s VEYSIN DIFFERENTIAL GEOMETRY R. Bott, Stable bundles revisited G. D'Ambra & M. Gromov, Lectures on transformation groups: geometry and dynamics 19 J. Kollar, Flips, flops, minimal models, etc. 113 R. M. Schoen, A report on some recent progress on non- linear problems in geometry 201 E. Witten, Two-dimensional gravity and intersection theory on moduli space 243 NUMBER 1 '1991 I • • SUPPLEMENT TO Tt-:JE JOURNA( . , OF DIFFERENTIAL GEOMETRY , h!/ / EDITORIABLO ARD JOURNALO F DIFFERENTIGAELO METRY Editors-in-Chief c.C .H SIUNG S.T .Y AU LehigUhn iversity HarvarUdn iversity BethlehePm ,A1 8015 CambridgMeA, 02138 Editors ROBERT L.B RYANT H. BLAINEL AWSON,J R. DukeU niversity StatUen iversoiftN ye w York DurhamN,C 27706 StonByr ookN,Y 11794 SIMON K.D ONALDSON RICHARD M. SCHOEN UniversoiftO yx ford StanfoUrndi versity OxfordE,n glanOdX I3 LB StanforCdA, 9 4305 AssociEadtiet ors ENRICAOR BARELLO MICHAEHL. F REEDMAN UniversdietgalS it uddii R oma UniversoiftC ya lifornia UniversiDtIap rA.:. oMro 2 LaJ ollCaA, 9 2093 00185R OI C;:l-HGEFUMMIO RI JEFFC HEEG Iltoyf S ciences StatUen i ,oyUan iversity StonByr e !,o4y6a4 J,A PAN Libraryo fC ongreCsast aloging-in-DPautbal ication ConferenocneG eometrayn dT opolog(ysI t 1:9 90:L ehigUhn iversity) Surveiynsd ifferengteioamle trpyr:o ceedionfgt sh eC onferenocneG eometrayn dT opol ogyh elda tH arvarUdn iversiAtpyr,i2 l7 -29I,9 90/sponsobryeL de higUhn iversiCt.yC ;. Hsiunegd,i tor-in-chief. p.c m.-(Supplemteont th eJ ournoafld ifferengteioamle trnyo;. I ) ISBN0 -8218-0168-6 IG.e ometrDyi,ff erential-Congr1e.Hss sieusn.Cg h,u an-Chi1h9,1 6-. IIL.e higUhn i - versitIyI.IT .i tleI.V .S eries. QA641.C61159 90 91-7410 516.63-'d c20 CIP No partosf t hipsu blicatmiaoyn b er eproducoerdt ransmititnea dn yf orm orb ya nym eanse,l ectroonrim ce chanicianlc,l udipnhgo tocopyirnegc,o rdionrg anyI nformation satnodrr eatgrei esvyaslt em without pewrrmiitstesnif oronm t he publisrh.e _, Pricpee rc opy$:6 0.0i0n stitut$i2o5n.,0f 0o rp ersonuasleo fi ndividuOarls. derssh oulbde a ddressteodt heA mericaMna thematiScoacli etPy.,O .B ox1 571, Alolr demrussb te a ccoamp AnnexS tatioPnr,o videnRcIe 0,2 901-157U1.,S .A. niebdyp aymeanntdc, h eschko ublemd a dpea yatbolt ehA em eriMcaatnh ematical Society. @ Copyright 1991L ehigUhn iversiAtlylr. i ghtrse served. ISSN1 052-9233 Printeidnt heU niteSdt ateosf A merica. Thep aperu sedi nt hiiss suies a cid-fraenedf allwsi thitnh eg uidelines § establisthoee dn surpee rmanencaen dd urability. SURVEYSI N DIFFERENTIAL GEOMETRY Proceedionfgt sh eC onference onG eometrayn dT opology Helda tH arvarUdn iversity Apri27l- 291,9 90 SponsorbeydL ehigUhn iversity NUMBER 1 1991 SUPPLEMENTT O THE JOURNAL OF DIFFERENTIAGLE OMETRY PREFACE As a service to the mathematics community, the editors of the Jour nal of Differential Geometry decided to organize a conference, to be held every three years, at which invited speakers would survey the field of dif ferential geometry and related subjects. In choosing the speakers, we look at geometry in a very broad way and hope geometers will go beyond the classical perception of other fields. (Among the speakers at our first con ference were people working in algebraic geometry, mathematical physics and other fields.) We decided that the speakers should be prominent spe cialists in their respective areas who are able to give a broad overview of recent trends and make predictions and suggestions for future research. We believe that these conferences will be useful gatherings for experts and nonexperts alike, and in particular for those who are isolated from the mainstream activity in geometry. Our first conference, held at Harvard University from April 27-29, 1990, was made possible by the generous funding of the Journal of Dif ferential Geometry. We are also grateful to Harvard University and the faculty of the Mathematics Department who were strongly supportive of this endeavor. Dean Michael Spence delivered the opening remarks com mending our efforts to bring together geometry and other fields. Central to the success of this conference were the speakers