DIFFERENTIAL OPERATORS ON VARIETIES BY Daniel M. Burns, Jr. A.B., University of Notre Dame (1967) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR TIIE DEGREE OF DOCTOR OF PHILOSOPHY at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June, 1972 Signature of AuthorD....e ,o . t. o .L.. ...... Department of M hematics, Certified by0. ...... . - 1-1 .. ;...'..... r.0" h' Thesis Accepted by .................................. 5 1972...... .... My 5, 1972 'rvisor•. 0 upervisor .......... Chairman, Departmental Committee on Graduate Students Archive IUN 29 1972 18,14AOiesss 2. DIFFERENTIAL OPERATORS ON VARIETIES By Daniel M. Burns, Jr. Submitted to the Department of Mathematics on May 5, 1972 in partial fulfillment of the requirements for the degree of Doctor of Philosophy. ABSTRACT Let X be a (countable, reduced) complex analytic variety, and let n : X + X be a resolution of singularities as constructed by Hironaka. The thesis initiates a study of differential operators on the variety X by pulling these operators back to operators with singular coefficients on R. A formal condition is given on a differential form w on X which guarantees that it is the pull-back of a differential form on X, provided X has isolated singularities: this extends a result of G. Glaeser, and uses methods due to H. Whitney and S. Lojasiewicz. Several applications of this result to differential operators and function theory are given. Thesis Supervisor: I. M. Singer Title: Norbert Wiener Professor of Mathematics 3. To Anne, Fritz Lang, and the MBTA Table of Contents Page Introduction 5 51. Preliminaries 11 §2. Operators on Varieties Definitions 30 §3. Blowing Down Forms 49 §4. DeRham Theory Applications 75 §5. Miscellaneous Applications; Operators on Curves §6. Two Examples 99 121 Introduction. This thesis represents an attempt to extend to singular complex analytic varieties the methods of partial differential equations. The results obtained are the initial steps in one approach at constructing this generalized elliptic operator theory. The basic example of geometric significance in the case of non-singular manifolds is, of course, the Laplacian of Hodge's theory of harmonic integrals. For the case of a singular variety, we again have a DeRham complex of C differential forms, the coboundary operator being the differential operator of exterior differentiation. The Poincare lemma fails in the singular case, however, and the cohomology of the DeRham complex may no longer be interpreted as the topological cohomology of the variety in question. It is a theorem, due to Bloom and Herrera [4], that this DeRham cohomology contains the topological cohomology as a direct summand, invariant under the ring structure on cohomology given by exterior multiplication of forms. Thus, the De Rham cohomology contains topological information about the variety, as well as some analytic invariants concerning the differentiable nature of the singularities of the variety. It is not known at present what these invariants are; in fact, it is not even known in general whether the DeRham cohomology for a compact variety are finite-dimensional. Trying to demonstrate the finite dimensionality of the DeRham cohomology by elliptic equation methods was one of the test questions in the author's mind when beginning to search for an elliptic operator theory on varieties. This modest desire was unfulfilled; one result contained in this report is the finite dimensionality of DeRham cohomology in the case of a compact variety with isolated singular points, but this is proved by methods other than elliptic equations. It is possible to show some simple singularities for which the Poincare lemma does hold, and we consider some of the difficulties involved in trying to construct harmonic integrals even in these simple cases: the problem, in general, is that we don't know at present how to construct formal adjoints for differential operators on varieties. What we shall consider in this report are the first steps towards implementing the following philosophy concerning operators on varieties: 1. lift all analysis problems from the variety X to a resolution of singularities 2. X of X. solve the partial differential equation problems on the manifold X by (hopefully) standard, non-singular elliptic theory 3. push the solutions down to obtaining solutions on X, when possible, X. We have good results on 1., which is quite simple, although the lifted operators one gets on the manifold X are, in general, "meromorphic". This meromorphy already forces us to very non-standard partial differential equation situations on even for when X X X, which are difficult to handle, with isolated singularities. For example, has dimension, we get operators with very singular lower order terms. In the case where X has isolated singularities, we can find simple conditions for pushing down solutions on X to X. In fact, the main positive result of the present work is a criterion for pushing down, or "blowing down" in more Italianate phrasing, a differential form from case of isolated singularities. X to X in the This criterion is simple enough to let us conclude several corollaries from it without much difficulty. is One of these corollaries a functional analytic measure of the "number" of differential operators on functions on a variety of dimension 1. In outline, then, this is the content of this thesis: Section 1 is preparatory, and mainly presents a generalization to varieties of a classical result of Emile Borel. Section 2 deals with definitions also, and contains the lifting procedure for a differential operator. Section 3 contains all of the blowing-down results, which extend a theorem in the local, non-singular case by Glaeser. The methods of this section, as well as Glaeser's, go back to the work of Whitney and Lojasiewicz on differentiable functions, as detailed in Malgrange's book [20]. Section 4 applies the blowing-down results to the smooth DeRham cohomology groups of varieties. The most complete result here is the finite-dimensionality of these groups for a compact X with isolated singularities. This case admits a direct comparison with analogous holomorphic cohomology, whose finite dimensionality has been proved by function theoretic arguments in [51. In section 5 we collect some more examples of applications of the results of section 3. The first result tells which holomorphic functions on a resolution singularities on X. X of an X with isolated blow down to holomorphic functions The second generalizes to such an X a theorem of Malgrange on ideals of differentiable functions generated by finitely many real analytic functions. The last result is the completeness of the space of smooth functions on a variety of dimension 1 in the topology given by the semi-norms any compact set D on X. K in X sup IDf(x)l, xeK for and any differential operator This is made possible by putting results of sections 2 and 3 together to construct a large supply of differential operators on such an X. The method employed fails in higher dimensions because the functions from functions on X: X have too great a codimension in the for example, they don't contain any principal ideals of the larger algebra of functions on X. Finally, section 6 uses the method of section 4 to calculate as explicitly as possible the DeRham complex of the two simplest examples of singular plane curves, in order to see what might be done about the missing Hodge theory for the DeRham complex. 10. The author would like to extend his gratitude and appreciation to his advisor, Professor I. M. Singer, as well as to Professor M. Artin, Professor V. Guillemin, Professor H. Hironaka, and Gerry Schwarz. 11. Preliminaries. §1. Part A of this section recalls basic definitions concerning complex analytic varieties and fixes notations. Part B generalizes a classical result of Emile Borel on formal power series to an analogue for singular varieties. Finally, part C contains a few necessary comments on globalizing results of Lojasiewicz and Whitney. A. A complex analytic variety is a Definition: X Hausdorff topological space together with a sheaf of germs of continuous functions such that every point x e X has an open neighborhood U with the 0X following property: •: U - Y CA , Y there is a homeomorphism where is a closed subset of A is a polydisc in A CN , defined by the vanishing functions. of a finite number of holomorphic -I is isomorphic to OXIU 4 Oy, where Furthermore, Oy = 0 /I Y is the ideal sheaf of germs vanishing on is : 0 A, and the sheaf of germs of holomorphic functions on Iy and Y. Throughout this thesis it is assumed that complex analytic varieties are countable at infinity, i.e., union of countably many compact subsets. the (We have also, in the definition given, assumed them reduced.) We also 12. use the terms complex analytic space, or variety in what follows. S(X) in X X, the points denotes the singular locus of at which X is not locally a complex manifold. X is a closed, nowhere dense subvariety of S(X) p. 111 and p. 141). ([12], By means of the local imbeddings € : U - Y A used above, one may also consider sheaves -l(EA/Jy), and EXIU AX IU = -1(AA/K) : here EA = sheaf of germs of differentiable functions on A(AA = sheaf of germs of C-valued real analytic functions on on Y A) and (respectively, vanishing on Y). J = ideal of Cw germs vanishing KA = ideal of real analytic germs These definitions make good sense, regardless of the local imbeddings chosen by means of the inverse mapping theorem ([12], p. 154). Thus, each complex analytic variety equipped with three sheaves of rings, X comes OX, AX and E X ' Each of these classes of functions has an associated notion of differential form. For general reference on forms, one may consult [11], 0.20.4, or [4]. For the 13. category of holomorphic functions, the definition adopted here is as follows: holomorphic 1-forms on sheaf S: U -(•/Iy - Y the sheaf of germs of X, denoted SX, is locally the 00* + O *d(Iy)), using the rotation A as before, and further: A = sheaf of germs holomorphic 1-forms on d A and is the exterior differentiation operator on 1-forms on A. i-forms = A Define the sheaf of germs of holomorphic = OX by convention.) Again, the inverse mapping theorem insures these local definitions are consistent. For the real analytic and Co categories, the definitions are slightly different: a local imbedding of and define X, 4 :U Y - i = sheaf of germs of N see [4]. Consider A, as before, Co i-forms which vanish when restricted to the submanifold Y - S(Y), which is dense in Y. Then define the sheaf of germs of differentiable i-forms on where E X by EXI U denotes germs of smooth i-forms on -1(E/Ny), A. may categorically imitate this definition to obtain One 14. i sheaves AX of real analytic i-forms on X. Again, the inverse mapping theorem proves independence of the local imbeddings. Given a map f : X - Y of complex varieties, i.e., the f-pull back of any holomorphic function is holomorphic, then Asf: f- Ey f induces pull-backs 0 E X , as well as for the holomorphic and real analytic categories. Note, also, that 'q AP x where is p Ap AX X defined as Es = X p+q=-s AX A = f APj q A p+q=s A decomposition q $ p+q=s of forms into x sometimes will sometimes be denoted wXx ideal. Let If ' Ex. Ai X,x R M EP xX A similar comment applies to Ex(X). EiX,x' AX X,x and B. X was, using the (dz, dF)-type on the polydisc A (similarly for The stalk EX will sometimes be denoted The stalk EPX'q be a local ring and m its maximal is a finitely generated R module, denote A by M the Hausdorff completion of M with respect to ) 15. A the Krull topology. a module over M is A A R, and M is A isomorphic to lim M/mn.M. The Krull topology on M is is the inverse limit of the discrete topologies on the M/mnM. on If C, and R/m R is a topological field, such as is an R/m algebra, then each is a finite dimensional vector space over R M/mn.M R/m. For complete, locally compact fields there is a unique topology of topological vector space over the spaces M/mnM. Thus, M R/m for also inherits the inductive A limit topology of these topologies. M topological vector space over In all our examples, R/m is just R/m. is thus a complete C, and in this case, each of the M/mnM's A above may be normed, and the inverse limit topology on M makes M a Frechet space. M/mnM is finite dimensional, it is a nuclear topological Furthermore, since each A vector space. Hence, M is also a nuclear space ([21], p. 103). Let's look at some examples, first for local rings of functions at non-singular points of a complex manifold. If we take R to be OCn, or simply 00, then A 00 = CI[z1,". in (Zl,...,Zn) n ]], , the algebra of formal power series where (l,..,,z n) are holomorphic 16. 0 coordinates centered at (i.e., zi( 0 ) = 0, i = 1,...,n). A The Krull topology on A 00 says that z if and only if the coefficient of f - f is zero, for all j fj f = Z 1 ... Z n n in 0O sufficiently large. in The A Frechet-nuclear topology on 00 is the topology of simple convergence of coefficients, i.e., za and only if the coefficient of to 0 in C j as we get f If R - - f is f if tends A or A A E n C,o gets large. in fj A = C[[Zl,...,Zn, z1,...,zn]] = E The same relation as above holds between the Krull and A A Frechet-nuclear topologies on A and Eo , and the isomorphisms above are topological isomorphisms (from Krull topology to Krull topology, etc., of course). Considering modules of germs differential forms, one gets, for example,3 like fdz is in ... C[[zl,...,Zn]] dz sn Cno , is composed of sums of terms 1 < i1 , < . < n, where f and the topology is the Krull or Frechet nuclear topology on the series. < i (n) possible coefficient 17. Returning to the general situation, given a module A M as before, there is always a map from M -+ M, which will be continuous, by definition of the Krull or Frechet nuclear topology on or M. If R is noetherian, e.g., A 00, then this natural map is injective, by Krull's theorem. The case of E0o The natural map, E is quite different, however. Eo is given by sending a germ of differentiable function at the origin into its complete Taylor series expansion. By the classical theorem of E. Borel, this map is surjective and its kernel is just M m k=1l k where , m is the maximal ideal of E . Our first proposition shows that this surjectivity property is true for all finitely generated Proposition 1.1: functions at in Eo = germs of differentiable Let Rn, let m M be a finitely generated Let sequence m 0 = 0 m k=l m m, * M- M - E0 modules. M -+ 0 be its maximal ideal. Eo is module. The exact where and all the arrows are the natural ones. 18. K O module, Eo finitely generated sequence, n k=l We first show that k=l mkM) = ~ ) is the mk*M = m.