The isomorphism problem for some universal operator algebras (Q635487): Difference between revisions

In this extensive paper, the authors declare that, ``[i]n some sense, we are attempting to develop noncommutative complex algebraic geometry.'' The diligent reader is rewarded, in the form of attractive theorems concerning isomorphisms of operator algebras associated to ideals of multivariable complex polynomials, with significant insight into what such a development might entail. The starting point is the observation that, if \(p\) is a polynomial in \(d\) variables, then the equation \(p(x_1, \dots, x_d)=0\) makes sense if the variables are taken to be bounded linear operators on Hilbert space. A finite-dimensional Hilbert space \(E\) with orthonormal basis \(\{e_1, \dots, e_d\}\) generates the free Fock space \({\mathcal F}(E)=\bigoplus_{n\geq 0}{E^{\otimes n}}\). The \(d\)-shift \(L=(L_1, \dots, L_d)\) on \({\mathcal F}(E)\) is defined by \(L_i(e_{\alpha_1}\otimes\cdots\otimes e_{\alpha_n}) = e_i\otimes e_{\alpha_1}\otimes\cdots\otimes e_{\alpha_n}\) for \(i=1, \dots, d\). For a polynomial \(p\) in \(d\) non-commuting variables with complex coefficients, take \(p(e)\) to denote the element of \({\mathcal F}(E)\) obtained by substituting the basis vectors \(\{e_i\}\) for the variables and replacing multiplication with \(\otimes\); for instance, if \(p(z)=z_1z_2-z_3z_1z_3\), then \(p(e)=e_1\otimes e_2-e_3\otimes e_1\otimes e_3\). A \textit{subproduct system} is a collection \(X=\{X(n)\}_{n\geq 0}\) of finite-dimensional Hilbert spaces such that \(X(0)={\mathbb C}\) and \(X(m+n)\subseteq X(m)\otimes X(n)\). Taking \(E=X(1)\), the space \({\mathcal F}_X:=\bigoplus_{n\geq 0}{X(n)}\) is a subspace of \({\mathcal F}(E)\). To a subproduct system \(X\), we associate the ideal of polynomials \(I^X=\text{span}\{p:p(e)\in E^{\otimes n}\ominus X(n) \text{ for\;some\;}n\}\). Indeed, there is a bijection between homogeneous ideals in \({\mathbb C}\langle z_1, \dots, z_d\rangle\) and subproduct systems \(X\) with \(X(1)\subseteq E\). Now let \({S_1}^X, \dots, {S_d}^X\) denote the compressions of \(L_1, \dots, L_d\) to \({\mathcal F}_X\). The main object studied in this paper is \({\mathcal A}_X\), defined to be the norm closed algebra generated by \(I, {S_1}^X, \dots, {S_d}^X\). In the case where the polynomials are in the commuting variables \(z_1, \dots, z_d\), the structures involved in this scenario are simpler, but the overall construction is similar and the same notation can be used. In the commutative setting, the \textit{radical} of an ideal of polynomials is defined by \(\sqrt{I}:=\{p:p^n\in I\;\text{for\;some\;}n\}\), and a radical ideal is one for which \(I=\sqrt{I}\). The ``main noncommutative result'' of the paper is Theorem 4.8, which asserts that the subproduct systems \(X\) and \(Y\) are isomorphic if, and only if, the corresponding algebras \(A_X\) and \(A_Y\) are isometrically isomorphic. To illustrate some of the subtlety of this result, Example 3.4 offers us two algebras \(A_X\) and \(A_Y\) that are isomorphic as algebras but not isometrically isomorphic; thus, the underlying subproduct systems \(X\) and \(Y\) are not isomorphic even though there is a unital algebraic automorphism mapping \(I^X\) onto \(I^Y\). From Section 5 on, the paper focuses on the commutative case, where, as Proposition 5.6 shows, when \(I=I^X\) is a radical homogeneous ideal, then \({\mathcal F}_X\) is a reproducing kernel Hilbert space on the set \(Z^0(I)\) and \(A_X\) is the norm closure of the polynomials in the algebra of multipliers on \({\mathcal F}_X\). (Here, \(Z^0(I)\) denotes the intersection of the open unit ball in \({\mathbb C}^d\) and the variety \(V(I):=\{z\in {\mathbb C}^d:p(z)=0\;\forall p\in I\}\).) Section 6 considers a broader class of reproducing kernel Hilbert spaces and culminates in Theorem 6.12, ``a projective Nullstellensatz for closed ideals'' in the closure of the polynomials in the multiplier algebra on such a space, and its Corollary 6.13, which demonstrates ``that in these algebras, any function that vanishes on a homogeneous algebraic variety can be approximated in the norm by polynomials vanishing on that variety.'' Sections 7 and 8 make up about one third of the paper with, as the authors put it, ``most of the hard work in the former, and the main results in the latter.'' In Theorem 7.17, the highlight of Section 7, the authors show that, when \(I\) and \(J\) are radical homogeneous ideals in \({\mathbb C}[z_1, \dots, z_d]\) such that the variety \(V(J)\) is \textit{tractable}, then every invertible linear map between \(V(I)\) and \(V(J)\) that is length-preserving on the varieties induces a completely bounded isomorphism from \(A_I\) onto \(A_J\). Among the classification results in Section 8, we find Theorem 8.2 -- when \(I\) and \(J\) are radical homogeneous ideals in \({\mathbb C}[z_1, \dots, z_d]\), then \(A_I\) and \(A_J\) are isometrically isomorphic if, and only if, there is a unitary transformation \(U\) on \({\mathbb C}^d\) taking \(V(J)\) onto \(V(I)\) -- and Theorem 8.5 -- when \(V(I)\) and \(V(J)\) are tractable, then \(A_I\) is isomorphic to \(A_J\) if, and only if, there is a linear map from \(V(I)\) onto \(V(J)\) that is length-preserving on \(V(I)\). There is much more in this paper to sink one's teeth into, so hopefully these highlights have whetted the reader's appetite.