Khintchine inequality

In mathematics, the Khintchine inequality, named after Aleksandr Khinchin and spelled in multiple ways in the Latin alphabet, is a theorem from probability, and is also frequently used in analysis.

Consider $N$ complex numbers $x_{1},\dots ,x_{N}\in \mathbb {C}$, which can be pictured as vectors in a plane. Now sample $N$ random signs $\epsilon _{1},\dots ,\epsilon _{N}\in \{-1,+1\}$, with equal independent probability. The inequality intuitively states that $${\Big |}\sum _{i}\epsilon _{i}x_{i}{\Big |}\approx {\sqrt {|x_{1}|^{2}+\cdots +|x_{N}|^{2}}}$$

Statement

Let $\{\varepsilon _{n}\}_{n=1}^{N}$ be i.i.d. random variables with $P(\varepsilon _{n}=\pm 1)={\frac {1}{2}}$ for $n=1,\ldots ,N$, i.e., a sequence with Rademacher distribution. Let $0<p<\infty$ and let $x_{1},\ldots ,x_{N}\in \mathbb {C}$. Then

$A_{p}\left(\sum _{n=1}^{N}|x_{n}|^{2}\right)^{1/2}\leq \left(\operatorname {E} \left|\sum _{n=1}^{N}\varepsilon _{n}x_{n}\right|^{p}\right)^{1/p}\leq B_{p}\left(\sum _{n=1}^{N}|x_{n}|^{2}\right)^{1/2}$

for some constants $A_{p},B_{p}>0$ depending only on $p$ (see Expected value for notation). More succinctly, $$\left(\operatorname {E} \left|\sum _{n=1}^{N}\varepsilon _{n}x_{n}\right|^{p}\right)^{1/p}\in [A_{p},B_{p}]$$for any sequence $x$ with unit $\ell ^{2}$ norm.

The sharp values of the constants $A_{p},B_{p}$ were found by Haagerup (Ref. 2; see Ref. 3 for a simpler proof). It is a simple matter to see that $A_{p}=1$ when $p\geq 2$, and $B_{p}=1$ when $0<p\leq 2$.

Haagerup found that

${\begin{aligned}A_{p}&={\begin{cases}2^{1/2-1/p}&0<p\leq p_{0},\\2^{1/2}(\Gamma ((p+1)/2)/{\sqrt {\pi }})^{1/p}&p_{0}<p<2\\1&2\leq p<\infty \end{cases}}\\&{\text{and}}\\B_{p}&={\begin{cases}1&0<p\leq 2\\2^{1/2}(\Gamma ((p+1)/2)/{\sqrt {\pi }})^{1/p}&2<p<\infty \end{cases}},\end{aligned}}$

where $p_{0}\approx 1.847$ and $\Gamma$ is the Gamma function. One may note in particular that $B_{p}$ matches exactly the moments of a normal distribution.

Uses in analysis

The uses of this inequality are not limited to applications in probability theory. One example of its use in analysis is the following: if we let $T$ be a linear operator between two L^p spaces $L^{p}(X,\mu )$ and $L^{p}(Y,\nu )$, $1<p<\infty$, with bounded norm $\|T\|<\infty$, then one can use Khintchine's inequality to show that

$\left\|\left(\sum _{n=1}^{N}|Tf_{n}|^{2}\right)^{1/2}\right\|_{L^{p}(Y,\nu )}\leq C_{p}\left\|\left(\sum _{n=1}^{N}|f_{n}|^{2}\right)^{1/2}\right\|_{L^{p}(X,\mu )}$

for some constant $C_{p}>0$ depending only on $p$ and $\|T\|$.

Generalizations

For the case of Rademacher random variables, Pawel Hitczenko showed that the sharpest version is:

$A\left({\sqrt {p}}\left(\sum _{n=b+1}^{N}x_{n}^{2}\right)^{1/2}+\sum _{n=1}^{b}x_{n}\right)\leq \left(\operatorname {E} \left|\sum _{n=1}^{N}\varepsilon _{n}x_{n}\right|^{p}\right)^{1/p}\leq B\left({\sqrt {p}}\left(\sum _{n=b+1}^{N}x_{n}^{2}\right)^{1/2}+\sum _{n=1}^{b}x_{n}\right)$

where $b=\lfloor p\rfloor$, and $A$ and $B$ are universal constants independent of $p$.

Here we assume that the $x_{i}$ are non-negative and non-increasing.

See also

• Marcinkiewicz–Zygmund inequality
• Burkholder-Davis-Gundy inequality

References

1. Thomas H. Wolff, "Lectures on Harmonic Analysis". American Mathematical Society, University Lecture Series vol. 29, 2003. ISBN 0-8218-3449-5
2. Uffe Haagerup, "The best constants in the Khintchine inequality", Studia Math. 70 (1981), no. 3, 231–283 (1982).
3. Fedor Nazarov and Anatoliy Podkorytov, "Ball, Haagerup, and distribution functions", Complex analysis, operators, and related topics, 247–267, Oper. Theory Adv. Appl., 113, Birkhäuser, Basel, 2000.