*M. p-l(mk.M) = n r. Consider (K + mk.E(r)). This k=l K + m .E(r) 0 which may be seen as The theorem of E. Borel says that follows: o o g given by theorem^k theorem, K' K' Consider the submodule (r) = E(r)/mE(r) E^(r) an exact E (r where is a M Eo, for some finite k=l last term is just o there is 0 M Er) direct sum of r-copies of p-1(( Since (mainly from [20], p. 73) Proof: = (K + m *E 0(r)/mE (r) (r). = K', where ) (K' + mk*E r) of By Krull's m = maximal ideal k=1 of E^ . 00 Taking inverse images in (r) K + m *E 0 (r) = k0 (r) (K + mk .E(r)). k=l1 it follows that n mkM = m .M. (r) , one gets Er Taking p of both sides, Thus, the kernel of the k=1 natural map M -+ M is just m M, and injection whose image is a dense subset. M/m M + M is an 19. The following sequence is exact: E)(r 1 tK' M/m M O - 0, three terms are and all A finitely generated Eo Taking Krull completions modules. A as E modules, which preserves exactness ([11, §3, no. 5), and since K' and are already complete Eo M/mmM §3, no. 1, cor. 1), we see that ([2], is A complete in the Krull topology as But module. E M/mNM this is the same as the Krull topology on as Eo module, since ^ ^k (Eo/m) @ k E (M/m M) = (E /M ) 0E (M/m M) o0 E = M/mkM. A Therefore, the natural inclusion isomorphism, since M/m M + M is an M/mNM is already complete. Before passing on, one should note that the proposition applies to the modules EX as in x 1A above. Given two local rings ideals m and R and m , an extension R with maximal : R - R of local 20. a ring homomorphism rings is ¢ Such a ¢(m) m A* A ¢ : R induces a map such that 4 A + R , continuous in the Krull topology or the Frechet nuclear topology, if R R and are R Similarly, if are C-algebras. R and r e R, m e M, : M M continuous map , then - and x : M modules respectively, and is a 4-homomorphism of modules, i.e. for M M M 4(rm) = ¢(r)*.(m), induces a natural, M . w : Y For example, if + X is a differentiable map of analytic spaces, then : EX,w(y 7 s *'s E ) are all is a local homomorphism and the R -homomorphisms. The following lemma is useful on several occasions. Lemma 1.2. 4 : E Eo - F F be Frechet spaces, and let be a dense subspace of Proof: open map. F E, be a surjective, continuous linear map. is injective, ¢,E 0 Let 4 E. If 4 restricted to Eo is a topological isomorphism. By the open mapping theorem, Hence, Let giving E 4 is an the induced topology from E, is a homeomorphism onto its image being given the subspace topology in 4(E o ) F. = Fo , Thus, 21. : F there is a continuous, linear map *.E 0 = Id that on The map E o. continuity to the closure of since on E. is dense in E Fo, i -* such E extends by which is all of Hence, 0 F, is injective E. One application is to the following theorem. Theorem 1.3: variety, and X Let a point of x be a complex analytic X. Then the natural map ^s A X,x + E X,x Proof: x is an isomorphism. The map is an isomorphism in the case of a regular point of X. For a general point since the problem is local about x, consider closed analytic set in a polydisc to the origin 0 of Cn. X,o 0 -+ As ES 0 +AX, EX, Ao X,o 6,o X,o , X x as a corresponding We have the following exact commutative diagram (see §lA): Ns A C Cn x s X, 22. Taking completions, and using the proof of the previous proposition, we get another diagram (exact, commutative): s N + X,o 0 S AmE ,o +· A A SEs AA,o A,o A Es X,o X,o 4- 4, 0 0 A A Hence, AX 0 SEs is X,o X,o A AS The map surjective. AX, ° 4 X,o s X,o is clearly continuous in the Frechet nuclear topology, and AX, is a dense subspace A of A3 X,O' if A 9 S By the previous lemma, the theorem is proved E S is injective, ie, S if A 00 S m *E X,o S{01 in EXo* Take a germ Let X = i=1 Xi wa As ,2' defined on an open set be a decomposition of X locally at U. 0 23. n : X -+ X Let into irredundant irreducible components. be a local resolution of singularities as in the result of Hironaka ([131). of in 0 Xi We may take open neighborhoods Ui ([121): so that (1) U } Xi " Ui (2) Let Ui - U i C S(X) = closure in X note that each Xi is a proper map, of 0 e X. is a connected complex manifold. -1 n- (Ui) of - (U i - U i n contains a neighborhood w e A Assume that m *E X,o x i C 7- (0)n Xi, s ~ ( Xsn *w E A. X,x i i = l,...,r. s mCO *E. X,x . since Hence, n i. in and let But this says that the i Hence, w E 0 is an isomorphism from w = 0 V Then, X n w By uniqueness of analytic continuation, for each n X,O' Taylor series of the analytic form Xi, Since is a connected manifold. n( Xi) S(X)); is Xs 0 w at 0 xi. on on V - X - n- (S(X)) -+ X - S(X). AX Xo' by by definition S(X)tf V, 24. It would, of course, be pleasing to construct a proof which doesn't use the resolution of singularities at such a basic level. i.e., For the case of functions, s = 0, a more elementary proof may be for constructed from theorem 3.4 of Chapter VI of [20]. This result of Malgrange is still very non-trivial, and it is not clear how one might extend Malgrange's method to the case of forms on a variety. The last proposition in the section relates the Frechet nuclear topology introduced above with the natural maps X sr E As Ey where : EX X,x Y,y' X,x -+ E Y,y fr(y) induced from the maps = X. one must consider several points in Xs and (or Xs k EX $ E 5YYi Xx i=l ) For later purposes, w-1(X) will denote the map from (or Esx -> k A $ E i=lYi y whoeth whose it h (or s "(yi ) respectively) E factor is the map EXx simultaneously, + E = x, for each Y, Xsr : EXx respectively), i = 1,...,k. E where 25. Proposition 1.4: map, where space. For Let w : Y -+ X x e X } {yl,..,Yk and sw image of the natural map S is an analytic X is a manifold, and Y be a differentiable C 7-1(x), S the A : EX X,x E has i=l closed image in the Frechet nuclear topology. Proof: (after [71, prop. VIII) The map A 7s is continuous in the Frechet nuclear topologies. By a 5s 7T basic theorem of Dieudonne and Schwartz [ 61, has closed range if and only if the dual map has weakly (EX,x)'. closed range in But, if considering a closed subvariety of a polydisc S by proposition 1.1 above, A Ci CN, EX 2N ( ) Now A ^ Xo ) A,o (EAo)', copies of i : (EsS (E and x = 0, is surjective A closed subspace. as AS E. and induces an inclusion and X )' (Es ,o )' as a is the direct sum of ^ (EA)' has the same algebraic and topological duals, namely EA since A (EA,o)' is just finite linear combinations of the Dirac function at 0 subspace of (EeA,o and its derivatives. )' is closed, hence Thus, any linear 26. k i : As (EE A s (ED)' i=lY,1 : (s)' Intersecting with (Es has closed image. But )' has closed range.. ,o it follows that )' AX, (Xs I), is Frechet nuclear, and Eo ([21], p. 147), implying that therefore reflexive This "weakly closed" is the same as "closed". concludes the proof. In this section we simply note that several C. theorems of stated locally for r-tuples of [-0] globally for manifolds and vector functions are valid The theorems of [20] we are interested in bundles. are Whitney's extension theorem (Chap. I, Theorem 4.1) and Lojasiewicz's theorem on regular situation of two closed analytic sets (Chap. IV, cor. 4.4). First of all, given a section f of an A EX module M on X, denote by by the natural map induced by f in Mx . Mx - Mx of the germ Now if X (E) fx This is to be considered "taking the Taylor series expansion" of let the image in A A Mx f(x) is a manifold, and E f at x. a vector bundle over X, be the sheaf of germs of smooth sections of E. 27. Choose local coordinates around trivialize E locally. E, section of f(x) is a smooth f is the Taylor series expansion of the r-tuple of functions given by to the local basis of and with r-tuples of formal power In this identification, if series. X As noted in section B above, (E)x this identifies in x E. f with respect Whitney's extension theorem gives necessary and sufficient conditions (called A regularity conditions) for a collection r-tuples of formal power series, for subset of {vx }x x e K K of a closed X, to arise as the collection of Taylor A series {f(x)})x on when X, of a smooth r-tuple of functions f K X is an open subset of Rn . However, it is easy to check that Whitney's conditions on the A collection {vx) x coordinates in K are still satisfied if we change X, and change the basis for the r-tuples differentiably: use Whitney's theorem to prove this invariance as follows. {vx xc K Interpreting as a collection of r-tuples of formal power series in one local trivialization, suppose these series satisfy Whitney's regularity conditions, and 28. A A hence, by Whitney's theorem, {vx x K for some smooth r-tuple of functions is {f(x)}xeK f. Changing coordinates and bases gives a new r-tuple of functions, f', but still representing the same section of {vx }x Hence, when one interprets the K E. as formal power series again, in the second trivialization, they A are already given as {f'(x)})x K and hence, using Whitney's theorem in the other direction, the new formal power series satisfy the regularity conditions. It is a simple matter to check that with this sort of invariance, one can use smooth partitions of unity to solve the corresponding global problem. Thus, we may A speak of a regular gield of E over X, for K {vx }x K of formal sections closed in a manifold X, and A Whitney's theorem says that Taylor series over {f(x)})xK {v XEK is given as the of a smooth section of E X. Lojasiewicz's theorem, when seen from the same point of view, says the following: Proposition 1.5 (Lojasiewicz): Let X be a real analytic manifold (countable at infinity), and let Y 29. Z and {w zz) are Whitney regular formal sections of Z A A over E the vector bundle Z, vt = wt, {Wz}ze {Vy }yy then X, and Y regular formal section over is also a Whitney Z. where y e Y, is a smooth section of fl w z = f2 (z), then there is a section of conditions, i.e., vt's z E Z, f f(y) A provided the ) E, and A A similarly, fl (y as vy In other words, if we can obtain for for A A t C Y {v } If be two closed analytic subsets. and f2 a smooth section, which satisfies both sets = vy and f(z) = wz, A and wt 's agree on Y Z. 30. §2. Operators on Varieties Definitions In this section several definitions of a differential operator on a variety (all equivalent) are reviewed at first. Then we examine what real vector fields may be lifted when we blow-up a complex sub-manifold of a complex manifold. Using the observations made, and the construction of the resolution of singularities given by Hironaka, it follows that any differential operator on a singular variety may be pulled back to a meromorphic operator (in an appropriate sense) on a resolution of singularities. Differential operators on varieties allow of several possible definitions, probably the most general being based on Grothendieck's definitions in ([11], IV.16.8), which make it possible to talk of a differential operator from one sheaf of modules to another. consider on a complex variety X We can many different sheaves of rings and their corresponding modules, i.e., the sheaves OX, AX and X considered in section 1. Each of these sheaves of rings has its own associated differential operators, analogous to holomorphic operators, and operators with real analytic or differentiable coefficients 31. Cn for domains in . While making a succession of definitions which are categorical in nature, we will RX simply speak of a sheaf of rings (Inductive definition) Definition 1. - which is C-linear, but not necessarily a N D homomorphism of RX-modules. operator from M is a differential if it is locally of finite N to order at each point in given i.e., X; any sufficiently small neighborhood that some integer k} {fl,*.f MJU - x denotes the commutator: If k containing of x [f 2 *'"* [fk, [fl' then Rx(Ux), f E RX(Ux), Ux x E X, and such k, given any k-tuple is the O-homomorphism from for M, N Let a homomorphism of sheaves D be RX-modules, and D : M A, respectively. O0, AA or which is is a X A, we also speak of closed subvariety of a polydisc RA where X, If for our interests. X R X = OX, AX or on NIU . a D] ] - D(Feo), a e M(U )). is the minimal such integer for a x, then K-1 ] [ , ] (Here [f, DI(a) = f*D(a) ... Ux is called the order of D 32. near x. That says that D D x clearly just is an Rx-module homomorphism near Definition 2: above. is of order 0 near (Jets definition) A C-linear sheaf map Let D : M - N x. M, N is be as a differential operator if it locally factors through the sheaf of germs of k-jets of sections of M, for some jk : M * Jk(M). means of the natural map M, RX-modules ([111], Jk(M) is Jk(M) k, by For defined categorically as in IV.16.7, where it is denoted pX(M); Jk(M) is closer to analytic and differential geometric conventions). Here jk is the map which sends a germ to its k-jet at each pt. If the postulated factoring is given by the D commutative