Some remarks concerning Voevodsky's nilpotence conjecture (Q1747191)

Voevodsky introduced the smash-nilpotency equivalence relation \(\sim_{\otimes\text{nil}}\) on algebraic cycles and conjectured that it agrees with the ordinary numerical equivalence relation \(\sim_{\text{num}}\). The nilpotency conjecture states explicitly that for a smooth projective \(k\)-scheme \(X\), the following holds true. \(V(X):\;\mathcal Z^\ast(X)_F/\sim_{\otimes\text{nil}}=Z^\ast(X)_F/\sim_{\text{num}}\) where \(k\) is a base field and \(F\) a field of coefficients of characteristic \(0\). The conjecture has been proved for curves, surfaces, and abelian 3-folds (when \(k\) is of characteristic \(0\)). The main objective in this article is to apply noncommutative geometry to prove a generalized conjecture, specializing to more cases in commutative geometry. A dg category \(\mathcal A\) is called smooth if it is perfect as a bimodule over itself, and proper if for any two objects \(x\) and \(y\) in \(\mathcal A\), \(\sum_i\dim H^i\mathcal A(x,y)<\infty.\) The Grothendieck group \(K_0(\mathcal A)\) of every smooth proper dg category \(\mathcal A\) has a \(\otimes\)-nilpotency equivalence relation \(\sim_{\otimes\text{nil}}\) and a numerical equivalence relation \(\sim_{\text{num}.}\) This is the noncommutative setting of the conjecture, which can be stated as \(V_{NC}(\mathcal A):\) \(K_0(\mathcal A)_F/\sim_{\otimes\text{nil}}=K_0(\mathcal A)_F/\sim_{\text{num}}\). The first main result is the reformulation of Voevodsky's conjecture: The conjecture \(V(X)\) is equivalent to Conjecture \(V_{NC}(\text{perf}_{\text{dg}}(X)).\) The authors apply this result to restrict to the following commutative problems. Quadric fibrations. Let \(S\) be a smooth projective \(k\)-scheme and let \(q:A\rightarrow S\) be a flat quadric fibration of relative dimension \(n\) with \(Q\) smooth. Let \(\mathcal C_0\) be the even part of the Clifford algebra associated to \(q\). When the discriminant divisor of \(q\) is smooth and \(n\) is even (resp. odd), there is a discriminant double cover \(\tilde S\rightarrow S\) (resp. a square root stack \(\hat S\)) equipped with an Azumaya algebra \(\mathcal B_0\). The second main result of the article makes it possible to decompose the Conjecture \(V(Q)\) into simpler pieces: (i) \(V(Q)\Leftrightarrow V_{\text{NC}}(\text{perf}_{\text{dg}}(S,\mathcal C_0))+V(S),\) (ii) When the discriminant divisor of \(q\) is smooth and \(n\) is even, \(V(Q)\Leftrightarrow V(\tilde S)+V(S),\) (iii) When the discriminant divisor of \(q\) is smooth and \(n\) is odd, we have \(V(Q)\Leftrightarrow V_{\text{NC}}(\text{perf}_{\text{dg}}(\tilde S,\mathcal B_0))+V(S).\) As a consequence, conjecture \(V(Q)\) becomes equivalent to \(V_{\text{NC}}(\text{perf}_{\text{dg}}(\hat S,\mathcal B_0))\) when \(\dim(S)\leq 2\), and this last conjecture is true for \(\dim(S)\leq 2.\) In the case of intersection of quadrics, let \(X\) be a smooth complete intersection of \(r\) quadric hypersurfaces in \(\mathbb P^m\). The linear span of these gives a hypersurface \(Q\subset\mathbb P^{r-1}\times\mathbb P^m\), and the projection to the first factor is a flat quadric fibration \(q:Q\rightarrow\mathbb P^{r-1}\) of relative dimension \(m-1.\) The authors prove that (i) \(V(X)\Leftrightarrow V_{NC}(\text{perf}_{\text{dg}}(\mathbb P^{r-1},\mathcal C_0)),\) (ii) when the discriminant divisor of \(q\) is smooth and \(m\) is odd, \(V(X)\Leftrightarrow V(\widetilde{\mathbb P^{r-1}}),\) implying that \(V(X)\) holds when \(r\leq 3,\) and (iii) when the discriminant divisor of \(q\) is smooth and \(m\) is even, \(V(X)\Leftrightarrow V_{NC}(\text{perf}_{\text{dg}}(\widehat{\mathbb P^{r-1}},\mathcal B_0)).