homotopy sphere. The Poincaré conjecture asserts that S 3 is the only such manifold.

(It seems that Poincaré himself did not explicitly formulate this as a conjecture.) As
we shall see shortly, in the nonsimply-connected case there are many examples of
closed 3 manifolds with isomorphic fundamental groups that are not diffeomorphic.
The Classification of 3-Manifolds — A Brief Overview
Prime Decomposition
The first reduction of the classification problem is due to Kneser around 1930. If
Allen Hatcher
a 3 manifold M contains an embedded sphere S 2 (disjoint from the boundary of M ,
if M has a nonempty boundary) separating M into two components, we can split M
along this S 2 into manifolds M1 and M2 each having this sphere as a component of
Our aim here is to sketch the very nice conjectural picture of the classification its boundary. We can then fill in these two boundary spheres with balls to produce
of closed orientable 3 manifolds that emerged around 1980 as a consequence of geo- manifolds N1 and N2 that are closed if M was closed. One says M is the connected
metric work of Thurston and more topological results of Jaco-Shalen and Johannson sum of N1 and N2 , or in symbols, M = N1 ] N2 . This splitting operation is commu-
at the end of a long chain of topological developments going back 50 years or more. tative by definition, and it is not hard to check that it is also associative. One rather
A pleasant feature of 3 manifolds, in contrast to higher dimensions, is that there trivial possibility for the splitting 2 sphere is as the boundary of a ball in M , and
is no essential difference between smooth, piecewise linear, and topological mani- this gives the decomposition M = M ] S 3 . If this is the only way that M splits as a
folds. It was shown by Bing and Moise in the 1950s that every topological 3-manifold connected sum, then M is said to be prime. This is equivalent to saying that every
can be triangulated as a simplicial complex whose combinatorial type is unique up separating 2 sphere in M bounds a ball in M .
to subdivision. And every triangulation of a 3 manifold can be taken to be a smooth A fundamental theorem of Alexander from 1924 says that every 2 sphere in S 3
triangulation in some differential structure on the manifold, unique up to diffeomor- bounds a ball on each side. (We are assuming smoothness here, which rules out local
phism. Thus every topological 3 manifold has a unique smooth structure, and the pathology as in the Alexander horned sphere.) Hence S 3 is prime. Note that if this
classifications up to diffeomorphism and homeomorphism coincide. In what follows were not true then there would be no prime 3 manifolds, although perhaps something
we will deal with smooth manifolds and diffeomorphisms between them. could be salvaged by redefining primeness to allow for units, as in algebra. Fortunately
When we say “manifold” we will always mean “connected manifold”. For the this is not necessary. Of course, if S 3 were not prime, then the Poincaré conjecture
sake of simplicity we restrict attention to orientable manifolds, although with more would be false since a connected summand of a simply-connected manifold is also
trouble the nonorientable case could be covered as well. The primary focus will be simply-connected, as π1 (M ] N) is the free product π1 (M) ∗ π1 (N) by van Kampen’s
on manifolds that are closed, that is, compact and without bounday, but from time to theorem.
time it will be natural to consider also compact manifolds with nonempty boundary.
Kneser’s Theorem. Every compact orientable 3 manifold M factors as a connected
The most powerful of the standard invariants of algebraic topology for distin-
sum of primes, M = P1 ] ··· ] Pn , and this decomposition is unique up to insertion or
guishing 3 manifolds is the fundamental group. This determines all the homology
deletion of S 3 summands.
groups of a closed orientable 3 manifold M . Namely, H1 (M) is the abelianization of
π1 (M) , and by Poincaré duality H2 (M) is isomorphic to H 1 (M) which is H1 (M) mod If one assumes the Poincaré conjecture, then the existence of a prime decompo-
torsion by the universal coefficient theorem. Since M is closed and orientable, H3 (M) sition follows easily from the algebra fact that a finitely-generated group cannot be
is Z . All higher homology groups are zero, of course. split as a free product of an arbitrarily large number of nontrivial factors. Kneser’s
In particular, if M is a simply-connected closed 3 manifold, then M has the same proof is noteworthy for being independent of the Poincaré conjecture. If there were
homology groups as the 3 sphere S 3 . In fact M is homotopy equivalent to S 3 . For homotopy spheres other than S 3 , they too could be decomposed uniquely as sums of
by the Hurewicz theorem π2 (M) = H2 (M) = 0 and π3 (M) = H3 (M) = Z . A generator prime homotopy spheres.