diagram M -- > N , then 4 is called jk Jk(M) the coefficient homomorphism of D. We are tacitly assuming, by referring to we may categorically construct a product sheaf of rings subvariety. of RX' RX xX , Jk(M), X x X that with a for which the diagonal is a This is clearly possible for our examples 33. Definition 3. Let (Ambient neighborhood definition) RX = OX, AX be RX-modules, where M, N Assume further that M and generated as RX-modules, N i.e., or X9 are locally finitely near any point x c X, we may find surjective sheaf homomorphisms : RX + M and r-copies : ED RX + N. s-copies x c X if a sufficiently small neighborhood of M 0 + Ker (T) is A embedded as a closed subvariety of a polydisc one has sheaf exact sequences on Thus, CN A: RA @ @ r-copies + 0 and 0 Ker (W) + @ RA It 0 s-copies M, I Here by 0 denote extension of to be sheaves on all of homomorphism D : M ~+ D of M to N M, N A. respectively A C-linear sheaf gives rise to in the representation above, and differential operator if locally at any D is a x E X, there 34. exist ¢ and 9 as above giving a commutative diagram: 0 - Ker (T) RA ÷ M + 0 O r-copies 0 - Ker (W) S ÷ Rs-copies s-copies where D is given by a system of operators on A, such that An O -differential operator D = Z a (z) lal<k a -- , RA-differential D : Ker (W) D where + Ker ( ). is one of the form z = (zl,...,zN) coordinates on A, a = (al,...,aN) and the are holomorphic functions on aa(z) are a An A -differential operator is a multi-index, D is given by , where the A. ~a+B D = b I +B <k (zs) azwhe 3a,B b 's a, are C-valued, real analytic functions; for A-differential operators, the b ,8's may be C. 35. The equivalence of these three definitions of differential operators when all three apply, is essentially given in [111, IV.16.8.8 (cf. also [31). The final definition we give seems more limited in scope, but is more suited for our philosophy of analyzing differential operator problems on resolutions of singularities. Let X be a complex analytic variety, and let ~7T X -) X be a resolution of singularities as constructed by Hironaka ([13], [14]). Y = n-1(S(X)). Let define a "real-semimeromorphic function" poles on Y to be a C function on small enough neighborhood on U (X - Y) as U of f First, on X with X - Y, which for a y e Y, may be written g s Cw(U) g/4, where ¢ and is a real analytic function with zeroes only contained in This definition is local, and S~ Y. will denote the sheaf X,Y of rings of terms of such functions on X. One may also speak of real-semimeromorphic forms with poles on replacing g Y, in the definition above by a smooth form w; denote the sheaf of i-forms of such a type by i S& X,Y A differential operator D : E., X S, X,Y (S~ = s X,Y is a C-linear sheaf X,Y . 36. homomorphism which is locally representable in a real analytic coordinate neighborhood f E jaj<k and where b - f , for as f E E.(U), b X a ax x = (Xl,...,X2n) coordinate functions on operator U D : Ei * Sj X,Y X U. (U), S.. a X,Y are the real analytic Similarly, a differential is a C-linear sheaf homomorphism which is locally given by a matrix of differential operators from E, S. - X free over E X all and . X,Y Note that Ei is locally X S.1 is locally free over X,Y S. , for X,Y i.) (Resolution definition) Definition 4. sheaf homomorphism A C-linear - E is a differential operator aif differential there exists operator D : EX operator if there exists a differential operator D : Ei X Si XIY such that the following diagram commutes, for all open sets U 37. D -- > E (U) EX(U) E(n-l(U))----> X Thus, for Sj (n-I(u)). X,Y * ( w)) ) = w c Ei(U), ,D(xi this case, we say that J7*(D(w)). lifts to D D on In X, and that Since 7 is almost everywhere, it follows that D is unique for a D blows-down to given D on D on X. an isomorphism X. This last definition of differential operator is also categorical in nature: an analogous construction may be performed on real analytic forms, and on holomorphic forms as well, though for holomorphic forms, you allow the operators D to have meromorphic coefficients, and not real-semimeromorphic coefficients. We wish to show that the last definition is equivalent with the preceding ones, when the definitions overlap. Once this is done, we will have shown that problems concerning differential operators on a variety, at least on its sheaves of forms, may be lifted to 38. "meromorphic" problems on the manifold X, where more or less conventional analysis methods may be applied. It remains, of course, even in these preliminary considerations, to show that relatively simple criteria exist for "blowing-down" the results of analysis on the manifold X: this problem is taken up in the next section. Proposition 2.1: equivalent. All four definitions above are In particular, any differential operator on smooth forms (in any of the senses given by the first three definitions) lifts to a real-semimeromorphic operator on smooth forms on a resolution of singularities X of X. Proof: We already know that the first three definitions are equivalent. that a differential operator Thus, it suffices to show D in the sense of definition 4, is also one in the sense of definition 1; and conversely, a differential operator in the sense of definition 3 is also one in the sense of definition 4. First consider a C-linear sheaf homomorphism D D:Ei X EJ X which fits into a commutative diagram: 39. -D> E (U) Ex(U) Ji r*f +X 7T (*) Ei(X -D > SY (U)) X,Y for every open set U X, for some fixed res olution as earlier, and r : X -I X operator on X ( -l(U D with poles on relatively compact, U i.e., w a real-semimer omorphic Now, if Y. U is is compact in X, then is a proper map . On r-1(U) is compact, since 7w-l~ ), the system of differential operators represented by have, therefore, a finite order, D say < k. But, one checks easily by an inductive calculation in local coordinates that for such a W f - f*'D - D(fw), E (w-(17(U)), X < k - 1. functions, any (i.e., D, the operator [f, D]), for a fix ed w E Ei(-1 (U)), Hence, given any X is of order (k+l)-tuple of smooth {fl,..,' k+l}, the operator [fl',2',.,[fk+1,D]...] is of order < - 1, i.e., it 40. } {gl, ""gk+l . let EX(U). I X 7r ([gl,[g 2,.* in S5 Egl,[g 2 '*"" (-1 (U)), EX(U), and in be a (k+l)-tuple of functions from From diagram (*) it = w Consider any is identically zero. is easy to see that [gk+l,D]... ]()w)) [gk+lD]". .1(Xin *w) = 0 where = for gi, = 1 , . . ,k+l. X,Y Since XJ we see that EJ(U). is injective from E (U) [gl' Since w g2***' '"] (m) [gk+l'D] was arbitrary, Sj (-1(U)), X,Y to = 0 in the conditions of definition 1 above are verified. D : E i -+ E EX Conversely, assume x is a differential x operator in the sense of definition 3 above (in of an ambient space). We want to show that, given a resolution of singularities Hironaka ([13] terms and [14]), real semi-meromorphic on i : X + X in the sense of then we can "lift" X with poles in D Y = to a D -1(S(X)). 41. D Recall "lift" means that for every open set In order to do this, we X. U fits into a diagram (*) examine a simple situation for lifting which will, in fact, suffice for our purposes when we've examined the process Hironaka uses to resolve singularities. Consider the process of blowing up a point (say the origin) in (zl,... In Cn. Cn x pn-1 n) coordinates on Cn, with Pn-, homogeneous coordinates on (wl,...,wn) and we take the subvariety defined by the equations: for all Cn i, j between Cn x pn-l, projection onto 1 . wiz j = ziwj, Call this locus be induced by w : Cn + Cn and let Cn n. and the It is a basic fact that the construction above is independent of the choice of coordinates in automorphism Cn , and thus given a holomorphic ¢ : Cn . Cn , holomorphic automorphism we may cover it by a ~n ~n C . 4 : C 0 E Cn , and we that the construction is local about may blow-up 0 in U and get any open neighborhood of in C . Now if fis 0 in U U, - Cn Note, also, , and where U = U is -1 (U) a holomorphic vector field 42. defined on a neighborhood U of 0 E Cn, for U relatively compact and small enough, there exists a : U + Cn one parameter family of maps C, i.e., defines and f E C (U) If C(0) = 0, 0 is defined. C(f)(p) = 5 (here then which f((p))j(, 0 , for p E U, lies in an E-disc about 0 in C). (0) = = 0 for all ý for which Putting the two observations above together, we see that there is a family of local 5 automorphisms ( Let : U - Cn covering the maps 45. be the holomorphic vector field which they define: (f)(p) = Since the 's i f(E (p))I= 0, cover the O&'s, we have 5 Thinking of the vector fields operators, note that ((w f) = W (C(f)), df*(E) = ý of the differential operator 5 ( and dw*,() = 5. as differential says that f E C (U). for , f EC (U). for to Thus ( is a'lift" U. Next, note that by complex conjugation, one may lift any anti-holomorphic vector field on vanishes at 0. D on which Note further, that if one can lift a differential operator a U D acting on functions on U U, then one can clearly lift the operator to f-D, 43. It follows, therefore, that any smooth f s C'(U). where vector field 5 on U vector field 5 on U, with poles only along In fact, lifts to a real-semimeromorphic = E ai (z,z) write i = a( azj z ++ i,j Since i - z bi 's are in on U. Set i- only along z zjezj., zi lift, and the C0(U), the operator 1 = Tr (¢); i (0), ('/i and meromorphic lifting of write b biz * E i,j EJ j and i + E bi(Z) i i = If ai, b i E C (U). where ( - (0): ýC aiz 's and lifts to a E' is real analytic and vanishes is to the required semi- U. More generally, we may blow-up along a linear subspace Ck x 0 just Ck x Cn . Introduce coordinates and get, for the blow-up along Ck,,for (yj*.*k) Ck in Ck Cn C Ck+n x pnl, given, as before, by ziwj = wizj, and no conditions on the yj's. Again, this is invariant with respect to coordinate changes which take Ck x 0 into itself. earlier shows that for j = ,...,k. and Yj Finally, if 5 to a Ck. on in on we may lift C k x C , for any vector field E's defined only on open neighborhoods of Ck+n < k of order Ck x Cn order O*D = n E C zi i=l D = acting on functions, U of 0 in to a real-semimeromorphic operator D Ck x Cn. in < 1. < 1, D on the opne neighborhood lifts n -(U) write is a holomorphic Ck x {0}, Any differential operator on f The process is again a local one, and applies to 0 E' lift to the blow up, aYj function which vanishes along (f*T)*E The same argument as This is so for operators of In fact, for an operator D of order lifts to an operator without poles, where . In general, N E Dk+Do, • if D where is of order < k, E2,'s are vector fields R=l the D 's of order (k- are operators of order 0. #k*D = Z 0k5 .D 2 k-1 .kD ()-D*D + o. < k - +).(kk-1 +kDo = E(D By induction, 1, and is Do D ) (RE) and - 45. k-I ( k-1 D) ¢kD, lift, for every k, k as does kD , and hence, as a sum of products of liftable terms, liftable, to a smooth operator D'. is also Again, ~ D'/ k = D is a real-semimeromorphic lift of D. Now consider a differential operator (in the sense of definition 3), In order to lift D : EX + E X D to a for D : E, X a complex variety. S.. X we may do so X,Y locally, and patch them together (since the lift to 1(U) I(U) C C X X of of a a D D mentioned earlier). defined defined on on U U X is unique, Let us consider a U open in as X such that: (i) U may be considered a closed subvariety of a polydisc (ii) in CN A, the closed subvariety given by S(X) C U functions (iii) A is defined by finitely many holomorphic fl'", fr" we have the following commutative diagram 46. T l(U) u U(S) ÷ A(s) =- (1) U(l) IT U - i where each i = 1) for a(i) is the blowing-up of S(i) . (or Here A( i - l) (or A, along a closed submanifold contained in S(U (i - 1 ) ) , and where U(i -l) A U, for 7 (i) U (i ) i = 1) is the proper transform of under the transformation is the map induced by a(i) . This diagram results from the method used by Hironaka to resolve the singularities of (iv) there is a differential operator which induces the operator order < k on D : EU - EU, L : EA and L - s'k.*L r = E f.I.. i i=l We will lifts to a smooth operator L' EA is of A. Given all of this, let prove that X, [13]. on 47. A, where s is the number of blow-ups used to resolve U. In fact, we have already shown this if f. each 1 Thus on Set A(1) o(1)*(f k *L = () lifts to a smooth operator ()() (since it is contained in S(U(1)) o(1)-(S(U))). on s, we see that L' on A. S(U(1)), (s-l)k*L(1) function which vanishes along U: A - a 1 (S(U)) lifts to the desired g E C (A) U, then a - S(U), defines an operator on A type. vanishing along Thus L' a U, also is an isomorphism 1 a•(S(U)) and that is *Sk*L X and thus takes a function U into a function of the same defines an operator lifts the operator is any L'(g) U - U an everywhere dense open subset of on therefore, to this is because to A(2) and proceeding by induction It is clear that if vanishes along and vanishes and ýsk.L lifts, (s-1) kL(l), an operator A(1), blown up to get A(l) L(l) = Z(1)*(f i)((1)*() i) is holomorphic on i ) on the submanifold of of since vanishes along the submanifold which is blown-up. Each s = 1, sk D, and D' D'/ýsk on is U, which a 4 8. real-semimeromorphic operator which lifts the operator D, = n (4). W where Since singular locus of U, 7- l (s(U)) D ¢ vanishes only along the has poles only in = Y CU. The proof of the proposition for forms of higher degree is similar, but messier. We will not use this part of the proposition later, so we will omit its proof here. 490 §3. Blowing Down Forms In this section, a few basic results are proved on "blowing-down" smooth functions and forms from a resolution of singularities to the resolved variety. Although the problem is a natural one, there is very little in the literature about it, except for a theorem of G. Glaeser [71, which will be our starting point: Theorem: Let U, V respectively, and let Rn , Rk be open sets in 4 : U -+ V be a real analytic map such that: (1) (2) is a closed subset of W(U) if K is a compact subset of then there is a compact set that (3) C(U), K' C U such ý(K') D K. de, is surjective on an open dense subset of If V U. f e E(U) respect to (In particular, n > k). satisfies the "2-point condition" with 4, then there is a g s E(V) such that 0*(g) = f. A function satisfies the "n-point condition" with respect to 0 if, for any x E O(U), and xl , . . . , x n C U 50. ¢(x with ) = ... 