BULLETIN OF THE AMERICAN MATHEMATICAL SOCIETY Volume 80, Number 3, May 1974

FOLIATIONS
BY H. BLAINE LAWSON, JR.1
TABLE OF CONTENTS

1. 2. 3. 4. 5. 6. 7. 8.

Definitions and general examples. Foliations of dimension-one. Higher dimensional foliations; integrability criteria. Foliations of codimension-one; existence theorems. Notions of equivalence; foliated cobordism groups. The general theory; classifying spaces and characteristic classes for foliations. Results on open manifolds; the classification theory of Gromov-Haefliger-Phillips. Results on closed manifolds; questions of compact leaves and stability.

Introduction. The study of foliations on manifolds has a long history in mathematics, even though it did not emerge as a distinct field until the appearance in the 1940's of the work of Ehresmann and Reeb. Since that time, the subject has enjoyed a rapid development, and, at the moment, it is the focus of a great deal of research activity. The purpose of this article is to provide an introduction to the subject and present a picture of the field as it is currently evolving. The treatment will by no means be exhaustive. My original objective was merely to summarize some recent developments in the specialized study of codimension-one foliations on compact manifolds. However, somewhere in the writing I succumbed to the temptation to continue on to interesting, related topics. The end product is essentially a general survey of new results in the field with, of course, the customary bias for areas of personal interest to the author. Since such articles are not written for the specialist, I have spent some time in introducing and motivating the subject. However, this article need not be read linearly. 1, 2, 3 and 5 fall into the category of "basic material." 4, 8 and the combination 6-7 are essentially independent of each other. I would like to thank Bill Thurston and Andr Haefliger for making several valuable suggestions for improving the manuscript.
An address delivered before the 78th Annual Meeting of the Society in Las Vegas, Nevada, on January 17, 1972 by invitation of the Committee to Select Hour Speakers for the Summer and Annual Meetings, under the title Foliations of compact manifolds', received by the editors May 7, 1973. AMS (MOS) subject classifications (1970). Primary 57D30. 1 Work partially supported by NSF grant GP29697.
Copyright American Mathematical Society 1974

369

370

H. B. LAWSON, JR.

[May

1. Definitions and general examples. A manifold is, roughly speaking, a space locally modelled on affine space; and a submanifold is a subset locally modelled on an affine subspace. In this spirit, a foliated manifold is a manifold modelled locally on an affine space decomposed into parallel affine subspaces. r DEFINITION 1. By a p-dimensional, class C foliation of an m-dimensional manifold M we mean a decomposition of M into a union of disjoint connected subsets {<&a}aeA, called the leaves of the foliation, with the following property: Every point in M has a neighborhood U and a system of local, class Cr coordinates x=(x x , , xm)\ U-*Rm such that for each leaf oSPa, the components of J7 nj?a are described by the equations xp+1=constant, , xm=constant.

FIGURE 1

We shall denote such a foliation by ^r={^(x}aLGA' It will often be more natural to refer to the codimension q=mp of 3F rather than to its dimension/?. Note that every leaf of fF is a /^-dimensional, embedded submanifold of M. The embedding, however, may not be proper; in fact, as we shall see, it is possible for a leaf to be dense. Local coordinates with the property mentioned in Definition 1 are said to be distinguished by the foliation. If x and y are two such coordinate systems defined in an open set J7<= M, then the functions giving the change of coordinates y^y^x1, , xm) must satisfy the equations (1.1) Syjdxj = 0 for 1 j ^ p < i ^ m in U. Hence, choosing a covering of M by distinguished local coordinates gives rise to a G-structure on M (cf. Chern [C]) where G^GL(m9 R) is the group of matrices with zeros in the lower left (mp) xp block. That is, G is the subgroup of GL(m, R) which preserves the linear subspace Rv={(x\ , * , ( ) , , 0)}^Rm. One of the reasons that foliations interest people in geometry is that they constitute a class of structures on manifolds which is complicated enough to shed light on the general situation but has certain geometric aspects that make it tractable.

1974]

FOLIATIONS

371

Foliations arise naturally in many ways in mathematics and it should be useful to examine some of the important cases. The first and simplest examples come from nonsingular differentiable mappings. A. Submersions. Let M and Q be manifolds of dimension m and q^m respectively, and let f\M-+Q be a submersion, that is, suppose that mnk(df)=q. It follows from the Implicit Function Theorem that induces a codimension-gr foliation on M where the leaves are defined to be the components of/ _ 1 (x) for x e Q. Differentiable fiber bundles are examples of this sort. Note that locally every foliation is defined by submersion. B. Bundles with discrete structure group. Let M-+VP be a differentiable fiber bundle with fiber Q. Recall that a bundle is defined by an open covering {Ua}aeA of P, diffeomorphisms K'.TT~1(U0)->U(XXQ, and transition functions gafi : Ua n Up->Diff(Q) such that ha o h~^\x, y)=(x, ga/j(x)(y)). If the transition functions are locally constant, the bundle is said to have discrete structure group. Note that under this assumption, the codimensionq (</=dim Q) foliations of 7r~1(U0) given by the submersions 7r _1 ((7 a )^ Ua x Q->Q fit together to give a foliation of M. Every such bundle can be constructed in the following way. Let 9?:7r1(jP)-^Diff() be a homomorphism and denote by P the universal covering space of P. Then TT^P) acts jointly on the product PxQ, and we define M=PxQl7T1(P). The action preserves the product structure, and so the product foliation of PxQ (arising from P x Q-+Q) projects to a foliation of M. This is the foliation we described above. Note that each leaf looks like a many-valued cross-section of the bundle M-^>?P. In fact, 7 restricted to any leaf is a covering map. To see this note that if J5? T is the leaf corresponding to Px{x}<^PxQ, then JSP^P/T^ where Tx= {gE7r1(P):(p(g)(x)=x}.

FIGURE 2

372

H. B. LAWSON, JR.

[May

The simplest example of a bundle of this sort is the Mbius band, M= RxRjZ where Z is generated by the map f(x,y)=(x+l, y). The lines y=constant project to a foliation of M by circles as in Figure 2. Note that the circles corresponding to y= c for c^O must go around the band twice before closing. The most common examples of bundles of this sort are flat vector bundles. In fact, any principal (/-bundle with a flat connection (cf. [KN]) is a bundle with discrete structure group. The vanishing of curvature is exactly the condition that the horizontal planes be tangent to a foliation. In a semilocal sense every foliation is a foliation of this sort. Specifically, the normal bundle to a leaf inherits a natural flat connection and corresponding discrete structure group. The resulting foliation of the normal bundle is the "first order part" of the foliation in a neighborhood of the leaf. In particular, the holonomy of this flat connection is the linear part of the "holonomy" of the foliation along the leaf (an important concept due originally to Ehresmann, cf. 8). C. Group actions. Let G be a Lie group acting differentiably on a manifold M. If we assume the action is locally free, that is, for each x e M the isotropy subgroup Gx={g e G:g(x)=x} is discrete, then the orbits of G form a foliation of M. When G is not compact these foliations can be quite complicated. A simple case of this type arises when G is a subgroup of a Lie group G'M, and the action is left multiplication. The leaves are then the left cosets of G in G'. If, for example, we let G=R be a noncompact 1-parameter subgroup of a torus (a line of irrational slope), then every leaf of the resulting foliation is dense. Related to this discussion is the notion in differential topology of the rank of a manifold. This is defined as the largest n such that there exists a locally free action of Rn on the manifold. (Alternatively, for compact manifolds, it is the maximum number of pointwise independent, commuting vector fields.) The determination of this invariant generally depends on a deep study of foliations. D. Differential equations. A foliation always appears as the family of solutions to some nonsingular system of differential equations. To study the foliation is to study the global behavior of the solutions. For instance a nonsingular system of ordinary differential equation, when reduced to first order, becomes a nonvanishing vectorfield.The local solutions (orbits of the local flow generated by the vector field) fit together to form a 1dimensional foliation. The study of the global aspects of this foliation go back to Poincar. One can analogously consider ordinary differential equations in the complex case (where dependence on the variables is holomorphic). One

1974]

FOLIATIONS

373

obtains nonsingular holomorphic vector fields and corresponding foliations by complex curves. The first approach to this subject from the point of view of foliations was made by Painlev who considered the important equation: y'=R(x, y) where R is a rational function in y with coefficients holomorphic in x. An exposition of this work and later generalizations can be found in [R3]. While foliations are themselves solutions to differential equations of a particular sort, they also occur in the intermediate stages of solving more complicated systems, where the leaves appear as characteristic manifolds. (See, for example, [STG, p. 135].) They also appear in the famous study, made by Anosov, of the general structure of certain systems of ordinary differential equations. (See [ANO], [AR].) E. Transversal mappings. Within the general category of foliated manifolds there is a class of natural mappings. r DEFINITION 2. Let M be a manifold with a codimension-gr, C foliation s ^F, and suppose f:N-^M is a mapping of class C , l^s^r, of a manifold N into M. Then is said to be transverse to IF if for all x e N, there exists a system of distinguished coordinates (x 1 , , xm) a t / ( x ) on M such that the map cp=(xm~Q+1 o,, xm of) is a submersion in a neighborhood of x. The above condition is independent of the distinguished coordinates chosen at y=f(x). In fact, if ry(^) denotes the vectors in Ty(M) tangent to the foliation, and if * : Tx(N)->Ty(M) denotes the linear map on tangent vectors induced by , then the condition of transversality at x is
that:

Ty(M) =

TV(JO

+ UTX{N).

It follows immediately from the definitions that if/: N^~M is transverse to a foliation J ^ = { ^ a } a e A on M as above, then induces a class Cs foliation *J^ on N where the leaves are defined as the components of/_1(.J?a) for OLE A. Note that codim(/*J r )=codim(J r ). In the special case that is a submersion, we can consider to be transverse to the trivial foliation of M by points. The induced foliation is the one described in part A above. More generally, suppose : N ^ M is a submersion and !F is any foliation of M. Then is transverse to & and Z * ^ is defined. Thus any codimension-^ foliation of M can be lifted to codimension- foliations of manifolds which fiber over M. We shall say that a submanifold N of M is transverse to a foliation &> if the inclusion map i:Nc+M is transverse to &r. To simplify language in the subsequent discussion, we make the convention that the word "smooth" means "of class C r " where r is an integer > 1 which is fixed in context.

374

H. B. LAWSON, JR.

[May

2. Foliations of dimension one. If a manifold M admits a foliation of dimension one, then the tangents to the leaves form a differentiable field of line elements on M. Conversely, every smooth line field on M is tangent to a one-dimensional foliation. To see this, observe that in a neighborhood of any point there is a smooth, nonvanishing vector field V which generates the line field. The integral curves of V (forgetting the parameter) give the foliation in this neighborhood. For future reference, we state this fact explicitly. 1. The one-dimensional C00-foliations of a manifold M are in a natural one-to-one correspondence with the set of C line fields on M.
LEMMA COROLLARY 1. Every open manifold {"open" means no component is compact) has a one-dimensional foliation. COROLLARY 2. A compact manifold has a one-dimensional foliation if and only if its Euler characteristic is zero.

Thus, the only compact surfaces with foliations are the torus and the Klein bottle. However, on these surfaces there is a rich variety of possibilities. There are the foliations of T2=R2/Z2 coming from families of parallel lines in R2 (cf. 1, C). More generally, if/: S^S1 is any orientation preserving diffeomorphism of the circle, we can consider the torus as a quotient T 2 = J R X 5 1 / Z where Z is generated by the diffeomorphism (t9 0)-Kf+l,(<?)). The foliation {Rx{6}}0eSi of RxS1 projects to a foliation of T2. (When is a rotation, we obtain the linear foliation above.) We can then modify this foliation by introducing a Reeb component at each fixed point of (cf. Figure 3 and 4 below). These constructions

FIGURE 3

produce essentially all foliations of T2 up to diffeomorphism. Thus, the study of these foliations is reduced to the study of Diff^S 1 ). It turns out that there is a radical difference here between the differentiability classes r = l and r^.2 [DE]. This difference is reflected throughout the study of foliations. As we shall soon see, there is a major distinction between foliations of dimension 1 and those of higher dimension. It is basically the distinction between ordinary and partial differential equations. The one-dimensional case has inherently more structure and is more properly studied from the

1974]

FOLIATIONS

375

viewpoint of dynamical systems (cf. [SM]). Nonetheless, the deep theorems in this special case often lead to important general results. This has been particularly true of the Poincar-Bendixson theorem and the Seifert conjecture as we shall see in 8. 3. Higher dimensional foliations; integrability criteria. Suppose that ^={j? a } a6 ^ is a foliation of dimension p>l on an m-manifold M. Then associated to F is a smooth field of p-planes tangent to the leaves, which we denote r(^). Consequently, in order that a manifold admit a foliation of dimension p, it must first admit a continuous field ofp-planes (or, by duality, afieldof (m/?)-planes). This is, of course, a nontrivial topological requirement. For example, S5 admits no continuous 2-plane (or 3-plane) fields. However, one might ask whether this is the only obstruction to finding foliations. If we are given a smooth field of r-planes r on M, can we, in analogy with Lemma 1, find a foliation SF such that r=r(^")? The answer in general is no. In order to solve the resulting system of partial differential equations, certain compatibility conditions (which result from the commutativity of second partial derivatives) must be satisfied. The condition can be stated as follows. Let $"(r) denote the set of vector fields V on M such that Vx e rx for all xeM. If for all K, We 2(r) we also have the Lie bracket [V, W] e^(r), then T is called integrable. This is equivalent to the condition that the ideal J(r) of exterior differential forms which vanish on r is closed under exterior differentiation. One of the classical theorems of analysis is the following.
THEOREM 1. r is the field of tangent planes to a foliation if and only if it is integrable.

This theorem is generally ascribed to Frobenius [F], although it has been pointed out by Milnor [M4] (and Frobenius, himself) that it can be found in the earlier work of A. Clebsch and F. Deahna. Observe that if J^ is a foliation of class Cr, then r(^") is of class Cr_1. Unfortunately, if r is an integrable plane field of class C r_1 , the associated foliation is, in general, only of class Cr_1. (When integrating one does not increase differentiability in the normal direction.) It is easy to see that most plane fields are not integrable. So if we are given an /--planefieldr, the natural question (cf. Haefliger [H2]) is whether T can be deformed to an integrable one. That is, does there exist a continuous family of r-plane fields rt9 O ^ f ^ l , such that T0=T and r1=r(^r) for some foliation J^? The answer, as pointed out by Bott, is no. To see why we must consider the normal bundle v(^r)=rJ-(^r) of the foliation, which is defined as the bundle of cotangent vectors which vanish on r.

