The absorption laws for the generalized inverses (Q2250231)

Let \(\mathcal H\) and \(\mathcal K\) be Hilbert spaces. The Moore-Penrose (generalized) inverse \(A^{\dagger}\) (if it exists) of a bounded linear operator \(A\) from \(\mathcal H\) into \(\mathcal K\) is the unique bounded linear operator \(X\) from \(\mathcal K\) into \(\mathcal H\) that satisfies the equations \(AXA=A\), \(XAX=X\), \((AX)^*=AX\) and \((XA)^*=XA\). It is well known that \(A^{\dagger}\) exists if and only if \(R(A)\) is closed. Let \(A\) and \(B\) be bounded linear operators from \(\mathcal H\) into \(\mathcal K\) with closed ranges. The authors study the question of when the following ``absorption'' law holds: \(A^{\dagger}+B^{\dagger}=A^{\dagger}(A+B) B^{\dagger}\). A characterization is given in terms of certain inclusions among the range spaces associated with \(A, B\) and their adjoint operators. The law is then extended to include various classes of generalized inverses. A necessary and sufficient condition is presented for each class. Some mixed laws are also studied. All the proofs rely rather heavily on block operator representations.