Let $E \subseteq \mathbb R$ have finite measure, and let $p_1,p_2 \in (1,\infty)$ with $p_1 < p_2$. We have $L^{p_2}(E) \subseteq L^{p_1}(E)$; let $i: L^{p_2}(E) \hookrightarrow L^{p_1}(E)$ denote the inclusion map. The dual map $i^*: L^{p_1}(E)^* \to L^{p_2}(E)^*$ is given by $$i^*(\phi) = \phi \circ i = \phi|_{L^{p_2}(E)}$$ By the Hahn-Banach Theorem, every bounded linear functional on a linear subspace of a normed linear space extends to a bounded linear functional on the entire space; therefore $i^*$ is a surjective map. Now let $q_1,q_2$ denote the respective conjugates of $p_1,p_2$, defined by $\frac1{p_i} + \frac1{q_i} = 1$. We have $q_2 < q_1$, and so $L^{q_1}(E) \subseteq L^{q_2}(E)$; let $j$ denote the inclusion map. Let $R_1:L^{q_1}(E) \cong L^{p_1}(E)^*$ denote the isomorphism afforded by the Riesz Representation Theorem; namely $R_1(g) = \left(L^{p_1}(E) \ni f \mapsto \int_E fg\right)$, and let $R_2$ be defined similarly. We have the following diagram:

$\require{AMScd}$ \begin{CD} L^{p_1}(E)^* @>i^*>> L^{p_2}(E)^*\\ @| @| \\ L^{q_1}(E) @>>j> L^{q_2}(E) \end{CD}

where the double vertical bars indicate the respective isomorphisms.

Q: Does this diagram commmute?

On the one hand, direct verification seems to suuggest that this is indeed the case: let $g \in L^{q_1}(E)$, then $(f \mapsto \int_E fg)$ is the linear functional $R_1(g)$ on $L^{p_1}(E)$; restricting this to $L^{p_2}(E)$ gives a linear functional on $L^{p_2}(E)$ which is given by integration against an $L^{q_2}(E)$ function (since $g \in L^{q_2}(E)$ also), and by Riesz Representation there is only one $g' \in L^{q_2}(E)$ which accomplishes this, namely $g' = g$ itself. Therefore, one would conclude that the diagram indeed commutes.

On the other hand, $R_2^{-1} \circ i^* \circ R_1$ is a composition of surjective maps and thus surjective, while the containment $L^{q_1}(E) \subseteq L^{q_2}(E)$ is strict; therefore it seems that the diagram can't possibly commute!

What is going on here?


$i^{\ast}$ is not surjective. Let $(X,\Vert\cdot\Vert_{X})$ be a normed space. Hahn-Banach theorem says that if $Y$ is a subspace of $X$ and $T:Y\rightarrow Z$ is linear and continuous with respect to the norm $\Vert\cdot\Vert_{X}$, then you can extend $T$ to a linear continuous function $T_{1}:X\rightarrow Z$. In this case given $T\in L^{p_{2}}(E)^{\ast}$, we have that $T:L^{p_{2}% }(E)\rightarrow\mathbb{R}$ is linear but it is continuous with respect to the norm in $\Vert\cdot\Vert_{L^{p_{2}}(E)}$ but not with respect to the norm in $L^{p_{1}}(E)$. So even if $L^{p_{2}}(E)\subset L^{p_{1}}(E)$ we cannot use Hahn-Banach to extend $T$ to an element of $L^{p_{1}}(E)^{\ast}$. More concretely, since $q_{2}<q_{1}$ you can find $g\in L^{q_{2}}(E)\setminus L^{q_{1}}(E)$. Hence the linear functional $f\mapsto\int_{E}fg$ belongs to $L^{p_{2}}(E)^{\ast}$ (by Holder's inequality) but it does not belong to $i^{\ast}(L^{p_{1}% }(E)^{\ast})$.

  • $\begingroup$ I was using Hahn-Banach to argue that every functional in $L^{p_2}(E)^*$ ought to extend to a functional in $L^{p_1}(E)^*$, and thus that every functional in $L^{p_2}(E)^*$ is the restriction of a functional in $L^{p_1}(E)^*$. Hence surjectivity of $i^*$. The problem is that the norms are different, as I described above. $\endgroup$ – silvascientist Feb 27 '17 at 19:35
  • $\begingroup$ OKI understand what you meant with Hahn-Banach now. Anyway, independently of any theorem, as I said, $i^*$ cannot be surjective, since a function $g$ in $L^{q_2}\setminus L^{q_1}$ will give a functional in the dual of $L^{p_2}$ which cannot be extended to the dual of $L^{p_1}$. And yes, the problem is that the norms are different. $\endgroup$ – Gio67 Feb 27 '17 at 19:48
  • $\begingroup$ Would you mind updating your answer to reflect the difficulty with Hahn-Banach? I'd like to have a single, canonical answer to award the bounty to. $\endgroup$ – silvascientist Mar 2 '17 at 19:20
  • $\begingroup$ sure. I'll change it later today $\endgroup$ – Gio67 Mar 2 '17 at 19:29
  • $\begingroup$ I changed it. It should be OK now $\endgroup$ – Gio67 Mar 2 '17 at 20:17

I've figured it out. The problem is a misapplication of the Hahn-Banach Theorem. This is actually really subtle. The problem is that we have a subspace with a different norm than the ambient space, namely the $||\cdot||_{p_2}$ norm on $L^{p_2}(E)$, compared with the $||\cdot||_{p_1}$ norm on $L^{p_1}(E)$. Any functional $T \in L^{p_2}(E)^*$ is bounded with respect to the $p_2$-norm, but it may not be bounded with respect to the $p_1$-norm on $L^{p_2}(E)$. For Hahn-Banach to apply we need our functional to be bounded with respect to the norm induced by the ambient space. This kind of functional may not be extended to be a functional on the entirety of $L^{p_1}(E)$, and so the surjectivity argument fails. If it were not for this subtlety, and for example if we could suppose that every $||\cdot||_{p_2}$-bounded functional was also $||\cdot||_{p_1}$-bounded, then Hahn-Banach would indeed apply and we could say that every functional in $L^{p_2}(E)$ could be extended to a functional in $L^{p_1}(E)$, and thus that every functional in $L^{p_2}(E)^*$ was the restriction of a functional in $L^{p_1}(E)$, implying surjectivity of $i^*$.


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.