Randomized methods to characterize large-scale vortical flow networks.

Area of Science:

• Fluid Dynamics
• Network Theory
• Computational Science

Background:

• Network theory reveals connectivity and dynamics in vortical flows.
• Analyzing high-dimensional turbulent flows using network methods is computationally expensive.
• Efficient approximation techniques are needed for large-scale flow analysis.

Purpose of the Study:

• To apply randomized linear algebra methods for efficient network-based analysis of vortical flows.
• To approximate key network quantities, like eigendecomposition, using randomized techniques.
• To reduce computational cost and memory requirements for high-dimensional flow analysis.

Main Methods:

• Utilized randomized linear algebra, specifically the Nyström method.
• Approximated the leading eigendecomposition of the adjacency matrix.
• Applied quasi-uniform and uniform column sampling strategies.

Main Results:

• Demonstrated significant computational savings and reduced memory usage.
• Successfully applied the technique to two-dimensional flow past an airfoil and two-dimensional turbulence.
• Found quasi-uniform column sampling to be more effective than uniform sampling at the same computational complexity.

Conclusions:

• Randomized methods offer an effective and efficient approach for network-based analysis of complex vortical flows.
• The Nyström method provides substantial computational and memory benefits for high-dimensional turbulent flows.
• Sampling strategies impact the performance of randomized network analysis.