I still didn't understand what is Second Dual Space. From class, I was told that: $X^{**}=\{g_{x}:x \in X\}$ which $g_{x}$ is a map from algebraic dual space $X^{*}$ to scalar field $K$ of vector space $X$. $g_x$ is defined by $g_x(f)=f(x)$. Then, I met $C$ which is called canonical mapping and map from vector space $X$ to second algebraic dual space $X^{**}$ defined by $Cx=g_x$. The book state that $C$ is injective but not always surjective. What I didn't understand is why $C$ isn't always surjective? From the definition of element in $X^{**}$ and from the statement that $C$ is injective, then $C$ must be surjective. Or I might be wrong define elements in $X^{**}$? And I'm using the book from Erwin Kreyszig - Introductory Functional Analysis with Applications

  • $\begingroup$ $X^{**}=\{g_{x}:x \in X\}$ is not the definition of the second dual space. The second dual space is the dual of the dual space. $\endgroup$ – Aweygan Oct 25 '16 at 16:13
  • $\begingroup$ @Aweygan right, I'm sorry. This book said it's Second Algebraic Dual Space. Is it same with Second Dual Space? $\endgroup$ – user379677 Oct 25 '16 at 16:16

You're wrong about the definition of $X^{**}$. From Kreyszig, page 107:

We may go one step further and consider the algebraic dual $(X^*)^*$ of $X^*$, whose elements are the linear functionals defined on $X^*$... and call it the second algebraic dual space of $X$.

He then goes on to explain that for any $x$, $g_x$ is an element of $X^{**}$.

Notably, however, there is nothing about the definition of $X^{**}$ which guarantees that every element $g \in X^{**}$, that is every $g:X^* \to \Bbb C$, has the form $g_x$ for some $x \in X$. In fact, if $X$ is an infinite dimensional normed space (and if we take the axiom of choice), then this is necessarily not the case.

For example: take $X\subset \ell^1$ to be the set of sequences for which all but finitely many elements are $0$.

Let $\{e_i\}_{i \in \Bbb N}$ be the canonical basis of $X$ (so for example, $e_1 = (1,0,0\dots)$); note that this is a Hamel basis for our space. Every functional $f$ is determined by its values over these elements, since $$ f(\sum_{i} \alpha_i e_i) = \sum_i \alpha_i f(e_i) $$ Thus, for every sequence $(\beta_i)_{i \in \Bbb N}$, there is an associated $f$ with $f(e_i) = \beta_i$.

Now, consider the subspace of $U \subset X^*$ consiting of $f \in X^*$ such that $$ \sup_{i \in \Bbb N}|f(e_i)| < \infty $$ We may define the linear map $g:U \to \Bbb C$ by $$ g(f) = \sup_{i \in \Bbb N} f(e_i) $$ and by the axiom of choice, we may write $X^* = U \oplus V$ for some subspace $V$, and we may extend $g$ to all of $X^*$ by defining $g|_V = 0$.

Note that $g \in X^{**}$, but there is no $x \in X$ such that for every $f\in X$, $g(f) = f(x)$.

  • $\begingroup$ Do you happen to know whether the last assertion requires (part of) the axiom of choice? $\endgroup$ – Daniel Fischer Oct 25 '16 at 16:30
  • $\begingroup$ @DanielFischer It almost certainly does; I'll say so. $\endgroup$ – Omnomnomnom Oct 25 '16 at 16:31
  • $\begingroup$ So, It's not necessary that every $g \in X^{**}$ has the form $g_x$? If it's so, then how we define $g(f)$ if it has no form $g_x$? I'm really confused. $\endgroup$ – user379677 Oct 25 '16 at 16:39
  • $\begingroup$ It's hard to come up with an explicit example, but I added one in my latest edit. $\endgroup$ – Omnomnomnom Oct 25 '16 at 16:54
  • $\begingroup$ Ah, now I understand. So, element in $X^{**}$ is a function that maps from $X^*$ to $K$ and the definition is not always $g(f)=f(x)$. Thank you very much! $\endgroup$ – user379677 Oct 25 '16 at 17:08

If we take $c_0 , \ell_1 , \ell_{\infty}. $ Then $c_0^{\ast} =\ell_1 , c_0^{\ast\ast}=\ell_1^{\ast}=\ell_{\infty}. $ Then the functional $\xi (x) =\sum_j x_j $ is not in the image $\kappa (c_0 )$ where $\kappa $ is the canonical mapping but $\xi \in c_0^{\ast\ast}=\ell_1^{\ast}=\ell_{\infty}.$

  • $\begingroup$ One question, what's the dimention of $c_0, l_1, l_{\infty}$? Are they infinite dimentional vector spaces? $\endgroup$ – user379677 Oct 25 '16 at 16:40
  • $\begingroup$ $\mbox{dim} (c_0 ) =\mathfrak{c}$ $\endgroup$ – MotylaNogaTomkaMazura Oct 25 '16 at 16:49
  • $\begingroup$ In this answer, you're using continuous dual as opposed to the algebraic dual. $\endgroup$ – Omnomnomnom Oct 25 '16 at 16:54

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.