and the Covariance Algebra

## The Crossed Product by a Partial Endomorphism

### November 27, 2006

Abstract
Given a local homeomorphism $\sigma :U\to X$  where $U\subseteq X$  is clopen and $X$  is a compact and Hausdorff topological space, we obtain the possible transfer operators ${L}_{\rho }$  which may occur for $\alpha :C\left(X\right)\to C\left(U\right)$  given by $\alpha \left(f\right)=f\circ \sigma$  . We obtain examples of partial dynamical systems $\left({X}_{A},{\sigma }_{A}\right)$  such that the construction of the covariance algebra ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)$  and the crossed product by partial endomorphism $\mathcal{O}\left({X}_{A},\alpha ,L\right)$  associated to this system are not equivalent, in the sense that there does not exists invertible function $\rho \in C\left(U\right)$  such that $\mathcal{O}\left({X}_{A},\alpha ,L\rho \right)={C}^{*}\left({X}_{A},\sigma \right)$  .

1 Introduction

We start with a summary of the construction of the crossed product by a partial endomorphism.
Details may be seen in [3. A partial ${C}^{*}$  -dynamical system $\left(A,\alpha ,L\right)$  consists of a (closed) ideal $I$  of a ${C}^{*}$  -algebra $A$  , a idempotent self-adjoint ideal $J$  of $I$  (not necessarily closed), a *-homomorphism $\alpha :A\to M\left(I\right)$  and a linear posisitve map (which preserves *) $L:J\to A$  such that $L\left(a\alpha \left(b\right)\right)=L\left(a\right)b$  for each $a\in J$  and $b\in A$  . The map $L$  is called transfer operator. Define in $J$  an inner product (which may be degenerated) by $\left(x,y\right)=L\left({x}^{*}y\right)$  . Then we obtain a inner product $〈,〉$  in the quotient ${J}_{0}=J/\left\{x\in J:L\left({x}^{*}x\right)=0\right\}$  defined by $〈\stackrel{~}{x},\stackrel{~}{y}〉=L\left({x}^{*}y\right)$  , which induces a norm $\parallel \parallel$  . Define $M={\overline{{J}_{0}}}^{\parallel \parallel }$  , which is a right Hilbert $A$  -module and also a left $A$  -module, where the left multiplication is defined by the *-homomorphism $\phi :A\to L\left(M\right)$  (the adjointable operators in $M$  ), where $\phi \left(a\right)\left(\stackrel{~}{x}\right)=\stackrel{~}{ax}$  for each $x\in J$  . The toeplitz algebra associated to $\left(A,\alpha ,L\right)$  is the universal ${C}^{*}$  -algebra $\mathcal{T}\left(A,\alpha ,L\right)$  generated by $A\cup M$  with the relations of $A$  , of $M$  , the bi-module products and ${m}^{*}n=〈m,n〉$  .
A redundancy in $\mathcal{T}\left(A,\alpha ,LL\right)$  is a pair $\left(a,k\right)\in A×\stackrel{^}{{K}_{1}}$  , ( $\stackrel{^}{{K}_{1}}=\overline{\text{span}}\left\{m{n}^{*},m,n\in M\right\}$  ), such that $am=km$  for every $m\in M$  . The Crossed Product by a Partial Endomorphism $\mathcal{O}\left(A,\alpha ,L\right)$  is the quotient of $\mathcal{T}\left(A,\alpha ,L\right)$  by the ideal generated by all the elements $a-k$  where $\left(a,k\right)$  is a redundancy and $a\in ker\left(\phi {\right)}^{-1}\cap {\phi }^{-1}\left(K\left(M\right)\right)$  .
In [3it was defined the algebra $\mathcal{O}\left(X,\alpha ,L\right)$  . This algebra is constructed from a partial ${C}^{*}$  -dynamical system $\left(C\left(X\right),\alpha ,L\right)$  induced by a local homeomorphism $\sigma :U\to X$  , where $U$  is an open subset of a compact topological Hausdorff space $X$  . More specifically,
 $\begin{array}{cccc}\alpha :& C\left(X\right)& \to & {C}^{b}\left(U\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\circ \sigma \end{array}$
where ${C}^{b}\left(U\right)$  is the space of all continuous bounded functions in $U$  and $L:{C}_{c}\left(U\right)\to C\left(X\right)$  ( ${C}_{c}\left(U\right)$  is the set of the continuous functions with compact support in $U$  ) is defined by $L\left(f\right)\left(x\right)=\left\{\begin{array}{cc}{\sum }_{y\in {\sigma }^{-1}\left(x\right)}f\left(y\right)& \text{if}x\in \sigma \left(U\right)\\ 0& \text{otherwise}\end{array}$  for every $f\in {C}_{c}\left(U\right)$  and $x\in X$  .
In [4it was defined the algebra ${C}^{*}\left(X,\alpha \right)$  , called covariance algebra. This algebra is also constructedfrom a partial dynamical system, that is, a continuous map $\sigma :U\to X$  where $X$  is a topologial compact Hausdorff space , $U$  is a clopen subset of $X$  and $\sigma \left(U\right)$  is open.
If we suppose that $\sigma :U\to X$  is an local homeomorphism, $U$  clopen (and so $\sigma \left(U\right)$  is always open) then $\left(X,\sigma \right)$  gives rise to two ${C}^{*}$  -algebras, the covariance algebra ${C}^{*}\left(X,\sigma \right)$  and the crossed product by a partial endomorphism $\mathcal{O}\left(X,\alpha ,L\right)$  .
In this paper we identify the transfer operators ${L}_{\rho }$  which may occur for $\alpha$  . Moreover we show that the constructions of covariance algebra and crossed product by partial endomoprhism are not equivalent, in the following sense: we obtain examples of partial dynamical systems $\left({X}_{A},{\sigma }_{A}\right)$  such that there does not exists invertible function $\rho$  such that $\mathcal{O}\left({X}_{A},\alpha ,L\rho \right)={C}^{*}\left(X,\alpha \right)$  .
Given ${C}^{*}$  -algebras $A$  and $B$  , if we write $A=B$  , we will say that $A$  and $B$  are *-isomorphic.
Acknowledgements The author wishes to express his thanks to R. Exel for the many stimulating conversations.

