Analyse fonctionnelle/Functional analysis
Transcription
Analyse fonctionnelle/Functional analysis
C. R. Acad. Sci. Paris, t. ???, Série I, p. 1–??, 2002 Analyse fonctionnelle/Functional analysis (/) - PXMA????.TEX - K-duality for pseudomanifolds with an isolated singularity Claire Debord a , Jean-Marie Lescure b , a Laboratoire de Mathématiques et Applications, Université de Haute-Alsace, 4 rue des Frères Lumière, 68093 Mulhouse cedex, France E-mail: [email protected] b Laboratoire de Mathématiques pures, Université Blaise Pascal, Complexe universitaire des Cézeaux, 124 Av. des Landais, 63177 Aubière cedex, France E-mail: [email protected] (Reçu le jour mois année, accepté après révision le jour mois année) Abstract. We associate to a pseudomanifold X with an isolated singularity a differentiable groupoid G which plays the role of the tangent space of X. We construct a Dirac element D and a Dual Dirac element λ which induce a Poincaré duality in K-theory between the C ∗ -algebras c 2002 Académie des sciences/Éditions scientifiques et médicales C(X) and C ∗ (G). Elsevier SAS K-dualité pour les pseudo-variétés à singularité isolée. Résumé. 1. Etant donnée une pseudo-variété X ayant une singularité isolée, nous lui associons un groupoı̈de différentiable G qui joue le rôle d’espace tangent à X. Nous construisons un élément Dirac D ainsi qu’un élément dual-Dirac λ qui induisent une dualité de Poincaré c 2002 Académie des sciences/Édien K-théorie entre les C ∗ -algèbres C(X) et C ∗ (G). tions scientifiques et médicales Elsevier SAS The tangent bundle of a singular manifold Let X be a pseudomanifold with an isolated singularity c, that is X = cL ∪ X1 , where X1 is a smooth compact manifold with boundary L glued along its boundary with the cone over L : cL = L×[0, 1]/L×{0}. The singularity c is then the image of L × {0} in cL. We denote by M = L×] − 1, 1] ∪ X1 the manifold obtained by gluing X1 with L×] − 1, 1] along the boundary. If y is a point of M or of X \ {c} which is in L×] − 1, 1[ we note yL ∈ L its tangential componant and ky ∈] − 1, 1[ its radial coordinate; the function ky is smoothly extended to X1 in such a way that ky > 1 on X1 . We set M + = {y ∈ M | ky > 0}, M − = {y ∈ M | ky < 0} and M + = {y ∈ M | ky > 0}. We suppose that the manifold M is equipped with a Riemannian metric whose injectivity radius is bigger than 1 and which is a product metric on L×] − 1, 1[. Note présentée par First name NAME S0764-4442(00)0????-?/FLA c 2002 Académie des sciences/Éditions scientifiques et médicales Elsevier SAS. Tous droits réservés. 1 C. Debord et J.M. Lescure We define the groupoid G, with source map s , range map r and space of units G(0) = M : s G = M− × M− ∪ TM+ ⇒ M . r It is the union of the bundle T M + = {(z, v) ∈ T M | z ∈ M + } equipped with its standard groupoid structure over M + and the pair groupoid M − × M − ⇒ M − . In order to equip G with a smooth structure, we take the usual structure of manifold on M − × M − and on T M + . We take a smooth positive function τ : ] − 1, +∞[→ R which satisfies τ −1 ({0}) = [0, +∞[. A local chart around boundary points of T M + is provided by the following map : (y, V ) 7→ (y, V ) if y ∈ M + EG : V(T M ) −→ V(G) , (y, V ) 7→ (y, expy (−τ (ky )V )) if y ∈ M − where exp is the exponential map of the Riemannian manifold M , V(T M ) = {(y, V ) ∈ T M ; −τ (ky )V ∈ dom(expy )} and V(G) is a neighborhood of G(0) in G. The groupoid G is called the tangent bundle of X. Following the construction of A. Connes for smooth manifolds [3], we define the tangent groupoid of X in the following way : G = M × M ×]0, 1] ∪ G × {0}⇒M × [0, 1] . The groupoid G is the union of the groupoid G×{0} ⇒ M ×{0} and the pair groupoid over M parametrized by ]0, 1]. We equip G with a structure of smooth groupoid similarly as we did for G. The groupoid G is amenable [1] so that its reduced C ∗ -algebra coincides with the maximal one and it is nuclear. The same occurs for G. We denote