The question of when to stop iterating is a matter of judgment. An insufficient number of iterations will result in an image that is not completely deconvolved. An excessive number will cause noise amplification, particularly in areas with faint, continuous details. General guidelines are as follows:
The processed image should obviously look sharper than the original version. Often the brighter areas will deconvolve first, followed by the fainter features.
Some images are much more difficult to deconvolve than others. Often a dozen iterations are sufficient; occasionally 40 or more are required. If the image converges too slowly, check the PSF and Noise model settings.
It is quite possible to overdo it. Once the image is sharpened, very little additional improvement will be seen. But the algorithm will continue to try to deconvolve the noise floor, which was not blurred by the PSF. The background will start looking noisy and ”curdled.”