376

H. B. LAWSON, JR.

[May

(Introducing a riemannian metric, v can be identified with the field of (m/?)-planes perpendicular to r.)
THEOREM 2 (BOTT [Bl]). Let v be the field of normal planes to a foliation of codimension-q on M, and denote by Font*(v)<^H*(M; R) the subring generated by the Pontryagin classes of v. Then

(2.1)

Pontfc(r) = 0 for k > 2q.

Condition (2.1) depends only on the homotopy class of v (in fact, only on its stable equivalence class as an abstract bundle), and must therefore hold for any bundle which can be deformed to the normal bundle of a foliation. The condition is nontrivial. As Bott shows, complex projective -space Pn(C), for n odd, admits a plane field of codimension 2. However, an argument using (2.1) and the ring structure of 7f*(P n (C); R) shows that no such plane field can be integrable. The proof of Theorem 2 proceeds by constructing o n n connection which is flat along the leaves. Condition (2.1) then follows directly from the Chern-Weil homomorphism. A number of results similar to Theorem 2 have been proven. For example, Bott also has a version for the complex analytic case, with a resulting condition on Chern classes. Joel Pasternak [PI], [P2] has improved the vanishing criterion (2.1) to the range k>q for foliations with certain metric properties. (Roughly, one assumes the leaves are locally a constant distance apart.) As a corollary, Pasternak concludes that if an m-manifold M admits an almost-free action of a compact pdimensional Lie group, then the real Pontryagin classes o f M must vanish in dimensions greater then mp. This result is surprisingly false for integral Pontryagin classes. Recently, Herbert Shulman [SH] has shown that there are secondary obstructions to integrability which are independent of Bott's primary ones. He proves that if v is the normal bundle to a foliation of codimension-q, then for all a, /?, y e Pont*(r) (real classesl) with deg(a fi)>2q and deg(j8 y)>2q, the Massey triple product (a, /?, y ) = 0 . In view of these results one might wonder whether there exist reasonable conditions sufficient to guarantee that a plane field is homotopic to a foliation. This problem will be discussed, at least for open manifolds, in 7. 4. Foliations of codimension-one. It should be noted that none of the sundry versions of Theorem 2 apply to foliations of codimension-one. In fact, by generalizing the immersion theory of Hirsch and Smale, A. Phillips proved the following striking result.
THEOREM 3 (PHILLIPS [PHI], [PH2]). On an open manifold codimension-one plane field is homotopic to a smooth foliation.

every

1974]

FOLIATIONS

377

Therefore, in analogy with Corollary 1 above, every open manifold admits a smooth, codimension-one foliation. For compact manifolds this is not true since a compact manifold admits a continuous codimensionone plane field if and only if its Euler characteristic vanishes. (This rules out, for example, all even-dimensional spheres but says nothing about odd-dimensional manifolds.) The principal conjecture, made by Emery Thomas [TS1], is the following analogue of Corollary 2 above.
CONJECTURE 1. A compact manifold admits a smooth, codimension-one foliation if and only if its Euler characteristic vanishes.

As we shall see there is now much more evidence to support this conjecture than existed when it was first made. It is not unreasonable therefore to make the following stronger conjecture (cf. Theorem 3).
CONJECTURE 2. On a compact manifold every codimension-one plane field is homotopic to a smooth foliation.

The remainder of this section will be devoted to these two questions. In the above conjectures the foliations were asked to be differentiable but not necessarily analytic. This is for a good reason. 4 (A. HAEFLIGER [HI]). A compact manifold with finite fundamental group has no real-analytic foliations of codimension-one.
THEOREM

This was one of the earliest results on the question of existence of foliations. The proof, in outline, is as follows. Suppose M is simplyconnected with an analytic, codimension-one foliation. Let y^M be a closed curve transversal to the foliation. (See Lemma 4 below.) Then y is the boundary of a mapping/: D2-+M which by general position arguments can be assumed to be "Morse regular" with respect to the foliation. That is, at any point/? e D2 consider a system of distinguished local coordinates (x 1 , , xn) where x w =constant defines the foliation. Then xn o f has nondegenerate (Morse-type) singular points. It can now be shown that the induced foliation (with singularities) on D2 contradicts analyticity. In particular, the nonanalytic phenomenon of "one-sided holonomy," pictured in Figure 4, must occur.

FIGURE 4

378

H. B. LAWSON, JR.

[May

Theorem 4 says nothing about C00 foliations, and indeed in 1944 (long before Haefliger's paper) G. Reeb had constructed a codimension-one exfoliation of S3 as follows. Consider the C00-foliation of the (x, j)-plane given by the lines x=c for \c\ ^ 1 together with the graphs of the functions y*=f(x)+c'9 1 < X < 1 and c'e R, where has the property that lim^i^!<k)(x)= oo for all k.

FIGURE 5

Consider now the foliation of the solid cylinder obtained by rotating the strip {(x, y) e R2 : 1 ^ x ^ 1} about the j-axis in 3-space. This foliation is invariant by vertical translations, and so we can obtain a foliation of the solid torus where each noncompact leaf has the form of a snake eternally eating its tail. The 3-sphere can be decomposed as two solid tori joined along their common 2-torus boundary. Indeed, if one removes the solid torus of rotation from Rz=S3~{oo} (cf. Figure 5), what remains is homeomorphic to a solid torus minus an interior point. (Consider the vertical coordinate axis as the core circle.) Gluing together two copies of our foliated solid torus gives a Reeb foliation of the 3-sphere. Note that gluing together the two solid tori by different diffeomorphisms of T2 gives codimension-one foliations of all the 3-dimensional lens spaces. Actually much more is true. Combining this idea of Reeb with a paper written by J. Alexander in 1923 proves the following.
THEOREM 5 (ALEXANDER, LICKORISH, NOVIKOV, REEB, AND ZEISCHANG).

Every compact orientable ?>-manifold has a smooth codimension-one foliation.

1974]

FOLIATIONS

379

Before entering a discussion of this theorem we shall make some useful general remarks. We begin by introducing the following important notion. DEFINITION 2. A compact manifold is said to have an Alexander decomposition if there exists a differentiable map F.M-+C such that: (i) The origin 0 e C is a regular value, i.e., there is a neighborhood of A=F~1(0) in which Fis a submersion. (ii) The map f=Fj\F\\MA-+S1 is a submersion. The submanifold ^4=F_1(0) is called the axis of the decomposition. It is not difficult to see (cf. [M2]) that for e>0 sufficiently small, the (tubular) neighborhood T(A)=F~1({z:\z\<8}) of the axis is diffeomorphic to A x D2. In particular A has a trivial normal bundle. Furthermore, the map : M A-+S1 is afiberbundle whosefibere ^ = / - 1 ( l ) is called the generator of the decomposition. One can check that for each 0, the closure of s/0=f~1(ei0) is a compact manifold with boundary ds/e*=A.

FIGURE 6

We can therefore think of Alexander decompositions as obtained by "spinning" the generator se about its boundary ds/=A (the axis). In fact, all manifolds with Alexander decompositions can be constructed as follows. Let se be a compact manifold with boundary ds/ and let d: s/->s/ be a diffeomorphism which is the identity in a neighborhood N(ds/) of ds/. The manifold M is then obtained from se x [0, 2TT] by first identifying (x, 0) with (d(x), 2TT) for xe s/9 and then for each x e ds/ identifying all the points (x, 0), 0^0^2TT. M has a natural smooth structure and an Alexander decomposition with function F defined as follows. Let cp:stf-+ jR+U{0} be a smooth function such that <p=l outside N(dsf) and <p(x) = distance^, ds/) near ds/ (where distance is defined with respect to some riemannian metric on s/). Then we set F(x, 6)=cp(x)et9. Two important examples of Alexander decompositions are the following.

380
EXAMPLE

H. B. LAWSON, JR.

[May

4A. Let M=S2n+1={Ze C n + 1 :|Z| = l} and let p(Z) be a polynomial such that Vp=(dpldZ0, , dpldZn)^0 for Z ^ O . Then F=p\S2n+1 gives an Alexander decomposition of *S2n+1. It is straightforward to see that 0 is a regular value ofF. Therefore we need only check that = Fj\F\\S2n+1A-^S1 is a submersion. When p is homogeneous of some degree v>0, this is easy. At any Z e S2n+1A consider the curve t\-+euZ. ThQnf(euZ)=eivtf(Z) and so maps the velocity vector of this curve at t=0 to a nonzero tangent vector to S 1 a t / ( Z ) . Thus, is a submersion. In the general case we must appeal to a theorem of Milnor [M3]. 1 EXAMPLE 4B. Let niE^S be a smooth, orientable fiber bundle with connected fiber J / 0 . Let a:S1-^E be a smooth cross-section and denote by T(a) a tubular neighborhood of oiS1) in E. Then * f = - T(a) is a manifold with boundary d^S1 x Sn~2 where 7r\{ei6} X Sn~2=eid. We then construct a compact manifold M by gluing & to D2xSn~2 along their common boundary by a diffeomorphism which sends {el0} x Sn~2 to itself for each 6. An Alexander decomposition F: M>C is given by
F{x\ w

- | ff W

for

\<p(z>, for x = (z, x') e Z)2 x S n ~ 2 ,

* G ^>

where cp:D2->R+ is a smooth function which = 1 in a neighborhood of zero and = l / | z | near dD2. In this case the generator stf is diffeomorphic to the fiber <s/0 with a disk removed, and the axis A is diffeomorphic to Sn~2. In fact A = { 0 } x S n - 2 c ^ x S n - 2 . The following is a sometimes useful observation.
LEMMA 2. Le Mfc, A:=0, 1, be compact m-manifolds with Alexander decompositions having axes Ak and generators <s/k respectively. Then the connected sum M^MX has an Alexander decomposition with axis A0#A1 and generator s^sx {connected sum at the boundary).

The proof is straightforward (cf. [DL]). The relevance of Alexander decompositions to foliations is given in the next proposition. We say that a codimension-one Cr-foliation of a manifold M is trivial at the boundary if each component of the boundary is a leaf and if the foliation extends to a Cr-collaring of M by defining the leaves in the collar dMx [0, 1] to be the components of dMx{t} for O r g ^ l . (See [Wl], [L].)
PROPOSITION 1. Let M be a compact manifold with an Alexander decomposition F. If the manifold F'1^) x D2 has a smooth, codimension-one foliation which is trivial at the boundary, then M has a smooth, codimensionone foliation.

19741 PROOF.

FOLIATIONS

381

As above we set A=F~1(0) and choose e > 0 sufficiently small

that: (i) r ^ 4 ) = F ~ H r W x 2 ) 2 , where D2r={z:\z\^r}9 for 0 < r ^ 2 e . (ii) F is a submersion in T2e(A). We now consider a foliation JF of C{0} defined as follows. Let F be a smooth vector field in C{0} such that V= 9/9r outside D ^ a n d F = 9/90 inside Z>*, and let the foliation F correspond to the orbits of the flow. Observe now that the map F\MA is transverse to F. In particular, F lifts F to a smooth codimension-one foliation of M Te(A) which is trivial at the boundary. By assumption we have a foliation of Te(A) which

FIGURE 7

is trivial at the boundary. Consequently, we may join the two to obtain a foliation of M. We are now in a position to discuss Theorem 5. The principal step in its proof is :
ALEXANDER'S THEOREM [A]. Every compact orientable 3-manifold has an Alexander decomposition.

The axis A of any such decomposition is a disjoint union of circles. Thus, A X D2 is a disjoint union of solid tori which, by Reeb's construction above, have codimension-one foliations which are trivial at the boundary. Applying Proposition 1 then proves Theorem 5. The case of nonorientable 3-manifolds is somewhat more complicated and has been done by John Wood. (See [Wl] for details.)
THEOREM 5' (ALEXANDER, LICKORISH, ETC. AND WOOD). Every compact 3-manifold has a smooth, codimension-one foliation.

382

H. B. LAWSON, JR.

[May

Curiously, the relevance of Alexander's Theorem to foliations has been noticed only recently. Proofs of Theorem 5 using surgery methods have been given by Lickorish [LI], Novikov, and Zeischang. The first major use of Alexander decompositions was made when the geometry of isolated singularities of algebraic hypersurfaces was used to construct foliations of higher dimensional spheres.
THEOREM 6 (LAWSON [L]). There exist smooth, codimension-one foliations of S2^ for k=l, 2, 3, .

Wefirstconsider the Alexander decomposition F of S5 given by setting F=p\S5 wherep is the polynomial
PROOF.

(See Example 4A above.) The axis A =F~1(0) is the inverse image under the Hopf map 7T:S 5 -KP 2 (C) of the nonsingular algebraic curve defined by p in the complex projective plane P2(C). By a classical formula of algebraic geometry (genus=\(d l)(d2) where d= degree of/?) this curve must be a torus. Projecting this torus onto one of its factors, we get a submersion A-^S1 and, thus, a submersion AxD2->S1xD2. We may now lift the Reeb foliation of S 1 x D2 to a codimension-one foliation of A x D2 which is trivial at the boundary. By Proposition 1 we then have a foliation of S5. Note. It has been pointed out by D. Tischler and A. Verjovsky that this foliation of S5 can be generalized as follows. Let i:S1xS1-+P2(C)bQ an embedding with homology degree m7*0. (That is, [/(51X*S1)]= mlP^C)] in H2(P2(C); Z)^Z.) Let denote the restriction of the Hopf bundle to the complement of /(S1 x S1). It is straightforward to see that the Chern class of | is a torsion class of order m. Thus, f has a finite structure group, and the total space of f has a foliation given by a closed 1-form. The rest of the construction is as before. In the special case m = l , | ^ (P^QiiSP-xSP-ytxS1. When m=3 we get the foliation above. To get the result for higher dimensional spheres, we establish an induction procedure. We begin by considering the Alexander decomposition of 2n+i gj v e n by ^ e complex polynomial
qn(Z) = Z20 + -2n+1

+ Zl

The axis AnS r\{qn=0} is diffeomorphic to the bundle of unit tangent vectors to the -sphere. To see this write Z e S2n+1<=Cn+1 as Z=Z+/Fwhere|Z| 2 +|r| 2 =l.Then,/7(Z)=|A r | 2 -|7| 2 +2/<Z, 7), and so An={(X, Y)\\X\2=l, |F| 2 =4 and (Y, X)=0}. The difleomorphism can now be explicitly constructed. However, all we need observe for our purposes is that the map Tr:An-+Sn given by 7T(X, Y)=Xis a submersion. This is easy to check.