respectively C ∗ (G) and C ∗ (G) these C ∗ -algebras. Moreover, up to isomorphisms, these C ∗ -algebras do not depend on the map τ used to define the smooth structure of G. 2. The Dirac element The partition M × {0} ∪ M ×]0, 1] of G (0) into a saturated open subset and a saturated closed subset induces the following exact sequence of C ∗ -algebras [5] : e 0 −−−−→ C ∗ (G|M ×]0,1] ) −−−−→ C ∗ (G) −−−0−→ C ∗ (G|M ×{0} ) = C ∗ (G) −−−−→ 0 where the first map is the inclusion and e0 is the evaluation map at 0. The C ∗ -algebra C ∗ (G|M ×]0,1] ) is isomorphic to K ⊗ C0 (]0, 1]) which is contractible. So, since C ∗ (G) is nuclear, the element [e0 ] of KK(C ∗ (G), C ∗ (G)) corresponding to e0 , is invertible. We denote [e0 ]−1 ∈ KK(C ∗ (G), C ∗ (G)) its inverse. Let e1 : C ∗ (G) → C ∗ (G|M ×{1} ) = K be the evaluation map at 1. We let b be the generator of KK(K, C). We set : ∂ = [e0 ]−1 ⊗ [e1 ] ⊗ b ∈ KK(C ∗ (G), C) . C ∗ (G) K The algebra C(X) maps to the center of the multiplier algebra of C ∗ (G). Let Ψ : C ∗ (G) ⊗ C(X) → C ∗ (G) be the morphism induced by multiplication and [Ψ] be the corresponding element in KK(C ∗ (G) ⊗ C(X), C ∗ (G)). The Dirac element is then defined as D = [Ψ] ⊗ ∂ ∈ KK(C ∗ (G) ⊗ C(X), C) . C ∗ (G) 2 K-duality for pseudomanifolds with an isolated singularity 3. The dual Dirac element We are looking for an element λ in KK(C, C ∗ (G) ⊗ C(X)), that is a continuous family (λy )y∈X of elements of KK(C, C ∗ (G)). In order to be dual to the Dirac element D, λ is constructed so that : (i) λ is in the image of (iO )∗ : KK(C, C ∗ (G × X|O )) → KK(C, C ∗ (G × X)) where O is an open subset of U = {(x, y) ∈ M × X | kx < 1, ky < 1} ∪ {(x, y) ∈ M × X | d(x, y) < 1}. (ii) The equality λ ⊗ ∂ = 1 ∈ K 0 (X) holds. C ∗ (G) We first assign to each point y of X an open subset Oy of M which is a ball centered on y when ky > 1, which is contained in M − when ky 6 1/2 and equal to M − when ky 6 ε (where 0 < ε < 1/2). Furthermore the set O = ∪y∈X Oy × {y} is an open subset of M × X contained in U. Notice that K(C ∗ (G|Oy )) ' Z for each y ∈ X. We pull back the vector bundle of differential forms on M to G and G using their range maps and then to G × X and G × X using the first projection. We denote all these bundles by the same letter Λ. The following step is the construction of a continuous family (βy )y∈X , where βy = (Fy , C ∗ (G|Oy , Λ)) is an element of E(C, C ∗ (G|Oy )) which class is a generator of K(C ∗ (G|Oy )) ' Z. When Oy is a subset of M + it is natural to state Fy = ay where ay ∈ Cb∞ (T ∗ Oy , EndΛ) is the symbol of a pseudo-differential operator Gy which satisfies : • Gy belongs to Ψ0c (Oy , Λ) + Cb∞ (Oy , EndΛ) and G2y − 1 belongs to Ψ−1 c (Oy , Λ) , − 0 Gy • Gy is of the form Gy = with respect to the usual decomposition of Λ = Λev ⊕ Λodd , G+ 0 y + ∞ • G+ y is surjective and Ker(Gy ) = C · y where y belongs to Cb (Oy , Λ), S • the family (Gy ) y∈X defines a continuous section of B(H) where H is the bundle y∈X L2 (Oy , Λ). ky >ε ky >ε The existence of such an operator is a consequence of theorem (19.2.12) of [6]. When y comes closer to the singularity c, we gradually pass from a situation where G|Oy = T Oy to a situation where G|Oy = Oy × Oy . The pseudo-differential calculus on groupoids, first introduced by A. Connes in [2] (see also [4]), enables us to construct a continuous family (Fy )y∈X,ky >ε such that up to a compact operator Fy = ay when G|Oy = T Oy and Fy = Gy when G|Oy = Oy × Oy . Thus we “replace” the symbol ay by the operator Gy . The last case is when Oy becomes equal to M − . We first choose an appropriate extension of Fy which belongs to B(L2 (Oy , Λ)) into an element of B(L2 (M − , Λ)). Afterwards, thanks to the properties of the operator Gy listed above, we can replace Fy by a constant operator Fc . We construct in this way a continuous family βy of elements of E(C, C ∗ (G|Oy )). This family induces an element β of KK(C, C ∗ (G × X|O )). We set λ = (iO )∗ (β). The dual-Dirac element λ satisfies the properties (i) and (ii) mentionned above. 