The Friedgut junta inequality says that if and
then there is a real-valued -junta such that
To prove it, put . Part (i), applied to each discrete derivative of a Boolean function , gives
On the other hand, expanding the noise stability in Fourier coefficients gives
Choose and define
The preceding bounds make the low-degree Fourier mass omitted by at most , while the hypothesis makes the high-degree mass at most . Thus . Finally,
which gives the asserted bound on .
Solved by gpt-5.6-sol high.
For a Boolean-valued , each discrete derivative of a Boolean function takes values in and has degree at most . If depends on coordinate , then is nonzero, so part (iii) gives
Since has degree at most , the Fourier formula for total influence and Parseval identity give
If coordinates affect , then , so . Thus is a -junta, which is the Nisan-Szegedy junta theorem.
Solved by gpt-5.6-sol high.
If , monotonicity already gives , so assume . Suppose for a contradiction that . By the mean value theorem, some satisfies
The Margulis-Russo formula identifies this derivative with the appropriately normalized total influence, so is bounded solely in terms of . The -biased Friedgut junta theorem then supplies, for any small , a Boolean -junta with and
Because is monotone, , hence when . It follows that
For some assignment on with , therefore, . Monotonicity and imply . Choose , set , and take . Then
contradicting -quasirandomness. Thus .
Solved by gpt-5.6-sol high.