On group gradings on PI-algebras. (Q2257283)

The development of the theory of algebras with polynomial identities in recent times has been closely associated to gradings on algebras and the corresponding graded polynomial identities. Let \(W\) be an associative algebra over a fixed field of characteristic 0 and denote by \(T(W)\) its T-ideal. Thus \(T(W)\) is the ideal of all polynomial identities for \(W\) in the free associative algebra with free generators \(x_1,x_2,\ldots\). Denote by \(P_n\) the span of \(x_1,\ldots,x_n\), it is well known that \(T(W)\) is generated as a T-ideal by all of its multilinear elements, that is by all polynomials lying in \(\bigcup_{n\geq 1}P_n\cap T(W)\). One of the most important numerical invariants of a PI algebra \(W\) is its codimension sequence \(c_n(W)\) where \(c_n(W)=\dim P_n/(P_n\cap T(W))\). Amitsur conjectured that the limit \(\exp(W)=\lim_{n\to\infty}c_n(W)^{1/n}\) exists for every PI algebra \(W\) and that this limit is always an integer. It is called the PI exponent of \(W\). The original Amitsur's conjecture was proved by Giambruno and Zaicev, and later on Giambruno, La Mattina and Aljadeff proved that it holds also for \(G\)-graded algebras assuming the group \(G\) is finite. Recall that the finiteness of the group \(G\) is essential in the above mentioned result. Now let \(G\) be an arbitrary group (possibly infinite) and let \(W=\bigoplus_{g\in G}W_g\) be \(G\)-graded. The grading is non-degenerate if for each positive integer \(r\) and for every \(g_1,\ldots,g_r\in G\) one has \(W_{g_1}\cdots W_{g_r}\neq 0\). The main theorem of the paper under review states that there exists an integer \(K\) such that for every PI algebra \(W\) graded non-degenerately by an arbitrary group \(G\) there exists an abelian subgroup \(U\leq G\) with \(|G:U|\leq\exp(W)^K\). Recall that a \(G\)-grading on \(W\) is strong if for every \(g,h\in W\) one has \(W_gW_h=W_{gh}\). A crossed product grading is one where each homogeneous component contains an invertible element. Clearly the latter condition implies the former, and it in turn implies non-degeneracy of the grading. On the other hand one may consider weaker conditions on the grading. These include the so-called bounded non-degenerate gradings: these are \(G\)-gradings on \(W\) such that there exists a constant \(r_0\) and for every \(r\leq r_0\) and every \(g_1,\ldots,g_r\) one has \(W_{g_1}\cdots W_{g_r}\neq 0\). A still weaker requirement on the grading is the connectedness: a grading is connected if \(W_g\neq 0\) for each \(g\in G\). The authors give an example of a PI algebra \(W\) with a bounded non-degenerate grading where their main theorem does not hold. Therefore one cannot weaken too much the conditions on the grading.