Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I am making some elementary mistakes here. Could you please help me point out the problems? Thank you very much!

Suppose on some space $H$ we have two inner products, which make $H$ after completion two real Hilbert spaces. Suppose that these two inner products are comparable:

$$ (f,f)_1 \le (f,f)_2,\quad \forall f\in H. $$

Denote these two Hilbert spaces by $H_1$ and $H_2$. It is clear that

$$ H_2 \subseteq H_1. $$

Let $H_1'$ and $H_2'$ be the dual spaces of $H_1$ and $H_2$, respectively. They are also Hilbert spaces. Noticing that the fact that

$$ ||u||_{H_i'} = \sup_{(x,x)_i \le 1} |u(x)|,\qquad i=1,2, $$

we have that

$$ H_1' \subseteq H_2'\:. $$

Being Hilbert spaces, $H_i \cong H_i'$ (i.e., $H_i$ is isomorphic to $H_i'$). Since the Hilbert spaces are real, we can identify $H_i'$ by $H_i$. Then the above two inclusions imply that

$$ H_1 = H_2 , $$

which cannot be true in general. What is wrong in my arguments? Thank you very much for your great help! :-)


share|cite|improve this question
When you identify $H_1$ with $H_1'$, it's with respect to the first inner product, whereas the identification between $H_2$ and $H_2'$ is with respect to the second inner product. We cannot do these identifications "in the same time". An example is $H=\ell^2$, and $V:=\{u\in H,\sum_n nu_n^2<\infty\}$ with the inner product $((u,v))=\sum_n nu_nv_n$. – Davide Giraudo Jan 28 '12 at 15:56
$H_1'\not \subseteq H_2'$. They are functionals on different spaces. – azarel Jan 28 '12 at 16:03
up vote 9 down vote accepted

A key question here is "What precisely does $H_2\subseteq H_1$ mean"? Let's begin with the identity map on the pre-Hilbert space $H$: $$\text{id}:(H,(\cdot,\cdot)_2)\to (H,(\cdot,\cdot)_1).$$ where we assume that, for all $h\in H$, we have $(h,h)_1\leq (h,h)_2$. This map lifts uniquely to the completions $$i: H_2\to H_1.$$ It is often overlooked, however, that the resulting map $i$ may not be injective. The map $i$ is injective if and only if the bilinear form $((h,h)_2, H)$ is closable on $H_1$.

Let's suppose that $i$ is injective so that we can really think of elements of $H_2$ as also belonging to $H_1$. We should consider $H_2\subseteq H_1$ as shorthand for $i:H_2\to H_1$.

Now $i$ is a linear map, and the dual map gives $i^\prime:H_1^\prime\to H_2^\prime$. This gives a precise meaning to $H_1^\prime\subseteq H_2^\prime$. Combining this with Riesz representations on $H_1$ and $H_2$, we can write $$H_1\overset{j}{\rightarrow} H_1^\prime \overset{i^\prime}{\rightarrow} H_2^\prime \overset{k}{\rightarrow} H_2.$$ I think that this is what you mean by $H_1\subseteq H_2$.

In short, all of the paradoxes disappear when you keep careful track of the mappings involved.

share|cite|improve this answer
Just a comment... the composition $ki'j$ is just the Hilbert space adjoint $i^*$. Then $i^*i$ is a positive operator which "implements" $(.|.)_1$ as $(i^*i(h)|h')_2 = (i(h)|i(h'))_1 = (h|h')_1$ for $h,h'\in H$. So we'd only have a "paradox" if $i^*i$ were the identity-- but we now see that this happens only when $(.|.)_1 = (.|.)_2$... – Matthew Daws Jan 28 '12 at 20:57
Thank Professors Byron Schmuland and Matthew Daws for your nice answer and nice comment. I am clear now. :-) – Anand Jan 29 '12 at 12:14

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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