On symmetric operators of higher degree and their applications (Q1076767)

Let \(\phi\) be a non-degenerate k-linear form on an n-dimensional complex vector space \(L_ n\). A linear transformation \(A:L_ n\to L_ n\) is called symmetric with respect to \(\phi\) if \(\phi (Ax_ 1,x_ 2,...,x_ k)=\phi (x_ 1,Ax_ 2,...,x_ k)\) for all \(x_ 1,...,x_ k\in L_ n\). Assume that \(k\geq 3\). Then the symmetric transformations with respect to \(\phi\) form a commutative algebra \(\Omega\). In many cases \(\Omega\) will be trivial, consisting merely of the scalar multiples of the identity. For the purposes of the present review, let us call \(\phi\) good if \(\Omega\) is non-trivial, excellent if \(\Omega\) contains an element A whose n eigenvalues are all different. The author shows that \(\phi\) is excellent if, and only if, \(L_ n\) has a basis \(f_ 1,...,f_ n\) such that \(\phi (f_{i_ l},...,f_{i_ k})=0\) unless \(i_ 1=...=i_ k\); and, in the case \(n=2\), he determines the precise conditions for \(\phi\) to be good. He also gives a constructive solution to the problem of extending a given \(\phi\) on \(L_ 2\) to an excellent \({\tilde \phi}\) on some \(L_ r\supseteq L_ 2\), the dimension r being kept as small as possible. This problem (for a k-linear \(\phi)\) is closely related to that of expressing a given complex polynomial \(P(t)=\sum^{k}_{j=0}a_ j\left( \begin{matrix} k\\ j\end{matrix} \right)t^{k-j}\) of degree at most k as a sum of r kth powers of linear polynomials. The author proves the following interesting Theorem. If \[ (*)\quad P(t)=\sum^{r}_{i=1}\mu_ i(t+\lambda_ i)^ k,\quad r\leq (k+1), \] where all \(\mu_ i\neq 0\) and all \(\lambda_ i\) are distinct, then the matrix \[ \begin{pmatrix} a_ 0 & a_ 1 &... & a_{k-r} \\ a_ 1 & a_ 2&... & a_{k-r+1} \\ \vdots &&&\vdots \\ a_ r & a_{r+1} &... & a_ k \end{pmatrix} \] has rank r and the determinant \[ \left|\begin{matrix} 1 & \lambda & \lambda^ 2 &... & \lambda^ r \\ a_ 0 & a_ 1 & a_ 2 &... & a_ r \\ a_ 1 & a_ 2 & a_ 3 &... & a_{r+1} & \\ \vdots & \vdots & \vdots && \vdots \\ a_{r-1} & a_ r & a_{r+1} &... & a_{2r-1} \end{matrix}\right| \] is a polynomial of degree r with distinct roots \(\lambda_ 1, ..., \lambda_ r\). Conversely, if these two conditions hold, then P(t) has a representation (*). The author points out that the special case \(k=3\), \(r=2\) yields Cardan's solution of the cubic.