Exactness of the Cuntz-Pimsner Construction

Menevşe Eryüzlü Paulovicks Department of Mathematics, University of Colorado Boulder, Boulder, CO 80309-0395 [email protected]

(Date: September 26, 2024)

Abstract.

In prior work we described how the Cuntz-Pimsner construction may be viewed as a functor. The domain of this functor is a category whose objects are $C^{*}$ -correspondences and morphisms are isomorphism classes of certain pairs comprised of a $C^{*}$ -correspondence and an isomorphism. The codomain is the well-studied category whose objects are $C^{*}$ -algebras and morphisms are isomorphism classes of $C^{*}$ -correspondences. In this paper we show that certain fundamental results in the theory of Cuntz-Pimsner algebras are direct consequences of the functoriality of the Cuntz-Pimsner construction. In addition, we describe exact sequences in the target and domain categories, and prove that the Cuntz-Pimsner functor is exact.

Key words and phrases:

C^{*}

-correspondence, Cuntz-Pimsner algebra, exact sequence, exact functor

2020 Mathematics Subject Classification:

Primary 46L08; Secondary 18B99

1. Introduction

In [3] we introduced a categorical framework for viewing the Cuntz-Pimsner construction as a functor, which thereby allows one to determine relationships between Cuntz-Pimsner algebras from relationships between the defining $C^{*}$ -correspondences. The domain of this functor is the category $\mathsf{C^{*}cor_{pair}}$ , introduced in [3], that has $C^{*}$ -correspondences as objects, and a morphism from ${}_{A}X_{A}$ to ${}_{B}Y_{B}$ is the isomorphism class of the pair ( ${}_{A}M_{B}$ , $U_{M}$ ), where ${}_{A}M_{B}$ is a $C^{*}$ -correspondence satisfying certain conditions, and

U_{M}:{}_{A}(X\otimes_{A}M)_{B}\rightarrow{}_{A}(M\otimes_{B}Y)_{B}

is a $C^{*}$ -correspondence isomorphism. The codomain category $\mathsf{C^{*}alg_{cor}}$ , which has sometimes been called the Enchilada Category in the literature, has $C^{*}$ -algebras as objects and isomorphism classes of $C^{*}$ -correspondences as morphisms. For any pair ( ${}_{A}M_{B}$ , $U_{M}$ ) one can construct a covariant representation ( $\pi,\Phi$ ) of ${}_{A}X_{A}$ on $\mathcal{K}(M\otimes_{B}\mathcal{O}_{Y})$ . Then the universal property of Cuntz-Pimsner algebras assures the existence of an associated homomorphism $\sigma_{(\pi,\Phi)}:\mathcal{O}_{X}\rightarrow\mathcal{K}(M\otimes_{B}\mathcal% {O}_{Y}),$ which allows us to view $M\otimes_{B}\mathcal{O}_{Y}$ as an $\mathcal{O}_{X}$ – $\mathcal{O}_{Y}$ -correspondence. It is shown in [3] that there exists a functor $\mathcal{E}$ from $\mathsf{C^{*}cor_{pair}}$ to $\mathsf{C^{*}alg_{cor}}$ that maps a $C^{*}$ -correspondence ${}_{A}X_{A}$ to its Cuntz-Pimsner algebra $\mathcal{O}_{X}$ , and a morphisms from ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ is mapped to the isomorphism class of an $\mathcal{O}_{X}$ – $\mathcal{O}_{Y}$ -correspondence. The first part of this paper is devoted to using this functor to obtain some well-known Cuntz-Pimsner algebra results. Specifically, if ( $\Upsilon,t$ ) is a universal covariant representation of ${}_{A}X_{A}$ , and $I$ is a positive $X$ -invariant ideal of $A$ , then the following hold:

(i)

$\mathcal{O}_{IX}$ is isomorphic to the smallest hereditary subalgebra of $\mathcal{O}_{X}$ containing $\Upsilon(I)$ .
(ii)

$\mathcal{O}_{IX}$ is Morita equivalent to the ideal $\left\langle\Upsilon(I)\right\rangle$ generated by $\Upsilon(I)$ in $\mathcal{O}_{X}$ .
(iii)

If ${}_{A}X_{A}$ is regular and $I$ is an $X$ -invariant ideal then $\mathcal{O}_{X}/\langle\Upsilon(I)\rangle\cong\mathcal{O}_{X/XI}.$

Items (i) and (ii) were proven by Katsura in [6, Proposition 9.3 and Proposition 9.5] by using what are called O-pairs. Item (iii) can be shown by combining [7, Lemma 6.3] and [5, Theorem 3.1]. Item (iii) can also be deduced by combining [6, Proposition 5.3] and [6, Proposition 8.5]; however, this requires a deep understanding of O-pairs and the properties of $C^{*}$ -algebras generated by such pairs. The first objective of this paper is to show that the functor established in [3] can be applied to obtain easier and more direct proofs of these three results.

The work that is presented in the second part of this paper was motivated by a question frequently asked by audience members when presenting the results of [3], namely: “Can one define exact sequences in the domain and codomain categories so that the Cuntz-Pimsner functor $\mathcal{E}$ is exact?” In order to answer this question, one needs to know what is meant by an exact sequence in both the domain and codomain categories. One of the difficulties in answering this question is that one can not identify images in either categories, and it is suspected that images may not exist in either categories. Therefore the usual “kernel-image definition” of exactness may not apply. To circumvent this obstruction in each category, we define $0\rightarrow A\xrightarrow{f}B\xrightarrow{g}C\to 0$ to be a short exact sequence when $f$ is a categorical kernel of $g$ and $g$ is a categorical cokernel of $f$ (see Definition 5.8 and Definition 5.10). We characterize this “kernel-cokernel definition” of exactness in each category, showing that kernel-cokernel exactness is a tractable concept in these categories. We prove that with the kernel-cokernel definition of short exact sequence, the Cuntz-Pimsner functor is exact. We end the paper by showing that as an immediate consequence of exactness one can obtain the results (i),(ii), and (iii) listed above for the case when ${}_{A}X_{A}$ is regular.

2. Preliminaries

Let $\mathcal{C}$ be a pointed category (a category with a zero object), and let $f:A\rightarrow B$ be a morphism. A kernel of $f$ is a pair $(K,k)$ consists of an object $K$ and a morphism $k:K\rightarrow A$ such that

•

$f\circ k=0$ ;
•

whenever a morphism $h:D\rightarrow A$ satisfies $f\circ k=0$ there exists a unique morphism $p:D\rightarrow K$ such that $k\circ p=h$ .

A cokernel of $f$ is a pair $(C,c)$ consists of an object $C$ and a morphism $c:B\rightarrow C$ such that

•

$c\circ f=0;$
•

whenever a morphism $h:B\rightarrow D$ satisfies $h\circ f=0$ there exists a unique morphism $p:C\rightarrow D$ such that $p\circ c=h.$

We say that $f$ is a monomorphism if for all morphisms $g:C\to A$ and $h:C\to A$ in $\mathcal{C}$ , we have $f\circ g=f\circ h$ implies $g=h$ . We say that $f$ is an epimorphism if for all morphisms $g:B\to C$ and $h:B\to C$ in $\mathcal{C}$ , we have $g\circ f=h\circ f$ implies $g=h$ .

A $C^{*}$ -correspondence ${}_{A}X_{B}$ is a right Hilbert $B$ -module equipped with a left action given by a homomorphism $\varphi_{X}:A\rightarrow\mathcal{L}(X),$ where $\mathcal{L}(X)$ denotes the $C^{*}$ -algebra of adjointable operators on $X$ . We denote the kernel of the left action homomorphism $\varphi_{X}$ by $\operatorname{Ker}\varphi_{X}$ . For a $C^{*}$ -correspondence ${}_{A}X_{B}$ define $A\cdot X=\{\varphi_{X}(a)x:a\in A,x\in X\}$ . The correspondence ${}_{A}X_{B}$ is called nondegenerate if $A\cdot X=X$ . In this paper all our correspondences will be nondegenerate by standing hypothesis. A $C^{*}$ -correspondence ${}_{A}X_{B}$ is called injective if the left action $\varphi_{X}:A\rightarrow\mathcal{L}(X)$ is injective; it is called proper if $\varphi_{X}(A)$ is contained in the $C^{*}$ -algebra $\mathcal{K}(X)$ of compact operators on $X$ . A $C^{*}$ -correspondence ${}_{A}X_{B}$ is called regular if it is both injective and proper. For a $C^{*}$ -correspondence ${}_{A}X_{B}$ we denote the closed span of $B$ -valued inner products $\langle X,X\rangle_{B}$ by $B_{X}$ . One of the $C^{*}$ -correspondence properties we use frequently in this paper is the following: let $I$ be an ideal of $B$ such that $B_{X}\subset I$ . Then, $X$ can be viewed as an $A$ – $I$ -correspondence [4, Lemma 3.2].

A $C^{*}$ -correspondence isomorphism from ${}_{A}X_{B}$ to ${}_{A}Y_{B}$ is a bijective linear map $\Phi:X\rightarrow Y$ satisfying

(i)

$\Phi(a\cdot x)=a\cdot\Phi(x)$ ,
(ii)

$\langle x,z\rangle_{B}=\langle\Phi(x),\Phi(z)\rangle_{B}$ ,

for all $a\in A$ , and $x,z\in X.$ We let $\mathop{Ad}\Phi:\mathcal{L}(X)\rightarrow\mathcal{L}(Y)$ denote the associated $C^{*}$ -algebra isomorphism defined by $\mathop{Ad}\Phi(T)=\Phi\circ T\circ\Phi^{-1}.$

The balanced tensor product $X\otimes_{B}Y$ of an $A-B$ correspondence $X$ and a $B-C$ correspondence $Y$ is formed as follows: the algebraic tensor product $X\odot Y$ is a pre-correspondence with the $A-C$ bimodule structure satisfying

a(x\otimes y)c=ax\otimes yc\qquad\text{for }a\in A,x\in X,y\in Y,c\in C,

and the unique $C$ -valued semi-inner product whose values on elementary tensors are given by

\langle x\otimes y,u\otimes v\rangle_{C}=\langle y,\langle x,u\rangle_{B}\cdot v% \rangle_{C}\qquad\text{for }x,u\in X,y,v\in Y.

This semi-inner product defines a $C$ -valued inner product on the quotient $X{\odot}_{B}Y$ of $X\odot Y$ by the subspace generated by elements of form

x\cdot b\otimes y-x\otimes\varphi_{Y}(b)y\qquad\text{($x\in X$, $y\in Y$, $b% \in B$) }.

The completion $X\otimes_{B}Y$ of $X{\odot}_{B}Y$ with respect to the norm coming from the $C$ -valued inner product is an $A-B$ correspondence, where the left action is given by

A\rightarrow\mathcal{L}(X\otimes_{B}Y),\qquad\text{$a\mapsto\varphi_{X}(a)% \otimes 1_{Y},$ }

for $a\in A.$ We denote the canonical image of $x\otimes y$ in $X\otimes_{B}Y$ by $x\otimes_{B}y$ .

