329–333. 258

83. R.L. Dobrushin, The description of a random field by means of conditionalprobabilities and conditions of its regularity, Theor. Probab. Appl., 13 (1968),pp. 197–224. 206

84. R.L. Dobrushin, Prescribing a system of random variables by conditional dis-tributions, Theor. Probab. Appl., 15 (1970), pp. 458–486. 206

85. R.L. Dobrushin and O. Hryniv, Fluctuations of shapes of large areas underpaths of random walks, Probab. Theory Relat. Fields, 105 (1996), pp. 423–458. 171

86. R.L. Dobrushin, R. Kotecky and S. Shlosman, Wulff Construction: aGlobal Shape from Local Interaction, AMS translation series, 104 (1992). 110, 165, 171

87. R.L. Dobrushin, S.B. Shlosman, Thermodynamic inequalities for the sur-face tension and the geometry of the Wulff construction, “Ideas and methodsin quantum and statistical physics” (Oslo, 1988, eds. Albeverio, Holden andLindstom), Cambridge Univ. Press, 1992, pp. 461–483. 157

88. M.D. Donsker and S.R.S. Varadhan, Large deviations from a hydrodynamicscaling limit, Commun. Pure Appl. Math., 42 (1989), pp. 243–270. 238

89. H. Doss and G. Royer, Processus de diffusion associe aux mesures de Gibbs

sur RZ

d

, Z. Wahr. verw. Gebiete, 46 (1978), pp. 107–124. 214, 21590. F. Dunlop, Stationary states and scaling shapes of one-dimensional interfaces,

J. Statis. Phys., 111 (2003), pp. 433–442. 124, 26191. F. Dunlop, P.A. Ferrari and L.R.G. Fontes, A dynamic one-dimensional

interface interacting with a wall, J. Statis. Phys., 107 (2002), pp. 705–727. 24392. F. Dunlop, J. Magnen and V. Rivasseau, Mass generation for an interface

in the mean field regime, Ann. Inst. H. Poincare Phys. Theor., 57 (1992),pp. 333–360. 197

93. F. Dunlop, J. Magnen, V. Rivasseau and P. Roche, Pinning of an inter-face by a weak potential, J. Statis. Phys., 66 (1992), pp. 71–98. 112, 196, 197

94. R. Durrett, B.L. Granovsky and S. Gueron, The equilibrium behavior ofreversible coagulation-fragmentation processes, J. Theoret. Probab., 12 (1999),pp. 447–474. 260

95. L.C. Evans and F. Rezakhanlou, A stochastic model for growing sandpilesand its continuum limit, Commun. Math. Phys., 197 (1998), pp. 325–345. 241


96. G. Eyink, J.L. Lebowitz and H. Spohn, Lattice gas models in contact withstochastic reservoirs: local equilibrium and relaxation to the steady state, Com-mun. Math. Phys., 140 (1991), pp. 119–131. 232

97. W.G. Faris and G. Jona-Lasinio, Large fluctuations for a nonlinear heatequation with noise, J. Phys. A: Math. Gen., 15 (1982), pp. 3025–3055. 258

98. P.A. Ferrari, L.R.G. Fontes, B. Niederhauser and M. Vachkovskaia,The serial harness interacting with a wall, Stoch. Proc. Appl., 114 (2004),pp. 175–190. 243

99. P.L. Ferrari and H. Spohn, Last branching in directed last passage percola-tion, Markov Processes Relat. Fields, 9 (2003), pp. 323–339. 261

100. R. Fernandez, J. Frohlich and A.D. Sokal, Random Walks, Critical Phe-nomena, and Triviality in Quantum Field Theory, Springer, Berlin, 1992. 158

101. M.E. Fisher, Walks, walls, wetting, and melting, J. Statis. Phys., 34 (1984),pp. 667–729. 112, 199

102. J.R. Fontaine, Non-perturbative methods for the study of massless models, in“Scaling and Self-similarity in Physics” (ed. Frohlich), 1983, pp. 203–226. 148, 151

103. L.R.G. Fontes, M. Vachkovskaia and A. Yambartsev, Repulsion of anevolving surface on random walls, math.PR/0405186, preprint, 2004. 243

