In the algebra of matrices $M_n(A)$ over a $C^*$ algebra $A$, consider the corner algebra $PM_n(A)P$ for a Hermitian projection $P\in M_n(A)$. Is there any condition known for $P$ to make $PM_n(A)P$ isomorphic to $A$? If not, where should I look for what is known on the structure of $PM_n(A)P$? (This question is aimed at linking the peoperties of $P$ to $PM_n(A)P$ in general, rather than for the case of specific $C^*$ algebras.)

The background: A Morita context from $A$ to itself can be viewed as a left finitely generated projective $A$-module $L$ whose left module maps are themselves isomorphic to $A$ - thus giving a bimodule structure. If $L$ has associated projection matrix $P$ in $K$-theory, then the left module maps are given by the corner algebra $PM_n(A)P$, thus the question of when this is isomorphic to $A$. (This is the noncommutative analogue of a line bundle in geometry.)

  • $\begingroup$ Larry Brown's paper, Stable isomorphism of hereditary subalgebras of C*-algebras. Pacific J. Math. 71 (1977), no. 2, 335–348, is probably a good place to start. $\endgroup$ Jun 9, 2014 at 22:33
  • $\begingroup$ Just got a copy of Brown's paper - I will read it! $\endgroup$ Jun 10, 2014 at 13:23

1 Answer 1


If $A$ is unital, then an obvious sufficient condition would be that $P$ is Murray-von-Neumann equivalent to the unit $1_A$ of $A$. (Where we view $A$ as sitting canonically in $M_n(A)$ as the upper left corner, or more algebraically, $A$ is identified with $e_{11}\otimes A\subset M_n(A)$).

Because if $v\in M_n(A)$ is a partial isometry with $v^*v=1_A$ and $vv^*=P$, then the map $A\to PM_n(A)P$ defined by $x\mapsto vxv^*$ yields an isomorphism.

  • $\begingroup$ This seems wrong to me. The map you define gives an isomorphism $M_n(A) \to PM_n(A)P$, rather than $A \to PM_n(A)P$. More generally, if two projections $P$ and $Q$ are Murray-von Neumann equivalent, then a partial isometry inducing the equivalence should induce an isomorphism between the corners $PM_n(A)P$ and $QM_n(A)Q$ (and there is nothing terribly special about the algebra $M_n(A)$ here). $\endgroup$ Jun 9, 2014 at 20:41
  • $\begingroup$ Note the identification of the unit $1_A$ with $e_{11}\otimes 1_A$, as I mentioned right at the beginning of my answer. You have $(e_{11}\otimes 1_A)M_n(A)(e_{11}\otimes 1_A)\cong A$, so what you say is exactly the point. $\endgroup$ Jun 10, 2014 at 7:10
  • $\begingroup$ Oops! My apologies for misreading $1_A$ for $1_{M_n(A)}$! $\endgroup$ Jun 10, 2014 at 11:26
  • $\begingroup$ That is a good point - hopefully there are other cases - there are classically line bundles not with the same K-theory class as the trivial bundle. (E.g. on the sphere.) $\endgroup$ Jun 10, 2014 at 13:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.