I am currently reading a book which adresses topics in Banach space theory and while i was reading i came across the following definition:

Suppose that $X$ is a normed space and $Y$ is a subspace of $X^*$.Let us consider a new norm on $X$ defined by $$\left\lVert{x}\right\rVert_Y=\sup\{|y^*(x)| : y^* \in Y \text{ , } \lVert{y^*}\rVert=1\}$$

And i was wondering why the above equation defines a norm in $X$ since i am not able to justify the implication $$\lVert{x}\rVert_Y=0 \implies x=0$$

Thanks in advance !

Edit: Actually i will post the whole definition so it could be more clear

Definition :

Suppose that $X$ is a normed space and $Y$ is a subspace of $X^*$.Let us consider a new norm on $X$ defined by $$\left\lVert{x}\right\rVert_Y=\sup\{|y^*(x)| : y^* \in Y \text{ , } > \lVert{y^*}\rVert=1\}$$If there is a constant $0<c\leq 1$ such that for all $x \in X$ we have $$c\lVert{x}\rVert\leq \lVert{x}\rVert_Y\leq \lVert{x}\rVert$$then $Y$ is said to be a c-norming subspace of $X^*$.

So i guess the author means that $\lVert{x}\rvert_Y$ is a norm when $Y$ is a c-norming subspace of $X^*$?

  • 1
    $\begingroup$ Take $X=\mathbb{R}^2$, $x=(1,0)$ and $Y = \{y^*\in X^* : y^*(\xi)=0$ for any $\xi=(\xi_1,0)\}$. Then $\lVert x\rVert_{Y}=0$. Is that really a definition in your book? $\endgroup$ – user539887 May 6 '18 at 9:45
  • $\begingroup$ i made an edit in my question $\endgroup$ – dem0nakos May 6 '18 at 9:48
  • $\begingroup$ Certainly that was the author's intent. $\endgroup$ – user539887 May 6 '18 at 9:49
  • $\begingroup$ No its ok , leave it , your example is correct $\endgroup$ – dem0nakos May 6 '18 at 9:49

It's indeed not true, think about finite dimensional $X$ or $Y=0$.

It's true, however, if $Y=X^*$ or if $Y$ is dense in $X^*$.
Without this assumption we can only state that it's a norm on the quotient space $X/(\bigcap_{f\in Y} \ker f) $.

To your edit: yes, it perfectly makes sense if $c\Vert x\Vert\le\Vert x\Vert_Y$ with a fixed $c>0$.


That implication need not be true in general. If you want to insist $Y \neq \{0\}$ try considering the case where $Y = \operatorname{span}(\phi)$ is one dimensional, for $\|\phi\| =1$ (and $\dim X > 1$). Then $\|x\|_Y = |\phi(x)|$. But $\phi$ must have non-trivial kernel so $\| \cdot \|_Y$ is not a norm in general.

In response to the edit: What's going on here is that in general $\| \cdot \|_Y$ defines a seminorm on $X$. However, whenever a seminorm is equivalent to a norm, as is the case if $Y$ is a $c$-norming subspace of $X^*$, then that seminorm is automatically actually a norm since $\|x\|_Y = 0 \implies \|x\| = 0$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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