Proposition 2.1.

[4, Proposition 3.1] For $C^{*}$ -correspondences ${}_{A}X_{B}$ and ${}_{B}Y_{C}$ we have

{}_{A}(X\otimes_{B}Y)_{B}\cong{}_{A}0_{B}\iff{}_{A}(X\otimes_{B}Y)_{B}={}_{A}0% _{B}\iff B_{X}\subset\operatorname{Ker}\varphi_{Y}.

Lemma 2.2 ([5]).

Let $X$ be a $C^{*}$ -correspondence over $A$ and let ${}_{A}Y_{B}$ be an injective $C^{*}$ -correspondence. Then the map $\iota:T\mapsto T\otimes 1_{Y}$ gives an isometric homomorphism of $\mathcal{L}(X)$ into $\mathcal{L}(X\otimes_{A}Y).$ If, in addition, $\varphi_{Y}(A)\subset\mathcal{K}(Y)$ , then $\iota$ embeds $\mathcal{K}(X)$ into $\mathcal{K}(X\otimes_{A}Y).$

A Hilbert bimodule ${}_{A}X_{B}$ is a $C^{*}$ -correspondence that is also equipped with an $A$ -valued inner product ${}_{A}\langle\cdot,\cdot\rangle$ , which satisfies

{}_{A}\langle a\cdot x,y\rangle=a\cdot{}_{A}\langle x,y\rangle\qquad\text{and % }\qquad{}_{A}\langle x,y\rangle^{*}={}_{A}\langle y,x\rangle

for all $a\in A,x,y\in X$ , as well as the compatibility property

{}_{A}\langle x,y\rangle\cdot z=x\cdot\langle y,z\rangle_{B}\qquad\text{for }x% ,y,z\in X.

A Hilbert bimodule ${}_{A}X_{B}$ is left-full if the closed span of ${}_{A}\langle X,X\rangle$ is all of $A$ .

An imprimitivity bimodule ${}_{A}X_{B}$ is a Hilbert bimodule that is full on both the left and the right. The identity correspondence on $A$ is the Hilbert bimodule ${}_{A}A_{A}$ where the bimodule structure is given by multiplication, and the inner products are given by

{}_{A}\langle a,b\rangle=ab^{*},\hskip 14.22636pt\langle a,b\rangle_{A}=a^{*}b% ,\qquad\text{for $a,b\in A$ }.

A representation $(\pi,t)$ of a $C^{*}$ -correspondence ${}_{A}X_{A}$ on a $C^{*}$ -algebra $B$ consists of a $*-$ homomorphism $\pi:A\rightarrow B$ and a linear map $t:X\rightarrow B$ such that

\pi(a)t(x)=t(\varphi_{X}(a)(x))\qquad\text{and }\qquad t(x)^{*}t(y)=\pi(% \langle x,y\rangle_{A}),

for $a\in A$ and $x,y\in X$ , where $\varphi_{X}$ is the left action homomorphism associated with ${}_{A}X_{A}$ . For any representation $(\pi,t)$ of ${}_{A}X_{A}$ on $B$ , there is an associated homomorphism $\psi_{t}:\mathcal{K}(X)\rightarrow B$ satisfying $\psi_{t}(\theta_{x,x^{\prime}})=t(x)t(x^{\prime})^{*}$ for $x,x^{\prime}\in X$ . The representation $(\pi,t)$ is called injective if $\pi$ is injective, in which case $t$ is an isometry. We denote the $C^{*}$ -algebra generated by the images of $\pi$ and $t$ in $B$ by $C^{*}$ $(\pi,t).$

Consider a $C^{*}$ -correspondence ${}_{A}X_{A}$ . The ideal $J_{X}$ is define as

	$\displaystyle J_{X}$	$\displaystyle=\varphi_{X}^{-1}(\mathcal{K}(X))\cap(\operatorname{Ker}\varphi_{% X})^{\perp}$
		$\displaystyle=\text{\{$a\in A$ : $\varphi_{X}(a)\in\mathcal{K}(X)$ and $ab=0$ % for all $b\in\operatorname{Ker}\varphi_{X}$}\},$

and is called the Katsura ideal. Notice here that for a regular ${}_{A}X_{A}$ we have $J_{X}=A.$

A representation $(\pi,t)$ of ${}_{A}X_{A}$ is called covariant if $\pi(a)=\Psi_{t}(\varphi_{X}(a))$ , for all $a\in J_{X}.$ The $C^{*}$ -algebra generated by the universal covariant representation of ${}_{A}X_{A}$ is called the Cuntz-Pimsner algebra $\mathcal{O}_{X}$ of ${}_{A}X_{A}$ .

3. Categories and the covariant representation

In this section we briefly explain the construction of the functor $\mathcal{E}$ defined in [3], and recall the related categories. The range category $\mathsf{C^{*}alg_{cor}}$ of $\mathcal{E}$ is sometimes called “the enchilada category” as in [4]. In this category our objects are $C^{*}$ -algebras, and a morphism from $A$ to $B$ is the isomorphism class of an $A$ – $B$ -correspondence. The composition [ ${}_{B}Y_{C}$ ] $\circ$ [ ${}_{A}X_{B}$ ] is the isomorphism class of the balanced tensor product ${}_{A}(X\otimes_{B}Y)_{C}$ ; the identity morphism on $A$ is the isomorphism class of the identity correspondence ${}_{A}A_{A}$ , and the zero morphism $A\rightarrow B$ is $[{}_{A}0_{B}].$ Note that a morphism [ ${}_{A}X_{B}$ ] is an isomorphism in $\mathsf{C^{*}alg_{cor}}$ if and only if ${}_{A}X_{B}$ is an imprimitivity bimodule [2, Proposition 2.6].

We need the following definition for the domain category.

Definition 3.1.

[3, Definition 3.1] For $C^{*}$ -correspondences ${}_{A}X_{A}$ , ${}_{B}Y_{B}$ , and $A$ – $B$ -correspondences ${}_{A}M_{B}$ , ${}_{A}N_{B}$ , let $U_{M}:X\otimes_{A}M\rightarrow M\otimes_{B}Y$ and $U_{N}:X\otimes_{A}N\rightarrow N\otimes_{B}Y$ be $A$ – $B$ -correspondence isomorphisms. The pairs ( ${}_{A}M_{B}$ , $U_{M}$ ) and ( ${}_{A}N_{B}$ , $U_{N}$ ) are called isomorphic if

•

there exists an isomorphism $\xi$ : ${}_{A}M_{B}$ $\rightarrow$ ${}_{A}N_{B}$ ; and

•

the diagram

commutes.

We denote the isomorphism class of the pair ( ${}_{A}M_{B}$ , $U_{M}$ ) by $[{}_{A}M_{B},U_{M}]$ .

Remark 3.2.

For a $C^{*}$ -correspondence ${}_{A}M_{B}$ , let $I$ and $J$ be ideals of $A$ and $B$ , respectively. We denote the map

M\otimes_{B}J\rightarrow MJ,\qquad\text{ $m\otimes_{B}j\mapsto m\cdot j$ }

by $\xi_{(r,M,J)}$ , where $m\in M$ , $j\in J$ . This map defines an $A$ – $B$ -correspondence isomorphism as well as an $A$ – $J$ -correspondence isomorphism. Similarly, we denote the map

I\otimes_{A}M\rightarrow IM,\qquad\text{ $i\otimes_{A}m\mapsto i\cdot m$ }

by $\xi_{(l,M,I)},$ where $m\in M$ , $i\in I.$ This map defines an $A$ – $B$ -correspondence isomorphism as well as an $I$ – $B$ -correspondence isomorphism.

Theorem 3.3 ([3, Theorem 3.2]).

There exists a category $\mathsf{C^{*}cor_{pair}}$ such that

•

objects are $C^{*}$ -correspondences ${}_{A}X_{A}$ ;
•

morphisms ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ are isomorphism classes [ ${}_{A}M_{B}$ , $U_{M}$ ] where $U_{M}$ denotes an $A$ – $B$ -correspondence isomorphism $X\otimes_{A}M\rightarrow M\otimes_{B}Y$ , and ${}_{A}M_{B}$ is a proper correspondence satisfying $J_{X}\cdot M\subset M\cdot J_{Y}$ ;

•

the composition [ ${}_{B}N_{C}$ , $U_{N}$ ] $\circ$ [ ${}_{A}M_{B}$ , $U_{M}$ ] is given by the isomorphism class

[

{}_{A}(M\otimes_{B}N)_{C}

U_{M\otimes_{B}N}

]

where $U_{M\otimes_{B}N}$ denotes the isomorphism $(1_{M}\otimes U_{N})(U_{M}\otimes 1_{N})$ ;

•

the identity morphism on ${}_{A}X_{A}$ is [ ${}_{A}A_{A}$ , $U_{A}$ ], where $U_{A}$ denotes the isomorphism $\xi_{l,X,A}^{-1}\circ\xi_{r,X,A}:X\otimes_{A}A\rightarrow A\otimes_{A}X$ .

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}}$ . Denote the universal covariant representation of ${}_{B}Y_{B}$ by $(\Upsilon,t).$ Let $V_{Y}:Y\otimes_{B}\mathcal{O}_{Y}\rightarrow\overline{t(Y)\mathcal{O}_{Y}}$ be the isomorphism determined on elementary tensors by

V_{Y}(y\otimes_{B}S)=t(y)S

for $y\in Y$ , $S\in\mathcal{O}_{Y}$ . Define $T:X\rightarrow\mathcal{L}(M,M\otimes_{B}Y)$ by

T(x)(m)=U_{M}(x\otimes_{A}m),

for $x\in X$ , $m\in M$ . Next, define a linear map $\Phi:X\rightarrow\mathcal{K}(M\otimes_{B}\mathcal{O}_{Y})$ by

\Phi(x)=(1_{M}\otimes V_{Y})(T(x)\otimes 1_{Y}).

and a homomorphism $\pi:A\rightarrow\mathcal{K}(M\otimes_{B}\mathcal{O}_{Y})$ by

\pi(a)=\varphi_{M}\otimes 1_{\mathcal{O}_{Y}}.

The pair $(\pi,\Phi)$ is a covariant representation of ${}_{A}X_{A}$ on $\mathcal{K}(M\otimes_{B}\mathcal{O}_{Y})$ [3, Proposition 4.2], and it is called the C-covariant representation of ${}_{A}X_{A}$ . It is injective when the homomorphism $\varphi_{M}$ is. By the universal property of $\mathcal{O}_{X}$ we obtain a $*$ -homomorphism $\sigma_{(\pi,\Phi)}:\mathcal{O}_{X}\rightarrow\mathcal{K}(M\otimes_{B}\mathcal% {O}_{Y})$ , which provides a left action of $\mathcal{O}_{X}$ on the Hilbert $\mathcal{O}_{Y}$ -module $M\otimes_{B}\mathcal{O}_{Y}$ and allows us to view $M\otimes_{B}\mathcal{O}_{Y}$ as a proper $\mathcal{O}_{X}$ – $\mathcal{O}_{Y}$ -correspondence. It is important to note that the C-covariant representation $(\pi,\Phi)$ admits a gauge action. Consequently, the homomorphism $\sigma_{(\pi,\Phi)}$ is an isomorphism onto $C^{*}$ $(\pi,\Phi)$ when ${}_{A}M_{B}$ is an injective $C^{*}$ -correspondence [3, Theorem 4.13].

