The \(\gamma\)-filtration and the Rost invariant (Q475834)

Let \(K_0(X)\) be the Grothendieck ring of coherent sheaves on a projective variety. \(K_0(X)\) has two filtrations , the \(\gamma\)-filtration \(\gamma_0\supseteq\gamma^1\supseteq\dots\) and the topological filtration \(\tau^0\supseteq\tau^1\supseteq\dots\) There are homomorphisms between the corresponding graded rings \(\bigoplus_{i\geq 0}\gamma^{i/i+1}\), \(\bigoplus_{i\geq 0}\tau^{i/i+1}\) and the Chow ring \(CH(X)\), which by Riemann-Roch become isomorphisms after tensoring with \(\mathbb Q\). Therefore, in order to compare these rings, one should examine their torsion subgroups. In the case \(X\) is the variety of Borel subgroups of a semisimple linear algebraic group \(G\) over an algebraically closed field \(k\), then, by a result of Grothendieck, \(\bigoplus_{i\geq 0}\tau^{i/i+1}\) and \(CH(X)\) are torsion free and isomorphic but \(\bigoplus_{i\geq 0}\gamma^{i/i+1}\) is not always torsion-free. More precisely \(\bigoplus_{i\geq 0}\gamma^{i/i+1}\) is torsion free if and only if the simply connected cover of \(G\) is isomorphic to a product of copies of \(\mathrm{SL}_n\) and \(\mathrm{Sp}_{2n}\), for various \(n\). The goal of this paper is to make Grothendieck's result more precise by calculating the torsion in some cases, by easing the hypotheses on \(G\) and \(k\). If \(k\) is arbitrary and \(G\) is split i.e \(G\) contains a maximal \(k\)-torus that is split over \(k\), one knows that \(\gamma^{0/1}\simeq \mathrm{CH}^0(X)\simeq\mathbb Z\) and \(\gamma^{1/2}\simeq \mathrm{CH}^1(X)\simeq\mathrm{Pic }X\), so these components are torsion free. Therefore the next case is \(\gamma^{2/3}\), for which the authors prove the following. Theorem 1. Let \(G\) be a split simple linear algebraic group over a field \(k\). Then the torsion subgroup of \(\gamma^{2/3}\) for the variety of Borel subgroups of \(G\) is \(\mathbb Z/N(G)\mathbb Z\), for \(N(G)\) the Dynkin index of \(GI\). In the case \(k\) is algebraically closed one gets the following corollary: Corollary 1. Let \(G\) be a semi simple linear algebraic group over an algebraically closed field \(k\). Then \(\bigoplus_{i\geq 0}\gamma^{i/i+1}\) is torsion free if and only if \(\gamma^{2/3}\) is torsion free, if and only if the simply connected cover of \(G\) is isomorphic to a product of copies of \(\mathrm{SL}_n\) and \(\mathrm{Sp}_{2n}\), for various \(n\). As far as the Chow ring is concerned, then \(\mathrm{CH}^0(X)\) and \(\mathrm{CH}^1(X)\) are known and the torsion in \(\mathrm{CH}^2(X)\) is known by the theory of the Rost invariant. However little is known about \(\mathrm{CH}^3(X)\) in general, not even if it is finitely generated. The paper contains a result to bound the torsion in \(\mathrm{CH}^3(X)\), which applies in particular if \(G\) is versal: the torsion part of \(\mathrm{CH}^3(X)\) can only consists of subgroups \(\mathbb Z/2^s\mathbb Z\) for \(s\leq 4\), \(\mathbb Z/3\mathbb Z\) or \(\mathbb Z/5\mathbb Z\).