Non-Gaussian fixed points of the block spin transformation. Hierarchical model approximation (Q795414)

The paper is concerned with the study of the critical point of the classical statistical systems with non-Gaussian long distance behaviour and of its scaling limit. An example of such systems is a lattice model with the Gibbs state given formally by \(Z^{-1}\exp [- \sum_{x}v(\phi_ x)d\mu_ G(\phi)]\), where \(\phi_ x\in {\mathbb{R}}^ N\), \(x\in {\mathbb{Z}}^ d\), \(d\mu_ G\) is the Gaussian measure with mean zero and covariance G with \(| G_{xy}| \sim | x-y|^{- \alpha}\) for large \(| x-y|\) and \(\alpha<{1\over2}d\). An example of \(v(\phi)\) is \(v(\phi)={1\over2}m^ 2\phi^ 2+\lambda(\phi^ 2)^ 2\), \(m^ 2\in {\mathbb{R}}\), \(\lambda>0.\) For the critical value of \(m^ 2\) when correlation lengths is infinite the system goes under renormalization group transformation to a fixed point. The perturbation methods for small \(\epsilon ={1\over2}d-\alpha\) or \(N^{-1}\) shows that the fixed point is non-Gaussian. The authors offer a procedure that controls corrections to the leading nontrivial order in \(\epsilon\) and \(N^{-1}\) for the renormalization group transformation. This is done for the hierarchical approximation, but they are convinced that the method works also for the general case. The idea is to find a neighbourhood of the approximate fixed point preserved by the renormalization group transformation except for one relevant direction, and then to show that the distance between two successive iterations within this neighbourhood shrinks at the critical point. In both cases (small \(\epsilon\) and large N) the analyticity techniques developed earlier by the authors are employed.