A new proof of the sharpness of the phase transition for Bernoulli percolation on \mathbb{Z}^{d}

# A new proof of the sharpness of the phase transition for Bernoulli percolation on \mathbb{Z}^{d}

Hugo Duminil-Copin and Vincent Tassion
July 8, 2019
###### Abstract

We provide a new proof of the sharpness of the phase transition for nearest-neighbour Bernoulli percolation. More precisely, we show that

• for p<p_{c}, the probability that the origin is connected by an open path to distance n decays exponentially fast in n.

• for p>p_{c}, the probability that the origin belongs to an infinite cluster satisfies the mean-field lower bound \theta(p)\ge\tfrac{p-p_{c}}{p(1-p_{c})}.

This note presents the argument of [DCT15], which is valid for long-range Bernoulli percolation (and for the Ising model) on arbitrary transitive graphs in the simpler framework of nearest-neighbour Bernoulli percolation on \mathbb{Z}^{d}.

## 1 Statement of the result

#### Notation.

Fix an integer d\ge 2. We consider the d-dimensional hypercubic lattice (\mathbb{Z}^{d},\mathbb{E}^{d}). Let \Lambda_{n}=\{-n,\ldots,n\}^{d}, and let \partial\Lambda_{n}:=\Lambda_{n}\setminus\Lambda_{n-1} be its vertex-boundary. Throughout this note, S always stands for a finite set of vertices containing the origin. Given such a set, we denote its edge-boundary by \Delta S, defined by all the edges \{x,y\} with x\in S and y\notin S.

Consider the Bernoulli bond percolation measure \mathbb{P}_{p} on \{0,1\}^{\mathbb{E}^{d}} for which each edge of \mathbb{E}^{d} is declared open with probability p and closed otherwise, independently for different edges.

Two vertices x and y are connected in S\subset V if there exists a path of vertices (v_{k})_{0\le k\le K} in S such that v_{0}=x, v_{K}=y, and \{v_{k},v_{k+1}\} is open for every 0\le k<K. We denote this event by x\lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}y. If S=\mathbb{Z}^{d}, we drop it from the notation. We set 0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}\infty (resp. 0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}% \partial\Lambda_{n}) if 0 is connected to infinity (resp. 0 is connected to a vertex in \partial\Lambda_{n}).

#### Phase transition.

A new idea of this paper is to use a different definition of the critical parameter than the standard one. This new definition relies on the following quantity. For p\in[0,1] and 0\in S\subset\mathbb{Z}^{d}, define

 \varphi_{p}(S):=p\tsum\slimits@_{\{x,y\}\in\Delta S}\mathbb{P}_{p}[0% \lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}x] (1.1)

and introduce the following quantities:

 \displaystyle\tilde{p}_{c} \displaystyle:=\sup\big{\{}p\in[0,1]\text{ s.t.\@ there exists a finite set $0\subset S\subset\mathbb{Z}^{d}$ with $\varphi_{p}(S)<1$}\big{\}}, (1.2) \displaystyle p_{c} \displaystyle:=\sup\{p\text{ s.t. $\mathbb{P}_{p}[0\leftarrow\mathrel{\mskip-3% .1mu }\rightarrow\infty]=0$}\}.

We are now in a position to state our main result.

###### Theorem 1.1.

For any d\ge 2, \tilde{p}_{c}=p_{c}. Furthermore,

1. For p<p_{c}, there exists c=c(p)>0 such that for every n\ge 1,

 \mathbb{P}_{p}[0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\partial\Lambda_{n% }]\le e^{-cn}.
2. For p>p_{c},

 \mathbb{P}_{p}[0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\infty]\ge\frac{p-% p_{c}}{p(1-p_{c})}.

Remarks.

1. We refer to [DCT15] for a detailed bibliography, and for a version of the proof valid in greater generality. The aim of this note is to provide a proof in the simplest framework.

2. Theorem 1.1 was proved by Aizenman and Barsky [AB87] in the more general framework of long-range percolation. In their proof, they consider an additional parameter h corresponding to an external field, and they derive the results from differential inequalities satisfied by the thermodynamical quantities of the model. A different proof, based on the geometric study of the pivotal edges, was obtained at the same time by Menshikov [Men86]. These two proofs are also presented in [Gri99].

