Sparse Matrix Binormalization on a GPU

In considering the problem of binormalizing a real symmetric algebraic system of equations, an iterative algorithm (BIN) was proposed by Livne and Golub. Numerical experiments and theoretical results show that this algorithm is capable of significantly reducing the condition number relative to other available algorithms. The BIN algorithm is approximately two to four times more expensive than “näıve” diagonal scaling algorithms. In this paper, an NVIDIA GeForce G80 GPU is used to mitigate this expense.