4. The Poincaré duality T HEOREM 4.1. – The Dirac element D and the dual-Dirac element λ induce a Poincaré duality between the C ∗ -algebras C ∗ (G) and C(X), that is : λ ⊗ D = 1C(X) ∈ KK(C(X), C(X)) and λ ⊗ D = 1C ∗ (G) ∈ KK(C ∗ (G), C ∗ (G)). C ∗ (G) C(X) Idea of the proof : Let us consider the morphisms Ψ0 , ∆0 : C ∗ (G × X) ⊗ C(X) → C ∗ (G) ⊗ C(X) given by Ψ0 (f ⊗ g ⊗ h) = Ψ(f ⊗ h) ⊗ g and ∆0 (f ⊗ g ⊗ h) = f ⊗ ∆(h ⊗ g) where ∆ : C(X) ⊗ C(X) → C(X) is the multiplication map. Their restrictions C ∗ (G × X|O ) ⊗ C(X) → C ∗ (G) ⊗ C(X) are homotopic, hence, since λ = (iO )∗ (β), the following equality holds : λ ⊗ [Ψ] = λ ⊗ [∆]. C ∗ (G) C(X) 3 C. Debord et J.M. Lescure This ensures that λ ⊗ D = (λ ⊗ ∂) ⊗ [∆] = 1C(X) . C ∗ (G) C ∗ (G) C(X) In order to show the second equality, we study the invariance of the element λ⊗C(X) [Ψ] under the flip automorphism f of C ∗ (G × G), that is the automorphism induced by (γ, η) ∈ G × G 7→ (η, γ). The motivation comes from the equality ((λ ⊗ [Ψ]) ⊗ C(X) C ∗ (G×G) [f ])) ⊗ C ∗ (G) ∂ = (λ ⊗ ∂) ⊗ [Ψ] = 1C ∗ (G) , which implies C ∗ (G) λ ⊗ D − 1C ∗ (G) = ((λ ⊗ [Ψ]) C(X) C(X) C(X) ⊗ [id −f ]) ⊗ C ∗ (G×G) ∂. C ∗ (G) If B is a symmetric geodesically convex neighborhood of the diagonal of M + × M + contained in the range of exp, then the flip automorphism of C ∗ (T B) is homotopic to identity. Let C = L×] − 1, 1[⊂ M and F = M × M \ C × C. We denote by r∗ : KK(C ∗ (G), C ∗ (G × G)) → KK(C ∗ (G), C ∗ (G × G|F )) the morphism induced by the restriction r. Using again that λ is in the image of (iO )∗ and that O ∩ M + × M + is a small enough neigborhood of the diagonal, we show that r∗ (λ⊗C(X) [Ψ]) is invariant under the flip automorphism of C ∗ (G × G|F ). Then, the long exact sequence in KK-theory associated to : iC×C r 0 −−−−→ C ∗ (G × G|C×C ) −−−−→ C ∗ (G × G) −−−−→ C ∗ (G × G|F ) −−−−→ 0 ensures that (λ⊗C(X) [Ψ])⊗C ∗ (G×G) [id −f ] belongs to the image of (iC×C )∗ : KK(C ∗ (G), C ∗ (G × G|C×C )) → KK(C ∗ (G), C ∗ (G×G)). This enables us to show that λ⊗C(X) D −1C ∗ (G) is in the image of (iC )∗ : KK(C ∗ (G), C ∗ (G|C )) → KK(C ∗ (G), C ∗ (G)) where iC is the inclusion morphism of C ∗ (G|C ) into C ∗ (G). On the other hand the equality λ⊗C ∗ (G) D = 1C(X) ensures that λ⊗C(X) D − 1C ∗ (G) is in the kernel of the map (·⊗C ∗ (G) D) : KK(C ∗ (G), C ∗ (G)) → KK(C ∗ (G) ⊗ C(X), C). To finish the proof we show that this map is injective in restriction to the image of (iC )∗ . This last point comes from the fact that the inclusion iK of C ∗ (G|M − ) ' K into C ∗ (G|C ) induces a KK-equivalence between K and C ∗ (G|C ); and (·⊗C ∗ (G) D) ◦ (iC )∗ ◦ (iK )∗ = e∗c where e∗c : KK(C ∗ (G), C) → KK(C ∗ (G) ⊗ C(X), C) comes from the evaluation map at c, ec : C(X) → C. 2 A consequence of the preceding theorem is a Poincaré duality between C0 (T ∗ M + ) and C0 (M + ). Thus we get a second Poincaré duality for manifolds with boundary, the first one has been stated by G. Kasparov in [7] for the algebras C0 (T ∗ M + ) and C(M + ). References [1] C. Anantharaman-Delaroche and J. Renault, Amenable groupoids, Monographies de l’Enseignement Mathématique, 36, L’Enseignement Mathématique, Genève (2000). [2] A. Connes, Sur la théorie non commutative de l’intégration, Lecture Notes in Math. 725, Springer (1979), 19143. [3] A. Connes, Noncommutative Geometry, Academic Press (1994). [4] A. Connes and G. Skandalis, The longitudinal index theorem for foliations, Publ. R.I.M.S. Kyoto Univ., vol 20 (1984), 1139-1183. [5] M. Hilsum and G. Skandalis, Morphismes K-orientés d’espaces de feuilles et fonctorialité en théorie de Kasparov , Ann. Sci. Ecole Norm. Sup., vol. 20:4 (1987), 325-390. [6] L. Hörmander, The analysis of linear partial differential operators, III, A series of comprehensive studies in mathematics, Springer-Verlag, Berlin (1985). [7] G.G. Kasparov, Equivariant KK-theory and the Novikov conjecture, Invent. Math. 91:1 (1988), 147-201. 4