Identifiability of homogeneous polynomials and Cremona transformations (Q2009900): Difference between revisions

Let \(F\) be a generic homogeneous polynomial in \(\mathbb C[x_0,\dots,x_n]_d\). The minimal value for which we can write \(F = L_1^{d_1}+\dots+L_h^{d_h}\), where the \(L_i\)'s are linear forms is called the Waring rank of \(F\) (or also its symmetric rank, since \(F\) can be viewed as a symmetric tendor). When such decomposition of \(F\) is unique, \(F\) is said to be \(h\)-identifiable and \(L_1^{d_1}+\dots+L_h^{d_h}\) is said to be a canonical form for \(F\). It is classically known that \(F\) is \(s\)-identifiable in the three following cases: \((n,d,s) = (1,2k-1,k)\), \((n,d,s) = (3,3,5)\), \((n,d,s) = (2,5,7)\); and it has been expected that those are the only cases for which a generic \(F\in \mathbb C[x_0,\dots,x_n]_d\) is identifiable, but there were only partial results about this conjecture: this paper gives eventually an affermative answer to this long-standing conjecture. The proof is a quite geometric one, since this problem relates to the properties of projections of Veronese varieties and in turn this is related to properties of linear systems in \(\mathbb P^n\) with fixed singularities. A decisive step in the proof is the following theorem (which is of interest also ``per se''): Theorem. Let \(\mathcal{L}_{n,d}(2^h)\) be the linear system of forms in \(\mathbb C[x_0,\dots,x_n]_d\) with \(h\) multiple points in general position in \(\mathbb P^n\) and let \(\varphi_{n,d,h}\) be its associated rational map. Then \(\varphi_{n,d,h}\) is a Cremona transform (i.e. dim\(\mathcal{L}_{n,d}(2^h)=n\) and \(\varphi_{n,d,h}:\mathbb P^n - - \rightarrow \mathbb P^n\) is birational) if and only if either \((n,d,h) = (1,2k+1,k)\), \((n,d,s) = (3,3,4)\), \((n,d,s) = (2,5,6)\). The proof of the main theorem also uses Alexander-Hirschowitz' results on the regularity of the linear systems \(\mathcal{L}_{n,d}(2^h)\) and the partial results which are known for the regularity of the linear systems \(\mathcal{L}_{2,d}(m_1,\dots,m_s)\) described by a conjecture that goes back to Beniamino Segre (1961).