of π3 (M) is represented by a map S 3 →M 3 of degree one, inducing an isomorphism At first glance one might think that the unique prime decomposition immediately
on H3 = π3 . This means we have a map S 3
→M of simply-connected simplicial com- reduced the classification of closed oriented 3 manifolds to the classification of the
plexes inducing isomorphisms on all homology groups, so by Whitehead’s theorem prime manifolds, but there is a small subtlety here: The prime factors Pi are uniquely
the map is a homotopy equivalence. Thus a simply-connected closed 3 manifold is a determined by M , but the converse need not be true. Given two manifolds P and Q
there are are two potentially different ways of forming their connected sum since after with finite π1 . This conjecture can be broken into two parts. The first is to show that
removing the interior of a ball from each of P and Q , one could glue the two resulting the universal cover of such a manifold is S 3 , which is the Poincaré conjecture, and then
boundary spheres together by a diffeomorphism that either preserves or reverses cho- the second part is to show that any free action of a finite group on S 3 is equivalent
sen orientations of the two 2 spheres. If either P or Q has a self-diffeomorphism that to an action by isometrics of S 3 with its standard metric. In particular this involves
reverses orientation, then the two gluings produce diffeomorphic connected sums, but showing that the group is a subgroup of SO(4) . There have been many partial results
otherwise they will not. One way to avoid this ambiguity it to talk about manifolds that on this second part over the years, but it has proved to be a very difficult problem.
are not just orientable but oriented. Then there is a unique way to form connected Type II: infinite cyclic fundamental group. It turns out that there is only one prime
sums respecting orientations, and Kneser’s theorem remains true for oriented mani- closed 3 manifold satisfying this condition, and that is S 1 × S 2 . This is also the only
folds. This reduces the classification problem to classifying oriented prime manifolds, orientable 3 manifold that is prime but not irreducible, where a 3 manifold M is
which in particular involves deciding for each prime orientable manifold P whether irreducible if every 2 sphere in M bounds a ball in M . For if M is reducible but prime
there is a diffeomorphism P →P that reverses orientation. it must contain a nonseparating S 2 . This has a product neighborhood S 2 × I , and the
It turns out that there exist many prime 3 manifolds that do not have orientation- union of this neighborhood with a tubular neighborhood of an arc joining S 2 × {0}
reversing self-diffeomorphisms. The two ways of forming the connected sum of two to S 2 × {1} in the complement of S 2 × I is diffeomorphic to the complement of a ball
such manifolds will produce nondiffeomorphic manifolds with isomorphic fundamen- in S 1 × S 2 . This says that M has S 1 × S 2 as a connected summand, so by primeness
tal groups since the isomorphism π1 (P ] Q) ≈ π1 (P ) ∗ π1 (Q) is valid no matter how M = S1 × S2 .
the two summands are glued together. Another special feature of S 1 × S 2 is that it is the only prime orientable 3 manifold
Prime 3 manifolds that are closed and orientable can be lumped broadly into with nontrivial π2 . This is a consequence of the Sphere Theorem, which says that for
three classes: an orientable 3 manifold M , if π2 (M) is nonzero then there is an embedded sphere
f is
Type I: finite fundamental group. For such a manifold M the universal cover M in M that represents a nontrivial element of π2 (M) . This sphere cannot bound a ball,
simply-connected and closed, hence a homotopy sphere. All the known examples so M is reducible, hence if it is prime it must be S 1 × S 2 .