Theorem 3.4 ([3, Theorem 5.1]).

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}}$ . Then the assignments ${}_{A}X_{A}$ $\mapsto\mathcal{O}_{X}$ on objects and

[{}_{A}M_{B},U_{M}]\mapsto[{}_{\mathcal{O}_{X}}(M\otimes_{B}\mathcal{O}_{Y})_{% \mathcal{O}_{Y}}]

on morphisms define a functor $\mathcal{E}$ from $\mathsf{C^{*}cor_{pair}}$ to $\mathsf{C^{*}alg_{cor}}$ .

Next Proposition is crucial for this paper.

Proposition 3.5.

Let $[{}_{A}M_{B},U_{M}]:$ ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}}$ , where ${}_{A}X_{A}$ is a regular correspondence, and let $\sigma:\mathcal{O}_{X}\rightarrow\mathcal{K}(M\otimes_{B}\mathcal{O}_{Y})$ be the associated homomorphism. Denote the universal covariant representation of ${}_{A}X_{A}$ by $(\Upsilon,t)$ . Then $\operatorname{Ker}\sigma$ is the ideal $\langle\Upsilon(\operatorname{Ker}\varphi_{M})\rangle$ generated by $\Upsilon(\operatorname{Ker}\varphi_{M})$ in $\mathcal{O}_{X}.$

Proof.

It suffices to show the equality $\operatorname{Ker}\sigma\cap\Upsilon(A)=\langle\Upsilon(\operatorname{Ker}% \varphi_{M})\rangle\cap\Upsilon(A)$ , since gauge invariant ideals of $\mathcal{O}_{X}$ are distinguished by their intersection with $\Upsilon(A)$ when ${}_{A}X_{A}$ is regular [6, Corollory 8.7]. One can easily verify that $\langle\Upsilon(\operatorname{Ker}\varphi_{M})\rangle\subset\operatorname{Ker}\sigma$ . Let $\Upsilon(a)\in\operatorname{Ker}\sigma$ . Then we have

0=\sigma(\Upsilon(a))=\varphi_{M}(a)\otimes 1_{\mathcal{O}_{Y}}.

This implies $\varphi_{M}(a)=0$ by Lemma 2.2. And thus, $a\in\operatorname{Ker}\varphi_{M}$ , which means $\Upsilon(a)\in\Upsilon(A)\cap\langle\Upsilon(\operatorname{Ker}\varphi_{M})\rangle.$ ∎

4. Invariant Ideals and Structure Theorems

Definition 4.1.

Let ${}_{A}X_{A}$ be a $C^{*}$ -correspondence. For an ideal $I$ of $A$ , define an ideal $X^{-1}(I)$ of $A$ by

\displaystyle X^{-1}(I)

\displaystyle=\{a\in A:\langle x,a\cdot y\rangle_{A}\in I\text{ for all }x,y% \in X\}.

An ideal $I$ of $A$ is said to be positive $X$ -invariant if $IX\subset XI$ , negative $X$ -invariant if $J_{X}\cap X^{-1}(I)\subset I,$ and $X$ -invariant if $I$ is both positive and negative invariant.

Note that $I$ is a positive $X$ -invariant ideal of $A$ if and only if $\langle X,IX\rangle_{A}\subset I.$ When that’s the case, we have $IX=IX\langle IX,IX\rangle_{A}\subset IXI$ . Therefore, we have the equality $IX=IXI$ . Consequently, the $I$ – $A$ -correspondence $IX$ can be viewed as a $C^{*}$ -correspondence over $I$ .

Lemma 4.2.

Let ${}_{A}X_{A}$ be a $C^{*}$ -correspondence and I be a positive X-invariant ideal of A. Denote the $I$ – $A$ -correspondence isomorphism $\xi_{(l,X,I)}^{-1}\circ\xi_{(r,IX,I)}:IX\otimes_{I}I\rightarrow I\otimes_{A}X$ by $U_{I}$ , where $\xi_{(l,X,I)}$ and $\xi_{(r,IX,I)}$ are the $I$ – $A$ -correspondence isomorphisms defined as in Remark 3.2. Then, the isomorphism class $[{}_{I}I_{A},U_{I}]:{}_{I}IX_{I}\rightarrow$ ${}_{A}X_{A}$ is a morphism in $\mathsf{C^{*}cor_{pair}}$ .

Proof.

It suffices to show $J_{IX}\cdot I\subset J_{X}$ , which follows immediately from the fact that $J_{IX}=I\cap J_{X}$ [6, Proposition 9.2]. ∎

Lemma 4.3.

For C*-algebras $A$ and $B$ , let $A\subset B$ . Then we have the $C^{*}$ -algebra isomorphism $\mathcal{K}(AB)\cong ABA$ , where $AB$ is viewed as a Hilbert $B$ -module .

Proof.

For any $x\in ABA$ , consider the operator $T_{x}:AB\rightarrow AB$ defined by $T_{x}(y)=xy$ , where $y\in AB.$ Then each $T_{x}$ is an element of $\mathcal{K}(AB)$ , and the map $L:ABA\rightarrow\mathcal{K}(AB)$ defined by $x\mapsto T_{x}$ is an injective $*$ -homomorphism. Now take any $\theta_{a_{1}b_{1},a_{2}b_{2}}\in\mathcal{K}(AB)$ . We have $\theta_{a_{1}b_{1},a_{2}b_{2}}=T_{a_{1}b_{1}b_{2}^{*}a_{2}^{*}}=L(a_{1}b_{1}b_% {2}^{*}a_{2}^{*})$ . And thus, $L$ is surjective. ∎

Theorem 4.4.

Let ${}_{A}X_{A}$ be a $C^{*}$ -correspondence and I be a positive X-invariant ideal of $A$ . Let $(\Upsilon,t)$ be the universal covariant representation of ${}_{A}X_{A}$ . Then we have the following:

(1)

$\mathcal{E}\left([{}_{I}I_{A},U_{I}]\right)=\left[{}_{\mathcal{O}_{IX}}(I% \otimes_{A}\mathcal{O}_{X})_{\mathcal{O}_{X}}\right]$ is an isomorphism class of a left-full Hilbert bimodule.
(2)

$\mathcal{O}_{IX}$ is isomorphic to the smallest hereditary subalgebra of $\mathcal{O}_{X}$ containing $\Upsilon(I)$ [6, Proposition 9.3].
(3)

$\mathcal{O}_{IX}$ is Morita equivalent to the ideal $\left\langle\Upsilon(I)\right\rangle$ generated by $\Upsilon(I)$ in $\mathcal{O}_{X}$ . [6, Proposition 9.5].

Proof.

Let $\xi:I\otimes_{A}\mathcal{O}_{X}\rightarrow\Upsilon(I)\mathcal{O}_{X}$ denote the Hilbert $\mathcal{O}_{X}$ -module isomorphism defined on elementary tensors by $i\otimes_{A}S\mapsto\Upsilon(i)S$ . Denote the C-covariant representation of ${}_{A}X_{A}$ by ( $\pi,\Phi$ ), and let $L:\Upsilon(I)\mathcal{O}_{X}\Upsilon(I)\rightarrow\mathcal{K}(\Upsilon(I)% \mathcal{O}_{X})$ be the isomorphism defined as in the proof of Lemma 4.3. Then we have the following diagram.

We claim that $\mathop{Ad}\xi^{-1}\circ L$ is an isomorphism onto $C^{*}$ ( $\pi,\Phi$ ). It suffices to show the equalities

\mathop{Ad}\xi[\Phi(ixj)]=L\left(\Upsilon(i)t(x)\Upsilon(j)\right)\text{ and }% \mathop{Ad}\xi[\pi(iaj)]=L\left(\Upsilon(iaj)\right),

for any $i,j\in I,x\in X,$ and $a\in A.$ Let $V:X\otimes_{A}\mathcal{O}_{X}\rightarrow\overline{t(X)\mathcal{O}_{X}}$ denote the $A-\mathcal{O}_{X}$ correspondence isomorphism defined on elementary tensors by $x\otimes_{A}S\mapsto t(x)S$ , for any $x\in X$ , $S\in\mathcal{O}_{X}$ . For $k\in I,S\in\mathcal{O}_{X}$ we have

$\displaystyle\xi\Phi(ixj)(k\otimes_{A}S)$	$\displaystyle=\xi(1_{I}\otimes V)U_{I}(ixj\otimes_{I}k)\otimes_{A}S$
	$\displaystyle=\xi(a\otimes_{A}t(z)S)$	(where $a\in I,z\in X$ with $az=ixjk$ )
	$\displaystyle=\Upsilon(a)t(z)S$
	$\displaystyle=t(ixjk)S.$

On the other hand, we have

L\left(\Upsilon(i)t(x)\Upsilon(j)\right)\xi(k\otimes_{A}S)=\Upsilon(i)t(x)% \Upsilon(j)\Upsilon(k)S=t(ixjk)S,

which proves the first equality. For the second equality we observe that

\xi\pi(iaj)(k\otimes_{A}S)=\xi(iajk\otimes_{A}S)=\Upsilon(iajk)S=L\left(% \Upsilon(iaj)\right)\xi(k\otimes_{A}S),

which proves our claim.

We may now conclude that the injective $*$ -homomorphism $\sigma:\mathcal{O}_{IX}\rightarrow\mathcal{K}(I\otimes_{A}\mathcal{O}_{X})$ is onto. And thus, the $C^{*}$ -correspondence ${}_{\mathcal{O}_{IX}}(I\otimes_{A}\mathcal{O}_{X})_{\mathcal{O}_{X}}$ is a left-full Hilbert bimodule, which implies $\mathcal{O}_{IX}$ and $\langle I\otimes_{A}\mathcal{O}_{X},I\otimes_{A}\mathcal{O}_{X}\rangle_{% \mathcal{O}_{X}}=\langle\Upsilon(I)\rangle$ are Morita equivalent $C^{*}$ -algebras. Moreover, by Lemma 4.3 we have $\mathcal{O}_{IX}\cong\mathcal{K}(I\otimes_{A}\mathcal{O}_{X})\cong\mathcal{K}(% \Upsilon(I)\mathcal{O}_{X})\cong\Upsilon(I)\mathcal{O}_{X}\Upsilon(I)$ , which proves item (2). ∎

Remark 4.5.

