Let $\Vert\cdot\Vert$ is a norm in $R^n$. Let $x_1,\dots,x_N$ non-independent Rademacher random variables random variables (variables which are uniform on $\{-1, 1\}$). . By $E$ we denote an expectation. There exists a constant $A_{p,q}$ depending only on $p,q$ such that for any vectors $u_1,\dots, u_N\in R^n$ $$ \left(E \left\Vert\sum_{i=1}^N x_iu_i\right\Vert^p\right)^\frac{1}{p}\leq A_{p,q} E \left(\left\Vert\sum_{i=1}^N x_iu_i\right\Vert^{q}\right)^{\frac 1q}, $$ for all $p,q \in [1, \infty)$. This is so called restricted Kahane's inequality.

What are possible applications of this inequality?

  • $\begingroup$ Certainly you need some assumption about the $x_i$. Maybe less than independence suffices but you can't let them be arbitrary. $\endgroup$ Commented Mar 29, 2018 at 19:09
  • $\begingroup$ The claim is false even for a single x, see my answer. $\endgroup$
    – Aryeh
    Commented Mar 30, 2018 at 4:55
  • 1
    $\begingroup$ Kahane's inequality asks that the $x_i$'s be Rademacher r.v.'s, i.e. uniform in $\{-1,1\}$. Without this assumption, as @Aryeh points out the statement in the question is false. $\endgroup$
    – Clement C.
    Commented Mar 30, 2018 at 17:43
  • 2
    $\begingroup$ This still doesn't make sense. For one thing the expectation sign needs to be inside the parenthesis on the right. For another the inequality still fails if you assume that $x_1, \ldots x_N$ are each Rademacher but not independent. You need to make some assumption about the dependence structure. For example the restricted Khintchine inequality considers $(x_1, \ldots, x_N)$ chosen uniformly from $\{-1,1\}^N$ such that $x_1 + \ldots + x_N = 0$. $\endgroup$ Commented Apr 3, 2018 at 5:31
  • $\begingroup$ It's been a week, and after many edits this question is still not making sense: namely, it is asking for applications of an inequality which does not hold. $\endgroup$
    – Clement C.
    Commented Apr 5, 2018 at 17:37

1 Answer 1


I don't think the claim is correct. Take $n=1$ (so the norm is just absolute value), $N=1$, and $u_1=1$. Let $X$ be any random variable with a finite first moment and infinite second moment (i.e., $E|X|=m<\infty$ while $E|X|^2=\infty$; one such example is the Pareto family https://math.stackexchange.com/questions/236181/example-of-a-general-random-variable-with-finite-mean-but-infinite-variance ). The claim would imply a finite $A_2$ for which $E|X|^2\le A_2^2 m^2$, which cannot be true.

  • $\begingroup$ You are right, thank you. I have corrected inequality, now it is true. $\endgroup$
    – user124297
    Commented Mar 29, 2018 at 14:05
  • $\begingroup$ Presumably, you'd want $A$ to depend on both $p$ and $q$. In my example, $p=2$ and $q=1$, so it still fails. $\endgroup$
    – Aryeh
    Commented Mar 29, 2018 at 14:12
  • $\begingroup$ Yes, thank you very much for your corrections. $\endgroup$
    – user124297
    Commented Mar 29, 2018 at 14:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.