1974]

FOLIATIONS

383

According to Proposition 1 the sphere S2n+1 has a foliation if there is a foliation of An x D2 (trivial at the boundary). Since there is a submersion 7TXl:AnxD2~>SnxD29 it is sufficient to find a foliation of SnxD2. However, the following was pointed out to me by Alberto Verjovsky. 3. SnxD2 has a smooth codimension-one foliation which is trivial at the boundary if and only if Sn+2 does.
LEMMA

To see this note that 5 w+2 =5' n xD 2 Ui) n + 1 x*S' 1 where the two pieces are joined along their boundary. If there is a foliation of SnxD2, we construct a Reeb foliation of Dn+1xS1 and, putting them together, get a foliation of Sn+2. Conversely, suppose Sn+2 has a codimension-one foliation. Then by Lemma 4 below, there is a circle S1(^Sn+2 embedded transversely to the foliation. There is then a tubular neighborhood TiS1) of S1 in 5W+2 and a diffeomorphism d: 7T(,S1)-*Z)n+1 X S1 which carries the foliation of TiS1) to the natural foliation of Dn+1xS1 by Dn+1x{eie} for 0^0^277. (See [Wl] for details.) We now modify the foliation in the interior of T(SX) by introducing a Reeb component. That is, we construct a new foliation of Dn+1 x S1 which agrees with the old one near the boundary and which contains a copy of Dn+1 x S1 with the Reeb foliation in its interior. This new foliation is carried to T(SX) by d. Removing the Reeb

Dn+1x5x
FIGURE 8

component gives us a foliation of Sn+2-Dn+1xS1^SnxD2 which is trivial at the boundary. We now proceed inductively as follows. 3 fol on S5=>3 fol on Ss X D2=> 3 fol on A3 x D2=>3 fol on 57=>3 fol on S5 X D2=> . In general, we see that if there is a smooth codimension-one foliation of Sn, then there is a

384

H. B. LAWSON, JR.

[May

smooth codimension-one foliation of S2"("-*)+*, for jfc=0, 1, 2, . This completes the proof of Theorem 6. We now turn our attention to the general case of odd-dimensional spheres. To handle this we must strengthen the induction argument. Let us suppose that there exist smooth, codimension-one foliations of S2k+1 for 2^k<n. The above arguments show that if n is odd, there is a codimension-one foliation of *S2n+1. If is even, the Alexander decomposition given by qn fails because the axis An fibers over an even-dimensional sphere. Hence, the key to doing the general case is to find an Alexander decomposition of 5 4 m + 1 for m^2, with axis which fibers over an odddimensional sphere. This was achieved simultaneously and independently by A. Durfee [D] and I. Tamura [Tl]. While their methods were quite different, the results were remarkably similar. They both constructed Alexander decompositions of S 4m+1 , ra:2, having as axis S2mxS2m"~1. Briefly, Durfee's argument is as follows. Consider the Alexander decompositions of 4W+1 coming from the polynomials
2m

p0(Z) = (Z0 + Zl)(Zl + Zl) + 2 Zl

2ra

Pl(z) = z\ + z\ + 2 ABy resolving the singularity and using some results from his thesis, Durfee shows that the axis ^ 0 = ^ 4 m + 1 n { ^ 0 = 0 } ^ 5 2 w x * S 2 w - 1 # 2 4 m + 1 where S 4 m + 1 is the Milnor exotic sphere. It is well known that the axis Ax= 4m+i n {^ i = Q}^_24m+i ^ orientations are chosen properly). Therefore, taking connected sums of the Alexander decompositions (cf. Lemma 1) gives the result. It is interesting to note that while Tamura used methods purely from differential topology, he also needed to cancel a Milnor exotic sphere from the product in the same way. We can now complete the induction begun above. Suppose there are foliations of S2^1 for 2<^k<n. Then there are foliations of 5 2 M X i ) 2 for 0^k<n. There is an Alexander decomposition of S2n+1 with axis A admitting a submersion A-^S2^1 (where 2k \=n or nl). Lifting the foliation of S^^xD2 to AxD2 and applying Proposition 1 gives a 2n+1 foliation of S . Since there is a foliation of S 5 , we get THEOREM 7 (DURFEE, TAMURA). Every odd-dimensional sphere admits a smooth codimension-one foliation. It has been remarked by Milnor that if S2W+1 has a codimension-one foliation, so does every exotic (2+l)-sphere. (See [L, Corollary 6].) The natural question at this point is what can be said about other odd-dimensional manifolds. The first result in this direction was the following.

1974]

FOLIATIONS

385

THEOREM 8 (N. A'CAMPO). Every compact simply-connected 5-manifold has a smooth codimension-one foliation.

The proof goes roughly as follows. In [BA] Barden constructs a sequence {Mk}%L0 of compact simply-connected 5-manifolds, where M0=S2xSz, and shows that every other compact, simply-connected 5-manifold is a connected sum of a finite number of Mk9s. A'Campo [AC] shows that each Mk for k^.1 has an Alexander decomposition of type 4B above. Hence, any connected sum of these has an Alexander decomposition with axis S3, and since there is a foliation of S3xD2, this connected sum has a codimension-one foliation. Suppose now that M=S2xSz#M' where M' has a codimension-one foliation F'. Then M is obtained from M' by removing a tubular neighborhood f of a smoothly embedded curve y:S1^M\ and sewing in a (foliated) copy of Ss x D 2 . If M' is simply-connected, we may assume y to be transverse to 3F' and then modify the foliation in M' 2?~ to be trivial at the boundary, as we did in proving Proposition 1. This produces a codimension-one foliation of M. Applying this procedure inductively handles the remaining cases. We note that this result was obtained for simply-connected, spin 5manifolds also by H. Shulman. One is now led to consider (n l)-connected (2rc+l)-manifolds. Using the geometry of algebraic hypersurface singularities and classification theorems, Durfee and Lawson [DL] proved the existence of foliations on a class of these which bound parallelizable manifolds. These results fairly well exhausted the theorems possible from known classification results, and some classification-free methods were needed. In particular it was natural to investigate the higher-dimensional versions f Alexander's Theorem. An extensive study of this was made by E. H. Winkelnkemper [WI], and we mention the relevant aspects of his work here.
WINKELNKEMPER'S THEOREM. Let M be a compact simply-connected manifold of dimension > 5 . 7/'?0mod4, then M has an Alexander decomposition. Ifn^O mod 4, M has an Alexander decomposition if and only if the signature of M is zero. Furthermore, the Alexander decompositions can always be chosen to have a simply-connected generator se where Hi(s\ Z)0for > [njl] and where the natural map H^s/; Z)->i/t.(Af ; Z) is*-an isomorphism for / < [njl].

This is a beautiful and quite powerful result. It essentially gives an inductive method for constructing manifolds, and it should prove to be a basic tool in differential topology. The application of this result to

386

H. B. LAWSON, JR.

[May

foliations was made by Tamura who obtained independently some partial results of the above type [T3]. Suppose n=2k+l^.7. By Lemma 3 and Theorem 7, if Mn has an Alexander decomposition with axis Sn~2, then Mn has a codimension-one foliation. Unfortunately, we cannot make this requirement on the axis in general. (It cannot be done on SnxS2 for /*>3, for example.) However, when M2k+1 is (k l)-connected, the generator se of the Alexander decomposition is a 2/:-disk with /c-handles attached. By attacking these handles one at a time with clever surgery techniques, Tamura [T4] succeeded in changing the Alexander decomposition into one with S2fc_1 as axis. This gave the following very pretty generalization of Theorems 5' and 8.
THEOREM 9 (TAMURA). Every compact (kl)-connected, manifold has a smooth codimension-one foliation.

(2/c+l)-

I have learned that, using Winkelnkemper's method of proof, M. Freedman [FR] gave an independent proof of this theorem at about the same time as Tamura. We note incidentally, that these last methods apply only to the cases 2A:+1>5. For dimensions 3 and 5 we still need the old proofs. Of course, this theorem immediately implies that there are foliations of a somewhat larger class of manifolds.
COROLLARY 3. Any manifold which fibers over an (n-l)-connected, (2n + l)-manifold, for example, all the classical groups and their associated Stiefel manifolds, have smooth codimension-one foliations.

The proof of Theorem 9 actually shows that every (nl)-connected, (2n + iymanifold is obtained from a bundle over a circle by performing a surgery on a cross-section. (See Example 4B.) Actually, a much larger class of manifolds (but not all) can be constructed this way, and every such manifold has a codimension-one foliation. It is interesting to note at this point that by returning to polynomials in Cn+i we can construct large numbers of distinct foliations on (1)connected (2+l)-manifolds. Consider for example the Alexander decompositions on S 3 given by p(Z0, Zt) = ZJ + Zf. When r and s are relatively prime, the axis is a torus knot of type (r, s)9 and the generator is a compact, orientable surface of genus (r1)0?1) punctured at one point. By taking connected sums, we get an infinite family of interesting Alexander decompositions (and, therefore, foliations) of any compact, orientable 3-manifold.

1974]

FOLIATIONS

387

The analogous remarks apply in higher dimensions where one considers the Brieskorn polynomials p(z)=z%+zl+ Yz\. For appropriate d, the axis is a sphere S2"-1 knotted in S 2 w f l . (See [L] for details.) At the moment Theorem 9 and its corollaries represent the state of our knowledge concerning Conjecture 1. However, one remark is in order here. Note that in the Winkelnkemper Theorem a requirement for a 4A;-manifold to have an Alexander decomposition is that its signature be zero. This requirement is independent of the requirement that the Euler characteristic vanish. For example, P 2 ( C ) # P 2 ( 0 # ^ X S^ftS1 X S 3 has signature 2 and Euler number 0. Thus we pose the following Problem 1. Does there exist a 4A>manifold of nonzero signature which admits a codimension-one foliation? An example of such an animal was constructed in [RI4], but an error has been found in the proof, so the question is still open. We turn our attention now to Conjecture 2. Most of the work on this problem has been done by John Wood [Wl]. To state his theorem we need the following definition. A codimension-one plane field r on a manifold M is said to be transversely orientable if there is a (nowhere zero) vector field V on M transverse to r, i.e., VP $ rP for all/? e M.
THEOREM 10 (WOOD). Every transversely orientable 2-plane field on a compact 3-manifold is homotopic to a foliation.

This theorem is impressive for two reasons. There are an infinite number of distinct homotopy classes of such 2-plane fields on a compact 3-manifold. Furthermore, as one can check locally, most 2-plane fields cannot be C-approximated by integrable ones. In fact, there are 2-plane fields which cannot lie within 90 of an integrable one. For example, let V be any nonvanishing, divergence-free vector on S 3 (e.g. the Hopf field V(Z)=iZ for Z G S 3 c: C 2 ). Then Fis not transverse to any codimension-one foliation. In general, V cannot be transverse to any compact surface 2 embedded in S3, for if cpt is the flow generated by V and if @+ is the component of 5 3 - S having V as interior normal at 3 ^ + = S , then <p t (^+)g^+ for / > 0 . Since cpt preserves volume this is impossible. However, by Novikov (see 8) every codimension-one foliation of Sz has a compact leaf. It follows that the 2-plane field T=V cannot lie uniformly within 90 of the plane field to a foliation. In higher dimensions Wood has proved the following result [Wl].
THEOREM 11 (WOOD). Let M be a compact manifold with a transversely orientable, codimension-one foliation &*. If it is possible to find a family of closed curves transverse to 3F', which generate H^M; Z), then every transversely orientable, codimension-one plane field is homotopic to a foliation. In particular this conclusion holds for all manifolds of the type M==NxS1t

388

H. B. LAWSON, JR.

[May

Combining Theorems 7 and 11 gives the following.

COROLLARY. Let M be a compact manifold which fibers over an odddimensional sphere. Then every transversely orientable, codimension-one plane field on M is homotopic to a foliation.

There are also results for transversely unorientable plane fields. For details see [Wl] and [W2]. However, Wood's general problem is still open. Problem 2. Show that Conjecture 2 holds for any manifold which admits a codimension-one foliation. Note. Since the writing of this article Paul Schweitzer has succeeded in removing the homology condition from Wood's Theorem 11, obtaining, however, only C foliations. Thurston then modified Schweitzer's procedure to produce C00 foliations. Thus, Problem 2 is solved in the transversely orientable case. Appendix to 4. We conclude this section with two items of separate interest. It was mentioned above that the continuous, integrable plane fields on Sz are not C-dense in the set of all plane fields. With our thoughts in this direction it is interesting to note that Rosenberg and Thurston [RT] have constructed a continuous integrable plane field on the 3-torus which cannot be approximated by an integrable plane field of class C2. (More recently, examples of integrable C 1 fields not approximable by integrable C 2 fields have been found.) We end by sketching the proof of the following useful fact (cf. [HI]).
LEMMA 4. Let M be a compact manifold with a foliation SP of codimension-one. Then there exists a smooth closed curve embedded in M transversely to JF\ PROOF. By passing, if necessary, to a two-sheeted covering of M, choose a vector field V transverse to the foliation. Fix any point and consider the integral curve of Vthrough it. Either it is closed or it eventually accumulates on itself. When the self-accumulation is sufficiently tight, the curve will return to the same distinguished coordinate neighborhood and can be easily closed while maintaining transversality.

