Article
Keywords:
imperfect conjugate gradient algorithm; symmetric, positive definite matrix; biorthogonalization; line searches; global efficiency
Summary:
A new biorthogonalization algorithm is defined which does not depend on the step-size used. The algorithm is suggested so as to minimize the total error after $n$ steps if imperfect steps are used. The majority of conjugate gradient algorithms are sensitive to the exactness of the line searches and this phenomenon may destroy the global efficiency of these algorithms.
References:
[1] M. R. Hestenes E. Stiefel:
The method of conjugate gradients for solving linear systems. J. Res. Nat. Bur. Standards, 49 (1952), 409-436.
DOI 10.6028/jres.049.044 |
MR 0060307
[2] R. Fletcher C. M. Reeves:
Function minimization by conjugate gradients. Соmр. J., 2 (1964), 149-154.
MR 0187375
[3] E. Polak G. Ribiere:
Note sur le Convergence des Methods de Directions Conjuges. Reone Fr. Int. Rech. Oper. 16R1 (1969), 35-43.
MR 0255025
[5] L. C. W. Dixon:
Conjugate Gradient algorithms: Quadratic termination properties without line searches. J. of Inst. of Math. and Applics, 15 (1975), 9-18.
DOI 10.1093/imamat/15.1.9 |
MR 0368429
[8] J. Stoer:
On the Relation between Quadratic Termination and Convergence Properties of Minimization Algorithms, Part I. Theory. Num. Math., 28 (1977) 343 - 366.
MR 0496670 |
Zbl 0366.65027
[9] P. Baptist J. Stoer:
On the Relation between Quadratic Termination and Convergence Properties of Minimization Algorithms, Part II, Applications. Num. Math.,28 (1977), 367-391
MR 0496671