Palestrante: Osni Marques - Lawrence Berkeley National Laboratory
Resumo: It is well known that the SVD of a matrix A, A=U*S*V^T, can be obtained from the eigenpairs of the matrix C_v=A^T*A or C_u=A*A^T. Alternatively, the SVD can be obtained from the eigenpairs of the augmented matrix C_uv= [ 0 A, A^T 0 ]. This presentation focuses on C_uv in the context of bidiagonal matrices and Sca/LAPACK's tridiagonal eigensolvers, and discusses accuracy and implementation issues.
Observação: o palestrante foi convidado pelo Centro de Ciências Matemáticas Aplicadas à Indústria (CEPID - CeMEAI).