1 gx c Ex = 4(x ) = x, such that (gCx) = there exists a f s Ex i i = 1,...,n. , i The "finite point condition" for a function a map ¢ ¢, for means that for every f n. f and satisfies the n-point condition We can also speak of the n-point condition for i-forms as well, replacing rings of formal power series with modules of formal i-forms, and i^A with c . Xi We shall first make two very simple extensions of Glaeser's result. The first generalizes this result to include forms. Proposition 3.1: w e ES(U) let on U Rn 1 < ... T c ES(V) y',...,yn Let < i - V be as above, and such that XSC*(T) P the same for s runs over all such = W. RP . V is a multi-index of degree < p ý. be the standard coordinates x' ... , and I = (il,...,i s ) 1< $ : U satisfy the 2-point condition for Then there exists Proof: Let set dx i = dx 1 ... S-multi-indices the s, i dx s dx 's If with If I form a 51. basis for on U, E s ( V) when over dy 's Similarly for E(V). J = (Jl,...,Js), 1 < jl With respect to these bases, write < .*. < js < n. xs( *) as a matrix, , X (* )dxI = Ea dy J where the I's a 's The sequences. are restricted to increasing J's and are (real-valued) real analytic A functions on U. Thus, a formal s-form T Z= I written gI , dx EX , x A^^ may be and AT 4 ( = Xs A T s Es I)aj dy in E , 0(y) = x. I,J w e ES(U), Given 2-point condition for in particular, that for every E fjdy = y e U. , w = E fjdy write w f a E(U). with respect to "y = (T XS ), The implies, (y) E y) Expanding this, we see (gI)a JI dy , for some gi's ^* ^ J Equating coefficients of E E(y). dy 's, we see f = E ^ I in E, Let for each N = ( y e U. ), S M = (n), s so that the )A (gI ) a 's define 52. A : a linear map $ E(U) -1 N copies matrix multiplication. is if and only if, for every point l , . . . , a M) L A gI E4 (y) Th. 1). ^* ^ Ey, i = ,...,M Since every power series gI on 0(y) of a V, the equations ^I fj = E I (gi)aj (fl,*..f M) say that the M-tuple of functions satisfy the conditions of Whitney's (fl',*,f and hence in the image of A in is the power series at differentiable function ^ by y E U, there exists ai = bi such that Exp. 25, ([18], A E(U) By a theorem of Malgrange, an (bl,...,bM) M-tuple of functions (a $ M copies M) E L. Thus, = f h a , theorem, with I h I e E(U). y E U, one gets: Passing to formal power series at ^ ^ fj = hla j = ^I ^* ^ Hence, that ^* 1(hI I de, - ^I (g )a ^ (g ^A , with ^AI ))a g A E 0(y). A = 0 in E . The assumption y is surjective on a dense subset of dually, that XS( same dense set. * ) U means, is injective on the C-image of that Thus, the matrix (a ) has to have a 53. N x N minor when evaluated at points non-zero arbitrarily close to any given arbitrarily close to Hence, there minor which is non-zero at points N x N is at least one y s U. d, is a This minor, call it y. real analytic function which is non-zero arbitrarily A A d c E y, and hence close to is non-zero (uniqueness of power series expansion), and is the corresponding (a ) minor of the matrix A However, A h C(h I - A* A )A ) )a ¢ (g with entries from = 0, implies ^ A* 4 (g l )) = 0, for every d(hI - E I, and hence A A E (gi) = 0, since hI - is an integral domain. Now the 2-point condition for x = 0(y) = ¢(y'), A A y (gI) A one set of says that if then we may simultaneously solve the A* A hI - equations w gl's = 0 and hI A - Ex . Thus, each Glaeser's theorem quoted above, where the I, T = E gdx I , gives with A in the 2-point condition with respect to each 4y ,(g) g 's w = are in hi satisfies 4, so that by h ! = ¢ (g!), E(V). for Setting 4, (T). I Q.E.D. 54. - theorem globalizes The next extension of Glaeser's the domain and range of the map in question. Proposition 3.2: manifolds, X, Y Let countable at infinity. be real analytic c : X -+ Y Let be a real analytic map such that (1) O(X) (2) for every compact is closed in compact (3) such that 0, then w = * (T), O(K') D K. where T e ES(y). It is easy to see, using a partition of unity, that the question is local on Y Y, i.e., we RP . By a theorem is an open set in of Grauert ([8]), we may also assume X properly imbedded as a real analytic submanifold of Let of U CRk a satisfies the 2-point condition with Proof: may assume there is X. w s ES(X) respect to K C O(X), is surjective on an open dense subset do4 of If K' CI X Y R ,$ large. be a real analytic tubular neighborhood X, with real analytic projection we consider the composed map i = n : U + X. If o n : U + Y, then P, U, Y satisfy the conditions of Glaeser's theorem. 55. By functoriality, the s-form s *((w) 2-point condition for the map J. By the previous Xs proposition, (W) = XSi T e ES(Y). some Since XAs w = injective, (T) Xs X sr = : Es(X) satisfies the XNs* (T), - for is Es(U) (T). The next extension of these methods is to the case w : X of a map - which is X a resolution of singularities. It would be reasonable to conjecture that a smooth form SES(X) T was the pull-back of a smooth form E ES(X), if w satisfied some simple formal condition D, on its Taylor series at points of e.g., if it We show here satisfied the "finite-point condition". that this is, in fact, so if all possible Taylor series obstructions to blowing-down vanish. Theorem 3.3: singularities in as above, ES, for every Proof: assume X Let w : X + and let y e D, then X be a resolution of w e ES(X). w = The question is local on sw(T), X, is a sub-variety of a polydise Assume first that X If w = 0 T E Es(X). i.e., we may A C CN is a resolution of singularities of the type constructed by Hironaka ([131). That is, 56. X is constructed by a succession of blowing-ups as follows: A(n) (n X = X = A . (n) 7(n) C1 X(1) ( (l) +7T ) x = x ( o) Each a(i) (0) A(i1) : A() closed submanifold of and X( i ) respect to is the blowing-up of a A(i -l ), contained in is the proper transform of a(i) a neighborhood of Let X U c A imbedded in function on in a neighborhood of A, supported in smooth s-form on A E,S for every X. Then X(i-1) A, and let X. p : U- Let p X be a U, and identically 1 u = A, and by construction, y e D. with be a smooth tubular be the retraction of the tube onto C S(X(i-1), (WSp is a (t) u = 0 y in Now, throw away the form u A and keep only the collection of formal forms uy, y c X. 57. Since it is the field of Taylor series of a smooth form, it is a regular Whitney field of formal forms, we'll simply call Set of A, and u. -1 (S(X)); Y = Y X = D. Y which is an analytic subvariety Consider the regular Whitney field of formal S-forms on Y which is identically A zero. Since u = 0 fields, call it v, D, the union of these two along is well-defined. theorem of Lojasiewicz (cf. [20], v is a regular Whitney field on a smooth form v and Y or prop. 1.5 above) By Whitney's extension theorem ([20], there is X A, the regular-situation are analytic subvarieties of implies that Since on field of Taylor series along A Y Y U X. or §lc above), such that X. v is its In particular, A v(y) = 0, for every y E Y. Thus 2-point condition for the map v a : A satisfies the * A, which is proper, holomorphic and biholomorphic on an open dense subset. Hence, v = Xs *(a), diagram: proposition 3.2 applies, for some a~ ES(A). and Consider the 58. Then X --- > A X -> A XSW*(T) s .s (a) i * (a) = si*(v) = m, concluding the proof. Now if w : X -+ X is a resolution of singularities not of the type constructed by Hironaka, construct the following situation: Z r-> Z > X' +p +7' X Here Z X' T-> X is a resolution of is the fibre product of Hironaka's type, X xX X', global Hironaka resolution of [14]. X Z, and Z -r-> is a as constructed in The maps in the diagram are all proper, holomorphic and even biholomorphic except at points over w Z as in the hypotheses of the theorem, has vanishing Taylor series along S(X). Xs(p-r) Given (W) (w-.pr)-1(S(X)), and 59. p' n : Z - is an isomorphism outside this nowhere X' dense subvariety. E ES(X'), c a unique S 5(p.r)*( Thus, ) = XS(p'*r) by proposition 3.2 above. M(), Now for i has vanishing Taylor series along 7- 1 (S(X)) = p'*r((r p r)- 1 (S(X))): we proved this in the course of proving proposition 3.1. Xs argument there shows 0 = p' setting r. ( q ) : ) -E In fact, the Es r' : X' -+ X Since is is injective, now a Hironaka-resolution, conclude, by the first half of the proof, that ý = Xs(T, *(T), through the diagram above, gives T E Es ( Xs W* ( ) = W. X). Chasing The above theorem has several consequences: Proposition 3 .4: Let X be a complex analytic variety with isolated singularities, and let * : X - X be a resolution of singularities. If wE E S(X) satisfies the finite-point condition for w, then s w = We will first Lemma 3.5: and x c X E (X). T T*(T), prove a lemma. Let X be a complex analytic space, a point such that X is irreducible at x. 60, Let r : X Xsi* : Xr :7E x X - + be a resolution of singularities. S-1 E Proof: y c 7( x). The proof consists of a comparison with the real analytic category. If x, there is nothing to prove. x c S(X). T is injective, for every y Then Since X X is non-singular at Thus, we may assume that is irreducible at x, we may assume X - S(X) is a connected complex manifold ([12]), shrinking X x, if necessary - this is clearly permissable, about since the proposition is a local one at X = r 1(X - S(X)) an element x e X T is a connected manifold. s E Ax Let U is connected. If Xs, (T) = 0 in as a form on some neighborhood of -l (U - (U n S(X))) -1(U), -1(U) T = 0 AS y. U; U - (U 0 S(X)) then Xs*() = Since is connected and dense in is connected. analytic continuation, Thus, Consider is defined on T We can assume that T. Thus be an open neighborhood of such that the germ of form we still call it x c X. By the uniqueness of As T(T) = 0 as a form on all of by definition is therefore on all of w-l(U). U - U r S(X), and 0 E AS Now consider the commutative diagram 0, 61. c_ A AS Cx r 7 Xsw1 A> AS 2Z> E x x s^ * J Trr As c Es y y where the horizontal arrows are natural inclusions. proposition 1.4, X r (Es) x is closed in given their natural Frechet topologies. * mapping, where : Ex Xsw mapping theorem, Es, both spaces y By the open (E x is x) is given the subspace topology A A of an open ½ (Es) x X s X By Es. y" As Now is x a dense subspace of Es and x' restricted to that subspace is injective. lemma 1.2, this means Asw By is injective on all of S E . X· Returning to the proof of the theorem, let N an isolated singularity of X. Write X = U x (i) X(i) i=l where each X(i) irreducible at and x(i)f X is a complex analytic space, x ) and non-singular away from = {x, for i X jB ([12]) x, - again, be 62. we have to shrink X Let singularity. about X(i) = X is a manifold and Let w e ES(X) with respect to for each exists a X have yi E X(i)n w-1({x}) X ^ c (i) 7 and then i. AS i. E 1. for all al denotes the natural inclusion, set If y s(w i)) for some i. S o X n (T) ) = (T 1 ), = my, -(x), A Xw7 such that x Es Y we The finite-point condition sA* Es(X) Xs( (iA is any other point in By the previous lemma, : w-1(x). be chosen Xsw Y(T) = yri Ex y E X( i) and hence ^ A* such that By naturality of y Let AS e Ex T A . i P J. for -, The finite-point condition says there says there is a E = i. induced by for each X n : each A ^( ^) - hi(w ) i X( by X, satisfy the finite-point condition 7. h i : X(i) If {x}), closure in -l(x(i) be the proper transforms of the is the only x to ensure x ^ ^ ^(i) hi(a) = T we have i.e., By proposition 1.1, A (a) = m in y ^ Es(X Xs())*()) W = X * (T) = y in ( )). A for any we may choose a 63. A ES(X) TE - Asw (T). ' = y : so that w W' = ((y), A T = T A x Hence, in W y x ES(X). = 0, Let for every and, by the previous theorem, for some si*(a), ES(X). oa Subtracting gives W = A r ( ca +t). Remark 1: The finite-point condition can be replaced by the max (2,N)-point condition, where N = number of locally irreducible components of at X x. Remark 2: In the case of a general singular X, the finite-point condition implies that for every there is a unique A A T ES(X) x ES(X), y for every y c w- (x). unique formal solution for T ^ (T) A = W y Thus, there is a at every point of We make the following conjecture: If w satisfies the finite-point condition for W = S* XSr -1 A in such that x e X, s w*(T), for some T E Es ( X). X. E S(X) w, then The conjecture, therefore, amounts to showing that the formal solution above is a differentiable solution, i.e., the problem solved in the non-singular case by Whitney's extension 64. theorem. As a final extension of the technique given above, we prove a Glaeser theroem for holomorphic maps to a variety X with isolated singularities. Theorem 3.6: Let f : X' analytic spaces, where and X' X be a map of complex has isolated singularities, Assume is arbitrary. X - f satisfies the following conditions: (1) f(X') (2) for every is closed in compact in K compact X' (3) X' - f-1 (S(X)) (4) df* w E ES(X'). T f(K') D K. is dense in X' If w X' - f-1 (S(X)). satisfies the finite point f, then w = As (T), e ES(X). Proof: K' is surjective on a dense set in the condition with respect to some f(X'), there is such that regular points of Let X Construct a fiber product square for 65. It 1 X X' xX - -- > X' +7f' +f > X X where is a resolution of singularities, and i7 are the natural induced maps. proper and surjective. and