No label defined (Q4934263)

The paper is written by the founder and leading expert in the theory of antieigenvalues, who introduced them for strongly accretive operators \(\text{Re }A> 0\) as a cosine of a maximal angle through which the operator \(A\) may turn the vector \[ \mu= \cos A=\inf_{x\in\partial(A)} {\text{Re}\langle Ax,x\rangle\over\|Ax\|\|x\|}, \] the vector \(x_\mu\) for which \[ {\text{Re}\langle Ax_\mu,x_\mu\rangle\over\|Ax\|\|x\|}= \mu \] is called the antieigenvector. The antieigenvalues appear naturally in gradient-based iterative algorithms for solutions of linear systems with positive operators \(Ax- b=0\). Solution to this equation is equivalent to minimization of the quadratic form \(Q(x)={1\over 2}\langle x,Ax\rangle-\langle b,x\rangle\). The method of steepest descent gives the iterations \[ x_{k+1}= x_k- {\|Ax_k- b\|^2(Ax_k- b)\over \langle A(Ax_k- b), Ax_k-b\rangle}, \] which are ``trigonometrically convergent'' to the solution \(x^*\) in \(A\)-norm \(\langle A(x_n-x^*), (x_k-x^*)\rangle= E_A(x_k)\): \[ E_A(x_k)\leq (1- \cos^2A) E_A(x_{k-1})\equiv\sin^2 AE_A(x_{k-1}). \] This geometrical interpretation of the Kantorovich result [\textit{L. V. Kantorovich}, ``Selected works, II''. Amsterdam (1996; Zbl 1061.01506)] was obtained by the authors one decade ago. A new result mentioned in the paper is concerning the antieigenvectors: they fulfill the ``Euler equation'' \[ 2\|Ax\|^2\|x\|^2(\text{Re }A) x-\|x\|^2\text{Re}\langle Ax,x\rangle A^*Ax-\|Ax\|^2\text{Re}\langle Ax,x\rangle x= 0. \] For the entire collection see [Zbl 0927.00013].