2

I have the following problem: Let $(a_n)_{n \in \mathbb{N}}$ be a sequence in $\ell^p (\mathbb{N})$ for some $p \in (0,2]$. Let $q \geq p$, $N \in \mathbb{N}$. Show: There exists a finite index set $I \subseteq \mathbb{N}$ of size $N$, such that $$\left(\sum_{n \in \mathbb{N} \setminus I} (a_n)^q \right)^{1/q} \leq || (a_n)_{n \in \mathbb{N}}||_{\ell^p} (N+1)^{1/q - 1/p}.$$ So the obvious choice for $I$ would be to pick the biggest values of $a_n$, since the sum converges these values cannot become arbitrary big. I know of a similar inequality regarding finite measure sets, see e.g. Inequalities in $l_p$ norm or An inequality about $L^p$ norm? , but this is a little bit different. What can I do here?

LinearAlgebruh
  • 1,038
  • 8
  • 16

1 Answers1

3

I will use an approach very similar to the one in the accepted answer by Syl.Qiu of An inequality about $L^p$ norm? which you had linked in your question. You also had the right idea to consider the biggest values of $a_n$, as this is what I'll be doing.

Let $I_0 := \emptyset$ and $i_1 := \min\{i \in \mathbb{N} \mid |a_i| = \sup |a_n|\}$, then define recursively $I_N := \{i_1 , \dots, i_N\}$ and $$i_{N+1} := \min\left\{i \in \mathbb{N} \setminus I_N \mid |a_i| = \sup_{n \in \mathbb{N}\setminus I_N} |a_n|\right\}$$ This is just to formalise the process of "picking the biggest values" by choosing an ordering.

Then $I_N$ is of size exactly $N$, even for $N = 0$, and we have: $$\begin{split} \left(\sum_{n \in \mathbb{N} \setminus I_N} |a_n|^q \right)^{\frac1q} &= \left(\sum_{n \in \mathbb{N} \setminus I_N} |a_n|^p |a_n|^{q-p} \right)^{\frac1q}\\ &\leq \left(\sum_{n \in \mathbb{N} \setminus I_N} |a_n|^p \right)^{\frac1q} \left(\sup_{n \in \mathbb{N} \setminus I_N}|a_n|^{q-p}\right)^{\frac1q}\\ &\leq \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1q} \bigg(\underbrace{\sup_{n \in \mathbb{N} \setminus I_N}|a_n|}_{=\, |a_{i_{N+1}}|}\bigg)^{(q - p) \cdot \frac1q \cdot \frac p p}\\ &\leq \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1q} \left(|a_{i_{N+1}}|^p\right)^{\frac{q-p}{pq}}\\ &\leq \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1q} \left(\frac{1}{N+1}\sum_{k=1}^{N+1}|a_{i_k}|^p\right)^{\frac1p - \frac1q}\\ &\leq \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1q} \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1p - \frac1q} (N+1)^{\frac1q - \frac1p}\\ &\leq \left(\sum_{n \in \mathbb{N}} |a_n|^p \right)^{\frac1p} (N+1)^{\frac1q - \frac1p}\end{split}$$ where we used the fact that $|a_{i_{N+1}}|^p \leq |a_{i_k}|^p$ for $k \leq N$, so that $(N+1) |a_{i_{N+1}}|^p \leq \sum_{k = 1}^{N+1} |a_{i_k}|^p$.

Bruno B
  • 7,655
  • Thanks for the answer. Im wondering: Do we need the condition $q \geq p$ and $0 < p \leq 2$? For me it looks like as if it should work for arbitrary non-negative $p,q$ – LinearAlgebruh Dec 20 '24 at 12:16
  • 1
    @LinearAlgebruh I wonder for the condition on $p$ yeah, but the fact that $q \geq p$ is used to have increasingness of $x \in [0, \infty) \mapsto x^{\frac1p - \frac1q}$, so that we can add all the missing terms of the series towards the end of the calculations. It's my fault for not having emphasized that in my answer to be fair. I also don't see what happens for $p = 0$ as the expression seems undefined, so I won't be able to help there. – Bruno B Dec 20 '24 at 12:32
  • 1
    $q \geq p$ is also used to move powers around on the $\sup$ towards the beginning, as you're only allowed to do that with non-negative powers for obvious reasons. Moreover, this cannot be true for $q < p$, otherwise the two topologies would be equivalent on the smaller space, and that seems impossible (you can probably use some density + completeness argument to show that it is indeed impossible? I don't remember). – Bruno B Dec 20 '24 at 12:39