Let ${}_{A}X_{A}$ be a $C^{*}$ -correspondence, and let $I$ be a positive $X$ -invariant ideal of $A$ . Let $p:A\rightarrow A/I$ and $q:X\rightarrow X/XI$ be the natural quotient maps. Then, $X/XI$ can be viewed as a $C^{*}$ -correspondence over $A/I$ with the module actions and the inner product are given by

p(a)\cdot q(x)\cdot q(a^{\prime})=q(axa^{\prime})\hskip 28.45274pt\langle q(x)% ,q(y)\rangle=p\left(\langle x,y\rangle_{A}\right),

for $a,a^{\prime}\in A$ , and $x,y\in X$ .

Assume ${}_{A}X_{A}$ is regular and $I$ is an $X$ -invariant ideal. Then $X/XI$ is a regular correspondence as well: properness of $X/XI$ is straightforward by construction. To see injectivity let $a\in A$ , and let $p(a)q(x)=0$ for all $x\in X$ . Then $ax\in XI$ for all $x\in X$ , which means $a\in X^{-1}(I)$ . Since $I$ is $X$ -invariant and ${}_{A}X_{A}$ is regular we have $X^{-1}(I)\subset I$ , and thus $p(a)=0$ .

Now, for a regular correspondence ${}_{A}X_{A}$ and an $X$ -invariant ideal $I$ , consider the isomorphisms

	$\displaystyle i_{1}:X\otimes_{A}A/I\rightarrow X/XI,\hskip 28.45274ptx\otimes_% {A}p(a)\mapsto q(xa)$
	$\displaystyle i_{2}:A/I{\otimes}_{A/I}X/XI\rightarrow X/XI,\hskip 28.45274ptp(% a)\otimes_{A/I}q(x)\mapsto q(ax),$

where $x\in X,a\in A.$ Then, [ $A/I,U_{A/I}$ ]: ${}_{A}X_{A}$ $\rightarrow{}_{A/I}(X/XI)_{A/I}$ is a morphism in $\mathsf{C^{*}cor_{pair}}$ , where $U_{A/I}:=i_{2}^{-1}\circ i_{1}$ .

Note that for any $x\in X,$ $a\in A$ , we have $U_{A/I}(x\otimes_{A}p(a))=p(a^{\prime})\otimes_{A/I}q(x^{\prime}),$ for some $a^{\prime}\in A,x^{\prime}\in X,$ satisfying $p(a^{\prime})q(x^{\prime})=q(x)p(a)$ : $i_{1}(x\otimes_{A}p(a))=q(xa)$ . Since $q(xa)$ is an element of the non-degenerate correspondence $X/XI$ , there exists $p(a^{\prime})\in A/I,q(x^{\prime})\in X/XI$ such that $p(a^{\prime})q(x^{\prime})=q(x)p(a)$ .

Theorem 4.6.

Let ${}_{A}X_{A}$ be a regular correspondence and let $I$ be an $X$ -invariant ideal. Then, we have the isomorphism $\mathcal{O}_{X}/\langle\Upsilon(I)\rangle\cong\mathcal{O}_{X/XI}$ .

Proof.

Let ( $\Upsilon,t$ ) and ( $\tilde{\Upsilon},\tilde{t}$ ) be universal covariant representations of $X$ and $X/XI$ , respectively. And, let $p:A\rightarrow A/I$ and $q:X\rightarrow X/XI$ be the quotient maps. The map $\xi:A/I\otimes_{A/I}\mathcal{O}_{X/XI}\rightarrow\mathcal{O}_{X/XI}$ defined on elementary tensors by $p(a)\otimes S\mapsto\tilde{\Upsilon}(p(a))S$ is a Hilbert- $\mathcal{O}_{X/XI}$ module isomorphism, and extends to a $C^{*}$ -algebra isomorphism $\mathop{Ad}\xi:\mathcal{K}(A/I\otimes_{A/I}\mathcal{O}_{X/XI})\rightarrow% \mathcal{K}(\mathcal{O}_{X/XI})$ . Let $L:\mathcal{O}_{X/XI}\rightarrow\mathcal{K}(\mathcal{O}_{X/XI})$ be the $C^{*}$ -algebra isomorphism defined by $L(S)T=ST$ for $S,T\in\mathcal{O}_{X/XI}$ . Denote the C-covariant representation of ${}_{A}X_{A}$ on $\mathcal{K}(A/I\otimes_{A/I}\mathcal{O}_{X/XI})$ by $(\pi,\Phi).$ Then we have the following diagram.

We claim $\mathop{Ad}\xi^{-1}\circ L$ is an isomorphism onto $C^{*}$ $(\pi,\Phi)$ . To prove our claim we first show $\tilde{t}(q(x))=L^{-1}\mathop{Ad}\xi(\Phi(x)),$ for $x\in X$ . Let $a\in A,S\in\mathcal{O}_{X/XI}.$ On one hand we have

L(\tilde{t}(q(x))\xi(p(a)\otimes_{A/I}S)=L(\tilde{t}(q(x))\tilde{\Upsilon}(p(a% ))S=\tilde{t}(q(x))\tilde{\Upsilon}(p(a))S=\tilde{t}(q(xa))S.

Now, let $V:X/XI\otimes_{A/I}\mathcal{O}_{X/XI}\rightarrow\mathcal{O}_{X/XI}$ denote the isomorphism defined on elementary tensors by $q(z)\otimes_{A/I}T\mapsto\tilde{t}(q(z))T$ , where $z\in X$ and $T\in\mathcal{O}_{X/XI}$ . We have

$\displaystyle\xi\Phi(x)(p(a)\otimes_{A/I}S)$	$\displaystyle=\xi(1_{A/I}\otimes V)(U_{A/I}(x\otimes_{A}p(a))\otimes_{A/I}S)$
	$\displaystyle=\xi(1_{A/I}\otimes V)(p(a^{\prime})\otimes_{A/I}q(x^{\prime})% \otimes_{A/I}S)$	where $q(x)p(a)=p(a^{\prime})q(x^{\prime})$ ¹¹1See the last paragraph of Remark 4.5.
	$\displaystyle=\xi\left[p(a^{\prime})\otimes_{A/I}\tilde{t}(q(x^{\prime}))S\right]$
	$\displaystyle=\tilde{t}(q(xa))S.$

It is easy to show $\tilde{\Upsilon}(p(a))=L^{-1}\mathop{Ad}\xi(\pi(a)),$ for any $a\in A$ , completing the proof of our claim. We may now conclude that $\sigma:\mathcal{O}_{X}\rightarrow\mathcal{K}(A/I\otimes_{A/I}\mathcal{O}_{X/XI})$ is surjective. Then by the first isomorphism theorem we have $\mathcal{O}_{X}/\operatorname{Ker}\sigma\cong\sigma(\mathcal{O}_{X})$ . By using Proposition 3.5 we obtain $\mathcal{O}_{X}/\langle\Upsilon(I)\rangle\cong\mathcal{K}(A/I\otimes_{A/I}% \mathcal{O}_{X/XI})\cong\mathcal{O}_{X/XI}$ .∎

We next give a factorization property in $\mathsf{C^{*}cor_{pair}}$ , which allows us to generalize the first item of Theorem 4.4. But first we need a Lemma.

Lemma 4.7.

For an $A$ – $B$ -correspondence isomorphism $U_{M}:X\otimes_{A}M\rightarrow M\otimes_{B}Y$ we have the following.

(1)

The ideal $B_{M}=\overline{\langle M,M\rangle_{B}}$ of $B$ is positive $Y$ -invariant.
(2)

$\operatorname{Ker}\varphi_{M}$ is a positive $X$ -invariant ideal of $A$ . If ${}_{A}X_{A}$ and ${}_{B}Y_{B}$ are regular correspondences, then $\operatorname{Ker}\varphi_{M}$ is an $X$ -invariant ideal.

Proof.

For the first item we compute

\langle Y,B_{M}\cdot Y\rangle_{B}=\langle M\otimes_{B}Y,M\otimes_{B}Y\rangle_{% B}=\langle X\otimes_{A}M,X\otimes_{A}M\rangle_{B}=\langle M,A_{X}\cdot M% \rangle_{B}\subset B_{M},

as desired. For the second item denote $\operatorname{Ker}\varphi_{M}$ by $K$ . We have

0=\langle K\cdot M\otimes_{B}Y,M\otimes_{B}Y\rangle_{B}=\langle K\cdot X% \otimes_{A}M,X\otimes_{A}M\rangle_{B}=\langle M,\langle KX,X\rangle_{A}\cdot M% \rangle_{B},

which implies $\langle KX,X\rangle_{A}\subset K$ , as desired. Now, assume ${}_{A}X_{A}$ and ${}_{B}Y_{B}$ are regular correspondences. Let $a\in X^{-1}(K)$ . Then, $\langle ax,x^{\prime}\rangle_{A}\in K$ for any $x,x^{\prime}\in X$ . This means

\langle ax\otimes_{A}m,x^{\prime}\otimes_{A}n\rangle_{B}=\langle m,\langle ax,% x^{\prime}\rangle_{A}\cdot n\rangle_{B}=0,

for any $x,x^{\prime}\in X$ and $m,n\in M$ . This implies $a\in\operatorname{Ker}\varphi_{X\otimes_{A}M}=\operatorname{Ker}\varphi_{M% \otimes_{B}Y}.$ Then for any $m,n\in M$ and $y,y^{\prime}\in Y$ , we have

0=\langle a\cdot m\otimes_{B}y,m^{\prime}\otimes_{B}y^{\prime}\rangle_{B}=% \langle y,\langle am,m^{\prime}\rangle_{B}\cdot y^{\prime}\rangle_{B},

which implies $\langle am,m^{\prime}\rangle_{B}\in\operatorname{Ker}\varphi_{Y}$ . Since ${}_{B}Y_{B}$ is a regular correspondence, we conclude that $a\in\operatorname{Ker}\varphi_{M}$ . ∎

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}}$ . The first item of Lemma 4.7 and Lemma 4.2 together imply that [ ${}_{B_{M}}(B_{M})_{B},U_{B_{M}}$ ]: ${}_{B_{M}}(B_{M}Y)_{B_{M}}\rightarrow$ ${}_{B}Y_{B}$ is a morphism in $\mathsf{C^{*}cor_{pair}}$ .

Proposition 4.8.

For any morphism [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ in $\mathsf{C^{*}cor_{pair}}$ , there exists a morphism $[{}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}]$ : ${}_{A}X_{A}$ $\rightarrow{{}_{B_{M}}(B_{M}Y)}_{B_{M}}$ such that the equality

[

{}_{A}M_{B},U_{M}

]=[

{}_{B_{M}}(B_{M})_{B},U_{B_{M}}

]

\circ

[

{}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}

]

holds.

Proof.

