Renormalization Group-Like Flows in Randomly Connected Tensor Networks

Abstract Randomly connected tensor networks (RCTN) are dynamical systems defined by summing over all the possible networks of tensors. Because of the absence of fixed lattice structure, RCTN is not expected to have renormalization procedures. In this paper, however, we consider RCTN with a real tensor, and it is proven that a Hamiltonian vector flow of a tensor model in the canonical formalism with a positive cosmological constant has the properties that a renormalization group (RG) flow of RCTN would have: The flow has fixed points on phase transition surfaces; every flow line is asymptotically terminated by fixed points at both ends, where an upstream fixed point has higher criticality than a downstream one; the flow goes along phase transition surfaces; there exists a function that monotonically decreases along the flow, analogously to the $a$- and $c$-functions of RG. A complete classification of fixed points is given. Although there are no cyclic flows in the strict sense, they exist, if infinitesimal jumps are allowed near fixed points.