Ph.D. Thesis Defense

Rohini Uma-Vaideswaran

(Faculty Advisor: Dr. Pui-Kuen Yeung)

"Investigations of Lagrangian Intermittency and Inertial Particle Dynamics using Extreme-scale Computing"

Thursday, July 16

10:00 a.m.

Montgomery Knight 317

Abstract: 

The Lagrangian perspective of following individual particle trajectories is fundamental to studies of turbulent dispersion and transport. In this thesis, a high-resolution direct numerical simulation (DNS) framework is developed and deployed to investigate intermittency and inertial particle transport in the Lagrangian frame of reference.

A GPU-accelerated algorithm is presented for tracking both fluid and inertial particles in DNS of isotropic turbulence, scaling up to 35 trillion grid points and tracking up to 1 billion particles, using the world's first exascale computer, Frontier (at Oak Ridge Leadership Computing Facility). Using cubic spline interpolation and a local decomposition with ghost layers, the algorithm achieves good weak and strong scaling near the machine's full capacity at a cost nearly independent of particle count. Recognizing that an accurate measure of communication performance is a key component of optimizing extreme scale pseudo spectral DNS algorithms (and consequently particle tracking), a theoretical model is developed for the peak performance of all-to-all communication. The model is applied to distributed transposes, such as those encountered in the distributed 3D FFTs required by pseudo spectral DNS, and validated on three different network architectures.

Using this framework, three research questions are pursued. First, direct numerical simulations of forced isotropic turbulence on Frontera (at Texas Advanced Computing Center) at Taylor-scale Reynolds numbers of 140 to 1300 are used to examine inertial range scaling of the second-order Lagrangian velocity structure function, by decomposing the velocity increment into convective and local contributions. Results show that a narrower range of temporal versus spatial scales, along with particle displacement effects, governs the observed scaling behavior.

Second, conditional statistics of the pressure Hessian contracted with the velocity gradient tensor given the second and third-order invariants of the latter are investigated to provide insights into the role of the pressure Hessian at high Reynolds numbers. As Reynolds number increases from 390 to 1600, the influence of the pressure Hessian becomes increasingly concentrated in the region of the invariant space where intense strain rates and enstrophy production coincide.

Finally, inertial particles with Stokes drag and added-mass effects are tracked across a range of Stokes numbers and added-mass coefficients at Taylor-scale Reynolds numbers of 390 and 650 to investigate the dynamics of preferential sampling and clustering. It is found that preferential sampling strength varies non-monotonically with Stokes number for both light and heavy particles, and that the peaks of spatial clustering and preferential sampling occur at different Stokes numbers.

Committee:
Dr. Pui-Kuen Yeung (advisor), School of Aerospace Engineering
Dr. Suresh Menon, School of Aerospace Engineering
Dr. Sedina Tsikata, School of Aerospace Engineering
Dr. Chris C. K. Lai, School of Civil and Environmental Engineering
Dr. Charles Meneveau, Johns Hopkins University