Here's Prob. 7, Sec. 3.8 in Introductory Functional Analysis With Applications by Erwine Kreyszig:

Show that the dual space $H^\prime$ of a Hilbert space $H$ is a Hilbert space with inner product $\langle \ \cdot \ , \ \cdot \ \rangle_1$ defined by $$\langle f_z, f_v \rangle_1 \ = \ \overline{\langle z, v \rangle} \ = \ \langle v, z \rangle,$$ where $f_z(x) = \langle x, z \rangle$ for all $x \in X$.

Now I know that each bounded linear functional $f \in H^\prime$ can be written as $f = f_z$, for a unique $z \in H$ with $\Vert z \Vert = \Vert f \Vert$.

The sum of two bounded linear functionals on any normed space is again a bounded linear functional, and so is the scalar multiple of any bounded linear functional.

Moreover, for each $z \in H$ and for each $w \in H$, we have $$\left( f_z + f_w \right) (x) = f_z(x) + f_w(x) = \langle x, z \rangle + \langle x, w \rangle = \langle x, z+w \rangle = f_{z+w}(x)$$ for all $x \in H$. So $f_{z+w} = f_z + f_w$.

For $z \in H$ and for any scalar $\alpha$, we have $$f_{\alpha z} (x) = \langle x, \alpha z \rangle = \overline{\alpha} \langle x, z \rangle = \overline{\alpha} f_z (x) \ \mbox{ for all } \ x \in H.$$ So, $f_{\alpha z} = \overline{\alpha } f_z$.

Thus the mapping $z \mapsto f_z$ of $H$ into $H^\prime$ is surjective, isometric, and conjugate linear.

Moreover, the set of all bounded linear functionals on any normed space is itself a normed space, rather a Banach space.

But how to show that the inner product given by Kreyszig is the (only) natural one (i.e. that which fits into what we already know from the normed space theory)?

  • $\begingroup$ @Daniel Fischer, can you please answer this question? $\endgroup$ May 31 '15 at 4:59

Let me clarify what the question is asking, since based on your comments you actually already know how to solve it.

As you know, for any complex normed space $(X, \|\cdot\|_X)$, the dual space $X'$ is the complex vector space of all continuous linear functionals $f : X \to \mathbb{C}$, with pointwise addition and scalar multiplication. We usually equip $X'$ with the "operator norm" $\|\cdot\|_{X'}$ defined by $$\|f\|_{X'} = \sup_{x \in X, \|x\|_X = 1} |f(x)|.$$

Now in case $X$ is a Hilbert space $(H, \langle \cdot, \cdot \rangle_H)$, we know from the Riesz representation theorem that every $f \in H'$ is of the form $f_z$ for some $z \in H$, where $f_z$ is defined by $f_z(v) = \langle v, z \rangle$. This suggests constructing an inner product on the complex vector space $H'$: Kreyszig calls this new inner product $\langle \cdot, \cdot \rangle_1$, and it is defined by $\langle f_z, f_v \rangle_1 = \langle v, z \rangle_H$. You should verify that $\langle \cdot, \cdot \rangle_1$ is an inner product (sesquilinear, Hermitian, positive definite) but this is very easy.

Now, on the complex vector space $H'$ we have a norm $\|\cdot\|_{H'}$ and an inner product $\langle \cdot, \cdot \rangle_1$. The question is asking you to show that the inner product $\langle \cdot, \cdot \rangle_1$ induces the norm $\|\cdot\|_{H'}$. That is, you must show that $\|f\|_{H'} = \sqrt{\langle f, f \rangle_1}$ for every $f \in H'$.

If you unwind this, it comes down to showing something like the following: for every $z \in H$, $$\|z\|_H = \sup_{v \in H, \|v\|=1} |\langle v,z \rangle|.$$ Prove this by proving two inequalities. For one of them, use Cauchy-Schwarz. For the other, consider an appropriate choice for $v$.

You only have to show this for the specific inner product $\langle \cdot, \cdot \rangle_1$ given in the question, not for an arbitrary inner product on $H'$. The question is not asking you to show that every inner product on $H'$ induces the norm $\|\cdot\|_{H'}$. This is trivially false; as a silly example, consider the inner product $\langle \cdot, \cdot\rangle_2$ defined by $\langle f_z, f_v \rangle_2 = 2 \langle v, z \rangle_H$. If you use the axiom of choice, you can produce other inner products on $H'$ that induce completely different (and uninteresting) topologies.

  • $\begingroup$ thank you for your answer, but this much I already know. But I'm not sure if I should be content with using the formula for the inner product on the dual space, or if I should start with an arbitrary inner product and show that it reduced to the formula given by Kreyszig. $\endgroup$ May 30 '15 at 7:45
  • $\begingroup$ I'm using the real and imaginary parts as my starting point, for the norm on $H^\prime$ does satisfy the parallelogram identity. I remember looking up an article on one of the MIT websites about how we can prove the four properties of an inner product satisfied by the formulas from a norm that satisfies the parallelogram identities. Can you please dig it up for me? $\endgroup$ May 30 '15 at 8:29
  • $\begingroup$ I'm still eagerly awaiting your replies to my two comments. $\endgroup$ May 31 '15 at 4:59
  • $\begingroup$ @SaaqibMahmuud: Please see my edit, especially the last paragraph. In the future I would suggest you avoid posting followup comments like "please respond to me" just to ping someone; many people find them annoying, like a small child on a car trip asking "Are we there yet" every two minutes. For the parallelogram law fact, see math.stackexchange.com/questions/21792/…, but we don't need to use it here. No, I cannot dig up the MIT reference for you; I leave that to your own research skills. $\endgroup$ May 31 '15 at 14:26

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.