3
are spherical 3 manifolds, of the form M = S /Γ for Γ a finite subgroup of SO(4) Type III: infinite noncyclic fundamental group. Such a manifold M is a K(π , 1) ,
acting freely on S 3 by rotations. Thus S 3 is the universal cover of M and Γ = π1 (M) . or in other words, its universal cover is contractible. More generally, any irreducible
Spherical 3 manifolds were explicitly classified in the 1930s, using the fact that SO(4) 3 manifold M , not necessarily closed, with π1 (M) infinite is a K(π , 1) . For the uni-
is a 2 sheeted covering group of SO(3)× SO(3) , so the finite subgroups of SO(4) can f is simply-connected and has trivial homology groups: By the Hurewicz
versal cover M
be determined from the well-known finite subgroups of SO(3) . The examples with theorem H2 (Mf) = π2 (Mf) and π2 (Mf) = π2 (M) = 0 by the Sphere Theorem and the
Γ cyclic are known as lens spaces, and there are also a few infinite families with Γ f) = 0 since M
irreducibility of M . Finally, H3 (M f is noncompact, and all higher homol-
noncylic, including Poincaré’s famous homology sphere, which can be defined as the ogy groups vanish since M f is a 3 manifold. So Whitehead’s theorem implies that Mf
coset space SO(3)/I where I is the group of rotational symmetries of the icosahedron, is contractible.
of order 60 . Since the universal cover of SO(3) is S 3 , a 2 sheeted cover, the quotient The homotopy type of a K(π , 1) is uniquely determined by its fundamental group,
SO(3)/I is S 3 /Γ for Γ the preimage of I in S 3 , a group of order 120 . so for a closed 3 manifold M to be a K(π , 1) imposes strong restrictions on the group
There is a lens space Lp/q for each fraction p/q between 0 and 1 . The funda- π1 (M) = π . For example, it implies that π1 (M) must be torsionfree, since the cov-
mental group of Lp/q is the cyclic group Zq , and two lens spaces Lp/q and Lp0 /q0 ering space of M corresponding to a nontrivial finite cyclic subgroup Zn would be
are diffeomorphic if and only if q = q0 and p 0 is congruent to ±p ±1 mod q . For a finite-dimensional K(Zn , 1) CW complex, but this cannot exist since the homology
example, if q is prime one obtains at least (q − 1)/4 nondiffeomorphic lens spaces groups of a K(Zn , 1) are nonzero in infinitely many dimensions, as one can choose
with the same fundamental group Zq . If orientations are taken into account, then the an infinite-dimensional lens space as a K(Zn , 1) . One can also see that the only free
condition p 0 ≡ ±p ±1 mod q becomes p 0 ≡ p ±1 mod q , so there are many lens spaces abelian group Zn that can occur as π1 (M) is Z3 , the fundamental group of the 3 torus,
without orientation-reversing self-diffeomorphisms, for example L1/3 . The spherical since the n torus is a K(Zn , 1) and this has Hn nonzero and Hi zero for i > n . In
manifolds that are not lens spaces are determined up to diffeomorphism by their particular, there is no closed orientable 3 manifold K(π , 1) with infinite cyclic fun-
fundamental group. damental group, which shows that S 1 × S 2 is indeed the only prime closed orientable
It is an old conjecture that spherical 3 manifolds are the only closed 3 manifolds 3 manifold with π1 infinite cyclic.
 Since the homotopy type of an irreducible closed K(π , 1) 3 manifold is deter- Seifert Manifolds
mined by its fundmental group, one may then ask whether the fundamental group There is a special family of 3 manifolds called Seifert manifolds for which an
in fact determines the manifold completely. This is the 3 dimensional case of the explicit classification was made early in the history of 3 manifolds, in the 1930s.
Borel conjecture, that a closed n manifold that is a K(π , 1) is determined up to These manifolds are “singular fiber bundles” with base space a compact surface and
homeomorphism by its fundamental group. (In high dimensions it is important to fibers circles. The fibering is locally trivial, as in an ordinary fiber bundle, except for
say “homeomorphism” here rather than “diffeomorphism”.) No counterexamples are a finite number of isolated “multiple” fibers where the local model is the following.
