THE POINCARE CONJECTURE

JOHN MILNOR

1. Introduction The topology of two-dimensional manifolds or surfaces was well understood in the 19th century. In fact there is a simple list of all possible smooth compact orientable surfaces. Any such surface has a well-dened genus g 0, which can be described intuitively as the number of holes; and two such surfaces can be put into a smooth one-to-one correspondence with each other if and only if they have the same genus.1 The corresponding question in higher dimensions is much more

Figure 1. Sketches of smooth surfaces of genus 0, 1, and 2. dicult. Henri Poincar was perhaps the rst to try to make a similar study e of three-dimensional manifolds. The most basic example of such a manifold is the three-dimensional unit sphere, that is, the locus of all points (x, y, z, w) in four-dimensional Euclidean space which have distance exactly 1 from the origin: x2 + y 2 + z 2 + w2 = 1. He noted that a distinguishing feature of the two-dimensional sphere is that every simple closed curve in the sphere can be deformed continuously to a point without leaving the sphere. In 1904, he asked a corresponding question in dimension 3. In more modern language, it can be phrased as follows:2 Question. If a compact three-dimensional manifold M 3 has the property that every simple closed curve within the manifold can be deformed continuously to a point, does it follow that M 3 is homeomorphic to the sphere S 3 ? He commented, with considerable foresight, Mais cette question nous entra nerait trop loin. Since then, the hypothesis that every simply connected closed 3-manifold is homeomorphic to the 3-sphere has been known as the Poincar Cone jecture. It has inspired topologists ever since, and attempts to prove it have led to many advances in our understanding of the topology of manifolds.
1For denitions and other background material, see, for example, [21] or [29], as well as [48]. 2See [36, pages 498 and 370]. To Poincar, manifolds were always smooth or polyhedral, so e

that his term homeomorphism referred to a smooth or piecewise linear homeomorphism.

JOHN MILNOR

2. Early Missteps From the rst, the apparently simple nature of this statement has led mathematicians to overreach. Four years earlier, in 1900, Poincar himself had been the e rst to err, stating a false theorem that can be phrased as follows. False Theorem. Every compact polyhedral manifold with the homology of an ndimensional sphere is actually homeomorphic to the n-dimensional sphere. But his 1904 paper provided a beautiful counterexample to this claim, based on the concept of fundamental group, which he had introduced earlier (see [36, pp. 189192 and 193288]). This example can be described geometrically as follows. Consider all possible regular icosahedra inscribed in the two-dimensional unit sphere. In order to specify one particular icosahedron in this family, we must provide three parameters. For example, two parameters are needed to specify a single vertex on the sphere, and then another parameter to specify the direction to a neighboring vertex. Thus each such icosahedron can be considered as a single point in the three-dimensional manifold M 3 consisting of all such icosahedra.3 This manifold meets Poincars preliminary criterion: By the methods of homology e theory, it cannot be distinguished from the three-dimensional sphere. However, he could prove that it is not a sphere by constructing a simple closed curve that cannot be deformed to a point within M 3 . The construction is not dicult: Choose some representative icosahedron and consider its images under rotation about one vertex through angles 0 2/5. This denes a simple closed curve in M 3 that cannot be deformed to a point.

Figure 2. The Whitehead link The next important false theorem was by Henry Whitehead in 1934 [52]. As part of a purported proof of the Poincar Conjecture, he claimed the sharper statee ment that every open three-dimensional manifold that is contractible (that can be continuously deformed to a point) is homeomorphic to Euclidean space. Following in Poincars footsteps, he then substantially increased our understanding of the e topology of manifolds by discovering a counterexample to his own theorem. His counterexample can be briey described as follows. Start with two disjoint solid tori T0 and T1 in the 3-sphere that are embedded as shown in Figure 2, so that each one individually is unknotted, but so that the two are linked together with linking number zero. Since T1 is unknotted, its complement T1 = S 3 interior(T1 )
3In more technical language, this M 3 can be dened as the coset space SO(3)/I where SO(3) 60 is the group of all rotations of Euclidean 3-space and where I60 is the subgroup consisting of the 60 rotations that carry a standard icosahedron to itself. The fundamental group 1 (M 3 ), consisting of all homotopy classes of loops from a point to itself within M 3 , is a perfect group of order 120.