2 Transfer operators of $X$  for $\alpha$

Let $\sigma :U\to X$  be a local homeomorphism and $U$  an open subset of the compact Hausdorff space $X$  . This local homeomorphism induces the *-homomorphism
 $\begin{array}{cccc}\alpha :& C\left(X\right)& \to & {C}^{b}\left(U\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\circ \sigma \end{array}$
. Given a positive function $\rho \in C\left(U\right)$  , for all $f\in {C}_{c}\left(U\right)$  we may define ${L}_{\rho }\left(f\right)\left(x\right)=\left\{\begin{array}{cc}{\sum }_{y\in {\sigma }^{-1}\left(x\right)}\rho \left(y\right)f\left(y\right)& \text{if}x\in \sigma \left(U\right)\\ 0& \text{otherwise}\end{array}$  for each $x\in X$  . Note that ${L}_{\rho }\left(f\right)=L\left(\rho f\right)$  , and since $\rho f\in {C}_{c}\left(U\right)$  and $L\left(\rho f\right)\in C\left(X\right)$  (see [3) then ${L}_{\rho }\left(f\right)$  in fact is an element of $C\left(X\right)$  . In this way we may define the map ${L}_{\rho }:{C}_{c}\left(U\right)\to C\left(X\right)$  , which is linear and positive (by the fact that $\rho$  is positive). It is easy to see that ${L}_{\rho }\left(f\alpha \left(g\right)\right)={L}_{\rho }\left(f\right)g$  for each $f\in {C}_{c}\left(U\right)$  and $g\in C\left(X\right)$  . The following proposition shows that if $U$  is clopen in $X$  then every transfer operator for $\alpha$  is of the form ${L}_{\rho }$  for some $\rho \in C\left(U\right)$  .
Proposition 2.1 Let $L:{C}_{c}\left(U\right)\to C\left(X\right)$  ( $U$  clopen in $X$  ) a transfer operator for $\alpha$  , that is, $L$  is linear, positive and $L\left(g\alpha \left(f\right)\right)=L\left(g\right)f$  for each $f\in C\left(X\right)$  and $g\in {C}_{c}\left(U\right)$  . Then there exists $\rho \in C\left(U\right)$  such that $L={L}_{\rho }$  .
Proof. Let $\left\{{V}_{i}{\right\}}_{i=1}^{n}$  be an open cover of $U$  such that ${\sigma }_{{|}_{{V}_{i}}}$  is a homeomorphism. (such cover exists because $U$  is compact and $\sigma$  is a local homeomorphism). For each $i$  take an open subset ${U}_{i}\subseteq {V}_{i}$  such that $\overline{{U}_{i}}\subseteq {V}_{i}$  and $\left\{{U}_{i}{\right\}}_{i}$  is also a cover for $U$  . Consider the partition of unity $\left\{{\phi }_{i}{\right\}}_{i}$  subordinated to $\left\{{U}_{i}{\right\}}_{i}$  and define ${\xi }_{i}=\sqrt{{\phi }_{i}}$  . Since ${\xi }_{i}$  is positive for each $i$  then $L\left({\xi }_{i}\right)$  is a positive function. Define $\rho ={\sum }_{i=1}^{n}\alpha \left(L\left({\xi }_{i}\right)\right){\xi }_{i}$  which is also positive. Given $f\in {C}_{c}\left(U\right)$  define for each $i$  , ${g}_{i}\left(x\right)=\left\{\begin{array}{cc}{\xi }_{i}\left({\sigma }^{-1}\left(x\right)\right)f\left({\sigma }^{-1}\left(x\right)\right)& x\in \sigma \left({V}_{i}\right)\\ 0& \text{otherwise}\end{array}$  Claim 1: ${g}_{i}\in C\left(X\right)$  for all $i$  Let ${x}_{j}\to x$  . Suppose $x\in \sigma \left({V}_{i}\right)$  . Since $\sigma \left({V}_{i}\right)$  is open we may suppose that ${x}_{j}\in \sigma \left({V}_{i}\right)$  for each $j$  . Since ${\sigma }_{{|}_{{V}_{i}}}$  is a homeomorphism then ${\sigma }^{-1}\left({x}_{j}\right)\to {\sigma }^{-1}\left(x\right)$  in ${V}_{i}$  and so ${g}_{i}\left({x}_{j}\right)=\left({\xi }_{i}f\right)\left({\sigma }^{-1}\left({x}_{j}\right)\right)\to \left({\xi }_{i}f\right)\left({\sigma }^{-1}\left(x\right)\right)={g}_{i}\left(x\right)$  . If $x/\in$  $\sigma \left({V}_{i}\right)$  then $x/\in \sigma \left(\overline{{U}_{i}}\right)$  , which is closed. Therefore we may suppose that ${x}_{j}/\in \sigma \left(\overline{{U}_{i}}\right)$  and so ${g}_{i}\left({x}_{j}\right)=0={g}_{i}\left(x\right)$  .
Claim 2: ${\xi }_{i}\alpha \left({g}_{i}\right)={\phi }_{i}f$  If $x/\in {U}_{i}$  then $\left({\xi }_{i}\alpha \left({g}_{i}\right)\right)\left(x\right)=0=\left({\phi }_{i}f\right)\left(x\right)$  . If $x\in {U}_{i}$  then $\alpha \left({g}_{i}\right)\left(x\right)={g}_{i}\left(\sigma \left(x\right)\right)={\xi }_{i}\left(x\right)f\left(x\right)$  and so ${\xi }_{i}\left(x\right)\alpha \left({g}_{i}\right)\left(x\right)={\xi }^{2}\left(x\right)f\left(x\right)=\phi \left(x\right)f\left(x\right)$  .