who prepared not only two hours' worth of lectures, but also the manuscripts published here. Their lectures were enthusiastically received by several hundred par ticipants. We would like to express our gratitude to them for their efforts. It is our hope that these conferences will become a tradition. We look forward to the next conference, which will take place at Harvard University in 1993. S. T. Yau H. Blaine Lawson, Jr. SURVEYS IN DIFFERENTIAL GEOMETRY 1 (1991) 1-18 STABLE BUNDLES REVISITED RAOUL BOTT The topological classification of complex vector bundles over a Riemann surface is of course very simple: they are classified by one integer c, (E) , corresponding to the first Chern class of E. On the other hand, a classi fication in the complex-analytic-or algebro-geometric-category leads to "continuous moduli" and subtle phenomena which have links with num ber theory, gauge theory, and conformal field theory. I will try to report briefly on some of these developments here. The simplest instance of our problem occurs when dim E = 1, that is, when we are dealing with complex analytic line bundles over M-a compact Riemann surface of the genus g. Because line bundles form a group under the tensor product L, L' ..,... L 0 L' , the set of isomorphism classes of line bundles J(X) will in this case inherit an abelian group structure, and the first fundamental theorem of the subject, going back to Riemann, Abel, and Jacoby, asserts that J(M) can be given the structure of a complex analytic abelian group. More precisely the first Chern class gives rise to a homomorphism J(M) -+ Z -+ 0 whose kernel Jo(M) is a complex analytic, and indeed algebraic, torus of dime = g : 0-+ Jo(M) -+ J(M) -+ Z -+ O. A proof of this exact sequence on the complex analytic level is actually quite easy, granted the basics of sheaf theory and Hodge theory. Indeed, the exact sequence O-+Ze-27+r1 C-+C * -+0 induces an analogous sequence on the corresponding sheafs of germs of functions, and hence leads to an exact sequence in cohomology: HO(M; ~.) -+ HI(M;~) -+ HI(M;~) -+ HI(M; ~.) ~ H2(M; ~). But HI(X; ~*) is seen to be precisely J(M), by passing from a line bun dle to its transition functions, while by the Dolbaux resolution, HI (M; ~) ::: HO, I (M) , that is, the vector space of type (0, 1) in HI (M ; JR) . Received May 29, 1990. RAOVLBOTT 2 Fm. aI I y, t h e us ·m th's sequence corIresponds. to c1 ' so that the ex. actness o f t h e coh omoI o gY sequence together with the ObV. IO.U S computatIOn that IfJ(M;~) = HO(M; C*) = C, leads to the descnptIOn of Jo(M) as the quotient 1 Jo(M) = HO,I(M)/H (M; Z) of a g-dimensional complex vector space by a lattice, i.e., a torus. The algebraic structure on Jo(M) is much less trivial. In any case, these classical results provide us with as beautiful an answer as we could hope for: this Jacobian J(M) varies functorially with M, and its own function theory is intimately related to that of M. In particular, there is a natural map M ~ J (M) 1 which maps a point P E M to the isomorphism class of the line bundle Lp on M determined by p as a divisor on M. (In terms of a local coordinate zp centered at p in M, Lp can also be described by the data: let V = {Vo' VI} be the cover of M consisting of a small disc, Vo, about p, and let VI = M - p. Let gu u = -L, so that the assignment 0' [ Zp 10 = zp' 11 == I defines a holomorphic section of Lp with precisely one zero at p. Thus c1 (Lp) = 1 , and so p ---> Lp takes values in the component J (M) of bundles with c = 1.) Iterating this map (g - I) times leads to 1 1 the diagram: Mx .. ·xM ---> J x··· x J !!!. J (M) ~ ~ g-1 (g-l) (g-l) 1 M(g-l) and so to a canonical map of the M(g-l) ,the (g - I)st symmetric power of Minto Jg_ 1 (M) . The image of this arrow is now of codimension 1 in Jg_1 (M) and so defines a divisor (the e-divisor), and hence a line bundle Yo, over Jg_1 (M). In the terminology of our century the classical e-functions of Jacoby of level k now appear as the space of holomorphic sections of the bundle Yo k over Jg_1 : e-functions of level k 2'" HO(Jg_1 (M); .fek). From the classical