3. In the definition of \tilde{p}_{c}, the set of parameters p such that there exists a finite set 0\subset S\subset\mathbb{Z}^{d} with \varphi_{p}(S)<1 is an open subset of [0,1]. Thus, \tilde{p}_{c} do not belong to this set.

We obtain that the expected size of the cluster of the origin satisfies that for every p>p_{c},

 \tsum\slimits@_{x\in\mathbb{Z}^{d}}\mathbb{P}_{p}[0\leftarrow\mathrel{\mskip-3% .1mu }\rightarrow x]\ge\tsum\slimits@_{n\ge 0}\varphi_{p}(\Lambda_{n})=+\infty.
4. Since \varphi_{p}(\{0\})=2dp, we obtain p_{c}\ge 1/{2d}.

5. Item 2 provides a mean-field lower bound for the infinite cluster density.

6. Theorem 1.1 implies that p_{c}\le 1/2 on \mathbb{Z}^{2}. Combined with Zhang’s argument [Gri99, Lemma 11.12], this shows that p_{c}=1/2.

## 2 Proof of the theorem

It is sufficient to show Items 1 and 2 with p_{c} replaced by \tilde{p}_{c} (since it immediately implies the equality p_{c}=\tilde{p}_{c}).

### 2.1 Proof of Item 1

The proof of Item 1 can be derived from the BK-inequality [vdBK85]. We present here an exploration argument, similar to the one in [Ham57], which does not rely on the BK-inequality. Let p<\tilde{p}_{c}. By definition, one can fix a finite set S containing the origin, such that \varphi_{p}(S)<1. Let L>0 such that S\subset\Lambda_{L-1}.

Let k\ge 1 and assume that the event 0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}% \partial\Lambda_{kL} holds. Let

 \mathrsfs{C}=\{z\in S:0\lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}z\}.

Since S\cap\partial\Lambda_{kL}=\emptyset, there exists an edge \{x,y\}\in\Delta S such that the following events occur:

• 0 is connected to x in S,

• \{x,y\} is open,

• y is connected to \partial\Lambda_{kL} in \mathrsfs{C}^{c}.

Using first the union bound, and then a decomposition with respect to possible values of \mathrsfs{C}, we find

 \displaystyle\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}\partial\Lambda_{kL}] (2.1) \displaystyle\le\tsum\slimits@_{\{x,y\}\in\Delta S}\tsum\slimits@_{C\subset S}% \mathbb{P}_{p}\big{[}\{0\lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}x,\mathrsfs{C}=C\}\cap\{\{x,y\}\text{ is open}\}% \cap\{y\lx@stackrel{{\scriptstyle\mathbb{Z}^{d}\setminus C}}{{\leftarrow% \mathrel{\mskip-3.1mu }\rightarrow}}\partial\Lambda_{kL}\}\big{]} (2.2) \displaystyle=p\tsum\slimits@_{\{x,y\}\in\Delta S}\tsum\slimits@_{C\subset S}% \mathbb{P}_{p}\big{[}0\lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}x,\mathrsfs{C}=C\big{]}\mathbb{P}_{p}\big{[}y% \lx@stackrel{{\scriptstyle\mathbb{Z}^{d}\setminus C}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}\partial\Lambda_{kL}\big{]}. (2.3)

In the second line, we used the fact that the three events depend on different sets of edges and are therefore independent. Since y\in\Lambda_{L}, one can bound \mathbb{P}_{p}[y\lx@stackrel{{\scriptstyle\mathbb{Z}^{d}\setminus C}}{{% \leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}\partial\Lambda_{kL}] by \mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu % }\rightarrow}}\partial\Lambda_{(k-1)L}] in the last expression. Hence, we find

 \mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu % }\rightarrow}}\partial\Lambda_{kL}]\le\varphi_{p}(S)\mathbb{P}_{p}\big{[}y% \lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}% \partial\Lambda_{(k-1)L}\big{]} (2.4)

which by induction gives

 \mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu % }\rightarrow}}\partial\Lambda_{kL}]\le\varphi_{p}(S)^{k-1}. (2.5)

This proves the desired exponential decay.

### 2.2 Proof of Item 2

Let us start by the following lemma providing a differential inequality valid for every p.

