The McEliece-Rodemich-Rumsey-Welch (MRRW) bound is an asymptotic upper bound for the rate of a binary code of
relative minimum distance
. Define the binary entropy function and an auxiliary function
by
|
(1)
|
with . For
, define
|
(2)
|
The first MRRW exponent is
|
(3)
|
and the optimized second MRRW exponent is
|
(4)
|
For , the classical bound is
(McEliece et al. 1977).
OpenAI (2026) enlarged the two classical variational families as follows. For ,
set
|
(5)
|
and
|
(6)
|
For parameters
|
(7)
| |||
|
(8)
|
define
|
(9)
| |||
|
(10)
| |||
|
(11)
| |||
|
(12)
| |||
|
(13)
|
Let be the set of quadruples
satisfying these ranges and
|
(14)
|
The constant-weight exponent is
|
(15)
|
Finally, the new binary-code exponent is
|
(16)
|
and OpenAI proved
|
(17)
|
The comparison is strict for every fixed in this interval. The boundary choices
and
recover the classical
and constant-weight families. Positive harmonic degrees
strictly improve every interior minimizing layer; when the endpoint
minimizes
, so that
, the whole-cube family instead gives the strict improvement.