(Stanford Real Analysis Qualifying Exam: Spring 2012) (Ideal time: 24 minutes)

5) Let $X$, $Y$ be separable reflexive Banach spaces. Let $P \colon X \to Y$ be a bounded linear map, and $P'\colon Y^* \to X^*$ its adjoint (transpose). Suppose that there exists a Banach space $Z$ and a compact map $\iota\colon Y^* \to Z$ such that there exists $C > 0$ such that every $\phi \in Y^*$ satisfies $$\Vert \phi \Vert_{Y^*} \leq C(\Vert P'\phi \Vert_{X^*} + \Vert \iota\phi \Vert_Z).$$

b) Let $V \subset Y^*$ be a closed subspace with $V \oplus \text{Ker}(P') = Y^*$. Show that there is $C'>0$ such that every $\phi \in V$ has $\Vert \phi \Vert_{Y^*} \leq C'\Vert P'\phi\Vert_{X^*}$.

c) Show that if $f \in Y$ has $\ell(f) = 0$ for every $\ell \in \text{Ker}(P')$, then there exists $u \in X$ with $Pu = f$.

Note: Part (a) asks the student to show that $\text{Ker}(P')$ is finite-dimensional, which I've done.

My attempts:

(b) I think $P'|_V\colon V \to \text{Im}(P'|_V)$ is a bijection, and hence has a (bounded) inverse map $Q$. Therefore, if $\phi \in V$, then $\Vert \iota \phi\Vert_Z \leq \Vert \iota Q \Vert \Vert P'|_V \phi\Vert_Z$, and we take $C' = \Vert \iota Q\Vert$. However, I'm not fully confident in this solution.

(c) I see that every $\ell \in \text{Ker}(P')$ satisfies $\Vert \ell \Vert_{Y^*} \leq C \Vert \iota \ell\Vert_Z \leq c\Vert \ell\Vert_{Y^*}$ for some $c > 0$. This seems useful, but I'm not sure where to go from here.

Maybe we can find $u \in X$ via some sort of "weak limit" or as an element of $X^{**}$? I figure we should use the finite-dimensionality of $\text{Ker}(P')$ and/or the compactness of $\iota$ somehow.

  • $\begingroup$ I think your attempt for (b) is correct: by taking a complement of $\ker P'$ and restricting $P'$ to its image (which is closed), you do create a bijection. However, I'm not sure why you need all the $\iota$'s in your proof though. For (c), assuming I'm interpreting the question correctly, can't you simply use the fact that $(\ker P')^\perp = \text{im} P$, which holds for all bounded linear operators $P$ with closed range? $\endgroup$
    – JHF
    Mar 26 '13 at 16:53
  • $\begingroup$ @JHF: Ah, yes, that should work for (c). Thanks. But why does $P$ have closed range? Is this because $V$ is closed? Also, if you write your comment as an answer, I'll accept it. $\endgroup$ Mar 26 '13 at 22:06
  • $\begingroup$ @JHF: I don't know what you mean by all the $\iota$'s in my proof for (b), though. (Oh, and I should have said take $C' = \text{max}(C, C\Vert \iota Q\Vert)$.) $\endgroup$ Mar 26 '13 at 22:08

(Continued from the comments above.)

(b) To clarify what I meant about the $\iota$'s, I just thought it might be simpler to note that for $\phi \in V$, $\phi = QP'|_V \phi$, so $\lVert \phi \rVert_{Y^*} \leq \lVert Q \rVert \lVert P' \phi \rVert_{X^*}$, but maybe I'm missing something?

(c) We use the assumption $\lVert \phi \rVert_{Y^*} \leq C(\lVert P'\phi \rVert_{X^*} + \lVert \iota\phi \rVert_Z)$ to show that $P$ (or equivalently, $P'$) has closed range. (Indeed, this assumption is equivalent to having finite-dimensional kernel and closed range.)

First, recall that $P'|_V$ is injective by construction. Let $(u_n)$ be a sequence in $V$ such that $P'u_n \to x$ in $X^*$. We want to show that $x \in \text{im } P'$. To do this, note that $(u_n)$ is bounded; otherwise, let $v_n = \frac{u_n}{\lVert u_n \rVert_{Y^*}}$, so we would have \begin{equation} \lVert v_n - v_m \rVert_{Y^*} \leq C(\lVert P' v_n - P' v_m \rVert_{X^*} + \lVert \iota v_n - \iota v_m \rVert_Z). \end{equation}

The right side converges to zero (along some subsequence) since $P' u_n$ has a limit, $(u_n)$ is unbounded, and $\iota$ is compact. So some subsequence of $(v_n)$ is Cauchy, hence has a limit $v$. However, we have $\lVert v \rVert_{Y^*} = 1$ but $P' v = 0$, contradicting injectivity.

Thus $(u_n)$ is bounded and therefore there is a convergent subsequence $(\iota u_n)$. Again using \begin{equation} \lVert u_n - u_m \rVert_{Y^*} \leq C(\lVert P' u_n - P' u_m \rVert_{X^*} + \lVert \iota u_n - \iota u_m \rVert_Z) \to 0, \end{equation} we see that $(u_n)$ has a Cauchy subsequence which converges to some $u \in Y^*$. Then $x = P'u$, so $\text{im } P'$ is closed as wanted.


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.