A note on condition numbers for generalized inverse \(A^{(2)}_{T,S}\) and constrained linear systems (Q613247)

The authors use the Schur decomposition to obtain the condition numbers of general inverses and constrained singular linear systems. Suppose \(A\in{\mathbb C}^{m\times n}\) with rank \(r\), \(T\) is a subspace of \({\mathbb C}^n\) of dimension \(s\leq r\), and \(S\) is a subspace of \({\mathbb C}^m\) of dimension \(m-n\). If \(X\in{\mathbb C}^{n\times m}\) satisfies \(XAX=X\), \(\mathcal{R}(X)=T\), \(\mathcal{N}(X)=S\), then \(X\) is called the generalized inverse \(A_{T,S}^{(2)}\) of \(A\). Given \(A,E\subset{\mathbb C}^{m\times n}\), \(T\subset{\mathbb C}^n\) and \(S\subset{\mathbb C}^m\), there exists \(G\in{\mathbb C}^{n\times n}\) such that \(\mathcal{R}(G)=T\) and \(\mathcal{N}(G)=S\). Assume that \(A\), \(E\) and \(G\) satisfy \(\mathcal{R}(AG)=\mathcal{R}((GA)^*)\), \(\mathcal{R}(GA)=\mathcal{R}((GA)^*)\), \(\mathcal{R}(E)=AT\) and \(\mathcal{R}(E^*)=A^*S^\perp\). Then \(\text{cond}^{TS}(A):=\lim_{\varepsilon\to 0}\sup_{\|E\|\leq\varepsilon\|A\|}\|(A+E)_{T,S}^{(2)}-A_{T,S}^{(2)}\|/\varepsilon\|A_{T,S}^{(2)}\|\) where the matrix norm is induced by the vector norm. In the 2-norm, \(\text{cond}_2^{TS}(A)=\|A\|_2\|A_{T,S}^{(2)}\|_2\) and in the Frobenius norm \(\text{cond}_F^{TS}(A)=\|A\|_F\|A_{T,S}^{(2)}\|_2^2/\|A_{T,S}^{(2)}\|_F\). If the singular linear system \(Ax=b\), \(b\in AT\), \(x\in T\), admits the unique solution \(x=A_{T,S}^{(2)}b\), its condition number is \(\text{cond}^{TS}(A,b):=\lim_{\varepsilon\to 0}\sup_{\|E\|\leq\varepsilon\|A\|,\|f\|\leq\varepsilon\|b\|}\|(A+E)_{T,S}^{(2)}(b+f)-A_{T,S}^{(2)}b\|/\varepsilon\|A_{T,S}^{(2)}b\|\), where \(f\in ET\). Numerical experiments are given.