James Garrison won second place in the Mathematics category at the Graduate Student Research Symposium

Congratulations to James Garrison on winning second place in the Mathematics category at the Graduate Student Research Symposium.

The title of his poster presentation was Randomized Preconditioned Cholesky QR in Mixed Precision.

The research behind the presentation involved analyzing rpCholesky-QR, a randomized preconditioned Cholesky-QR algorithm for computing the thin QR factorization of full rank, real matrices. The perturbation analysis is transparent and identifies clearly all factors that contribute to error amplification; it requires only a minimal amount of assumptions and produces interpretable bounds, rather than first-order estimates. We adapt our perturbation analysis in to show that preconditioning a Cholesky-QR algorithm in low precision does not affect the accuracy when the input matrix has condition number less than 10^8, and we implement our mixed-precision-Cholesky-QR style algorithm on GPUs in C++ to evaluate its speed.