Fig. 1. Paths of 3D examples for (a) m=1 and (b) m=2 solved by zeroing-Tk and paths for (c) m=1 and (d) m=2 solved by PrSP-Tk. Red arrows: the path in the non-negative space. Dark green dashed doted arrows: the projections of the Newton algorithm projecting points into the solution space. Blue dashed dotted arrows: the zeroing steps projecting points into non-negative space from the solution space. Black arrows: the path of points in the solution space. Mauve dashed arrows: the accelerating step in the solution space.

Fig. 2. (a) Map of the number of iterations needed before the last path could be considered linear for different eigenvalue pairs. (b) The blue curve denotes the maximum number of iterations for given λ1. The black curve denotes the corresponding eigenvalue λi for the maximum number of iterations.

Fig. 3. Reconstruction results of the homogeneous experiment with an EED of 3 mm on the excitation plane by using (a) re-L1-NCG, (b) L1-StOMP, (c) IRL1, (d) zeroing-Tk, and (e) PrSP-Tk, respectively. (f) Profiles along the yellow dashed lines in (a)–(e).

Fig. 4. Reconstruction results of the homogeneous experiment with an EED of 1.5 mm on the excitation plane by using (a) re-L1-NCG, (b) L1-StOMP, (c) IRL1, (d) zeroing-Tk, and (e) PrSP-Tk, respectively. (f) Profiles along the yellow dashed lines in (a)–(e).

Fig. 5. Reconstruction results of the heterogeneous experiment with an EED of 4 mm on the excitation plane by using (a) IRL1 and (b) PrSP-Tk, respectively. (c) Profiles along the yellow dashed lines in (a) and (b).