Figure 5.

Number of iterations needed per Bregman iteration. The circulant preconditioner reduces the number of iterations considerably compared with the nonpreconditioned case. The Jacobi preconditioner does not reduce the number of iterations due to the poor approximation of the system matrix’ inverse. (A) depicts the iterations for Set 1: $\left(\mu =1·{10}^{-3},\lambda =4·{10}^{-3},\gamma =1·{10}^{-3}\right)$, whereas (B) depicts the iterations for Set 1, Set 2: $\left(\mu =1·{10}^{-2},\lambda =4·{10}^{-3},\gamma =1·{10}^{-3}\right)$, and Set 3: $\left(\mu =1·{10}^{-3},\lambda =4·{10}^{-3},\gamma =4·{10}^{-3}\right)$ The preconditioner shows the largest speed up factor when the regularization parameters are well‐balanced. (C) Shown are the number of iterations needed per Bregman iteration with and without coil compression applied. The solid lines and the dashed lines depict the results with and without coil compression, respectively.