 THE POINCARE CONJECTURE

is another unknotted solid torus that contains T0 . Choose a homeomorphism h of the 3-sphere that maps T0 onto this larger solid torus T1 . Then we can inductively construct solid tori T0 T1 T2 in S 3 by setting Tj+1 = h(Tj ). The union M 3 = Tj of this increasing sequence is the required Whitehead counterexample, a contractible manifold that is not homeomorphic to Euclidean space. To see that 1 (M 3 ) = 0, note that every closed loop in T0 can be shrunk to a point (after perhaps crossing through itself) within the larger solid torus T1 . But every closed loop in M 3 must be contained in some Tj , and hence can be shrunk to a point within Tj+1 M 3 . On the other hand, M 3 is not homeomorphic to Euclidean 3-space since, if K M 3 is any compact subset large enough to contain T0 , one can prove that the dierence set M 3 K is not simply connected. Since this time, many false proofs of the Poincar Conjecture have been proposed, e some of them relying on errors that are rather subtle and dicult to detect. For a delightful presentation of some of the pitfalls of three-dimensional topology, see [4]. 3. Higher Dimensions The late 1950s and early 1960s saw an avalanche of progress with the discovery that higher-dimensional manifolds are actually easier to work with than threedimensional ones. One reason for this is the following: The fundamental group plays an important role in all dimensions even when it is trivial, and relations between generators of the fundamental group correspond to two-dimensional disks, mapped into the manifold. In dimension 5 or greater, such disks can be put into general position so that they are disjoint from each other, with no self-intersections, but in dimension 3 or 4 it may not be possible to avoid intersections, leading to serious diculties. Stephen Smale announced a proof of the Poincar Conjecture in high dimensions e in 1960 [41]. He was quickly followed by John Stallings, who used a completely dierent method [43], and by Andrew Wallace, who had been working along lines quite similar to those of Smale [51]. Let me rst describe the Stallings result, which has a weaker hypothesis and easier proof, but also a weaker conclusion. He assumed that the dimension is seven or more, but Christopher Zeeman later extended his argument to dimensions 5 and 6 [54]. StallingsZeeman Theorem. If M n is a nite simplicial complex of dimension n 5 that has the homotopy type4 of the sphere S n and is locally piecewise linearly homeomorphic to the Euclidean space Rn , then M n is homeomorphic to S n under a homeomorphism that is piecewise linear except at a single point. In other words, the complement M n (point) is piecewise linearly homeomorphic to Rn . The method of proof consists of pushing all of the diculties o toward a single point; hence there can be no control near that point.
4In order to check that a manifold M n has the same homotopy type as the sphere S n , we must check not only that it is simply connected, 1 (M n ) = 0, but also that it has the same homology as the sphere. The example of the product S 2 S 2 shows that it is not enough to assume that 1 (M n ) = 0 when n > 3.

JOHN MILNOR

The Smale proof, and the closely related proof given shortly afterward by Wallace, depended rather on dierentiable methods, building a manifold up inductively, starting with an n-dimensional ball, by successively adding handles. Here a k-handle can be added to a manifold M n with boundary by rst attaching a k-dimensional cell, using an attaching homeomorphism from the (k 1)-dimensional boundary sphere into the boundary of M n , and then thickening and smoothing corners so as to obtain a larger manifold with boundary. The proof is carried out by rearranging and canceling such handles. (Compare the presentation in [24].)

Figure 3. A three-dimensional ball with a 1-handle attached Smale Theorem. If M n is a dierentiable homotopy sphere of dimension n 5, then M n is homeomorphic to S n . In fact, M n is dieomorphic to a manifold obtained by gluing together the boundaries of two closed n-balls under a suitable dieomorphism. This was also proved by Wallace, at least for n 6. (It should be noted that the ve-dimensional case is particularly dicult.) The much more dicult four-dimensional case had to wait twenty years, for the work of Michael Freedman [8]. Here the dierentiable methods used by Smale and Wallace and the piecewise linear methods used by Stallings and Zeeman do not work at all. Freedman used wildly non-dierentiable methods, not only to prove the four-dimensional Poincar Conjecture for topological manifolds, but also to give e a complete classication of all closed simply connected topological 4-manifolds. The integral cohomology group H 2 of such a manifold is free abelian. Freedman needed just two invariants: The cup product : H 2 H 2 H 4 Z is a symmetric = bilinear form with determinant 1, while the KirbySiebenmann invariant is an integer mod 2 that vanishes if and only if the product manifold M 4 R can be given a dierentiable structure. Freedman Theorem. Two closed simply connected 4-manifolds are homeomorphic if and only if they have the same bilinear form and the same KirbySiebenmann invariant . Any can be realized by such a manifold. If (x x) is odd for some x H 2 , then either value of can be realized also. However, if (x x) is always even, then is determined by , being congruent to one eighth of the signature of .