5. Notions of equivalence. The collection of different foliations on a given manifold is, in general, tremendous, and in order to make a quantitative approach to the subject it is necessary to establish some broad notions of equivalence. We shall list some of the important ones here for future reference. Let e^o and J ^ be two codimension-y exfoliations on an ra-manifold M. A. J^o and J^i are said to be Cs-conjugate (O^s^r) if there exists a dififeomorphism of M, of class C s , which maps the leaves of ^0 onto the

1974]

FOLIATIONS

389

leaves of SFX. If ^ = 0 , J^ 0 and 3FX are topologically conjugate. If s=r and the diffeomorphism is isotopic to the identity, the foliations are completely equivalent. B. ^" 0 and !FX a r e s a id to be homotopic if there is a continuous family of integrable plane fields rt, 0 ^ 5 = 1 , such that r 0 = r ( J r 0 ) and r 1 = r ( J r 1 ) . (We assume that each tFt gives a foliation of class Cr.) Problem 3. Do there exist nonhomotopic foliations with homotopic plane fields ? C. J^o a n d ^ i a r e called integrably homotopic if there exists a codimension-*/, Cr foliation J^ of the product M x [0, 1] which is transverse to the slice Mx {t} for each te [0, 1], and which induces the foliation J^ 0 on Mx{0} and ^r1 on M x { l } (by intersection of these slices with the leaves of J^). Note that if J^ 0 and 8FX are integrably homotopic, they are homotopic. The converse is not true. In fact, if M is compact, then #*0 and SFX are integrably homotopic if and only if they are completely equivalent. (See [M4].) Thus, any two foliations of the torus by "parallel lines" are homotopic but they are integrably homotopic iff they have the same slope. If we drop the requirement that J^ be transverse to M x {t} for 0<f < 1 in the above definition, the foliations J ^ and J ^ are then called concordant. A more general notion than this is the following. D. Let MQ and M be two closed, oriented m-manifolds with codimension-q, exfoliations. Then these foliated manifolds are said to be foliated cobordant if there is a compact, oriented (m+l)-manifold Jl with boundary d^=M1M0 and with a codimension-^, CMbliation IF transverse to the boundary and inducing the given foliation there. The resulting foliated cobordism classes form a group under disjoint union, which we denote J ^ f l ^ . In the case q=l the group operation has an interpretation similar to that of connected sum. Let M0 and Mx be compact, oriented manifolds with codimension-1 foliations and choose embedded closed curves YO^MQ, y^M^ transverse to these foliations. (See Lemma 4 above.) Let JT0 and Jfx be tubular neighborhoods of y0 and yx sufficiently thin that there exist diffeomorphisms^.: t yT fc ->5 1 xZ) m ~ 1 , k=0, 1, mapping yk to S1 X 0 and sending the foliation of Jfk onto the foliation of S1 X Z>m_1 by the disks {r}xD m _ 1 . We now glue M0y0 to M1y1 along the sets ^h7k by the diffeomorphism df^ h of0:^V0y^^iy-i where h:S1xDm-1^S1xDm~1 is given by h(t, *)=(*, (l-\x\)xj\x\) a n d / T 1 o/ 0 is assumed to be orientation reversing. The resulting manifold, which we denote M0icMl9 is again compact, oriented and has a codimension-1 foliation. Of course, the manifold MQ^MX depends on many choices (of curves,

390

H. B. LAWSON, JR.

[May

diffeomorphisms, etc.), however, the resulting foliated cobordism class is independent of these choices. In fact, MoicM1 is foliated cobordant to the disjoint union; so for any ic operation, [M0itM1]=[M0]+[M1], 'mtFlrmtl. As a first approximation to J^ one can study the bordism groups of g-plane fields. For some nice results in this direction see Koschorke [KO]. It has been proven by W. Thurston that ^ Q J i l ^ r 7 | T r , Tr] where r r =Diff[(S 1 ) is the group of orientation preserving Cr diffeomorphisms of the circle. Hence, by a result of M. Herman [HER], J ^ Q ^ O . In contrast it has also been proven by Thurston [Th2], using the characteristic class of Godbillon-Vey (see 5), that for r ^ 2 there is a surjective homomorphism (5.1) &aitl-+R. Hence, even with this highly indiscriminant notion of equivalence there exist uncountably many distinct codimension-1 foliations of three-manifolds. In fact, Thurston constructs these noncobordant foliations all on Sz. Problem 4. Determine the kernel of the homomorphism (5.1). In particular, decide whether it is zero. It has been shown by Rosenberg and Thurston [RT] that the natural map (5.2) &a?tl-+&Q!itl

has a nonzero kernel. In light of the results of Mather on BT\ (see next section), it is plausible to conjecture that ^Q^tX=0. Problem 5. Determine whether J ^ Q ^ ^ O , or at least whether the map (5.2) is surjective. We finally mention the general problem. Problem 6. Determine the structure of ^Q^^ for m > 3 . 6. The general theory. One very effective technique of modern mathematics is that of reformulating a given problem in the category of topological spaces and continuous maps where one has available the machinery of algebraic topology. This approach has enjoyed a certain amount of success in the study of foliations, and has led to constructions which should be of* central importance for many questions in the study of analysis on manifolds. The work is due to Haefliger, and the primary reference for details omitted in the following discussion is his article [H4]. To make a homotopy-theoretic approach to foliations it is necessary to generalize (or soften) the definition. Observe that any codimension-^, Cr foliation (for r > 0 ) on a manifold M can be presented in the following way. There is an open covering {0^iej of M and a family of Cr maps {fihei* where/,.:^.^ is a submersion, with the following property. For each i, j si and each x e 6tC\6j there is a Cr diffeomorphism y^ from a

1974]

FOLIATIONS

391 such that

neighborhood off^x)

to a neighborhood off^x)

(6.1) Si = VU in a neighborhood of x. Furthermore, for any x e 6iC>i&jr\0k we have (6-2) yl^y^oy^ in a neighborhood of fi(x). Our "cocycle" conditions (6.1) and (6.2) guarantee that foliations defined locally by the submersions ft piece together to give a global foliation of M. If we begin with a foliation presented as in Definition 1, we obtain functions as above by projecting the local distinguished coordinate maps onto the last q coordinates. Note that if we set (6.3) %l = dyl for all i,j e I, we obtain the transition functions for the normal bundle to the foliation (cf. Steenrod [ST]). Note furthermore that if each ft above is a local diffeomorphism (i.e., the foliation has dimension 0) then the foliated structure defined is just the differentiable structure of M, and the bundle given by (6.3) is the tangent bundle of M (thought of as the normal bundle to the "point" foliation). We now broaden the definition of a foliation by allowing the local submersions to be arbitrary continuous maps. In this way we can define foliations on general topological spaces. Before stating this explicitly we make one observation (following Haefliger) for the sake of elegance. For each x e (Pi9 y% is the identity map on a neighborhood off^x). Hence, if we know the map x^y^] (=germ of the identity map atf^x)), we certainly know fi(x); and we may replace (x) by y%. Doing this reduces (6.1) and (6.2) to the single cocycle condition (6.2). Denote by Vra the set of germs of local Cr diffeomorphisms of Rq (homeomorphisms, if r = 0 ) . For y e Fra let a(y) e RQ denote the source of y9 and r(y) its target. Whenever, o , (y 2 )=T(y 1 ), the composition y2 o yx is defined. There is a natural topology on Trq such that the inverse map, ^H->7 _1 , and the composition map, where it is defined, are continuous. (The topology is the usual germ, or "sheaf-like", topology.) Identifying a point x eRq with the germ of the identity map at x gives a topological embedding of B? into r ; . DEFINITION 3. Let X be a topological space. A codimension-^, Cr Haefliger cocycle over an open covering ^ = { 0 J e / of Xis an assignment to each pair i,j G I of a continuous map y ^ i ^ n ^ - ^ r j such that for all i,j9 k el Yki(x) = ykj(x) o yH{x) for x G (9^(9^(9^

392

H. B. LAWSON, JR.

[May

Two such cocycles over open coverings tt and U' are said to be equivalent if they extend to a cocycle on the disjoint union ll\J^W'. (We allow &i=&j for ijj in the definition above.) An equivalence class of cocycles is called a codimension-<7, Cr Haefliger structure on X. Note that if J^ is a Haefliger structure on X a n d / : Y-^Xis a continuous map, then induces a natural Haefliger structure *Jf on F. Clearly there are many uninteresting Haefliger structures on any given space, so we introduce an equivalence relation which is sensitive to certain important properties. Two Haefliger structures ^Q and ^ f 1 on X are said to be concordant if there is a Haefliger structure M" on Xx [0, 1] such that jtfk=i*jl? where ik:X-+Xx [0, 1] by ik(x)=(x, k) for k=0, 1. Suppose that Jtf* is a codimension-*/, C r (r>0) Haefliger structure on X, represented by the cocycle {yi0) over W={@i}iei. Then we can associate to J^ a ^-dimensional vector bundle v(J^) over X given by the local transition functions gij(x)=dyij(x) in a neighborhood of x e tf^ntf^. i>pf) is called the normal bundle of ^f. (For r = 0 , one gets normal microbundles.) One can easily check that concordant structures have the same normal bundle. Before proceeding let us examine some examples of codimension-1, C00 Haefliger structures on a manifold M. EXAMPLE 6A. Any codimension-1, C00 foliation of M. EXAMPLE 6B. Any continuous function : M-+R. (%={M} and y = i o where i'.R-^Y is the natural embedding.) We may approximate by a smooth Morse function f'\M-+R so that the resulting structure is concordant to the one given b y / . EXAMPLE 6C. Consider a foliation with singularities given by functions fi'>i~^R, as in the beginning of the section, except that e a c h / is allowed isolated nondegenerate critical points. This gives a Haefliger structure in the obvious way. We remark that every codimension-1, C00 Haefliger structure on M is concordant to one of type 6C. Note that the differentiable structure on Mm is itself a codimension-m, C Haefliger structure on M. The notion of Haefliger structures fits into a beautiful, unified theory. Each T is an example of a topological groupoid. A groupoid is a category in which the morphisms are invertible (and, generally, the objects in the category are identified with the units). A topological groupoid is a groupoid with a topology in which the maps yv-*y-x and (yl9 y'2)^yi y% (on the set where it is defined) are continuous. The following are two important classes of these objects. EXAMPLE 6D. Every topological group is a topological groupoid. (Here the category has only one object.) r EXAMPLE 6E. Let f be a pseudogroup of local C diffeomorphisms of

1974]

FOLIATIONS

393

Rq, for example: the pseudogroup Fg(C) of local biholomorphisms in CQ^R2q9 the pseudogroup f rQ(a) preserving a fixed symplectic form a on Rq, etc. (See [KN, Chapter I].) Associated to f is a topological groupoid r e r j , namely, the germs of elements in F. By replacing FrQ with a general topological groupoid V in Definition 3 we obtain the notion of a Vstructure on X. If T is a topological group, a T-structure on Xh (the equivalence class of) a principal T-bundle over X. Suppose T comes from f as in Example 6E. Then every F-structure is, in particular, a T-structure. For example, every complex structure on a manifold M2m is a r w (C)-structure. However, a T-structure, in this case, is best thought of as a generalized f-foliation (a complex analytic foliation, a symplectic foliation, a foliated foliation or "multifoliation", etc.). Note that if F is defined as in Example 6E, then there is associated to r a subgroup G^GLg(R), defined by taking the differentials of the elements in F which fix 0 e RQ. To any T-cocycle {y{j} we can associate a G-cocycle {gtj} by setting gij(x)=dyij(x) where each gu can be considered as acting at 0 by parallel translation in Rq. This process associates to the T-structure a G-structure, that is, a principal G-bundle which is the normal bundle to the T-foliation with structure group reduced to G. Having introduced and motivated T-structures, we are in a position to state some results. Using categorical arguments (or, more directly, by using a generalization of Milnor's join construction for topological groups, cf. [BL]), Haefliger proves the following [H4]. PROPOSITION 2. Associated to any topological groupoid F there is a topological space BF, equipped with a F-structure J f r , such that: (i) To each F-structure F on a paracompact X there is a continuous map f.X-^BF such that M>=f*jev. (ii) Two F-structures ffl and Jf\ on a paracompact X are concordant if and only if the associated maps f $ and f are homotopic. BF is called the classifying space for T, and a mapf:X->BF inducing a T-structure J4? on X is called the classifying map for Jtf. The proposition states simply that concordance classes of T-structures on X correspond bijectively to [X, BF], the homotopy classes of maps X-+BF. For topological groups this theorem is classical. Let us return to Haefliger structures. The space BFrQ has a restructure and thus for r > 0 a normal bundle. Let v:BFrQ-+BGLQ denote the map classifying the normal bundle. (This map can be obtained also by noting that taking the differential gives a continuous homomorphism FrQ-+GLQ. v is then the induced map on classifying spaces.) Let J^ be

394

H. B. LAWSON, JR.

[May

a Haefliger structure on a paracompact X with normal bundle v(#?). I f / i s the classifying map for ^ and v(f) the classifying map for v(J^), then the following diagram commutes. X f-+BYrQ

BGLq We can now formulate in a homotopy-theoretic setting the general question discussed in previous sections, namely: Q. When is a given plane field homotopic to a foliation! We replace this question with the following. HQ. When can a given map n:X-+BGLa be lifted to a mapf:X->BTrQ such that n=v <>? We shall see in the next section that for open manifolds an answer to HQ gives an answer to Q. Moreover, the question HQ can be dealt with by the methods of homotopy theory. To do this it is necessary to study the homotopy-theoretic fiber FTrq of the map v. (The space FTl can be thought of as the classifying space for Haefliger structures with a framing or trivialization of the normal bundle.) The obstructions to lifting the map n will be cohomology classes with (twisted) coefficients in the homotopy groups of FFrq. Any knowledge of the homotopy-type of these basic spaces BYrq and FVrQ is of enormous interest in geometry. The first results of this type are the following.
THEOREM

12

(HAEFLIGER

[H3]). For l ^ r ^ o o , ir/Frj) = 0

(6.5) for 0^j^,q.

In the real analytic case (r=co), (6.5) holds for

O^j^ql.

COROLLARY 4. If a paracompact X has the homotopy type of a kdimensional complex, where k^q+l, then the concordance classes of TJ structures ( l ^ r ^ o o ) correspond bijectively to the equivalence classes of bundles over X,

We recall that Bott has found necessary topological conditions for a bundle to be equivalent to an integrable one (Theorem 2). One might expect this condition to carry over to the homotopy-theoretic case, i.e. from question Q to HQ. Indeed, it does.
THEOREMH2.

Themapv*:Hk(BGLQ;

R)->H\BTQ; R) is zero for k>2q.

The fascinating thing is that over finite fields quite the reverse is true.

1974]

FOLIATIONS

395
For q^.2, the map

THEOREM 13 (BOTT AND HEITSCH [BHI], [HI1]).

v*:H*(BGLQ; ZP)-^H*(Brq; ZP) is infective for any prime p. Hence, v* is also infective on integral cohomology. Combining Theorems H2 and 13 gives
COROLLARY 5 (BOTT-HEITSCH). For q^2, the group H^^BT^ Z) is not finitely generated for 2k>q. Thus some homotopy group TT^FY^ is not finitely generated.

Thus, as one might have expected, the spaces BTrq are tremendous. The theorems above do not apply to the codimension-one case. There is much known, however, about this case, and the results vary drastically with the differentiability. The first theorem reflects the rigidity of analytic foliations that we witnessed in Theorem 4.
THEOREM 14 (HAEFLIGER [H4]). The space FT has the homotopy-type of an Eilenberg-Mac Lane space K(TT, 1) where TT is an uncountable, torsionfree , perfect group (i.e., ^FT^^TT, Tr^r^OJM).