known in any dimension. In the 3 dimensional case, Waldhausen proved the con- Start with a product D 2 × I fibered by the intervals {x}× I . If we glue the two ends
jecture for a large class of manifolds known as Haken manifolds. These are the ir- D 2 × {0} and D 1 × {1} together by the identity map of D 2 we would then have the
reducible compact orientable 3 manifolds M that contain an embedded nonsimply- standard product fibering of the solid torus D 2 × S 1 . Instead of doing this, we glue
connected compact orientable surface S for which the map π1 (S)→π1 (M) induced the two ends together using a rotation of D 2 through an angle of 2π p/q for some
by the inclusion S >M is injective. Such a surface S is said to be incompressible. pair of relatively prime integers p and q with q ≥ 2 . The resulting quotient manifold
In case S has a nonempty boundary ∂S one requires that S is properly embedded is still D 2 × S 1 and the core interval {0}× I closes up to form a circle, but all the other
in M , meaning that S ∩ ∂M = ∂S . (A properly embedded disk that does not split off intervals {x}× I do not immediately close to form circles. Instead one has to follow
a ball from M is usually considered to be incompressible as well.) It turns out that along q of these intervals before they close up to a circle. Thus we have the solid
for a Haken manifold M it is possible to performing a finite sequence of splitting op- torus decomposed into disjoint circles which are all approximately parallel locally,
erations along incompressible surfaces, so that in the end the manifold M has been but the central circle has multiplicty q in the sense that the projection of each nearby
reduced to a finite collection of disjoint balls. This sequence of splittings is called circle onto this core circle is a q -to- 1 covering space. If the core circle is deleted one
a hierarchy for M . In favorable cases a hierarchy can be used to construct proofs would have an actual fiber bundle, so there is just a single isolated singular fiber in
by induction over the successive steps in the hierarchy, and in particular this is how this fibering of D 2 × S 1 by circles. In a compact Seifert manifold M one can have a
Waldhausen’s theorem is proved. finite number of such singular fibers, all disjoint from ∂M , which, if it is nonempty,
It is still not clear whether a “typical” closed irreducible 3 manifold with infinite consists of tori with product fiberings by circles. The base space of the fibering is the
π1 is a Haken manifold. Some manifolds that are Haken manifolds are: quotient space of M obtained by identifying each circle fiber to a point. This is still a
compact surface, even at the images of the singular fibers. We are assuming that M
Products F × S 1 where F is a closed orientable surface. is orientable, but the base surface can be orientable or not. Since the singular fibers
1 are isolated and project to isolated points in the base surface, they have no effect on
More generally, fiber bundles over S with fiber a closed orientable surface.
orientability of the base.
More generally still, irreducible M with H1 (M) infinite.
The data specifying a Seifert fibering consists of:
There are various infinite families of closed irreducible 3 manifolds with infinite π1 The topological type of the base surface.
that have been shown not to be Haken manifolds, but they all seem to be manifolds The number of multiple fibers and the fractions p/q specifying the local fibering
that are in some sense “small”. Perhaps all manifolds that are sufficiently large are near each multiple fiber. In fact the value of p/q modulo 1 is enough to specify
Haken. (Indeed, the original name for Haken manifolds was “sufficiently large”, be- the local fibering.
fore Haken became famous for the Four-Color Theorem.) It is still an open question In case the base surface is closed, an “Euler number” specifying how twisted the
whether closed irreducible 3 manifolds with infinite π1 always have a finite-sheeted fibering is. As with ordinary circle bundles, this can be defined as the obstruction
covering space that is Haken. to a section of the bundle.
Even if the Borel conjecture were proved for Type III prime 3 manifolds, this Thus one obtains an explicit classification of all the different Seifert fiberings. It turns
would still be far from an explicit classification. One would want to know exactly out that in most cases the Seifert fibering of a Seifert manifold is unique up to diffeo-
which groups occur as fundamental groups of these manifolds, and one would want morphism, and in fact up to isotopy. The exceptions can be listed quite explicitly, so
to have an efficient way of distinguishing one such group from another. But perhaps one has a very concrete classification of all Seifert manifolds up to diffeomorphism.
there are just too many different manifolds to make such an explicit classification This includes also the information of which ones admit orientation-reversing diffeo-
feasible. morphisms.
 It is a happy accident that all the spherical 3 manifolds are Seifert manifolds. uniqueness it is essential to choose the collection of Tj ’s to be minimal. This subtlety
The base surface is S 2 in each case, and there are at most three multiple fibers. The in the uniqueness statement is probably the reason why this theorem was discovered
Type II manifold S 1 × S 2 is of course also a Seifert manifold, via the product fibering. only in the 1970s, since in hindsight one can see that it could have been proved in the
Most Seifert manifolds are of Type III. The only Seifert manifold that is not prime is 1930s when Seifert manifolds were first studied.
