In this paper, we compute the group of homotopy equivalences of closed, smooth, simply connected 6-manifolds whose integral homology groups are isomorphic to whose of the complex projective space CP3, building on computations of their mapping class groups.
Introduction
For a topological space X, the group of homotopy equivalences haut(X) is an object of significant interest in homotopy theory, and has been widely studied in the literature; see, for example, the survey article [2]. For a compact manifold M, the group haut(M) also plays an important role in manifold topology, particularly in the classification of manifolds and the study of group actions, see [3].
For simply connected 4- and 5-manifolds, the group haut(M) has been computed by Milnor [13] and Baues-Buth [6], respectively. In dimension 6, however, the situation is less well understood. The group haut(S2×S4) was computed by Sawashita [17] as part of a broader study of the group of homotopy equivalences of products of spheres. Partial information on the group of homotopy equivalences of S2-bundles over S4 has been obtained by Yamaguchi [21].
As a generalization of the results mentioned above, in this paper we compute the group haut(M) for closed, smooth, simply connected 6-manifolds M with H∗(M)≅H∗(CP3) (referred to as simply connected homology CP3 for short). For this class of manifolds, classifications up to diffeomorphisms and up to homotopy equivalences have long been known. We briefly recall these results below.
By the theory of Wall ([19 Theorem 3]) and Jupp ([9 Theorem 1]), the diffeomorphism type of an oriented, simply connected 6-manifold M with torsion-free homology is determined by the third Betti number b3(M), the cubic form
H2(M)×H2(M)×H2(M)→Z,(x,y,z)↦⟨x∪y∪z,[M]⟩
the second Stiefel-Whitney class w2(M) and the first Pontrjagin class p1(M).
When M is a homology CP3, the cubic form and p1(M) can be described in terms of numeric invariants as follows. Let x∈H2(M), y∈H4(M) be generators. Then there exists an integer d such x2=dy. If M is nonspin, then d is even. By choosing the sign of y, we may assume d≥0. There exists an integer l such that the first Pontrjagin class is given by
p1(M)=4(d+6l)y,if M is spin;p1(M)=(d+48l)y,if M is non-spin.
When d=0, we choose the sign of y so that l≥0.
The homotopy classification of this class of 6-manifolds has been established by several authors (see [4 Proposition 1.9] and the discussion therein). We summarize these results in the theorem below.
Theorem 1.1. Let M be a simply connected homology CP3. Then M admits an orientation-reversing diffeomorphism.
If M is spin, there is a one-to-one correspondence between the set of diffeomorphism classes of such manifolds and the set
{(d,l)∈Z×Z∣d≥0,and l≥0 when d=0}.
If d is odd, the homotopy type of M is determined by d; if d is even, the homotopy type of M is determined by d together with d/2+l(mod2).
If M is non-spin, there is a one-to-one correspondence between the set of diffeomorphism classes and the set
{(d,l)∈2Z×Z∣d≥0,and l≥0 when d=0}.
In this case, the homotopy type of M is determined by d.
We now give some example. In the spin case, the manifold with d=l=0 is S2×S4, while the manifold with d=0 and l=1 is CP3; in the non-spin case, the manifold with d=l=0 is the nontrivial S4-bundle over S2, denoted by S4×S2.
Another class of geometrically interesting examples is given by the total spaces Xn of S2-bundle over S4 with characteristic map n∈π3SO(3)≅Z. These manifolds are spin and satisfy d=n and l=0. As a consequence of the homotopy classification, if M is spin with d odd, or with both d and l even, then M is homotopy equivalent to the sphere bundle Xd.
A standard method in homotopy theory for studying the group of homotopy equivalences of an n-dimensional simply connected Poincaré complex X is provided by the exact sequence of Barcus and Barratt [5 Theorem 6.1] (see also [3, Proposition (1.3)])
Im{πn(X(n−1))→πn(X)}d∗haut(X)j∗haut(X(n−1)).
where X(n−1) denotes the (n−1)-skeleton of X. The homomorphism
d∗:Im{πn(X(n−1))→πn(X)}→haut(X)
is induced by the composition of the coaction μ:X→X∨Sn with a map Sn→X representing an element of Im{πn(X(n−1))→πn(X)}. There are clear descriptions of Imj∗ and the action of Imj∗ on Imd∗. The difficulty lies in the determination of the kernel of d∗, which involves the computation of a certain homomorphism
π1(Map∗(X(n−1),X),i)→πn(X).
This method has been applied to the class of 6-manifolds under consideration by several authors. For X=S2×S4, Sawashita showed that this sequence extends to a split short exact sequence ([17 Theorem 4.3])
0→π6(S2×S4)→haut(S2×S4)→(Z/2)3→1.
For X=Xd, Yamaguchi obtained an exact sequence ([21 Theorem 1.1])
π6(Xd)→haut(Xd)→Gd→1(1)
with Gd≅Z/2 when d is odd, and Gd≅Z/2×Z/2 when d is even. He also computed the group π6(Xd):
π6(Xd)≅⎩⎨⎧0Z/3Z/2×Z/(12,d/2)if (d,6)=1,if (d,6)=3,if d is even.(2)
However, since the kernel of the homomorphism π6(Xd)→haut(Xd) was not determined (except for the cases where π6(Xd)=0), the problem of computing haut(Xd) remained unsolved.
In this paper, we study the group of homotopy equivalences from a different perspective. In manifold topology, the mapping class group MCG(M) of a smooth manifold M plays an analogous role to that of the group of homotopy equivalences in homotopy theory, and is itself an object of considerable interest. The group haut(M) is related to group MCG(M) via the natural forgetful homomorphism ψ:MCG(M)→haut(M).