formulas one can then also compute the number of e such functions to be kg: dim HO(Jg_1( M); .fek) = kg. STABLE BUNDLES REVISITED 3 To summarize, for n = 1 , the classical theory teaches us that the moduli space of line bundles is A) a complex variety each of whose components is a torus of dim 2g over R so that its Poincare series is given by Pr{Jk(M)} = (1 + t)2g , and B) the component Jg_I (M) carries a natural line bundle ~, with e. dimHo{Jg_I(M); ~k} = The primary difficulty of extending these results to higher-dimensional vector bundles is that the problem immediately becomes infinite dimen sional. This comes about because bundles E of dim> 1 will in general have many nontrivial automorphisms, and to properly deal with this situ ation the Mumford notion of stability is indispensable. The condition is as follows: E is stable if and only if for any subbundle FeE, (1) cl(F) cl(E) dimF < dimE' Let me illustrate the power of this condition, say for a bundle E of dim 2. First, remark that E always admits a line subbundle LeE and so fits into an exact sequence O---'>L---'>E---'>Q---'>O with Land Q line bundles. Indeed, any rational section of E will determine such an L, and of course the Chern classes of the three constituents are related by CI (E) = cI (L) + cI (Q). Now, if E is stable, we will have cl(L) < cl(E)/2 whence cl(Q) > c (E)/2, and it follows that the sequence cannot split if E is stable! I Similarly a stable E can have no nontrivial automorphism. Indeed, consider an automorphism rp: E E. ---'> The characteristic polynomial of rp is clearly constant because M is compact. If the eigenvalues of rp differ, rp will split E into two line bundles, which is ruled out by stability. Hence the eigenvalues are equal to A, say, so that '" = rp - Al will have to be a nilpotent endomorphism: ",2 = O. Now if '" is not identically 0, and has constant rank, then we 4 RAOUL BOTT have an exact sequence o 'II L EL 0 ----t ----t ----t where L = ker 1fI. But by stability c (L) < c (E)/2, which rules out this I I alternative. Hence IfI == A . 1. q.e.d. The nonconstant rank case is eliminated similarly. In any case, armed with this concept, the structure of vector bundles now becomes tractable. First of all, Mumford shows that the isomorphism classes of stable bundles do admit a natural structure as a smooth algebraic variety In (M) , which varies functorially with M. The tangent space to In(M) at E is given by n I TE(J (M) ~ H (M; EndE). By Riemann-Roch and stability (dimHo = I!) we see that In(M) has dimension n2 (g - 1) + 1. Just as for J = J I , these spaces fall into components J;(M) , k = c (E) according to the topological type of E, I but these will not be isomorphic in general. In fact, J; is compact {:} nand k are relatively prime. In short, for n > 1, the "true" analogues of the Jacobian J(M) are the J,n(M) , with (n, r) = 1. When (n, r) i- 1, one usually compactifies J,n(M) by semistable bundles (replace < by :S;) modulo a certain equiv alence relation, to get a complete, but in general singular, variety, and in the sequel J,n (M) will always denote this compactified version. Actually this stability notion also helps to explain the structure of all bundles over M. For instance, suppose E has dimension two but is not semistable. Then one finds that E contains a unique line bundle LE C E of maximal cI ' so that we can attach to E a unique exact sequence o LI E L2 O. ----t ----t ----t ----t If we set l1i = cI (Li), then the bundles of this type are therefore natu rally parametrized by the set ~: JJ1,(M) x JJ12(M) x HI(M; L; 0 LI), ul + u2 = cl(E); 1112: 112' where the last factor measures the extension determined by E. But note that this is no longer a true parametrization of isomorphism classes. For, as is easily seen, two bundles whose extensions are multiples of each other, 1]1 = A1]2' Ai- 0 in HI (M; L; 0 LI) , will lead to isomorphic bundles. Still this seems to be the best one can do while staying in a Hausdorff framework, and it is therefore plausible to consider the disjoint union .