3 3
RP ] RP , the sum of two copies of real projective 3 space. (This manifold happens If the collection of tori Tj is nonempty, the manifolds Mi will have torus boundary
to have S 1 × S 2 as a 2 sheeted covering space, the only instance of a prime manifold components. Unlike in the prime decomposition where we split along spheres and
covering a nonprime manifold.) there was a canonical way to fill in the newly-created boundary spheres with balls,
Torus Decomposition when we split along tori there is no canonical way to fill in the resulting boundary
tori and thereby stay within the realm of closed manifolds if the original M was a
After the prime decomposition, it suffices to classify irreducible 3 manifolds.
closed manifold. The natural thing to try is to fill in the new boundary tori with
These cannot be simplified by splitting along spheres, but one may ask whether they
solid tori S 1 × D 2 , but there are infinitely many essentially different ways to do this
can be simplified by splitting along the next-simplest surfaces, embedded tori. Split-
since the glueing is achieved by a diffeomorphism of a torus and the group of isotopy
ting along a torus that lies inside a ball or bounds a solid torus is not likely to produce
3 classes of diffeomorphisms of a torus is GL2 (Z) , and only a relatively small number
a simpler manifold, as one can see already in the case of knotted tori in S . Such tori
of these diffeomorphisms extend over a solid torus. (We are essentially talking about
are obviously compressible, however, so it seems more promising to try splitting along
Dehn surgery here, which will be discussed in more detail later.) So it is best just to
incompressible tori. In fact, in an irreducible manifold M any embedded torus that
leave the Mi as manifolds with boundary tori. This means that even if one is primarily
does not lie in a ball in M and does not bound a solid torus in M must be incompress-
interested in closed manifolds, one is really forced to broaden one’s domain to include
ible. This is an easy consequence of the Loop Theorem, which says that for a properly
manifolds with boundary tori.
embedded compact orientable surface S ⊂ M , if the map π1 (S)→π1 (M) is not in-
jective then there is an embedded disk D ⊂ M with D ∩ S = ∂D such that the circle The manifold M determines the Mi ’s uniquely, but there are many choices for

∂D represents a nontrivial element of the kernel of the map π1 (S)→π1 (M) . (Often how to glue the Mi ’s together to reconstruct M . In the prime decomposition one had

this theorem is stated in the equivalent form that S is equal to ∂M or a subsurface only to worry about orientations to specify how to glue the pieces together, but here

of ∂M .) the glueings are by diffeomorphisms of tori, so again there is a wide choice of ele-

Splitting along incompressible tori turns out to work very nicely: ments of GL2 (Z) , or SL2 (Z) if orientations are specified, determining how to glue the
pieces Mi together. For a complete classification of the various manifolds that can
Torus Decomposition Theorem (Jaco-Shalen, Johannson). If M is an irre- be obtained by glueing together a fixed collection of Mi ’s one needs to know which
ducible compact orientable manifold, then there is a collection of disjoint incompress- collections of diffeomorphisms of the boundary tori of each Mi extend to diffeomor-
ible tori T1 , ··· , Tn in M such that splitting M along the union of these tori produces phisms of Mi . In case all the Mi ’s happen to be Seifert manifolds this can be figured
manifolds Mi which are either Seifert-fibered or atoroidal — every incompressible out, so the classification of the resulting manifolds M is known explicitly. This was
torus in Mi is isotopic to a torus component of ∂Mi . Furthermore, a minimal such first worked out by Waldhausen, who called these manifolds graph manifolds, in ref-
collection of tori Tj is unique up to isotopy in M . erence to the graph that describes the combinatorial pattern for glueing together the
The collection of tori Tj could be empty. This will happen if M itself is Seifert- Mi ’s, the graph having a vertex for each Mi and an edge for each Tj .