As an example, for (n−1)-connected 2n-dimensional almost parallelizable manifolds, both MCG(M) and haut(M) have been computed by various authors, and a detailed comparison between them is given in [3 §10].
In this paper, we analyze the forgetful homomorphism ψ for simply connected homology CP3. We determine the kernel and cokernel of ψ:MCG(M)→haut(M), and then use our knowledge of MCG(M), obtained in [12], to compute the group haut(M). The manifold-theoretic perspective adopted in this paper also allows us to describe the generators of haut(M) geometrically.
Before stating the main theorems, we present the construction of some self-homotopy equivalences of M, which serve as generators of haut(M). Let μ:M→M∨S6 be the pinch map which collapses the boundary of an embedded 6-disk to a point. Let ξ:S6→M be the composite of maps S6→S2 and S2→M representing a generator of π6(S2)≅Z/12 and π2(M)≅Z, respectively. When d is even, it is shown in [21 Proposition 5.3] that π6(Xd) contains a direct summand which is mapped isomorphically to π6(S4)≅Z/2 under the bundle projection Xd→S4. When M is spin with both d and ℓ even, by the homotopy classification, M is homotopy equivalent to Xd. Let η:S6→M represent the corresponding non-trivial element in π6(M). Define homotopy equivalences fξ and fη
which are diffeomorphisms of M supported in a tubular neighborhood of an embedded S3 in M.
Proposition 1.2.fξ is homotopic to tϵ; fη is not homotopic to a diffeomorphism.
For integers m and n, we write (m,n) for their greatest common divisor.
Theorem 1.3. Let M be a simply connected spin homology CP3 with d=0. If d is odd, then
haut(M)≅Z/(3,d)⋊Z/2,
the generator of Z/(3,d) is represented by fξ, the generator of Z/2 is represented by an orientation-reversing self-diffeomorphism of M, and the action of Z/2 on Z/(3,d) is by the multiplication of −1.
If d is even, then there is a short exact sequence
1⟶G⋊Z/2⟶haut(M)⟶Z/2⟶1,(3)
where
G≅{Z/2×Z/(6,d/2)Z/2×Z/(12,d/2)if d≡0(mod8) and l≡0(mod2),otherwise.
In the group G, the generator of Z/2 is represented by tρ, the generator of the other direct summand is represented by fξ. The generator of factor Z/2 in the semi-direct product is represented by an orientation-reversing smooth conjugation on M. The action of Z/2 on G is by the multiplication by −1. If l is even, then the exact sequence (3) splits, with a splitting given by fη.
Remark 1.4. Together with Sawashita’s computation of haut(S2×S4), the above theorem provides a description of haut(M) for all simply connected spin homology CP3’s, except for the unique homotopy type with d=0 and l odd (see Theorem 1.1). In the remaining case, we show that haut(M) contains a subgroup of index 2 or 4 isomorphic to (Z/2×Z/6)⋊Z/2; see Theorem 4.1 (5).
As a corollary, the exact sequence (1) extends to a split short exact sequence.
Corollary 1.5. Let Xd be the total space of the S2-bundle over S4 with characteristic class d∈π3SO(3). Assume that d=0. Then there is a split short exact sequence
0→π6(Xd)→haut(Xd)→Gd→1.
From this perspective, the generators of the corresponding groups are described as follows: when d is odd, a generator of π6(Xd) is represented by the Dehn twist tϵ, and a splitting of Gd≅Z/2 is given by the fiberwise antipodal map fa. When d is even, a generator of the direct summand Z/(12,d/2)<π6(Xd) is represented by the Dehn twist tϵ, while a generator of the direct summand Z/2<π6(Xd) is given by fη. A splitting of Gd≅Z/2×Z/2 is provided by fa and the Dehn twist tρ. The action of Gd on π6(Xd) is given by
where fa∗ and tρ∗ stand for the induced isomorphism on π6(Xd).
Theorem 1.6. Let M be a simply connected non-spin homology CP3. If d=0, then there are short exact sequences
1→H→haut(M)→Z/2→1
1→G→H→Z/2→1.
The group G is isomorphic to Z/4×Z/(6,d/2), the action of Z/2 on G is given by multiplication by −1.
If d=0, then there is a short exact sequence
0→(Z/4×Z/6)⋊(Z/2×Z/2)→haut(M)→Z/2→1.
In the semi-direct product, the actions of (1,0) and (0,1) on Z/4×Z/6 are both given by multiplication by −1.
The methods used in this paper are primarily drawn from manifold topology. In §2, we first recall the results on the Torelli group T(M) from [12], and then extend them to the full mapping class group MCG(M). In § 3, we study the forgetful homomorphism ψ0:T(M)→haut0(M) using the Browder–Novikov–Sullivan–Wall surgery exact sequence. Finally, in §4, we state and prove Theorem 4.1 and 4.2.
Acknowledgement. The second author is partially supported by NSFC 12471069.
The full mapping class group MCG(M)
The Torelli group T(M)
The Torelli group T(M). The Torelli group T(M) is an important subgroup of the mapping class group MCG(M). It consists of isotopy classes of diffeomorphisms of M that induce the identity on H∗(M). In [12], the Torelli groups of a more general class of 6-manifolds were computed, namely, simply connected 6-manifolds whose homology groups are isomorphic to H∗(CP3#r(S3×S3)). In this subsection, we review this computation in the case of simply connected homology CP3’s. For further details, see [12 §6–8, §10].
