A Subspace Preconditioned LSQR Gauss-Newton Method with a Constrained Line Search Path Applied to 3D Biomedical Microwave Imaging

Figure 3

Illustration of the constrained line search path: (a) the starting point lies far from the boundaries of the optimization domain, such that the new search path from (23) coincides with the original path from (3) up to large ; (b) the starting point lies close to a boundary of the optimization domain, such that the new search path from (23) starts running along the projection of (3) on this boundary already for small .

(a)

(b)