The geometry of basic, approximate, and minimum-norm solutions of linear equations (Q1805212)

For an equation \(Ax = b\) with a rectangular matrix \(A\), the basic solutions are defined as the solutions of subsystems corresponding to maximal nonsingular submatrices of \(A\). Their convex hull is denoted by \(C\). The scaled \(\ell_ p\)-approximate solutions are the minimizers of \(\| M (Ax - b) \|_ p\). The authors prove: The set of all scaled \(\ell_ p\)-approximate solutions, where \(M\) runs over all positive diagonal matrices, is independent of \(p\) for \(1 < p < \infty\). The set of all scaled least- squares solutions has the same closure as the set of solutions of \(\min (f(| Ax - b |))\), where \(f\) runs over all strictly isotone functions. For \(1 \leq p_ 1, p_ 2 < \infty\), the set \(C\) contains all minimum -\(\ell_{p_ 2}\)-norm \(\ell_{p_ 1}\)-approximate solutions. For \(A\) of full row rank and \(f\) isotone, there exists a solution of \(\min \{f (| x |) : Ax = b\}\) in \(C\). If \(f\) is strictly isotone, then every solution lies in \(C\).