Let M be a simply connected homology CP3, D⊂M be a fixed embedded 6-disk. Let T(M,D) be the relative Torelli group, i.e., the group of isotopy classes of diffeomorphisms f:M→M such that f∣D=id and f∗=id on H∗(M). It is shown in [12 Theorem 9.4, §10.3] that the homomorphism T(M,D)→T(M) is an isomorphism. Therefore we will identify these two groups in the sequel. There is an invariant S:T(M,D)→Z3/L(M) constructed from the mapping torus Mf=M×[0,1]/(x,0)∼(f(x),1), which is shown to be an isomorphism.
The definitions of the invariant are slightly different for spin and nonspin manifolds. We start with spin manifolds. Let B=CP∞×K(Z,4)×BSpin, p:B→BSpin→BSO be the canonical projection. Then the normal Gauss map ν:Mf→BSO has a lifting g:Mf→B constructed as follows. Fix an orientation of M and a generator x∈H2(M)=H2(Mf), take a map Mf→CP∞ representing x. For d=0, let y=x2/d be the generator of H4(M); for d=0, l=0, let y=p1(M)/24l∈H4(M); for d=l=0, take y∈H4(M) an arbitrary generator. Since f induces the identity on H∗(M), the Wang sequence implies that H4(Mf)→H4(M) is an isomorphism. Let y′∈H4(Mf) be the preimage of y, then x2=dy′ and p1(Mf)=4(d+6l)y′. Take a map Mf→K(Z,4) representing y′. Finally choose the spin structure on the stable normal bundle of Mf, which restricts to the trivial spin structure on the preferred embedding D×S1⊂Mf. This induces a map Mf→BSpin. Putting all these maps together we get g:Mf→B which is a lift of the normal Gauss map Mf→BSO. Note that all maps involved are unique up to homotopy in the appropriate sense.
The pair (Mf,g) represents a bordism class in the B-bordism group Ω7(B,p)=Ω7Spin(CP∞×K(Z,4)). By a standard computation with the Atiyah-Hirzebruch spectral sequence one shows that this group is zero ([11 Theorem 6]). Therefore there exists an 8-dimensional normal B-bordism G:W→B such that ∂(W,G)=(Mf,g). After taking connected sum with quaternionic projective planes HP2’s (or HP2’s with opposite orientation) we may assume that the signature of W is zero.
Let x∈H2(B) be the pull-back of the canonical generator of H2(CP∞), y∈H4(B) be the pull-back of the canonical generator of H4(K(Z,4)) and 2p1∈H4(B) be the pull-back of the universal spin Pontrjagin class in H4(BSpin). Let
α=G∗(x2−dy),β=G∗(2p1+2(d+6l)y)∈H4(W).
Then by construction, the restriction of α and β on ∂W=Mf are zero, so we have preimages α,β∈H4(W,∂W) of α and β respectively. Now define characteristic numbers
By comparing T(M,D) and the mapping class group of (S2×D4,∂), and taking consideration of the choices of the B-bordism W one has a homomorphism
S=[(s1/8,s2/2,s3)]:T(M,D)→Z3/L(M),
where L(M) is the lattice generated by the characteristic numbers of elements in the subgroup of the 8-dimensional B-bordism group Ω8(B)=Ω8spin(CP∞×K(Z,4)), consisting of manifolds with zero signature. Using the Atiyah-Hirzebruch spectral sequence one can show that this group is isomorphic to Z6 with explicit generators e1,⋯,e6[11 Theorem 10]. The characteristic numbers of these generators are given in the following table, where k=(d+6l),
The lattice L(M)⊂Z3 is generated by column vectors. In [12 Theorem 6.1] it is shown that S is an isomorphism, and the groups T(M) are given there. In particular, there is a surjective homomorphism Z/28×Z/12×Z/2→T(M).
Remark 2.1. In the definition of the S-invariant, we fix an orientation of M and a generator x∈H2(M). From the definition it is ready to see that if we reverse the orientation then the invariant changes by a sign, while if we use the generator −x then the invariant remains unchanged.
A diffeomorphism f:M→M is disk supported if f=id outside an embedded closed 6-dimensional disk D⊂M. Isotopy classes of disk supported diffeomorphisms form a subgroup K⊂T(M). The quotient group T(M)/K will be relevant in the computation of haut(M) (see Proposition 3.2). Let MCG(D6,∂) be the mapping classes group of the 6-disk relative to boundary. Fix an embedding D6⊂M, then there is a homomorphism MCG(D6,∂)→T(M), extending a diffeomorphism of D6 by the identity outside the embedded D6 to a diffeomorphism of M. By the disk theorem, the image of this isomorphism is exactly the subgroup K. The group MCG(D6,∂) is isomorphic to the Kervaire-Milnor group Θ7 of 7-dimensional homotopy spheres, which is isomorphic to Z/28, with a preferred generator 1 represented by the boundary of the E8-plumbing. The S-invariant of the image of 1∈Z/28 in T(M) equals [(1,0,0)]. Therefore from Table 2.1 we get the quotient group T(M)/K.
e1
e2
e3
e4
e5
e6
s1/8
−28
0
0
(k2−k)/2
k2
0
s2/2
0
12
0
(d2−d(2k−1))/2
d2−2dk
k−d
s3
0
0
2
d(2k−1)
4dk
−2k
Table 2.1.
Proposition 2.2. Let M be a simply connected spin homology CP3.
