Talk by Jennifer Scott (University of Reading and STFC Rutherford Appleton Laboratory)
Автор: ENLA Seminar
Загружено: 2022-02-02
Просмотров: 338
Randomised preconditioning for least squares problems in numerical weather prediction
We consider the large sparse weighted least squares problems that arise in the solution of weak constraint four-dimensional variational data assimilation, a method of significant interest for numerical weather prediction. In this talk, we focus on preconditioning the normal equations. This is challenging because of the size of the system, the cost of products with the system matrix, and the need to severely limit the number of iterations of the CG solver to ensure a forecast is obtained in a timely manner. Exploiting the recent resurgence of randomised methods, we propose using randomised methods to develop new preconditioners that are simple to compute and to apply. We illustrate their effectiveness using a model problem and look at how the number of observations of the dynamical system influences performance.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