£,(2) = J}(M) Il~, 11 = (uI ' u2) with 11] +!1-2 = r, 1112: 112' J1 STABLE BUNDLES REVISITED as some sort of covering of the set of all complex 2-dimensional bundles with c = rover M. In the algebraic category this data leads to what I 1 believe is called the "stack of bundles" over M. In any case, this stratification plays an essential role in the approaches that Harder [5], Harder & Narasimhan [6], and Atiyah & Bott [1] took to the problem of computing H*(J,n). The Harder-Narasimhan method is number-theoretical and has as its denouement the beautiful counting formula which is a function field analogue (A. Weil, Tamagawa) of a cor responding counting formula going back to Minkowski and C. L. Siegel in the number field case: '"' 1 1 (n1-l)(g-l) L.IAutEI = q -1 q (M(2) ... ,"~(n). E Here on the left one is counting the isomorphism classes of vector bundles (of fixed determinant!) over a curve of genus g, defined over a finite field Fq , each vector bundle contributing the reciprocal of the number of automorphisms it has. On the right the answer is given in terms of the (-function of M: n 2q(1 - wq-s) r _ 1 I '>M(s) - (1 _ q-S)( 1 _ q . q-S) , the wi(M) being intrinsically defined algebraic integers with IWil = q1/2 , and in terms of these {Wi} the number of rational points on the Jacobian J(M) of M is given by: 2g IJ(M)I = IT(1 - w). 1 Using these formulas it is an easy but beautiful computation to find the contribution of each stratum in .#1,(2) to the left-hand side of the Minkowski-Siegel-Weil formula. For ~(2) this then leads to the relation: N + IJ 12 . ~ I _ I . 3q-3 2 (2) q_1 (_1)2 L. 2,+2-g - (q-l) q (M()' q ,=0 q and this relation in turn enabled Harder to compute IJ 121 by subtraction, and hence, via the Weil conjectures, to obtain a hold on the Poincare poly nomial of J2! Actually these computations were done before the solution 1 of the Weyl conjectures by Deligne and Grothendieck, so that at that time they merely used this example to check these Weyl conjectures against the computations Newstead [10] had made for p((JI2), using a topological method which will be explained below. 6 RAOUL BOTT Fi.r st, however, one partm. g comment on t he' " covenng " /rut(r 2) • It 'IS unfortunately not true that the vector spaces HI (X; L; 0 Ll) in Cu are constant in dimension: For large genus this vector space jumps as Ll and L2 vary over J and J . A proper algebro-geometric model of the "set U1 U2 of all bundles" should therefore, I think, be built out of the virtual bundles with fiber Hl(X; L; 0 Ll) - HO(X; L; 0 Ll). These bundles have a constant (even if negative!) dimension, and the disjoint union J?) Il Il .Ji('2) V2Il· .. = ~ now precisely corresponds to the left-hand side of (2), if we agree to count the number of points of a virtual vector space, V, by qdim V Indeed, with this understanding, the expression cc IJ21+ 2)~1 1 i=l precisely goes over into the left-hand side of (2), of course multiplied by IJI, because we have not fixed the determinant in the present discussion. Now in topology it often happens that we are given a decomposition of a space Y, by strata of the form: Il Il ·Il Y = Yo WI W2 .. Wk ••. with the ~ vector bundles over Yi , so that Y is finally put together by successively attaching the boundary sphere-bundle of ~ to what has already been built. Indeed, any smooth function f on a manifold Y with reasonable critical behavior induces such a decomposition. Here Yo is the absolute minimum of f, and ~ is the "negative bundle" of f over the critical set Yi of "index i". Thus these ~ are spanned by the direction of steepest descent for f along Yi . What is novel in the algebraic framework about our ~(2) is therefore that the bundles V/l are virtual, and actually get more and more nega tive as J1 -> ()(). In topology such a state of affairs is usually remedied by "suspending" to convert the virtual bundles into honest ones. In the present situation that would indicate that one has to suspend more and more as the strata are added-in short, one seems to be building some thing "down from ()()", a phenomenon often present in the heuristics of modern physics.
Description: