Introduction
Which groups occur as fundamental groups of closed manifolds satisfying a given curvature condition is a classical question, and in dimension four the answer is known at both ends of the scale of positivity conditions. For positive scalar curvature, Gromov and Lawson [11 ] and Schoen and Yau [28 ] proved that the condition is preserved under surgery of codimension at least three, and Carr [5 Corollary 2] proved that every finitely presented group is the fundamental group of a closed four-manifold of positive scalar curvature. Carr’s manifolds are boundaries of regular neighborhoods of finite two-dimensional complexes in R 5 \mathbb{R}^{5} R 5 . They can also be obtained from a connected sum of copies of S 1 × S 3 S^{1} \times S^{3} S 1 × S 3 by surgery along circles that represent the relators of a presentation. Since the fundamental group of a closed manifold is always finitely presented, positive scalar curvature imposes no restriction at all. Much stronger conditions, by contrast, leave very little room. For positive isotropic curvature, introduced by Micallef and Moore [22 ] , Fraser [10 ] proved that in dimensions at least five the fundamental group of a closed manifold contains no subgroup isomorphic to Z ⊕ Z \mathbb{Z} \oplus\mathbb{Z} Z ⊕ Z , and in dimension four the classification of Chen, Tang and Zhu [6, Corollary 1] implies that the fundamental group is virtually free , that is, it contains a free subgroup of finite index.
In this paper we determine the fundamental groups for a condition that lies strictly between these two. Let ( M , g ) (M,g) ( M , g ) be a Riemannian four-manifold. The biorthogonal curvature of a two-plane σ ⊂ T p M \sigma\subset T_pM σ ⊂ T p M is the average
K g ⊥ ( σ ) = 1 2 ( K g ( σ ) + K g ( σ ⊥ ) ) , K_g^\perp(\sigma) = \frac{1}{2}\left(K_g(\sigma) + K_g(\sigma^\perp)\right), K g ⊥ ( σ ) = 2 1 ( K g ( σ ) + K g ( σ ⊥ ) ) ,
and we put
k g ( p ) = min σ ⊂ T p M K g ⊥ ( σ ) . k_g(p) = \min_{\sigma\subset T_pM} K_g^\perp(\sigma). k g ( p ) = σ ⊂ T p M min K g ⊥ ( σ ) .
where K g K_g K g denotes the sectional curvature and σ ⊥ \sigma^\perp σ ⊥ the orthogonal complement of σ \sigma σ in T p M T_pM T p M . We say that g g g has positive biorthogonal curvature if k g > 0 k_g > 0 k g > 0 everywhere. In an orthonormal frame, the scalar curvature s g s_g s g is four times the sum of the biorthogonal curvatures of the three pairs of complementary coordinate planes, and hence
K g > 0 ⟹ k g > 0 ⟹ s g > 0. K_g > 0 \quad\Longrightarrow\quad k_g > 0 \quad\Longrightarrow\quad s_g > 0. K g > 0 ⟹ k g > 0 ⟹ s g > 0.
Neither implication can be reversed. Biorthogonal curvature was studied by Seaman [29 ] in connection with orthogonal pinching, and by Costa [8 ] and Costa–Ribeiro [9 ] in connection with modified Yamabe invariants. On the existence side, Bettiol [3 ] constructed metrics with k g > 0 k_g > 0 k g > 0 on S 2 × S 2 S^2 \times S^2 S 2 × S 2 . In [4 ] he used Hoelzel’s surgery criterion [18 ] to prove that the condition is preserved under connected sums, and he classified, up to homeomorphism, the closed simply connected four-manifolds that admit it; the list is the same as for positive scalar curvature once the smooth structure is allowed to change. Beyond the simply connected case, he asked the following question [4 Section 3] .
Question 1.1. Is every finitely presented group the fundamental group of a closed four-manifold with positive biorthogonal curvature?
Bettiol pointed out that an affirmative answer would follow if the condition were stable under the corresponding surgeries, to which Hoelzel’s criterion does not apply. He singled out S 2 × T 2 S^2 \times T^2 S 2 × T 2 as the crucial test case, and the question whether S 2 × T 2 S^2 \times T^2 S 2 × T 2 admits positive biorthogonal curvature is also Question 4 in the problem list of Morgan and Pansu [25 ] . We settled this test case in [7 ] by constructing explicit metrics of positive biorthogonal curvature on S 2 × T 2 S^2 \times T^2 S 2 × T 2 , and more generally on every smooth oriented S 2 S^2 S 2 -bundle over a closed oriented surface. Since Z \mathbb{Z} Z is not virtually free, the biorthogonal and the isotropic conditions already part ways on S 2 × T 2 S^2 \times T^2 S 2 × T 2 . The purpose of the present paper is to answer Bettiol’s question in full.