Let ${}_{A}M^{\prime}_{B_{M}}$ be the Hilbert $B$ -module $M$ viewed as $A$ – $B_{M}$ -correspondence. Consider the following $C^{*}$ -correspondence isomorphisms:

	$\displaystyle\iota:{}_{A}(M\otimes_{B}B_{M})_{B_{M}}\rightarrow{}_{A}M^{\prime% }_{B_{M}},$	$\displaystyle m\otimes_{B}b\mapsto m\cdot b$
	$\displaystyle l:{}_{A}(M^{\prime}\otimes_{B_{M}}B_{M})_{B}\rightarrow{}_{A}M_{% B},$	$\displaystyle m\otimes_{B_{M}}b\mapsto m\cdot b$
	$\displaystyle j:{}_{B_{M}}(B_{M}\otimes_{B}Y)_{B}\rightarrow{}_{B_{M}}(B_{M}Y)% _{B},$	$\displaystyle b\otimes_{B}y\mapsto b\cdot y$
	$\displaystyle k:{}_{B_{M}}(B_{M}Y\otimes_{B}B_{M})_{B_{M}}\rightarrow{}_{B_{M}% }(B_{M}Y)_{B_{M}},$	$\displaystyle\xi\otimes_{B}b\mapsto\xi\cdot b$

where $b\in B_{M},m\in M,y\in Y,$ and $\xi\in B_{M}Y.$ Let $U_{M^{\prime}}$ be the composition of the $A-B_{M}$ correspondence isomorphisms

To prove [ ${}_{B_{M}}(B_{M})_{B},U_{B_{M}}$ ] $\circ$ $[{}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}]=[{}_{A}M_{B},U_{M}]$ we show that the diagram

commutes. Take an elementary tensor $x\otimes_{A}m\in X\otimes_{A}M^{\prime}$ . By Cohen-Hewitt factorization theorem there exist $m^{\prime}\in M,b^{\prime}\in B_{M}$ such that $m=m^{\prime}\cdot b^{\prime}.$ Then we have

	$\displaystyle U_{M^{\prime}}(x\otimes_{A}m)=(1_{M^{\prime}}\otimes k)(1_{M^{% \prime}}\otimes j\otimes 1_{B_{M}})(l^{-1}\otimes 1_{Y}\otimes 1_{B_{M}})(U_{M% }\otimes 1_{B_{M}})(1_{X}\otimes\iota^{-1})(x\otimes_{A}m)$
	$\displaystyle=(1_{M}\otimes k)(1_{M}\otimes j\otimes 1_{B_{M}})(l^{-1}\otimes 1% _{Y}\otimes 1_{B_{M}})(U_{M}\otimes 1_{B_{M}})(x\otimes_{A}m^{\prime}\otimes_{% B}b^{\prime})$
	$\displaystyle=(1_{M}\otimes k)(1_{M}\otimes j\otimes 1_{B_{M}})(l^{-1}\otimes 1% _{Y}\otimes 1_{B_{M}})\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}m_{i}^{n}% \otimes_{B}y_{i}^{n}\otimes_{B}b^{\prime},$

where $\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}m_{i}^{n}\otimes_{B}y_{i}^{n}=U_{M}% (x\otimes_{A}m^{\prime}).$ Again by Cohen-Hewitt factorization theorem, there exist $\xi_{i}^{n}\in M,c_{i}^{n}\in B_{M}$ such that $m_{i}^{n}=\xi_{i}^{n}\cdot c_{i}^{n}$ . Then we may continue our computation as

	$\displaystyle=(1_{M}\otimes k)(1_{M}\otimes j\otimes 1_{B_{M}})\lim_{n% \rightarrow\infty}\sum_{i=1}^{N_{n}}\xi_{i}^{n}\otimes_{B_{M}}c_{i}^{n}\otimes% _{B}y_{i}^{n}\otimes_{B}b^{\prime}$
	$\displaystyle=(1_{M}\otimes k)\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}\xi_{% i}^{n}\otimes_{B_{M}}c_{i}^{n}\cdot y_{i}^{n}\otimes_{B}b^{\prime}$
	$\displaystyle=\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}\xi_{i}^{n}\otimes_{B% _{M}}c_{i}^{n}\cdot y_{i}^{n}\cdot b^{\prime}$

Then, for the elementary tensor $x\otimes_{A}m\otimes_{B}b$ of $X\otimes_{A}M\otimes_{B}B_{M}$ we have

	$\displaystyle(l\otimes 1_{Y})(1_{M}\otimes U_{B_{M}})(U_{M^{\prime}}\otimes 1_% {B_{M}})(x\otimes_{A}m\otimes_{B}b)$
	$\displaystyle=(l\otimes 1_{Y})(1_{M}\otimes U_{B_{M}})\lim_{n\rightarrow\infty% }\sum_{i=1}^{N_{n}}\xi_{i}^{n}\otimes_{B_{M}}c_{i}^{n}\cdot y_{i}^{n}\cdot b^{% \prime}\otimes_{B}b$
	$\displaystyle=\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}m_{i}^{n}\otimes_{B}y% _{i}^{n}\cdot b^{\prime}b$
	$\displaystyle=U_{M}(x\otimes_{A}m^{\prime})b^{\prime}b$
	$\displaystyle=U_{M}(1_{X}\otimes l)(x\otimes_{A}m\otimes_{B_{M}}b),$

as desired.∎

Corollary 4.9.

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}}$ . If ${}_{A}M_{B}$ is a left-full Hilbert bimodule, then so is the associated correspondence ${}_{\mathcal{O}_{X}}(M\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}$ .

Proof.

By Proposition 4.8 we have [ ${}_{B_{M}}(B_{M})_{B},U_{B_{M}}$ ] $\circ$ [ ${}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}$ ] $=$ [ ${}_{A}M_{B},U_{M}$ ], and thus

[{}_{\mathcal{O}_{X}}(M\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}]=[{}_{% \mathcal{O}_{B_{M}Y}}(B_{M}\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}]\circ% [{}_{\mathcal{O}_{X}}(M^{\prime}\otimes_{B_{M}}\mathcal{O}_{B_{M}Y})_{\mathcal% {O}_{B_{M}Y}}].

Since ${}_{A}M^{\prime}_{B_{M}}$ is an imprimitivity bimodule, [ ${}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}$ ] is an isomorphism in $\mathsf{C^{*}cor_{pair}}$ , and thus, $\mathcal{E}\left([{}_{A}M^{\prime}_{B_{M}},U_{M^{\prime}}]\right)=[{}_{% \mathcal{O}_{X}}(M^{\prime}\otimes_{B_{M}}\mathcal{O}_{B_{M}Y})_{\mathcal{O}_{% B_{M}Y}}]$ is an isomorphism in $\mathsf{C^{*}alg_{cor}}$ . This means ${}_{\mathcal{O}_{X}}(M^{\prime}\otimes_{B_{M}}\mathcal{O}_{B_{M}Y})_{\mathcal{% O}_{B_{M}Y}}$ is an imprimitivity bimodule. We also know by Theorem 4.4 that ${}_{\mathcal{O}_{B_{M}Y}}(B_{M}\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}$ is a left-full Hilbert bimodule. Hence, ${}_{\mathcal{O}_{X}}(M\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}$ is a left-full Hilbert bimodule. ∎

5. Exactness

We denote by $\mathsf{C^{*}cor_{pair}^{reg}}$ the subcategory of $\mathsf{C^{*}cor_{pair}}$ where all objects are regular $C^{*}$ -correspondences. Every morphism in $\mathsf{C^{*}cor_{pair}}$ has a kernel; however, not every morphism has a cokernel. We show in this section that every kernel in $\mathsf{C^{*}cor_{pair}^{reg}}$ has a cokernel. This observation leads us to study exactness in the subcategory $\mathsf{C^{*}cor_{pair}^{reg}}$ instead of $\mathsf{C^{*}cor_{pair}}$ .

To study kernels in $\mathsf{C^{*}cor_{pair}^{reg}}$ we need some understanding of monomorphisms in this category. Following Lemma is necessary for this purpose.

Lemma 5.1.

Let $\mu:M\otimes_{B}N\rightarrow M^{\prime}\otimes_{B}N$ be an $A$ – $C$ -correspondence isomorphism where $M$ and $M^{\prime}$ are $A$ – $B$ -correspondences, and ${}_{B}N_{C}$ is a left-full Hilbert bimodule. Then, there exists an isomorphism $\iota:$ ${}_{A}M_{B}$ $\rightarrow$ ${}_{A}M^{\prime}_{B}$ such that $\iota\otimes 1_{N}$ = $\mu$ .

Proof.

Since ${}_{B}N_{C}$ is a left-full Hilbert bimodule, there exists a $C$ – $B$ -correspondence $\tilde{N}$ and a $B$ – $B$ -correspondence isomorphism

j:N\otimes_{C}\tilde{N}\rightarrow B,\text{ }n_{1}\otimes_{C}\tilde{n_{2}}% \mapsto{}_{B}\langle n_{1},n_{2}\rangle,

where $n_{1},n_{2}\in N.$ Define an isomorphism $\iota:$ ${}_{A}M_{B}$ $\rightarrow$ ${}_{A}M^{\prime}_{B}$ by

\iota=\xi_{(r,M^{\prime},B)}\left(1_{M^{\prime}}\otimes j\right)\left(\mu% \otimes 1_{\tilde{N}}\right)\left(1_{M}\otimes j^{-1}\right)\left(\xi_{(r,M,B)% }\right)^{-1},

where $\xi_{(r,M,B)}$ and $\xi_{(r,M^{\prime},B)}$ are the $A$ – $B$ -correspondence isomorphisms defined as in Remark 3.2. It suffices to use elementary tensors to verify the equality $\iota\otimes 1_{N}$ = $\mu$ . Let $m^{\prime}\in M^{\prime}$ and $n_{1},n_{2},n_{3}\in N$ . Then we have

	$\displaystyle\left(\xi_{(r,M^{\prime},B)}\otimes 1_{N}\right)$	$\displaystyle\left(1_{M^{\prime}}\otimes j\otimes 1_{N}\right)(m^{\prime}% \otimes_{B}n_{1}\otimes_{C}\tilde{n_{2}}\otimes_{B}n_{3})$
		$\displaystyle=\left(\xi_{(r,M^{\prime},B)}\otimes 1_{N}\right)m^{\prime}% \otimes_{B}{}_{B}\langle n_{1},n_{2}\rangle\otimes_{B}n_{3}$
		$\displaystyle=m^{\prime}{}_{B}\langle n_{1},n_{2}\rangle\otimes_{B}n_{3}$
		$\displaystyle=m^{\prime}\otimes_{B}n_{1}\langle n_{2},n_{3}\rangle_{C}.$

This shows that for any $x\in M\otimes_{B}N,$ and $n,n^{\prime}\in N$ we have

\left(\xi_{(r,M^{\prime},B)}\otimes 1_{N}\right)\left(1_{M^{\prime}}\otimes j% \otimes 1_{N}\right)\left(\mu\otimes 1_{\tilde{N}}\otimes 1_{N}\right)(x% \otimes_{C}\tilde{n}\otimes_{B}n^{\prime})=\mu(x)\langle n,n^{\prime}\rangle_{C}

and

\mu\left(\xi_{(r,M,B)}\otimes 1_{N}\right)\left(1_{M}\otimes j\otimes 1_{N}% \right)(x\otimes_{C}\tilde{n}\otimes_{B}n^{\prime})=\mu(x)\langle n,n^{\prime}% \rangle_{C},

as desired.∎

Proposition 5.2.