fibered or atoroidal. It is possible to characterize the tori Tj intrinsically as the in- After the Torus Decomposition Theorem, the remaining big problem is to classify
compressible tori T ⊂ M that are isolated in the sense that every other incompressible “general” prime 3 manifolds, those which are atoroidal and not Seifert-fibered. Exam-
torus in M can be isotoped to be disjoint from T . Notice the strength of the unique- ples of such manifolds can be found among the manifolds M which fiber over S 1 with
ness statement: up to isotopy, not just up to diffeomorphism of M . This differs from fiber a closed orientable surface F of genus at least 2 . Such a bundle is determined by
the prime decomposition where the spheres giving a splitting into primes are not at a diffeomorphism ϕ : F →F . If ϕ is isotopic to a diffeomorphism of finite order, then
all unique, even up to diffeomorphism of the manifold. Only the prime factors are M has an evident Seifert fibering, with circle fibers transverse to the surface fibers F .
unique. Another special case is if ϕ leaves invariant some finite collection of disjoint nontriv-
The Mi ’s that are Seifert-fibered could be further split along incompressible tori ial circles in F , since such a collection gives rises to a set of incompressible tori in M
into atoroidal pieces, but the resulting atoroidal pieces are usually not unique. To get transverse to the fibers. In all remaining cases it turns out that M is atoroidal and
not a Seifert manifold. These “general” ϕ ’s are the subject of Thurston’s theory of distance away from a really explicit classification of 3 manifolds since the number of
pseudo-Anosov surface diffeomorphisms. hyperbolic manifolds is so large.
There are many more examples. For example, the complement of an open tubular Dehn Surgery
neighborhood of a knot in S 3 is an irreducible manifold with torus boundary, and this
It is tempting to try to get a more complete picture of the set of all closed ori-
manifold is atoroidal and not Seifert-fibered for perhaps 99 percent of the first million
entable 3 manifolds by putting some sort of global structure on this set. The number
knots.
of compact 3 manifolds is countable since there are just countably many finite sim-
Geometric examples, analogous to spherical manifolds, are the hyperbolic man- plicial complexes, so perhaps there is some kind of “variety” whose rational points
ifolds, the quotients of hyperbolic 3 space H3 obtained by factoring out the action correspond bijectively, in some meaningful way, with all the diffeomorphism classes
of a group Γ of isometries of H3 , where the action is free and the quotient space of closed orientable 3 manifolds.
H3 /Γ is compact, hence a closed manifold. More generally, one can allow the quotient One possible way to implement this vague idea is to use Dehn surgery, which is
H3 /Γ to be noncompact but have finite volume, since in this case the quotient is the defined in the following way. Fix a closed orientable manifold M and choose a link
interior of a compact manifold whose boundary consists of tori, such as would arise L in M consisting of n disjoint embedded circles L1 , ··· , Ln . These have disjoint
in a nontrivial torus decomposition of a closed manifold. tubular neighborhoods N(Li ) that are solid tori S 1 × D 2 . Remove the interiors of these
Hyperbolic manifolds are always irreducible and atoroidal, and no hyperbolic solid tori from M and then glue the solid tori back in by means of diffeomorphisms
manifold can be Seifert-fibered. Amazingly, Thurston conjectured that the converse ∂N(Li )→∂(M − int(N(Li ))) . One can think of glueing in a solid torus S 1 × D 2 as
is also true: first glueing in a meridian disk {x}× D 2 and then glueing in a ball. All that matters is
Hyperbolization Conjecture: Every irreducible atoroidal closed 3 manifold that is how the meridian disk is glued in, since glueing in a ball is canonical. To specify how
not Seifert-fibered is hyperbolic. Furthermore, the interior of every compact irreducible the disk is glued in it suffices to specify the curve its boundary attaches to, and in a
atoroidal nonSeifert-fibered 3 manifold whose boundary consists of tori is hyperbolic. torus this curve will be, up to isotopy, p times a meridian plus q times a longitude
for some pair (p, q) of relatively prime integers, which can conveniently be regarded
Thurston proved this in many cases, for example for nonclosed manifolds, sur-
as a fraction, or slope, p/q , possibly 1/0 . Doing this for each solid torus N(Li ) in
face bundles, and more generally all Haken manifolds. In addition to these general
turn, we obtain a manifold ML (p1 /q1 , ··· , pn /qn ) which is said to be obtained from
theorems many concrete examples have been worked out using computer programs
M by Dehn surgery on L . This gives a set of manifolds parametrized by the rational
that have been developed for this purpose. The evidence in favor of the conjecture
n torus (Q ∪ {1/0})n . To give this set of manifolds a name, let us call it a “Dehn
seems quite strong.