let Let Z = the closure of X x subvariety of . p Z Z + X' Note also that U = X x X X' U in i', f' w' (f.7')- X xX X' Z is 1 (S(X)), is a and we get a diagram q X' S•+f' +1f X+X where p is a resolution of singularities, and the maps in the square are inherited from the previous square. The map q is proper and by condition (3) above, surjective. Pull back the form functoriality, for the map a down to f a w to y = Xs(q-p) w on Z, and by satisfies the finite point condition of complex manifolds. In order to blow X, it is first necessary to check 66. (1)- conditions f(Z) = f'(Z), since First, {(xi,xi)} i, Let of proposition 3.2 (3) be a sequence in converge to x C f'(Z). Since ki in f(X'), f(xi) 's converge to x - is in {Tr(xi)} U {wI(x)} since the X 7(x), and assumption (2), there is such that yj c K' such that X. We want Z, and a point w(y) £ S(X), if necessary, 7-l(s(x)). i. for every Hence, The set is compact, and contained in are in f(X') f(X'), the a compact set f(y~) = w(xi)'s is closed. {7(xi )} J {7r(x)} C f(K'). that Z {(xi, xi)}, r(x i ) SS(X), we assume that is surjective. if and only if U we replace the sequence one for which each p in is dense in is in Z in (y,y') U x above. K' C X' such Thus, there are r(xi), for every Passing to a subsequence if necessary, we can assume Y! +' y' (xi, y ) (x, y') E K', since is in is in U, Z, K' is compact. (Xi, yj) and - Now each (x, y') f'((x, y')) = x. implies 67. Next, let K C f(Z) find a compact is proper, a such that K' C Z, D be an open subset of f'(Z) KE K (0 D w(KE) C f(X). ) = f(K') K C f'(K') is dense in will f'(Z). is dense in KE Let K C KE , such that 7r(E). = K", compact in {xi I such that p K . and Consider By assumption (2), there is a compact such that and let f(y KE - is compact. q -1 (K') Since 7-1 (S(X)) C X.) still stands for K' C X' f(K) D K. compact, with -l (f(X')) - D n 7-1 (f(X')) (Here We will 1 (f(X') - f(X') n -l S(X)) = f'(U) = do, K C Z be a compact set. xi Let x be a point of be a sequence of points in x. + w(xi )' Z. Consider yi s K' Choose for every i. Each KC - KS r K, D so that (xi , y1) is in K", and, passing to a subsequence if necessary, we assume yj + y' in K'. f'((x, y')) = x, So (xi, YI) Hence + want a point z' is surjective at So, let z df, and is surjective be a point of arbitrarily close to z'. s K', K C f'(K"). Finally, it remains to show almost everywhere. (x, y') z Z, and we such that df* The following are closed, nowhere 68. Z : f-1 (D), (q*p)-1(S(X')), dense subvarieties of and p-1(S(Z)). (f-1(D) of the theorem.) is nowhere dense by condition (3) Consequently, we may assume the complement of their union. close to x' such that is surjective at y'. + F(O) (x, y') E X - S(X), 0 of y' and df, such that f(O) EX - S(X). E X x X' X x X X' > X x = If is contained in the definition of fibre-product, near map arbitrarily is the projection of a holomorphic fibre bundle over then By By the implicit function theorem, there is a neighborhood flo : 0 y' E X' x = f(y') is in x' = q.p(z). Let assumption (4) above, there is a z K- (x) X xX X'. (x, y') e X, By the is just the projection of the pullback of the bundle flo : 0 - f(O) over -1 7(f(O)), and hence neighborhood. dft is surjective on that Such points (x, y') are in U, and may be chosen arbitrarily close to the original z, verifying condition (3) of proposition 3.2. Thus, we may blow a down to a form the lemma below says that components of X. Thus, f(Z) T T on X. is a union of connected is uniquely determined on 69. these components, and arbitrary elsewhere on fix T on those components not in follows. Let T1 E ES(X) for every x e S(X), for every x' e f-1(x). X. To f(Z), proceed as be a smooth s-form such that, S = Xf (,x) A As x ' in E x ,(X'), This is possible formally, by the finite-point condition, and the existence of T, locally follows from proposition 1.I , and globally by a partition of unity. T Let s-form blown down from a be equal to the unique on the components in and on the other components equal to Sn*(T 1 ). an easy formal calculation to check that this blow down to f(Z), It is T will X. Lemma 3.7: Let f : Z -+ X be a holomorphic map of complex manifolds such that: (1) f(Z) is closed in (2) for every compact K' C Z (3) Then f(Z) df* X K compact with in f(Z), there is a f(K') D K, is surjective on a dense set in is open in X, i.e., of all connected components of f(Z) X Z, is the union which intersect it. 70. Remarks: The condition as is shown by the map (x, y) f(Z) be closed is necessary, (x, xy) - from C2 to This is just an "affine piece" of the blow-up of the C2 origin in The map f need not be an open map: X = C , Z = C2 = C2 with the origin blown-up, and f = the projection of all of f- 1 C2 , but f again, let Z onto X. The image of f is is not an open map at any point of (0). If f is assumed proper, subvariety of f(Z) is an analytic X, by the proper mapping theorem. Condition (3) then says that at a point dimension of f(Z) = dimension of x E f(Z), the X, hence f(Z) = X Condition (2) is probably not necessary, and nearby. could probably be removed by use of a dimension theory argument, Proof: Suppose, first of all, that at the dimension of f-l(x), the map X f were 1. x e f(Z), Then, at any point of would be open. be the origin of a small disc, and For, taking z E f-1(x) x the origin of a small ball mapping into the disc by condition (3), there is a 1-complex dimensional disc to C2 71. through z non-trivial. disc to C f such that restricted to this disc is But a non-trivial holomorphic map of the has open image. For higher dimensions, we cut things down to a one dimensional situation. By condition (3), f such that there is an open dense set restricted to are no boundary points of and let x E f(Z) a small open ball both Let f(U) and xl E B(x) 0 U is open. f(Z). is closed; let B(x) x, it must intersect around f(U), 2 -x l x2 E B(x) 0 (X - f(Z)). )j0 < t {xl + t(x ) Take a new ball 2(xl, x 2 ) 2 -xl)10 K < t < to} which contains centered at and call it the origin 0 of our ball in of the real line set of Z £(x such that l , 2 ) o4 f(Z) be the xl. x X 0 B(x) Then £(xl,x 2 ) r (X - f(Z)). in L = complex line through in K = 9(xl,x < 11. is a limit point of B Taking in non-empty open sets. and connected component of x I + to(x 2 -xl Assume the contrary, x e X - f(TZ). X - f(Z) {xl + t(x We show there be such that Consider the real line segment given by UC Z + to(x2-xl)2 X. Let which is the complexification x 2 ) C B. Let K' f(K') D K. Let {t i } be a compact be a 72. K A B sequence of distinct points in t i + 0 E B, and let for each i. z i E K' such that be such that ti f(zi ) Passing to a subsequence, if necessary, we may assume the zi's zO E K', converge to a and f(z ) = 0. Consider the closed subvariety f-1(L) of f-1(B), -l we have z o and z i : f (L), for every i. Let be representatives of the local irreducible c1 z 0 . Since at f- (L) components of U Yj forms an Y1,***,Ys j=l1 f- in zo open neighborhood of zi's; say contains infinitely many of the f : Y1 - is a map such that L non-constant near of If f(z o ) = 0 f(Y1) Y1 . and Now is f contains a neighborhood 0 e L, we are done, since this would contradict the 0 fact that is a limit of points in By shrinking consider zo n around Y1 zo L (X - f(Z)). if ne cessary, we may the center of local coordinates n = dimension of (l,.,n) zo . Y (L), at least one Z at (n zo, such that the projection ( 1"**,d), d = dimension of Y1 at zo, 73. restricted to Y1, represents of a neighborhood p : Y1 - Let B' B' of Y1 0 in as a branched cover (l'''***'k)-space be the projection. Let ([12]). D = the closed, nowhere dense analytic subvariety such that p : Y Y, - P 1 (D) -p(D) - B' - D Y 1. is dense in Even after shrinking, zi 's converging to Y1 z0 . analytic subvariety of > 1, and hence p(f-l(0)) is a topological covering; Y1- Note that contains infinitely many Y1 through f -1(0) > 1. neighborhood in 0 Therefore, in Y1 zo is dense in which is of codimension through f-1 (0) Hence, is a closed subvariety of B" in p of B" -1 (L') such that 0 p- (0) = z o . such that B' is a closed of codimension Y 1. Now through Hence, there is a B', and a complex line at C zo - . C L' L' n (B" n (D u p(f-l(0)))) = {0}. is a curve in a neighborhood of f- 1 (0)r) p-1(L') = z o . Let a a local irreducible component of this curve at let 0, z C = zo, and be a resolution of the possible singular point Then nC), and f(C) = f.r(C), and o f frw(i(z )) o = 0 in L. 74. Hence, by the 1-dimensional case mentioned initially, f(C) contains a neighborhood of 0 e L; as already mentioned, this leads to a contradiction, which proves the lemma. As a final remark, the above lemma extends to the case where f Z and X are complex analytic spaces, and satisfies the conditions of the last theorem. conclusion is then that irreducible components of f(Z) X. The is a union of global 75. §4. DeRham Theory Applications. In this section we study the smooth deRham cohomology HDR(X) especially for an X X such that X, with isolated singularities. msE X Definition: on of the analytic variety sheaf of germs of forms 0, x for every is clearly a sub-complex of EX. x e S(X). w mSE There is a natural exact sequence 0 -+ ms Ex E- + E aS -+ 0 A where EX, along S(X). then is to be thought of as formal forms in S EX S For example, if is supported at X has one singular point x, x, and there its stalk is Ex(X), the formal Poincare complex of in general, if Y C X EX, S is supported on is any subset of of germs of functions for every y E Y. Then f X X, let such that my EX is X at S(X).) my f x. (Clearly, More generally, be the EX-ideal = 0 in Ey a subcomplex of EX, 76. and the quotient complex is supported on closed. and Y For example, if X Y, if Y is is a non-singular manifold a closed submanifold, one can take global sections and get a short exact sequence of complexes: 0 + myE(X) + EX(X) - EX,y(X) - 0. The arguments given below will apply to the calculation of the long exact cohomology sequence associated to these complexes, and it is the topological cohomology exact sequence of the pair (X, Y) with C-coefficients. Generally speaking, the approach to controlling HDR(X) the groups given here is to use the blowing-down results of the previous section to control H (msEX(X)). By a comparison with the holomorphic category, some measure of control on H (EX,S(X)) is obtained, for with isolated singularities. consider the complexes First, 7 : X - X -I 1 Let be a resolution of singularities, as constructed by Hironaka. w mS*EX(X). (S(X)) is a divisor with normal crossings, i.e., given by local equation (zl,...,zn) We may thus assume that z 1 ...zj = 0, where is some appropriately chosen local X 77. coordinate system on X. By the blowing-down theorem of the previous section, : mSE(X) Asr + mDE"(X) X is an isomorphism, where D = w-1(S(X)). H (mSE(X)) ý)) = H*(mDE(X)), Thus, s since is a map of X complexes. Now the complex mDEZ(X) X is the complex of global mDE'. sections of the complex of sheaves -- exact sequence: E. X Consider the _A 0 + mDE - X The complex X EX + X - EX 0. X,D is, of course, exact at the stalk level On degrees > 0) - this is just the Poincare lemma - and on is a fine resolution of the constant sheaf E. X X. The complex mD*E, C is clearly also a resolution X of C at points not in and let such that (z1 ,...zn) D D. Let y be a point of D, be local coordinates centered at is defined by z ... zk = 0, locally, and y 78. z j = x j + iy, where each being real coordinates at near may be written t xj ' s the y. and yJ' s Any smooth s-form w = E a , dxI dx w where I I,J and J are increasing multi-indices, each ranging over 1,...,n, mD-E" X the of III + IJI = S. and near a y, If w is a section of and we write out w as above, each of 's has vanishing Taylor series at every point D. It is,therefore, easy to see that r aI j = the n 's j H ((x S j=l is IJ where we may choose a uniquely determined smooth function. r kN Write 2) to be arbitrary non-negative integers, and IJ then yj2)nj n2 ) 2 + (y N + (yj ) 2 N, and set a = be the ball of radius c about 2 R= ((x N N*a J=1 Now, let in xj . B yJ-coordinates, E'(B ) the complex y The Poincare lemma says that is exact except in dimension 0, X where its cohomology is exactly the constants, C. More explicitly, the lemma is proved by constructing operators ks : Es (B X ) Es - X (B E) as follows: In a 79. = III + IJI = s, define ks(a ,Jdx Idy) 1 (I 1 j(tz)t a 0 z dt)( k+x k (-) x k dx J m+k+lk dy + (-1) k=1 I I-i where k dx = dx i ... dx k. k-1 dxik+l k+l dx ... dx , and Extend this definition linearly and check that on Es(B ) X Es(B ): X Id , for s > 1 dksS + ks+d s+1 where 6y Co X aI J = = ks ra Id-6y, for s = 0, is the Dirac function at w E mDEs(BE) N , J, o I,J I for every E k (a jdx IdyJ), s Js I,J d,J J dxI dy , a N. each y. If then each Considering ) = k (a s jdx dy TI 3 is a smooth 1 (f 1 a (tz)tS- 1 dt)ti j) s-I form. Finally, x k=1 1I dy J-J similarly for to all of J = (J al,Jdxldy , I = (il,..., i) basic s-form where I -k dx dy ) 80. 