###### Lemma 2.1.

Let p\in[0,1] and n\ge 1,

 \frac{d}{dp}\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}\partial\Lambda_{n}]\ge\frac{1}{p(1-p)}\cdot\inf_{% \begin{subarray}{c}S\subset\Lambda_{n}\\ 0\in S\end{subarray}}\varphi_{p}(S)\cdot\big{(}1-\mathbb{P}_{p}[0\lx@stackrel{% {\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}\partial\Lambda% _{n}]\big{)}. (2.6)

Let us first see how it implies Item 2 of Theorem 1.1. Integrating the differential inequality (2.6) between \tilde{p}_{c} and p>\tilde{p}_{c} implies that for every n\ge 1, \mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{\mskip-3.1mu % }\rightarrow}}\partial\Lambda_{n}]\ge\frac{p-\tilde{p}_{c}}{p(1-\tilde{p}_{c})}. By letting n tend to infinity, we obtain the desired lower bound on \mathbb{P}_{p}[0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\infty].

###### Proof of Lemma 2.1.

Recall that \{x,y\} is pivotal for the configuration \omega and the event \{0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\partial\Lambda_{n}\} if \omega_{\{x,y\}}\notin\{0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\partial% \Lambda_{n}\} and \omega^{\{x,y\}}\in\{0\leftarrow\mathrel{\mskip-3.1mu }\rightarrow\partial% \Lambda_{n}\}. (The configuration \omega_{\{x,y\}}, resp. \omega^{\{x,y\}}, coincides with \omega except that the edge \{x,y\} is closed, resp. open.) By Russo’s formula (see [Gri99, Section 2.4]), we have

 \displaystyle\frac{d}{dp}\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{% \leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}\partial\Lambda_{n}] \displaystyle=\tsum\slimits@_{e\subset\Lambda_{n}}\mathbb{P}_{p}\big{[}\text{$% e$ is pivotal}\big{]} (2.7) \displaystyle=\frac{1}{1-p}\tsum\slimits@_{e\subset\Lambda_{n}}\mathbb{P}_{p}% \big{[}\text{$e$ is pivotal},\,0\overset{}{\mathrel{\mathchoice{\ooalign{$% \hfil\displaystyle/\hfil$\cr$\displaystyle\leftarrow\mathrel{\mskip-3.1mu }% \rightarrow$}}{\ooalign{$\hfil\textstyle/\hfil$\cr$\textstyle\leftarrow% \mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$\hfil\scriptstyle/\hfil$\cr$% \scriptstyle\leftarrow\mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$\hfil% \scriptscriptstyle/\hfil$\cr$\scriptscriptstyle\leftarrow\mathrel{\mskip-3.1mu% }\rightarrow$}}}}\partial\Lambda_{n}\big{]}. (2.8)

Define the following random subset of \Lambda_{n}:

 \mathrsfs{S}:=\{x\in\Lambda_{n}\text{ such that }x\mathrel{\mathchoice{% \ooalign{$\hfil\displaystyle/\hfil$\cr$\displaystyle\leftarrow\mathrel{\mskip-% 3.1mu }\rightarrow$}}{\ooalign{$\hfil\textstyle/\hfil$\cr$\textstyle\leftarrow% \mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$\hfil\scriptstyle/\hfil$\cr$% \scriptstyle\leftarrow\mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$\hfil% \scriptscriptstyle/\hfil$\cr$\scriptscriptstyle\leftarrow\mathrel{\mskip-3.1mu% }\rightarrow$}}}\partial\Lambda_{n}\}.

The boundary of \mathrsfs{S} corresponds to the outmost blocking surface (which can be obtained by exploring from the outside the set of vertices connected to the boundary). When 0 is not connected to \partial\Lambda_{n}, the set \mathrsfs{S} is always a subset of \Lambda_{n} containing the origin. By summing over the possible values for \mathrsfs{S}, we obtain

 \frac{d}{dp}\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}\partial\Lambda_{n}]=\frac{1}{1-p}\tsum\slimits@_{% \begin{subarray}{c}S\subset\Lambda_{n}\\ 0\in S\end{subarray}}\tsum\slimits@_{e\subset\Lambda_{n}}\mathbb{P}_{p}\big{[}% \text{$e$ is pivotal},\,\mathrsfs{S}=S\big{]} (2.9)

