Deconvolution is performed using an iterative procedure. This means it starts with a blank image and adjusts it in steps until the desired output is obtained. The decision on when to stop iterating is a difficult one, and is the subject of some debate (see below).
The image may not converge if the input parameters have not been set properly. This usually becomes apparent quickly as the image fades out, bounces back and forth between two states, or just remains fuzzy. The rate of convergence varies from image to image, depending on the distribution of information in the image. A dozen iterations may be required for the image to converge.
The image should be noticeably converging after about five iterations. If not, consider changing the Photoelectrons per ADU, noise model, or point-spread function. The most common reason for convergence failure with the Poisson noise model is an incorrect setting for Photoelectrons per ADU. For the uniform noise model, try adjusting the noise level setting. Increasing the uniform noise level or background level will often make it easier for the algorithm to converge. Note that increasing the uniform noise level or background level too much may reduce the effectiveness of the deconvolution.