16 Mikhail Borovoi and Cyril DemarcheThen we get a commutative diagramH 1 × Z m′′ Z(16)H × Ym Ywhere the map m ′′ still satisfies <strong>for</strong>mula (15) and now, <strong>for</strong> all z ∈ Z, we havem ′′ (e, z) = z. (17)We wish <<strong>strong</strong>>to</<strong>strong</strong>> prove that m ′′ is a left group action of H 1 on Z.Since m: H × Y → Y is a left action, we haveτ(m ′′ (h 1 h 2 , z)) = τ(m ′′ (h 1 , m ′′ (h 2 , z))) <strong>for</strong> h 1 , h 2 ∈ H 1 , z ∈ Z,where τ : Z → Y is the canonical map. Since τ : Z → Y is a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under G m ,there is a canonical mapZ × Y Z → G m ,(z 1 , z 2 ) ↦→ z 1 z −12 .We obtain a morphism of k-varietiessuch thatϕ: H 1 × H 1 × Z → G m (h 1 , h 2 , z) ↦→ ϕ z (h 1 , h 2 )m ′′ (h 1 h 2 , z) = ϕ z (h 1 , h 2 ).m ′′ (h 1 , m ′′ (h 2 , z)) <strong>for</strong> h 1 , h 2 ∈ H 1 , z ∈ Z.Then (17) implies thatϕ z (h, e) = 1 and ϕ z (e, h) = 1.By Rosenlicht’s lemma (see [37], Theorem 3, see also [38], Lemma 6.5), the mapϕ has <<strong>strong</strong>>to</<strong>strong</strong>> be trivial, i.e. ϕ z (h 1 , h 2 ) = 1 <strong>for</strong> all z, h 1 , h 2 . There<strong>for</strong>e we havem ′′ (h 1 h 2 , z) = m ′′ (h 1 , m ′′ (h 2 , z)) . (18)Formulas (17) and (18) show that m ′′ is a left group action of H 1 on Z. Sincem ′′ satisfies (15), we have <strong>for</strong> t ∈ G m , z ∈ Zm ′′ (te, z) = t.m ′′ (e, z) = t.z,hence the action m ′′ extends the action of G m on Z. From diagram (16) with m ′′instead of m ′ we see that the action m ′′ induces the action m of H on Y .
<<strong>strong</strong>>Manin</<strong>strong</strong>> <<strong>strong</strong>>obstruction</<strong>strong</strong>> <<strong>strong</strong>>to</<strong>strong</strong>> <strong>strong</strong> <strong>approximation</strong> <strong>for</strong> <strong>homogeneous</strong> <strong>spaces</strong> 17Consider the following commutative diagram (see (16)):H 1 × k Z φ Z Z × X Z(19)H × k Yφ Y Y × X Y ,where φ Z (h 1 , z) := (m ′′ (h 1 , z), z), φ Y (h, y) := (m(h, y), y), and the unnamed morphismsare the natural ones. Since Y → X is a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under H, the morphism φ Yis an isomorphism. The group G m × G m acts on H 1 × k Z via (t 1 , t 2 ).(h 1 , z) :=(t 1 h 1 , t 2 .z), making H 1 × k Z → H × k Y in<<strong>strong</strong>>to</<strong>strong</strong>> a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under G m × G m . We definean action of G m × G m on Z × X Z by(t 1 , t 2 ).(z 1 , z 2 ) := ((t 1 t 2 ).z 1 , t 2 .z 2 ),then Z × X Z → Y × X Y is a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under G m × G m . By <strong>for</strong>mula (15) the map φ Zin (19) is a morphism of <<strong>strong</strong>>to</<strong>strong</strong>>rsors under G m × G m compatible with the isomorphismφ Y of k-varieties. There<strong>for</strong>e the map φ Z is an isomorphism of k-varieties, whichproves that the action m ′′ makes Z → X in<<strong>strong</strong>>to</<strong>strong</strong>> a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under H 1 .Let us prove the uniqueness of the class of the extension H 1 . If H 2 is a centralextension of H that satisfies the conditions of the lemma, then the analogues ofdiagram (16) and <strong>for</strong>mula (15) with H 2 instead of H 1 define an isomorphism ofH × Y -<<strong>strong</strong>>to</<strong>strong</strong>>rsors under G m between the push-<strong>for</strong>ward of H 2 × Z by the morphismG m × G m → G m and the <<strong>strong</strong>>to</<strong>strong</strong>>rsor m ∗ Z. There<strong>for</strong>e, we getm ∗ [Z] = p ∗ H[H 2 ] + p ∗ Y [Z].Comparing with (13), we see that p ∗ H [H 2] = p ∗ H [H 1]. Since p ∗ H + p∗ Y : Pic(H) ⊕Pic(Y ) → Pic(H ×Y ) is an isomorphism, we see that p ∗ H : Pic(H) → Pic(H ×Y ) isan embedding, hence [H 2 ] = [H 1 ], which completes the proof of Lemma 2.13.2.15. Proof of Theorem 2.8: Top row of the diagram. First we prove that the<<strong>strong</strong>>to</<strong>strong</strong>>p row in diagram (8) is a complex. Let p ∈ Pic(Y ) and let us prove that∆ Y/X (ϕ 1 (p)) = 0. Let Z → Y be a <<strong>strong</strong>>to</<strong>strong</strong>>rsor under G m such that [Z] = p ∈ Pic(Y ).Let p ′ := ϕ 1 (p) ∈ Pic(H), and 1 → G m → H 1 → H → 1 be a central extensionof H by G m corresponding <<strong>strong</strong>>to</<strong>strong</strong>> p ′ via the isomorphism Ext c k(H, G m ) ∼ = Pic(H).Then ∆ Y/X (p ′ ) is equal (by definition) <<strong>strong</strong>>to</<strong>strong</strong>> ∂ H1 ([Y ]) ∈ H 2 (X, G m ), where ∂ H1 :H 1 (X, H) → H 2 (X, G m ) is the coboundary map coming from the extension H 1 ,and [Y ] is the class of the <<strong>strong</strong>>to</<strong>strong</strong>>rsor Y → X in H 1 (X, H).Lemma 2.13 implies that the class [Y ] ∈ H 1 (X, H) is in the image of the mapH 1 (X, H 1 ) → H 1 (X, H). From exact sequence (7) we see that the class ∂ H1 ([Y ])is trivial in H 2 (X, G m ), i.e. ∆ Y/X (p ′ ) = ∂ H1 ([Y ]) = 0 ∈ H 2 (X, G m ), hencePic(Y ) −→ j∗Pic(H) ∆ Y /X−−−−→ Br(X) is a complex.