Hoffmann's conjecture for totally singular forms of prime degree (Q738833)

In the present paper the author continues his study of so-called quasilinear \(p\)-forms that he began in [Compos. Math. 149, No. 3, 333--355 (2013; Zbl 1315.11025); J. Reine Angew. Math. 713, 49--83 (2016; Zbl 1343.11047)] building on work by the reviewer [Contemp. Math. 344, 135--183 (2004; Zbl 1074.11023)]. Let \(F\) be a field of characteristic \(p>0\). A quasilinear \(p\)-form (or \(p\)-form for short) on a finite-dimensional \(F\)-vector space \(V\) is a map \(\phi:V\to F\) satisfying \(\phi(\lambda v)=\lambda^p\phi(v)\) and \(\phi(v+w)=\phi(v)+\phi(w)\) for all \(\lambda\in F\) and all \(v,w\in V\). Isomorphism, isotropy, subform, etc., are defined in analogy to the theory of quadratic forms. Up to isomorphism, nonzero such forms a classified by their dimension and their value set \(D(\phi)=\{\phi(v)\mid v\in V\}\) which is an \(F^p\)-vector subspace of \(F\) of dimension \(\leq\dim(\phi)=\dim V\), and one defines the defect index \(i_0(\phi):=\dim V-\dim_{F^p}D(\phi)\). For \(p=2\), the theory of \(p\)-forms is in fact the theory of totally singular quadratic forms. In analogy to Pfister forms in the theory of bilinear and quadratic forms over fields, one can define quasi-Pfister forms which are \(p\)-forms \(\pi\) of dimension \(p^s\) for some \(s\geq 0\) and such that \(D(\pi)\) is a field, in which case \(\pi\) is called an \(s\)-fold quasi-Pfister form. A quasi-Pfister neighbor \(\phi\) is a \(p\)-form that is similar to a subform of some quasi-Pfister form \(\pi\) such that \(p\dim\phi>\dim\pi\). To each \(p\)-form \(\phi\), one obtains a projective hypersurface \(X_{\phi}\subset \mathbb P(V)\) given by the equation \(\phi =0\). \(X_{\phi}\) is integral if the anisotropic part of \(\phi\) has dimension \(\geq 2\), in which case one defines the function field of the \(p\)-form \(F(\phi):=F(X_{\phi})\). In analogy to Knebusch's generic splitting tower for (nonsingular) quadratic forms, one can define the Knebusch splitting tower of a \(p\)-form \(\phi\) by putting \(F_0:=F\) and \(\phi_0:=\phi_{\mathrm{an}}\), its anisotropic part. If \(\dim\phi_0\geq 2\), then \(\phi_0\) becomes isotropic over \(F_1:=F_0(\phi_0)\) and one puts \(\phi_1:=((\phi_0)_{F_1})_{\mathrm{an}}\). Continuing like this, one obtains a field tower \(F_0\subset F_1\subset\ldots \subset F_h\) and anisotropic forms \(\phi_i\) over \(F_i\) with \(h\) being the first index for which \(\dim\phi_h=1\). \(h\) is called the height of \(\phi\) and the Knebusch splitting pattern is given by the sequence of the values \(i_r(\phi)=i_0((\phi_{r-1})_{F_r})\), \(1\leq r\leq h\). In the present paper, the author answers various questions raised in the above mentioned papers concerning the splitting pattern and the first index \(i_1(\phi)\). In his paper [Zbl 1343.11047], he characterized quasi-Pfister neighbors by their Knebusch splitting pattern. The main theorem (Theorem 5.1) of the present paper deals with the case of non-neighbors. It states that if \(\phi\) is a \(p\)-form of dimension \(>2\) over \(F\) that is not a quasi-Pfister neighbor, and if \(s\) is the smallest integer with \(p^s\geq i_1(\phi)\), then \(\phi_1\) is divisible by an \(s\)-fold quasi-Pfister form. From this, the author derives a complete description of the possible Knebusch splitting patterns of \(p\)-forms. In the case \(p=2\), i.e., for totally singular quadratic forms, this implies that \(i_1(\phi)-1\) is the remainder modulo \(2^s\) of \(\dim\phi -1\) for some \(s>\log_2(\dim\phi)\). It should be noted that for (nonsingular) quadratic forms in characteristic not \(2\) and with \(i_1\) denoting the first Witt index, this latter statement was first conjectured by the reviewer and fully proved by \textit{N. A. Karpenko} [Invent. Math. 153, No. 2, 455--462 (2003; Zbl 1032.11016)]. It is still open if this result also holds in characteristic \(2\) in all generality for quadratic forms that are not totally singular. A further application concerns the canonical dimension \(\mathrm{cdim}(X)\) of a \(p\)-hypersurface \(X\), where \(\mathrm{cdim}(X)\) is defined to be the minimum dimension of the image of a rational map from \(X\) to itself. It is shown that there exists an integer \(s\geq 0\) such that \(p^s-1\leq\dim X<p^s+\mathrm{cdim}(X)\) and that \(p^s\) divides \(\mathrm{cdim}(X)+1\). Another application concerns maximal splitting. Let \(\phi\) be an anisotropic \(p\)-form with \(p^n<\dim\phi \leq p^{n+1}\). Then one always has \(i_1(\phi)\leq \dim\phi-p^n\), and \(\phi\) is said to have maximal splitting if equality holds. Quasi-Pfister neighbors always have maximal splitting. The author shows that the converse also holds provided \(p^n+p^{n-1}<\dim\phi\leq p^{n+1}\) for \(p>2\) and \(2^n+2^{n-2}<\dim\phi\leq 2^{n+1}\) for \(p=2\), thus solving a conjecture by the reviewer [loc. cit.].