0 a I j(tz)tS-ldt = , = 0 1 4N(tz)a• ,j(tz)ts-ldt 1 t 2Nr N(z)a N (tz)ts- dt 1 = Since kS N is arbitrary, CO q maps m t 2 Nr+sa NJ(tz)dt. (z) summing over 00 . E (B ) into m Sco0 mDE(B ), X is exact in all dimensions. e + 0, (B ). Since 6 X identically zero on as shows that -E X I,J rDEL(B ) the complex X is Taking direct limits over c, of the complexes considered, gives that the complex of sheaves oo mDE is exact at the stalk level X indimensions > 0,and the sheaf itresolves in dimension 0 is = the constant sheaf C X-D extended by to all of 0 C on X - D X. Consider again the exact sequence: 0 CO mDE: M E E: X X - E B XD - 0. (") 81. Hence, for y e D, 0 E" mDE + X,y E" X,y + 0 is exact in dimensions greater than in dimension 0. Thus, E_ an X,D,y The sequence of cohomology shows that sequence. is E0 X,D,y 0, and has kernel C is a fine resolution of X,D = constant sheaf of X. C on D extended by 0 to all Taking global sections and passing to cohomology on (*), gives ... Hi(mDE(X)) X Hi(E: (X)) X,D Hi(E(X)) R X ... By the general result on fine resolutions of sheaves, Hi(E'(X)) X - Hi(X;C) 1i(E" (X)) = Hi(D;C), and X,D and therefore, from the long exact sequence of topological cohomology for the pair Hi(mD(E'(X)) Hi (X,D;C) In particular, since Hi(mSEX(X)) SX (X,D), it follows that 1= Hi(mDE:(X)) Hi(X-S(X);C) c X H(X-S(X);C). - Hi(msE X)), X( and carries only topological 82. information about the regular points of Proposition 4.1: In particular, for * X (mSE(X)) X. Thus, H• (X-S(X);C). H (mSEX(X)) compact, is finite-dimensional. H (EX(X)) = HDR(X), In order to control remains to find something out about This we are only able to do for it H (E XS(X)). X with isolated X with one singular singularities. Consider the local case of point x . Again, singularities, with crossings. x : 0 r : X -+ X D = denotes a resolution of ) -l(x a divisor with normal Consider the exact sequence of stalks at ) mx *E•, E;,X E~,Xo with a result of T. Bloom ([51, 0. In analogy with prop. 3.1), we show the following: Proposition 4.2 : H (E an isolated singular point of X H (E' X. ) for 83. Proof: mx Ex 0 ,E ) =0. X x0 We'll show H (m o x0 = lim mx E(U), where U runs over a basis Since lim is 0 4. 0 U of neighborhoods of H (m E O xo . exact, we have ) = lim H (m0 E(U)) = lim H (U,x ;C), + o + o U U the last equality by the previous proposition. H (U,x ;C) = 0 are two ways to show basis of neighborhoods of xo . may be triangulated ([17]), call this triangulation T. neighborhood of Tn, x0 triangulation of subdivision of at in X T. H (m. X0 "E X X, w-l(U) U in a One way, you note that with Let xo Un = open star where Tn is the U 's The are a basis of neighborhoods x . ) = 0. Hence, xo, they H (Un,xo;C) = 0, A second way to see this is to 0 U X as a vertex; xo, and since they are star-shaped about note that as in for all by the n-th iterated barycentric are contractible onto and There runs over a neighborhood base of x runs over a neighborhood basis of D 84. in X. H (U,x ;C) Fit = H (7( (U),D;C) into the long exact sequence - Hi ... Taking lim r-1(U),D;C) Hi (U); ) - ... Hi(D;C) preserves exactness: U ... lirm H (Ux v ;C) ) lim H (-i ( 1(U);C) - H1i(D;C) - By continuity properties of sheaf cohomology, however, the limit map lim Hi(-1(U);C) + Hi(D;C) is an U isomorphism, and exactness says It is H (Ex lim Hi(U,x 0 ;C) = 0. ) which can be controlled for an isolated singular point. To do so, let xo 2,x denote the stalk-complex of germs of Grauert-Grothendieck holomorphic differential forms ([4]). We recall here two theorems concerning this complex: Theorem BH ([4]1): ) H (Qx X,xO is finite dimensional, . 85. for x E X an isolated singularity. Theorem B ([51): X0 o X H ( H (' ~ for an isolated singularity. An intermediate complex is needed to compare H (,x ) H (E" ). with Let K" C Q* denote the subcomplex of sheaves defined by KX (U) = {W i (U)0p(w) X for for every U restriction map. sends Ki = 0 open in 2iX(U - Ug 1 S(X))}, X in X, and p The differential d denotes the on 2X clearly into itself. X' forms on X, is the quotient complex First note that each Since i X coherent. is a coherent O -module: X is coherent, it suffices to show KX is To see this, consider a resolution of singularities sheaf W'i X R'/K'. *(Q.i ) X f : X on X. + X, and consider the direct image Since 7 is a proper holomorphic 86. 0, map, and X a coherent 0O-module, a theorem of X Grauert ([ 9 ]1) r,(Q ) implies For an open set U, is coherent on x Q i(T-I(u)), and 7,(Qi)(u) X A)T Xi 7 X i) induces a Qhomomorphism . Now l(Kof OX-modules i(U)) = 0 X . (T-l(U)), - 1 (U), and r-l(U - U n S(X)) since is Qi , in is dense in the sheaf of germs of holomorohic sections of a vector bundle on a manifold. isomorphism, if then Conversely, t : T-1 (U - U n"S(X)) -+ U - U r S(X) since w = 0 Thus, K If it (w) = 0 U - U n on i > dim X. singularity is an - Urn S(X)), w e Ki(U). and hence is coherent. is a regular point, it follows directly S(X). 1 Ki = 0, and X,x Note also that Again, suppose xo . -1(U ), T from the definition that supported on on S(X), which says = ker ( x e X X. Since Ki X i K X = 0 is this for has one isolated is coherent and supported 87. at xo, we have, by the analytic Nullstellensatz, that K 0 is a finite dimensional vector space over Xx'0 for each i, (Note that K• Xx dimesion o imbeddig Xax i > imbedding dimension of X for X,x at C, x .) Consider the short exact sequence of complexes: 0 + K'Xx 'o -+*Xx ' + Xx (#) + 0 Taking cohomology, the finiteness of KX X, xo and Theorem BH above together imply: Theorem BH': for x0 e X H (0•" Xx ) is finite dimensional, an isolated singularity. One may also pass to completions as in sequence (#), 0 Xx -modules and get induced differentials in the limit such that the following diagram commutes: 0 - K Xx -+ 0 ,x II 0 -+ Ký '0 - '" X,x -+ 0 (I ') 4.x X,X '0 XIx 'O + 0 88. Here we've used the exactness of the completion functor ([11 or [221), and the fact that some power of the maximal ideal of OX K annihilates Xx0 X• Xo. the inverse system for the completion of so that K~ , consists XxO of isomorphisms beyond a finite stage. Taking cohomology, using Theorem B above and the five lemma, H (5 Theorem B': Xx ) = H (I " Xx 0 ), yields: for x e X an isolated singular point. Xx C Clearly, 'X 0 on X at Also, Aso• xo xo where E wher E , 'X = germs of forms of type (i,0), as defined in section 1. -Ii where the bar denotes complex 0X is a module over ~X,Xx S germs at X' X ,x Cxo C EXo,i Ex 0 ' Xxx °0 x0 conjugation. Eio = ring of x 0X o of anti-holomorphic functions on : is a natural map 1 X,x 0 Xx +E^j Xx '0 ' X. There where the tensor product is just the algebraic tensor product of two C-vector spaces. 2 EX, The map 4 takes wl and then projects them into "2 to Xo 89. where W whereXx 2 Note that the X,x Qi 's Xx are all modules of real analytic forms as in section 1. We wish to show that if the map n : X let ¢ X - is locally irreducible at just defined is injective. xo$ To do this, let be a resolution of the singularity at nw-1 (x). x le0 X xo, and Consider the following diagram: ÷EAi ,9J Xx xi Xx 0 xJn * W* •i• xi+j * + ® X,x o XIX X,Xo As in the proof of Theorem 1.3, each vertical map is injective, and the top horizontal map (defined as was above) X is injective, is non-singular. as is immediate to see, Consequently, an analogous The space ^ij EXx : Q 1 xo 0 since 4 is injective. 'J X o A There is 0 ^iJ X, x may be naturally considered a Frechet nuclear topological vector space (cf., section 1 above). For the same reasons as there, the spaces 90. i 3 X,x o0 and sort. also have natural topologies of this X,X0 0' X,x If we endow r-topology or completion, with either the Xx e-topology, we therefore get the same ^ 0, 1 denoted which is then also x Xo' ox a Frechet nuclear topological vector space ([21]). an example, as noted in section 1, for at xo, with local coordinates at Xo = E , ° Exn x o' = c[[zl",..,z = oX,x non-singular (zl,...,zn) = C[[zl,...,Zn; A A X 1, As centered Zl,...,Zn]], x X,xo = c[[Z1 ,..., 11], each with the topology of pointwise convergence of It's coefficients. A AA Sgo O easy to check here that Xx O0 , and generally from this fact it = EX follows that O Q Xx 0 0 o fX O X~xo that this is also true for least if S: oi X,x o X is =^E ,j Xx x irreducible at Xx X,x0 X,x •X O We wish to show 0o singular in x . X, at The map is continuous, and consequently, 91. A we get a map : X,X X,x ' 9x0 O Proposition 4.3: xo, then : X If ^ at ^iXEx$,x j A is locally irreducible ^ Xx '0 X,x o also continuous. o + ^iXJ i9 Xx is a topological isomorphism. Proof: Embed X locally at A subvariety of a polydisc x = 0. x in some as a closed CN , and assume Consider the following commutative diagram A,o % A,o +restriction A^Ao AO• 4- Q AIs 00 •0j Al +restriction +restriction Ai,j X,o X,o X,o X,o X,o All three vertical arrows are onto: the left-hand side was shown to be so in section 1, the right-hand side is elementary, and the middle one follows from exactness properties of tensoring topologically with a Frechet nuclear topological vector space [10]. that Ai QAo Al,o ^X, Xio (Note first is surjective, by exactness of 92. 0A,o modules [22].) completion for finitely generated A By commutativity,$ Amapping 4 mapping theorem, 'j X,o ,i ® Xo above. is theorem, open. But 4 and Xo X,o0 ,i X,o -,j X,o restricted to 4), which was shown to be injective is just 4 By lemma 1.2, is an isomorphism. A I Consider the double complex At QXO X,o the differential from A is d 8 1, and from Sit i Xo X,o A^ AI 0'i 0 X,o X,o (X' X,o where ' ' i+l oX,o j + ti ^ 0 A X,o to Xj 0 to X,o i (-1) By the open A ,i X,o dense in must be surjective. X,o ^ A j+1 X,o A @ d. 4 The isomorphism A of complexes : A above is an isomorphism A " @0-' X,o X,o E considering the X,o double complex as a single complex in the standard way. C Corollary 4.q: 0 E X A H (Ek ) H(^" X,o ) an isolated singular point, and irreducible at Proof: X H ( " ), X,o for locally 0. First of all, it should be noted that all the differentials for the complexes above are continuous 93. in the Frechet nuclear topologies. and B', each of Theorems BIH' Secondly, by H ($"') X,o and H (A' ) X,o (which are isomorphic, of course) are finite dimensional. ^ Consequently, the differentials of the A 0 X,o ' and X,o 0' complexes are homomorphisms of Frechet nuclear topological vector spaces (i.e., they each have closed But then the corollary follows by Grothendieck's images). KtInneth formula for such complexes of topological vector spaces [101]. As noted in the proof above, we have the Corollary 4.5: 0 H (E o ) an isolated singularity of irreducible at is finite dimensional, for X, and X locally 0. Next, a simple induction procedure will show that the last corollary is true, for any isolated singular point. First, we prove a formal analogue of the Nullstellensatz. X Let sets in a neighborhood such that X /) Y = {0}. of germs of C which vanish on and A Y be complex analytic of the origin Let JX C EA 0 in CN denote the ideal functions on a neighborhood of the origin X; Jy is similarly defined for Y. 94, 00oo As above, is the ideal of germs at moEA,o 0 functions ) is with all derivatives at the origin vanishing. Proposition 4.6: + Jy + ME ,o /(JX E finite dimensional. Proof: 0 A,o First, let defined by X IX, Iy denote the ideals in The analytic Y, respectively. and 0A Nullstellensatz says that o/(Ix + Iy) dimensional, and hence, so is 7A o/(TX + is finite Again, y). denotes complex comjugation. E Next, Ao /(JA r + J J 00o Y + mEE 0 A,0 A ) E A A, /(Jr + J) A Y -- where J X = cp eX i,o X = J A X X and likewise for EAo/(IX + Iy + Let n o A ,o /mEA o X is the +Eo Ao X 6o Y, it suffices to show X + ) E A, is finite dimensional. be chosen large enough so that m(OAo )nc IX + Iy coordinates for IX + Iy. + mEA N C , If (zl,...,zN) therefore z1 are the standard 1 - N ...zN is in Consequently, any monomial in the zi's and 95. zj's of degree > 2n + -Y)'EA,o. + X + I Since such monomials generate 2n m(E , , we have Ato eo the ideal Theorem 4.7: then will be contained in H (E ,o Proof: ) If 0 c X m(E )2n C is an isolated singular point, is finite dimensional. The proof proceeds by induction on the number of local irreducible components of X (I + Iy + IX + Iv)*E X Y XA(o X at is irreducible, we are done by the corollary Suppose S X = V i=l , and irreducible at s > 1, where the X. X i ( Xj = {0}, 0, and 0. If 4 .5-above. are distinct for ifj. s-1 Let Y = U Xi . Since 0 is an isolated singularity i=l of X, we may assume Y t X s = {0). Consider the short exact sequence of complexes: 0 -+ M' (,) - E -+ 0 consists o ofwhich Yoforms germs vanish when restricted E,, consists of germs of forms which vanish when restricted to Y. Hence, there is a short exact sequence 96. o - M' + -+ Q' -+ 0 E* (*)s S,O Now a germ w in is in EX M if there is a S,O 0 in in CN neighborhood of imbedded at to XsS 0 ) (we still consider such that w m(EX sufficiently large. m(EAo )n Y. n for C Mi , We S,O Consider the case i = 0. is onto, for an f E m(EX )n m(Ex + )n.Ei so X is the restriction of a smooth form which vanishes on wish to see that ,Oso CN Since )n s,O we may extend it to preceeding proposition, F = F 1 + F 2 + F3, extension of if n f is As in the large enough, F 1 E JX, where F 1 = 0, i.e., We may assume )n. F e m(E A F2 E Jy, F2 + F 3 F 3 Em (E ,). also is an to a neighborhood of 0, The Whitney A field of Taylor series Whitney field, and is F 3 (x), for x e Xs, in EA,o, i.e., 0 is a regular F3(0) = 0. By the same regular situation argument as in section 3, Theorem 3.3, find a smooth function neighborhood of for x e Xs, for • s , 0 and nd in F ' 3 CN F ' on a N^ such that F 3 (x) = F 3 (x), is identically zero on Y. 97. F3 - F ' Consequently, is an extension of F2 + F ' and is zero on 0 to a neighborhood of f vanishes on Y. F2 + F 3 X, and so Thus, Cn in )ntC Mo m(EX S,O for n large enough. i, For higher )n*E m(E X S,O form may be written as a sum of germs of the S,O f*w, f E m(E X any extension of w simply note that every germ in w restricts to w )n and w S,O E Let W be S,O to a neighborhood of in EX and take 0 e CN , i.e., F an S,O extension of which vanishes on f an extension of f*w Y. which vanishes on Then F*w is Y. The preceeding argument implies that the complex in sequence (*)S above, is vector spaces, H (Q') Since H (M') Hence, ) H (E' s,o Q" a complex of finite dimensional is finite dimensional. is finite dimensional, by corollary 4.4, is finite dimensional, Returning to (*)y, is finite dimensional by induction hypothesis. H (E' o YHence, H (Eo) is also finite dimensionalo Hence, H (E' X~o) isalso finite dimensional, 98. Let Corollary 4.8: X be a compact analytic space with isolated singular points. cohomology of X is Then the C -deRham finite dimensional. As earlier in this section, consider the Proof: short exact sequence 0 Since Since + + E ES S(X) E E S (X) = XS xeS(X) X~x 2 S(X) dimensional. isomorphic to dimensional. of msE X 0. H (E~ S (X)) XPS H (E = xcS(X) ,). X is a finite set, this last space is finite As noted earlier H (mSEX(X)) is naturally H (X,S(X);C), which is also finite Hence, HI (EX(X)), the C X, is finite dimensional. deRham cohomology 99. §5. Miscellaneous Applications; Operators on Curves This section deals with miscellaneous applications of the blowing down results of section 3. The first half contains an extension of Malgrange's theorem on ideals of differentiable functions to varieties with It also contains a characterization isolated singularities. of holomorphic functions on a resolution of singularities of a variety with isolated singularities come from holomorphic functions on the variety: this is a direct consequence of §3 and a theorem of Malgrange [191. These results would follow for varieties with arbitrary singularities, if the blowing down conjecture in §3 were true. The second half deals with results which are more exceptional in nature. It is shown that E(X), for X a curve, is complete in the topology generated by differential operators on X. In fact, the method works for a variety with isolated singularities and a finite resolution (which will then not be the Hironaka resoltuion, in general). in dimensions > 1. Finite resolutions are rare In the considerations here, finiteness of the resolution map is used to construct a large family of differential operators on the variety in question. 