We say that a bounded operator $\varphi:X\to Y$ is

  • $c$-topologically injective if $\Vert\varphi(x)\Vert\geq c\Vert x\Vert$ for all $x\in X$
  • $c$-topologically surjective if for all $y\in Y$ there exist $x\in X$ such that $\varphi(x)=y$ and $\Vert x\Vert\leq c\Vert y\Vert$

I have already proved that $$ \varphi \;\text{$c$-topologically injective}\Longleftrightarrow\varphi^*\; \text{$c^{-1}$-topologically surjective}\\ \varphi \;\text{$c$-topologically surjective}\Longrightarrow\varphi^*\; \text{$c^{-1}$-topologically injective} $$ and got stuck at proving that $$ \varphi \;\text{$c$-topologically surjective}\Longleftarrow\varphi^*\; \text{$c^{-1}$-topologically injective}\tag{1} $$


Thank to rayuela, now I can perform the proof. Let $\varphi^*$ is $c^{-1}$ topologically injective. Assume that $\mathrm{cl}_Y(\varphi(cB_X))\not{\supset} B_Y$, then we have some $y_0\in B_Y\setminus\mathrm{cl}_Y(\varphi(cB_X))$. By geometric form of Hahn Banch theorem there exist $g\in Y^*$ and $\gamma_2>\gamma_1>0$ such that $$ \mathrm{Re}(g_0(y_0))>\gamma_2>\gamma_1>\mathrm{Re}(g_0(y)) $$ for all $y\in \mathrm{cl}_Y(\varphi(cB_X))$. By considering $g=2(\gamma_1+\gamma_2)^{-1}g_0$ we get $$ \mathrm{Re}(g(y_0))>1>\mathrm{Re}(g(y)) $$ for all $y\in \varphi(B_X)$. For all $\lambda\in\mathbb{C}$ with $|\lambda|<1$ and $y\in\varphi(B_X)$ we have $\lambda y\in\varphi(B_X)$, so $$ |g(y_0)|>\mathrm{Re}(g(y_0))>1>|g(y)| $$ for all $y\in \varphi(cB_X)$. Since $|g(y_0)|>1$ while $y_0\in B_Y$, then $\Vert g\Vert> 1$. On the other hand for all $x\in c B_X$ we have $|g(\varphi(x))|<1$, i.e. $|\varphi^*(g)(x)|<1$. So $\Vert\varphi^*(g)\Vert<c$ and from $c^{-1}$-topological injectivity of $\varphi^*$ we get $\Vert g\Vert\leq c\Vert\varphi^*(g)\Vert\leq 1$. Contradiction, so $B_Y\subset\mathrm{cl}_Y(\varphi(cB_X))$. By open mapping theorem $\varphi(cB_X)\supset B_Y$ which is equivalent to $c$-topological surjectivity of $\varphi$.

Thank you for taking time.


See Rudin, Functional analysis, Lemma 4.13.

You were not so far off track: Instead of arguing by contradiction, argue directly. Let $A$ be the closure of $\varphi(B_X)$ and suppose $y_0 \in Y \setminus A$. Since $A$ is closed, convex and balanced we can apply Hahn-Banach and find a functional $g \in Y^\ast$ such that $$ \lvert \langle g, a \rangle\rvert \leq 1 \lt \lvert \langle g,y_0\rangle\rvert $$ for all $a \in A$. For $x \in B_X$ we have $\varphi(x) \in A$, so $\lvert \langle \varphi^\ast g,x \rangle \rvert = \lvert \langle g,\varphi(x)\rangle\rvert \leq 1$. Taking the supremum over $x \in B_X$ we deduce $1 \geq \lVert \varphi^\ast g\rVert \geq c^{-1} \lVert g \rVert$ where the last inequality follows from $c^{-1}$-topological injectivity. This is as far as you got.

But now $$ c\lVert y_0\rVert \geq \lVert g\rVert \lVert y_0\rVert \geq \lvert \langle g,y_0\rangle\rvert > 1 $$ and $\lVert y_0 \rVert \gt c^{-1}$. This means that $c^{-1}B_Y \subseteq A$ and now use completeness of $X$ and the usual Banach-Schauder back-and-forth argument in the proof of the open mapping theorem shows that $c\varphi(B_X) \supseteq B_Y$ which is what you want.

To see that you need completeness of $X$ consider the following: suppose $\varphi \colon X \to Y$ is $c$-topologically surjective. Restricting $\varphi$ to a dense subspace $D$ of $X$ will not change the adjoint: $D^\ast = X^\ast$ and $(\varphi|_D)^\ast = \varphi^\ast$. However, restricting will affect $c$-topologial surjectivity.

  • $\begingroup$ Thank you! I relized mistake in my proof. So a small fix will salvage it. Also I'm gratefull to you for this simple counterexample. $\endgroup$ – Norbert Apr 5 '13 at 7:54

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.