Anomalies on codimension growth of algebras. (Q302305)

The algebras considered in the paper under review are assumed to be over a fixed field of characteristic 0, and are not necessarily associative. If \(A\) is an algebra one denotes by \(Id(A)\) the ideal of all polynomial identities of \(A\) in the free (nonassociative) algebra. One of the principal numerical invariants attached to a PI algebra \(A\) is its codimension sequence \(c_n(A)\) defined as \(c_n(A)=\dim P_n/(P_n\cap Id(A))\) where \(P_n\) is the vector space of all multilinear polynomials of degree \(n\) in the variables \(x_1,\ldots,x_n\) in the free algebra. Computing the codimension sequence for a given algebra \(A\) is, as a rule, extremely difficult; the codimensions of very few algebras are known. Thus one wants to study the asymptotic behaviour of \(c_n(A)\). If \(A\) is associative and satisfies a polynomial identity, a celebrated result of A. Regev asserts that \(c_n(A)\) is exponentially bounded, that is \(c_n(A)\leq d^n\) for some constant \(d\). Later on Amitsur conjectured that again in the associative case, the limit \(\lim_{n\to\infty} (c_n(A))^{1/n}\), capped the PI exponent \(\exp(A)\) of \(A\), always exists and is an integer. The conjecture was confirmed in 1999 by the authors of the present paper. More precise asymptotics of the codimension sequence of an associative algebra were obtained as well. The best known estimates are due to the authors of the present paper who showed in 2014 that if \(A\) is associative and PI then \(C_1n^kd^n\leq c_n(A)\leq C_2 n^kd^n\) where \(d=\exp(A)\), \(C_1\), \(C_2\) and \(k\) are constants, and \(k\) is a half integer. If \(A\) is associative, Lie or Jordan algebra such that its codimension sequence is polynomially bounded then it is well known that there exist an integer \(k\) and a real number \(q\) such that \(c_n(A)=qn^k+O(k-1)\) (a theorem of Drensky). The codimension sequence of a non-associative algebra may behave not so nicely as in the associative case. Even for Lie and Jordan algebras the codimension sequence of a PI algebra may grow faster than any exponential function. Moreover the PI exponent may not exist. Even if it exists it may be any given real number \(d>1\). The paper under review gives more evidence of such a ``wild'' behaviour of the codimension sequence of a non-associative algebra. The authors give in the present paper a construction of a family of algebras \(B_a\), for every real number \(a>3\) such that the codimension sequence of \(B_a\) is polynomially bounded and \(a\leq\limsup\log_n(c_n(B_a))\leq a+1\) but \(\liminf\log_n(c_n(B_a))\leq 3\). Thus the limit \(\lim_{n\to\infty}(\log_n(c_n(B_a)))\) does not exist. Furthermore the authors give an example of an algebra \(B\) such that \(\exp(B)\) exists and equals 2, and for every \(k\geq 1\) there exist infinitely many integers \(n\) with \(c_n(B)>n^k2^n\). It follows that \(\limsup\log_n (c_n(B)/\exp(B)^n))=\infty\).