Proof. From Table 2.1 we see that the finite abelian group T(M)/K has only 2- and 3-torsions, with exponents at most 2 and 1, respectively. To determined the exponents we only need to take tensor product with Z/4 or Z/3, respectively. Then an elementary calculation shows this proposition. □
When M is nonspin, we still take B=CP∞×K(Z,4)×BSpin, the fibration p is
p:B=CP∞×K(Z,4)×BSpin→BSO×BSO⟶⊕BSO,
where CP∞→BSO is the classifying map of the canonical complex line bundle η over CP∞ and K(Z,4)×BSpin→BSpin→BSO is the composition of natural projections. Then similar to the spin case, for a diffeomorphism f, we have a lifting g:Mf→B of the normal Gauss map of Mf. The corresponding bordism group Ω∗(B,p) is the twisted spin cobordism group Ω∗spin(CP∞×K(Z,4);η). This group vanishes ([11 Theorem 6]). Let xˉ∈H2(CP∞), yˉ∈H4(K(Z,4)) and 2p1∈H4(BSpin) be as in the spin case. Let G:W8→B be a coboundary of g:Mf→B with signature zero, let α=G∗(xˉ2−dyˉ), β=G∗(2p1+(d+24l)yˉ)∈H4(W), then they have preimages αˉ,βˉ∈H4(W,∂W) respectively. Define characteristic numbers
Let λ=σ1+3σ2+σ3, μ=(6σ1+10σ2+3σ3)/4, then the S-invariant
S=[(μ,λ)]:T(M,D)→Z2/L(M)
is an isomorphism, where L(M) is the lattice generated by the column vectors of the following table.
μ
6⋅28
0
−δ2−48δl+2δ+432l2−36l
−2δ2−96δl−864l2
3(δ−4l)
λ
0
−4
−4δ2−48δl+2δ+576l2−48l
−2(4δ2+48δl−576l2)
6(δ−4l)
The image of the preferred generator of MCG(D6,∂)≅Z/28 in T(M) has S-invariant [(−6,−12)]. From this one may compute the quotient group T(M)/K.
Proposition 2.3. Let M be a simply connected nonspin homology CP3, T(M)/K is given in the following table.
d
12∣d
3∣d,4∤d
3∤d,4∣d
3∤d,4∤d
T(M)/K
Z/6×Z/4
Z/3×Z/4
Z/2×Z/4
Z/4
Dehn twists
Let i:S2→M represent a generator of π2(M)≅Z, φ:S3→S2 be the Hopf fiber map. Then i∘φ:S3→M represents a generator of π3(M)≅Z. Fix a thickened embedded 3-sphere S3×D3⊂M, such that S3×{0} represents this generator of π3(M). For any α∈π3SO(4), choose a representing smooth map f:D3→SO(4), which maps an open neighborhood of ∂D3 to I. The Dehn twist tα:M→M is defined as the diffeomorphism which maps the complement of S3×D3 by the identity and (x,y)∈S3×D3 to (f(y)x,y).
From the exact sequence of homotopy groups of fiber bundle
SO(3)→SO(4)→S3
it can be shown that the group π3(SO(4)) is isomorphic to Z⊕Z, with generators represented by ϵ:S3→SO(3)→SO(4), which is the composite of the projection S3→SO(3) with the inclusion, and ρ:S3→SO(4), which is the inclusion of the unit quaternions S3 to SO(4). The corresponding Dehn twists are denoted by tϵ and tρ, respectively.
Lemma 2.4. (1) The group T(M)/K is generated by tϵ and tρ; the order of tρ is at most 2.
(2) Let J:π3SO(3)→π6(S3) be the J-homomorphism, ξ=i∘φ∘J(ϵ):S6→M. Then the homotopy equivalence fξ=(id∨ξ)∘μ:M→M∨S6→M is homotopic to tϵ.
Proof. The statements in (1) is shown in [12 §7]. To show (2), consider the natural inclusion map SO(3)→G+(S3), where G+(3) is the space of base point and orientation preserving homotopy equivalences of S3. The induced homomorphism is the J-homomorphism
J:π3SO(3)→π3G+(3)≅π6(S3).
Therefore regarded as a diffeomorphism of S3×D3, the Dehn twist tϵ is homotopic to
(id∨J(ϵ))∘μ:S3×D3→(S3×D3)∨S6→S3×D3,
where the map on S6 is the composition of J(ϵ):S6→S3×{0} and the inclusion S3×{0}→S3×D3. As a diffeomorphism of M, the Dehn twist tϵ is homotopic to (id∨ξ)∘μ.
The full mapping class group MCG(M)
In this subsection we compute the full mapping class group MCG(M), in which orientation reversing diffeomorphisms are contained as well.
Theorem 2.5. Let M be a simply connected homology CP3 which is not diffeomorphic to S2×S4 or S4×S2. Then there is a short exact sequence
1→T(M)→MCG(M)→Z/2→1,(4)
where the action of Z/2 on T(M) is by −1. This short exact sequence splits when M is spin, hence MCG(M)=T(M)⋊Z/2.
Proof. Denote by MCG+(M) the oriented mapping class group of M. Then there is an exact sequence
1→MCG+(M)→MCG(M)→Z/2.
It is assumed that M is not diffeomorphic to S2×S4 or S4×S2. Therefore either d=0 or l=0. If d=0, then clearly T(M)=MCG+(M). If d=0 but l=0, then p1(M)=0. In this case since a diffeomorphism preserves p1(M), it induces the identity on H4(M), therefore T(M)=MCG+(M). Moreover, by the classification theorem (Theorem 1.1), there exists an orientation-reversing self-diffeomorphism of M, therefore we have a short exact sequence
1→T(M)→MCG(M)→Z/2→1.