Since $\phi$  is partition of unity then $f={\sum }_{i=1}^{n}{\phi }_{i}f={\sum }_{i=1}^{n}{\xi }_{i}\alpha \left({g}_{i}\right)$  , where the last equality follows by claim 2. Then $L\left(f\right)={\sum }_{i=1}^{n}L\left({\xi }_{i}\alpha \left({g}_{i}\right)\right)={\sum }_{i=1}^{n}L\left({\xi }_{i}\right){g}_{i}.$  We show that $L={L}_{\rho }$  . If $x/\in \sigma \left(U\right)$  then ${L}_{\rho }\left(f\right)\left(x\right)=0=L\left(f\right)\left(x\right)$  by definition.
Given $x\in \sigma \left(U\right)$  , ${L}_{\rho }\left(f\right)\left(x\right)={\sum }_{y\in {\sigma }^{-1}\left(x\right)}\rho \left(y\right)f\left(y\right)={\sum }_{y\in {\sigma }^{-1}\left(x\right)}{\sum }_{i=1}^{n}\alpha \left(L\left({\xi }_{i}\right)\right)\left(y\right){\xi }_{i}\left(y\right)f\left(y\right)=$  $={\sum }_{y\in {\sigma }^{-1}\left(x\right)}{\sum }_{i:y\in {U}_{i}}L\left({\xi }_{i}\right)\left(x\right){\xi }_{i}\left(y\right)f\left(y\right).$  On the other hand, $L\left(f\right)\left(x\right)={\sum }_{i=1}^{n}L\left({\xi }_{i}\right)\left(x\right){g}_{i}\left(x\right)={\sum }_{i:x\in \sigma \left({U}_{i}\right)}L\left({\xi }_{i}\right)\left(x\right){\xi }_{i}\left({\sigma }^{-1}\left(x\right)\right)f\left({\sigma }^{-1}\left(x\right)\right).$  To see that those two expressions are equal note that the summands are the same.
$\square$  Denote by ${M}_{\rho }$  the Hilbert bi-module generated by ${C}_{c}\left(U\right)$  with the inner product given by ${L}_{\rho }$  and by $\stackrel{^}{{K}_{1\rho }}$  the algebra generated by $n{m}^{*}$  in $\mathcal{T}\left(X,\alpha ,L\rho \right)$  . Moreover, denote by ${\phi }_{\rho }:C\left(X\right)\to L\left({M}_{\rho }\right)$  the *-homomorphism given by the left product of $A$  by ${M}_{\rho }$  .
Lemma 2.2 Let $\rho ,{\rho }^{\prime }\in C\left(U\right)$  positive functions. If $ker\left(\rho \right)=ker\left({\rho }^{\prime }\right)$  then $ker\left({\phi }_{\rho }\right)=ker\left({\phi }_{{\rho }^{\prime }}\right)$  .
Proof. Let $f\in C\left(X\right)$  . Then $f\in ker\left({\phi }_{\rho }\right)⇔fm=0$  for each $m\in {M}_{\rho }$  $⇔$  $\stackrel{~}{fg}=f\stackrel{~}{g}=0$  for each $g\in {C}_{c}\left(U\right)$  .
It is easy to check that $\stackrel{~}{fg}=0$  in ${M}_{\rho }$  if and only if $\rho fg=0$  . Then $f\in ker\left({\phi }_{\rho }\right)$  if and only if $\rho fg=0$  for each $g\in {C}_{c}\left(U\right)$  . In the same way, $f\in ker\left({\phi }_{\rho }^{\prime }\right)$  if and only if ${\rho }^{\prime }fg=0$  for each $g\in {C}_{c}\left(U\right)$  . Since $ker\left(\rho \right)=ker\left({\rho }^{\prime }\right)$  then $\rho fg=0$  if and only if ${\rho }^{\prime }fg=0$  for each $g\in {C}_{c}\left(U\right)$  .
$\square$
Proposition 2.3 If $\rho$  and ${\rho }^{\prime }$  are elements of $C\left(U\right)$  such that there exists $r\in C\left(U\right)$  such that $r\left(x\right)\ne 0$  for each $x\in U$  and $\rho =r{\rho }^{\prime }$  then $\mathcal{O}\left(X,\alpha ,L\rho \right)$  and $\mathcal{O}\left(X,\alpha ,L{\rho }^{\prime }\right)$  are *-isomorphic.
Proof. Let us define a *-homomorphism from $\mathcal{O}\left(X,\alpha ,L\rho \right)$  to $\mathcal{O}\left(X,\alpha ,L{\rho }^{\prime }\right)$  . Define
 $\begin{array}{cccc}{\psi }_{1}:& C\left(X\right)& \to & \mathcal{T}\left(X,\alpha ,L{\rho }^{\prime }\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\end{array}$
. Let $\xi =\sqrt{r}$  , and note that for each $g\in {C}_{c}\left(U\right)$  , $\parallel \stackrel{~}{g}{\parallel }_{\rho }^{2}=\parallel {L}_{\rho }\left({g}^{*}g\right)\parallel =\parallel L\left(\rho {g}^{*}g\right)\parallel =\parallel L\left(r{\rho }^{\prime }{g}^{*}g\right)\parallel =\parallel {L}_{{\rho }^{\prime }}\left(\left(\xi g{\right)}^{*}\xi g\right)\parallel =\parallel \stackrel{~}{\xi g}{\parallel }_{{\rho }^{\prime }}^{2}.