Erwin Kreyszig's _Introductory Functional Analysis With Applications_: Section 2.3, Prob. 14

888 Views Asked by At

Here's problem 14 in the Problem Set immediately following Section 2.3 in the book, Introductory Functional Analysis With Applications by Erwin Kreyszig.

Let $Y$ be a closed subspace of a normed space $(X, ||\cdot||)$. Show that a norm $||\cdot||_0$ on the quotient space $X/Y$ can be defined as follows: $$||\hat{x}||_0 \colon= \inf_{x\in\hat{x}} ||x||$$ for all $\hat{x} \in X/Y$.

Now my question is why do we require the subspace $Y$ to be closed? It is my feeling that it is to ensure that the norm on the quotient space satisfies the property $$||\hat{x}||_0 = 0 \mbox{ if and only if } \hat{x} = \hat{0},$$ but I'm not completely sure.

What if $Y$ is not a closed subspace?

And, how to demonstrate all the properties of the norm in case $Y$ is a closed subspace?

2

There are 2 best solutions below

0
On BEST ANSWER

Let us take a closer look at $\lVert\,\cdot\,\rVert_0$. Since for $x\in X$ we have $\hat{x} = \{ x-y : y \in Y\}$, we can write

$$\lVert \hat{x}\rVert_0 = \inf_{z\in\hat{x}} \lVert z\rVert = \inf \{ \lVert x-y\rVert : y \in Y\}.\tag{1}$$

So we have $\lVert \hat{x}\rVert_0 = 0$ if and only if for every $\varepsilon > 0$ there exists a $y_\varepsilon \in Y$ with $\lVert x-y_\varepsilon\rVert < \varepsilon$. We can write the latter as

$$\bigl(\forall\varepsilon > 0\bigr)\bigl(B_\varepsilon(x)\cap Y \neq\varnothing\bigr),\tag{2}$$

where $B_r(x) := \{ z\in X : \lVert x-z\rVert < r\}$ is the open ball with radius $r$ and centre $x$. But $(2)$ is just the criterion for $x \in \overline{Y}$, either by definition of, or by one of the first and most important criteria for, the closure of a set in a topological space.

The right hand side of $(1)$ is, for general subsets $Y\subset X$, the definition of the distance of the point $x$ to the subset $Y$, and we have

$$\overline{A} = \{ x \in S : \operatorname{dist}(x,A) = 0\}$$

for any subset $A$ of a metric space $S$.

Thus we have seen that

$$\{ x\in X : \lVert\hat{x}\rVert_0 = 0\} = \{ x\in X : \operatorname{dist}(x,Y) = 0\} = \overline{Y}.$$

But $\{ x\in X : \hat{x} = 0_{X/Y}\} = Y$, so if $Y$ is not closed, then there are elements $\xi$ of $X/Y$ with $\lVert\xi\rVert_0 = 0$ but $\xi \neq 0$, namely the elements $\xi = \hat{x}$ for $x\in \overline{Y}\setminus Y$.

Hence $\lVert\,\cdot\,\rVert_0$ can only be a norm on $X/Y$ if $Y$ is closed.

In all cases, however, $\lVert\,\cdot\,\rVert_0$ is a seminorm (inducing the quotient topology) on $X/Y$.

It is immediate that $\lVert\hat{x}\rVert_0 \geqslant 0$ for all $\hat{x}$, and that $\lVert\hat{0}\rVert_0 = 0$. The latter gives us the homogeneity

$$\lVert \alpha\cdot\hat{x}\rVert_0 = \lvert\alpha\rvert\cdot\lVert\hat{x}\rVert_0\tag{3}$$

for the case $\alpha = 0$, so we only need to prove it for $\alpha \in K\setminus \{0\}$, where $K$ is the scalar field (which needs to have a valuation $\lvert\,\cdot\,\rvert$ for $(3)$ to make sense; typically, $K$ is $\mathbb{C}$ or $\mathbb{R}$). For $\alpha\neq 0$, we have

