Melonic turbulence (Q2304968)

A resonant non-linear equation \[ i\frac{d\alpha_j}{dt}=\sum_{S=j}^\infty\sum_{k=0}^S C^S_{jk}\bar{\alpha}_{S-j}(t)\alpha_k(t)\alpha_{S-k}(t) \tag{1} \] which arise in weakly non-liner analysis of PDEs whose frequency spectra of linearized perturbations are highly resonant. This is illustrated taking linearized problem of one-dimensional Schrödinger equation adopting random tensor theory in the Introduction. The main topic is the turbulent properties of solutions of (1), which can be solved as a power series in time. When randomize this power series (cf. [\textit{R. M. May}, Stability and complexity in model ecosystems. With a new introduction by the author. 2nd ed. Princeton, NJ: Princeton University Press (2001; Zbl 1044.92047)]), Feynman-like perturbation theory emerges for averaged Sobolev norms that quantify the strength of the turbulent cascade. It is shown, in the limit of many initially excited low-living model, the dominant terms at each order of this perturbation are given by the melonic graphs (\S4, Definition 1, cf. [\textit{R. Gurau}, Random tensors. Oxford: Oxford University Press (2017; Zbl 1371.81007)]). The Authors say melonic approximation already known as Dirac Interaction Approximation [\textit{R. H. Kraichnan}, J. Math. Phys. 2, 124--148 (1961; Zbl 0268.76038)], it has not been applied to the strongly resonant systems considered in this paper. For the model (1), this paper proves the melonic approximation display energy cascade in the sense of Sobolev norm growth, at lease within a certain initial time interval (\S2 Theorems 1 and 2 proved in \S4). The Sobolev norms are introduced by $S_\gamma(t)=\sum_r r^\gamma\bar{\alpha}_r(t)\alpha_r(t)=\sum_{n\ge 0}s_{\gamma,n}t^n$. Imposing the covariance (\S2 (11)), the family $\{C^S_{jk}\}_{S\in\mathbb{N}}$ can be regrded as a random Gaussian ensemble, for which initial conditions for $\alpha_j$ are drawn (\S2 (12), (13), (14)). Then the main quantities studied in this paper are the averaged Sobolev norms $\bar{S}_\gamma(t)=(S_\gamma(t))_{C,\alpha}=\sum_{n \text{ even}}\bar{s}_{\gamma,n}t^n$. The corresponding quantities obtained from melonic approximation are denoted by $S_\gamma^{melo}(t)$, etc. Then the main theorems of this paper are \textbf{Theorem 1}. The dominant growth as $N\to \infty$ for the averaged Sobolev norm $\bar{S}_\gamma(t)$ are exactly the melonic graphs and the corresponding approximation $S_\gamma^{melo}(t)$ is an analytic function of time in a disk $|T|<\rho$ of finite number $\rho>0$. \textbf{Theorem 2}. For any $\gamma>1$ there exists a constant $\delta$ such that $S_\gamma^{melo}(t)$ growth monotonically in time for $t\in[0,\delta]$. After explaining how to randomize power series in \S2.2, tree expansion of Feynman-like perturbation is explained in \S2.2--2.3. Then from the Sobolev norm, amplitudes $A_r(T)$ and $A_r(G)$ of a tree $T$ and graph $G$ are defined (\S2 (26), (37)). \S3 gives explicit computations of amplitude at order $t^2$, the first nontrivial order at perturbation theory. As conclusion, there are three types of graphs, which have nontrivial amplitudes and other types of graphs have trivial amplitudes. Based on this fact, melonic graphs are defined in \S4 (Definition 1). $A_r(G)$ can be explaced in terms of the face momentum (Lemma 3 proved in Appendix A). By this fact, $A_r(G)$ at fixed order $n$ and as $N\to \infty$, $A_r(G)$ is bounded from above by $N^{-d(G}$ (Proposition 1). Then showing melonic graphs are all dominant and are only dominat graphs (Proposition 3) and local analyticity of $S_\gamma^{melo}(t)$ (Proposition 4), Theorems 1 and 2 are proved in \S4, the last section.