Is the following assertion and the proof below correct,
or am I missing something very important?

Moreover, would the corollaries be correct then?

Besides, I would also appreciate a lot any comment, advice, suggestion about the mathematical writing like structure, titles, phrasing, layout, arguments, etc. :)


Assertion: Given C*-algebras $A$ and $D$ and let $I\trianglelefteq A$ be a closed two-sided ideal in $A$.
Consider their algebraic tensor products $A\odot D$ and $I\odot D$, respectively.

Let $B(H)$ denote the bounded operators over a Hilbert space $H$.

Then every *-morphism $\varphi:I\odot D\to B(H)$ extends to $$ \begin{matrix} A\odot D & \xrightarrow{\Phi} & B(H) \\ \uparrow & \nearrow\varphi & \\ I\odot D \end{matrix}$$ for a unique *-morphism $\Phi:A\odot D\to B(H)$.


As an interesting application of the assertion we would obtain the following

Corollary: Given C*-algebras $A$ and $B$.
Denote their multiplier algebras by $\mathcal{M}A$ and $\mathcal{M}B$, respectively.
Consider their algebraic tensor products $A\odot B$ and $\mathcal{M}A\odot\mathcal{M}B$.

Let $B(H)$ denote the bounded operators over a Hilbert space $H$.

Then every *-morphism $\varphi:A\odot B\to B(H)$ extends to $$ \begin{matrix} \mathcal{M}A\odot \mathcal{M}B & \xrightarrow{\Phi} & B(H) \\ \uparrow & \nearrow\varphi & \\ A\odot B \end{matrix}$$ for a unique *-morphism $\Phi:A\odot D\to B(H)$.

as well as a nice proof for the restriction theorem:

Corollary: Given C*-algebras $A$ and $B$.
Consider their algebraic tensor product $A\odot B$.

Let $B(H)$ denote the bounded operators over a Hilbert space $H$.

Then every *-morphism $\varphi:A\odot B\to B(H)$ restricts to $$ \begin{matrix} A\odot B & \xrightarrow{\varphi} & B(H) \\ \uparrow & \nearrow\varphi_A\cdot\varphi_B & \\ A\times B \end{matrix}$$ for unique *-morphisms $\varphi_A:A\to B(H)$ and $\varphi_B:B\to B(H)$.

Moreover, $\varphi_A$ and $\varphi_B$ commute.


Proof: a) Existence:

Though the idea of the construction is pretty simple, the proof can get pretty lengthy in detail.
Thus let us split the proof into several parts:

i.) Reduction to non-degenerate *-morphism:

Denote the kernel of the image by $$\mathcal{N}(I\odot D):=\{v\in H:\varphi(I\odot D)v=0\}=\bigcap_{y\in I\odot D}\mathcal{N}y$$ while the range of the image by $$\mathcal{R}(I\odot D):=\varphi(I\odot D)H=\bigcup_{y\in I\odot D}\mathcal{R}y$$ and denote its closed linear span by $H_\varphi:=\overline{\operatorname{span}}\mathcal{R}(I\odot D)$.

Then we obtain the orthogonal decomposition $$H=H_\varphi\oplus H_\varphi^\bot=\overline{\operatorname{span}}\mathcal{R}(I\odot D)\oplus\mathcal{N}(I\odot D)$$ and whence the matrix representation $$\varphi=\begin{pmatrix}\pi&0\\0&0\end{pmatrix}$$ where $\pi:I\odot D\to B(H_\varphi)$ is a non-degenerate *-morphism.

For shorthand notation redefine $H:=H_\varphi$.

Thus the problem reduces to extending a non-degenerate *-morphism $\pi:I\odot D\to B(H)$.

ii.) Construction:

Fix an element $d_0\in D$ and define the map $\Pi(\_\odot d_0):A\to B(H)$ by $$\Pi(a\odot d_0)\pi(i\odot d)v:=\pi(ai\odot d_0d)v$$ for $a\in A$, $i\in I$, $d\in D$ and $v\in H$ and extend linearly.

iii.) Well-Definedness:

This is indeed well-defined by the following standard argument,
which is actually the heart of the proof:

As a first instance, note that for $d_0\geq0$ the linear map $$\pi(\_\odot d_0):I\to B(H):i\mapsto\pi(i\odot d_0)$$ is positive between C*-algebras and as such bounded.

