Gradient descent based wavefront reconstruction using multiple wavelength measurements for a scalar and a vector ZWFS. The input wavefront is described by a set of modal coefficients, θ. The current estimate is propagated through a model of the considered ZWFS which produces the expect output intensity images at the considered wavelengths. These are compared to the measured intensity images and the similarity is captured in a cost function J(θ). The gradients of this cost function with respect to θ are calculated by a back-propagation process and these are used to update the current estimate by using an optimizer such as the Newton-CG.