Suppose $T:\ell_1\to\ell_1$ is a continuous linear operator. When can we say that $T$ is a dual, or adjoint, of an operator on $c_0$? In other words, under what conditions can we find a continuous linear $S:c_0\to c_0$ such that $T=S^*$?

Recall that $c_0$ is a subspace of $\ell_\infty=\ell_1^*$, and so we can talk about the restriction $T^*|_{c_0}$ of $T^*:\ell_\infty\to\ell_\infty$ to $c_0$.

Observation. If $T^*c_0\subseteq c_0$ then $T=S^*$ for some continuous linear $S:c_0\to c_0$.

Proof. Notice that $(T^*|_{c_0})^*$ is a continuous linear operator acting on $\ell_1$. We claim $T=(T^*|_{c_0})^*$. If not, then there is $z^*\in\ell_1$ so that $Tz^*\neq(T^*|_{c_0})^*z^*$. Hence, there is $y^{**}\in c_0\subset\ell_\infty$ such that $(Tz^*)(y^{**})\neq((T^*|_{c_0})^*z^*)(y^{**})$. However,


a contradiction. $\square$

So, using the above, we could look for conditions to guarantee that $T^*c_0\subseteq c_0$. Perhaps it might help to try to adapt the above proof to work for certain kinds of transformations, say $U,V:\ell_\infty\to\ell_\infty$ with $VT^*Uc_0\subseteq c_0$.

Or, we could try an entirely different approach. Suggestions are welcome!

P.S. It would be fantastic if it turned out that if $T$ fails to be strictly singular (or have countable spectrum) then $T=S^*$ for some $S$. However, that seems too much to hope for.

  • $\begingroup$ Some thoughts: In general, a bounded operator between dual spaces is an adjoint operator iff it is weak*-weak* continuous. Also of use may be the fact that operators from $\ell_1$ to a Banach space $X$ are characterized by the bounded sequences in $X$ (given a bounded sequence $(x_i)$ in $X$, map $e_i$ to $x_i$). So in order to be an adjoint operator, $(T e_i)$ must be weak* null. An example: define $T:\ell_1\rightarrow\ell_1$ via $Te_1=e_1$ and $Te_j=e_1+e_j$, $j>1$. $T$ is continuous, in fact an isomorphism, but not weak*-weak* continuous. $\endgroup$ – David Mitra Aug 24 '14 at 21:30
  • $\begingroup$ If $(x_n^*)$ is weak* null in $\ell_1$ then $Se_i=x_i^*$ gives an operator from $\ell_1$ to $\ell_1$. One can define a bounded operator from $c_0$ to $c_0$ via $T x=(x_n^* x)$. One then has $T^*=S$, no? $\endgroup$ – David Mitra Aug 24 '14 at 22:00

We have the following characterization of adjoint operators:

Suppose that $X$ and $Y$ are normed spaces. If $T:X\rightarrow Y$ is a bounded linear operator, then $T^*$ is weak*-weak* continuous. Conversely, if $S$ is a weak*-weak* continuous linear operator from $Y^*$ to $X^*$, then there is a bounded linear operator $T:X\rightarrow Y$ such that $T^*=S$.

For a proof, see Robert E. Megginson's An Introduction to Banach Space Theory, Theorem 3.1.11.

The following two facts will also be useful here:

The bounded linear operators from $\ell_1$ to a Banach space $X$ correspond to the bounded sequences in $X$. The correspondence is given by $Te_i^*=x_i$ for the bounded sequence $(x_i)$, where $e_i^*$ is the $i^{\rm th}$ standard unit vector in $\ell_1$.

The bounded linear operators from a Banach space $X$ to $c_0$ correspond to the weak*-null sequences in $X^*$. The correspondence is given by $Tx=(x_i^* x)$ for the weak*-null sequence $(x_i^*)$.

From the first (and easy) part of the characterization of adjoint operators, one sees that if $T:\ell_1\rightarrow\ell_1$ is the adjoint of a bounded operator, then $(Te_i^*)$ is weak*-null.

Conversely, suppose $(x_i^*)$ is a weak*-null sequence in $\ell_1$. Then $(x_i^*)$ is bounded in $\ell_1$; so, the linear operator $T:\ell_1\rightarrow\ell_1$ defined by $Te_i^*=x_i^*$ is bounded. Also, the linear operator $S:c_0\rightarrow c_0$ defined by $Sx=(x_i^* x)$ is bounded. One can easily show $S^* e_i^* =x_i^*$ for each $i$:
$$ (S^* e_i^*)e_j=e_i^*(Se_j)=e_i^*(x_n^* e_j)=x_i^*e_j,\text{ for all }j. $$ It follows that, $S^*=T$.

To summarize, an operator $T:\ell_1\rightarrow\ell_1$ is the adjoint of a bounded operator if and only if $(Te_i^*)$ is weak*-null.

  • $\begingroup$ Wow, that's pretty sweet, thank you! I will try to work with this, to get what I need. $\endgroup$ – Ben W Aug 25 '14 at 1:48

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.