104. J. Fritz, An Introduction to the Theory of Hydrodynamic Limits, Lectures inMathematical Sciences 18, The University of Tokyo, Tokyo 2001. 261

105. J. Fritz, Entropy pairs and compensated compactness for weakly asymmetricsystems, Proceedings of Shonan/Kyoto meetings “Stochastic Analysis on LargeScale Interacting Systems” (2002, eds. Funaki and Osada), Adv. Stud. PureMath., 39, Math. Soc. Japan, 2004, pp. 143–171. 261

106. J. Frohlich, C.E. Pfister and T. Spencer, On the statistical mechanicsof surfaces, Stochastic processes in quantum theory and statistical physics(Marseille, 1981), Lect. Notes Phys., 173, pp. 169–199, Springer, 1982. 118

107. J. Frohlich and T. Spencer, The Kosterlitz-Thouless transition in two-dimensional abelian spin systems and the Coulomb gas, Commun. Math. Phys.,81 (1981), pp. 527–602. 158

108. T. Funaki, Random motion of strings and related stochastic evolution equa-tions, Nagoya Math. J., 89 (1983), pp. 129–193. 156

109. T. Funaki, The reversible measures of multi-dimensional Ginzburg-Landautype continuum model, Osaka J. Math., 28 (1991), pp. 463–494. 211, 215

110. T. Funaki, The scaling limit for a stochastic PDE and the separation of phases,Probab. Theory Relat. Fields, 102 (1995), pp. 221–288. 256

111. T. Funaki, SPDE approach and log-Sobolev inequalities for continuum fieldwith two-body interactions, Proceedings of Japan-German Seminar, 1996, Hi-roshima, unpublished manuscript. 234

112. T. Funaki, Singular limit for reaction-diffusion equation with self-similarGaussian noise, Proceedings of Taniguchi symposium “New Trends in Sto-chastic Analysis” (eds. Elworthy, Kusuoka and Shigekawa), World Sci., 1997,pp. 132–152. 256

113. T. Funaki, Stochastic models for phase separation and evolution equations ofinterfaces, Sugaku Expositions, 16 (2003), pp. 97–116 (In Japanese: 50 (1998),pp. 68–85). 254, 255

114. T. Funaki, Free boundary problem from stochastic lattice gas model, Ann. Inst.H. Poincare, 35 (1999), pp. 573–603. 232, 250

115. T. Funaki, Singular limit for stochastic reaction-diffusion equation and gen-eration of random interfaces, Acta Math. Sinica, 15 (1999), pp. 407–438. 257

268 T. Funaki

116. T. Funaki, Recent results on the Ginzburg-Landau ∇φ interface model,In: “Hydrodynamic Limits and Related Topics”, edited by S. Feng, A.T.Lawniczak and S.R.S. Varadhan, Fields Institute Communications and Mono-graph Series, 2000, pp. 71–81. 113

117. T. Funaki, Hydrodynamic limit for ∇φ interface model on a wall, Probab.Theory Relat. Fields, 126 (2003), pp. 155–183. 113, 240

118. T. Funaki, Zero temperature limit for interacting Brownian particles, I. Mo-tion of a single body, Ann. Probab., 32 (2004), pp. 1201–1227. 250, 252, 253

119. T. Funaki, Zero temperature limit for interacting Brownian particles, II. Co-agulation in one dimension, Ann. Probab., 32 (2004), pp. 1228–1246. 250, 254

120. T. Funaki and H. Nagai, Degenerative convergence of diffusion process to-ward a submanifold by strong drift, Stochastics, 44 (1993), pp. 1–25. 258

121. T. Funaki and T. Nishikawa, Large deviations for the Ginzburg-Landau ∇φinterface model, Probab. Theory Relat. Fields, 120 (2001), pp. 535–568. 113, 237

122. T. Funaki and S. Olla, Fluctuations for ∇φ interface model on a wall, Stoch.Proc. Appl., 94 (2001), pp. 1–27. 113, 241, 242

123. T. Funaki and H. Sakagawa, Large deviations for ∇ϕ interface model andderivation of free boundary problems, Proceedings of Shonan/Kyoto meetings“Stochastic Analysis on Large Scale Interacting Systems” (2002, eds. Funakiand Osada), Adv. Stud. Pure Math., 39, Math. Soc. Japan, 2004, pp. 173–211. 112, 167, 169, 181, 182

