Consider the probability spaceW={−1, 1} n with the uniform (=product) measure. Letf: W →R be a function. Letf=Σf IXI be its unique expression as a multilinear polynomial whereX I=Π i∈I x i. For 1≤m≤n let\(f_{\hat m} \)=Σ|I|=m f IXI. LetT ɛ (f)=Σf Iɛ|I| X I where 0<ɛ<1 is a constant. A hypercontractive inequality, proven by Bonami and independently by Beckner, states that $$\left| {T_\varepsilon \left( f \right)} \right|_2 \le \left| f \right|_{1 + \varepsilon ^2 } $$ This inequality has been used in several papers dealing with combinatorial and probabilistic problems. It is equivalent to the following inequality via duality: For anyq≥2 $$\left| {f_{\hat m} } \right|_q \le \left( {\sqrt {q - 1} } \right)^m \left| {f_{\hat m} } \right|_2 $$ In this paper we prove a special case with a slightly weaker constant, which is sufficient for most applications. We show $$\left| {f_{\hat m} } \right|_4 \le c^m \left| {f_{\hat m} } \right|_2 $$ where\(c = \sqrt[4]{{28}}\). Our proof uses probabilistic arguments, and a generalization of Shearer’s Entropy Lemma, which is of interest in its own right.
Read full abstract