$  So we may define ${\psi }_{2}:{M}_{\rho }\to \mathcal{T}\left(X,\alpha ,L{\rho }^{\prime }\right)$  by ${\psi }_{2}\left(\stackrel{~}{g}\right)=\stackrel{~}{\xi g}$  . Let ${\psi }_{3}={\psi }_{1}\cup {\psi }_{2}$  . We show that ${\psi }_{3}$  extends to $\mathcal{T}\left(X,\alpha ,L\rho \right)$  . For each $f\in C\left(X\right)$  and $g\in {C}_{c}\left(U\right)$  we have ${\psi }_{3}\left(f\right){\psi }_{3}\left(\stackrel{~}{g}\right)=f\stackrel{~}{\xi g}=\stackrel{~}{\xi fg}={\psi }_{3}\left(\stackrel{~}{fg}\right)$  and ${\psi }_{3}\left(\stackrel{~}{g}\right){\psi }_{3}\left(f\right)=\stackrel{~}{\xi g}f=\stackrel{~}{\xi g\alpha \left(f\right)}={\psi }_{3}\left(\stackrel{~}{g\alpha f}\right).$  Moreover, if $h\in {C}_{c}\left(U\right)$  then ${\psi }_{3}\left(\stackrel{~}{g}{\right)}^{*}{\psi }_{3}\left(\stackrel{~}{h}\right)={\stackrel{~}{\xi g}}^{*}\stackrel{~}{\xi h}={L}_{{\rho }^{\prime }}\left(\left(\xi g{\right)}^{*}\xi h\right)={L}_{{\rho }^{\prime }r}\left({g}^{*}h\right)={L}_{\rho }\left({g}^{*}h\right)={\psi }_{3}\left({L}_{\rho }\left({g}^{*}h\right)\right).$  So ${\psi }_{3}$  axtends to $\mathcal{T}\left(X,\alpha ,L\rho \right)$  . Let $\left(f,k\right)\in C\left(X\right)×\stackrel{^}{{K}_{1\rho }}$  a redundancy with $f\in ker\left({\phi }_{\rho }{\right)}^{\perp }\cap {\phi }_{\rho }^{-1}\left(K\left({M}_{\rho }\right)\right)$  . Since ${\psi }_{3}\left({M}_{\rho }\right)\subseteq {M}_{{\rho }^{\prime }}$  it follows that ${\psi }_{3}\left(k\right)\in \stackrel{^}{{K}_{1{\rho }^{\prime }}}$  and so $\left({\psi }_{3}\left(f\right),{\psi }_{3}\left(k\right)\right)\in C\left(X\right)×\stackrel{^}{{K}_{1{\rho }^{\prime }}}$  . Moreover, given $g\in {C}_{c}\left(U\right)$  then ${\xi }^{-1}g\in {C}_{c}\left(U\right)$  and ${\psi }_{3}\left(\stackrel{~}{{\xi }^{-1}g}\right)=\stackrel{~}{g}$  from where ${\psi }_{3}\left({M}_{\rho }\right)$  is dense in ${M}_{{\rho }^{\prime }}$  , and so, since $fm=km$  for each $m\in {M}_{\rho }$  then ${\psi }_{3}\left(f\right)n={\psi }_{3}\left(k\right)n$  for every $n\in {M}_{{\rho }^{\prime }}$  . Therefore $\left({\psi }_{3}\left(f\right),{\psi }_{3}\left(k\right)\right)$  is a redundancy. Since $f\in ker\left({\phi }_{\rho }{\right)}^{\perp }$  , by the previous lemma, ${\psi }_{3}\left(f\right)\in ker\left({\phi }_{{\rho }^{\prime }}{\right)}^{\perp }$  . Then, since $\left({\psi }_{3}\left(f\right),{\psi }_{3}\left(k\right)\right)$  is a redundancy of $\mathcal{T}\left(X,\alpha ,L\right)$  then by [ 3 :2.6], ${\psi }_{3}\left(f\right)\in {\phi }^{-1}\left(K\left({M}_{{\rho }^{\prime }}\right)\right)$  . So ${\psi }_{3}\left(f\right)\in ker\left({\phi }_{{\rho }^{\prime }}{\right)}^{\perp }\cap {\phi }_{{\rho }^{\prime }}^{-1}\left(K\left({M}_{{\rho }^{\prime }}\right)\right)$  . This shows that if $\phi$  is the quotient *-homomorphism from $\mathcal{T}\left(X,\alpha ,L\right)$  in $\mathcal{O}\left(X,\alpha ,L\right)$  then $\phi \circ {\psi }_{3}:\mathcal{T}\left(X,\alpha ,L\rho \right)\to \mathcal{O}\left(X,\alpha ,L\right)$  is a homomorphism which vanishes on all the elements of the form $\left(a-k\right)$  where $\left(a,k\right)$  is a redundancy and $a\in {\phi }_{\rho }^{-1}\left(K\left({M}_{\rho }\right)\right)\cap ker\left({\phi }_{\rho }{\right)}^{\perp }$  . So we obtain a *-homomorphism
 $\begin{array}{cccc}\psi :& \mathcal{O}\left(X,\alpha ,L\rho \right)& \to & \mathcal{O}\left(X,\alpha ,L{\rho }^{\prime }\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\end{array}$
 $\begin{array}{cccc}& \stackrel{~}{g}& ↦& \stackrel{~}{\xi g}\end{array}$
. In the same way we may define the *-homomorphism
 $\begin{array}{cccc}{\psi }_{0}:& \mathcal{O}\left(X,\alpha ,L{\rho }^{\prime }\right)& \to & \mathcal{O}\left(X,\alpha ,L\rho \right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\end{array}$
 $\begin{array}{cccc}& \stackrel{~}{g}& ↦& \stackrel{~}{{\xi }^{-1}g}\end{array}$
. Note that ${\psi }_{0}$  is the inverse of $\psi$  , showing that tha algebras are *-isomorphic.
$\square$
Corollary 2.4 If $\rho \in C\left(U\right)$  is a positive function such that $\rho \left(x\right)\ne 0$  for all $x\in U$  then $\mathcal{O}\left(X,\alpha ,L\rho \right)$  is *-isomorphic to $\mathcal{O}\left(X,\alpha ,L\right)$  .
Proof. Note that the transfer operator $L$  associated to the algebra $\mathcal{O}\left(X,\alpha ,L\right)$  is the operator ${L}_{{1}_{U}}$  . Since $\rho ={1}_{U}$  is invertible, taking $r={\rho }^{-1}$  , by the previous proposition follows the corollary.
$\square$