124. T. Funaki and H. Spohn, Motion by mean curvature from the Ginzburg-Landau ∇φ interface model, Commun. Math. Phys., 185 (1997), pp. 1–36. 112, 151, 154, 157, 158, 208, 209, 2

125. T. Funaki and K. Uchiyama, From Micro to Macro 1, Mathematical Theoryof Interface Models, in Japanese. Springer-Verlag Tokyo, 2002, 283+xi pages. 113

126. T. Funaki, K. Uchiyama and H.T. Yau, Hydrodynamic limit for lattice gasreversible under Bernoulli measures, “Nonlinear Stochastic PDE’s: Hydrody-namic Limit and Burgers’ Turbulence” (eds. Funaki and Woyczynski), IMAvolume (Univ. Minnesota) 77 (1995), pp. 1–40, Springer. 234

127. K. Gawedzki and A. Kupiainen, Renormalization group study of a criticallattice model I, II, Commun. Math. Phys., 82 (1981), pp. 407–433 and 83(1982), pp. 469–492. 151, 215

128. H.-O. Georgii, Gibbs Measures and Phase Transitions, Walter, Berlin NewYork, 1988. 119, 122

129. G. Giacomin, Onset and structure of interfaces in a Kawasaki+Glauber in-teracting particle system, Probab. Theory Relat. Fields, 103 (1995), pp. 1–24. 249

130. G. Giacomin, Anharmonic lattices, random walks and random interfaces, inRecent research developments in statistical physics, Transworld research net-work, 1 (2000), pp. 97–118. 113, 150, 151

131. G. Giacomin, On stochastic domination in the Brascamp-Lieb framework,Math. Proc. Cambridge Philos. Soc., 134 (2003), pp. 507–514. 148

132. G. Giacomin, Limit theorems for random interfaces of Ginzburg-Landautype, Stochastic partial differential equations and applications (Trento, 2002),pp. 235–253, Lecture Notes in Pure and Appl. Math., 227, Dekker, New York,2002. 113

133. G. Giacomin, Aspects of statistical mechanics of random surfaces, Notes ofthe lectures given at IHP, fall 2001, preprint, 2002. 113

134. G. Giacomin, J.L. Lebowitz and E. Presutti, Deterministic and stochastichydrodynamic equations arising from simple microscopic model systems, In:Stochastic Partial Differential Equations: Six Perspectives (eds R. Carmonaand B. Rozovskii), pp. 107–152, Math. Surveys Monogr., 64, AMS, 1999. 250


135. G. Giacomin, S. Olla and H. Spohn, Equilibrium fluctuations for ∇ϕ in-terface model, Ann. Probab., 29 (2001), pp. 1138–1172. 113, 143, 149, 157, 204, 235, 236

136. M.-H. Giga and Y. Giga, Evolving graphs by singular weighted curvature,Arch. Rational Mech. Anal., 141 (1998), pp. 117–198. 232

137. M.-H. Giga and Y. Giga, A PDE Approach for Motion of Phase-Boundariesby a Singular Interfacial Energy, Proceedings of Shonan/Kyoto meetings“Stochastic Analysis on Large Scale Interacting Systems” (2002, eds. Funakiand Osada), Adv. Stud. Pure Math., 39, Math. Soc. Japan, 2004, pp. 213–232. 232

138. J. Glimm and A. Jaffe, Quantum Physics, a Functional Integral Point ofView, Springer, 1981. 137, 148, 156

139. J. Gravner and J. Quastel, Internal DLA and the Stefan problem, Ann.Probab., 28 (2000), pp. 1528–1562. 250

140. S. Gueron and Y. Rubinstein, The minimal reversible coagulation-fragmentation process having no factorized coagulation and fragmentation rates,Markov Processes Relat. Fields, 6 (2000), pp. 257–264. 260

141. M.Z. Guo, G.C. Papanicolaou and S.R.S. Varadhan, Nonlinear diffusionlimit for a system with nearest neighbor interactions, Commun. Math. Phys.,118 (1988), pp. 31–59. 153, 163, 220, 225, 226, 227

