What exactly does the option “Use max FSC over classes for filtering” do in heterogeneous refinement?
My assumption is that it means that all the classes are lowpass filtered at each iteration to the highest resolution attained by any class at the end of the previous iteration. Is this correct? Are the per-class FSCs still used to limit the resolution used for alignment in the subsequent iteration, or is that also based on the maximum resolution of any of the classes?
And if this option is switched off, does it then filter and align using the per-class FSCs?
Your assumption is correct – activating this parameter will filter each class by the maximum FSC value across all classes, for each resolution shell. The main use for this is to prevent the case where one class advances to a much higher resolution than all others, causing more particles to be incorrectly classified into that class because of the information imbalance between classes. If it’s off, the per-class FSC will be used to filter each volume.
Regardless of whether this parameter is on, the alignment resolution will always be taken as the max of the 0.143 FSC threshold value across all classes. This is because we must consider the same number of fourier coefficients when aligning/classifying particles for each class, in order to ensure that we can compare values of the posterior across classes in a fair manner.