Note that FT is the classifying space for transversely oriented, codimension-one analytic Haefliger structures. Therefore, the concordance classes of such structures on X correspond bijectively to the homomorphisms TT1(X)-+TT. In particular, if TT^X) is a torsion group, then every T^-structure is analytically concordant to the trivial one. Some of the deepest work in the codimension-one case is due to Mather [MAI], [MA2] who proved by means of explicit geometric constructions that the homology of FY{, 0 ^ r 5 ^ co, is isomorphic to the homology of a double complex obtained by iterating the bar construction on the group Gr=T)\ftrK(R) of Cr-diffeomorphisms of R with compact support. From this he obtains a spectral sequence whose E1 term can be expressed in terms of H*(Gr) and which converges to H*(FTrx\ Z). For all the corollaries of this see the announcement [MAI]. We mention here two important consequences.
THEOREM 15 (MATHER). The space FTl is contractible. Hence, any two codimension-one topological Haefliger structures are topo logically concordant. THEOREM

16

(MATHER).
r

For l^r^co,
r r

there is a natural isomorphism

G l[G ,G ]^H2(FT{;Z). In particular,

i/ 2 (Frf;Z)(^7T 2 (Frn) = 0.
Together with Theorem 12, Theorem 16 shows that FT is 2-connected,

396

H. B. LAWSON, JR.

[May

Utilizing the new cohomology invariant of Godbillon-Vey, Thurston proved the following.
THEOREM

17

(THURSTON

[Th2]). For r^2, there is a surjective map

This contrasts dramatically with Theorems 14 and 15, and shows that even the spaces BT{ are very large. Note. This result has since been generalized by Thurston to give surjective homomorphisms H2q+1(FYrq; Z)-+R for all q^l, r^.2. Theorem 17 still leaves open the following. Problem 7. Determine the homotopy groups rr^FYl) for y = 4 , 5, 6, and 2 ^ r ^ o o . To begin, determine whether the above foliations of S2n+1 can be extended as Haefliger structures over the ball >2n+2. Very recently Thurston has succeeded in extending Mather's constructions to foliations of general codimension. He has developed a theory which relates the classifying spaces for certain classical pseudogroups (Example 6E above) to diffeomorphism groups of manifolds. To state his results I will need some definitions. Let G be a topological group, and let G denote the same group with the discrete topology. The identity homomorphism G->G is continuous and has a homotopy theoretic fiber denoted G~~. G~ in isfact a group. To see this we recall that G~ = i(g> 7) e G X G 7 : r (0) = g and y(l) = e) where / = [0, 1] and e e G is the identity. The multiplication on G ~ is the one inherited from GxGz. The classifying space for G~ arises in the long exact sequence of fibrations >G~~>G->G-+BG~~-+BG (=K(G, l))-> BG-+ We can now state one of the principal results. For a topological space X, let lqX denote the gr-fold loops on X.
THEOREM 18 (THURSTON [TH4]). For O^r^co there exists a continuous map 5(Diff^(i?a)~)-^iQQ,(/rrg) which induces an isomorphism on integral homology.

This leads to generalizations of Theorems 15 and 16 above. Using the result of Mather [MA3] that H*(BDix(Rq))=0, Thurston obtains
COROLLARY

6. For allq, the space FYQQ is contractible.

By factoring the above map through B(Dif[r(TQ)~~) and applying Herman's theorem [HER], he obtains
COROLLARY

7. For allq^ 1,

rr^FV^^O.

1974]

FOLIATIONS

397

This corollary shows that the dimension in Bott's vanishing result (Theorem H2 above) is best possible. Let K* be a 4-dimensional finite simplicial complex and f.K*-+B02 a map such that *px ^ 0 in H*(K*; Z) where p1 denotes the first universal Pontryagin class. Since TT3(FT^)=09 we know from obstruction theory that there exists a lifting off, : K^-^BY^ such that v f=f Hence, v^p^O. By considering products of K*, we get that v*(p$)9*0 in H*\BT?k; R). The results above leave some interesting questions. Problem 8. Is the fiber FT 2^-connected? Problem 9. What is the homotopy type of FF% for q> 1 ? It is interesting to note that the homology of FYrq is also related to that of the diffeomorphism group of a manifold. 19 (THURSTON). Let Mq be a compact differentiate manifold of dimension q, and let i:Rq->MQ be an embedding given, say, by a local coordinate chart. Then for any r, O ^ r ^ o o , the map
THEOREM

i*:Hk(B(DirK(R)-);

Z)-* Hk(B(Diffr(MT~);

Z)

is an isomorphism up to and including the first dimension k where the groups are nonzero. Combining this result with a theorem of Epstein [EP1] gives 8. For any compact manifold MQ, the connected component of the identity in Diff^^M^) is a simple group.
COROLLARY

One of the first things that occurs to a topologist when classifying spaces are mentioned is characteristic classes. Given any topological groupoid r , the ring of universal characteristic classes for Y is defined simply as H*(BY; Z). Note that for a given a eH*(BY; Z), every restructure J f o n a paracompact X has associated a well-defined characteristic class a(Jf ? )=/*(a) where : X->BV classifies Jf. From the above theorems we know there are rather a lot of universal characteristic classes for Haefliger structures. Theorem 10 states that up to dimension q the I^-characteristic classes are exactly the Pontryagin classes of the normal bundle. (Thus, for differentiate structures on manifolds we are getting nothing new.) However, in higher dimensions, new classes occur. Of course, the mere existence of such classes has little value without some method of computing them for a given foliation. The first formulas of this sort were given by Godbillon and Vey [GV] and by Bott [B3]. The Godbillon-Vey construction goes as follows. Let J5" be a codimension-^ foliation on a manifold M and suppose r(J^) is orientable. Then ZF is defined by a global decomposable ^-form Q. (Let {{&iy Xi))iei t>e a locally finite cover of distinguished coordinate charts

398

H. B. LAWSON, JR.

[May

on M with a smooth partition of unity {p J . Then, set Q = 2 Pi;dx?~q+1 A ' ' A dxT.)
iel

Since, ii is integrable dQ,=dAQ> where 0 is a one-form on M. The (2#+1)form y=0AddQ is closed, and its de Rham cohomology class \y\{^) e H2q+1(M; R) is independent of all the choices involved in defining it. It depends only on 3F. This construction can be generalized to arbitrary restructures for r ^ 2 , as a mixed de Rham-Cech cohomology class, and thus gives an element in H2Q+1(BTrq; R). If M is the unit tangent sphere bundle to a compact manifold of constant negative curvature, and if 3F is the codimension- Anosov foliation arising from the geodesic flow (flow X stable foliation; see [AR]), then [y](^)^0 in H2q+1(M;R). This was first noticed by Roussarie and Thurston. It follows that none of these classes is trivial. Note that the Godbillon-Vey class is, in fact, a cobordism invariant of codimension-^ foliations of compact (2+l)-manifolds (cf. 5). The Godbillon-Vey construction gives us one computable characteristic class for Yra. What we would like optimally is a generalization of the Chern-Weil construction for GLQ. (See [KN, vol. 2].) That is, we would like an abstract graded differential algebra with the property that for any codimension-^ foliation F on a manifold Af, there is a g.d.a. homomorphism into the de Rham algebra on M, defined in terms of J^, such that the induced map on cohomology factors through a universal map into H*(BVrq; R). This was already accomplished by Godbillon and Vey for q= 1. The algebra they discovered was the Gelfand-Fuchs Lie coalgebra of formal vector fields in one variable. This construction has been generalized to arbitrary codimension by Bott and Haefliger as follows. Consider the graded differential algebras (over R) :
W

Oq

= A(W1? UZ, ' , W2[<z/2]-l) PQ(Cl> ' * ' , CQ)

with duCi for odd / and dc~0

for all /; and

Wq = A(ul9 u2, - , uq) Pq(cl9 , ci) with du Ci and dc~0 for z = l , , q, where: deg u~2i 1 ; deg c~2i; A denotes exterior algebra; and Pq denotes the polynomial algebra in the c/s modulo elements of total degree >2q. The cohomology of Wq is the Gelfand-Fuchs cohomology of the Lie algebra of formal vector fields in q variables. We note that the ring structure at the cohomology level is trivial, that is, all cup products are zero. THEOREM 20 (BOTT-HAEFLIGER-GODBILLON-VEY [BH], [GH]). There are homomorphisms (p:H*(WOq) -> H*(BTrq; R), cp:H*(WQ) -> H*(FTrQ; R)

19741

FOLIATIONS

399

for r^.2 with the following property. If IF is a codimension~q9 Cr foliation of a manifold M9 there is a g.d.a. homomorphism (p<r:WOQ->A*(M) into the de Rham algebra on M, defined in terms of the differential geometry of IF and unique up to chain homotopy, such that on cohomology we have y = / * o cp where f:M-+BTrq classifies F. If the normal bundle of F is trivial, there is a homomorphism with analogous properties. For the explicit geometric construction of cp^ and <py in terms ot connections on v(F), see Conlon's notes from Bott's lectures [B2]. The diagram H*(WOq) - % H*(BTrq; R) H*(Wq)+H*(FT'Q;K) where/? is the natural map and i* is induced by inclusion, is commutative. If a foliation has trivial normal bundle, v(F), the theorem says that y factors through H*(Wq) (on the cohomology level). Note that if TT:P(F)-> M is the principal bundle associated to v(F), then TT*F on P{F) always has a trivial normal bundle. Hence, there is always a diagram: H*(WOq) - ^ - > H*(M\ R) H*(Wq) ^ > H*(P(F); R) which factors through the diagram above. The appearance of Gelfand-Fuchs cohomology has a direct analogy in the cohomology of topological groups. Bott and Haefliger have shown that the notion of continuous (or differential) cohomology for Lie groups has a natural generalization to the Lie groupoids coming from pseudogroups in Rq (cf. Example 6E), and that in analogy with the work of Van Est, this continuous cohomology is just the Gelfand-Fuchs cohomology. (See [H6] for an excellent exposition.) Thus, H*(Wq) is naturally the continuous cohomology of Trq, and H*(WOq) is the continuous cohomology of (TJ, Oq) (i.e., the cohomology of continuous, Oa-invariant cochains). A fundamental consequence of this result is that any characteristic class of foliations which is constructed locally in the algebra of differential

400

H. B. LAWSON, JR.

[May

forms will be, universally, in the image of cp (or cp). However, H*(BTrq; R) in general contains much more than the image of cp. (See Theorem 17.) Some of the remaining, "discontinuous" cohomology is related to classes for bundles with discrete structure groups. An alternative approach to the cohomology of BVrq has been taken by J. Simons, who has avoided the business of passing to the normal bundle by using circle coefficients. Continuing his work with S. S. Chern [CS], he has developed a theory which associates to a principal bundle with connection a family of characteristic homomorphisms from the integral cycles on a manifold to S1. For the Bott connection on a normal bundle to a foliation, the Simons characters of degree >2q define cohomology classes independent of the choice of Bott connection. His result can be described as follows. From the Bockstein sequence for the coefficient homomorphism O-^Z^R-^S1-^, we get the commutative diagram: H2k~\BGLq\ S1) - ^ > H2k(BGLQ; Z) - 1 > H2\BGLq\ R)

(6.6)

[s
-U

jo
H2k(BTrq;R).

H2k~HBYq\ S1) -^-> H2k(BYrq\Z)

For k>q, the right vertical map is zero by Theorem H2, and the middle map s is an injection by Theorem 13. By exactness, s(H2k(BGLq; Z))<= iH^-^BV^ S1)). Simons defines an extension Kf of H2k(BGLq; Z) and a map S such that the diagram Kf > H2k(BGLQ; Z) S1) - 1 * H2k(BTrq;Z)

H2k-\Brrq;

commutes. The extension is defined as follows. Let I2k(GLq) denote the homogeneous polynomials of degree k on the Lie algebra gla of GLq, which are invariant by the adjoint representation. (Recall that 7*(GL a )^ * l A , , cq] where d e t ( A / - ^ l ) = ^ - c 1 ( ^ ) ^ - 1 + - - + (-\ycq(A) for A G glq.) There is a natural map j:I2k(GLq) -> H2k(BGLq; R) given by the Chern-Weil homomorphism. We define the Simons ring Kqk to be the kernel of the homomorphism : H2k(BGLQ; Z) x I2k(GLq) - ^ - > H2k(BGLq; R) where / is given in (6.6). Note that if p e I2k(GLq) involves some cm for m odd, then ( 0 , ^ ) 6 ^ * .

1974] THEOREM

FOLIATIONS

401

21

(SIMONS

[SI]). For k>q andr^.2 there is a homomorphism SiKf-^H^iBT^S1)

with the following property. If IF is a codimension-q, Crfoliation of a manifold M, there is a homomorphism S^'.Kf-^H^-^MiS1) defined in terms of the geometry of F', such that S#r=f* o 5 where : M-> BTq classifies &. The Godbillon-Vey class arises in both theorems above. It is given as the cohomology class of ux<S>cl e WOq in Theorem 20. Reduced mod Z, it is the class 5(0, cj+1) in Theorem 21. It should be pointed out that, potentially, the homomorphisms op and 5 carry some essentially distinct classes. That is, if both cp and 5 are injective, then neither of the subgroups image(5) or image(p o ^ ) , where p is reduction mod Z, is contained in the other. The extent to which these homomorphisms are nontrivial is largely an open question. Some things are known. For example, one can easily check that H*(W01) has only two nonzero elements, the usual one in dimension zero and the Godbillon-Vey class in dimension three. Hence, cp\H*(WO^H*{BTrx\ R) for 2<><;oo is injective. Bott and Haefliger have certain partial results in this direction for q^2. (For example, uxc\ and uxc% are linearly independent in H5(BFl; R).) Since the map s in (6.6) is injective and the projection Klk-+H2k(BrrQ; Z) is surjective, the map 5 is nonzero for even k. Furthermore, by computation of the Godbillon-Vey class, 5(0, cl+l) is not zero for all q. However, the following question remains open. Problem 10. Are the maps <p, y and 5 defined in Theorems 20 and 21 injective? ADDED IN PROOF. It has been brought to my attention that some of these results on H*(BTQ) are also contained in the work of Kamber and Tondeur [KT]. It is interesting to note that most of the classes of dimension >2q+l discussed in these theorems depend only on the homotopy class of the foliation in the sense of 5B. (See [HI2].) For classes in dimension 2q+l this is not true. Before leaving this subject we remark that, in general, whenever T comes from a pseudogroup acting on Rq, as in Example 6E, the structure of FT is of great interest. A number of people are currently studying these spaces. Pasternak [P2] has results on FRTrq, the classifying space for foliations with bundle-like metrics. Landweber [LA] has shown that FVq(C) is (q l)-connected, thereby obtaining results on the question of when an

402

H. B. LAWSON, JR.

[May

almost complex structure is homotopic to an integrable one. Bott ([B3] and [B4]) has results on H*(BTq(C); R) and TT2Q+1(BTQ(C)). Thurston [TH4] has theorems similar to Theorems 18 and 19 above for volumepreserving diffeomorphisms and for diffeomorphisms preserving a symplectic form. The interested reader should also see [H4] for a discussion of further results on these spaces. 7. Results on open manifolds. There is a tremendous difference between the theory of foliations on open manifolds and that on closed manifolds. In the open case it is difficult to obtain qualitative results. One can easily appreciate this by observing that any two codimension-<7 foliations of Rn are integrably homotopic. To prove this it suffices to show that any ^ on Rn is integrably homotopic to the foliation of a distinguished neighborhood U^Rn where U^Rn. Let RnxR have the product foliation, and choose an embedding O : Rn x R^Rn x R such that (x, t)=((pt(x), t) where <p0=identity and (p1:Rn->U is a homeomorphism. Then RnxR, with the foliation O * ^ , is the desired integrable homotopy. This construction illustrates a powerful technique which is not available on compact manifolds, namely, that of "pushing the difficulties out to infinity". This technique makes it possible to obtain beautiful quantitative results. The general theory which applies to quantitative questions on open manifolds had its inception with the immersion theory of Smale and Hirsch and has undergone a long development, culminating recently in the work of Gromov. (See [PHI], [PH4], [PH5], [G], [H5].) The principle is: "Questions in analysis involving open conditions on open manifolds are always solvable". The general procedure of proof is to decompose the manifold as a handlebody; prove the result first for the disk, and then proceed to attach handles and argue inductively. We shall discuss how this theorem together with the results in 6 can be used to classify foliations on open manifolds. To begin it is necessary to present a foliation by an "open condition" as follows. Let N be a manifold with a smooth, codimension- foliation 3F. For any manifold M we define the space Trans(M, N^) of smooth maps f\M->N which are transverse to 3F. The space is given the usual topology for smooth maps. Of course, for G Trans (M, N&), * J r is a codimension-^ foliation of M9 and Trans(.M, N#) is open in the space of all maps from M to N. We now consider the associated space Epi(r(M), v{^)) of continuous bundle maps from the tangent bundle T(M) of M to the normal bundle K ^ ^ T X A O M J ^ ) of the foliation, which is an epimorphism on each fiber. This space is given the compact open topology. There is a natural continuous map (7.1) D:Trans(M, JV -> Epi(TTM), v{&))

1974]

FOLIATIONS

403

given by D(f)=p
P\T{N)-^T{N)IT{^)^V{^)

o df where df: T(M)-+T(N) is the differential o f / and

is projection.