142. M.E. Gurtin, Thermomechanics of Evolving Phase Boundaries in the Plane,Oxford Press, 1993. 258

143. B. Helffer, Remarks on decay of correlations and Witten Laplacians,Brascamp-Lieb inequalities and semiclassical limit, J. Funct. Anal., 155 (1998),pp. 571–586. 148

144. B. Helffer and J. Sjostrand, On the correlation for Kac-like models in theconvex case, J. Statis. Phys., 74 (1994), pp. 349–409. 142

145. C. Hiergeist and R. Lipowsky, Local contacts of membranes and strings,Physica A, 244 (1997), pp. 164–175. 118, 196

146. Y. Higuchi, J. Murai and J. Wang, The Dobrushin-Hryniv theoryfor the two-dimensional lattice Widom-Rowlinson model, Proceedings ofShonan/Kyoto meetings “Stochastic Analysis on Large Scale Interacting Sys-tems” (2002, eds. Funaki and Osada), Adv. Stud. Pure Math., 39, Math. Soc.Japan, 2004, pp. 233–281. 171

147. P.C. Hohenberg and B.I. Halperin, Theory of dynamic critical phenomena,Rev. Mod. Phys., 49 (1977), pp. 435–475. 232

148. C.D. Howard, Models of first-passage percolation, in “Probability on Dis-crete Structures” (ed. Kesten), Encyclopaedia Math. Sci., 110 (2004), Springer,pp. 125–173. 261

149. N. Ikeda and S. Watanabe, Stochastic Differential Equations and DiffusionProcesses, North Holland/Kodansha, 1981 (2nd edition 1989). 245, 246

150. D. Ioffe, Large deviations for the 2D Ising model: a lower bound withoutcluster expansions, J. Statis. Phys., 74 (1994), pp. 411–432. 171

151. D. Ioffe, Exact large deviation bounds up to Tc for the Ising model in twodimensions, Probab. Theory Relat. Fields, 102 (1995), pp. 313–330. 171

152. D. Ioffe and R.H. Schonmann, Dobrushin-Kotecky-Shlosman theorem up tothe critical temperature, Commun. Math. Phys., 199 (1998), pp. 117–167. 110, 171

153. D. Ioffe and Y. Velenik, A note on the decay of correlations under δ-pinning, Probab. Theory Relat. Fields, 116 (2000), pp. 379–389. 112, 197, 198

154. Y. Isozaki and N. Yoshida, Weakly pinned random walk on the wall: pathwisedescriptions of the phase transition, Stoch. Proc. Appl., 96 (2001), pp. 261–284. 199, 202

270 T. Funaki

155. K. Ito and H.P. McKean, Potentials and the random walk, Illinois J. Math.,4 (1960), pp. 119–132. 132

156. K. Iwata, Reversible measures of a P (φ)1-time evolution, “Probabilistic Meth-ods in Mathematical Physics” (eds. Ito and Ikeda), Proceedings, Katata Kyoto1985, pp. 195–209, Kinokuniya, Tokyo, 1987. 215

157. K. Johansson, Transversal fluctuations for increasing subsequences on theplane, Probab. Theory Relat. Fields, 116 (2000), pp. 445–456. 261

158. K. Johansson, Shape fluctuations and random matrices, Commun. Math.Phys., 209 (2000), pp. 437–476. 261

159. K. Johansson, Random growth and random matrices, European Congressof Mathematics, Vol. I (Barcelona, 2000), pp. 445–456, Progr. Math., 201,Birkhauser, Basel, 2001. 261

160. K. Johansson, Non-intersecting paths, random tilings and random matrices,Probab. Theory Relat. Fields, 123 (2002), pp. 225–280. 261

161. M. Kardar, G. Parisi and Y.-C. Zhang, Dynamical scaling of growing in-terfaces, Phys. Rev. Lett., 56 (1986), pp. 889–892. 262

162. T. Kasai, Curvature flow added random forces, Nagoya Univ. Master Thesis,1994. 258

163. M.A. Katsoulakis and P.E. Souganidis, Interacting particle systems andgeneralized evolution of fronts, Arch. Rat. Mech. Anal., 127 (1994), pp. 133–157. 249

