Rost nilpotence for twisted Milnor hypersurfaces

Indeed, for any \psi\in \operatorname{End} (M), we have \psi=\psi_1\oplus\ldots \oplus\psi_r, where each \psi_i\in \operatorname{End}(M_i). Hence, we may assume M is the motive of some 0-dimensional variety X. But the restriction morphism res_{l/k} is known to be injective in this case (the Chow group of X is free).

Following the proof of [VZ08, Prop.3.1] the restriction morphism factors as the composite

where the middle term denotes endomorphisms in the category of Chow motives over the relative base X, the first map is injective, and the second map is induced by passing to the generic point of X. Observe that the kernel of res_{k(X)/X} is the nilpotent ideal by the proof of [VZ08, Lem.3.2] extended in [Gi25, §3.5].

Now the kernel of the restriction res_{l/k}\colon \operatorname{End}_k(\operatorname{M}(X)) \to \operatorname{End}_l(\operatorname{M}(X_l)) is contained in the kernel of res_{F/k}, where F contains both l and k(X), so is nilpotent.

From [XZ24, §2] it follows that the motive \operatorname{M}(Y_A) splits as

where the motive N is supported only in the middle codimension (over the algebraic closure of k, N splits as a direct sum of Tate motives twisted by n-2).

Since \operatorname{M}(Y_A) is supported in dimension zero only by the motive of a Severi-Brauer variety \operatorname{M}(\operatorname{SB}(A)), the latter must contain a Tate motive (in dimension 0) over the generic point k(Y_A). Therefore, by [Ka96] the algebra A splits over k(Y_A). Hence the variety Y_A turns over k(Y_A) into the variety Y, and the strong Rost Nilpotence holds for Y_A generically. To finish we apply Lemma 3.

Let U be the upper motive of \operatorname{M}(Y_A; \mathbb{F}_p). It is isomorphic to the indecomposable motive \operatorname{M}(\operatorname{SB}(D_p);\mathbb{F}_p) up to a shift. Observe that \dim(\operatorname{SB}(D_p))=d-1. Being an indecomposable lower summand, the dual motive U^* of U is isomorphic to U(\dim(Y_A)-d-1) by the Krull-Schmidt property (see [Ka10, Prop.5.2]). By [Ka10, Thm.5.1], the canonical p-dimension of Y_A is then equal to the dimension of U that is d-1.