recommend change to start of diff() so it works generally

[uoa1184615]

Currently, diff(x) and diff(x,2) cause an error. This simple change to AutoDiff.m appears to at least allow diff() to do its job properly.

        function x = diff(x, n, dim)
            if nargin<3, dim=1; end %AJR, 28 May 2026
            if nargin<2, n=1;   end %AJR, 28 May 2026
            if n>1 %AJR, 28 May 2026, recursion probably inefficient
                x = diff(x,n-1,dim);
            end
            if issparse(x.values)

[uoa1184615]

Remembered that diff() operates on the first non-singleton dimension, so change my suggested start of AutoDiff:diff() to more like this:

        function x = diff(x, n, dim)
            if nargin<3, dim=min(find(size(x)>1)); end %AJR, 31 May 2026
            if nargin<2, n=1;   end %AJR, 28 May 2026
            if n>1 %AJR, 28 May 2026, recursion probably inefficient
                x = diff(x,n-1,dim);
            end
            if issparse(x.values)

[uoa1184615]

Also recommend the following tweaks to AutoDiff.m in order to avoid output of x info

        function x = real(x)
            x.derivatives = real(x.derivatives);%added ; AJR, 3 Jun 2026
            x.values = real(x.values);
        end

        function x = imag(x)
            x.derivatives = imag(x.derivatives);%added ; AJR, 3 Jun 2026
            x.values = imag(x.values);
        end

[uoa1184615]

Question: any chance of getting AutoDiff to encompass fft() and ifft()?

[uoa1184615]

On fft() and ifft(): have found that Chris Noble has coded these within github.com/noblec04/MatlabGP, to be included in AutoDiff(). I suggest including the following vectorised versions to speed-up by factor of roughly 10 (on my laptop).

        function y = fft(x,varargin)
        % fft(), C.2025 by Chris Noble, github.com/noblec04/MatlabGP
        % revised AJR 7 Jun 2026.  Question: perhaps force sparse derivs?
            y.values = fft(x.values,varargin{:});
            nz = size(x.derivatives,2);
            if 0 % original version by CN
            for i = 1:nz
                dxi = reshape(full(x.derivatives(:,i)),size(x.values));
                dyi = fft(dxi,varargin{:});
                y.derivatives(:,i) = dyi(:);
            end%for
            else % AJR, vectorized version may be roughly 10x faster
                Dxi = reshape(full(x.derivatives),[size(x.values) nz]);
                Dxi = fft(Dxi,varargin{:});
                y.derivatives = reshape(Dxi,size(x.derivatives));
            end%if
            y = AutoDiff(y);
        end

        function y = ifft(x,varargin)
        % ifft(), C.2025 by Chris Noble, github.com/noblec04/MatlabGP
        % revised AJR 7 Jun 2026.  Question: perhaps force sparse derivs?
            y.values = ifft(x.values,varargin{:});
            nz = size(x.derivatives,2);
            if 0 % original version by CN
            for i = 1:nz
                dxi = reshape(full(x.derivatives(:,i)),size(x.values));
                dyi = ifft(dxi,varargin{:});
                y.derivatives(:,i) = dyi(:);
            end%for
            else % AJR, vectorized version may be roughly 10x faster
                Dxi = reshape(full(x.derivatives),[size(x.values) nz]);
                Dxi = ifft(Dxi,varargin{:});
                y.derivatives = reshape(Dxi,size(x.derivatives));
            end%if
            y = AutoDiff(y);
        end