A bit simpler than what you did is...
A = max(a21,a21').^2;
A(1:6:end) = -A(1:6:end);
A
A =
-1 4 9 16 25
4 -4 9 16 25
9 9 -9 16 25
16 16 16 -16 25
25 25 25 25 -25
with no loop required. It does require at least R2016b though to work. It should be doable in one line of code too, but sometimes the extra effort just makes for unreadable code too.
Sigh. You insist on a one-liner? ;-) I suppose this works.
A = toeplitz([-1 1 1 1 1]).*max(a21,a21').^2;
but it would help to know what toeplitz does. As it is, if you are not familiar with toeplitz, then the latter form I wrote tends to look rather strange.
In either case, it is pretty simple to fix in the case where a21 is some vector of completely arbitrary length.