Note that the space Epi is considerably larger than Trans. If M is parallelizable, for example, there always exists a map in Epi(r(M), vF)), where IF is the foliation of RQ by points (q^dim M). However, if M is compact, a corresponding transverse map, i.e., a submersion M-+Rq, does not exist. The startling fact is that for open manifolds it does exist.
THEOREM 22 (GROMOV-PHILLIPS [G], [PH4]). For any open manifold M the map (l.l)isa weak homotopy equivalence, that is, it induces isomorphisms on all homotopy and homology groups.

In particular, D establishes a one-to-one correspondence between the connected components (i.e., 7r0) of the two spaces. Thus, we conclude the following.
COROLLARY 9. If M is an open manifold, then every codimension-q plane field T=ker(/?) for ft e Epi(r(M), v{^)) is homotopic to a foliation.

It should be noted that Theorem 22 contains the Smale-Hirsch immersion theorem and the Phillips submersion theorem. For the immersion theorem, let M be the normal bundle. In light of Corollary 9, our problem now is: given r on M, find N, J** and ft so that r = k e r ft. Following an idea of Milnor [M4], we can do this canonically by defining N to be the total space of the vector bundle v= T(M)jr. We say that a ^-dimensional vector bundle TT : V-+M is offoliated type if there exists a smooth codimension-^ foliation of V whose leaves are everywhere transverse to the fibers. (Thus for J? e 3F, TT\3?\3?-+M is a local diffeomorphism.)
COROLLARY 10. Let M be an open manifold. Then a continuous codimension-q plane field T on M is homotopic to a smooth foliation if and only if the vector bundle v=T{M)jr is of foliated type. PROOF. If T = T ( ^ ) , choose a riemannian metric on M and identify v with T1. The exponential map e:v->M is transverse on a neighborhood U of the zero-section. Shrink v into U and pull back the foliation e*IF to all of v. If v is of foliated type with a foliation &', consider the epimorphic bundle map T(M)->T(V)IT(<F) obtained as the composition.

T(M) -^v-^->

T(v) - ^ >

T{V)JT{^)

where them's are natural projections and z is the canonical embedding of v along the zero-section. Applying Corollary 9 concludes the proof. We are now in a position to describe how one passes from general

404

H. B. LAWSON, JR.

[May

Haefliger structures on M to honest foliations. The key observation is the following.

LEMMA 4. A vector bundle v over a manifold M is the normal bundle to a Haefliger structure on M if and only if it is of foliated type.

If v is of foliated type, consider the Haefliger structure Jtf9 on M induced from the foliation on v by the zero section M->v. Clearly,
PROOF.

Conversely, suppose v is the normal bundle of a Haefliger structure J^ on M defined by a cocycle yi over a cover %={Ot}iei. We define a new manifold J^(M), a normal thickening of Af, as follows. Assume ll is locally finite. Then for i e , define Ui to be a small neighborhood of the graph Y~{x,f{x)}^OixRq. We identify a neighborhood of (x, fix)) e I\<= ui with a neighborhood of (x9f(x)) e V^ JJi by the diffeomorphism (x, y)-+(x, yH - y). The foliations of the C//s obtained by projection on Rq fit together to give a foliation 8F of J^(M). Choosing one of the obvious diffeomorphisms v-*Jf{M) to lift 3F shows that v is of foliated type. Combining Corollary 10 with Lemma 4 shows that: a codimension-^ plane field T on M is homotopic to a foliation iff v=T(M)\r is the normal bundle of a Haefliger structure on M, that is, iff the map cpv:M-+BGLQ, classifying v, lifts to a map cpp\M->BYrq. It is, furthermore, possible to show that two foliations with the same normal bundle determine homotopic liftings if and only if they are integrably homotopic. The general classification theorem can now be stated. For p+q=m let Bp\BGLpXBGLq-+BGLm be the map induced by the standard homomorphism p : GLV x GLq->GLm.
THEOREM 23 (HAEFLIGER [H3]). Let M be an open manifold of dimension m. Then the integrable homotopy classes of codimension-q, Cr foliations are in one-to-one correspondence with homotopy classes of liftings <p^ X q)T of the map cpT in the diagram : 2?r, x BGLm_Q
<PF*<PT _ |
v x M

X BGLm_Q BGLm where cpT classifies the tangent bundle of M. Note that lifting cpT over Bp to cpv X cpT corresponds to the topological problem of finding of codimension-*/ plane field on M. The second lifting, of cpv over v to (p#r, is the problem of integrating the given (homotopy class of the) plane field. Of course, obstructions to this second lifting will, in

1974]

FOLIATIONS

405

general, exist. However, in time they should become reasonably computable. The beauty of this theorem is that the general machinery developed to handle the homotopy question HQ in 6 can now be brought to bear on the analytical question Q for open manifolds. For example, in view of Theorem 12 and Corollary 7 we have the following.
COROLLARY 11. Let M be an open manifold having the homotopy-type of a k-dimensional CW complex. Then if k^q+2, every codimension-q plane field on M is homotopic to a foliation which is uniquely determined up to integrable homotopy.

8. Results on closed manifolds. In studying foliations on open manifolds, one finds wonderful classification theorems of a general type, but practically no qualitative results. The situation for compact manifolds is exactly the reverse. There are few general existence theorems. However, there is a vast literature concerned with deep special results. I shall mention here some of the interesting theorems and open questions. The reader should also consult [H2], [Rl], [R3], [R03] and [TS1]. The existence of foliations of codimension-one has been discussed in 4. For foliations of higher codimension, very little is known. One exception is the sphere S1. By Theorem 6 there is a codimension-one foliation. Alberto Verjovsky [V2] has constructed a foliation of codimension 3 and Jose Arraut [AR] has constructed one of codimension 5. The classical Hopf fiberings S7-+S* and S7->P3(C) have codimensions 4 and 6 respectively. It remains only to find a foliation of codimension-2. The existence of this foliation would answer for S7 the following general question of Reeb. Problem 11. Does every sphere that admits a codimension-^ plane field admit a codimension-*/ foliation? ADDED IN PROOF. Very recently Thurston has established a startling and beautiful quantitative theory for foliations on compact manifolds in codimension ^ 2 (cf. The theory of foliations of codimension greater than one, Comment. Math. Helv. (to appear)). The fundamental result is the following.
THEOREM. Let r be a plane field of codimension q^.2 on a compact manifold M. Then r is homotopic to a foliation of class Cr Orgrrg oo, on M if and only ifr1 is the normal bundle to a Haefliger structure on M (i.e., if and only if the classifying map n:M-+BGLqfor r1 can be lifted to BTrQ, cf. 6). Furthermore if r is already integrable in a neighborhood of some compact Set K<= M, then the homotopy can be chosen to be constant on K.

There are several immediate consequences of this theorem. From Corollary 7 we have : (1) Every 2-plane field on a manifold of dimension ^ 4 is homotopic to a

406

H. B. LAWSON, JR.

[May

foliation. (In fact, Thurston has extended his methods to make this true also for 3-manifolds, thereby eliminating the condition of transverse orientability in Wood's Theorem.) From Corollary 6 we have : (2) Every plane field of codimension g:2 on a manifold is homotopic to a C foliation (which is actually Lipschitz with C leaves). (3) The span of any q-frame field (q^2) on a manifold is homotopic to the normal bundle of a foliation. In particular, this result answers Reeb's question for S7 and gives the following partial answer for general spheres. (4) Every plane field on Sn of codimension ^n/2 is homotopic to a foliation. While on this subject we shall mention a closely related question. Following Kodaira and Spencer [KS] we can define a multifoliation of a manifold M to be a (finite) collection of foliations J ^ , , <Fr on M such that all intersections are "transverse", that is, at any point of M, c o d i m O r J ^ r v "C\T^'i)=codim(T<Fii)'\ h c o d i m ( T ^ ) for all subsets {il9 , is}<^{\9 , r}. (Thus, these intersections themselves determine foliations.) The multifoliation is called total if r=dim(M), and codim(J r ,)=r1 for each i. In this case, the tangent bundle of M is reduced to a sum of line bundles. D. Tischler [Til] has proven that every 3-manifold which is the total space of a principal circle bundle over an oriented surface admits a total multifoliation. In particular, this includes Sz, so in analogy with Reeb's question above, we pose the following problem. Problem 12. Does S7 admit a total multifoliation? Problems 11 and 12 can, of course, be restated for general manifolds. For a discussion of further problems of this type, see [TS1]. One of the most fundamental classical results in the study of dynamical systems in the Poincar-Bendixson Theorem which asserts that for a C 1 flow on S2 every minimal closed invariant set is either a fixed point or a closed orbit. A natural generalization of this result to foliations would assert that for a C 1 foliation of Sn, every minimal, closed, invariant (a union of leaves) set is a closed leaf. This is far from true. (See, for example, [S2], [R3], [RR1], and [RA].) However, the question of the existence of a closed leaf remains open, and one of the most celebrated results in the theory of foliations is the following.
THEOREM 24 (NOVIKOV [N]). Let M be a compact ^-dimensional manifold with a smooth codimension-one foliation J r , and suppose that any of the following conditions are satisfied: (1) 7TX(M) is finite,

(2) 7T 2 (^)^0,

1974]

FOLIATIONS

407

(3) there exists a closed curve in M transverse to IF which is nullhomotopic, or (4) there exists a leaf & e IF such that the map 7r1(JSf)->77'1(M) has nontrivial kernel. Then IF has a closed leaf In fact, except for case (2), the foliation contains a Reeb component. In case (2) either there is a Reeb component or M is a finite quotient of a fiber bundle over S1 (i.e., all leaves are compact and =S2 or P2(R)). 12. Every smooth codimension-one foliation of S3 has a Reeb component.
COROLLARY

This leads to the following natural question. Problem 13. Does every smooth, codimension-one foliation of *S2w+1, for w > l , have a closed leaf? One partial result of this type has been obtained by Joe Plante [PLI]. He shows that if the leaves satisfy certain intrinsic growth conditions, then the strong Poincar-Benedixson Theorem holds, i.e., every minimal set is a closed leaf. His methods give nice results for actions of nilpotent Lie groups on simply-connected manifolds. Sacksteder [S3] has given a condition in terms of linear holonomy groups, which guarantees the existence of a compact leaf. Note that if a codimension-one foliation of Sn has a closed leaf, then it certainly does not have a dense leaf. However, it is not even known that foliations of Sn with dense leaves cannot exist. CONJECTURE. Let M be a compact manifold which admits a smooth codimension-one foliation having a dense leaf. Then, HX(M\ R)y*0. It should be noted that Hector [HEC] has recently constructed a codimension-one foliation of euclidean space in which every leaf is dense. ADDED IN PROOF. Paul Schweitzer has recently shown that if a manifold of dimension ^ 5 admits a smooth, codimension-one foliation, then it also admits a codimension-one, C foliation having no compact leaves. One might naturally ask whether Novikov's theorem extends to a larger class of 3-manifolds. In particular, could it be true for a manifold which is a K(TT, 1)? This question was recently answered by W. Thurston who considered 3-manifolds which are circle bundles over surfaces. Here it is sometimes possible to construct foliations without compact leaves by the methods of 1B. However, this is the only allowable procedure.
THEOREM 25 (W. THURSTON [TH1]). Let M be the total space of an oriented circle bundle I over an oriented surface H^S1xS1, and let IF be a C 2 , transversely oriented foliation of M. Then either !F is isotopic to a foliation transverse to the fibers of'| or IF has a compact leaf. In particular

408

H. B. LAWSON, JR.

[May

* l%(f)l>lz(^)l> where % denotes Euler characteristic, then ^ must have a compact leaf. The second part of this theorem follows from a pretty result of John Wood [W3] that a circle bundle f over a compact surface 2 admits a foliation transverse to the fibers if and only if |#()|^min{0, #(2)}. (Here the bundle must be " 2 orientable", i.e., the total space of f must be orientable.) The question of closed leaves for foliations of higher codimension goes back to Seifert who posed the following well-known 3 SEIFERT CONJECTURE. Every nonsingular vector field on S has a closed integral curve. Seifert proved the conjecture for vector fields C-close to the Hopf vector field (or, more generally, C close to the vector field of any Seifert fibering of SB). Novikov [N] proved the conjecture for fields transverse to a codimension-one foliation of Sz. Pugh [PU] proved the conjecture for generic vector fields. However, in a remarkable paper [SC], Paul Schweitzer has recently shown the following. 26 (P. SCHWEITZER). Every C1 foliation of codimension ^ 2 on a manifold is homotopic to a C1 foliation with no compact leaves.
THEOREM

13. Every homotopy class of nonvanishing C1 vector fields on S contains a counterexample to the Seifert conjecture.
COROLLARY
3

Schweitzer's foliations actually have tangent plane fields of class C 1 but definitely not of class C 2 . Hence, the conjecture remains open for vector fields of class Cr where r^.2 or r=co. In the case that compact leaves do exist, it is natural to ask about their topological type. The most classical and important result of this type is the Reeb Stability Theorem ([Rl]). 27 (G. REEB). Let ^ be a Cr foliation on a manifold M where r ^ O , and suppose that J^ has a compact leaf 3? with finite fundamental group. Then every neighborhood of'jSf contains an invariant neighborhood U with the property that each leaf 3" c: U is a finite covering space of ?.
THEOREM

Thus, if oaf is simply connected, [/has the form 3?xDq with the foliation {JS? x {p}}vez>: I n general, a finite covering of U is of this form, where JSP is replaced by the universal covering of Sf. Note that M need not be compact, and the codimension does not matter. If we tighten these requirements we get stronger results.
COROLLARY 14. Let M be a compact manifold with a smooth, codimension-one, transversely orientable foliation F. If IF has a compact leaf J?