\begin{align} \lVert\alpha\hat{x}\rVert_0 = \lVert\widehat{\alpha x}\rVert_0 &= \inf \{ \lVert \alpha x-y\rVert : y\in Y\}\\ & = \inf \{ \lVert \alpha x - \alpha (\alpha^{-1}y)\rVert : y \in Y\}\\ &= \inf \{ \lVert \alpha (x-\alpha^{-1}y)\rVert : y \in Y\}\\ &= \inf \{ \lvert\alpha\cdot\lVert x-\alpha^{-1}y\rVert : y \in Y\}\\ &= \lvert\alpha\rvert\cdot \inf \{ \lVert x-\alpha^{-1}y\rVert : y\in Y\}\\ &= \lvert\alpha\rvert\cdot \inf \{ \lVert x- y'\rVert : y' \in Y\}\\ &= \lvert\alpha\rvert\cdot \lVert \hat{x}\rVert_0, \end{align}

where in the penultimate line we used that $\alpha^{-1}Y = Y$. That establishes $(3)$ for nonzero $\alpha$, and together with the remark before, $(3)$ is established for all $\alpha \in K$ and $\hat{x} \in X/Y$.

It remains to check the triangle inequality for $\lVert\,\cdot\,\rVert_0$. Fix $x_1,x_2 \in X$ arbitrarily, and let $\varepsilon > 0$. By definition of $\inf$, there are $y_1,y_2 \in Y$ with

$$\lVert x_i - y_i\rVert < \lVert \hat{x}_i\rVert_0 + \frac{\varepsilon}{2}$$

for $i\in \{1,2\}$, and thus

\begin{align} \lVert \hat{x}_1 + \hat{x}_2\rVert_0 &= \inf \{ \lVert x_1 + x_2 - y\rVert : y \in Y\}\\ &= \inf \left\{ \lVert (x_1 - y_1) + (x_2 - y_2) - (y-y_1-y_2)\rVert : y\in Y\right\}\\ &\leqslant \lVert (x_1-y_1) + (x_2 -y_2)\rVert\\ &\leqslant \lVert x_1 -y_1\rVert + \lVert x_2 - y_2\rVert\\ & < \lVert \hat{x}_1\rVert_0 + \frac{\varepsilon}{2} + \lVert\hat{x}_2\rVert_0 + \frac{\varepsilon}{2}\\ &= \lVert\hat{x}_1\rVert_0 + \lVert\hat{x}_2\rVert_0 + \varepsilon. \end{align}

Since the inequality holds for every $\varepsilon > 0$, taking the infimum over the right hand side we conclude that indeed

$$\lVert\hat{x}_1 + \hat{x}_2\rVert_0 \leqslant \lVert \hat{x}_1\rVert_0 + \lVert\hat{x}_2\rVert_0.\tag{4}$$

Since $x_1,x_2$ were arbitrary, the triangle inequality $(4)$ for $\lVert\,\cdot\,\rVert_0$ is established.

We have seen above that $\lVert\hat{x}\rVert_0 = 0 \iff x\in \overline{Y}$, so the seminorm $\lVert\,\cdot\,\rVert_0$ is a norm on $X/Y$ if and only if $Y$ is closed.

It should be noted that in proving that $\lVert\,\cdot\,\rVert_0$ is a seminorm, only the seminorm properties of $\lVert\,\cdot\,\rVert$ were used, so we have the useful fact that every seminorm $p$ on $X$ induces a seminorm $p_0$ on the quotient $X/Y$, for every subspace $Y\subset X$, closed or not.

0
On

To summarize Daniel Fischer's answer for hasty readers:

Your feeling is right, $\|~\|_0$ is always a semi-norm, and the closedness of $Y$ is equivalent to $\|~\|_0$ being a norm, i.e. satifying moreover $$\|\hat x\|_0 = 0\implies\hat{x} = \hat0,$$ or equivalently (by definition of $d(x,Y)$) $$d(x,Y)=0\implies x\in Y,$$ otherly written: $$\overline Y\subset Y.$$