Multiplicities in the trace cocharacter sequence of two \(4\times 4\) matrices. (Q861968)

The ring of generic matrices is one of the most important and extensively studied objects in PI theory. Let \(X_1,\dots,X_d\) be \(d\) generic \(n\times n\) matrices over a field \(K\) of characteristic 0. (This means that the entries of these matrices are \(dn^2\) independent commuting variables.) Let \(C=C_d\) be the (commutative) \(K\)-algebra generated by the traces of all products \(X_{i_1}\cdots X_{i_k}\), \(k\geq 1\); \(C\) is called the pure trace algebra. One identifies the elements of \(C\) with the corresponding scalar matrices and considers the \(K\)-algebra \(T=T_d\) generated by \(C\) and by \(X_1,\dots,X_d\). The latter is called the mixed trace algebra. It is well known that \(C\) is the algebra of invariants of the general linear group \(\text{GL}_n(K)\) acting on \(d\) \(n\times n\) matrices by simultaneous conjugation. On the other hand \(T\) is the algebra of matrix concomitants. Since both \(C\) and \(T\) are graded algebras (by the multidegree) one is led to consider their Hilbert series \(H(C)\) and \(H(T)\). Write \(H(C)=\sum m_\lambda(C)S_\lambda(t_1,\dots,t_d)\) where \(S_\lambda\) are the Schur functions and \(m_\lambda(C)\) are their multiplicities, and analogously for \(H(T)\). Here the partitions \(\lambda\) have at most \(d\) parts. The nonnegative integers \(m_\lambda(C)\) and \(m_\lambda(T)\) are of significant importance both in ring theory and in combinatorics. They coincide with the multiplicities of the irreducible \(\text{Sym}_k\)-cocharacters in the pure trace and mixed trace cases. Thus their knowledge could provide us with information about the ordinary cocharacters of the \(n\times n\) matrices. However, the multiplicities \(m_\lambda(C)\) and \(m_\lambda(T)\) are known only in very few special cases: when \(n=2\), any \(d\); \(n=3\) and \(d\leq 3\), \(n=4\) and \(d=2\). The main contribution of the paper under review consists in presenting an explicit form of the generating functions for \(m_\lambda(C)\) and \(m_\lambda(T)\) when \(n=4\), \(d=2\) and \(\lambda=(\lambda_1,\lambda_2)\) is a partition of two parts. The formulas are quite involved; the authors give the asymptotics of \(m_\lambda(C)\) and \(m_\lambda(T)\). It turns out that \(m_\lambda(C)\) and \(m_\lambda(T)\) behave like polynomials of degree 14 in \(\lambda_1\) and \(\lambda_2\).