Approximately order zero maps between \(\mathrm{C}^\ast \)-algebras (Q2020095)

Let \(\mathcal{A}\) and \(\mathcal{B}\) be two C\(^*\)-algebras, \(\phi:\mathcal{A}\to \mathcal{B}\) a linear bounded map, and if \(\mathcal{A}\) has a unit \(\mathbf{1}\) then set \(h=\phi(\mathbf{1})\). The paper under review deals with maps \(\phi\) which approximately preserve orthogonality. In one of his seminal papers, \textit{R. V. Kadison} [Ann. Math. (2) 54, 325--338 (1951; Zbl 0045.06201)] has shown that surjective linear isometries between unital C\(^*\)-algebras are the composition of a Jordan \(^*\)-isomorphism and the multiplication by a unitary element. Although orthogonality preserving linear maps clearly need not be isometries, they turned out to be of a similar form. In fact, \textit{M. Wolff} [Arch. Math. 62, No. 3, 248--253 (1994; Zbl 0803.46069)] has shown that if \(\mathcal{A}\) is unital and if \(\phi\) preserves self-adjointness and orthogonality of self-adjoint elements, then it is a Jordan \(^*\)-homomorphism compressed (or weighted) by \(h\), more precisely, the image of \(\phi\) lies in \(\mathcal{C}=\overline{h\{h\}'}\) and there is a Jordan \(^*\)-homomorphism \(S\) from \(\mathcal{A}\) into the multiplier algebra \(\mathcal{M}(\mathcal{C})\) such that \(\phi(\cdot)=hS(\cdot)\). (As pointed out by Wolff, examples of (even commutative) algebras show that in general \(S\) cannot be expected to map into \(\mathcal{B}\).) The author cites and uses a stability result of \textit{J. Alaminos} et al. [Isr. J. Math. 178, 1--28 (2010; Zbl 1209.47013)] which, roughly speaking, states (amongst others) that if \(\mathcal{A}\) is unital and if \(\phi\) preserves zero products (of arbitrary elements of \(\mathcal{A}\)), then \(\phi\) is multiplicative ``up to \(h\)'' in the sense that \(\phi(x)\phi(y)=\phi(xy)h\) (in particular, \(\phi\) is a compressed Jordan \(^*\)-homomorphism), and which states moreover that this fact remains stable for almost zero-product-preserving maps. The paper under review addresses the question of stability under small perturbations under the weaker hypothesis that \(\phi\) almost preserves only orthogonality of positive elements. To this end, the author considers, for \(\varepsilon\ge0\), the four notions \begin{itemize} \item \(\phi\) is an \(\varepsilon\)-order zero map if \(\|\phi(x)\phi(y)\|\le\varepsilon\|x\| \|y\|\) for all positive orthogonal elements \(x,y\in \mathcal{A}\). (Two elements \(x,y\in \mathcal{A}\) are called orthogonal, \(x\perp y\), if \(xy=yx=x^*y=xy^*=0\), which is stronger than the usual C\(^*\)-orthogonality \(x\perp_{C^*} y\) demanding \(x^*y=xy^*=0\); but since only orthogonality of positive elements is considered in the paper, both notions amount to the same. Moreover, easy arguments show that if \(\phi\) is \(\varepsilon\)-order zero, then it is \(16\varepsilon\)-order zero for all \(x,y\in \mathcal{A}\), \(x\perp y\).) \item \(\phi\) is \(\varepsilon\)-self-adjoint if \(\|\phi(x^*)-\phi(x)^*\|\le\varepsilon\|x\|\) for all \(x\in \mathcal{A}\). \item \(\phi\) is \(\varepsilon\)-disjointness preserving if it is both \(\varepsilon\)-order zero and \(\varepsilon\)-self-adjoint. \item \(\phi\) is an \(\varepsilon\)-Jordan \(^*\)-homomorphism if it is \(\varepsilon\)-self-adjoint and if \(\|\phi(x^2)-\phi(x)^2\|\le\varepsilon\|x\|^2\) for all \(x\in \mathcal{A}\). \end{itemize} A first satisfactory result is that if \(\mathcal{A}\) is unital and if \(\phi\) is \(\varepsilon\)-order zero, then \(\phi\) is almost a compressed Jordan homomorphism, more precisely, \(\|\phi(x)^2-h\phi(x^2)\|\le108\varepsilon\|x\|^2\) for all \(x\in \mathcal{A}\). (It is shown that if \(\mathcal{A}\) is nonunital, then \(\phi\) can be extended to an \(\varepsilon\)-order zero map on the unitization of \(\mathcal{A}\).) Then the author continues to address the more delicate question whether an \(\varepsilon\)-disjointness preserving \(\phi\) can be approximated by a compressed Jordan homomorphism in the sense that \(\phi=\phi_s+\phi_r\) with small \(\phi_s\) (i.e., \(\|\phi_s\|\to0\) as \(\varepsilon\to0\)) and such that \(\phi_r\) is a compressed unital Jordan homomorphism. There is the structural result that the range of \(\phi\) is close to the hereditary subalgebra \(\overline{h \mathcal{B} h}\). In order to achieve the decomposition \(\phi=\phi_s+\phi_r\) one runs into two technical obstacles concerning commutation relations for spectral projections: in general almost commuting operators (in fact matrices) need not be close to commuting ones and in general the commutator of polynomials of \(h\) and elements of \(\mathcal{B}\) becomes unbounded as the degree of the polynomial increases. Therefore the decomposition \(\phi=\phi_s+\phi_r\) is achieved when \(h\) is an algebraic element of \(\mathcal{B}\), for example when \(\mathcal{B}\) is finite dimensional; it is also achieved when \(\mathcal{A}\) is commutative and \(\phi\) is surjective. Attention is paid to obtain concrete constants. Reviewer's question: Could it be useful to use ultraproduct techniques in order to get perturbation results (at the cost of concrete constants)?