1974]

FOLIATIONS

409

with finite fundamental group, then M is a fiber bundle over S1 with fiber (and F is the associated foliation). Theorem 25 can be strengthened to include all nearby foliations.
THEOREM 27' (G. REEB). Let M, F and be as in Theorem 25. Then given a neighborhood i^ of , there is a neighborhood & of F in the space of Cr foliations of M and a neighborhood U of jSf in M such that every i ? ' e F' e & with P' C\ll^ 0 is a finite covering space of' JSf.

There is a version of Theorem 25, due to B. Reinhart [RI1], for foliations with metric properties, that is, for foliations whose holonomy (see below) is distance-preserving in some riemannian metric. Such a metric is called bundle-like. For these foliations there is also an associated global stability theorem, similar to Corollary 14, which holds in general codimensions [RI3]. Related to these results is a deep theorem of Sacksteder. To state it, we need to introduce the important notion, due to Ehresmann, of holonomy. Let F be a codimension-g foliation on M, and at a fixed p e M choose an embedding cp\Dq->M transverse to F with cp(0)=p. To each oriented loop y based at p and lying in the leaf j? containing p, we associate a (germ of) a diffeomorphism hy of a neighborhood of 0 in Dq to another neighborhood of 0 as follows. Choose a transverse map O : Dq X S^M with the property that O(0, 0) = y(0) and O|Z>*x{0} = <p. For xeDq, q 1 1 follow the curve of the foliation induced on D X S as 0 traverses S . For x sufficiently close to 0, it is possible to pass completely around S1 and return to a new point h7(x) e Dq. The map hy depends only on the homotopy class of y in J?, and is called the holonomy map along y. We get a homomorphism, ^(JSf, p)->Trq, the germs of local diffeomorphisms of Rq which fix 0. The image is the holonomy group dit p. If 7r1(5f,/?) = {l}, then the holonomy group at/7 is trivial.
THEOREM 28 (SACKSTEDER [S3]). Let M be a compact manifold with a codimension-one, exfoliation F, and suppose that all the holonomy groups of F are finite. Then there is a riemannian metric on M invariant under holonomy, and M is covered by Rx? where is either a leaf or a two-fold cover of a leaf in F. In particular, if F is orientable and all the holonomy groups are trivial, then there is a fixed-point free flow on M which preserves F, i.e., maps leaves to leaves for all time.

Sacksteder also shows in [S3] that if Mm has an orientable, codimension1 foliation F, given by the orbits of a locally free Rm~x action, and if no leaves of F are compact, then every leaf is dense and H^MiR^O.

410

H. B. LAWSON, JR.

[May

Note. There was, incidentally, a minor error in Lemma 14.1(h) of [S3]. It was pointed out by E. Vidal and A. Vasquez that under the assumptions of Theorem 28, one can conclude that H^M; R)?*0 only if IF is orientable. Note, however that if !F is not orientable, its lift to an appropriate 2-sheeted covering surface is. The example of Vasquez is M= S1 X S2jZ2 where Z2 is generated by oc(z, x)=(z, x) for (z, x)eS1xS2<^ CxRz and where J^ lifts to the foliation {{z} x S2}zSi of S1 X S2. It is important to point out that neither the metric nor the flow in Theorem 28 are, in general, smooth. One must first change the differentiable structure of M. Once the change is made, the flow can be viewed as coming from a smooth vector field transverse to IF. The existence of such a vector field V means that !F can be defined by a closed one-form. Let be the 1-form which vanishes on IF and has the property co(K)=l. Then one can check that d = A(Lv) where L denotes Lie derivative. Since IF is invariant by V, Lvco=0. If IF comes from a submersion M-+S1 it is defined by a closed 1-form. The converse is not true. (Consider parallel foliations on a torus.) However, it is nearly true.
THEOREM 29 (TISCHLER [TI2]). If M admits a codimension-one foliation defined by a closedform, then there is afibering M-+S1 which is close to IF.

Much of the discussion in this chapter is concerned with when a foliation must have a compact leaf. One might equally well ask what can be said of a foliation all of whose leaves are compact. In codimension-one, the holonomy groups must all be finite, and by Sacksteder (or, in this case, the more elementary arguments of Reeb) some 2-sheeted cover of the foliation is a fibration over S1. In higher codimensions the problem is much harder. For example, Reeb [Rl, pp. 113-115] has given an example of an open manifold with a foliation by compact leaves such that the union of leaves meeting a certain compact set K is not compact. It is reasonable therefore to assume the manifold is also compact before asking for strong consequences. The best result of this type is the following deep theorem. 30 (EPSTEIN [EP2]). Every Cr foliation ( l ^ r ^ o o ) of a compact orientable three-manifold, possibly with boundary, by circles is Crconjugate to a Seifert fibration. That is, every leaf has a saturated neighborhood U^S1xD2 where the foliation is Cr-conjugate to the orbits of the action <pt(eid,z) = (eiit+e\eii/a)t-z)
THEOREM

on SxxD2for

relatively prime integers p andq.

Outside this result little is known about the general problem.

19743

FOLIATIONS

411

Problem 14. Characterize the foliations of compact manifolds in which every leaf is compact. A fundamental area of study in the field of foliations surrounds the notion of stability. A smooth foliation F is said to be Cr-stable if every foliation F' such that r(F') is sufficiently close to r(F) in the C r -topology, is C-conjugate to F. Unstable foliations exist. (Again consider parallel foliations on a torus.) The interesting problem is the following. Problem 15. Find conditions which guarantee that a foliation is stable. Reeb's Theorem 25 together with Corollary 14 gives a result of this type. Furthermore, Rosenberg and Roussarie have recently shown the following.
THEOREM 31 (ROSENBERG AND ROUSSARIE [RR4], [ROU]). No codimension-one foliation of S3 is (C00) stable. The only stable foliations of S1xS2 are those conjugate to the foliation {{0}xS2:6 G S1}, modified by introducing a finite number of hyperbolic Reeb components along closed transversals.

There are also results on stability of intrinsic components. One could analogously ask for specific properties of a foliation to be stable, for example, the property of having a compact leaf. Theorem 27' falls into this category and has been recently generalized by Hirsch. 32 (M. HIRSCH [HR]). Let F be a Cr foliation ( r ^ l ) on a manifold M and suppose there is a compact leaf J? e F and an element
THEOREM a e TT-JXJS?) such that

(1) a is in the center ofit^SP). (2) The differential of the holonomy map of en does not have 1 as an eigenvalue. Then for all e sufficiently small, there is a neighborhood OofF in the space of Crfoliations ofM, such that for each F' e &, there is a compact leaf ' e F' and a map f.3?-+3?' such that dist(x, (x))<e (in a fixed riemannian metric on M). Moreover, " is unique andf is a homotopy equivalence. Condition (1) in this theorem can be replaced by: " a belongs to a nilpotent subgroup of finite index". A version of the result is also true for the case r = 0 . (See [HR].) There is an area in the study of foliations concerned with questions of the following general type. What are the topological conditions necessary for a manifold to admit a foliation of prescribed type? For example, Rosenberg [R02] has shown that a (not necessarily compact) Z-manifold, C2foliated by planes, has the property that any embedded 2-sphere bounds an embedded ball. Moreover, from basic results of [ROl], the following theorem (due to Rosenberg and Sondow [R02] for m = 3 , Joubert and

412

H. B. LAWSON, JR.

[May

Moussu [JM] for m = 4 , and Rosenberg [R02] together with results of C. T. C. Wall for m |>5) can be proved. 33. If a closed m-manifold M admits a codimension-one, C2 foliation where every leaf is homeomorphic to JRW-1, then M is homeomorphic to an m-torus, S1 X X S1.
THEOREM

Deep results in a similar vein have been obtained by Yerjovsky [VI] for codimension-one, Anosov foliations. (These foliations are not C2.) Theorem 33 for the case ra=3 has been extended by Rosenberg, Roussarie and Chatelet to foliations of 3-manifolds with boundary [RR2], [RR3], [CR]. This work includes a classification of Reeb foliations up to C conjugation, and, in particular, the following generalization of the Denjoy theorem: Every C2foliation of the 3-torus by 2-planes is C conjugate to a linearfoliation. (M. Herman has shown this to be false for C foliations. The C 1 case appears to be unknown.) This work, together with the results of Moussu and Roussaire [MR1], gives a classification of C 2 foliations of TB with no Reeb components as follows : For each such foliation there is a decomposition of T3 into submanifolds with boundary Al9 , Ap, Bl9 , Bq where A^B0^T2xI and where, up to C conjugacy: the foliation on Ai is a suspension of the Reeb foliation of S1X , and the foliation on B is transverse to the fibers of the projection T2xl->T2 (and, thus, given by a homomorphism ^(r 2 )Diff + (/), cf. 1). There is an entire literature on noncompact group actions (in particular, Inactions) on manifolds, which we shall not discuss. However, for a presentation of the results of Lima, Novikov, and Rosenberg-RoussarieWeil on the rank of 3-manifolds (cf. 1C), see [TS1]. We have also omitted mentioning questions of deformations of foliations (see [KS], [H2] and [H4]), and questions of the applications of elliptic operator theory (see Reinhart [RI2]).
REFERENCES

[AC] N. A'Campo, Feuilletages de codimension 1 sur les varits de dimension 5, Comment. Math. Helv. 47 (1973), 54-65. [AR] R. Abraham. J. Robbin and A. Kelley, Transversal mappings and flows, Benjamin, New York, 1967. MR 39 #2181. [A] J. W. Alexander, A lemma on systems of knotted curves, Proc. Nat. Acad. Sci. U.S.A. 9 (1923), 93-95. [ANO] D. V. Anosov, Geodesic flows on closed Riemannian manifolds of negative curvature, Trudy Mat. Inst. Steklov. 90 (1967)=Proc. Steklov Inst. Math. 90 (1967). MR 36 #7157; 39 #3527. [AR] J. L. Arraut, A 2-dimensiona I foliation of S\ Topology 12 (1973), 243-246. [BA] D. Barden, Simply connected five manifolds, Ann. of Math. (2) 82 (1965), 365385. MR 32 #1714. [BAU] P. Baum, Structure offoliation singularities, Brown University (preprint).

19741

FOLIATIONS

413

[BB] P. Baum and R. Bott, Singularities of holomorphic foliations, J. Differential Geometry 7 (1972), 279-342. [BR] 1. N. Bernstein and B. I. Rosenfeld, Sur les classes charactristiques des feuilletages, Functional Analysis 6 (1972), 68-69. [Bl] R. Bott, On a topological obstruction to integrability, Proc. Sympos. Pure Math., vol. 16, Amer. Math. Soc, Providence, R.I., 1970, pp. 127-131. MR 42 #1155. [B2] , Lectures on characteristic classes and foliations, Notes by Lawrence Conlon, Mexico, 1971, Springer Lecture Notes, Springer N.Y., 1973. [B3] , On topological obstructions to integrability, Proc. Internat. Congress Math. (Nice, 1970), vol. 1, Gauthier-Villars, Paris, 1971, pp. 27-36. [B4] , The Lefschetz formula and exotic characteristic classes, Proc. of the Differential Geometry Conference, Rome, 1971. [BH] R. Bott and A. Haefliger, On characteristic classes of V-foliations, Bull. Amer. Math. Soc. 78 (1972), 1039-1044. [BHI] R. Bott and J. Heitsch, A remark on the integral cohomology ofBYQ, Topology 11 (1972), 141-146. MR 45 #2738. [BL] J. P. Buffet and J. C. Lor, Une construction d'un universel pour une classe assez large de Y-structures, C. R. Acad. Sci. Paris Sr. A-B 270 (1970), A640-A642. MR 42 #6823. [CE] B. Cenkl, Foliations and connections with zero torsion, Northeastern University preprint (to appear). [CR] G. Chatelet and H. Rosenberg, Un theorem de conjugaison des feuilletages, Ann. Inst. Fourier (Grenoble) (1971), 95-107. [C] S. S. Chern, The geometry of G-structures, Bull. Amer. Math. Soc. 72 (1966), 167-219. MR 33 #661. [CS] S. S. Chern and J. Simons, Characteristic forms and transgression. I, Berkeley preprint, (to appear). [COI] L. Conlon, Transversely parallelizable foliations of codimension two, Washington University, St. Louis, Missouri (preprint). [C02] , Foliations and locally free transformation groups of codimension two, Washington University, St. Louis, Missouri (preprint). [DE] A. Denjoy, Sur les courbes dfinies par les equations diffrentielles la surface du tore, J. Math. Pures Appl. 11 (1932), 333-375. [D] A. H. Durfee, Foliations of odd-dimensional spheres, Ann. of Math. (2) 96 (1972), 407-411. [DL] A. H. Durfee and H. B. Lawson, Jr., Fibered knots and foliations of highly connected manifolds, Invent. Math. 17 (1972), 203-215. [EP] D. Epstein, The simplicity of certain groups of homeomorphisms, Compositio Math. 22 (1970), 165-173. MR 42 #2491. [ER] C. Ehresmann and G. Reeb, Sur les champs d'lments de contact de dimension p compltement intgrable dans une varit continuement differentiable, C. R. Acad. Sci. Paris 218 (1944), 955-957. MR 7, 327. [El] C. Ehresmann, Sur la thorie des varits feuilletes, Univ. Roma 1st. Naz Alta Mat. Rend. Mat. e Appl. (5) 10 (1951), 64-82. MR 13, 870. [E2] , Les connexions infinitsimales dans un espace fibre differentiable, Colloque de topologie (espaces fibres), Bruxelles, 1950; Georges Thone, Lige; Masson, Paris, 1951, pp. 29-55. MR 13, 159. [E3] , Introduction la thorie des structures infinitsimales et des pseudogroupes de Lie. Gomtrie diffrentielle, Colloq. Internat, du Centre National de la Recherche Scientifique, Strasbourg, 1953, Centre National de la Recherche Scientifique, Paris, 1953, pp. 97-110. MR 16, 75.

