The merit factor problem asks for binary sequences with small off-zero correlations. Let be the set of sequences
with
and
for
, ...,
, and define
|
(1)
|
for , 1, ...,
. Then the merit factor problem
requires the minimization of
over
for a fixed positive integer
.
For , 2, ..., the first few minima are 5, 10, 18, 27,
43, 52, 72, ... (OEIS A091386).
Finding optimal sequences is computationally difficult (Borwein and Bailey 2003, p. 6).
For a sequence of length
, its merit factor is
|
(2)
|
where the
are its aperiodic autocorrelations.
The zero-shift value
is excluded from this denominator.
Consequently, maximizing
is equivalent to the minimization above, where
is fixed. Changing every sign preserves
, so requiring
does not affect the extremum.
For the associated Littlewood polynomial , Parseval's
theorem gives
|
(3)
|
Thus large merit factor is equivalent to a normalized fourth raw moment close to 1 (Jedwab et al. 2013). OpenAI (2026) announced that the
largest merit factor at length tends to infinity as
through all integers.
This follows already from the reported construction with
for every fixed
and all sufficiently large
. Indeed, the fourth raw moment
is at most
, so
|
(4)
|
This consequence does not require the stronger two-sided uniform claim about Littlewood polynomials.