Let [ ${}_{B}N_{C}$ , $U_{N}$ ]: ${}_{B}Y_{B}$ $\rightarrow$ ${}_{C}Z_{C}$ be a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ . If ${}_{B}N_{C}$ is a left-full Hilbert bimodule, then [ ${}_{B}N_{C}$ , $U_{N}$ ] is a monomorphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ .

Proof.

Let [ ${}_{A}M_{B}$ , $U_{M}$ ], [ ${}_{A}M^{\prime}_{B}$ , $U_{M^{\prime}}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be morphisms in $\mathsf{C^{*}cor_{pair}^{reg}}$ satisfying

\text{[${}_{B}N_{C}$, $U_{N}$] $\circ$ [${}_{A}M_{B}$, $U_{M}$] = [${}_{B}N_{C% }$, $U_{N}$] $\circ$ [${}_{A}M^{\prime}_{B}$, $U_{M^{\prime}}$]}.

Then, there exists an isomorphism $\mu:M\otimes_{B}N\rightarrow M^{\prime}\otimes_{B}N$ with the commutative diagram

Since ${}_{B}N_{C}$ is a left-full Hilbert bimodule, by Lemma 5.1, there exists an isomorphism $\iota:$ ${}_{A}M_{B}$ $\rightarrow$ ${}_{A}M^{\prime}_{B}$ such that $\mu=\iota\otimes 1_{N}$ . We aim to show that the diagram

commutes.

By the first diagram above, we have

	$\displaystyle(1_{M^{\prime}}\otimes U_{N})(U_{M^{\prime}}\otimes 1_{N})(1_{X}% \otimes\iota\otimes 1_{N})$	$\displaystyle=(\iota\otimes 1_{N}\otimes 1_{Z})(1_{M}\otimes U_{N})(U_{M}% \otimes 1_{N})$
		$\displaystyle=(1_{M^{\prime}}\otimes U_{N})(\iota\otimes 1_{Y}\otimes 1_{N})(U% _{M}\otimes 1_{N}),$

which implies the equality

(U_{M^{\prime}}\otimes 1_{N})(1_{X}\otimes\iota\otimes 1_{N})=(\iota\otimes 1_% {Y}\otimes 1_{N})(U_{M}\otimes 1_{N}).

Since ${}_{B}N_{C}$ is an injective correspondence, by Lemma 2.2, we have

U_{M^{\prime}}(1_{X}\otimes\iota)=(\iota\otimes 1_{Y})U_{M},

completing the proof. ∎

Remark 5.3.

Let $U_{M}:X\otimes_{A}M\rightarrow M\otimes_{B}Y$ be an $A$ – $B$ -correspondence isomorphism. We know by Lemma 4.7 that $\operatorname{Ker}\varphi_{M}$ is a positive $X$ -invariant ideal of $A$ . And thus we may view $KX$ as a $C^{*}$ -correspondence over $K$ , where $K$ denotes the ideal $\operatorname{Ker}\varphi_{M}.$ Then, as described in Lemma 4.2, $[{}_{K}K_{A},U_{K}]$ : ${}_{K}{KX}_{K}$ $\rightarrow$ ${}_{A}X_{A}$ is a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ where $U_{K}(kx\otimes_{K}k^{\prime})=k\otimes_{A}xk^{\prime}$ , for any $k,k^{\prime}\in K$ and $x\in X$ .

We are now ready to determine kernels in $\mathsf{C^{*}cor_{pair}^{reg}}$ .

Theorem 5.4.

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ . Let $K$ denote the kernel of the homomorphism $\varphi_{M}:A\rightarrow\mathcal{K}(M).$ Then, the object ${}_{K}(KX)_{K}$ paired with the morphism $[{}_{K}K_{A},U_{K}]:$ ${}_{K}{KX}_{K}$ $\rightarrow$ ${}_{A}X_{A}$ is a kernel of [ ${}_{A}M_{B}$ , $U_{M}$ ].

Proof.

We must show the following:

(1)

[ ${}_{A}M_{B}$ , $U_{M}$ ] $\circ$ [ ${}_{K}K_{A}$ , $U_{K}$ ] = [0, $0_{KX,Y}$ ]; and
(2)

assume [ ${}_{C}N_{A}$ , $U_{N}$ ]: ${}_{C}Z_{C}$ $\rightarrow$ ${}_{A}X_{A}$ is a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ satisfying the equality [ ${}_{A}M_{B}$ , $U_{M}$ ] $\circ$ [ ${}_{C}N_{A}$ , $U_{N}$ ] = [0, $0_{Z,Y}$ ] . Then, there exists a unique morphism [ ${}_{C}T_{K}$ , $U_{T}$ ]: ${}_{C}Z_{C}$ $\rightarrow$ ${}_{K}{KX}_{K}$ such that [ ${}_{K}K_{A}$ , $U_{K}$ ] $\circ$ [ ${}_{C}T_{K}$ , $U_{T}$ ]=[ ${}_{C}N_{A}$ , $U_{N}$ ].

Item (1) is folklore. For (2), notice that since $N\otimes_{A}M\cong 0$ we have $\langle N,N\rangle_{A}\subset K$ . Thus we may view ${}_{C}N_{A}$ as a $C$ – $K$ -correspondence, which we denote by $N^{\prime}$ . Moreover, we have the isomorphisms

\iota:{}_{C}N\otimes_{A}K_{K}\rightarrow{}_{C}N^{\prime}_{K}\hskip 56.9055ptn% \otimes_{A}k\mapsto n\cdot k

and

j:{}_{C}N^{\prime}\otimes_{K}K_{A}\rightarrow\textnormal{${}_{C}N_{A}$}\hskip 5% 6.9055ptn\otimes_{K}k\mapsto n\cdot k

for $n\in N$ , $k\in K$ . Now let $U_{N^{\prime}}$ be the $C$ – $K$ -correspondence isomorphism

where $\xi_{l}$ is the $K$ – $A$ -correspondence isomorphism $\xi_{(l,X,K)}:K\otimes_{A}X\rightarrow KX$ , and $\xi_{r}$ is the $K$ – $K$ -correspondence isomorphism $\xi_{(r,KX,K)}:KX\otimes_{A}K\rightarrow KX$ , i.e.,

U_{N^{\prime}}:=[1_{N^{\prime}}\otimes\xi_{r}][1_{N^{\prime}}\otimes\xi_{l}% \otimes 1_{K}][j^{-1}\otimes 1_{X}\otimes 1_{K}][U_{N}\otimes 1_{K}][1_{Z}% \otimes\iota^{-1}].

We show that [ ${}_{C}(N^{\prime}\otimes_{K}K)_{A}$ , $(1_{N^{\prime}}\otimes U_{K})(U_{N^{\prime}}\otimes 1_{K})$ ]=[ ${}_{C}N_{A}$ , $U_{N}$ ], i.e., the diagram

commutes. Consider an elementary tensor $n\otimes_{A}x\otimes_{A}k_{1}\otimes_{K}k_{2}$ of ${}_{A}(N\otimes_{A}X\otimes_{A}K\otimes_{K}K)_{A}$ . By Cohen-Hewitt factorization theorem we have $n=n^{\prime}\cdot k^{\prime}$ for some $n^{\prime}\in N$ and $k^{\prime}\in\langle N,N\rangle_{A}\subset K$ . Then,

	$\displaystyle(j\otimes 1_{X})(1_{N^{\prime}}\otimes U_{K})$	$\displaystyle(1_{N^{\prime}}\otimes\xi_{r}\otimes 1_{K})(1_{N^{\prime}}\otimes% \xi_{l}\otimes 1_{K}\otimes 1_{K})(j^{-1}\otimes 1_{X}\otimes 1_{K}\otimes 1_{% K})(n\otimes_{A}x\otimes_{A}k_{1}\otimes_{K}k_{2})$
		$\displaystyle=(j\otimes 1_{X})(1_{N^{\prime}}\otimes U_{K})(1_{N^{\prime}}% \otimes\xi_{r}\otimes 1_{K})(n^{\prime}\otimes_{K}k^{\prime}x\otimes_{A}k_{1}% \otimes_{K}k_{2})$
		$\displaystyle=(j\otimes 1_{X})(1_{N^{\prime}}\otimes U_{K})(n^{\prime}\otimes_% {K}k^{\prime}xk_{1}\otimes_{K}k_{2})$
		$\displaystyle=(j\otimes 1_{X})(n^{\prime}\otimes_{K}k^{\prime}\otimes_{A}xk_{1% }k_{2})$
		$\displaystyle=n\otimes_{A}xk_{1}k_{2}.$

On the other hand, it is not hard to seee that

U_{N}(1_{Z}\otimes j)(1_{Z}\otimes\iota\otimes 1_{K})(U_{N}^{-1}\otimes 1_{K}% \otimes 1_{K})(n\otimes_{A}x\otimes_{A}k_{1}\otimes_{K}k_{2})=n\otimes_{A}xk_{% 1}k_{2}.

Uniqueness of the morphism [ ${}_{C}N^{\prime}_{K}$ , $U_{N^{\prime}}$ ] follows from Proposition 5.2, since ${}_{K}K_{A}$ is a left-full Hilbert bimodule. ∎

We next study cokernels in $\mathsf{C^{*}cor_{pair}^{reg}}$ .

Lemma 5.5.

Let ${}_{C}X_{B}$ and ${}_{C}Y_{B}$ be $C^{*}$ -correspondences. Let ${}_{A}C_{C}$ be the $C^{*}$ -correspondence where the left action is determined by the surjective map $\pi:A\rightarrow C$ . If there exists an $A$ – $B$ -correspondence isomorphism $U:C\otimes_{C}X\rightarrow C\otimes_{C}Y$ , then there exists an isomorphism $V:{}_{C}X_{B}\rightarrow{}_{C}Y_{B}$ such that $1_{C}\otimes V=U$ .

Proof.