164. M.A. Katsoulakis and P.E. Souganidis, Generalized motion by mean cur-vature as a macroscopic limit of stochastic Ising models with long range inter-actions and Glauber dynamics, Commun. Math. Phys., 169 (1995), pp. 61–97. 249

165. G.S. Katzenberger, Solutions of a stochastic differential equation forced ontoa manifold by a large drift, Ann. Probab., 19 (1991), pp. 1587–1628. 258

166. K. Kawasaki and T. Ohta, Kinetic drumhead model of interface I, Prog.Theoret. Phys., 67 (1982), pp. 147–163. 257, 258

167. R. Kenyon, The planar dimer model with boundary: a survey, “Directionsin Mathematical Quasicrystals” (eds. Baake and Moody), Amer. Math. Soc.,Providence, RI 2000, pp. 307–328. 259

168. R. Kenyon, Dominos and the Gaussian free field, Ann. Probab., 29 (2001),pp. 1128–1137. 172

169. S.V. Kerov, Asymptotic Representation Theory of the Symmetric Group andits Applications in Analysis, Translations of Math. Monographs, 219, Amer.Math. Soc., Providence, RI, 2003, xvi+201 pp. 258

170. C. Kipnis and C. Landim, Scaling Limits of Interacting Particle Systems,Springer, 1999. 220

171. C. Kipnis, C. Landim and S. Olla, Macroscopic properties of a station-ary non-equilibrium distribution for a non-gradient interacting particle system,Ann. Inst. H. Poincare, 31 (1995), pp. 191–221. 232

172. C. Kipnis, S. Olla and S.R.S. Varadhan, Hydrodynamics and large devi-ation for simple exclusion processes, Commun. Pure Appl. Math., 42 (1989),pp. 115–137. 238

173. C. Kipnis and S.R.S. Varadhan, Central limit theorem for additive func-tionals of reversible Markov processes and applications to simple exclusions,Commun. Math. Phys., 104 (1986), pp. 1–19. 236

174. K. Komoriya, Free boundary problem from two component system on Z, toappear in J. Math. Sci. Univ. Tokyo, 12 (2005), pp. 141–163. 250


175. S.M. Kozlov, The method of averaging and walks in inhomogeneous environ-ments, Russian Math. Surveys, 40 (1985), pp. 73–145. 212

176. J. Krug and H. Spohn, Universality classes for deterministic surface growth,Phys. Rev. A, 38 (1988), pp. 4271–4283. 262

177. J. Krug and H. Spohn, Kinetic roughening of growing surfaces, in “Solids Farfrom Equilibrium” (ed. Godreche), Cambridge Univ. Press, 1991, pp. 479–582. 261

178. N.V. Krylov and B.L. Rozovski, Stochastic evolution equations, J. SovietMath., 16 (1981), pp. 1233–1277. 221

179. H. Kunsch, Decay of correlations under Dobrushin’s uniqueness condition andits applications, Commun. Math. Phys., 84 (1982), pp. 207–222. 206

180. O.A. Ladyzenskaya, V.A. Solonnikov and N.N. Ural’ceva, Linear andquasilinear equations of parabolic type, Translations of Math. Monographs, 23,Amer. Math. Soc., Providence, RI, 1968. 223

181. C. Landim, S. Olla and S. Volchan, Driven tracer particle in one di-mensional symmetric simple exclusion, Commun. Math. Phys., 192 (1998),pp. 287–307. 250

182. C. Landim, S. Olla and H.T. Yau, First order correction for the hydro-dynamic limit of asymmetric simple exclusion processes in dimension d ≥ 3,Commun. Pure Appl. Math., 50 (1997), pp. 149–203. 261

183. C. Landim, S. Sethuraman and S. Varadhan, Spectral gap for zero-rangedynamics, Ann. Probab., 24 (1996), pp. 1871–1902. 234

184. C. Landim and H.-T. Yau, Large deviations of interacting particle systemsin infinite volume, Commun. Pure Appl. Math., 48 (1995), pp. 339–379 238

