Meaning
Phase retrieval in coherent diffraction imaging relies on computationally solving the phase problem from measured intensity data. The family of iterative transform algorithms works by repeatedly moving a wavefront estimate between the spatial domain and the Fourier domain using Fourier transforms. At each domain, the estimate is modified to comply with known physical constraints, such as the measured intensity or object support.
This iterative process continues until the difference between the estimated and measured intensities falls below a defined threshold.
Constraint Enforcement
Domain constraints are applied at each step to guide the estimated field toward the true solution. In the Fourier domain, the magnitude of the calculated wavefront is replaced by the square root of the measured diffraction intensity, while the phase is kept. In the spatial domain, constraints like positivity or a support mask are enforced.
These iterative transform algorithms use these updates to progressively eliminate unphysical phase solutions.
Convergence Speed
Progress of the reconstruction depends on the specific algorithm variant used, such as the error-reduction or hybrid input-output method. The error-reduction method is stable but converges slowly, while the hybrid input-output method is faster and better at escaping local minima. Combining these methods in a scheduled sequence yields the most reliable results.
Stagnation Risk
Algorithm execution can become trapped in local minima, producing reconstruction artifacts that resemble a partially solved phase map. This stagnation occurs when the support constraint is too loose or the measured data is noisy. To overcome this, random initializations or dynamic support adjustments are applied.
If the stagnation persists, the reconstructed phase remains ambiguous, which prevents the accurate determination of the sample’s structure. Therefore, monitoring the error metrics during execution is essential to detect stagnation early.