Since T(M) is an abelian group, there is an action of Z/2 on T(M), given by conjugation: let f be an orientation-reversing diffeomorphism, [g]∈T(M), then the action of 1∈Z/2 on [g] is [f−1∘g∘f]∈T(M). Observe that there is an orientation-preserving diffeomorphism between the mapping tori
Mf−1∘g∘f≅Mg−1,[(x,t)]↦[(f(x),1−t)]
This implies that f−1∘g∘f and g−1 have the same S-invariants, therefore [f−1∘g∘f]=[g−1]∈T(M). This shows that the action of Z/2 on T(M) is by −1.
Next we show that when M is spin the exact sequence is split. If l is even, it’s known that there exists a free involution on M (see [14 Theorem 1.2] and the discussion there). By the Lefschetz fixed-point theorem, this free involution is orientation-reversing. This gives a splitting. If d is even, then by [15 Corollary 1.5], there exists a conjugation on M, which is a special type of involution. Such a conjugation has a 3-dimensional fixed-point set, hence must be orientation-reversing. This gives a splitting. If both d and l are odd, then from the tables in [12 Theorem 6.1] one sees that the Torelli group T(M) has odd order. Therefore H2(Z/2;T(M))=0. This implies that the sequence (4) splits. □
Theorem 2.6. Let M=S2×S4 or S4×S2, then
MCG(M)=T(M)⋊(Z/2×Z/2)
where the actions of the standard generators of Z/2×Z/2 on T(M) are by −1.
Proof. If M=S2×S4, then clearly we have a short exact sequence
1⟶T(M)⟶MCG(M)⟶Z/2×Z/2⟶1
where the homomorphism MCG(M)→Z/2×Z/2 is given by the actions on H2(M) and H4(M). Apparently the elements (1,0) and (0,1)∈Z/2×Z/2 have liftings (−id,id):S2×S4→S2×S4 and (id,−id):S2×S4→S2×S4 respectively. Since these are orientation reversing diffeomorphisms, as discussed in the proof of last theorem, their actions on T(M) are by −1.
When M=S4×S2, the non-trivial linear S4-bundle over S2, we construct two smooth involutions as follows. The first involution f1 is the fiberwise antipodal map, then f1 induces the identity on H2 and −id on H4. To construct the second involution f2, we use the decomposition S4×S2=(S4×D2)1∪α(S4×D2)2, with α a map S1→SO(5) representing the nontrivial element in π1SO(5)≅Z/2. Consider the space
SO(5)Z/2={A∈SO(5)∣At=A}.
Such matrices have eigenvalues ±1. It’s not hard to see that this space is a union of Grassmannians of even dimensional linear subspaces in R5. Especially we have an inclusion G4(R5)≅RP4⊂SO(5)Z/2⊂SO(5), inducing isomorphism on π1. We may choose α to be a loop in SO(5)Z/2. Then we may define an involution f2:S4×S2→S4×S2, by interchanging the two copies S4×D2: (x,y)1↦(x,y)2, (x,y)2↦(x,y)1. Since for any y∈S1, α(y)2=I, this is a well-defined diffeomorphism. Clearly f2 induces −id on H2 and the identity on H4, and f1 and f2 commute. Therefore we have MCG(S4×S2)=T(S4×S2)⋊(Z/2×Z/2), with the actions of (1,0) and (0,1) are by −1. □
The forgetful homomorphism
The full mapping class group MCG(M) and the group of self-equivalences haut(M) are related by the forgetful homomorphism ψ:MCG(M)→haut(M). Let haut0(M) be the subgroup of haut(M) consisting of homotopy equivalences which act trivially on H∗(M), then ψ0:T(M)→haut0(M). In this section we analyze this homomorphism by surgery theory.
Proposition 3.1. Let M be a simply connected homology CP3, ψ0:T(M)→haut0(M) be the forgetful homomorphism. When d is odd, ψ0 is surjective; when d is even, the image of ψ0 is an index 2 subgroup of haut0(M).
Proof. Denote by S(M) the smooth structure set of M. Elements of S(M) are equivalence classes of homotopy equivalences f:X→M from a closed smooth manifold X to M. Two pairs (X1,f1) and (X2,f2) are equivalent if there exists a diffeomorphism ϕ:X1→X2 such that f1 is homotopic to f2∘ϕ. There is a map α:haut0(M)→S(M) sending a homotopy equivalence f:M→M to [M,f]∈S(M). By definition, α(f)=α(g) if and only if f and g are in the same coset of Imψ0. Therefore we have an injective map between sets
α:haut0(M)/Imψ0→S(M)
The image of this map can be analyzed via the surgery exact sequence ([10 Theorem 14.8])
L7(Z)→S(M)ηN(M)θL6(Z).
The surgery obstruction group L7(Z) is zero, hence η is injective. Elements in N(M) are represented by degree 1 normal maps
νN↓⏐NfˉfE↓⏐M
More precisely, the set N(M) can be described as
N(M)=⨁Ω6(M;E)1/Aut(E),
where the direct sum takes over all isomorphism classes of oriented stable vector bundles E over M, for the details see [10 §14.5].
We first analyze the group Aut(E), which is the group of homotopy classes of bundle isomorphisms of E. Fix a k-dimensional representative ξ of E with k≫0, let P→M be the SO(k)-principal bundle associated to ξ. Then the group Aut(ξ) is identified with π0(G(P)), where G(P) is the gauge group of the principal bundle P. It is known that the gauge group G(P) is homotopy equivalent to the based loop space ΩMap(M,BSO(k)) of the space of continuous maps from M to BSO(k), with base point a classifying map φ:M→BSO(k) of ξ (see [1]). Therefore Aut(ξ) is isomorphic to π1(Map(M,BSO(k)),φ). From the fiber sequence
ΩBSO(k)→Map∗(M,BSO(k))→Map(M,BSO(k))evBSO(k)
where Map∗ denotes the space of based continuous maps, and ev is the evaluation map at the base point x0∈M, one has an exact sequence
By the adjunction formula, the space Map∗(S1,Map∗(M,BSO(k))φ) is homeomorphic to the space of continuous maps from S1×M/S1×{x0} to BSO(k) whose restriction on {s0}×M is φ, where s0∈S1 is the base point. Notice that all the groups
vanish. Elementary obstruction theory shows that all such maps are homotopic relative to {s0}×M. Therefore the exact sequence (5) implies Aut(ξ)=0. By passing to the limit one has Aut(E)=0. Hence N(M)=⨁Ω6(M;E)1.