\) The last conjecture holds when \(r\leq 3\) and \(k\) is algebraically closed. In the case of linear sections of Grassmannians: (I) Let \(X_L\) be a generic linear section of codimension \(r\) of the Grassmannian \(\text{Gr}(2,W)\) with \(W=k^6\), under the Plücker embedding, and let \(Y_L\) be the corresponding dual linear section of the cubic Pfaffian \(\text{Pf}(4,W^\ast)\subset\mathbb P(\bigwedge^2 W^\ast).\) (II) Let \(X_L\) be a generic linear section of codimension \(r\) of the Grassmannian \(\text{Gr}(2,W)\), \(W=k^7\), under the Plücker embedding, let \(Y_L\) be the corresponding dual linear section of the cubic Pfaffian \(\text{Pf}(4,W^\ast)\subset\mathbb P(\bigwedge^2 W^\ast).\) In this case, the following is proved: When \(X_L\) and \(Y_L\) are smooth, \(V(X_L)\Leftrightarrow V(Y_L).\) This conjecture holds when \(r\leq 6\) for class (I) and when \(r\leq 6\) and \(8\leq r\leq 10\) for class (II). This result proves Voevodsky's nilpotency conjecture in new cases. In the case of linear sections of determinantal varieties, let \(U\) and \(V\) be \(k\)-vector spaces of dimensions \(m\leq n\) respectively and \(0<r<m\) an integer. Let \(Z_{m,n}^r\subset\mathbb P(U\otimes V)\) be the universal determinantal variety defined by the locus of matrices \(M:U^\ast\rightarrow V\) of rank at most \(r\). Let \(\mathcal X_{m,n}^r=\mathbb P(\mathcal U\otimes V)\rightarrow\text{Gr}(r,U)\), \(\mathcal U\) the tautological subbundle on \(\text{Gr}(r,U)\), be the Springer resolution. The classes of schemes considered in this situation are \(X_L\), a generic section of codimension \(c\) of \(\mathcal X_{m,n}^r\) under the map \(\mathcal X^r_{m,n}\rightarrow\mathbb P(U\otimes V),\) and \(Y_L\), the corresponding dual linear section of \(\mathcal X^{m-r}_{m,n}\) under the map \(\mathcal X^{m-r}_{m,n}\rightarrow\mathbb P(U^\ast\otimes V^\ast).\) The article list which special cases this includes for various values of \(m,n,r,\) and the main result in this case is that \(V(X_L)\Leftrightarrow V(Y_L)\) and that this conjecture holds when \(\dim(X_L)\leq 2\) or \(\dim(Y_L)\leq 2\). For homological projective duality, a generalization of the result for linear sections of Grassmannians is presented: For \(X\) smooth, projective, with ample line bundle \(\mathcal O(1)\) and \(Y\rightarrow\mathbb P^1(V^\ast)\) the associated morphism, assume there is a Lefschetz decomposition \(\text{Perf}(X)=\langle\mathbb A_0,\dots,\mathbb A_n\rangle\) with respect to \(\mathcal O_X(1)\). Also assume \(V_{\text{NC}}(\mathbb A_0^{\text{dg}})\) to be true. Let \(L\subset V\) be a subspace such that the linear sections \(X_L=X\times_{\mathbb P(V)}\mathbb P(L)\) and \(Y_L=Y\times_{\mathbb P(V^\ast)}\mathbb P(L^\bot)\) are of the expected dimensions. Then under the assumption that \(X_L\), \(Y_L\) are smooth, \(V(X_L)\Leftrightarrow V(Y_L).\) The final case studied in this article, is the result for Moishezon Manifolds. Artin and Mumford give bundles \(Y\rightarrow\mathbb P^2\) as examples of unirational varieties that are not rational. Letting \(X\rightarrow Y\) be a small resolution, \(X\) is a smooth Moishezon manifold. The result in this case is that the conjecture \(V_{\text{NC}}(\text{perf}_{\text{dg}}(X))\) holds for the above resolutions. The article is well written, highly advanced, and makes no attempt to be self contained on the algebraic geometric part (as the title indicates). It includes the necessary categorical definitions, and the proofs are detailed and precise.