3 Relationship between the Covariance Algebra and the Crossed Product by Partial Endomorphism

We show here that given a partial dynamical system $\sigma :U\to X$  , where $U$  is clopen, there exists an other partial dynamical system $\stackrel{~}{\sigma }:\stackrel{~}{U}\to \stackrel{~}{X}$  (called in [4the $\sigma$  -extension of $X$  ) such that ${C}^{*}\left(\sigma ,X\right)=\mathcal{O}\left(\stackrel{~}{X},\alpha ,L\right)$  . Moreover, if $\sigma$  is injective then ${C}^{*}\left(\sigma ,X\right)=\mathcal{O}\left(X,\alpha ,L\right)$  .

3.1 The Covariance Algebra as an Crossed Product by a Partial Endomorphism

Let us start with a summary of the construction of the covariance algebra. Let $\sigma :U\to X$  a continuous map, $U\subseteq X$  clopen, $X$  compact Hausdorff and $\sigma \left(U\right)$  open. Denote $\sigma \left(U\right)={U}_{-1}$  . Consider the space $X\cup \left\{0\right\}$  , where $\left\{0\right\}$  is a symbol, which we define to be clopen. So $X\cup \left\{0\right\}$  is a compact and Hausdorff space.
Define $\stackrel{~}{X}\subset {\prod }_{i=0}^{\infty }X\cup \left\{0\right\}$  , $\stackrel{~}{X}={\bigcup }_{N=0}^{\infty }{X}_{N}\cup {X}_{\infty }$  onde ${X}_{N}=\left\{\left({x}_{0},{x}_{1},...,{x}_{N},0,0,...\right):\sigma \left({x}_{i}\right)={x}_{i-1}\text{e}{x}_{N}/\in {U}_{-1}\right\}$  and ${X}_{\infty }=\left\{\left({x}_{0},{x}_{1},...\right):\sigma \left({x}_{i}\right)={x}_{i-1}\right\}.$  In $\stackrel{~}{X}$  we consider the product topology induced from $\prod X\cup \left\{0\right\}$  .
By  4  $\left[:2.2\right]$  $\stackrel{~}{X}$  is compact. Define
 $\begin{array}{cccc}\Phi :& \stackrel{~}{X}& \to & X\end{array}$
 $\begin{array}{cccc}& \left({x}_{0},{x}_{1},...\right)& ↦& {x}_{0}\end{array}$
which is continuous and surjective. Consider the clopen subsets $\stackrel{~}{U}={\Phi }^{-1}\left(U\right)$  and $\stackrel{~}{{U}_{-1}}={\Phi }^{-1}\left({U}_{-1}\right)$  and the continuous map
 $\begin{array}{cccc}\stackrel{~}{\sigma }:& \stackrel{~}{U}& \to & \stackrel{~}{{U}_{-1}}\end{array}$
 $\begin{array}{cccc}& \left({x}_{0},{x}_{1},...\right)& ↦& \left(\sigma \left({x}_{0}\right),{x}_{0},{x}_{1},...\right)\end{array}$
. Those maps satisfies the relation $\Phi \left(\stackrel{~}{\sigma }\left(\stackrel{~}{x}\right)\right)=\sigma \left(\Phi \left(\stackrel{~}{x}\right)\right).$  Note that $\stackrel{~}{\sigma }$  is in fact an homeomorphism. This homeomorphism induces the *-isomorphism
 $\begin{array}{cccc}\theta :& C\left(\stackrel{~}{{U}_{-1}}\right)& \to & C\left(\stackrel{~}{U}\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\circ \stackrel{~}{\sigma }\end{array}$
. So we may consider the partial crossed product $C\left(\stackrel{~}{X}\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}$  (see [1).
Definition 3.1 ( 4 : 4.2) The covariance algebra associated to the partial system $\left(X,\sigma \right)$  is the algebra $C\left(\stackrel{~}{X}\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}$  and will be denoted ${C}^{*}\left(X,\sigma \right)$  .
Lemma 3.2 If $\sigma :U\to X$  is injective, $U$  clopen and ${U}_{-1}$  open then $C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}=\mathcal{O}\left(X,\alpha ,L\right)$  , where $\theta :C\left({U}_{-1}\right)\to C\left(U\right)$  is given by $\theta \left(f\right)=f\circ \sigma$  .
