I haven't used PDL at all, but maybe you have enough dimensions in your piddle to make the following approach efficient:

- Create a matrix to calculate the shifted, negative piddle (can be cached)
- Multiply the source piddle by the matrix to create the shifted negative source piddle
- Add the source piddle and the new piddle

The matrix would be (or, would be)

`( 0 0 0 ...)
( -1 0 0 ...)
( 0 -1 0 ...)
( 0 0 -1 ...)
...
`

Reading from the PDL documentation, basically the following code should work (while PDL is installing):

`# 4 dimensions
my $matrix = pdl([[0,0,0,0],[-1,0,0,0],[0,-1,0,0],[0,0,-1,0]]);
my $pdl = pdl(...);
my $diff = $pdl x $matrix; # not sure about whether to use left- or ri
+ght-multiplication here
my $res = $pdl + $diff;
`

Comment onRe: PDL: efficient self-referencing math?SelectorDownloadCode