Proper identities, Lie identities and exponential codimension growth. (Q950212)

Assume \(F\) is a fixed field of characteristic 0, and let \(F(X)\) be the free associative algebra freely generated over \(F\) by the infinite set \(X\). It is well known that the polynomial identities of an associative algebra are determined by the multilinear ones. If one considers unitary algebras, \(1\in A\), then the identities of \(A\) are determined by the proper ones, that is by the polynomials that are linear combinations of products of Lie commutators. One of the most important characteristics of the PI algebra \(A\) is its sequence of codimensions \(c_n(A)=\dim P_n/(P_n\cap T(A))\) where \(T(A)\) is the T-ideal of \(A\) and \(P_n\) is the vector subspace of \(F(X)\) consisting of all multilinear elements on \(x_1,\dots,x_n\). A celebrated theorem of Regev states that if \(A\) satisfies a nontrivial identity then \(c_n(A)\) is exponentially bounded. Later on the authors of the paper under review proved [in Adv. Math. 140, No. 2, 145-155 (1998; Zbl 0920.16012) and ibid. 142, No. 2, 221-243 (1999; Zbl 0920.16013)] that if \(A\) satisfies a nontrivial identity then the limit \(\lim_{n\to\infty}(c_n(A))^{1/n}\) exists and is always an integer, called the PI exponent \(\exp(A)\) of \(A\). In the paper under review the authors deal with similar problems concerning the sequences of Lie codimensions and of proper codimensions of an associative algebra \(A\). Recall that \(A\) can be viewed as a Lie algebra with respect to the usual Lie bracket. Assume \(V_n\) is the vector space of the Lie multilinear elements in the variables \(x_1,\dots,x_n\), one defines the \(n\)-th Lie codimension of \(A\) as \(c^L_n(A)=\dim V_n/(V_n\cap T(A))\). (Here one considers the free Lie algebra embedded in \(F(X)\).) Analogously let \(\Gamma_n\) be the vector space in \(F(X)\) of all proper multilinear elements in \(x_1,\dots,x_n\). Then the \(n\)-th proper codimension \(c^p_n(A)=\dim\Gamma_n/(\Gamma_n\cap T(A))\). It is known that the inequalities \(c^L_n(A)\leq c^p_n(A)\leq c_n(A)\) hold. As in the ordinary case one may define the lower and the upper Lie exponents of \(A\) by \(\liminf_{n\to\infty}(c^L_n(A))^{1/n}\) and by \(\limsup_{n\to\infty}(c^L_n(A))^{1/n}\). If these two coincide they are called the Lie exponent of \(A\), \(\exp^L(A)\). In the same manner one defines the proper exponent of \(A\), \(c^p(A)\). The behaviour of these invariants for Lie algebras is not that smooth as in the associative case. It is known that \(c_n(A)\) may grow overexponentially for a Lie algebra \(A\); even if it grows exponentially the rate of growth is not always an integer. However if \(\dim A<\infty\) then \(c^L(A)\) exists and is an integer. The paper under review deals mainly with finitely generated PI algebras. It is proved for such algebras that \(\exp^L(A)\) and \(\exp^p(A)\) do exist, and are integers. Furthermore, they either are equal to \(\exp(A)\) or to \(\exp(A)-1\). If \(A\) is not finitely generated the situation is more complicated. The authors construct the algebra \(A=G\oplus B\oplus N\) such that \(\exp^p(A)\) does not exist. Here \(G\) is the infinite dimensional Grassmann algebra, \(\dim B<\infty\) and the Jacobson radical of \(B\) is of codimension 1, and \(N\) is nilpotent. On the other hand the authors give a general positive result about the proper exponent. Namely they prove that if \(\exp(A)=d\) then \(\exp^p(A)\) does exist and equals either \(d\) or \(d-1\), unless \(d=2\) and \(A\) is of the type \(G\oplus B\oplus N\) as above.