Marginal triviality of the scaling limits of critical 4D Ising and \(\phi_4^4\) models (Q2042072)

The authors prove that the scaling limit of the 4D Ising type models at the critical point is Gaussian. More specifically, for the 4D nearest neighbor Ising model, let \(\Lambda_R=[-R, R]^4\cap\mathbb{Z}^4\). Also, \(\sigma_x\in\{-1, +1\}\) for each \(x\in\Lambda_R\). Thus, \(\sigma\in\{-1, +1\}^{\Lambda_R}\). For each \(g:\{-1, +1\}^{\Lambda_R}\to\mathbb{R}\), define a correlation function as \[\langle g\rangle_\beta= \lim_{R\to+\infty} \frac{\sum_{\sigma\in\{-1,+1\}^{\Lambda_R}} e^{\beta\sum_{|x-y|=1,x,y\in \Lambda_R}\sigma_x\sigma_y}g(\sigma)} {\sum_{\sigma\in\{-1,+1\}^{\Lambda_R}}e^{\beta\sum_{|x-y|=1,x,y\in \Lambda_R}\sigma_x\sigma_y}}\] for positive number \(\beta\). In this definition, each of the functions \(g:\{-1, +1\}^{\Lambda_R}\to \mathbb{R}\) is a restriction from a single function \(g:\{-1, +1\}^{\mathbb{Z}^4}\to\mathbb{R}\). With modest assumption on the function \(g\), the existence of this infinite limit can be established by correlation inequalities (Theorem 4.2.3, page 60 by \textit{J. Glimm} and \textit{A. Jaffe} [Quantum physics. A functional point of view, Springer-Vert (1987)]). To state the result by the authors, first let \(\Sigma_{L, \beta}=\langle(\sum_{x\in \Lambda_L}\sigma_x)^2\rangle_\beta\) for \(0<L\). Let \(f:\mathbb{R}^4\to\mathbb{R}\) be a continuous function with compact support. \(T_{f,L}(\sigma)= \frac{1}{\sqrt{\Sigma_{L,\beta}}}\sum_{x\in\mathbb{Z}^4}f(\frac{x}{L})\sigma_x\). Then, they prove (Proposition 1.4, page 171 of this paper) \[\lim_{L\to+\infty}\langle e^{zT_{f,L}(\sigma)\frac{z^2}{2}\langle T_{f, L}(\sigma)^2\rangle_{\beta_c}}\rangle_{\beta_c}= 1\tag{1}\] at the critical point \(\beta_c\) (please see the discussion near Equation (1.19), page 169), \(z\in\mathbb{R}\). Their result extends to more general models (Proposition 7.2, page 219 of this paper), but for this review, the focus will be the nearest neighbor Ising models at the critical point \(\beta_c\). An idea of their proof is as follows. In equation (1), only the linear term \(z\) and the quadratic term \(z^2\) are present. This is the mark of the Gaussian measure. In general, higher orders terms \(z^4,z^6,\dots\) would appear. The odd terms do not appear due to the symmetry property of the Ising models. The authors show that all the higher-order terms vanish in the limit \(L\to+\infty\). There is an underlying idea/theorem due to \textit{C. Newman} (Theorem 3, page 2 [Commun. Math. Phys. 41, No. 1, 1--9 (1975)]) who showed that for \(\varphi^4\) models, if the fourth-order term is zero, then the sixth-order term is zero, and all higher-order terms are also zero. Indeed, this is the strategy that they invoke. To understand the \(z^4\) term, the main quantity of interest is the 4-point Ursell function \[U^{\beta_c}_4(x,y,z,t)= \langle\sigma_x\sigma_y\sigma_z\sigma_t\rangle_{\beta_c}- [\langle\sigma_x\sigma_y\rangle_{\beta_c}\cdot\langle\sigma_z\sigma_t\rangle_{\beta_c}+\langle\sigma_x\sigma_z\rangle_{\beta_c}\cdot \langle \sigma_y\sigma_t\rangle_{\beta_c}\cdot\langle\sigma_z \sigma_z\rangle_{\beta_c}].\tag{2}\] An estimate for this 4-point Ursell function was already proved by the first author [Commun. Math. Phys. 86, 1--48 (1982; Zbl 0533.58034)] in dimensions greater than 4 (see inequality (1.22) of this paper). In dimension 4, that estimate is not strong enough, and one of the main innovations in this work by the authors is to strengthen that inequality (Theorem 1.3, page 171) in order to prove equation (1) given above. The first issue is to deal with the subtraction that appears in this formula given in (2). Probability measure spaces are defined (Definition 3.1, page 174, equations (3.8) and (3.9), page 176) so that this 4-point Ursell function is expressed in terms of probabilities on these probability measure spaces (equation (3.11), page 177). The 4-point Ursell function expressed in terms of these probability measure spaces allows they authors to localize the main difficulty to understanding the long-distance behavior of \(\langle\sigma_x\sigma_y\rangle_{\beta_c}\) as \(|x-y|\to+\infty\) (Assumption 4.1, page 179 and the subsequent removal of this assumption, Theorem 5.12, page 202) in estimating this 4-point Ursell function. An important aspect of this reduction process is to estimate the deviation of sets within these probability spaces to be independent (Proposition 4.6, page 186). Once the man estimate (Theorem 1.3, page 171) for the 4-point Ursell function is proved, the estimates for the 4-point Ursell function are used to bound the higher-order terms \(z^6,z^8,\dots\) (inequalities (6.44), page 215), and the main result (1) is established. This result by the authors is an important advancement for at least two reasons. First, the nearest neighbor Ising model is one of the fundamental statistical mechanical models. Second, this result does not require any ``weak'' coupling. Prior to this work, the first author [loc. cit.] and \textit{J. Fröhlich} [Nucl. Phys. B 200, 281--296 (1982)] established similar results in dimensions greater than 4. There have been numerous works using the perturbative renormalization group methods in the ``weak'' coupling regime of the \(\varphi^4\) models in dimension 4 (for example [\textit{K. Gawȩdzki} and \textit{A. Kupiainen}, J. Stat. Phys. 35, 267--284 (1984; Zbl 0589.60094); \textit{R. Bauerschmidt} et al., J. Stat. Phys. 157, No. 4--5, 692--742 (2014; Zbl 1308.82026)]).