Minimizing convex quadratic with variable precision Krylov methods

Iterative algorithms for the solution of convex quadratic optimization problems are investigated, which exploit inaccurate matrix-vector products. Theoretical bounds on the performance of a Conjugate Gradients and a Full-Orthormalization methods are derived, the necessary quantities occurring in the theoretical bounds estimated and new practical algorithms derived. Numerical experiments suggest that the new methods have significant potential, including in the steadily more important context of multi-precision computations.

[1]  Valeria Simoncini,et al.  Theory of Inexact Krylov Subspace Methods and Applications to Scientific Computing , 2003, SIAM J. Sci. Comput..

[2]  Serge Gratton,et al.  Linear regression models, least-squares problems, normal equations, and stopping criteria for the conjugate gradient method , 2012, Comput. Phys. Commun..

[3]  Andrew J. Wathen,et al.  Stopping criteria for iterations in finite element methods , 2005, Numerische Mathematik.

[4]  Iain S. Duff,et al.  Stopping Criteria for Iterative Solvers , 1992, SIAM J. Matrix Anal. Appl..

[5]  M. Arioli,et al.  A stopping criterion for the conjugate gradient algorithm in a finite element method framework , 2000, Numerische Mathematik.

[6]  Gerard L. G. Sleijpen,et al.  Inexact Krylov Subspace Methods for Linear Systems , 2004, SIAM J. Matrix Anal. Appl..

[7]  M. Hestenes,et al.  Methods of conjugate gradients for solving linear systems , 1952 .

[8]  Mark Horowitz,et al.  Energy-Efficient Floating-Point Unit Design , 2011, IEEE Transactions on Computers.

[9]  Yousef Saad,et al.  Iterative methods for sparse linear systems , 2003 .

[10]  Mario Arioli,et al.  Generalized Golub-Kahan Bidiagonalization and Stopping Criteria , 2013, SIAM J. Matrix Anal. Appl..

[11]  C. Kelley.  Iterative Methods for Linear and Nonlinear Equations , 1987 .

[12]  Serge Gratton,et al.  Guaranteeing the convergence of the saddle formulation for weakly constrained 4D‐Var data assimilation , 2017, Quarterly Journal of the Royal Meteorological Society.

[13]  Serge Gratton,et al.  Range-Space Variants and Inexact Matrix-Vector Products in Krylov Solvers for Linear Systems Arising from Inverse Problems , 2011, SIAM J. Matrix Anal. Appl..

[14]  J. van den Eshof,et al.  Relaxation strategies for nested Krylov methods , 2003 .

[15]  Mark Horowitz,et al.  FPMax: a 106GFLOPS/W at 217GFLOPS/mm2 Single-Precision FPU, and a 43.7GFLOPS/W at 74.6GFLOPS/mm2 Double-Precision FPU, in 28nm UTBB FDSOI , 2016, ArXiv.

[16]  Robert H. Halstead,et al.  Matrix Computations , 2011, Encyclopedia of Parallel Computing.