Next we analyze the map η. The element η([M,f])∈N(M) is represented by the degree 1 normal map (fˉ,f)
νM↓⏐Mfˉfg∗(νM)↓⏐M
where g is a homotopy inverse of f. By the minimal cell structure theorem, M is homotopy equivalent to Pd∪e6, where Pd=S2∪de4, η:S3→S2 the Hopf map. The Puppe sequence of the cofibration sequence S5→Pd→M→S6, together with the fact π5(BO)=π6(BO)=0 implies that the restriction map [M,BO]→[Pd,BO] is an isomorphism. A further analysis of the Puppe sequence of the cofibration sequence S3→S2→Pd→S4 shows that a stable vector bundle ξ over Pd is determined by w2(ξ) and p1(ξ). Since g acts trivially on homology, g∗(νM) is isomorphic to νM. Therefore η([M,f]) is in the subset Ω6(M;νM)1 of N(M).
The bordism group Ω6(M;νM) is identified by the Thom-Pontrjagin construction with the homotopy group π6(Th(νM)) of the Thom spectrum Th(νM), whose n-th space is the Thom space of the normal bundle of M⊂R6+n. We compute the homotopy group of the spectrum by the Atiyah-Hirzebruch spectral sequence. The Ep,q2-terms Ep,q2 are identified via the Thom isomorphism with Hp(M;πqS). The differential d2:E6,12→E4,22 is dual to
Sq2+w2(M)∪−:H4(M;Z/2)→H6(M;Z/2);
d2:E6,02→E4,12 is the composition of the mod 2 reduction and the dual of Sq2+w2(M)∪−:H4(M;Z/2)→H6(M;Z/2) ([18 Lemma]). Using the same idea in the proof of Lemma in [18], it can be shown that the differential d2:E4,22→E2,32 is the composition of the dual of Sq2+w2(M)∪−:H2(M;Z/2)→H4(M;Z/2) and the homomorphism j∗:H2(M;Z/2)→H2(M;Z/24) induced by the inclusion of coefficients. (The differential d2 is a stable homology operation. Such operations are classified by
[HZ/2,Σ2HZ/24]≅H2(HZ/2;Z/24)≅Z/2{j∗∘(Sq2)∗},
and the non-triviality of d2 can be detected by CP2, whose stable homotopy groups can be calculated using the cofiber sequence S2→CP2→S4.) A straightforward calculation shows that d2:E6,12→E4,22 and d2:E6,02→E4,12 are trivial; d2:E4,22→E2,32 is nontrivial when d is odd, and trivial when d is even.
Therefore, when d is odd, we get a short exact sequence
0→Ω6fr→Ω6(M;νM)→H6(M)→0.
The surgery obstruction map θ:Ω6fr→L6(Z) is an isomorphism. The normal map (fˉ,f) has vanishing surgery obstruction and the degree of f is 1. Therefore the image of η∘αˉ consists of one element. Since η is injective, this shows that haut0(M)/Imψ0 consists of one element. When d is even, there are short exact sequences
0→G→Ω6(M;νM)→H6(M)→0
0→Ω6fr→G→Z/2→0.
Therefore there are two bordism classes of degree one normal maps with trivial surgery obstruction. Such an element is represented by
νN↓⏐NgˉgνM↓⏐M
with g a homotopy equivalence. By the classification of homology CP3 (Theorem 1.1), N is diffeomorphism to M. Therefore, the image of η∘αˉ consists of two elements. This shows that Imψ0 is a subgroup of index 2 in haut0(M). □
Proposition 3.2.The kernel of ψ0 is the subgroup of T(M) consisting of disk-supported diffeomorphisms.
Proof. Clearly any disk-supported diffeomorphism is in the kernel of ψ0 by the Alexander’s trick. On the other hand, given a diffeomorphism f which is homotopic to the identity, let h:M×[0,1]→M×[0,1] be a homotopy, with h0=id and h1=f, then [M×[0,1],h] can be viewed as an element in the relative structure set S∂(M×[0,1]). Consider the surgery exact sequence
L8(Z)→S∂(M×[0,1])→[ΣM,G/O]→L7(Z).
The homomorphism L8(Z)→S∂(M×[0,1]) is given by taking connected sum of some 7-dimensional homotopy sphere Σ with the base point [M×[0,1],id] in S∂(M×[0,1]). By obstruction theory it’s easy to see that [ΣM,G/O]=0 (since πi(G/O)=0 for i=1,3,5,7). Therefore there is a diffeomorphism g:M×[0,1]→(M×[0,1])#Σ such that g∣M×0=id, g∣M×1=f. Composed with the diffeomorphism ϕ:(M×[0,1])#Σ→M×[0,1], where ϕ∣M×0 is the disk supported diffeomorphism corresponding to Σ and ϕ∣M×1=id, we have a diffeomorphism k:M×[0,1]→M×[0,1] such that k∣M×0 is the disk supported diffeomorphism corresponding to Σ, and k∣M×1=f. This shows that f is pseudo-isotopic to a disk supported diffeomorphism. By the pseudo-isotopy theorem of Cerf [8], f is isotopic to a disk supported diffeomorphism.
