Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 28: Fluid Physics and Turbulence
DY 28.2: Talk
Tuesday, March 10, 2026, 14:30–14:45, ZEU/0118
Compression, simulation, and synthesis of turbulent flows — •Stefano Pisoni1,2, Raghavendra Peddinti2, Siddhartha Morales2, Egor Tiunov2, and Leandro Aolita2 — 1TUHH, Hamburg, Germany — 2TII Abu, Dhabi, UAE
Numerical simulations of turbulent fluids are paramount to real-life applications. However, they are also computationally challenging due to the intrinsically non-linear dynamics, which requires a very high spatial resolution to accurately describe them. A promising idea is to represent flows on a discrete mesh using tensor trains (TTs), where the values of the velocity field are encoded as a product of matrices (also known as Matrix Product States). This representation features an exponential compression of the number of parameters, under the assumption of low inter-scale correlations. However, it is yet not clear how the achieved compression of TTs is affected by the complexity of the flows. In fact, no TT fluid solver has been extensively validated in a fully developed turbulent regime yet. We fill this gap by analyzing TTs as an Ansatz to compress, simulate, and generate 3D snapshots with turbulent-like features. We first investigate the effect of TT compression on key turbulence statistical signatures. Second, we present a TT solver to time evolve a 3D fluid fields according to the incompressible Navier-Stokes equations. Third, we develop a memory-efficient TT algorithm to generate artificial snapshots displaying turbulent-like features. In all three cases we observe that the memory-efficient TT representation captures the relevant features of turbulent flows, offering a powerful quantum-inspired toolkit for their computational treatment.
Keywords: Tensor Network; Turbulence; Fluid Dynamics; Matrix Product States; Tensor Train