185. G.F. Lawler, Intersections of Random Walks, Birkhauser, Boston, 1991. 132186. J.L. Lebowitz and C. Maes, The effect of an external field on an interface,

entropic repulsion, J. Statis. Phys., 46 (1987), pp. 39–49. 112, 193187. P. Lenz and R. Lipowsky, Stability of droplets and channels on homogeneous

and structured surfaces, Eur. Phys. J. E, 1 (2000), pp. 249–262. 203188. P. Lenz, W. Fenzl and R. Lipowsky, Wetting of ring-shaped surface do-

mains, Europhys. Lett., 53 (2001), pp. 618–624. 203189. T.M. Liggett, Interacting Particle Systems, Springer, 1985. 249190. T.M. Liggett, Stochastic interacting systems: contact, voter and exclusion

processes, Springer, 1999. 249191. P.L. Lions and A.-S. Sznitman, Stochastic differential equations with reflect-

ing boundary conditions, Commun. Pure Appl. Math., 37 (1984), pp. 511–537. 239192. R. Lipowsky, Structured surfaces and morphological wetting transitions, In-

terface Science 9 (2001), pp. 105–115. 203193. R. Lipowsky, Morphological wetting transitions at chemically structured sur-

faces, Current Opinion in Colloid and Interface Science 6 (2001), pp. 40–48. 203194. S. Lu, Hydrodynamic scaling limits with deterministic initial configurations,

Ann. Probab., 23 (1995), pp. 1831–1852. 231195. S.L. Lu and H.-T. Yau, Spectral gap and logarithmic Sobolev inequality for

Kawasaki and Glauber dynamics, Commun. Math. Phys., 156 (1993), pp. 399–433. 234

196. J. Magnen and R. Seneor, The infrared behavior of (∇ϕ)43, Ann. Phys., 152(1984), pp. 136–202. 151

197. J.B. Martin, Limiting shape for directed percolation models, Ann. Probab.,32 (2004), pp. 2908–2937. 261

272 T. Funaki

198. P. Massart, Concentration inequalities and model selection, “Ecole d’Ete deProbabilites de Saint-Flour XXXIII-2003” (ed. Picard), Lect. Notes Math.,Springer, 2005. 148

199. A. Messager, S. Miracle-Sole and J. Ruiz, Convexity properties of thesurface tension and equilibrium crystals, J. Statis. Phys., 67 (1992), pp. 449–470. 157

200. A.A. Mogul’skii, Large deviations for trajectories of multi-dimensional ran-dom walks, Theory Probab. Appl., 21 (1976), pp. 300–315. 163, 177

201. S. Muller, Variational models for microstructure and phase transitions, In:Calculus of variations and geometric evolution problems (Cetraro, 1996), Lect.Notes Math., 1713 (1999), Springer, pp. 85–210. 216

202. A. Naddaf and T. Spencer, On homogenization and scaling limit of somegradient perturbations of a massless free field, Commun. Math. Phys., 183(1997), pp. 55–84. 112, 132, 137, 150, 204, 236

203. T. Nishikawa, Hydrodynamic limit for the Ginzburg-Landau ∇φ interfacemodel with a conservation law, J. Math. Sci. Univ. Tokyo., 9 (2002), pp. 481–519. 233

204. T. Nishikawa, Hydrodynamic limit for the Ginzburg-Landau ∇φ interfacemodel with boundary conditions, Probab. Theory Relat. Fields, 127 (2003),pp. 205–227. 218

205. Y. Nishiura, Far-from-equilibrium dynamics, Translations of Math. Mono-graphs, 209, Iwanami Series in Modern Math., Amer. Math. Soc., Providence,RI, 2002, xx+311 pp. 235

206. J. Norris, Smoluchowski’s coagulation equation: uniqueness, non-uniquenessand a hydrodynamic limit for the stochastic coalescent, Ann. Appl. Prob., 9(1999), pp. 78–109. 260

207. J. Norris, Cluster coagulation, Commun. Math. Phys., 209 (2000), pp. 407–435. 260

208. D. Nualart and E. Pardoux, White noise driven quasilinear SPDEs withreflection, Probab. Theory Relat. Fields, 93 (1992), pp. 77–89. 241

209. T. Ohta, Instability of interfaces and pattern formation, Frontiers of Physics,10, Kyoritsu, 1985, in Japanese. 258

210. T. Ohta, Mathematical theory on interfacial dynamics, Nihon-Hyoronsha,1997, in Japanese. 113

211. T. Ohta, D. Jasnow and K. Kawasaki, Universal scaling in the motion ofrandom interfaces, Physical Review Letters, 49 (1982), pp. 1223–1226. 258