For a more general discussion of the homomorphism T(M)→haut0(M) in higher dimensions see [7 Proposition II.9]. □
Determination of haut(M)
Let ψ:MCG(M)→haut(M) be the forgetful homomorphism, haut0(M) be the group of homotopy equivalences which acts trivially on homology.
Theorem 4.1. Let M be a simply connected spin homology CP3.
If d is odd, then ψ is surjective,
haut(M)=haut0(M)×Z/2,haut0(M)≅Z/(3,d),
the action of Z/2 on haut0(M) is by −1.
If d is even and d=0, then there is a short exact sequence
1→Imψ→haut(M)→Z/2→1,
where Imψ is isomorphic to G×Z/2, and
G≅{Z/2×Z/(6,d/2)Z/2×Z/(12,d/2)if d=0(mod8) and l=0(mod2)otherwise
The action of Z/2 on G is by −1, and G⊂haut0(M) is a subgroup of index 2.
If d=l=0, i.e., M=S2×S4, then there is a short exact sequence
1→Imψ→haut(M)→Z/2→1,
where Imψ is isomorphic to G×(Z/2×Z/2), G≅Z/2×Z/12, the actions of (1,0) and (0,1) on G are by −1. G⊂haut0(M) is a subgroup of index 2.
If d=0, l is even and non-zero, then Imψ is a subgroup of index 4, isomorphic to G×Z/2, G≅Z/2×Z/12, the action of Z/2 on G is by −1. G⊂haut0(M) is a subgroup of index 2.
If d=0 and l is odd, then Imψ is a subgroup of index 2 or 4, isomorphic to (Z/2×Z/6)×Z/2, the action of Z/2 is by −1.
Theorem 4.2. Let M be a simply connected nonspin homology CP3.
If d=0, then there are short exact sequences
1→Imψ→haut(M)→Z/2→1
1→G→Imψ→Z/2→1,
where G is isomorphic to
12∣d
3∣d,4∤d
3∤d,4∣d
otherwise
G
Z/6×Z/4
Z/3×Z/4
Z/2×Z/4
Z/4
Table 1.
The action of Z/2 on G is by −1. G⊂haut0(M) is a subgroup of index 2.
If d=l=0, i.e. M=S4×S2, then there is a short exact sequence
1→Imψ→haut(M)→Z/2→1,
where Imψ is isomorphic to G×(Z/2×Z/2), G≅Z/6×Z/4, the actions of (1,0) and (0,1) on G are by −1. G⊂haut0(M) is a subgroup of index 2.
If d=0 and l=0, then Imψ is a subgroup of index 4, and there is a short exact sequence
0→G→Imψ→Z/2→0,
where G≅Z/6×Z/4, the action of Z/2 on G is by −1. G⊂haut0(M) is a subgroup of index 2
Proof of Theorem 4.1 and Theorem 4.2. First assume d=0. Then by Theorem 2.5 we have the following commutative diagram, where rows and columns are exact sequences
When M is spin, by Theorem 2.5 the upper exact sequence splits. If d is odd, by Proposition 3.1ψ0 is surjective. This implies that ψ is surjective, and haut(M)=haut0(M)⋊Z/2, with haut0(M)=Imψ0. If d is even and d=0, by Proposition 3.1Imψ0 is a subgroup of index 2 in haut0(M). Hence Imψ is a subgroup of index 2 in haut(M), and isomorphic to Imψ0⋊Z/2. By Proposition 3.2Imψ0 is isomorphic to T(M)/K, whose isomorphism class is given in Proposition 2.2.
When M is non-spin, d is even. By Proposition 3.1Imψ0 is a subgroup of index 2 in haut0(M). Hence Imψ is a subgroup of index 2 in haut(M). There is a short exact sequence
where the homomorphisms from MCG(M) or haut(M) to Z/2×Z/2 are induced by the actions on H2(M) and H4(M), and the two rows are split short exact sequences. By Proposition 3.1Imψ0 is a subgroup of index 2 in haut0(M), hence Imψ is a subgroup of index 2 in haut(M), isomorphic to Imψ0⋊(Z/2×Z/2), with Imψ0≅Z/2×Z/12 if M=S2×S4, or Imψ0≅Z/6×Z/4 if M=S4×S2.
If M is spin with l even and l=0, or if M is nonspin with l=0, then we have the following commutative diagram
where the first row is by Theorem 2.5, the second row by the fact that M is homotopy equivalent to S2×S4 or S4×S2, and the homomorphism Z/2→Z/2×Z/2 maps 1 to (1,0). This shows that Imψ has index 4 in haut(M). When M is spin, the upper exact sequence splits (Theorem 2.5), therefore Imψ is isomorphic to Imψ0⋊Z/2, with Imψ0≅Z/2×Z/12. When M is nonspin, there is a short exact sequence
1→Imψ0→Imψ→Z/2→1
with Imψ0≅Z/6.
If M is spin and l is odd, it is not known to the authors whether the image of the homomorphism haut(M)→Z/2×Z/2 in diagram (6) is Z/2 or Z/2×Z/2. Therefore Imψ is a subgroup of index 2 or 4 in haut(M).
