Multilinear polynomials are surjective on algebras with surjective inner derivations (Q2204841)

Let \(F\langle X\rangle\) be the free associative algebra freely generated by the set \(X\) over a field \(F\), \(X=\{X_1, X_2, \ldots\}\). If \(f\in F\langle X\rangle\) is a polynomial and \(A\) an \(F\)-algebra then one denotes \(f(A)\) the image of \(f\) on \(A\), that is the set of all evaluations \(f(a_1,\ldots,a_n)\), \(a_i\in A\). Assume that \(f\) is multilinear in \(X_1\), \dots, \(X_n\). A well-known conjecture proposed by Kaplansky and L'vov is that \(f(A)\) is a vector subspace of \(A\) whenever \(A=M_k(F)\), the full matrix algebra over \(f\). It is also well known that such a vector subspace can be either 0 (then \(f\) is a polynomial identity), or \(F\) (then \(f\) is a central polynomial), or \(sl_k(F)\), or else \(M_k(F)\). This conjecture holds for \(2\times 2\) matrices by \textit{A. Kanel-Belov} et al. [Proc. Am. Math. Soc. 140, No. 2, 465--478 (2012; Zbl 1241.16017)], whenever the base field \(F\) is quadratically closed. The paper under review goes in a different direction related to that conjecture. Let \(A\) be an algebra and suppose \(A\) has an inner derivation that is surjective. Note that the matrix algebras over \(F\) admit no such inner derivations as their images are always contained in \(sl_n\). In fact no finite dimensional algebra can have a surjective inner derivation: if it is given by \(b\mapsto [b,a]\) for some fixed \(a\in A\), and if \(a\ne 0\) then the kernel of the derivation is nontrivial. A similar conclusion can be drawn for an arbitrary PI algebra, as it was proved by \textit{A. Ya. Belov} [Sib. Mat. Zh. 44, No. 6, 1239--1254 (2003; Zbl 1054.16015); translation in Sib. Math. J. 44, No. 6, 969--980 (2003)]. This very neatly written paper establishes the following theorem. Suppose \(A\) is a unitary \(F\)-algebra that admits a surjective inner derivation. If \(f\) is a nonzero multilinear polynomial then \(f(A)=A\). Two corollaries are worth mentioning. If \(V\) is an infinite dimensional vector space over \(F\) (or more generally over a division algebra over \(F\)) and \(E\) is the algebra of the endomorphisms of \(V\) then \(f(E)=E\). Also if \(W_n\) is the \(n\)th Weyl algebra then \(f(W_n)=W_n\).