Trying to figure out how different fMRI softwares will model events that have a 0 second duration.
SPM will include those in its design matrix. From memory I think those are included and given a duration equal to the time bin duration of the temporal resolution at which the convolution with HRF is done.
I have no knowledge of the internals of AFNI and FSL (or other tools that do fMRI analysis).
For FSL, I think they treat 0-duration events as 0.05s (or at least they did way back in 2005). Here’s a post on their mailing list that might be helpful:
In AFNI, users specify the actual basis function. For an “instantaneous” event, I would expect one to use SPMG1 (or 2 or 3), GAM, or BLOCK with a short duration.