Proof. Define ${\psi }_{1}:C\left(X\right)\cup M\to C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}$  by ${\psi }_{1}\left(f\right)=f{\delta }_{0}$  and ${\psi }_{1}\left(\stackrel{~}{{1}_{U}}\right)={1}_{U}{\delta }_{1}$  . It is easy to check that ${\psi }_{1}$  axtends to $\mathcal{T}\left(X,\alpha ,L\right)$  . We show that ${\Psi }_{1}$  vanishes on the redundancies. Let $\left(f,k\right)$  redundancy with $f\in ker\left(\phi {\right)}^{\perp }\cap {\phi }^{-1}\left(K\left(M\right)\right)$  . By [ 3 : 2.6], $f\in C\left(U\right)$  . Then ${\psi }_{1}\left(\stackrel{~}{f}\right){\psi }_{1}\left(\stackrel{~}{{1}_{U}}{\right)}^{*}=f{\delta }_{1}{1}_{{U}_{-1}}{\delta }_{-1}=\theta \left({\theta }^{-1}\left(f\right){1}_{{U}_{-1}}\right){\delta }_{0}={\psi }_{1}\left(f\right)$  . Take $\left({k}_{n}{\right)}_{n}\subseteq \stackrel{^}{{K}_{1}}$  , ${k}_{n}={\sum }_{i}{m}_{ni}{l}_{ni}^{*}$  where ${m}_{ni},{l}_{ni}\in M$  . Then $\left({\psi }_{1}\left(f\right)-\psi \left(k\right)\right)\left({\psi }_{1}\left(f\right)-\psi \left(k\right){\right)}^{*}=\left({\psi }_{1}\left(f\right)-{\psi }_{1}\left(k\right)\right){\psi }_{1}\left(f-k\right)={\psi }_{1}\left(f-k\right)\left({\psi }_{1}\left(\stackrel{~}{f}{\stackrel{~}{{1}_{U}}}^{*}\right)-{\psi }_{1}\left(k\right){\right)}^{*}=$  $=\psi \left(f-k\right)\left(\stackrel{~}{{1}_{U}}{\stackrel{~}{f}}^{*}-k\right)={lim}_{n\to \infty }\psi \left(f-k\right)\left(\stackrel{~}{{1}_{U}}{\stackrel{~}{f}}^{*}-{k}_{n}\right)=0.$  The last equality follows by the fact that $\left(f-k\right)m=0$  for each $m\in M$  . So, by passage to the quotient we may consider $\psi :\mathcal{O}\left(X,\alpha ,L\right)\to C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}$  . By the other hand, define
 $\begin{array}{cccc}{\psi }_{0}:& C\left(X\right)& \to & \mathcal{O}\left(X,\alpha ,L\right)\end{array}$
 $\begin{array}{cccc}& f& ↦& f\end{array}$
which is a *-homomorphism. Note that for each $f\in C\left({U}_{-1}\right)$  , $\stackrel{~}{{1}_{U}}{\psi }_{0}\left(f\right){\stackrel{~}{{1}_{U}}}^{*}=\stackrel{~}{{1}_{U}\alpha \left(f\right)}{\stackrel{~}{{1}_{U}}}^{*}={1}_{U}\alpha \left(f\right)=\theta \left(f\right)={\psi }_{0}\left(\theta \left(f\right)\right)$  and moreower $\stackrel{~}{{1}_{U}}$  is a partial isometry such that $\stackrel{~}{{1}_{U}}{\stackrel{~}{{1}_{U}}}^{*}={1}_{U}$  and ${\stackrel{~}{{1}_{U}}}^{*}\stackrel{~}{{1}_{U}}={1}_{{U}_{-1}}$  . Then, since $\left({\psi }_{0},\stackrel{~}{{1}_{U}}\right)$  is a covariant representation of $C\left(X\right)$  in $\mathcal{O}\left(X,\alpha ,L\right)$  , there exists a *-homomorphism ${\psi }^{\prime }:C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}\to \mathcal{O}\left(X,\alpha ,L\right)$  such that ${\psi }^{\prime }\left(f{\delta }_{n}\right)=f{\stackrel{~}{{1}_{U}}}^{n}$  (see [ 1 : 5]). The *-homomorphisms $\psi$  and ${\psi }^{\prime }$  are inverses of each other, and so the algebras are *-isomorphic.
$\square$
Corollary 3.3 ${C}^{*}\left(X,\sigma \right)=\mathcal{O}\left(\stackrel{~}{X},\alpha ,L\right)$
Proof. Follows by the definition of covariance algebras and by the previous lemma.
$\square$  By the following proposition, if $\sigma$  is injective then the constructions of covariance algebra and crossed product by partial endomorphism are equivalent.
Proposition 3.4 If $\sigma :U\to X$  is injective then ${C}^{*}\left(X,\sigma \right)=\mathcal{O}\left(X,\alpha ,L\right)$  .
Proof. By [ 4 : 2.3] the map
 $\begin{array}{cccc}\Phi :& \stackrel{~}{X}& \to & X\end{array}$
 $\begin{array}{cccc}& \left({x}_{0},{x}_{1},...\right)& ↦& {x}_{0}\end{array}$
is a homeomorphism. Moreover, since $\Phi \circ \stackrel{~}{\sigma }=\sigma \circ \Phi$  then $C\left(\stackrel{~}{X}\right){\mathbb{\setminus }}_{\stackrel{~}{\theta }}\mathbb{Z}=C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}$  . By the previous lemma $C\left(X\right){\mathbb{\setminus }}_{\theta }\mathbb{Z}=\mathcal{O}\left(X,\alpha ,L\right)$  .
$\square$