variety” V (M, L) . (There is an ambiguity in the choice of longitudes, but this just
Hyperbolic manifolds must have infinite fundamental group since they have fi- corresponds to a simple change of coordinates in V (M, L) .)
nite volume but their universal cover H3 has infinite volume. This means that the As an example, if we take M to be a product bundle F × S 1 with base F a compact
hyperbolization conjecture implies the Poincaré conjecture, as there are no coun- orientable surface and we choose L to be a collection of n fibers {x}× S 1 , then the
terexamples to the Poincaré conjecture among Seifert manifolds, and manifolds that Seifert manifolds obtained by replacing these fibers by possibly multiple fibers are
contain incompressible tori have infinite fundamental group. One could make a more almost exactly the same as the manifolds obtained by Dehn surgery on L . The only
restricted form of the hyperbolization conjecture that would not imply the Poincaré exceptions are the Dehn surgeries in which a meridian disk is glued in along a curve
conjecture by adding the hypothesis of infinite fundamental group. isotopic to a circle fiber of F × S 1 . These exceptional surgeries produce connected
If two hyperbolic 3 manifolds have isomorphic fundamental groups, then they sums of Seifert manifolds with S 1 × S 2 . In a similar fashion, the Seifert manifolds
are in fact isometric, according to the Mostow Rigidity Theorem. In particular, this over a nonorientable base surface can be obtained by Dehn surgery on fibers of a
says that hyperbolic structures are unique, if they exist. It also means that the hy- twisted circle bundle.
perbolization conjecture implies the Borel conjecture in dimension 3 , that K(π , 1) The first theorem about Dehn surgery is that if one fixes M and varies the link L
3 manifolds are determined up to homeomorphism by their fundamental groups. and the slopes pi /qi , then one obtains all closed orientable 3 manifolds. In particular
Much is known about hyperbolic manifolds, so the hyperbolization conjecture, this happens when M = S 3 and one is doing Dehn surgery on links in the usual
if true, would give a great deal of information about individual 3 manifolds. How- sense. No single V (M, L) can contain all 3 manifolds since the homology groups of
ever, even if the hyperbolization conjecture were proved, we would still be a some manifolds in V (M, L) have bounded ranks. Any two V (M, L) ’s are contained in a third,
so the union of all Dehn varieties can be thought of as a sort of an infinite-dimensional
Dehn variety which consists of all closed orientable 3 manifolds.
The complicating factor with Dehn varieties is of course that the surjections
(Q ∪ {1/0})n →V (M, L) , (p1 /q1 , ··· , pn /qn ) , ML (p1 /q1 , ··· , pn /qn ) , need not be
injective, so Dehn varieties are not just rational tori but some kind of singular objects.
A special case is Dehn curves, the varieties V (M, L) when L has just one compo-
nent. Two manifolds in the same Dehn curve have a lot in common: the complement
of a knot, in fact. Here are some interesting theorems and conjectures:
No Dehn curve passes through S 3 more than once. (Theorem of Gordon-Luecke)
More generally, no Dehn curve through S 3 passes through a homotopy sphere at
another point of the curve. (The Property P conjecture)
There is a unique Dehn curve though S 3 and S 1 × S 2 , the Dehn curve for the trivial
knot in S 3 . (Property R, proved by Gabai. The other points on this curve are lens
spaces.)
Every other Dehn curve through S 3 passes through manifolds with finite π1 at
most three times. (Theorem of Culler-Gordon-Luecke-Shalen. There are examples
showing the number “three” is best possible.)

Not much seems to have been done yet with Dehn varieties of higher dimension.

References for further reading

As a primary source for the more basic topological aspects of the theory, including
prime decomposition, torus decomposition, and Seifert manifolds, we suggest

A Hatcher. Basic Topology of 3-Manifolds. Unpublished notes available online at

J Hempel. 3-Manifolds. Annals of Math Studies 86. Princeton University Press,

A book that contains an exposition of the classification of spherical 3 manifolds is

W Thurston. Three-Dimensional Geometry and Topology. Princeton University