Building a better nonuniform fast Fourier transform
The NUFFT allows Fourier analysis of data on non-uniform points at close-to-FFT speeds. I will overview its applications in science and engineering, including the 3D reconstruction of proteins from sets of 2D cryo electron microscopy images. The main part of the talk explains what happens "under the hood" in our new implementation (FINUFFT). This includes 1) a simpler spreading kernel that accelerates run-times for the same accuracy, while preserving a rigorous error analysis, and 2) smart multi-threaded blocking. Along the way we will discover how the nationally known bluegrass fiddler Tex Logan fits into the story.
Joint work with Jeremy Magland, Ludvig af Klinteberg, and Yu-Hsuan Shih.
Alex Barnett is group leader in numerical analysis at the Center for Computational Mathematics at the Flatiron Institute, a division of the Simons Foundation. He was a faculty member in the mathematics department at Dartmouth College for 12 years, becoming a full professor in 2017. He obtained his Ph.D. in physics at Harvard University, followed by a postdoctoral fellowship in radiology at Massachusetts General Hospital and a Courant Instructorship at New York University. His research interests include partial differential equations, integral equations, fast algorithms, wave scattering, fluid flow, imaging, neuroscience, quantum chaos, scientific computing and software libraries.