On condition numbers of a nondefective multiple eigenvalue of a nonsymmetric matrix pencil (Q445818): Difference between revisions

Let \(A,B\in\mathbb{C}^{n\times n}\), and let \(\{A,B\}\) be a regular matrix pencil. Assume that \(\lambda_1\) is a nondefective multiple eigenvalue of \(\{A,B\}\) with multiplicity~\(r\). That is, there exist invertible matrices \(X,Y\in\mathbb{C}^{n\times n}\) such that \(Y^TAX=\lambda_1I_r\oplus A_2\), \(Y^TBX=I_r\oplus B_2\), and \(\lambda_1\) is not an eigenvalue of the generalized eigenvalue problem \(A_2x=\lambda B_2x\). The authors study a condition number of~\(\lambda_1\), defined by \[ c([A,B],\lambda_1)=\sup_{\Phi,\Psi\in\mathbb{C}^{n\times n},\|[\Phi,\Psi]\|_F=1}\rho\big(Y_1^T(\Phi-\lambda_1\Psi)X_1\big). \] Here \(X=[X_1,X_2]\), \(Y=[Y_1,Y_2]\), \(X_1,Y_1\in\mathbb{C}^{n\times r}\), \(\rho\) stands for the spectral radius, and \(\|.\|_F\) is the Frobenius norm. More generally, they consider \[ \tilde{c}([A,B],\lambda_1)=\sup_{\Phi,\Psi\in\mathbb{C}^{n\times n},\|[\Phi,\Psi]\|=1}\rho\big(Y_1^T(\Phi-\lambda_1\Psi)X_1\big), \] where \(\|.\|\) is a normalized unitarily invariant norm. (``Normalized'' means that if \(\mathrm{rank}\,M=1\), then \(\|M\|=\|M\|_2\), the spectral norm.) The condition numbers~\(\tilde{c}\) describe the ``worst case'' sensitivity of~\(\lambda_1\). Since a perturbation of~\(\lambda_1\) generally produces \(r\) simple eigenvalues with different sensitivities, the authors also study \(r\) condition numbers, defined by \[ k_i([A,B],\lambda_1)=\sup_{\Phi,\Psi\in\mathbb{C}^{n\times n},\|[\Phi,\Psi]\|=1}|\mu_i\big(Y_1^T(\Phi-\lambda_1\Psi)X_1\big)|,\quad i=1,\dots,r. \] Here \(\mu_1(M),\dots,\mu_r(M)\) denote the eigenvalues of \(M\in\mathbb{C}^{r\times r}\), ordered \(|\mu_1(M)|\geq\dots\geq|\mu_r(M)|\).