The alignment mask is use during alignment to define the part of the template cube that will actually be used to align particle.
This is the default mode in Dynamo The template is rotated and shifted (virtually, through fourier acceleration). At each posible combination of rotation angles and shift, the mask is also rotated and shifted, defining a moving region inside the template. The rotated and shifted template is compared to the data particle only inside this moving region. This means that Dynamo will try to fit texture like the one contained in the template inside the data, avoiding to drive the alignment by searching for the shape of the template inside the data. This later effect can arise if global correlation is used, i.e., if the mask is directly applied onto the template, and then the masked template compared to the data. In doing so, the zeros outside of the mask would also get compared against the particle.
There is no need to define a smooth mask: the mask is never applied onto the template (never gets pixelwise multiplied against the template). The alignment mask is rather used to define the region inside which template and particle will be compared. This moving region is never produced explicitly: the effect of selecting a moving area is coded inside the Fourier acceleration.
Switching local correlation off
It is possible to switch the local correlation off for certain rounds. In this case, computation time should increase 60% at the cost of a less accurate algorithm: Dynamo will actually just multiply the template with the mask. In this case, it is advised to use a smooth mask, i.e., one with a smooth decay of the boundaries.
In order to to this from dcp, you need to go to Numerical Parameters, then switch on the "advanced" button, make sure that the Cross Correlation checkbox is checked and then locate the parameter "normalized cross correlation" . Put it to zero in the rounds that you want.
Alignment mask during averaging
The alignment mask is used once again during the averaging step. After rotating and shifting the particle along the alignment parameters, the full particle is normalized using the values found inside the mask. The particle is normalized in order to ensure a zero mean and standard deviation one inside the region defined by the alignment mask.
No further use of the alignment mask is performed: not for FSC computations, or filtering.
Alan M. Roseman, Particle finding in electron micrographs using a fast local correlation algorithm, Ultramicroscopy 94 (2003) 225–236