Correlation length of the two-dimensional random field Ising model via greedy lattice animal (Q6117108)

In this paper, a notion of correlation length for the two-dimensional random field Ising model with boundary conditions is considered and shown to scale as \(\exp(\Theta(\varepsilon^{4/3}))\) in the zero-temperature regime, with the upper bound remaining valid for all temperatures. The two-dimensional random field Ising model (RFIM) with boundary conditions is a random measure on \(\mathbb{Z}^2\) defined via a Hamiltonian that consists of three terms: the `usual' Ising interaction inside the box \(\Lambda_N\) of side length \(2N\) centred at the origin, a boundary interaction summing over products of neighbouring spins lying on either side of the boundary of \(\Lambda_N\) and finally a random forcing term, here given in terms of a field of i.i.d Gaussian random variables \(\varepsilon h_v\) of variance \(\varepsilon^2\). There are two ways the boundary conditions can enter the Hamiltonian, either with a positive or a negative sign. Formally, the two Hamiltonians are defined as \[ H^\pm(\sigma,\Lambda_n,\varepsilon)=-\left(\sum_{\substack{u\sim v\\ u,v\in\Lambda_N}} \pm \sum_{\substack{u\sim v\\ u\in\Lambda_N\\ v\notin \Lambda_N}} + \sum_{u\in\Lambda_N} \varepsilon h_u \sigma_u \right). \] where \(u\sim v\) means that vertices \(u\) and \(v\) are neighbours in \(\mathbb {Z}^2\), i.e., one coordinate is equal and the other differs by 1. The RFIM random measure is now defined as the Gibbs measure \(\mu_{\beta,\Lambda_N,\varepsilon}^\pm\) on \(\{-1,1\}^{\mathbb{Z}^2}\) with inverse temperature \(\beta\): \[ \mu_{\beta,\Lambda_N,\varepsilon}^\pm(\sigma) \frac1Z \exp\left\{-\beta H^\pm(\sigma,\Lambda_n,\varepsilon) \right\}. \] The notion of correlation length for this random measure is defined to be critical scale where the influence of the random field on the spin at zero is equivalent to the influence of the boundary conditions on the spin at zero. More precisely, let \(m_{\beta,\Lambda_N,\varepsilon}\) be the expected differences of the first moment of the spin at zero under \(\mu^+\) and \(\mu^-\) and fix \(m\in (0,1)\). Then the correlation length \(\psi(\beta,m,\varepsilon)\) is defined as the smallest box size \(N\) such that \(m_{\beta,\Lambda_N,\varepsilon} \leq m\). (If \(m_{\beta,\Lambda_N,\varepsilon}\geq m\) for all \(N\in\mathbb N\) then \(\psi(\beta,m,\varepsilon)=\infty\) by definition.) The main result of the paper can now be stated in terms of upper and lower bounds on the growth of \(\Psi(\beta,m,\varepsilon)\) as \(\varepsilon\) goes to zero: for each \(m\in (0,1)\) there exists a constant \(C(m)>0\) such that \(\Psi(\beta,m,\varepsilon)\leq \exp (C(m)\varepsilon^{-4/3})\) for any temperature \(\beta\), with a corresponding lower bound by \(\exp (C(m)^{-1}\varepsilon^{-4/3})\) that holds in the zero-temperature setting, i.e. when \(\beta=\infty\). The result is proved by relating the correlation length to the value of the \textit{greedy lattice animal} normalised by its boundary, via a geometric argument. The latter is defined as the maximum of \(1/|\partial A| \sum_{v\in A} h_v\), as \(A\) ranges over connected (respectively simply connected) subsets of \(\Lambda_N\) that contain the origin (the \textit{lattice animals}). It had previously been known that both greedy lattice animals grow in expectation with \(N\) as \((\log N)^{3/4}\).