 THE POINCARE CONJECTURE

In particular, if M 4 is a homotopy sphere, then H 2 = 0 and = 0, so M 4 is homeomorphic to S 4 . It should be noted that the piecewise linear or dierentiable theories in dimension 4 are much more dicult. It is not known whether every smooth homotopy 4-sphere is dieomorphic to S 4 ; it is not known which 4manifolds with = 0 actually possess dierentiable structures; and it is not known when this structure is essentially unique. The major results on these questions are due to Simon Donaldson [7]. As one indication of the complications, Freedman showed, using Donaldsons work, that R4 admits uncountably many inequivalent dierentiable structures. (Compare [12].) In dimension 3, the discrepancies between topological, piecewise linear, and differentiable theories disappear (see [18], [28], and [26]). However, diculties with the fundamental group become severe. 4. The Thurston Geometrization Conjecture In the two-dimensional case, each smooth compact surface can be given a beautiful geometrical structure, as a round sphere in the genus zero case, as a at torus in the genus 1 case, and as a surface of constant negative curvature when the genus is 2 or more. A far-reaching conjecture by William Thurston in 1983 claims that something similar is true in dimension 3 [46]. This conjecture asserts that every compact orientable three-dimensional manifold can be cut up along 2-spheres and tori so as to decompose into essentially unique pieces, each of which has a simple geometrical structure. There are eight possible three-dimensional geometries in Thurstons program. Six of these are now well understood,5 and there has been a great deal of progress with the geometry of constant negative curvature.6 The eighth geometry, however, corresponding to constant positive curvature, remains largely untouched. For this geometry, we have the following extension of the Poincar Conjecture. e Thurston Elliptization Conjecture. Every closed 3-manifold with nite fundamental group has a metric of constant positive curvature and hence is homeomorphic to a quotient S 3 /, where SO(4) is a nite group of rotations that acts freely on S 3 . The Poincar Conjecture corresponds to the special case where the group = e 1 (M 3 ) is trivial. The possible subgroups SO(4) were classied long ago by [19] (compare [23]), but this conjecture remains wide open. 5. Approaches through Differential Geometry and Differential Equations7 In recent years there have been several attacks on the geometrization problem (and hence on the Poincar Conjecture) based on a study of the geometry of the e innite dimensional space consisting of all Riemannian metrics on a given smooth three-dimensional manifold.
5See, for example, [13], [3], [38, 39, 40], [49], [9], and [6]. 6See [44], [27], [47], [22], and [30]. The pioneering papers by [14] and [50] provided the basis

for much of this work. 7Added in 2004

JOHN MILNOR

By denition, the length of a path on a Riemannian manifold is computed, in terms of the metric tensor gij , as the integral ds =

gij dxi dxj .

From the rst and second derivatives of this metric tensor, one can compute the Ricci curvature tensor Rij , and the scalar curvature R. (As an example, for the at Euclidean space one gets Rij = R = 0, while for a round three-dimensional sphere of radius r, one gets Ricci curvature Rij = 2gij /r2 and scalar curvature R = 6/r2 .) One approach by Michael Anderson, based on ideas of Hidehiko Yamabe [53], studies the total scalar curvature R dV as a functional on the space of all M3 smooth unit volume Riemannian metrics. The critical points of this functional are the metrics of constant curvature (see [1]). A dierent approach, initiated by Richard Hamilton studies the Ricci ow [15, 16, 17], that is, the solutions to the dierential equation dgij = 2Rij . dt In other words, the metric is required to change with time so that distances decrease in directions of positive curvature. This is essentially a parabolic dierential equationa and behaves much like the heat equation studied by physicists: If we heat one end of a cold rod, then the heat will gradually ow throughout the rod until it attains an even temperature. Similarly, a naive hope for 3-manifolds with nite fundamental group might have been that, under the Ricci ow, positive curvature would tend to spread out until, in the limit (after rescaling to constant size), the manifold would attain constant curvature. If we start with a 3-manifold of positive Ricci curvature, Hamilton was able to carry out this program and construct a metric of constant curvature, thus solving a very special case of the Elliptization Conjecture. However, in the general case, there are very serious diculties, since this ow may tend toward singularities.8 I want to thank many mathematicians who helped me with this report. May 2000, revised June 2004 References
[1] M.T. Anderson, Scalar curvature, metric degenerations and the static vacuum Einstein equations on 3-manifolds, Geom. Funct. Anal. 9 (1999), 855963 and 11 (2001) 273381. See also: Scalar curvature and the existence of geometric structures on 3-manifolds, J. reine angew. Math. 553 (2002), 125182 and 563 (2003), 115195. [2] M.T. Anderson, Geometrization of 3-manifolds via the Ricci ow, Notices AMS 51 (2004), 184193. [3] L. Auslander and F.E.A. Johnson, On a conjecture of C.T.C. Wall, J. Lond. Math. Soc. 14 (1976), 331332. [4] R.H. Bing, Some aspects of the topology of 3-manifolds related to the Poincar conjecture, e in Lectures on Modern Mathematics II (T. L. Saaty, ed.), Wiley, New York, 1964. [5] J. Birman, Poincars conjecture and the homeotopy group of a closed, orientable 2-manifold, e J. Austral. Math. Soc. 17 (1974), 214221.
8Grisha Perelman, in St. Petersburg, has posted three preprints on arXiv.org which go a long way toward resolving these diculties, and in fact claim to prove the full geometrization conjecture [32, 33, 34]. These preprints have generated a great deal of interest. (Compare [2] and [25], as well as the website http://www.math.lsa.umich.edu/research/ricciow/perelman.html organized by B. Kleiner and J. Lott.) However, full details have not appeared.

