-2

How to prove $|x+y|^q+|x-y|^q \le 2(|x|^p+|y|^p)^{q/p},p\in (1,2),pq=p+q$

This inequality looks like a generalization of the Minkowski inequality. I have tried two methods: one is to break it down and apply scaling separately, which obviously doesn't work; the other is to attempt using the auxiliary inequality $( s^q + t^q \leq (s^2 + t^2)^{q/2} )$, but that also failed.

This problem is still different from the inequality below, $2(|a|^p + |b|^p) \leq |a + b|^p + |a - b|^p \leq 2^{p-1}(|a|^p + |b|^p)$,as the original inequality involves powers of $p $ and $q$.. I appreciate the answer below for pointing out the origin of the inequality. The two cases of Clarkson's inequalities on Wikipedia happen to correspond exactly to these two problems.

Jack N.
  • 15

1 Answers1

1

If you know about the complex interpolation, the following proof will work. In fact, this proof is a standard way to prove Clarkson's inequality, which is an $L^p$-generalization of this inequality.

The inequality is equivalent to $$\left[\left|\frac{x+y}{2}\right|^q+\left|\frac{x-y}{2}\right|^q \right]^{1/q}\leq\left[\frac{|x|^p+|y|^p}{2} \right]^{1/p} \tag{1}$$ for $1<p<2$. So let us begin by first proving this inequality for the extremal cases $(p,q)=(1,\infty)$ and $(2,2)$. The first case follows from the triangle inequality: $$\left|\frac{x\pm y}{2}\right|\leq \frac{|x|+|y|}{2}, $$ and the second case is just the parallelogram law. Now the general case will follow by interpolating these two results. Define $$ T(x,y)=\left(\frac{x+y}{2},\frac{x-y}{2} \right)\qquad (x,y\in\mathbb{C}). $$ Then the above estimates say that $T: \ell^q(\{1,2\})\to\ell^p_\nu\{(1,2)\}$ is bounded for $p=1$ and $p=2$ with operator norm $1$; here, $\ell^p_\nu$ denotes the $\ell^p$ space equipped with the measure $\nu(\{1\})=\nu(\{2\})=1/2$. It follows from the complex interpolation that $T:\ell^q(\{1,2\})\to\ell^p_\nu\{(1,2)\}$ is bounded and has norm $1$ for every $1\leq p\leq 2$, which proves the assertion.

By the way, since we are dealing with finite sequence spaces $\ell^p(\{1,2\})$, I think that there might be an elementary interpolation argument which avoids using the Riesz-Thorin theorem, but I'm not sure. (A direct application of M. Riesz's convexity theorem would work, of course, but it seems no big different to appealing to Riesz-Thorin.)