Polynomial codimension growth and the Specht problem (Q330176)

The paper under review studies polynomial identities of nonassociative algebras over a field \(F\) of characteristic 0. In characteristic 0, every ideal of identities in the free nonassociative algebra \(F\{X\}\) is generated as a T-ideal by its multilinear elements. If \(P_n\) is the vector space of the multilinear polynomials in the variables \(x_1\), \dots, \(x_n\) in \(F\{X\}\) then the symmetric group acts on \(P_n\) by permuting the variables. Clearly the identities for an algebra \(A\) that lie in \(P_n\) form an \(S_n\)-submodule; one studies the factor-module \(P_n(A)=P_n/P_n\cap Id(A)\) where \(Id(A)\) is the T-ideal of \(A\). Thus one employs the representation theory of \(S_n\) to the study of the polynomial identities of an algebra. One of the most important numerical invariants of a T-ideal is its codimension sequence \(c_n(A)=\dim P_n(A)\). It is clear that if \(A\) is associative then \(c_n(A)\leq n!\); if \(A\) is a Lie algebra then \(c_n(A)\leq (n-1)!\) and in general \(c_n(A)\leq {1\over n} {2n-2\choose n-1} n!\) the latter being an overexponential function. It is well known that if \(A\) is associative PI then there exists a constant \(d\) such that \(c_n(A)\leq d^n\); if \(A\) is finite dimensional (not necessarily associative) then \(c_n(A)\leq (\dim A)^{n+1}\). If there are constants \(\alpha\) and \(t\) such that \(c_n(A)\leq \alpha n^t\) then \(A\) is of polynomial growth. This is a significant restriction on \(A\); in the associative, Lie and Jordan cases the polynomial growth is equivalent to \(c_n(A)=qn^t+O(n^{t-1})\), \(q>0\) and \(t\) a positive integer. It follows \(\lim\log_n c_n(A) = t\) must be an integer in the associative, Lie and Jordan cases as well as for Poisson algebras. On the other hand for general nonassociative algebras this need not hold. There are examples of nonassociative algebras with \(t=2.5\) and even examples where the limit \(\lim\log_n c_n(A)\) does not exist at all. From the opposite side if the codimension growth of an algebra \(A\) is less than linear then necessarily \(c_n(A)\leq 1\) for large \(n\). Thus studying the varieties of nonassociative algebras having slow codimension growth may reveal interesting information. In this direction the authors of the paper under review obtain the a series of nice examples. They exhibit, for each real \(\alpha\), \(0<\alpha<1\), a commutative and metabelian algebra \(A_\alpha\) such that \(c_n(A_\alpha\) behaves asymptotically as \(n^{3+\alpha}\). It is also interesting to study how the slow codimension growth may affect the identities satisfied by the algebras. Clearly there is no hope that one may describe explicitly all varieties of algebras with polynomial growth of their codimensions. The authors of the paper go one step further in this direction. Namely they exhibit a variety of commutative (nonassociative) algebras whose codimensions are polynomially bounded but do not satisfy the Specht property. We recall that in the associative case such an example does not exist as proved by Kemer. In fact Kemer proved that a variety of associative algebras with polynomial growth of its codimensions is necessarily Spechtian, and later on he developed the structure theory of the ideals of identities of associative algebras in characteristic 0. One of the consequences of Kemer's theory was the positive solution of the Specht problem for associative algebras. Moreover \textit{I. I. Benediktovich} and \textit{A. E. Zalesskiĭ} [Izv. Akad. Nauk BSSR, Ser. Fiz.-Mat. Nauk 1980, No. 3, 5--10 (1980; Zbl 0434.17009)] described the ideals of identities of Lie algebras with polynomial growth, and proved that these are Spechtian.