3
$\begingroup$

Inspired by this MSE question we ask the following question:

Is there a noncommutative $C^*$-algebra $A$ for which the following identity holds for all $x,y \in A$?

$$e^{(xy-yx)}= e^xe^y e^{-x}e^{-y}$$

That is $$e^{[x,y]}=[e^x,e^y]$$ where the bracket on the left-hand side is the algebra commutator, and the bracket on the right-hand side denotes the group commutator.

$\endgroup$
13
  • 1
    $\begingroup$ Thanks for the clarification. I have reworded the title in case anyone makes the same error as I did $\endgroup$
    – Yemon Choi
    Oct 13, 2019 at 18:46
  • 2
    $\begingroup$ I would expect that the Baker-Campbell-Hausdorff formula could be used to say that $A$ is commutative; failing that, that for every $X, Y$ we have that $[X, Y]$ commutes with $X$ and $Y$. Have you tried it? $\endgroup$
    – user44191
    Oct 13, 2019 at 21:05
  • 5
    $\begingroup$ Taylor at order 3 $[+O(\|x\|^4+\|y\|^4)]$ yields that a unital Banach algebra with this identity has to be 2-step nilpotent (that is $[x,y]$) is central for all $x,y$); indeed it says that $[x,[x,y]]$ is zero, which by polarization gives the result. $\endgroup$
    – YCor
    Oct 13, 2019 at 21:12
  • 2
    $\begingroup$ @YCor I think it may be worth noting that the Jacobi identity is needed, that is, that polarization doesn't quite give the result directly. The derivation is straightforward, but not immediate. $\endgroup$
    – user44191
    Oct 14, 2019 at 4:04
  • 2
    $\begingroup$ Conversely BCH for a unital Banach algebra satisfying $[x,[x,y]]=0$ identically writes as $\exp(x)\exp(y)=\exp(x+y+\frac12[xy])$ and this gives the commutator identity. So a unital Banach algebra satisfies the given commutator identity iff its underlying Lie algebra is 2-step nilpotent. What remains is to determine whether a $C^*$-algebra whose underlying Lie algebra is 2-step nilpotent is necessarily commutative. $\endgroup$
    – YCor
    Oct 14, 2019 at 8:45

1 Answer 1

14
$\begingroup$

Yes:

A $C^*$-algebra satisfies the identity $e^{[xy-yx]}=e^xe^ye^{-x}e^{-y}$ iff it is commutative.

This follows from two independent facts (I write $[x,y]=xy-yx$)

1) A (real/complex) unital Banach algebra satisfies the identity $e^{[xy-yx]}=e^xe^ye^{-x}e^{-y}$ $\Leftrightarrow$ it satisfies the identity $[x,[x,y]]=0$ $\Leftrightarrow$ it satisfies the identity $[x,[y,z]]$, i.e., its underlying Lie algebra is 2-step nilpotent.

2) A $C^*$-algebra satisfying the identity $[x,[x,y]]=0$ is commutative.

For (1), one obtains that the main implication, namely the identity implies $[x,[x,y]]=0$, by a 3rd order Taylor expansion: in every Banach algebra one has $$e^{xy-yx}=1+xy-yx+o(\|x\|^3+\|y\|^3);$$ $$e^xe^ye^{-x}e^{-y}=1+xy-yx+\frac12([x+y,[x,y]])+o(\|x\|^3+\|y\|^3),$$ so the commutator identity forces $[x+y,[x,y]]=0$ for all $x,y$, and hence $[x,[x,y]$ identically vanishes by homogeneity.

(As mentioned in the comments, a simple argument then implies $[x,[y,z]]=0$ for all $x,y,z$, but this is not needed to run the argument.)

Conversely, if $[x,[x,y]]$ is identically zero, the Baker-Campbell-Hausdorff formula reads as: for all $x,y$, one has $\exp(x)\exp(y)=\exp(x+y+(1/2)[x,y])$. The commutator identity follows.

Let us pass to (2). Let $A$ be a $C^*$-algebra satisfying the PI-identity $[x,[x,y]]=0$. First, if $A$ is finite-dimensional, it is a product of matrix algebras, and hence is commutative (since the given identity fails for $M_n(\mathbf{C})$ for $n\ge 2$). In general, I need the last emphasized statement in my answer here: $A$ has a family $(J_i)$ of finite-codimensional 2-sided closed $*$-ideals with trivial intersection (so $A$ embeds into the product $\prod_i A/J_i$). By the finite-dimensional case, $A/J_i$ is commutative for every $i$. Hence $A$ is commutative.

$\endgroup$
2
  • $\begingroup$ I think your 3rd order Taylor expansion can't be right; there should be some symmetry between $x$ and $y$, which should force a term of the form $[x, [x, y]]$; this can also be seen by looking at the inverse (which takes the form $e^y e^x e^{-y} e^{-x}$). This can be dealt with by slightly more careful manipulation of the coefficients of $x$ and $y$, though. $\endgroup$
    – user44191
    Oct 14, 2019 at 16:18
  • 1
    $\begingroup$ @user44191 Thanks! I had actually selected one of two homogeneous terms in my draft, then forgot the other one. It indeed yields $[x+y,[x,y]]$. Corrected. $\endgroup$
    – YCor
    Oct 14, 2019 at 16:45

Your Answer

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

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