Lie, Jordan and proper codimensions of associative algebras. (Q1025426)

Let \(A\) be an associative algebra over a field \(F\) of characteristic 0. The study of the polynomial identities of \(A\) can be reduced to the multilinear ones. Let \(P_n\) be the vector space in the free associative algebra spanned by all multilinear elements on \(x_1,\dots,x_n\). It is immediate that \(P_n\) is a module over the symmetric group \(S_n\); in fact it is isomorphic to the regular one. So one studies the identities of \(A\) that lie in \(P_n\). Let \(T(A)\) be the T-ideal of \(A\) (that is the ideal of all identities of \(A\)). Since \(T(A)\cap P_n\) tends to become huge when \(n\) is large one is led to study the quotient \(P_n(A)=P_n/(P_n\cap T(A))\). Clearly \(P_n(A)\) is an \(S_n\)-module. Its dimension \(c_n(A)=\dim P_n(A)\) is the \(n\)-th codimension of \(A\). The codimension sequence of \(A\) is among the most important numerical invariants of the polynomial identities of \(A\). Let \(\exp(A)=\lim_{n\to\infty}c_n(A)^{1/n}\) stand for the PI exponent of \(A\). A celebrated result of the same authors gives that the exponent always exists and is an integer. One may study other kinds of codimensions. Let \(\Gamma_n\) be the subspace of \(P_n\) spanned by all proper (commutator) elements. This means that \(\Gamma_n\) is spanned by all products of commutators (of length at least two). The importance of \(\Gamma_n\) is justified by the fact that when \(1\in A\) then \(T(A)\) is determined by its proper multilinear elements. Clearly \(\Gamma_n\) is an \(S_n\)-module. One defines \(\Gamma_n(A)=\Gamma_n/(\Gamma_n\cap T(A))\) and the \(n\)-th proper codimension \(c_n^p=\dim\Gamma_n(A)\). One may turn \(A\) into a Lie or Jordan algebra in the following standard way. Introduce on the vector space of \(A\) the bracket operation \([a,b]=ab-ba\) and denote the resulting Lie algebra by \(A^-\). Analogously if \(a\circ b=(ab+ba)/2\) then \(A\) becomes a Jordan algebra denoted by \(A^+\). The notions of multilinear Lie and Jordan polynomials in \(P_n\) are the natural ones. Denote the corresponding codimensions as \(c_n^L(A)\) and as \(c_n^J(A)\), respectively. It is well known (and easy to see) that \(c_n^L(A)\leq c_n^p(A)\leq c_n(A)\) and \(c_n^J(A)\leq c_n(A)\). The paper under review studies the relations among the above types of codimensions. As a first step the authors of the paper under review compute the proper, Lie and Jordan exponents of the algebras of block triangular matrices in terms of the sizes of the blocks. It turns out that these are, as in the ordinary case, integers. It turns out that \(\exp^L(R)=\exp^p(R)=\exp^J(R)-1=\exp(R)-1\) if \(R\) is a block triangular matrix algebra. The above result is extended to finitely generated PI algebras \(A\). The authors prove that for such algebras \(\exp^J(A)=\exp(A)\). Moreover, \(\exp^L(A)=\exp^p(A)\) equals either \(\exp(A)\) or \(\exp(A)-1\). If \(1\in A\) then the latter case holds. The most general result in the paper is that if \(A\) is not necessarily finitely generated and \(\exp(A)=d\geq 1\) then \(\exp^p(A)\) exists and equals either \(d\) or \(d-1\), with one exception. It is when \(d=2\) and \(A\) satisfies the same identities as \(G\oplus B\oplus N\) where \(G\) is the infinite dimensional Grassmann algebra, \(B\) is finite dimensional and with Jacobson radical of codimension 1, and \(N\) is nilpotent.