Kreyszig's Functional analysis seems to introduce the hilbert adjoint operator by means of an explicit representation. I haven't seen this anywhere else and I would like to confirm this explicit solution for the Hilbert adjoint operator, that is $T^*:H_2\rightarrow H_1$ such that $\langle Tx,y\rangle=\langle x, T^*y\rangle, \forall x\in H_1, y\in H_2$.

The following discussion is pieced together from 2 theorems from Kreyszig's Introduction to Functional Analysis text.

First, statements of the 2 Theorems whose results piece together into the adjoint operator. I've left out some of their details that I didn't deem relevant.

3.8-1 Riesz's Theorem (Functionals on Hilber spaces): Every bounded linear functional $h$ on a Hilbert space $H$ can be represented in terms of the inner product, namely, $$f\left(x\right)=\left\langle x,z\right\rangle $$ where $z$ is unique and given by $$z=\frac{\overline{f\left(z_{0}\right)}}{\left\langle z_{0},z_{0}\right\rangle }z_{0},\hspace{1em}z_{0}\in\mathcal{N}\left(f\right)^{\perp} $$

($\mathcal{N(f)}$ is the null space of $f$. $z_0$ is chosen as such so that $\langle x, z \rangle = 0 $ for all $x : f(x)=0$)

3.8-4 Riesz representation: Let $H_{1} , H_{2} $ Hilbert spaces and $h:\ H_{1}\times H_{2}\rightarrow K $ a bounded sesquilinear form. Then $h$ has a representation $$h\left(x,y\right)=\left\langle Sx,y\right\rangle $$

The proof from the text considers $\overline{h\left(x,y\right)} $, keeps $x$ fixed (so as if $\overline{h\left(x,y\right)}$ is a function of 1 variable, $y$ ), and then applies Riesz's Theorem (above) such that $$\overline{h\left(x,y\right)}=\left\langle y,z\right\rangle ,\hspace{1em}z=\frac{h\left(x,z_{0}\right)}{\left\langle z_{0},z_{0}\right\rangle }z_{0},\hspace{1em}z_{0}\in\mathcal{N}\left(h\left(x,y\right)\right)^{\perp} $$

and taking conjugates:$$ h\left(x,y\right)=\left\langle z,y\right\rangle $$

and now we let $z=Sx$

and there we have the unique representation.

Now an explicit form for $T^{\star}$: Working backwards from the previous proof (replacing $S$ with $T^{*}$ , $x$ with $y$ ) I have $$T^{*}y=z=\frac{h\left(y,z_{0}\right)}{\left\langle z_{0},z_{0}\right\rangle }z_0,\hspace{1em}z_{0}\in\mathcal{N}\left(h\left(y,x\right)\right)^{\perp} $$

with$$h\left(y,x\right)=\left\langle z,x\right\rangle =\left\langle T^{*}y,x\right\rangle $$

which, by definition of the Hilbert adjoint operator $$\cdots=\left\langle y,Tx\right\rangle $$

so does this give us an explicit form for the adjoing operator? i.e. $$T^{*}y=\frac{\left\langle y,Tx\right\rangle }{\left\langle z_{0},z_{0}\right\rangle }z_0,\hspace{1em}z_{0}\in\mathcal{N}\left(\left\langle y,Tx\right\rangle \right)^{\perp} $$

Is this a correct "explicit" form of the adjoint operator? the hard part seems to just find the $z_0$. If so, why can't I find this form of the adjoint operator anywhere? It seems difficult to "solve" for $T^*$ straight from the definition $\langle Tx,y\rangle=\langle x, T^*y\rangle$ (like as they did in here: Finding the adjoint of an operator, or as I've seen $V^* = A^H$ proved for matrices)

Thanks for your help.


In case anyone's still reading this, here's an elaboration of Martin Argerami's answer. In short, your formula isn't in any way a formula for actually computing anything, but rather an encapsulation of the proof of Riesz's theorem. Fleshing things out, if $f \in H^\ast$ is non-zero:

  1. On entirely abstract grounds, $\mathbb{C} = \operatorname{Ran}(f) \cong H/\operatorname{Ker}(f) \cong \operatorname{Ker}(f)^\perp$, so that $\operatorname{Ker}(f)^\perp$ is $1$-dimensional, and hence $\operatorname{Ker}(f)^\perp = \mathbb{C} z_0$ for any non-zero $z_0 \in \operatorname{Ker}(f)^\perp$, so fix some (viz, any) such $z_0$ as a basis for $\operatorname{Ker}(f)^\perp$. Note that this is all that one knows about what $z_0$ is.
  2. Writing $H = \operatorname{Ker}(f) \oplus \operatorname{Ker}(f)^\perp$, we see that $f$ is entirely determined by its restriction to $\operatorname{Ker}(f)^\perp$. However, $\operatorname{Ker}(f)^\perp$ is $1$-dimensional, so by finite-dimensional linear algebra, its dual $\left(\operatorname{Ker}(f)^\perp\right)^\ast$ is also $1$-dimensional, and thus spanned by any non-zero functional, e.g., $f_0 : x \mapsto \left\langle x,z_0 \right\rangle$. Thus, there exists some constant $\alpha \in \mathbb{C}$ such that $f|_{\operatorname{Ker}(f)^\perp} = \alpha f_0$. Plugging in our distinguished (but not explicitly known!) vector $z_0$, we find that $f(z_0) = \alpha f_0(z_0) = \alpha \left\langle z_0,z_0\right\rangle$, and hence that $$\alpha = \frac{f(z_0)}{\left\langle z_0,z_0 \right\rangle}.$$
  3. Putting everything together, one has that $f(x) = \left\langle x, \overline{\alpha}z_0 \right\rangle$ for $x \in \operatorname{Ker}(f)^\perp$, and hence for all $x \in H$, yielding your formula.

So, to cut a long story short, the formula you gave for $z$ depends entirely on a vector $z_0$ whose existence, guaranteed on purely abstract grounds, is the only thing we know about it. As a result, unless by some miracle you should happen to know an explicit non-zero vector in $\operatorname{Ker}(f)^\perp$, in which case you can take that as your $z_0$, the formula for $z$ is of no real computational use to you at all. In particular, it cannot possibly help you get an explicit formula for the adjoint of an operator.


Note that your $T^*$ has one-dimensional range (scalar multiples of $z_0$), which is not the case unless $T$ is one-dimensional to start with.

I cannot really follow what you are trying to do in your proof. In particular you make $z_0$ appear as of nowhere.

  • $\begingroup$ Thanks - I just edited my question to make it (hopefully) clearer $\endgroup$ – chris Dec 3 '12 at 19:49
  • $\begingroup$ Note that your "proof" of the Riesz Representation is not such. You write twice formulas where the left-hand-side is a vector and the right-hand-side is a number. And you are still introducing $z_0$ without saying what it is. $\endgroup$ – Martin Argerami Dec 3 '12 at 19:57
  • $\begingroup$ Yes of course - just found the typo, I forgot the $z_0$ on the rhs of the fraction of inner products. Thanks for pointing that out! $\endgroup$ – chris Dec 4 '12 at 15:13
  • 1
    $\begingroup$ Still, there's nothing really "explicit" about your formula. You are hiding information in "$z_0$". $\endgroup$ – Martin Argerami Dec 4 '12 at 15:17

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.