Overview

BcdiMeso implements phase retrieval in operator-style format. This means that the multiplication (*) and power (^) operators are used to apply operators to some current state. This may look like the following:

state = State(intens, gVecs, recSupport, x, y, z, rho, ux, uy, uz)
optimizeState = OptimizeState(state, primitiveRecipLattice, numPeaks)

optimizeState^100 * state

This short script applies 100 stochastic gradient descent iterations iterations. This makes it easy to implement very complex recipes for phase retrieval algorithms.

API

BcdiMeso.State — Type

State(intensities, gVecs, recSupport, x, y, z, rho, ux, uy, uz)

Create the reconstruction state. intensities is a vector of fully measured diffraction peaks, gVecs is a vector of peak locations, and recSupport is a vector of masks over the intensities that removes those intenities from the reconstruction process. The positions of real space points (x, y, and z) must be passed in as well as the magnitude of the electron density rho and the displacement field (ux, uy, and uz).

The initialization process shifts each peak to be centered (i.e. the center of mass of the peak is moved to the center of the image).

source

BcdiMeso.OptimizeState — Type

OptimizeState(state, primitiveRecipLattice, numPeaks)

Create an object that performs an iteration of stochastic gradient descent. numPeaks number of peaks are selected randomly. One step of gradient descent is taken using the More-Thuente linesearch.

This implimentation takes into account effects of small angle measurement usually ignored in the BCDI problem. This is described in [1] although this implimentation will be faster because a NUFFT is used instead of many FFTs.

source