Sparse block factorization of saddle point matrices

The factorization method presented in this paper takes advantage of the special structures and properties of saddle point matrices. A variant of Gaussian elimination equivalent to the Cholesky's factorization is suggested and implemented for factorizing the saddle point matrices block-wise with small blocks of orders 1 and 2. The Gaussian elimination applied to these small blocks on block level also induces a block 3×33×3 structured factorization of which the blocks have special properties. We compare the new block factorization with the Schilders' factorization in terms of sparsity and computational complexity. The factorization can be used as a direct method, and also anticipate for preconditioning techniques.