TCSLib

18 Boolean Analysis — Simple

18.1 Overview

This module develops the “simple” hypercontractivity results for real-valued Boolean functions \(f : \{ 0,1\} ^n \to \mathbb {R}\), that is, the results obtainable by direct induction on coordinates without the full general machinery. It builds up the \((2,4)\), \((4/3,2)\), \((2,2k)\), and \((p,2)\) hypercontractivity inequalities for the noise operator \(T_\rho \), together with the supporting character, cardinality, Fourier, and combinatorial lemmas used along the way.

18.2 Declarations

Lemma 18.1 Character of a lifted set ignores the last coordinate

For a subset \(S \subseteq \mathrm{Fin}\, n\), a point \(x \in \{ 0,1\} ^n\), and a bit \(b\), the parity character of the lifted set equals the character of \(S\) on \(x\):

\[ \chi _{\, \mathrm{castSucc}(S)}(\mathrm{snoc}\, x\, b) = \chi _S(x). \]

That is, appending a last coordinate \(b\) and not including it in the set leaves the character unchanged.

Lemma 18.2 Character of a lifted set with the last coordinate

For \(S \subseteq \mathrm{Fin}\, n\), a point \(x\), and a bit \(b\), adjoining the last coordinate to the lifted set multiplies the character by the sign of \(b\):

\[ \chi _{\, \mathrm{castSucc}(S)\cup \{ \mathrm{last}\, n\} }(\mathrm{snoc}\, x\, b) = (-1)^{b}\, \chi _S(x), \]

where \((-1)^b\) denotes boolToSign \(b\).

Lemma 18.3 Partition of a sum over subsets of \(\mathrm{Fin}(n+1)\)

For any \(\varphi \) on subsets of \(\mathrm{Fin}(n+1)\), the total sum splits according to whether the last coordinate belongs to the subset:

\[ \sum _{S \subseteq \mathrm{Fin}(n+1)} \varphi (S) = \sum _{T \subseteq \mathrm{Fin}\, n} \varphi (\mathrm{castSucc}(T)) + \sum _{T \subseteq \mathrm{Fin}\, n} \varphi (\mathrm{castSucc}(T)\cup \{ \mathrm{last}\, n\} ). \]
Lemma 18.4 Cardinality of a lifted set
#

Lifting a set along \(\mathrm{castSucc}\) preserves its cardinality: \(|\mathrm{castSucc}(S)| = |S|\).

Lemma 18.5 Cardinality of a lifted set with the last coordinate

Adjoining the (necessarily new) last coordinate to a lifted set increases its cardinality by one: \(|\mathrm{castSucc}(S)\cup \{ \mathrm{last}\, n\} | = |S| + 1\).

For \(f : \{ 0,1\} ^{n+1} \to \mathbb {R}\) the noise operator splits over the last coordinate:

\[ T_\rho f(\mathrm{snoc}\, x\, b) = T_\rho (\mathrm{avgLast}\, f)(x) + (-1)^b\, \rho \, T_\rho (\mathrm{diffLast}\, f)(x), \]

where \(\mathrm{avgLast}\, f\) and \(\mathrm{diffLast}\, f\) are the average and difference of \(f\) over the last bit.

Lemma 18.7 Fourth moment decomposition with the noise operator

For \(f : \{ 0,1\} ^{n+1} \to \mathbb {R}\), writing \(g = \mathrm{avgLast}\, f\) and \(h = \mathrm{diffLast}\, f\), the fourth moment of \(T_\rho f\) decomposes as

\[ \mathbb {E}[(T_\rho f)^4] = \mathbb {E}[(T_\rho g)^4] + 6\rho ^2\, \mathbb {E}[(T_\rho g)^2 (T_\rho h)^2] + \rho ^4\, \mathbb {E}[(T_\rho h)^4]. \]

For any Boolean function \(f : \{ 0,1\} ^n \to \mathbb {R}\) and noise parameter \(\rho \) with \(\rho ^2 \le 1/3\) (equivalently \(\left\lvert \rho \right\rvert \le 1/\sqrt3\)),

\[ \mathbb {E}[(T_\rho f)^4] \le \bigl(\mathbb {E}[f^2]\bigr)^2, \]

that is \(\lVert T_\rho f\rVert _4 \le \lVert f\rVert _2\).

Lemma 18.9 Hölder’s inequality for \(p = 4/3\), \(q = 4\)

For Boolean functions \(f, g\),

\[ \langle f, g \rangle \le \bigl(\mathbb {E}[\, \left\lvert f\right\rvert ^{4/3}\, ]\bigr)^{3/4}\, \bigl(\mathbb {E}[\, \left\lvert g\right\rvert ^{4}\, ]\bigr)^{1/4}. \]
Theorem 18.10 \((4/3,2)\)-Hypercontractivity with \(\rho = 1/\sqrt3\)

Boolean functions are \((4/3, 2)\)-hypercontractive with parameter \(1/\sqrt3\): for any \(f : \{ 0,1\} ^n \to \mathbb {R}\),

\[ \bigl(\mathbb {E}[(T_{1/\sqrt3} f)^2]\bigr)^{1/2} \le \bigl(\mathbb {E}[\, \left\lvert f\right\rvert ^{4/3}\, ]\bigr)^{3/4}. \]
Lemma 18.11 \(L^2\) norm of \(T_\rho f\) in Fourier space

The squared \(L^2\) norm of \(T_\rho f\) expands over the Fourier spectrum:

\[ \langle T_\rho f, T_\rho f \rangle = \sum _{S \subseteq \mathrm{Fin}\, n} \rho ^{2\left\lvert S\right\rvert }\, \widehat{f}(S)^2. \]
Theorem 18.12 Contractivity (\(q = 2\) case)

For \(\rho ^2 \le 1\) and any Boolean function \(f\),

\[ \mathbb {E}[(T_\rho f)^2] \le \mathbb {E}[f^2]. \]

This is the \(q = 2\) case of hypercontractivity.