 THE POINCARE CONJECTURE

[6] A. Casson and D. Jungreis, Convergence groups and Seifert bered 3-manifolds, Invent. Math. 118 (1994), 441456. [7] S.K. Donaldson, Self-dual connections and the topology of smooth 4-manifolds, Bull. Amer. Math. Soc. 8 (1983), 8183. [8] M.H. Freedman, The topology of four-dimensional manifolds, J. Di. Geom. 17 (1982), 357 453. [9] D. Gabai, Convergence groups are Fuchsian groups, Ann. Math. 136 (1992), 447510. [10] D. Gabai, Valentin Poenarus program for the Poincar conjecture, in Geometry, topology, e & physics, Conf. Proc. Lecture Notes Geom. Topology, VI, Internat. Press, Cambridge, MA, 1995, 139166. [11] D. Gillman and D. Rolfsen, The Zeeman conjecture for standard spines is equivalent to the Poincar conjecture, Topology 22 (1983), 315323. e [12] R. Gompf, An exotic menagerie, J. Dierential Geom. 37 (1993) 199223. [13] C. Gordon and W. Heil, Cyclic normal subgroups of fundamental groups of 3-manifolds, Topology 14 (1975), 305309. [14] W. Haken, Uber das Homomorphieproblem der 3-Mannigfaltigkeiten I, Math. Z. 80 (1962), o 89120. [15] R.S. Hamilton, Three-manifolds with positive Ricci curvature, J. Dierential Geom. 17 (1982), 255306. [16] R.S. Hamilton, The formation of singularities in the Ricci ow, in Surveys in dierential geometry, Vol. II (Cambridge, MA, 1993), Internat. Press, Cambridge, MA, 1995, 7136. [17] R.S. Hamilton, Non-singular solutions of the Ricci ow on three-manifolds Comm. Anal. Geom. 7 (1999), 695729. [18] M. Hirsch, Obstruction theories for smoothing manifolds and maps, Bull. Amer. Math. Soc. 69 (1963), 352-356. [19] H. Hopf, Zum CliordKleinschen Raumproblem, Math. Ann. 95 (1925-26) 313-319. [20] W. Jakobsche, The Bing-Borsuk conjecture is stronger than the Poincar conjecture, Fund. e Math. 106 (1980), 127134. [21] W.S. Massey, Algebraic Topology: An Introduction, Harcourt Brace, New York, 1967; Springer, New York 1977; or A Basic Course in Algebraic Topology, Springer, New York, 1991. [22] C. McMullen, Riemann surfaces and geometrization of 3-manifolds, Bull. Amer. Math. Soc. 27 (1992), 207216. [23] J. Milnor, Groups which act on S n without xed points, Amer. J. Math. 79 (1957), 623630. [24] J. Milnor (with L. Siebenmann and J. Sondow), Lectures on the h-Cobordism Theorem, Princeton Math. Notes, Princeton University Press, Princeton, 1965. [25] J. Milnor, Towards the Poincar conjecture and the classication of 3-manifolds, Notices e AMS 50 (2003), 12261233. [26] E.E. Moise, Geometric Topology in Dimensions 2 and 3, Springer, New York, 1977. [27] J. Morgan, On Thurstons uniformization theorem for three-dimensional manifolds, in The Smith Conjecture (H. Bass and J. Morgan, eds.), Pure and Appl. Math. 112, Academic Press, New York, 1984, 37125. [28] J. Munkres, Obstructions to the smoothing of piecewise-dierentiable homeomorphisms, Ann. Math. 72 (1960), 521554. [29] J. Munkres, Topology: A First Course, PrenticeHall, Englewood Clis, NJ, 1975. [30] J.-P. Otal, The hyperbolization theorem for bered 3-manifolds, translated from the 1996 French original by Leslie D. Kay, SMF/AMS Texts and Monographs 7, American Mathematical Society, Providence, RI; Socit Mathatique de France, Paris, 2001. e e [31] C. Papakyriakopoulos, A reduction of the Poincar conjecture to group theoretic conjectures, e Ann. Math. 77 (1963), 250305. [32] G. Perelman, The entropy formula for the Ricci ow and its geometric applications, arXiv: math.DG/0211159v1, 11 Nov 2002. [33] G. Perelman, Ricci ow with surgery on three-manifolds, arXiv: math.DG/0303109, 10 Mar 2003. [34] G. Perelman, Finite extinction time for the solutions to the Ricci ow on certain threemanifolds, arXiv: math.DG/0307245, 17 Jul 2003.