212. C.-E. Pfister and Y. Velenik, Interface, surface tension and reentrant pin-ning transition in the 2D Ising model, Commun. Math. Phys., 204 (1999),pp. 269–312. 174, 175, 198, 202

213. A. Pisztora, Surface order large deviations for Ising, Potts and percolationmodels, Probab. Theory Relat. Fields, 104 (1996), pp. 427–466. 171

214. B. Pittel, On a likely shape of the random Ferrers diagram, Adv. Appl. Math.,18 (1997), pp. 432–488. 260

215. M. Prahofer and H. Spohn, Current fluctuations for the totally asymmet-ric simple exclusion process, in “In and Out of Equilibrium (ed. Sidoravicius,Mambucaba, 2000)”, pp. 185–204, Progr. Probab., 51, Birkhauser, 2002. 261

216. M. Prahofer and H. Spohn, Scale invariance of the PNG droplet and theAiry process, J. Statist. Phys., 108 (2002), pp. 1071–1106. 261

217. J. Quastel, Diffusion of color in the simple exclusion process, Commun. PureAppl. Math. 45, 623–679 (1992). 250


218. M. Reed and B. Simon, Methods of Modern Mathematical Physics, vol. I:Functional Analysis, revised and enlarged edition; vol.II: Fourier Analysis, Self-Adjointness, Academic Press, New York, 1980, 1975. 229

219. F. Rezakhanlou, Hydrodynamic limit for a system with finite range interac-tions, Commun. Math. Phys., 129 (1990), pp. 445–480. 232

220. F. Rezakhanlou, Hydrodynamic limit for attractive particle systems on Zd,

Commun. Math. Phys., 140 (1991), pp. 417–448. 232221. F. Rezakhanlou, Continuum limit for some growth models, Stoch. Proc.

Appl., 101 (2002), pp. 1–41. 241222. F. Rezakhanlou, Continuum limit for some growth models II, Ann. Probab.,

29 (2001), pp. 1329–1372. 241223. D. Ruelle, Statistical Mechanics: Rigorous Results, W. A. Benjamin, 1969. 119224. Y. Saito, Statistical Physics of Crystal Growth, World Sci., 1996. 113225. H. Sakagawa, The reversible measures of interacting diffusion system with

plural conservation laws, Markov Process. Related Fields, 7 (2001), pp. 289–300. 215

226. H. Sakagawa, Entropic repulsion for a Gaussian lattice field with finite rangeinteraction, J. Math. Phys., 44 (2003), pp. 2939–2951. 196

227. H. Sakagawa, Entropic repulsion for two dimensional multi-layered harmoniccrystals, J. Statist. Phys., 114 (2004), pp. 37–49. 196

228. H. Sakagawa, Entropic repulsion for a high dimensional Gaussian lattice fieldbetween two walls, preprint, 2004. 196

229. T. Seppalainen, Existence of hydrodynamics for the totally asymmetric simpleK-exclusion process, Ann. Probab., 27 (1999), pp. 361–415. 241

230. S. Sheffield, Random surfaces: large deviations principles and gra-dient Gibbs measure classifications, PhD thesis, Stanford Univ., 2003,arXiv:math.PR/0304049. 117, 152, 155, 168, 215

231. T. Shiga and A. Shimizu, Infinite-dimensional stochastic differential equa-tions and their applications, J. Math. Kyoto Univ., 20 (1980), pp. 395–416. 214, 215

232. I. Shigekawa, Stochastic Analysis, Translations of Math. Monographs, 224,Iwanami Series in Modern Math., Amer. Math. Soc., Providence, RI, 2004,182 pp. 131

233. S. Shlosman, The Wulff construction in statistical mechanics and combina-torics, Russian Math. Surveys, 56 (2001), pp. 709–738. 172, 258, 259

234. B. Simon, The P (φ)2 Euclidean (Quantum) Field Theory, Princeton Univ.Press, 1974. 137

235. B. Simon, Functional Integration and Quantum Physics, Academic Press, 1979. 148236. Ya.G. Sinai, Distribution of some functionals of the integral of a random walk,

Theoret. Math. Phys., 90 (1992), pp. 219–241. 196237. J. Sjostrand, Correlation asymptotics and Witten Laplacians, St. Petersburg

Math. J., 8 (1997), pp. 123–147. 142238. F. Spitzer, Principles of Random Walk, Nostrand, 1964. 132, 133239. H. Spohn, Large Scale Dynamics of Interacting Particles, Springer, 1991. 220240. H. Spohn, Interface motion in models with stochastic dynamics, J. Statis.

