Power-central polynomials on matrices. (Q908322)

Let \(K\) be an infinite field of characteristic either 0 or prime to \(n!\), and denote by \(R=M_n(K)\) the full matrix algebra of size \(n\) over \(K\). For a polynomial in the free associative algebra \(p(x_1,\ldots,x_m)\) one denotes by \(\mathrm{Im}(p)\) the set \(\{p(a_1,\ldots,a_m)\mid a_i\in R\}\). A long-standing problem due to Kaplansky asks whether for \(p\) multilinear, \(\mathrm{Im}(p)\) can be only among 0, \(K\), \(sl_n(K)\) or \(M_n(K)\). In the first case \(p\) is a polynomial identity for \(R\), in the second it is a central polynomial. The authors prove that if \(p\) is multilinear, and is neither an identity nor a central polynomial then \(\mathrm{Im}(p)\) contains an element \(\alpha_ne_{n1}+\sum_{i=1}^{n-1}\alpha_ie_{i,i+1}\). Thus if one considers the Zariski closure of \(\mathrm{Im}(p)\) one has that for \(p\) neither PI nor central \(\mathrm{Im}(p)\) is at least \(n^2-n+2\)-dimensional. A polynomial is \(v\)-central if \(p^v\) is central for \(R\) and \(v\) is the least positive integer with this property. It is power-central if it is \(v\)-central for some \(v>1\). It was known in the 19-th century that \([x_1,x_2]\) is 2-central for \(M_2(K)\). It is also known that for \(n\geq 3\), if \(p\) is multilinear and \(K\) contains no \(v\)-th roots of 1 then \(p\) cannot be \(v\)-central. The authors prove a stronger result: they show that for \(n\geq 4\) there are no multilinear power-central polynomials. Moreover if \(n=3\) then necessarily \(v=3\) (but the case is still open). The authors establish additional properties of \(v\)-central polynomials. Namely they prove that if \(K\) is the field of the rationals then there are no \(v\)-central polynomials for \(M_n(K)\) apart from the case when \(v\) is a product of distinct odd primes and \(n=mv\) where \(m\) and \(v\) are coprime. Moreover when \(n\) is even the authors show there do not exist 4-central polynomials.