should the ADF calculation be done in the multislice calculator?

[tpchuckles]

Copying in bits of our email discussion here, since I don't see it publicly documented anywhere, and searching my email is a tragic mess.

currently: ADF is done by post-processing the exit wave to sum over an area of reciprocal space.
proposed: propagate, sum, save sum and throw out the rest, before moving to the next timestep or probe position, and do not cache the full exit wave.

~ current approach may not be intuitive to some (e.g., abTEM allows you to define a detector and then the signal on that detector is calculated as multislice is performed). "how do i calculate an ADF image" "well first you need your full exit waves" might be odd to some. although this is what happens inside the microscope (full wave is propagated, detector sums over a given area)...
~ saving off the full exit wave, just so we can sum over an area of reciprocal space, might be wasteful (disk space, memory usage, etc). although this would limit the ability to reprocess, if that's something people care about (e.g. to change your collection angle, you would need to reslice).
~ one may be interested in different regimes: central few Brillouin zones for TACAW, plus a HAADF image. the ability to do both (sum for ADF, then crop in reciprocal space for efficiency for TACAW) would be nice.

[tpchuckles]

summing and then cropping before saving the MultisliceCalculator.wavefunction_data would also help with memory usage for large high-res ADF scans (where many probe positions and many kx,ky points will mean wavefunction_data is enormous.

memmapping (suggestion #43) won't completely solve this problem on its own, because it might still require gargantuan files on disk.

edit: nope: exit_waves_batch (even before wavefunction_data) is going to be enormous too. exit_waves_batch is approximately x,y,kx,ky, whereas wavefunction_data is collected across frames, x,y,t,kx,ky

[tpchuckles]

ADF option added via commits 1b9e137 and 98848e4

calculator creates a dummy HAADFData object, asks it for the mask, applies the mask any time frame_data is encountered (whether it's from cache, while looping probes, or calculated for all probes simultaneously). then calculator returns the HAADFData object in addition to(?) the WFData object.

is it weird to change what we return depending on whether you use the "ADF=True" argument? maybe, but packing the HAADFData into the WFData object or something feels messier.

edit: also adding store_full option so the user can specify if they want to skip creating exitwave_data? (user might want to do ADF on the fly and still get the full x,y,kx,ky cube, e.g., cropped tacaw + ADF. but another user might only be interested in ADF, so the full cube is just a waste of ram and/or disk space once the ADF is computed).