Research Interests:

Numerical analysis, matrix theory, high-performance scientific computing, and parallel computing.

Current Projects:

Theory and algorithm development for large matrix computation problems arising in computational science and engineering.

Education:

Ph.D., University of Maryland at College Park, 1991
M.S., Academia Sinica, Beijing, China, 1983

To solve larger and more complex problems arising in computational science and engineering, numerical algorithms are re-examined in multiple performance aspects such as real-time efficiency, scalability, and programmability, as well as numerical accuracy and stability. Professor Xiaobai Sun’s recent work has been focused on understanding, characterizing and developing new and fast algorithms to solve large matrix computation problems, removing obstacles in the dogmatic use of conventional/convenient assumptions and approaches to handle emerging, non-traditional problems. The emerging computational problems include, in particular, those arising from Professor Sun’s collaboration work with other researchers on image restoration and identification, telecommunication network performance analysis, design and development of integrated sensing and processing systems.
Professor Sun’s research has contributed to the design of some new block algorithms and parallel algorithms. For instance, she and her colleagues have developed a framework, called the successive band reduction (SBR) approach, for efficiently reducing a large matrix to a condensed form on high-performance computers with memory hierarchy. A software package for the serial implementation of SBR is developed, and the development of a parallel counterpart is in progress.

Matrix theory has been found to be powerful in our work. To simplify the analysis of block algorithms with orthogonal transformations, we have developed a uniform representation, called the basis-kernel representation, for general orthogonal matrices. To ease and improve the implementation of blocked algorithms for high-performance, we have developed a unified theory for aggregating various transformations used in matrix computations.

Numerical algorithm development and computing environment development have mutual impacts on each other. Professor Sun is also involved in cooperative research on programming support and system support for high-performance matrix computations.

Related Links
http://www.cs.duke.edu/~xiaobai/