Consider the natural $A$ – $B$ -correspondence isomorphisms

	$\displaystyle\iota_{C,X}:C\otimes_{C}X\rightarrow X\text{}$	$\displaystyle c\otimes_{C}x\mapsto c\cdot x$
	$\displaystyle\iota_{C,Y}:C\otimes_{C}Y\rightarrow Y\text{}$	$\displaystyle c\otimes_{C}y\mapsto c\cdot y$

[4, Lemma 3.3] tells us that the map $\iota_{C,Y}\circ U\circ\iota_{C,X}^{-1}:{}_{A}X_{B}\rightarrow{}_{A}Y_{B}$ preserves the left $C$ -module structure and thus, provides an isomorphism ${}_{C}X_{B}\rightarrow{}_{C}Y_{B}$ . We observe that $1_{C}\otimes\iota_{C,Y}U\iota_{C,X}^{-1}=U$ : let $c,c^{\prime}\in C,x\in X.$ Notice that since $U(c^{\prime}\otimes_{C}x)=\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}c_{i}^{n}% \otimes_{C}y_{i}^{n}$ for $c_{i}^{n}\in C,y_{i}^{n}\in Y$ , we have

	$\displaystyle(1_{C}\otimes\iota_{C,Y})(1_{C}\otimes U)(c\otimes_{C}c^{\prime}% \otimes_{C}x)$	$\displaystyle=c\otimes_{C}\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}c_{i}^{n}% \cdot y_{i}^{n}$
		$\displaystyle=\lim_{n\rightarrow\infty}\sum_{i=1}^{N_{n}}cc_{i}^{n}\otimes_{C}% y_{i}^{n}$
		$\displaystyle=U(1_{C}\otimes\iota_{C,X})(c\otimes_{C}c^{\prime}\otimes_{C}x),$

as desired. ∎

Proposition 5.6.

Let ${}_{A}X_{A}$ be a regular $C^{*}$ -correspondence and let $I$ be an $X$ -invariant ideal of $A$ . Then, $[{}_{A}(A/I)_{A/I},U_{A/I}]$ : ${}_{A}X_{A}$ $\rightarrow{}_{A/I}(X/XI)_{A/I}$ is an epimorphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ .

Proof.

Assume there exist morphisms [ ${}_{A/I}M_{B}$ , $U_{M}$ ], [ ${}_{A/I}N_{B}$ , $U_{N}$ ]: ${}_{A/I}(X/XI)_{A/I}\rightarrow$ ${}_{B}Y_{B}$ in $\mathsf{C^{*}cor_{pair}}$ such that

\text{[${}_{A/I}M_{B}$, $U_{M}$]$\circ$ [${}_{A}(A/I)_{A/I}$, $U_{A/I}$] = [${% }_{A/I}N_{B}$, $U_{N}$]$\circ$ [${}_{A}(A/I)_{A/I}$, $U_{A/I}$]}.

Then, there exists an $A$ – $B$ -correspondence isomorphism

\mu:A/I\otimes_{A/I}M\rightarrow A/I\otimes_{A/I}N

making the diagram

commute.

Since the $C^{*}$ -correspondence ${}_{A}(A/I)_{A/I}$ comes from the surjective homomorphism $A\rightarrow A/I$ , by Lemma 5.5, there exists an isomorphism $\xi:$ ${}_{A/I}M_{B}$ $\rightarrow$ ${}_{A/I}N_{B}$ such that $\mu=1_{A/I}\otimes\xi$ . Then, by the diagram above, we have

	$\displaystyle(1_{A/I}\otimes\xi\otimes 1_{Y})(1_{A/I}\otimes U_{M})(U_{A/I}% \otimes 1_{M})$	$\displaystyle=(1_{A/I}\otimes U_{N})(U_{A/I}\otimes 1_{N})(1_{X}\otimes 1_{A/I% }\otimes\xi)$
		$\displaystyle=(1_{A/I}\otimes U_{N})(1_{A/I}\otimes 1_{X/XI}\otimes\xi)(U_{A/I% }\otimes 1_{M}),$

which means $1_{A/I}\otimes(\xi\otimes 1_{Y})U_{M}=1_{A/I}\otimes U_{N}(1_{X/XI}\otimes\xi)$ . Since [ ${}_{A}(A/I)_{A/I}$ ] is an epimorphism in $\mathsf{C^{*}alg_{cor}}$ , we may now conclude the equality $(\xi\otimes 1_{Y})U_{M}=U_{N}(1_{X/XI}\otimes\xi)$ , which implies [ ${}_{A/I}M_{B}$ , $U_{M}$ ]=[ ${}_{A/I}N_{B}$ , $U_{N}$ ]. ∎

Theorem 5.7.

Let [ ${}_{A}M_{B}$ , $U_{M}$ ]: ${}_{A}X_{A}$ $\rightarrow$ ${}_{B}Y_{B}$ be a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ . If $B_{M}$ is a $Y$ -invariant ideal of $B$ , then a cokernel of [ ${}_{A}M_{B}$ , $U_{M}$ ] is $[{}_{B}(B/B_{M})_{B/B_{M}},U_{B/B_{M}}]$ : ${}_{B}Y_{B}$ $\rightarrow$ ${}_{B/B_{M}}(Y/YB_{M})_{B/B_{M}}$ .

Proof.

We must show the following:

(1)

$[{}_{B}(B/B_{M})_{B/B_{M}},U_{B/B_{M}}]$ $\circ$ [ ${}_{A}M_{B}$ , $U_{M}$ ] = [0, $0_{X,Y/YB_{M}}$ ]; and
(2)

assume [ ${}_{B}N_{C}$ , $U_{N}$ ]: ${}_{B}Y_{B}$ $\rightarrow$ ${}_{C}Z_{C}$ is a morphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ satisfying the equality [ ${}_{B}N_{C}$ , $U_{N}$ ] $\circ$ [ ${}_{A}M_{B}$ , $U_{M}$ ] = [0, $0_{X,Z}$ ] . Then, there exists a unique morphism [ ${}_{B/B_{M}}T_{C}$ , $U_{T}$ ]: ${}_{B/B_{M}}(Y/YB_{M})_{B/B_{M}}$ $\rightarrow$ ${}_{C}Z_{C}$ satisfying the equality [ ${}_{B}N_{C}$ , $U_{N}$ ] $=$ [ ${}_{B/B_{M}}T_{C}$ , $U_{T}$ ] $\circ$ [ ${}_{B}(B/B_{M})_{B/B_{M}}$ , $U_{B/B_{M}}$ ].

The first item is easy to verify. Let [ ${}_{B}N_{C}$ , $U_{N}$ ]: ${}_{B}Y_{B}$ $\rightarrow$ ${}_{C}Z_{C}$ be a morphism described as in the second item. Then ${}_{B}N_{C}$ can be viewed as a $B/B_{M}$ – $C$ -correspondence [4, Lemma 3.3], which we denote by $N^{\prime}$ . Now, let $\xi$ be the $C^{*}$ -correspondence isomorphism ${}_{B}(B/B_{M}\otimes_{B/B_{M}}N^{\prime})_{C}\rightarrow$ ${}_{B}N_{C}$ ; and consider the $B$ – $C$ -correspondence isomorphism

(\xi^{-1}\otimes 1_{Z})U_{N}(1_{Y}\otimes\xi)(U_{B/B_{M}}^{-1}\otimes 1_{N^{% \prime}}):B/B_{M}\otimes_{B/B_{M}}Y/YB_{M}\otimes_{B/B_{M}}N^{\prime}% \rightarrow B/B_{M}\otimes_{B/B_{M}}N^{\prime}\otimes_{C}Z.

By Lemma 5.5 there exists an isomorphism

U_{N^{\prime}}:{}_{B/B_{M}}(Y/YB_{M}\otimes_{B/B_{M}}N^{\prime})_{C}% \rightarrow{}_{B/B_{M}}(N^{\prime}\otimes_{C}Z)_{C}

such that $1_{B/B_{M}}\otimes U_{N^{\prime}}=(\xi^{-1}\otimes 1_{Z})U_{N}(1_{Y}\otimes\xi% )(U_{B/B_{M}}^{-1}\otimes 1_{N^{\prime}})$ . One can now see that the diagram

commutes. The uniqueness of [ ${}_{B/B_{M}}N^{\prime}_{C}$ , $U_{N^{\prime}}$ ] follows from Proposition 5.6.∎

Let [ ${}_{A}X_{B}$ ] be a morphism in $\mathsf{C^{*}alg_{cor}}$ . A kernel of [ ${}_{A}X_{B}$ ] is the pair $(K,[{}_{K}K_{A}])$ , where $K$ denotes the kernel of $\varphi_{X}:A\rightarrow\mathcal{L}(X)$ [4, Theorem 3.11]. A cokernel of [ ${}_{A}X_{B}$ ] is the pair $(B/B_{X},[{}_{B}(B/B_{X})_{B/B_{X}}])$ [4, Corollary 3.12].

Definition 5.8.

A sequence $0\rightarrow A\xrightarrow{[{}_{A}X_{B}]}B\xrightarrow{[{}_{B}Y_{C}]}C\rightarrow 0$ in $\mathsf{C^{*}alg_{cor}}$ is exact if the pair $(A,[{}_{A}X_{B}])$ is a kernel of $[{}_{B}Y_{C}]$ and the pair $(B,[{}_{B}Y_{C}])$ is a cokernel of $[{}_{A}X_{B}]$ .

Proposition 5.9.

A sequence $0\rightarrow A\xrightarrow{[{}_{A}X_{B}]}B\xrightarrow{[{}_{B}Y_{C}]}C\rightarrow 0$ in $\mathsf{C^{*}alg_{cor}}$ is exact if and only if the following three holds.

(1)

${}_{A}X_{B}$ is a left-full Hilbert bimodule;
(2)

$B_{X}=K$ , where $K$ denotes the kernel of $\varphi_{Y}:B\rightarrow\mathcal{L}(Y)$ .
(3)

Hilbert $C$ -module $Y$ viewed as a $B/K$ – $C$ -correspondence ${}_{B/K}Y^{\prime}_{C}$ is an imprimitivity bimodule.

Proof.

Assume we have (1)-(3). In $\mathsf{C^{*}alg_{cor}}$ , we know that kernel of [ ${}_{B}Y_{C}$ ] is the pair $(K,[{}_{K}K_{B}])$ , where $K$ denotes the kernel of $\varphi_{Y}:B\rightarrow\mathcal{L}(Y)$ . On the other hand, item (2) implies that [ ${}_{B}Y_{C}$ ] $\circ$ [ ${}_{A}X_{B}$ ]=[ ${}_{A}0_{C}$ ]. Then, by the universal property of kernels there exists a morphism from $A$ to $K$ which [ ${}_{A}X_{B}$ ] factors through. As shown in [4, Theorem 3.9] this unique morphism is [ ${}_{A}X^{\prime}_{K}$ ] where $X$ is just $X^{\prime}$ viewed as an $A$ – $K$ -correspondence. Since ${}_{A}X^{\prime}_{K}$ is an imprimitivity bimodule we have that [ ${}_{A}X^{\prime}_{K}$ ] is an isomorphism in $\mathsf{C^{*}alg_{cor}}$ . It remains to show that [ ${}_{B}Y_{C}$ ] is a cokernel of [ ${}_{A}X_{B}$ ]. We know that a cokernel of [ ${}_{A}X_{B}$ ] is [ ${}_{B}B/K_{B/K}$ ], and since [ ${}_{B}Y_{C}$ ] $\circ$ [ ${}_{A}X_{B}$ ]=[ ${}_{A}0_{C}$ ], by the universal property of cokernels there exists a unique morphism which [ ${}_{B}Y_{C}$ ] factors through. As shown in [4, Proposition 3.11], this unique morphism is [ ${}_{B/K}Y^{\prime}_{C}$ ], which is an isomorphism in $\mathsf{C^{*}alg_{cor}}$ by item (3).