Phys., 71 (1993), pp. 1081–1132. 158, 232, 233, 249, 257241. A. Stohr, Uber einige lineare partielle Differenzengleichungen mit konstanten

Koeffizienten, I, II, III, Math. Nachr., 3 (1950), pp. 208–242, pp. 295–315,pp. 330–357. 133

242. D.W. Stroock and S.R.S. Varadhan, Diffusion processes with boundaryconditions, Commun. Pure Appl. Math., 24 (1971), 147–225. 245

274 T. Funaki

243. S. Takanobu and S. Watanabe, On the existence and uniqueness of diffu-sion processes with Wentzell’s boundary conditions, J. Math. Kyoto Univ., 28(1988), pp. 71–80. 245

244. H. Tanaka, Stochastic differential equations with reflecting boundary conditionin convex regions, Hiroshima Math. J., 9 (1979), pp. 163–177. 239

245. H. Tanemura and N. Yoshida, Localization transition of d-friendly walkers,Probab. Theory Relat. Fields., 125 (2003), pp. 593–608. 202

246. S.R.S. Varadhan, Large Deviations and Applications, SIAM, 1984. 177, 180247. S.R.S. Varadhan, Nonlinear diffusion limit for a system with nearest neigh-

bor interactions - II, “Asymptotic problems in probability theory: stochasticmodels and diffusions on fractals” (eds. Elworthy and Ikeda), Longman, 1993,pp. 75–128. 220

248. S.R.S. Varadhan, Large deviations for the asymmetric simple exclusionprocess, Proceedings of Shonan/Kyoto meetings “Stochastic Analysis on LargeScale Interacting Systems” (2002, eds. Funaki and Osada), Adv. Stud. PureMath., 39, Math. Soc. Japan, 2004, pp. 1–27. 261

249. Y. Velenik, Entropic repulsion of an interface in an external field, Probab.Theory Relat. Fields, 129 (2004), pp. 83–112. 202

250. A.M. Vershik, Statistical mechanics of combinatorial partitions, and theirlimit shapes, Funct. Anal. Appl., 30 (1996), pp. 90–105. 258, 259

251. A. Visintin, Models of Phase Transitions, Birkhauser, 1996. 235252. G.S. Weiss, A free boundary problem for non-radial-symmetric quasi-linear

elliptic equations, Adv. Math. Sci. Appl., 5 (1995), pp. 497–555. 172253. J. Wloka, Partial differential equations, Cambridge Univ. Press, 1987. 168254. G. Wulff, Zur Frage der Geschwindigkeit des Wachsthums und der Auflosung

der Krystallflachen, Z. Krystallogr., 34 (1901), pp. 449–530. 109255. K. Yosida, Functional Analysis, 4th edition, Springer, 1974. 221256. L. Zambotti, Fluctuations for a ∇φ interface model with repulsion from a

wall, Probab. Theory Relat. Fields, 129 (2004), pp. 315–339. 242257. M. Zhu, Equilibrium fluctuations for one-dimensional Ginzburg-Landau lattice

model, Nagoya Math. J., 117 (1990), pp. 63–92. 214258. M. Zhu, The reversible measures of a conservative system with finite range

interactions, “Nonlinear Stochastic PDEs: Hydrodynamic Limit and Burgers’Turbulence” (eds. Funaki and Woyczynski), IMA volume 77 (1995), Univ.Minnesota, Springer, New York, pp. 53–64. 215

Documents

Part II Tadahisa Funaki Stochastic Interface Modelsfunaki/SF/LNM1869.pdf · 2005-11-29 · Stochastic Interface Models 111 can be formulated in the framework of classical limit theorems