Observe that on the event \mathrsfs{S}=S, the pivotal edges are the edges \{x,y\}\in\Delta S such that 0 is connected to x in S. This implies that

 \frac{d}{dp}\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{\leftarrow\mathrel{% \mskip-3.1mu }\rightarrow}}\partial\Lambda_{n}]=\frac{1}{1-p}\tsum\slimits@_{% \begin{subarray}{c}S\subset\Lambda_{n}\\ 0\in S\end{subarray}}\tsum\slimits@_{\{x,y\}\in\Delta S}\mathbb{P}_{p}\big{[}0% \lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}% x,\,\mathrsfs{S}=S\big{]}. (2.10)

The event \{\mathrsfs{S}=S\} is measurable with respect to the configuration outside S and is therefore independent of \{0\lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{\mskip-3.1mu }% \rightarrow}}x\}. We obtain

 \displaystyle\frac{d}{dp}\mathbb{P}_{p}[0\lx@stackrel{{\scriptstyle}}{{% \leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}\partial\Lambda_{n}] \displaystyle=\frac{1}{1-p}\tsum\slimits@_{\begin{subarray}{c}S\subset\Lambda_% {n}\\ 0\in S\end{subarray}}\tsum\slimits@_{\{x,y\}\in\Delta S}\mathbb{P}_{p}\big{[}0% \lx@stackrel{{\scriptstyle S}}{{\leftarrow\mathrel{\mskip-3.1mu }\rightarrow}}% x\big{]}\mathbb{P}_{p}\big{[}\mathrsfs{S}=S\big{]} (2.11) \displaystyle=\frac{1}{p(1-p)}\tsum\slimits@_{\begin{subarray}{c}S\subset% \Lambda_{n}\\ 0\in S\end{subarray}}\varphi_{p}(S)\mathbb{P}_{p}\big{[}\mathrsfs{S}=S\big{]} (2.12) \displaystyle\ge\frac{1}{p(1-p)}\inf_{\begin{subarray}{c}S\subset\Lambda_{n}\\ 0\in S\end{subarray}}\varphi_{p}(S)\cdot\mathbb{P}_{p}\big{[}0\overset{}{% \mathrel{\mathchoice{\ooalign{$\hfil\displaystyle/\hfil$\cr$\displaystyle% \leftarrow\mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$\hfil\textstyle/% \hfil$\cr$\textstyle\leftarrow\mathrel{\mskip-3.1mu }\rightarrow$}}{\ooalign{$% \hfil\scriptstyle/\hfil$\cr$\scriptstyle\leftarrow\mathrel{\mskip-3.1mu }% \rightarrow$}}{\ooalign{$\hfil\scriptscriptstyle/\hfil$\cr$\scriptscriptstyle% \leftarrow\mathrel{\mskip-3.1mu }\rightarrow$}}}}\partial\Lambda_{n}], (2.13)

as desired. ∎

#### Acknowledgments

This work was supported by a grant from the Swiss FNS and the NCCR SwissMap also founded by the swiss NSF.

## References

• [AB87] Michael Aizenman and David J. Barsky. Sharpness of the phase transition in percolation models. Comm. Math. Phys., 108(3):489–526, 1987.
• [DCT15] Hugo Duminil-Copin and Vincent Tassion. A new proof of the sharpness of the phase transition for Bernoulli percolation and the Ising model. preprint, 2015.
• [Gri99] Geoffrey Grimmett. Percolation, volume 321 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, second edition, 1999.
• [Ham57] J. M. Hammersley. Percolation processes: Lower bounds for the critical probability. Ann. Math. Statist., 28:790–795, 1957.
• [Men86] M. V. Menshikov. Coincidence of critical points in percolation problems. Dokl. Akad. Nauk SSSR, 288(6):1308–1311, 1986.
• [vdBK85] J. van den Berg and H. Kesten. Inequalities with applications to percolation and reliability. J. Appl. Probab., 22(3):556–569, 1985.

Département de Mathématiques Université de Genève Genève, Switzerland E-mail: hugo.duminil@unige.ch, vincent.tassion@unige.ch

You are adding the first comment!
How to quickly get a good reply:
• Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
• Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
• Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters