An almost perfect nonlinear (APN) function is a function for which
|
(1)
|
has at most two solutions for every
and every
(Nyberg 1994). All additions are in the vector
space over the two-element finite field. Solutions
occur in pairs
,
so this is the smallest possible differential uniformity. The condition controls
differential behavior and differs from the affine approximation measured by Boolean
function nonlinearity.
The Dobbertin APN function on ,
where
and
, is
|
(2)
|
This family was introduced by Dobbertin (2001). For , define its Walsh coefficient by
|
(3)
|
Kölsch (2026) determines these coefficients through elliptic-curve point counts and Kloosterman sums. In particular, its linearity, the maximum of over
and all
,
is exactly
|
(4)
|
The set of absolute Walsh values occurring is
|
(5)
|
where
for odd
and
for even
. This lists the values, without claiming their multiplicities.
For example, at
the coefficient is
.
Kölsch (2026) reports that ChatGPT suggested the use of Dwork's rationality
theorem in the simplified proof.