100. A. The blowing-down technique of section 3, together with a result of Malgrange, shows which holomorphic functions on a resolution of singularities blow-down to holomorphic functions on the resolved variety, if the original variety has isolated singularities. Proposition 5.1. Let X be a complex analytic space n : X -+ X with isolated singularities, and let resolution of singularities. function on X, then function on X, if and only if If f condition with respect to the map Remark: is a holomorphic f = 7 (g), with f be a g a holomorphic satisfies the finite-point 7. This proposition clearly has interest only in the case where the isolated singular points of X are not normal points on the variety, since for such points, any holomorphic function on X blows down to a weakly holomorphic continuous function, and normality would say that such a function is, in fact, holomorphic. Proof: a g i :: XX-- on X One simply applies proposition 3.4 to find such that -l(S(X)) (3(X)) + XX -~ f = (g), and S(X) S(X) is is aa holomorphic holomorphic isomorphism, isomorphism, g e E(X). Since 101. g is holomorphic on the regular points of a result of Malgrange ([19]) X. However, says that such a function g is actually holomorphic. Next, we show that Malgrange's theorem concerning ideals generated by finitely many real analytic functions is also true in the ring E(X), for X with isolated singularities. Proposition 5.2. Let be real-valued, gl'*'*''s real analytic functions on X, isolated singular points. a complex variety with f e E(X) A function may be A written A f = hlg 1 + ... + hsgs, in Ex(X), with A hi C E (X). f (i.e., is in the ideal generated by if it is so formally at every point.) gl,**.gs A Furthermore, if that hi(x) = 0, Proof: is local on = = ,...,s, in h i's so Ex(X). A partition of unity shows that the question X, so we may assume point, call it f() f(x) = 0, we may choose the 0. A f(0) = h 1 g(0) + ... X has just one singular By assumption, A A(0), + hgs (0), for some for some hi, i hi, i = 1...,s$ 102. in Eo(X). Let hlo ,...,h A A and so that f - h 1 0 be chosen to be real-valued h (0) = hi, i = 1,...,s. _ s 0 gs = f' Then still satisfies the A assumptions of the proposition, Let 7 : X -> X and consider The gi's A f'(0) = 0 e E (X). and be a resolution of the singularity at f' = T (f'), gi = are real analytic, and to the ideal in E(X) i = 1,,...,s, (gi), i f' generated by the r . Furthermore, w- 1 (0). Taylore series along gi's f' E(X), = hlgl f' such that furthermore, we may assume the Taylor series along hi's -1 (0). may all be written hi's in E(X), and f = hhg1 + ... h (0) + hsgs, + hi ' s has vanishing hl',...,h s +"' + hs'gs, and hi's have vanishing By Theorem 3.3 in section 3, hi = 7 (hi), = 0. with at every By a theorem of Malgrange ([201 Chap. VI, Th. 1.2'), we may find in on X. belongs formally point, as is seen simply by pulling-back the above by means of 0, i = l,,.,,s, Finally, h 2i = h + hi, i = 1,...,s. with 103. The proposition above is also true for complex valued, real analytic functions, and for submodules of E(V) = C sections of C vector bundle V over X, since theorems similar to the Malgrange theorem quoted above are true for these extended situations. The point of the method is made by the example given. Clearly such results go through for more general provided an adequate blowing-down theorem for 7 X, : X -+ X is known. The following result will prove useful later in this section, and is a natural complement to theorem 3.6 above (cf. [71). Proposition 5.3: theorem 3. above. f Let : X' sf *(E Then s + X (X)) be as in is closed in Es(X'). Proof; is The subspace Xsf *(Es(X)), by proposition 3.4, determined as the space of all forms satisfying the finite point condition for points of X' ¢ : E s (X') 4 such that t ( Es , (X') i=l xi f. Let x1l ,...,xt t be x = f(x 1 ) = ... = f(xt). be the natural projection, Let 104. which is continuous and onto. A f (Ex(X)) Si= C it E. call it Then t ^. O Esx ,(X') i ~-1(E) By proposition 1.4, is a closed subspace, call is a closed subspace of V(x 1 ,...,xt). ES(X'), w e ES(x') To say that satisfies the finite point condition is to say that w{E IV(x 1 ,...,xt)If(x 1 ) = ... is equal to Since this Xsf*(Es(X)), and is also closed, the proposition is B. = f(xt)}. proved. We'd like to look now at a simple observation concerning differential operators which, in light of the blowing-down result above, will provide us with "many" differential operators for some special spaces 7 : X - X be a resolution of singularities, X. X Let with isolated singularities. Proposition 5.4: Let D' : E(X) differential operator on X. D' differential operator on X D - be a blows down to a if D' finite point condition with respect to For a differential operator such as point condition means that, given E(X) X1 , ... satisfies the v. D', the finite ,xt in X 105. operator = w(x t ) = x, the induced formal = ... wT(x) such that A t A D' : $ E. (X) ~ i=1x t ^ of $ E A* ^ S(E (X)) i=1 -LI~I~ + imposed on i=1 i xi into itself. (X) y imply, by proposition 3.4, that D' D' : r (E(X)) - By proposition 2.1, there *(E(X)). is a differential operator D = D' A E, (X) takes the subspace One simply notes that the formal conditions Proof: *1 T t $ on iT D i.e., E(X), on D' X such that blows down to X. An analogous result holds, of course, for real meromorphic operators as in section 2 above, with poles along sums of w-1(S(X)). D 's, D' : E~(X) - The formal conditions concern finite where, for such an operator Q(E,(X)), where Q(E,(X)) of quotients of the integral domain Theorem 5.5: and let w : X a function + Let X f e E(X) X denotes the field E,(X). x be of complex dimension 1, be its normalization. There exists such that for any differential 106. operator D' on Proof: feD' X, blows down to X. The theorem is clearly local on we may look at just one singular point in O, Let X on the analogous sheaf for X. is a short exact sequence of coherent 0 Since 7 so OX + > 7~(0) X + The following OX modules: Q + 0 1 is an isomorphism when restricted to 7 is an isomorphism outside of at 0 alone. annihilates 0, Hence, there is an Q, i.e., n and Q X - w- (0) is supported such that S(0, m(OX,o)n.*(0,)o Choose finitely many generators fl,...,fs of m(OX,o)n o) Ve may have to shrink This implies that f = fl.l + .. 7 (f) = f 0 X about 0 . m(O X $ )n such that their common zeros are exactly the point that 0. X, call it be the sheaf of germs of holomorphic functions OX X; X, 0. to do this.) is the only point where + fs' x vanishes. We want to show has the property that f*E(X) T (E(X)). 107. Let {Xl*,...,m} =.- suffices to show, for every ^ ^ h Eo(X) E ^A* By proposition 3.4, it 1(0). g e E(X), that there is an ^ m A E. (X) 7 (h) cE such that i=1 is equal to x. 1 f.g(xi ). 9 By the choice of the above, we fi's i=1 m know that fi.( s 0.. i=1 f. = SXO A fi'*(O0)o C (OXo), where X Hence, passing to closures, fi ( @ O.. i=l ~ ) C~ (E (X)) C A* mA fi*( C (fi). ) X,x i $ E. (X). i=l Xi A Similarly, e •, ~ ) C7 (Eo(X)). i=l X,x i Hence, ) X,x i mA fC( 0O ~ i=l X,x i ^ U ) X,xi 7 (E (X)), and passing to closures, yields A m fe( $ A A* E~ (X)) C i=1 x0 r (E (X)), and the desired property of is demonstrated. The theorem follows simply by noting that E(X) foD' maps into * (E(X)), or using the previous proposition. The previous theorem says, in effect, that there are "many" differential operators on a complex curve. f 108. One way of measuring how many differential operators there are on an analytic space is to consider the topology on which they generate, and compare it with the E(X) Explicitly, the differential E(X). usual topology on operator topology is given by a family of semi-norms p = p(D,K), where and K D is a differential operator on is a compact set of X, X, and where p(f,g) = sup IDf - Dgl , for f and g e E(X). xeK Denote the topological vector space topology by E(X) with this It is easy to see that the identity D. E(X) map is continuous from this is E(X) an isomorphism. - Since E(X)D, and we ask whether E(X) is a Frechet space, isomorphism would imply that a countable family of the p's above would generate the differential operator topology. One could also define the differential operator topology by a countable family of semi-norms if one knew that the differential operators are countably generated as E(X)-module, i.e., if there exists a family of differential operators {Di}icZ , differential operator and D D = Z fiDi, on E(X), for for every fi's E E(X), 109. all but finitely many identically zero. For an X with isolated singularities, such a countable family exists. We won't use this fact in general, so only sketch a proof: first, note that the underlying real analytic space of a complex analytic space with isolated singularities is coherent (cf. [191). For such a space, the sheaves of real analytic differential operators are coherent, too, and hence finitely generated locally. Finally, a result of J. M. Kantor ([16]) says that the real analytic differential operators generate all C0 differential operators over for X coherent (This last is a consequence of Malgrange's real-analytic. result that E(X), E (Rn) is flat over Ao(Rn): [20], Ch. VII, Cor. 1.12). If the topology of E(X)D were generated by a countable would be a Frechet family of semi-norms, then E(X)D space if it were complete. Finally, if E(X)D is a Frechet space, the open mapping theorem would say that E(X) - E(X)D is a topological isomorphism, and hence the E(X)D topology would inherit several useful properties from E(X), e.g., reflexivity, nuclearity, etc. 110. This chain of hypothetical reasoning proves valid for X of dimension 1. In higher dimensions, Kantor seems to have a counter-example to the "completeness" For dimension 1, the stage of the above argument. completeness follows from the following simple lemma about C functions on the disc. Lemma 5.6: be a sequence oi {fi=1, Let about C" functions on the unit disc A = A(0,1) C1 . the coordinate for If and if, sup xnK K on z = x + iy for all re m,n A, E(A), then the Remark: in C, sufficiently large, 2k 2k r2kDfn - r2kDf m compact in is 0 < and {fi ) for arbitrary D E > 0, a differential operator are a Cauchy sequence in Note here that k E(A). is a fixed, positive integer. Proof: is where Clearly, the only place there is a problem r = 0, i.e., at the origin. For convenience, by radial expansion, we may assume all our hypotheses are satisfied on A' = disc of radius 2 about 0 in C'. 111. This done, we only have to show that the hypotheses imply - sup ID(f provided n,m that course, integers. 2k the r fm) < c, E > 0, for arbitrary n XcT sufficiently large. s+t a D = ' for We may assume, of and s t non-negative The proof simply consists of "integrating out" factor from the estimates of the hypothesis. In what follows, stands for a positive 0(n,m) function, arbitrarily small for n,m sufficiently large; it needn't stand for the same function from line to line, however. sup E.g., our hypothesis is that Ir2 kD(fn = 0(n,m), f for any and we wish D, XCA to conclude that sup ID(fn - f ) = 0(n,m). To prove the desired estimate, note first that, if one can estimate a (f x (fn fm) and (fn - replacing the sequence {Yy- , fm ) {fi } by 0(n,m) on T., then of the hypothesis by ~-i { DRr one can proceed to estimate higher derivatives by induction. Secondly, note that if A1, A2 are bounded or 112. functions on T - or T (0), and if then n - f ) are estimated as X1 Dl(f ) f - + X2 D2 (f D2 and Dl(fn - f m ) differential operators, then if D2 (f D1 0(n,m) - and T on are ~ - or (0), can also be estimated f) Thus, it suffices to show we can estimate similarly. S(f - fm ) for then m 1 (f n - fm) m rhave and 0(n,m) as -- (0), (f SCfn fm on for then we have estimated cos and as e r (f sin 6 (f 0(n,m) on n - f ) n - fm ) a- sin 6 ~(fn r m - Cos r 8 T n - fm n ) m -_ f m (fn (0); hence, by continuity, on f) m T, proving the lemma. So, starting, from the estimates of the hypothesis, rfn we want to show how to estimate 1 F 7TnC(fn- fm) as O(n,m) on T - (0). - f) Thus, and 113. sup XE3 and Ir sup xe• 2k x( f - r 2 k 3 (f T fn = O(n,m) m) - O(n,m). - fm ) By the reasoning used above, this implies sup Ir2k (f - fm )I xcl-(O) sup r2k cos e a(f - fm) + r 2 ksin e (fn - f ) O(n,m) and sup Ir 2 k ( )- (f - f )I xce-(O) sup Ir2kCos 8 af - fm ) r2ksin 0 (f - f )I O(n,m). The same reasoning shows, by induction, that we can estimate (for N > 0, integer): = 114. 