414

H. B. LAWSON, JR.

[May

[E4] C. Ehresmann, Structures locales, Ann. Mat. Pura Appl. (4) 36 (1954), 133-142. MR 16, 504. [ES] C. Ehresmann and Shih Weisher, Sur les espaces feuillets: Thorme de stabilit, C. R. Acad. Sci. Paris 243 (1956), 344-346. MR 18, 751. [F] F. G. Frobenius, Ueber das Pfaffsche Problem, J. Reine Angew Math. 82 (1877), 267-282 (Gesamelte Abh. I, 286-301). [FR] M. Freedman, Thesis, Princeton University, Princeton, N.J., 1973. [GV] C. Godbillon and J. Vey, Un invariant des feuilletages de codimension 1, C. R. Acad. Sci. Paris Sr. A-B 273 (1971), A92-A95. MR 44 #1046. [G] M. L. Gromov, Stable mappings offoliations into manifolds, Izv. Akad. Nauk SSSR Ser. Mat. 33 (1969), 707-734=Math. USSR Izv. 3 (1969), 671-694. MR 41 #7708. [HI] A. Haefliger, Structures feuilletes et cohomologie valeur dans un faisceau de groupods, Comment. Math. Helv. 32 (1958), 249-329. [H2] , Varits feuillets, Ann. Scuola Norm. Sup. Pisa (3) 16 (1962), 367-397. MR 32 #6487. [H3] , Feuilletages sur les varits ouvertes, Topology 9 (1970), 183-194. MR 41 #7709. [H4] , Homotopy and integrability, Manifolds-Amsterdam 1970, Lecture Notes in Math., vol. 197, Springer, Berlin, 1971, pp. 133-163. MR 44 #2251. [H5] , Lectures on Gromov's theorem, Liverpool Singularities Symposium, Lecture Notes in Math., no. 209, Springer-Verlag, Berlin, 1971, pp. 128-141. [H6] , Sur les classes caractristiques des feuilletages, Sminaire Bourbaki, 1971-72, no. 412. [HI1] J. Heitsch, The cohomologies of classifying spaces for foliations, Thesis, University of Chicago, Chicago, 111., 1971. [HI2] , Deformations of secondary characteristic classes, Berkeley preprint, (to appear). [HEC] G. Hector, Sur un thorme de structure des feuilletages de codimension un, Thesis, Univ. de Strasbourg, 1972. [HEC2] , Sur le type des feuilletages transverses de Rz, C. R. Acad. Sci. Paris Sr, A-B 273 (1971), A810-A813. MR 44 #4773. [HER] M. Herman, Simplicit du groupe des diffomorphismes de classe C00, isotope l'identit, du tore de dimension n, C. R. Acad. Sci. Paris Sr. A-B 273 (1971), A232A234. MR 44 #4788. [HR] M. Hirsch, Stabitity of compact leaves offoliations, Proc. Internat. Conf. on Dynamical Systems, Salvador, Brazil, 1971, Academic Press, N.Y. (to appear). [JM] G. Joubert and R. Moussu, Feuilletage sans holonomie d'une varit ferme, C. R. Acad. Sci. Paris Sr. A-B 270 (1970), A507-A509. MR 41 #2702. [KT] F. W. Kamber and P. Tondeur, Characteristic invariants of foliated bundles, Univ. of Illinois (preprint). [KN] S. Kobayschi and K. Nomizu, Foundations of differential geometry, Vol. II, Interscience Tracts in Pure and Appl. Math., no. 15, Interscience, New York, 1969. MR 38 #6501. [KS] K. Kodaira and D. Spencer, Multifoliated structures, Ann. of Math. (2) 74 (1961), 52-100. MR 26 #5595. [KOI] U. Koschorke, Concordance andbordism of line elementfields,Rutgers Univ. (preprint). [K02] , Singularities andbordism ofq-planefieldsand o f foliations (to appear). [LA] P. Landweber, Complex structures on open manifolds, Rutgers Univ. (to appear).

1974]

FOLIATIONS

415

[L] H. B. Lawson, Jr. Codimension-one foliations of spheres, Ann. of Math. (2) 94 (1971), 494-503. MR 44 #4774. [LI] W. B. R. Lickorish, A foliation for 3-manifolds, Ann. of Math. (2)82(1965), 414-420. MR 32 #6488. [LM] E. Lima, Commuting vector fields on S\ Ann. of Math. (2) 81 (1965), 70-81. MR 30 #1517. [MAI] J. Mather, On Haefliger's classifying space. I, Bull. Amer. Math. Soc. 77 (1971), 1111-1115. MR 44 #1047. [MA2] , On Haefliger's classifying space. II, Harvard Univ. (to appear). [MA3] , The vanishing of the homology of certain groups of diffeomorphisms, Topology 10 (1971), 297-298. MR 44 #5973. [MA4] , Integrability in codimension-one, Comment. Math. Helv. 48 (1973), 195-233. [Ml] J. Milnor, Morse theory, Ann. of Math. Studies, no. 51, Princeton Univ. Press, Princeton, N.J., 1963. MR 29 #634. [M2] , Topology from the differentiate viewpoint, Benjamin, New York, 1967. [M3] , Singular points on complex hypersurfaces, Ann. of Math. Studies, no. 61, Princeton Univ. Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1968. MR 39 #969. [M4] , Foliations and foliated vector bundles, mimeographed notes, 1970. [MOI] R. Moussu, Sur les feuilletages de codimension un, Thesis, Orsay, France, 1971. [M02] , Feuilletage presque sans holonomie, C. R. Acad. Sci. Paris Sr. A-B 272 (1971), A114-A117. MR 43 #4060. [MP] R. Moussu and F. Pelletier, Sur le thorme de Poincar-Bendixon, Universit de Dijon (preprint). [MR1] R. Moussu and R. Roussarie, Une condition suffisante pour qu'un feuilletage soit sans holonomie, C. R. Acad. Sci. Paris Sr. A-B 271 (1970), A240-A243. MR 43 #2732. [MR2] , Relations de conjugasion et de eobordisme entre certains feuilletages, Inst. Hautes Etudes Sci. Publ Math, (to appear). [N] S. P. Novikov, Topology of foliations, Trudy Moskov. Mat. Obsc. 14 (1965), 248-278=Trans. Moscow Math. Soc. (1965), 268-304. MR 34 #824. [PI] J. Pasternak, Foliations and compact Lie group actions, Comment. Math. Helv. 46 (1971), 467-477. MR 45 #9353. [P2] , Riemannian Haefliger structures, Univ. of Rochester preprint (to appear). [PHI] A. Phillips, Submersions of open manifolds, Topology 6 (1967), 171-206. MR 34 #8420. [PH2] , Foliations of open manifolds. I, Comment. Math. Helv 43 (1968), 204-211. MR 37 #4829. [PH3] , Foliations of open manifolds. II, Comment. Math. Helv. 44 (1969), 367-370. MR 40 #6579. [PH4] , Smooth maps transverse to a foliation, Bull. Amer. Math. Soc. 76 (1970), 792-797. MR 41 #7711. [PH5] , Smooth maps of constant rank (to appear). [PLI] J. Plante, A generalization of the Poincar-Bendixon Theorem for foliations of codimension-one, Topology 12 (1973), 177-182. [PL2] , Asymptotic properties of foliations (to appear). [PO] V. Ponaru, Homotopy theory and differentiable singularities, ManifoldsAmsterdam 1970 (Proc. Nuffic Summer School), Lecture Notes in Math., vol. 197, Springer, Berlin, 1971, pp. 106-132. MR 44 #2250.

416

H. B. LAWSON, JR.

[May

[PU] C. Pugh, The closing lemma, Amer. J. Math. 89 (1967), 956-1009. MR 37 #2256. [Ql] N. van Que, Feuilletage singularits de varits de dimension 3 (Thorme de John Wood\ J. Differential Geometry 6 (1972), 473^78. [Q2] N. van Que and E. G. Wagneur, Foliations with singularities of 3-manifolds, Univ. of Montral (preprint). [RA] B. Raymond, Thesis, Orsay, 1970. [RI] G. Reeb, Sur certaines proprits topologiques des varits feuilletes, Actualits Sci. Indust., no. 1183=Publ. Inst. Math. Univ. Strasbourg 11, Hermann, Paris, 1952, pp. 91-154, 157-158. MR 14, 1113. [R2] , Sur les structures feuilletes de codimension un et sur un thorme de M. A. Denjoy, Ann. Inst. Fourier (Grenoble) 11 (1961), 185-200. MR 24 #A1135. [R3] , Feuillages, rsultats anciens et nouveaux, Univ. of Montreal, 1972. [Ril] B. L. Reinhart, Foliated manifolds with bundle-like metrics, Ann. of Math (2) 69 (1959), 119-132. MR 21 #6004. [RI2] , Harmonic integrals on foliated manifolds, Amer. J. Math. 81 (1959), 529-536. MR 21 #6005. [RI3] , Closed metric foliations, Michigan Math. J. 8 (1961), 7-9. MR 22 #11342. [RI4] , Characteristic numbers offoliated manifolds, Topology 6 (1967), 467471. MR 36 #4579. [RI5] , Cobordism and foliations, Ann. Inst. Fourier (Grenoble) 14 (1964), fasc. 1, 49-52. MR 29 #6501. [RI6] , Automorphisms and integrability of plane fields, J. Differential Geometry 6 (1971), 263-266. [RI7] , Indices for foliations of the 2-dimensional torus, Univ. of Maryland preprint (to appear), [RW] B. L. Reinhart and J. Wood, A metric formula for the Godbillon-Vey invariant or foliations (to appear). [RBI] S. A. Robertson, Grid manifolds, J. Differential Geometry 4 (1970), 245-253, MR 42 #3711. [RB2] , Symmetric foliations, University of Southampton (preprint). [RB3] , Parallelfoliations ofpseudoriemannian manifolds, University of Southampton (preprint). [ROI] H. Rosenberg, Actions ofRn on manifolds, Comment. Math. Helv. 41 (1966/ 67), 170-178. MR 34 #6794. [R02] , Foliations by planes, Topology 6 (1967), 131-138. MR 37 #3595. [R03] , The qualitative theory of foliations, Univ. of Montreal, 1973. [RRl] H. Rosenberg and R. Roussarie, Les feuilles exceptionnelles ne sont pas exceptionelles, Comment. Math. Helv. (1971), 43-49. [RR2] , Reeb foliations, Ann. of Math. (2) 91 (1970), 1-24. MR 41 #2704. [RR3] , Topological equivalence of Reeb foliations, Topology 9 (1970), 231242. MR 41 #7712. [RR4] , Some remarks on stability of foliations (to appear). [RRW] H. Rosenberg, R. Roussarie and D. Weil, A classification of closed orientable ^manifolds of rank two, Ann. of Math. (2) 91 (1970), 449-464. MR 42 #5280. [RT] H. Rosenberg and W. Thurston, Some remarks on foliations, Proc. Internat. Conf. on Dynamical Systems, Salvador, Brazil, 1971 (to appear). [ROU] R. Roussarie, Plongements dans les varits feuilletes et classification des feuilletages sans holonomie, Topology (to appear).

1974]

FOLIATIONS

417

[ROU2] R. Roussarie, Sur les feuilletages des varits de dimension trois, Ann. Inst. Fourier (Grenoble) 21 (1971). [ROU3] , Phnomnes de stabilit et d'instabilit dans les feuilletages, Orsay (preprint). [SI] R. Sacksteder, Some properties of foliations, Ann. Inst. Fourier (Grenoble) 14 (1964), fasc. 1, 31-35. MR 29 #4067. [S2] , On the existence of exceptional leaves in foliations of codimension-one, Ann. Inst. Fourier (Grenoble) 14 (1964), fasc. 2, 221-225. MR 30 #4267. [S3] , Foliations and pseudo-groups, Amer. J. Math. 87 (1965), 79-102. MR 30 #4268. [SS] R. Sacksteder and A. Schwartz, Limit sets for foliations, Ann. Inst. Fourier (Grenoble) 15 (1965), fasc. 2, 201-213. MR 32 #6489. [SC] P. A. Schweitzer, Counter-examples to the Seifert conjecture and opening closed leaves offoliations, Ann. of Math, (to appear). [SE] H. Seifert, Closed integral curves in 3-space and isotopic two-dimensional deformations, Proc. Amer. Math. Soc. 1 (1950), 287-302. MR 12, 273. [SH] H. Shulman, Characteristic classes and foliations, Thesis, Univ. of California, Berkeley, Calif., 1972. [SI] J. Simons, Characteristic forms and transgression. II: Characters associated to a connection, preprint S.U.N.Y., Stony Brook. [ST] N. Steenrod, The topology of fibre bundles, Princeton Math. Series, vol. 14, Princeton Univ. Press, Princeton, N.J., 1951. MR 12, 522. [STG] S. Sternberg, Lectures on differential geometry, Prentice-Hall, Englewood Cliffs, N.J., 1964. MR 33 #1797. [Tl] I. Tamura, Every odd dimensional homotopy sphere has a foliation of codimension one, Comment. Math. Helv. 47 (1972), 73-79. [T2] , Foliations of total spaces of sphere bundles over spheres (to appear). [T3] , Spinnable structures on differentiable manifolds, Proc. Japan Acad. 48 (1972), 293-296. [T4] , Foliations and spinnable structures on manifolds (to appear). [TS1] E. Thomas, Vectorfieldson manifolds, Bull. Amer. Math. Soc. 75 (1969), 643683. MR 39 #3522. [TS2] , Secondary obstructions to integrability, Proc. Conf. on Dynamical Systems and Foliations, Bahia, 1971. [TH1] W. Thurston, Foliations of 3-manifolds which are circle bundles, Thesis, University of California, Berkeley, Calif., 1972. [TH2] , Noncobordant foliations of S3, Bull. Amer. Math. Soc. 78 (1972), 511-514. MR 45 #7741. [TH3] , I.A.S. preprint (to appear). [TH4] , Foliations and groups of diffeomorphisms, I.A.S. preprint (to appear). [Til] D. Tischler, Totally parallelizable 3-manifolds, Topological Dynamics (Sympos., Colorado State Univ., Ft. Collins, Colo., 1967), Benjamin, New York, 1968, pp. 471492. MR 38 #3884. [TI2] , On fibering certain foliated manifolds over S1, Topology 9 (1970), 153-154. MR 41 #1069. [VI] A. Verjovsky, Thesis, Brown University. [Y2] , presented at the International Conference on Dynamical Systems, Salvador, Brazil, 1971, Bol. Soc. Mat. Mexicana 18 (1973). [WI] H. E. Winkelnkemper, Manifolds as open books, Bull. Amer. Math. Soc. 79 (1973), 45-51.

418

H. B. LAWSON, JR.

[Wl] J. Wood, Foliations on 1-manifolds, Ann. of Math. (2) 89 (1969), 336-358. MR 40 #2123. [\Y2] , Foliations of codimension-one, Bull. Amer. Math. Soc. 76 (1970), 1107-1111. MR 41 #7713. [W3] , Bundles with totally disconnected structure group. Comment. Math. Helv. 46 (1971), 257-273. MR 45 #2732. 1 1 [W4] 9 Foliated S bundles and diffeomorphisms of S , Dynamical Systems, Academic Press, New York 1973.
SCHOOL OF MATHEMATICS, INSTITUTE FOR ADVANCED STUDY, PRINCETON, NEW JERSEY

08540 Current address: Department of Mathematics, University of California, Berkeley, California 94720