On a question concerning generic matrices over the integers (Q1065904)

Let \(\{X_ 1,X_ 2,...,X_ n\}\) be the generic matrices over the ring R. Consider the natural homomorphism \(\phi\) : \(Z\{\) \(X_ 1,...,X_ m\}\to Z/pZ\{X_ 1,...,X_ m\}\). The problem considered is whether the kernel K of \(\phi\) is precisely \(pZ\{X_ 1,...,X_ m\}\). (This problem was raised by Procesi). It is known to be true for \(m=2\) and \(2\times 2\) matrices (a result of Formanek). In the paper it is shown, by exhibiting a specific polynomial of degree 6 for \(m=5\) and \(2\times 2\) matrices, that the result does not hold in general. The polynomial is obtained by taking the commutator G of \(F=\sum_{\sigma \in S_ n}(- 1)^{\sigma}X_{\sigma_ 1}YX_{\sigma_ 2}X_{\sigma_ 3}X_{\sigma_ 4}\). The latter polynomial F is central in characteristic 2 but not in characteristic 0. Therefore \(G=[F,X_ 1]\) is zero only in characteristic 2.