JOHN MILNOR

[35] V. Ponaru, A program for the Poincar conjecture and some of its ramications, in Topics e e in low-dimensional topology (University Park, PA, 1996), World Sci. Publishing, River Edge, NJ, 1999, 6588. [36] H. Poincar, uvres, Tome VI, GauthierVillars, Paris, 1953. e [37] C. Rourke, Algorithms to disprove the Poincar conjecture, Turkish J. Math. 21 (1997), e 99110. [38] P. Scott, A new proof of the annulus and torus theorems, Amer. J. Math. 102 (1980), 241 277. [39] P. Scott, There are no fake Seifert bre spaces with innite 1 , Ann. Math. 117 (1983), 3570. [40] P. Scott, The geometries of 3-manifolds, Bull. Lond. Math. Soc. 15 (1983), 401487. [41] S. Smale, Generalized Poincars conjecture in dimensions greater than four, Ann. Math. 74 e (1961), 391406. (See also: Bull. Amer. Math. Soc. 66 (1960), 373375.) [42] S. Smale, The story of the higher dimensional Poincar conjecture (What actually happened e on the beaches of Rio), Math. Intelligencer 12, no. 2 (1990), 4451. [43] J. Stallings, Polyhedral homotopy spheres, Bull. Amer. Math. Soc. 66 (1960), 485488. [44] D. Sullivan, Travaux de Thurston sur les groupes quasi-fuchsiens et sur les varits hyperee boliques de dimension 3 bres sur le cercle, Sm. Bourbaki 554, Lecture Notes Math. 842, e e Springer, New York, 1981. [45] T.L. Thickstun, Open acyclic 3-manifolds, a loop theorem and the Poincar conjecture, Bull. e Amer. Math. Soc. (N.S.) 4 (1981), 192194. [46] W.P. Thurston, Three dimensional manifolds, Kleinian groups and hyperbolic geometry, in The Mathematical heritage of Henri Poincar, Proc. Symp. Pure Math. 39 (1983), Part 1. e (Also in Bull. Amer. Math. Soc. 6 (1982), 357381.) [47] W.P. Thurston, Hyperbolic structures on 3-manifolds, I, deformation of acyclic manifolds, Ann. Math. 124 (1986), 203246 [48] W.P. Thurston, Three-Dimensional Geometry and Topology, Vol. 1, ed. by Silvio Levy, Princeton Mathematical Series 35, Princeton University Press, Princeton, 1997. [49] P. Tukia, Homeomorphic conjugates of Fuchsian groups, J. Reine Angew. Math. 391 (1988), 154. [50] F. Waldhausen, On irreducible 3-manifolds which are suciently large, Ann. Math. 87 (1968), 5688. [51] A. Wallace, Modications and cobounding manifolds, II, J. Math. Mech 10 (1961), 773809. [52] J.H.C. Whitehead, Mathematical Works, Volume II, Pergamon Press, New York, 1962. (See pages 21-50.) [53] H. Yamabe, On a deformation of Riemannian structures on compact manifolds, Osaka Math. J. 12 (1960), 2137. [54] E.C. Zeeman, The Poincar conjecture for n 5 , in Topology of 3-Manifolds and Related e Topics PrenticeHall, Englewood Clis, NJ, 1962, 198204. (See also Bull. Amer. Math. Soc. 67 (1961), 270.)

(Note: For a representative collection of attacks on the Poincar Conjecture, see e [31], [5], [20], [45], [11], [10], [37], and [35].)