“…Alternatively, we can also frame phase retrieval as a nonlinear minimization problem, where we minimize an error metric using a gradient-based approach. The gradient-based approach is flexible and can include in the forward model a large variety of the physical phenomena related to the probing light (such as partial coherence [17], source fluctuations [18], and errors in positions [19,20]), or the detection process (such as the measurement noise [21,22] and the finite size of the pixel [23]). As such, this method has been the focus of much recent literature, leading to the development of steepest descent methods [16,24,25], conjugate gradient methods [4,16,26], Gauss-Newton methods [27], and quasi-Newton methods [28].…”