For the other direction, assume [ ${}_{A}X_{B}$ ] is a kernel of [ ${}_{B}Y_{C}$ ] and [ ${}_{B}Y_{C}$ ] is a cokernel of [ ${}_{A}X_{B}$ ]. Since [ ${}_{A}X_{B}$ ] is a kernel of [ ${}_{B}Y_{C}$ ], the correspondence ${}_{A}X_{K}$ in the kernel factorization ${}_{A}X_{B}\cong{}_{A}X\otimes_{K}K_{B}$ must be an imprimitivity bimodule, which means ${}_{A}X_{B}$ is a left-full Hilbert bimodule, giving us item (1). Moreover, ${}_{A}X_{K}$ being an imprimitivity bimodule implies that $B_{X}=K$ , which proves item (2). Since [ ${}_{B}Y_{C}$ ] is a cokernel of [ ${}_{A}X_{B}$ ], the correspondence ${}_{B/K}Y^{\prime}_{C}$ in the cokernel factorization ${}_{B}B/K\otimes_{B/K}Y^{\prime}_{C}\cong{}_{B}Y_{C}$ must be an imprimitivity bimodule, concluding the proof. ∎

Definition 5.10.

A sequence

0\rightarrow{}_{A}X_{A}\xrightarrow{[{}_{A}M_{B},U_{M}]}{}_{B}Y_{B}% \xrightarrow{[{}_{B}N_{C},U_{N}]}{}_{C}Z_{C}\rightarrow 0

in $\mathsf{C^{*}cor_{pair}^{reg}}$ is called exact if the pair $\left({}_{A}X_{A},[{}_{A}M_{B},U_{M}]\right)$ is a kernel of the morphism $[{}_{B}N_{C},U_{N}]:{}_{B}Y_{B}\rightarrow{}_{C}Z_{C}$ ; and the pair $\left({}_{B}Y_{B},[{}_{B}N_{C},U_{N}]\right)$ is a cokernel of the morphism $[{}_{A}M_{B},U_{M}]:{}_{A}X_{A}\rightarrow{}_{B}Y_{B}.$

Note that $[{}_{A}M_{B},U_{M}]$ being a kernel of $[{}_{B}N_{C},U_{N}]$ in the sequence above implies that $B_{M}=\operatorname{Ker}\varphi_{N}$ . Then, Lemma 4.7 allows us to conclude that $B_{M}$ is a $Y$ -invariant ideal of $B$ , and thus, cokernel of $[{}_{A}M_{B},U_{M}]$ exists.

Corollary 5.11.

A sequence

0\rightarrow{}_{A}X_{A}\xrightarrow{[{}_{A}M_{B},U_{M}]}{}_{B}Y_{B}% \xrightarrow{[{}_{B}N_{C},U_{N}]}{}_{C}Z_{C}\rightarrow 0

is exact in $\mathsf{C^{*}cor_{pair}^{reg}}$ if and only if the following holds.

(1)

${}_{A}M_{B}$ is a left-full Hilbert bimodule;
(2)

$B_{M}=K$ , where $K$ denotes the kernel of $\varphi_{N}:B\rightarrow\mathcal{L}(N)$ ;
(3)

Hilbert $C$ -module $N$ viewed as a $B/K$ – $C$ -correspondence ${}_{B/K}N^{\prime}_{C}$ is an imprimitivity bimodule.

We omit the proof of Corollary 5.11 since it can be shown by following the proof of Proposition 5.9.

Theorem 5.12.

The restriction of the functor $\mathcal{E}$ to the category $\mathsf{C^{*}cor_{pair}^{reg}}$ is exact.

Proof.

Let the sequence

0\rightarrow{}_{A}X_{A}\xrightarrow{[{}_{A}M_{B},U_{M}]}{}_{B}Y_{B}% \xrightarrow{[{}_{B}N_{C},U_{N}]}{}_{C}Z_{C}\rightarrow 0

in $\mathsf{C^{*}cor_{pair}^{reg}}$ be exact. Denote $\operatorname{Ker}\varphi_{N}$ by $K$ . We know that ${}_{A}M_{B}$ is a left-full Hilbert bimodule, the correspondence ${}_{B/K}N^{\prime}_{C}$ is an imprimitivity bimodule, and we have the equality $B_{M}=K$ . The functor $\mathcal{E}$ maps this sequence to

0\rightarrow\mathcal{O}_{X}\xrightarrow{[{}_{\mathcal{O}_{X}}(M\otimes_{B}% \mathcal{O}_{Y})_{\mathcal{O}_{Y}}]}\mathcal{O}_{Y}\xrightarrow{[{}_{\mathcal{% O}_{Y}}(N\otimes_{C}\mathcal{O}_{Z})_{\mathcal{O}_{Z}}]}\mathcal{O}_{Z}% \rightarrow 0.

By Corollary 4.9, we have that ${}_{\mathcal{O}_{X}}(M\otimes_{B}\mathcal{O}_{Y})_{\mathcal{O}_{Y}}$ is a left-full Hilbert bimodule. Since [ ${}_{B/K}N^{\prime}_{C},U_{N^{\prime}}$ ]: ${}_{B/K}Y/YK_{B/K}\rightarrow{}_{C}Z_{C}$ is an isomorphism in $\mathsf{C^{*}cor_{pair}^{reg}}$ , we have that $\mathcal{E}([{}_{B/K}N^{\prime}_{C},U_{N^{\prime}}])=[{}_{\mathcal{O}_{Y/YK}}(% N^{\prime}\otimes_{C}\mathcal{O}_{Z})_{\mathcal{O}_{Z}}]$ is an isomorphism in $\mathsf{C^{*}alg_{cor}}$ , and thus ${}_{\mathcal{O}_{Y/YK}}(N^{\prime}\otimes_{C}\mathcal{O}_{Z})_{\mathcal{O}_{Z}}$ is an imprimitivity bimodule. It remains to prove that $\langle M\otimes_{B}\mathcal{O}_{Y},M\otimes_{B}\mathcal{O}_{Y}\rangle_{% \mathcal{O}_{Y}}=\operatorname{Ker}\sigma$ , where $\sigma:\mathcal{O}_{Y}\rightarrow\mathcal{K}(N\otimes_{C}\mathcal{O}_{Z})$ is the left action homomorphism associated to the correspondence ${}_{\mathcal{O}_{Y}}(N\otimes_{C}\mathcal{O}_{Z})_{\mathcal{O}_{Z}}.$ Let $(\Upsilon,t)$ denote the universal covariant representation of ${}_{B}Y_{B}$ . Then, by Proposition 3.5, we have

\langle M\otimes_{B}\mathcal{O}_{Y},M\otimes_{B}\mathcal{O}_{Y}\rangle_{% \mathcal{O}_{Y}}=\langle\mathcal{O}_{Y},K\cdot\mathcal{O}_{Y}\rangle=\langle% \Upsilon(K)\rangle=\operatorname{Ker}\sigma,

as desired. ∎

Example 5.13.

By using Theorem 5.12 we can easily see Theorem 4.6 and Theorem 4.4 for the case when ${}_{A}X_{A}$ is a regular correspondence: let $I$ be an $X$ -invariant ideal of $A$ . Then, the sequence

0\rightarrow{}_{I}IX_{I}\xrightarrow{[{}_{I}I_{A},U_{I}]}{}_{A}X_{A}% \xrightarrow{[{}_{A}A/I_{A/I},U_{A/I}]}{}_{A/I}X/XI_{A/I}\rightarrow 0

is exact in $\mathsf{C^{*}cor_{pair}^{reg}}$ . And thus, the sequence

0\rightarrow{}_{\mathcal{O}_{IX}}\xrightarrow{[I\otimes_{A}\mathcal{O}_{X}]}% \mathcal{O}_{X}\xrightarrow{[A/I\otimes_{A/I}\mathcal{O}_{X/XI}]}\mathcal{O}_{% X/XI}\rightarrow 0

is exact in $\mathsf{C^{*}alg_{cor}}$ . This implies by Proposition 5.9 that ${}_{\mathcal{O}_{IX}}(I\otimes_{A}\mathcal{O}_{X})_{\mathcal{O}_{X}}$ is a left-full Hilbert bimodule, which means ${}_{\mathcal{O}_{IX}}(I\otimes_{A}\mathcal{O}_{X})_{\langle\Upsilon(I)\rangle}$ is an imprimitivity bimodule, where ( $\Upsilon,t$ ) is the universal covariant representation of ${}_{A}X_{A}$ . Consequently, we have $\mathcal{O}_{IX}\cong\mathcal{K}(I\otimes_{A}\mathcal{O}_{X})\cong\Upsilon(I)% \mathcal{O}_{X}\Upsilon(I)$ . On the other hand, again by Proposition 5.9, we know that ${}_{\mathcal{O}_{X}/\langle\Upsilon(I)\rangle}(A/I\otimes_{A/I}\mathcal{O}_{X/% XI})_{\mathcal{O}_{X/XI}}$ is an imprimitivity bimodule. This allows us to conclude the isomorphism $\mathcal{O}_{X}/\langle\Upsilon(I)\rangle\cong\mathcal{K}(A/I\otimes_{A/I}% \mathcal{O}_{X/XI})\cong\mathcal{O}_{X/XI}$ .

References

[1] S. Echterhoff, S. Kaliszewski, J. Quigg and I. Raeburn, A categorical approach to imprimitivity theorems for $C^{*}$ -dynamical systems, Mem. Amer. Math. Soc. 180 (2006), no. 850, viii+169 pp.
[2] S. Echterhoff, S. Kaliszewski, J. Quigg and I. Raeburn, Naturality and induced representations, Bull. Austral. Math. Soc. 61 (2000), 415–438.
[3] M. Eryüzlü, Passing $C^{*}$ -correspondence relations to the Cuntz-Pimsner algebras, Münster J. of Math. 15 (2022), 441-471.
[4] M. Eryüzlü, S. Kaliszewski and J. Quigg, Exact sequences in the enchilada category, Theory Appl. Categ. 35 (2020), 350–370.
[5] N. J. Fowler, P. S. Muhly and I. Raeburn, Representations of Cuntz-Pimsner algebras, Indiana Univ. Math. J. 52 (2003), no. 3, 569–605.
[6] T. Katsura, Ideal structure of $C^{*}$ -algebras associated with $C^{*}$ -correspondences, Pacific J. Math. 230 (2007), 107–145.
[7] P. Muhly and M. Tomforde, Adding tails to $C^{*}$ -correspondences, Doc. Math. 9 (2004), 79–106. .