2 sk sup (fn Ir2 k aN sup xc~-(O) ar since r2 fm )l - 1 a m ) I = O(n,m), n 2k k aN+l cos e r -N+ a a - + sin e r -JT- 2k aN 2kaN+ (1 a ) rN+1 r 2k = a r 2k aN ax + cos 6 r 2k ar rN 2 aN -sin e r 2k sr ar N ar ar ar and = O(n,m) ar xE2-(O) and N Ir2 k In the last step, one should note, from the induction hypotheses, that if r2kk sup •n m) - = (n,m), 0r XCE-(0) then similarly, sup Ir2k aN 3r XCs-(O) ax the corresponding statement for and is also true. p2k Now integrate For a point in T - r m)' = 0(n,m), x(fn - in place of out of these estimates: (0), with polar coordinates (r,e), 115. •1 - -(rfn(r,e) ar r fm(r,e)) ai i+ - fm(p,6))dp + 1 ap + - fm(1,6))M. ar By hypothesis, i 1 sup T (fn - fm) I= 0(n,m), ar r=1 and thus, i S--(fn ar - fm < I r < / r 0(n m)dp P 2k fm)dpl n - i+ + ap - + O(n,m) O(n,m)(1 + -2ki r where we've used starting from + O(n,m) r2ki 3'2ri+l (f n ar - ) f ) < O(n,m). i = 2k + 1, we then have shown Now 116. a2k+l a 2k+l ar - fm) I (fn 1 < O(n,m)(l + r 2k-r Hence, repeating the process, gives that S2k r Sar 2(f -f <J -1 2k+1 2k+l f n-f m) I dp fn-fm)(1,O) n mn ap r 1 < f O(n,m)(l + -- 1•)dp + 0(n,m) P 1 = O(n,m)(l + r 1 r2k- The process clearly may be repeated inductively, and after 2k repetitions, provides the estimate: 2 S--rar ( (f - fm) O(n,m)(log r- 1 ) + 0(n,m), Hence, integrating once more, Tr- n - fmm I a (f ) < O(n,m)Ir = o(n,m), log r -1 - ri + O(n,m) 117. since - r log r 0 has limit as r The same 0. - procedure obviously may be used to show that 2k 2 k+l Ir2k using that, - fm)J < 0(n,m), I!.(fn 1 1 2 " Hence, the lemma ( < 0(n,m). - is proved. As a consequence of the lemma, it is easy to deduce the following theorem. Theorem 5.7: analytic space, X Let W : X - be a 1-dimensional complex X Then the normalization. there is a countable family of differential operators {Di) on X which blow down to the usual topology of Proof: a function X, E(X). Near a singular point f = f theorem 5.5 above. and which generate x0 of X, choose as constructed in the proof of Let F be a function in E(X) which has the following properties: (1) (2) for every x0 E S(X), a neighborhood of x . the only zeroes of F F agrees with are the xo 's fx in S(X). 118. Proposition 5.4 says that for any F = 7 (F). Let D differential operator {(FD proven if we show that of the topology of except in i) generates the topology which generate {Di) for a family of operators E(X), F-D the operator Hence, the claim of the theorem is X. blows down to X, on There is clearly no problem E(X). a neighborhood of a point Near such a point, write x 7-1(S(X)). in 2k F = r -g, where ie z = re = x + iy g(x) X 0: and where form of is a local coordinate centered at F x, this follows from the explicit given in a neighborhood of w(x) in the proof of theorem 5.5. Hence, we only want to show that operators of the form r 2kD on C generate the usual topology functions in a neighborhood of x. But this is, of course, precisely what the previous lemma states, and the proof is complete. Corollary 5.8: E(X) D is complete, for X of dimension 1, and hence, the topology generated by differential operators is the same as the ordinary topology on E(X). 119. Proof: {Di} and let X Let : X i on Let X, {Di } i.e., and generate the topology of X be the corresponding family of operators 7 (D h) = Di(7 h), i > 0. and every be the normalization again, be a family (countable) of operators on which blow down to E(X). X - {fj. If for every h e E(X), E(X) is a sequence in such that = 0(n,m), sup IDi(fn - fm)l XEK K for any X, then it follows that compact sup IDi(f 0(n,m), - f) xcK for any K that the sequence it has a limit f e T (E(X)). {fj} If E(X) in f e E(X). j, and since where f Di's, the last By definition of the every X, compact in Since n (E(X)) f = T (f), j = f, for every J. 3' estimate says Hence, is Cauchy. f. E for (nE(X)), is closed in E(X), f e E(X), + E(X)D, which proves completeness. f. f in Since we only used 120. a countable family of mapping theorem says isomorphism. Di's, as noted earlier, the open E(X) -÷ E(X)D is a topological 121. §6. Two Examples This section is devoted to two simple examples of we calculate the DeRham some of the foregoing theory: cohomology of the two simples singular plane curve, and examine the possibilities for Hodge operators on these curves. The two curves we shall look at are the projective plane curves X1 , X2 X1 : y 2 given by the affine equations: p2 2 = x (x - 1), 2 = x3 . 2 Each has the origin 0 E C in the above affine representation as its only singularity. simple double point at the origin, and Pictorially, we have: X1 X2 has a a cusp. 122. lemma at the origin: satisfies the Poincare X1 It is easy to check that in fact, locally at 0, X1 looks like two straight lines intersecting transversally, L1 call them given by any pair of f2 fl fl(0) = f2 (0). conditions on a pair of smooth forms W2 together" to a smooth form Ei = Ei L1, fl for L1 E EL 1 •Oo and w for , E L2,o ,1 is a closed 1-form, then for EX 1 o L1 and There are no wl on LI and L2 . X 1 . Thus, on i > 0. wl = dfl with If and w = wl w 22 m2 = df2' lemma f2 e EL 22 9o, by the Poincare If we choose f 1 ( 0 ) = 0 = f 2 (0) (we may change them by constants), f on of degree greater than zero to "piece L2 on is X1 function on functions CO such that L2 on A Cc L 2. and fl L = fl is, therefore, equal to and fl then there is an = f2' and df m. Since the Poincare lemma holds at all points of Xl, the smooth DeRham cohomology of X1 is just the 123. topological cohomology, dimensions although H (X1;C), 0, 1, 2, and 0 HDR(X) which is C in for higher dimensions. Thus, in general is "bigger" than the topological cohomology of X, it is not "big enough" to regain for singular varieties either Poincare Duality or the Hodge (p,q)-decomposition for the cohomology of a projective manifold: that DR(X1) for HDR (X) we may not write HI,o(X1) = Ho,1(X2 H•R(X) dim HDR(X1) HDR(X = is one implies is zero and clearly 1 ,0(X 1) @ Hol(X1 ) with . In the case of non-singular manifolds, Poincare duality is closely related to Hodge's *-operator, which is, in effect, based on local or infinitesimal Poincare duality in the exterior algebra of a vector space. The *-operator, too, is very closely related to the standard adjoint of the exterior differentiation operator d. Perhaps, in general, one should not expect to be able to create a theory of harmonic integrals (which is basically using an adjoint operator to construct a Hodge decomposition) in a space where Poincare duality 124. In our example is not satisfied. p1, is X1 it looks very unlikely that a "good" or X1 actually an immersion of 2 P , into to construct a formal adjoint for d 2 is we might hope on by means XI This will work on of the induced metric. X1 . X1--X1 C P Since natural adjoint will exist. on X 1 , whose resolution X1 , but not In fact, the picture is the following E(X 1) d :,E1(X 1 -> E2 ( 1) t 7T E(Xl) --> E1(X1) where n (E(X 1 )) d-> E (X1 ) has codimension 1 in the metric mentioned above, we construct and Laplaceans ~i EIi on each gives an operator on E i (X1 ). X1 , For which for in fact, give harmonic representatives for for trivial reasons. For i = 1, Zi since the harmonic 1-forms represent Using E(X 1 ). d s on X1' i = 1 and 2, i = 2 does, H2DR (Xl1) has no kernel, H1 (X1 ;C) = 0. 125. H (X1;C) = HDR(X) But is one-dimensional. The missing kernel which is supposed to represent H (X1;C) comes from the fact that d : (X) may be considered a differential operator on E(X 1) X1 only if we restrict it to acting on a codimension 1 subspace of EI(X1) , namely d-l(n*(E(X 1 ))). A one dimensional complement does not admit the Hodge decomposition on as a decomposition on X1 Xl, and this complement represents H 1 (X1 ;C). It is hard to imagine a more natural attempt at a d and it is harder still to see how to repair this d , so that it alters the Hodge decomposition by a 1-dimensional subspace. The example something about. X2 is a little bit harder to say One can check directly that it, too, satisfies the Poincare lemma at its singular point, or one can do it in the following way. X2 is irreducible at 0, Note first that so that the methods of section 4 apply to show that H (E 2 2,o0 ,( H ( 2X ) X2,o 6 H ( 'i X2,o ) = H (~ X• 2,o ) H ( , .X ), 2,o in 126. the notation of that section. H (Q ) So, we'd like to be zero in dimensions > 0, i.e., in 2,o dimension 1. It is a general calculation of Brieskorn and Mumford ([51, p. 132) which says that dim H1 ( 2 , o ) = 0, of the quotient in 0 2 C , conclude 0 + where Since K' H1 Q2, x 2 .0 2 , (f, -f, f ,o 0. at ) = dimension over C dim H2 (j W )/(-, and X2,o ) = 0, X2,o - X2 To recall the sequence 0, and it is easy to see that K' QX2,o. It is easy to K0 = 0, K 1 = {a(2xdy-3ydx)+bx(2xdy-3ydx)Ia,bEC}, K2 = {a dxdy + bx dxdyla,bEC}. d(2xdy - of the two ideals dim H 2 (nX, 2 ) = 0. 2,o the complex of torsion forms in check that -) is the defining equation of f = x3 -S, y2 , X2 ,o o 3ydx) = 5dxdy it follows that H (K') sequence of cohomology, and = 0. H1 P Since d(x(2xdy - 3ydx)) = 7xdxdy, Hence, by the long exact 2,o ) = 0. is 127. * X2 Thus, for X2 Since to is irreducible at X2 = P'. Thus X2 a - H (X2 ;C). HDR(X2) we also have 0, it is homeomorphic does satisfy Poincare duality. It is not possible to analyze quite so directly the relation of the DeRham complex of X1 as for and X1 and that of X2 Is it possible that there earlier. of d is a special formal adjoint X2 X2 on d which induces a Hodge decomposition on the subspaces of forms coming from X2 ? There are other reasons to believe there might be "elliptic" operators on X 2. X2 has a good "symbol space" in the sense of Bloom ([23]) if t - (t2 ,t3 ) at is a uniformization at meromorphic operator D =-- d 2 - 2d 2 d- 0. For example, 0, then the is an operator dt on X2 near 0 (i.e., it blows down to X 2 ), and its leading term has a non-zero coefficient. D*D O= has a leading term but the author is unable at present a2 , which is elliptic, to handle the singular but the author is unable at present to handle the singular 128. lower order perturbations of this leading term. This operator has the trivial, but encouraging property that it is formally onto at A A : EX2 0, i.e., the induced operator A + EX2 is surjective: in the non-singular case this is an immediate consequence for any elliptic operator of the local solvability properties of such operators. More examples of such operators are given in [15], which has some of the best constructive examples of operators on varieties available at present. 129. Biographical Note Dan Burns, Jr. was born in Brooklyn, New York in 1946. He graduated from Regis High School in New York in 1963 and graduated from the University of Notre Dame in 1967. Since 1967 he has been a graduate student at M.I.T., except for the 1969-70 academic year, during which he taught mathematics at the Boston Public Latin School. He was supported by a National Science Foundation Fellowship from 1967 to 1969. in 1969. He married Anne Cronin of Milton, Mass. 130. Bibliography [11 N. Bourbaki, Topologie ge6nrale, Chap. II, Hermann, Paris, 1961. [2] , Algebre commutative, Chap. III, Hermann, Paris, 1961. [31 T. Bloom, Differential operators on complex spaces, preprint. [4] T. Bloom and M. Herrera, DeRham cohomology of an analytic space, Invent. Math. 7 (1969), pp. 275-296. [5] E. Brieskorn, Die Monodromie der isolierten Singularit.ten von Hyperfl.chen, Manuscripta Math. 2 (1970), pp. 103-161. [6] J. Dieudonn6 and L. Schwartz, La dualite dans les espaces 5 et dZ,9, Ann. Inst. Four. I (1949), pp. 61-101. [7] G. Glaeser, Fonctions composees differentiables, Ann. of Math. 77 (1963), pp. 193-209. [8] H. Grauert, On Levi's problem and the imbedding of real analytic manifolds, Ann. of Math. 68 (1958), pp. 460-472. [9] , Ein Theorem der analytischen Garbentheorie, Publ. Math. I.H.E.S., no. 5, 1960. 131. [101 A. Grothendieck, Seminaire Schwartz, 1953/54, especially expos' 24. 11 , _ El6ments de geometrie algebrique, Publ. Math. I.H.E.S., nos. 4, 8, ... , [12] 1960 - . R. Gunning and H. Rossi, Analytic Functions of Several Complex Variables, Prentice-Hall, Englewood Cliffs (1965). [131 H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic zero, I-II, Ann. of Math. 79 (1962), pp. 109-326. [14] , Bimeromorphic smoothing of a complex-analytic space, mimeographed lectures, Harvard Univ., 1971. [15] M. Jaffe, Thesis, Brandeis Univ., 1972. [16] J.-M. Kantor, Operateurs differentiels a coefficients derivables sur les espaces analytiques reels, [17] preprint. S. Lojasiewicz, Triangulation of semi-analytic sets, Ann. Sc. Norm. Sup. Pisa, ser. III, v. XVIII (1964), pp. 449-474. [18] B. Malgrange, Seminaire Schwartz, 1959/60, exposes 21-25. 132. [19] B. Malgrange, Sur les fonctions differentiables et les ensembles analytiques, Bull. Soc. Math. France 91 (1963), pp. 113-127. , [20] Ideals of Differentiable Functions, Oxford Univ. Press, Bombay, 1966. [21] H. Schaeffer, Topological Vector Spaces, MacMillan, New York, 1966. [22] J. P. Serre, Geom6trie alg6brique et geom6trie analytique, Ann. Inst. Fourier 6 (1956), Spp. 1-42. [23] I. M. Singer, Future extensions of index theory and elliptic operators, in Prospects in Mathematics, Princeton Univ. Press, Princeton, 1971.