But a C*-algebra is spanned by its positive elements,
and thus $\pi(\_\odot d_0)$ is bounded also for $d_0\in D$ not necessarily positive.

With this observation in hand we obtain that $$\pi(eai\odot d_0)\xrightarrow{e\to1}\pi(ai\odot d_0)$$ for $a\in A$, $i\in I$ and any approximate unit $e\in I$ for $I$.

Let now $a\in A$ and $\sum_ki_k\odot d_kv_k\in\operatorname{span}\pi(I\odot D)H$.
Moreover let $e\in I$ be a, now bounded, approximate unit for $I$. Then \begin{align*} \|\Pi(a\odot d_0)\sum_k\pi(i_k\odot d_k)v_k\|=&\|\sum_k\pi(ai_k\odot d_0d_k)v_k\|\\ =&\|\lim_e\sum_k\pi(eai_k\odot d_0d_k)v_k\|\\ =&\lim_e\|\pi(ea\odot d_0)\sum_k\pi(i_k\odot d_k)v_k\|\\ \leq&\sup_e\|\pi(\_\odot d_0)\|\cdot\|ea\|\cdot\|\sum_k\pi(i_k\odot d_k)v_k\|\\ \leq&\|\pi(\_\odot d_0)\|\cdot\sup_e\|e\|\cdot\|a\|\cdot\|\sum_k\pi(i_k\odot d_k)v_k\|, \end{align*} whence the operator $\Pi(a\odot d_0)$ is well-defined and bounded on $\operatorname{span}\pi(I\odot D)H$,
and thus, since $\pi$ is non-degenerate, extends to a bounded operator on $H$.

Hence, the map $\Pi(\_\odot d_0)$ is well-defined.

iv.) *-Morphism:

We may now vary the previously fixed element $d_0\in D$ as well to obtain a bilinear map $$\Pi:A\times D\to B(H):(a,d)\mapsto \Pi(a\odot d)$$ and so to obtain a linear map, also denoted by $$\Pi:A\odot D\to B(H):a\odot d\mapsto\Pi(a\odot d).$$ The resulting map is easily seen to be a *-morphism:

For $a_1,a_2\in A$, $d_1,d_2\in D$ as well as $\pi(i\odot d)v\in\operatorname{span}\pi(I\odot D)H$ we have \begin{align*} \Pi(a_1\odot d_1)\Pi(a_2\odot d_2)\pi(i\odot d)v=&\pi(a_1a_2i\odot d_1d_2d)v\\ =&\Pi(a_1a_2\odot d_1d_2)\pi(i\odot d)v, \end{align*} and thus, again since $\pi$ is non-degenerate, $$\Pi(a_1\odot d_1)\Pi(a_2\odot d_2)=\Pi(a_1a_2\odot d_1d_2).$$

Involutivity follows a similar vein.

b) Uniqueness: Here, I'm not sure about the degenerate case!


There is no uniqeness in general and I think that existence is a standard fact. Indeed, since $\varphi$ is a $\ast$-homomorphism, it extends to a $\ast$-homomorphism $\widetilde{\varphi}: I\otimes_{\mathrm{max}} D \to B(H)$. Now $I\otimes_{\mathrm{max}} D$ is an ideal in $A\otimes_{\mathrm{max}} D$, and representations always extend from ideals (using approximate identity). Therefore we get a $\ast$-homomorphism $\widetilde{\Phi}: A \otimes_{\mathrm{max}} D \to B(H)$ which restricts to $\Phi: A \odot D \to B(H)$.

As for uniqueness, it is correct to say that degenerate case is the problem. We will just consider $D=\mathbb{C}$. Suppose that $A = B\oplus C$ is a direct sum of $C^{\ast}$-algebras and let $\varphi: B \to B(H)$ be a representation and then view it as a representation on $H\oplus K$, where $K$ is some Hilbert space. Then any representation $\sigma: C\to B(K)$ would provide an extension of $\varphi$ of the form $\varphi\oplus\sigma$ and this can be highly non-unique.


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.