3.2 Cuntz-Krieger algebras

We show examples of partial dynamical system ${\sigma }_{A}:U\to {X}_{A}$  such that there does not exists an invertible function $\rho \in C\left(U\right)$  such that $\mathcal{O}\left(X,\alpha ,L\rho \right)$  and ${C}^{*}\left(X,\alpha \right)$  are *-isomorphic. The examples are based on the Cuntz-Krieger algebras.
Let $A$  be a $n×n$  matrix with ${a}_{i,j}\in \left\{0,1\right\}$  . Denote by $Gr\left(A\right)$  the directed graph of $A$  , that is, the vertex set is $\left\{1,...,n\right\}$  and $A\left(i,j\right)$  is the number of oriented edges from $i$  to $j$  . A path is a sequence ${x}_{1},...,{x}_{m}$  such that $A\left({x}_{i},{x}_{i+1}\right)=1$  for each $i$  . The graph $Gr\left(A\right)$  is transitive if for each $i$  and $j$  there exists a path from $i$  to $j$  , that is, a path ${x}_{1},...,{x}_{m}$  such that ${x}_{1}=i$  and ${x}_{m}=j$  . The graph is a cicle if for each $i$  there exists only one $j$  such that $A\left(i,j\right)=1$  .
Let ${X}_{A}=\left\{x=\left({x}_{1},{x}_{2},..\right)\in \left\{1,...n{\right\}}^{\mathbb{N}}:A\left({x}_{i},{x}_{i+1}\right)=1\forall i\right\}\subseteq \left\{1,...,n{\right\}}^{\mathbb{N}}$  and
 $\begin{array}{cccc}{\sigma }_{A}:& {X}_{A}& \to & {X}_{A}\end{array}$
 $\begin{array}{cccc}& \left({x}_{0},{x}_{1},...\right)& ↦& \left({x}_{1},{x}_{2},...\right)\end{array}$
. Consider the set $\overline{{X}_{A}}=\left\{\left({x}_{i}{\right)}_{i\in \mathbb{Z}}\in \left\{1,...,n{\right\}}^{\mathbb{Z}}:A\left({x}_{i},{x}_{i}+1\right)=1\forall i\right\}\subseteq \left\{1,...,n{\right\}}^{\mathbb{Z}}$  and the map $\overline{{\sigma }_{A}}:\overline{{X}_{A}}\to \overline{{X}_{A}}$  defined by $\overline{{\sigma }_{A}}\left(\left({x}_{i}{\right)}_{i\in Z}\right)=\left({x}_{i+1}{\right)}_{i\in \mathbb{Z}}$  . It is showed in [ 4 : 2.8] that there exists a homeomorphism $\Phi :\stackrel{~}{{X}_{A}}\to \overline{{X}_{A}}$  such that $\Phi \circ \stackrel{~}{{\sigma }_{A}}=\overline{{\sigma }_{A}}\circ \Phi$  . Therefore $\mathcal{O}\left(\stackrel{~}{{X}_{A}},\alpha ,L\right)=\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  and so ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)=\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  . So we may analyze the ideal structure of ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)$  by using the theory developed for $\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  in [3. This theory is based on the $\overline{{\sigma }_{A}}$  , ${\overline{{\sigma }_{A}}}^{-1}$  invariant open subsets of $\overline{{X}_{A}}$  . (In a system $\sigma :U\to X$  , a subset $V\subseteq X$  is $\sigma ,{\sigma }^{-1}$  invariant if $\sigma \left(U\cap V\right)\subseteq V$  and ${\sigma }^{-1}\left(V\right)\subseteq V$  ).
Proposition 3.5 If $Gr\left(A\right)$  is transitive and is not a cicle then there exists at least one open non trivial $\overline{{\sigma }_{A}}$  , ${\overline{{\sigma }_{A}}}^{-1}$  invariant subset of $\overline{{X}_{A}}$  .
Proof. Let $r={x}_{1},{x}_{2},...,{x}_{n}$  an admissible word (that is, $A\left({x}_{i},{x}_{i+1}\right)=1$  for each $i$  ). Let ${V}_{r}=\left\{x\in \overline{{X}_{A}}:r\in x\right\}$  . Note that ${V}_{r}$  is open and $\overline{{\sigma }_{A}}$  , ${\overline{{\sigma }_{A}}}^{-1}$  invariant. We show that there exists a such non trivial ${V}_{r}$  . Take ${x}_{1}\in \left\{1,...,n\right\}$  . Consider an admissible word ${x}_{1},...,{x}_{m}$  where ${x}_{j}\ne {x}_{1}$  for each $j>1$  and $A\left({x}_{m},{x}_{1}\right)=1$  . Such word exists because $Gr\left(A\right)$  is transitive. Let $r={x}_{1},...,{x}_{m},{x}_{1}$  . Then $y=\left(...,{x}_{m},\stackrel{\circ }{{x}_{1}},{x}_{2},...,{x}_{m},{x}_{1},{x}_{2},...\right)\in {V}_{r}$  where $\stackrel{\circ }{{x}_{1}}$  means ${y}_{0}={x}_{1}$  .
We conclude the proof by showing that ${V}_{r}\ne \overline{{X}_{A}}$  . Suppose that there exists ${y}_{0}\in \left\{1,...,n\right\}$  with ${y}_{0}/\in \left\{{x}_{1},...,{x}_{m}\right\}$  . Let ${x}_{1},{y}_{1},...,{y}_{t},{y}_{0},{s}_{1},...,{s}_{l}$  an admissible word such that ${y}_{j}\ne {x}_{1}$  and ${s}_{j}\ne {x}_{1}$  for each $j$  and $A\left({s}_{l},{x}_{1}\right)=1$  . Then $\left(...{s}_{l},\stackrel{\circ }{{x}_{1}},{y}_{1},...,{y}_{t},{y}_{0},{s}_{1},...,{s}_{l},{x}_{1}...\right)/\in {V}_{r}.$  If $\left\{{x}_{1},...,{x}_{m}\right\}=\left\{1,...,n\right\}$  , since $Gr\left(A\right)$  is not a cicle, for some ${x}_{i}$  there exists ${x}_{t}$  such that $A\left({x}_{i},{x}_{t}\right)=1$  and ${x}_{t}\ne {x}_{i+1}$  . (if $i=m$  consider ${x}_{i+1}={x}_{1}$  ). If ${x}_{t}={x}_{1}$  (and so $i\ne m$  ) consider an admissible word ${x}_{1},...,{x}_{i},{x}_{1}$  and note that $\left(...,{x}_{i},\stackrel{\circ }{{x}_{1}},{x}_{2},...,{x}_{i},{x}_{1},...\right)/\in {V}_{r}.$  If ${x}_{t}\ne {x}_{1}$  consider an admissible word ${x}_{1},{x}_{2},...,{x}_{i},{x}_{t},{y}_{1},...,{y}_{l}$  such that ${y}_{j}\ne {x}_{1}$  and $A\left({y}_{l},{x}_{1}\right)=1$  (if there does not exists ${y}_{1}\ne {x}_{1}$  such that $A\left({x}_{t},{y}_{1}\right)=1$  then ${y}_{1},...,{y}_{l}$  is the empty word) and so $\left(...,{y}_{l},\stackrel{\circ }{{x}_{1}},{x}_{2},...{x}_{i},{x}_{t},{y}_{1},...,{y}_{l},{y}_{1},...\right)/\in {V}_{r}.$  So ${V}_{r}\ne \overline{{X}_{A}}$  .
$\square$  Now we analyse the ${\sigma }_{A},{\sigma }_{A}^{-1}$  invariant subsets of ${X}_{A}$  .
