One could write an algorithm for implementing functions of thisnode exchanges in the manner of the following pseudo-code.
loop over n x@n = y@f(n) end
It remains to sort out what is the relationship between this algorithm, and the split algorithm in section 8. In this we need to consider why we use the approach in section 8.
One assumption is that every process receives. There is also the identity case where p ``sends'' to p.
In some cases it may be more efficient to distribute the calculation of f(me) and gather the results.