Proof of Proposition 1.2. The first statement is shown in Lemma 2.4. The second statement is shown by examining the codimension 4 splitting invariant. It suffices to consider the case M=Xd, the S2-bundle over S4 with characteristic map d∈π3SO(3). Let f be a smooth map, transverse to a fiber S2⊂Xd, and homotopic to fη. Then the preimage F=f−1(S2) is a framed surface in Xd. By inspecting the construction of fη one sees that F is a disjoint union of the fiber S2 and g−1(x0), where g:S6→S4 is a smooth map representing the non-trivial element of π6(S4) and x0∈S4 is a regular value of g. By the Thom-Pontrjagin construction, the set of concordance classes of framed surfaces in Xd is isomorphic to the homotopy set [Xd,S4]. Therefore in order to show that f is not homotopic to a diffeomorphism, it suffices to show that the map π6(S4)→[Xd,S4] induced by the collapsing map Xd→S6 is injective. From the cofiber sequence
S5bPd→Xd→S6ΣbΣPd
one has an exact sequence [ΣPd,S4]→π6(S4)→[Xd,S4]. When d is odd, ΣPd is homotopy equivalent to ΣCP2 and it is easy to show [ΣCP2,S4]=0. When d is even, it is known that Σb is d-multiple of a certain element in π6(ΣPd) ([21 Lemma 4.5]). Since π6(S4)≅Z/2, the map [ΣPd,S4]→π6(S4) must be trivial.
Proof of Theorem 1.3 and 1.6. The statements are rephrasing of Theorem 4.1 and 4.2. It remains to show that when d and l are even there is a splitting. Since fη represents an element of order 2 in haut(M), and by Proposition 1.2fη is not homotopic to a diffeomorphism, it provides a splitting of the exact sequence
1→MCG(M)→haut(M)→Z/2→1.
Proof of Corollary 1.5. By comparing the cardinalities of the group G in Theorem 1.3 and the group π6(Xd) in formula (2), one sees that the map π6(Xd)→haut(Xd) is injective. Therefore exact sequence (1) is extended to a short exact sequence
0→π6(Xd)→haut(Xd)→Gd→1.
Note that homotopy equivalences constructed from π6(Xd) are all orientation-preserving. When d is odd, Gd≅Z/2, a splitting is given by the fiberwise antipodal map fa, which is orientation-reversing. When d is even, Gd≅Z/2×Z/2, a preimage of (1,0)∈Z/2×Z/2 in haut(Xd) is given by fa. Since in this case Imψ is isomorphic to Z/2×Z/(12,d/2), the order of the Dehn twist tρ must be 2 (Lemma 2.4), and is not in the image of π6(Xd). It is a preimage of (0,1)∈Z/2×Z/2. □
Tool Use. In this paper no instrument, software or AI was use.
HLM, School of Mathematics and Systems Science, Chinese Academy of Sciences; School of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China
[1]M. Atiyah, R. Bott, The Yang–Mills equations over Riemann surfaces, Philos. Trans. R Soc. Lond. Ser. A Math. Phys. Eng. Sci. 308 (1983), 523-615.
[2]M. Arkowitz, The group of self-homotopy equivalences—a survey, Lecture Notes in Math., vol. 1425, Springer, 1990, pp. 170-203.DOI
[3]H. Baues, On the group of homotopy equivalences of a manifold, Trans. A.M.S., vol. 348, No. 12, December 1996, 4737-4773.DOI
[4]H. Baues, The degree of maps between certain 6-manifolds, Compositio Mathematica 110: 51-64, 1998.DOI
[5]W. D. Barcus, M. G. Barratt, On the homotopy classification of the extension a fixed map, Trans. Amer. Math. Soc. 88 (1958), 57-74.DOI
[6]H. Baues, J. Buth, On the group of homotopy equivalences of simply connected five manifolds, Math. Z., 222, 1996, 573-614.DOI
[7]G. Brumfiel, Homotopy equivalences of almost smooth manifolds, Comment. Math. Helv. 46 (1971), 381-407.
[8]J. Cerf, La strafification naturelle des espaces de fonctions différentiables réelles et le théorème de la pseudo-isotopie, Inst. Hautes Etudes Sci. Publ. Math. No. 39 (1970) 5-173.DOI
[9]P. E. Jupp, Classification of certain 6-manifolds, Proc. Camb. Phil. Soc. (1973), 73, 293-300.
[10]M. Kreck and W. Lück, The Novikov Conjecture: Geometry and Algebra, Birkhäuser Verlag, 2005.DOI
[11]M. Kreck, On the classification of 1-connected 7-manifolds with torsion free second homology, J. Topology 11 (2018) No. 3, 720-738.DOI
[12]M. Kreck and Y. Su, Mapping class group of manifolds which look like 3-dimensional complete intersections, Duke Math. J. 174 (2025), no. 3, 501-574.
[13]J. W. Milnor, On simply-connected 4-manifolds, Symposium International de Topologia Algebraica, Mexico (1958), 122-128.
[14]M. Olbermann, Involutions on S^6 with 3-dimensional fixed point set, Algebr. Geom. Topol. 10 (2010), no. 4, 1905-1932.DOI
[15]M. Olbermann, Conjugations on 6-manifolds with free integral homology, Math. Ann. 353 (2012), no. 1, 65-93.arxiv.org/abs/1001.0911
[17]N. Sawashita, On the group of self-equivalences of the product of spheres, Horishima Math. J. 5 (1975), 69-86.DOI
[18]P. Teichner, On the signature of four-manifolds with universal cover spin, Math. Ann. 295, 745-759 (1993).DOI
[19]C. T. C. Wall, Classification problems in differential topology V, Invent. math. 1, 355-374 (1966).DOI
[21]K. Yamaguchi, The group of self-homotopy equivalences of S²-bundles over S⁴. I, Kodai Math. J. 9(1986), 308-326.DOI