Proposition 3.6 If $Gr\left(A\right)$  is transitive and is not a cicle the the unique open ${\sigma }_{A}$  -invariant subset of ${X}_{A}$  are $\varnothing$  and ${X}_{A}$  .
Proof. Let $V\subseteq {X}_{A}$  an open nonempty ${\sigma }_{A}$  invariant susbet of ${X}_{A}$  . Let $x\in V$  and ${V}_{m}$  a open neighbourhood of $x$  , ${V}_{m}\subseteq V$  , ${V}_{m}=\left\{y\in {X}_{A}:{x}_{i}={y}_{i}\text{for each}1\le i\le m\right\}.$  Given $z\in {X}_{A}$  take $r={r}_{1},...,{r}_{t}$  a path from ${x}_{m}$  to ${z}_{1}$  . Then $s=\left({x}_{1},...,{x}_{m},{r}_{2},...,{r}_{t-1},{z}_{1},{z}_{2},...\right)\in {V}_{m}$  and since $V$  is ${\sigma }_{A}$  invariant then $z={\sigma }_{A}^{m+t-2}\left(s\right)\in V$  . So $V={X}_{A}$  .
$\square$  According [3a partial dynamical system $\sigma :U\to X$  is topologically free if the closure of ${V}^{i,j}=\left\{x\in U:{\sigma }^{i}\left(x\right)={\sigma }^{j}\left(x\right)\right\}$  has empty interior for each $i,j\in \mathbb{N}$  , $i\ne j$  .
Proposition 3.7 If $Gr\left(A\right)$  is transitive and is not a cicle then $\left({X}_{A},{\sigma }_{A}\right)$  is topologically free.
Proof. Suppose that $\overline{{V}^{i,j}}$  has nonempty interior and $i  , $j=i+k$  . Let ${x}^{\prime }$  be an interior point of $\overline{{V}^{i,j}}$  and ${V}_{{x}^{\prime }}\subseteq \overline{{V}^{i,j}}$  open neighbourhood of ${x}^{\prime }$  . Take $x\in {V}^{i,j}\cap {V}_{{x}^{\prime }}$  . Since ${\sigma }_{A}^{i}\left(x\right)={\sigma }_{A}^{j}\left(x\right)$  then ${z}_{i+t}={z}_{j+t}$  for each $t\in \mathbb{N}$  and since $j=i+k$  then $x=\left({x}_{1},...,{x}_{i-1},r,r,...\right)$  where $r={x}_{i}{x}_{i+1}...{x}_{i+k-1}$  . Consider the open subset ${V}_{m}=\left\{z\in {X}_{A}:{z}_{i}={x}_{i},1\le i\le m\right\}$  where $m$  is such that $m\ge i+k$  and ${V}_{m}\subseteq {V}_{{x}^{\prime }}$  . Then, if $y\in {V}_{m}$  with $y\in {V}^{i,j}$  then $y=x$  . Therefore ${V}_{m}=\left\{x\right\}$  . We show that there exists $z\in {V}_{m}$  with $z\ne x$  , and that will be a contradiction. Suppose ${y}_{0}\in \left\{1,...,n\right\}$  and ${y}_{0}/\in \left\{{x}_{i},...,{x}_{i+k-1}\right\}$  . Take a path $s={s}_{1},...,{s}_{t}$  from ${x}_{i}$  to ${x}_{i+k-1}$  such that ${s}_{j}={y}_{0}$  for some $j$  . Then $z=\left({x}_{1},...,{x}_{i-1},r,r,...,r,s,s,...\right)\in {V}_{m}$  (where $r$  is repeated $m$  times) but $z\ne x$  .
Suppose $\left\{1,...,n\right\}=\left\{{x}_{i},...,{x}_{i+k-1}\right\}$  . Since $Gr\left(A\right)$  is not a cicle then for some ${x}_{j}$  there exists ${x}_{t}\ne {x}_{j+1}$  (consider ${x}_{j+1}={x}_{i}$  if $j=i+k-1$  ) such that $A\left({x}_{j},{x}_{t}\right)=1$  . Let $s$  be a path from ${x}_{t}$  to ${x}_{i+k-1}$  and define $p={x}_{i},...,{x}_{j},{x}_{t},s$  . Then $z=\left({x}_{1},...,{x}_{i-1},r,r,...,r,{x}_{i},...,{x}_{j},{x}_{t},p,p,p,,,...\right)\in {V}_{m}$  (where $p$  is repeated $m$  times) but $z\ne x$  . So, it is showed that there exists $z\in {V}_{m}$  , $z\ne x$  . Therefore, $\overline{{V}^{i,j}}$  has empty interior for each $i,j$  .
$\square$
Theorem 3.8 If $Gr\left(A\right)$  is transitive and is not a cicle then ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)$  and $\mathcal{O}\left({X}_{A},\alpha ,L\right)$  are not *-isomorphic ${C}^{*}$  -algebras.
Proof. By  3.2    , ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)=\mathcal{O}\left(\stackrel{~}{{X}_{A}},\alpha ,L\right)$  and since $\mathcal{O}\left(\stackrel{~}{{X}_{A}},\alpha ,L\right)=\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  then ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)=\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  . By  3.5    , $\overline{{X}_{A}}$  has at least one non trivial open $\overline{{\sigma }_{A}},{\overline{{\sigma }_{A}}}^{-1}$  invariant subset and by [ 3 : 3.9] $\mathcal{O}\left(\overline{{X}_{A}},\alpha ,L\right)$  has at least on non trivial ideal. On the other hand, by  3.6    , $\left({X}_{A},{\sigma }_{A}\right)$  has no open ${\sigma }_{A},{\sigma }_{A}^{-1}$  invariant subsets and by  3.7    , $\left({X}_{A},{\sigma }_{A}\right)$  is topologically free. By [ 3 : 4.8] $\mathcal{O}\left({X}_{A},\alpha ,L\right)$  is simple. So ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)$  and $\mathcal{O}\left({X}_{A},\alpha ,L\right)$  are not *-isomorphic.
$\square$
Corollary 3.9 If $Gr\left(A\right)$  is transitive and is not a cicle then there does not exists transfer operator ${L}_{\rho }$  , with $\rho \left(x\right)\ne 0$  for each $x\in U$  such that ${C}^{*}\left({X}_{A},{\sigma }_{A}\right)$  and $\mathcal{O}\left({X}_{A},\alpha ,L\right)$  are *-isomorphic ${C}^{*}$  -algebras.
Proof. Follows by the previous theorem an by  2.4 .
$\square$  References

1. R. Exel, Circle actions on C ${}^{*}$  -algebras, partial automorphisms and a generalized Pimsner-Voiculescu exact sequence, J. Funct. Anal. 122, (1994), 361-401.
2. R. Exel, A new look at the crossed-product of a ${C}^{*}$  -algebra by an endomorphism, Ergodic Theory Dynam. Systems, Vol. 23 (2003), pp. 1733-1750.
3. R. Exel and D. Royer, The crossed product by a partial endomorphism, Preprint 2004, math.OA/0410192.
4. B. K. Kwasniewski, Covariance algebra of a partial dynamical system, Preprint 2004, math.OA/0407352

Departamento de Matemática, Universidade Federal de Santa Catarina, Brasil. E-mail: royer@mtm.ufsc.br