# Nufftn yields only zeros for nonuniform grid. Bug?

1 view (last 30 days)
schlapp on 28 Apr 2020
Edited: schlapp on 29 Apr 2020
Following up a previous question from myself on this site, I decided to ask again about my second point more precisely.
My main issue is that in my test-case of a simple Gaussian the built-in MATLAB function 'nufftn' only gives a matrix of zeros as a result when using the nonuniform grid 'pcrop.mat', provided in the attachment.
in other words:
fcrop = exp(-(pcrop.^2 + pcrop'.^2 + pcrop.*pcrop'));
ftcrop = reshape(nufftn(fcrop,{pcrop,pcrop'},{pcrop,pcrop'})/2/pi,length(pcrop),length(pcrop));
yields 'ftcrop' as 2d matrix full of zeros.
Performing the same task with a interpolated uniform grid
puni=[pcrop(1):2*pcrop(end)/(length(pcrop)-1):pcrop(end)];
funi = exp(-(puni.^2 + puni'.^2 + puni.*puni'));
ftuni = reshape(nufftn(funi,{puni/sqrt(2*pi),puni/sqrt(2*pi)'},{puni/sqrt(2*pi),puni/sqrt(2*pi)'})/2/pi*(2*puni(end)/length(puni))^2,length(puni),length(puni));
gives the correct matrix 'ftuni'.
Is this a bug, or am I using the function in a wrong way? Unfortunately I dont get any error messages.

Chris Turnes on 29 Apr 2020
Yes, this appears to be a bug. Please open a case with technical support.
schlapp on 29 Apr 2020
Thank you for the feedback, I will open a case then. I guess this can happen with new features :)