André Freitas

Turbulence is defined by a set of emergent and singular phenomena: a constant-flux energy cascade, intermittency, a dissipation rate that survives the vanishing of viscosity, and the spontaneous loss of determinism of the inviscid limit. Which physical ingredients produce them remains unclear, because in fluids they always come together with incompressibility, spatial structure and vortex dynamics. Here we show that a hierarchy of inertia, coupled by the purely geometric nonlinearity of rigid constraints, is enough. A planar chain of rigid pendula whose lengths and masses decrease geometrically, coupled only through nearest-neighbor constraints, develops a far-from-equilibrium cascade that replaces thermal equipartition, with statistically local and time-irreversible energy transfer despite intrinsically nonlocal coupling. Its fluctuations are intermittent, yet self-similarity is restored by a hidden scale symmetry. As the viscosity vanishes, the dissipation rate saturates at a finite value, and thermal noise consistent with fluctuation-dissipation balance is amplified from the smallest to the largest scales within a fraction of a large-scale period. Because the hierarchy can be tuned and the same dynamics can be realized mechanically or electrically, turbulence becomes a property that can be designed and measured in the laboratory.

We introduce a Mori–Zwanzig graph neural network (MZ–GNN) framework for learning reduced-order Lagrangian dynamics of tracer particles in homogeneous isotropic turbulence. The model represents particle acceleration as a finite-memory expansion over present and delayed particle-neighborhood graphs, with each memory contribution parameterized by an equivariant message-passing graph neural network. By construction, the architecture respects the relevant physical symmetries of the problem, including permutation equivariance, Galilean invariance, and equivariance under rotations and reflections. Trained on direct numerical simulation data, the model is rolled out autoregressively and evaluated on observables that are not imposed during training. We show that memory is essential for recovering the intermittent, heavy-tailed acceleration statistics, and that the learned dynamics accurately reproduce single-particle dispersion, pair-dispersion statistics, and four-particle tetrad geometry. Our results establish a physically structured, scalable route to data-driven multi-particle simulation of turbulent transport, and a template for learning reduced dynamics of correlated, symmetry-rich particle systems.

Deterministic closures for coarse-grained turbulence models help reproduce mean statistics, but often fail to capture the finite-time growth of uncertainty. Using the framework of shell models as a quantitative multi-scale testbed, we compare fully resolved simulations with large-eddy simulations using either stochastic or deterministic subgrid closures. While in the fully resolved system a single microscopic perturbation is rapidly amplified by strongly chaotic dynamics, truncation produces a strong delay and suppression of variance growth when uncertainty is introduced through initial condition perturbations only. We show that a data-driven Langevin-type stochastic closure restores the correct timing and magnitude of variance growth across scales, demonstrating that sustained stochasticity is essential for predictability in reduced turbulent dynamics.