Lemma 18.13 Binomial coefficient bound
#

For natural numbers \(k \ge 1\) and \(j \le k\),

\[ \binom {2k}{2j} \le \binom {k}{j}\, (2k-1)^j. \]
Lemma 18.14 \(q\)-th moment decomposition along the last coordinate

For \(f : \{ 0,1\} ^{n+1} \to \mathbb {R}\), writing \(g = \mathrm{avgLast}\, f\) and \(h = \mathrm{diffLast}\, f\), the \(q\)-th moment decomposes as

\[ \mathbb {E}[f^q] = \mathbb {E}\! \left[\frac{(g+h)^q + (g-h)^q}{2}\right]. \]

For \(f : \{ 0,1\} ^{n+1} \to \mathbb {R}\), writing \(g = T_\rho (\mathrm{avgLast}\, f)\) and \(h = T_\rho (\mathrm{diffLast}\, f)\), the \(q\)-th moment of \(T_\rho f\) decomposes as

\[ \mathbb {E}[(T_\rho f)^q] = \mathbb {E}\! \left[\frac{(g + \rho h)^q + (g - \rho h)^q}{2}\right]. \]

For any Boolean function \(f : \{ 0,1\} ^n \to \mathbb {R}\), integer \(k \ge 1\), and noise parameter \(\rho \) with \(\rho ^2 \le 1/(2k-1)\),

\[ \mathbb {E}[(T_\rho f)^{2k}] \le \bigl(\mathbb {E}[f^2]\bigr)^k, \]

equivalently \(\lVert T_\rho f\rVert _{2k} \le \lVert f\rVert _2\).

Theorem 18.17 \((2,q)\)-Hypercontractivity for even \(q\)

For an even integer \(q \ge 2\) and noise parameter \(\rho \) with \(\rho ^2 \le 1/(q-1)\),

\[ \mathbb {E}[(T_\rho f)^q] \le \bigl(\mathbb {E}[f^2]\bigr)^{q/2}. \]

This is the \((2,2k)\) theorem restated in terms of an even exponent \(q = 2k\).

Theorem 18.18 \((2,2)\)-Hypercontractivity

For \(\rho ^2 \le 1\) and any Boolean function \(f\),

\[ \mathbb {E}[(T_\rho f)^2] \le \bigl(\mathbb {E}[f^2]\bigr)^1, \]

the \(q = 2\) instance, which is just contractivity.

Theorem 18.19 \((2,6)\)-Hypercontractivity

For \(\rho ^2 \le 1/5\) and any Boolean function \(f\),

\[ \mathbb {E}[(T_\rho f)^6] \le \bigl(\mathbb {E}[f^2]\bigr)^3. \]
Theorem 18.20 \((2,2k)\)-Hypercontractivity in norm form

For \(k \ge 1\) and noise parameter \(\rho \) with \(\rho ^2 \le 1/(2k-1)\), the \((2,2k)\) bound in the form \(\lVert T_\rho f\rVert _{2k} \le \lVert f\rVert _2\):

\[ \bigl(\mathbb {E}[(T_\rho f)^{2k}]\bigr)^{1/(2k)} \le \bigl(\mathbb {E}[f^2]\bigr)^{1/2}. \]
Lemma 18.21 Inner product equals expected square

The self inner product of a Boolean function equals the expectation of its square: \(\langle f, f \rangle = \mathbb {E}[f^2]\).

Lemma 18.22 Nonnegativity of the expected square of \(T_\rho f\)

For any \(\rho \) and Boolean function \(f\), \(\mathbb {E}[(T_\rho f)^2] \ge 0\).

Lemma 18.23 Nonnegativity of an expected absolute power

For any real exponent \(p\) and Boolean function \(f\), \(\mathbb {E}[\, \left\lvert f\right\rvert ^p\, ] \ge 0\).

Lemma 18.24 Composition of noise operators

Composing two noise operators multiplies their parameters: \(T_\rho (T_\sigma f) = T_{\rho \sigma } f\).

Let \(1 \lt p\), \(q \ge 2\) with \(1/p + 1/q = 1\). Given a Hölder inequality for the pair \((p,q)\) and a \((2,q)\)-hypercontractivity bound for \(T_\rho \), one concludes the dual \((p,2)\) bound:

\[ \bigl(\mathbb {E}[(T_\rho f)^2]\bigr)^{1/2} \le \bigl(\mathbb {E}[\, \left\lvert f\right\rvert ^p\, ]\bigr)^{1/p}. \]

18.3 Additional declarations

Lemma 18.26 Key algebraic inequality closing the recurrence

Let \(a, b, A, B, C, \rho \) be reals with \(a \ge 0\), \(b \ge 0\), \(B \ge 0\), and suppose \(A \le a^2\), \(B \le b^2\), \(C^2 \le AB\), and \(\rho ^2 \le 1/3\). Then

\[ A + 6\rho ^2 C + \rho ^4 B \; \le \; (a+b)^2 . \]

This is the algebraic step that makes the induction in the \((2,4)\)-hypercontractivity argument close.