Main results
Theorem 1.2. A group Γ \Gamma Γ is isomorphic to the fundamental group of a closed connected oriented smooth four-manifold admitting a metric of positive biorthogonal curvature if and only if Γ \Gamma Γ is finitely presented.
Only the “if” part has geometric content, since a closed smooth manifold has a finite triangulation and its fundamental group is therefore finitely presented. Thus positive biorthogonal curvature and positive scalar curvature allow exactly the same fundamental groups.
Our construction also gives some control over the topology of the manifold. For a finitely presented group Γ \Gamma Γ we put b 1 ( Γ ) = dim Q H 1 ( Γ ; Q ) b_1(\Gamma) = \dim_{\mathbb{Q}} H_1(\Gamma;\mathbb{Q}) b 1 ( Γ ) = dim Q H 1 ( Γ ; Q ) . For a closed oriented four-manifold M M M we write χ ( M ) \chi(M) χ ( M ) for its Euler characteristic, b j ( M ) b_j(M) b j ( M ) for its rational Betti numbers, and sign ( M ) \operatorname{sign}(M) sign ( M ) for the signature of the intersection form on H 2 ( M ; Q ) H_2(M;\mathbb{Q}) H 2 ( M ; Q ) . Recall that an oriented manifold is spin if its second Stiefel–Whitney class vanishes, and that a manifold is stably parallelizable if the sum of its tangent bundle and a trivial line bundle is trivial. A stably parallelizable closed oriented four-manifold is spin, and its signature vanishes because its first Pontryagin class is zero [24 ] .
Corollary 1.3. Suppose that Γ \Gamma Γ has a presentation with p p p generators and q q q relators, none of which is trivial in the free group on the generators. Then the manifold M M M in Theorem 1.2 can be chosen to be stably parallelizable, with
χ ( M ) = 2 ( 1 − p + q ) , b 2 ( M ) = 2 ( q − p + b 1 ( Γ ) ) . \chi(M) = 2(1-p+q), \qquad b_2(M) = 2(q-p+b_1(\Gamma)). χ ( M ) = 2 ( 1 − p + q ) , b 2 ( M ) = 2 ( q − p + b 1 ( Γ )) .
The manifold M M M is the boundary of a five-dimensional handlebody that is built from the presentation, and it is of the kind that Carr used for positive scalar curvature; see Remarks 5.2 and 5.3. For positive scalar curvature the corresponding statement was recorded by Mantione and Torres [21 Theorem 6] .
Behind Theorem 1.2 is a surgery result, which we state separately. Let γ \gamma γ be an embedded circle in an oriented four-manifold M M M . Its normal bundle is trivial and has exactly two oriented trivializations up to homotopy, the normal framings of γ \gamma γ . Surgery along γ \gamma γ removes an open tubular neighborhood S 1 × int D 3 S^1 \times\operatorname{int} D^3 S 1 × int D 3 of γ \gamma γ and glues in D 2 × S 2 D^2 \times S^2 D 2 × S 2 along the common boundary S 1 × S 2 S^1 \times S^2 S 1 × S 2 ; the result depends on γ \gamma γ and on the chosen framing.
Theorem 1.4. Let M M M be a closed connected oriented smooth four-manifold that admits a metric of positive biorthogonal curvature. Then every manifold obtained from M M M by surgery along finitely many pairwise disjoint embedded circles, with arbitrary normal framings, also admits a metric of positive biorthogonal curvature.
Bettiol’s connected-sum theorem [4 Proposition 3.1] takes care of surgery in codimension four. Together with Theorem 1.4 it shows that the class of closed oriented four-manifolds that admit positive biorthogonal curvature is closed under surgery of codimension at least three. The class of manifolds that admit positive scalar curvature has the same property by the theorem of Gromov–Lawson and Schoen–Yau. There is a difference, however. The surgery theorem for positive scalar curvature modifies a given metric near the surgery sphere only, whereas Theorem 1.4 is a statement about diffeomorphism types. In its proof we first move the circle, by an isotopy of a stabilization of M M M , to a closed geodesic of a suitable metric, and we do not claim that a given metric can be modified near a given circle only; compare Subsection 6.3.
Theorem 1.4 does not follow from the general surgery theory for curvature conditions. The model for a surgery of codimension three is S 2 × R 2 S^2 \times\mathbb{R}^2 S 2 × R 2 , and its mixed planes are flat and have flat complements. Its curvature operator therefore lies on the boundary of the cone of operators with positive biorthogonal curvature, whereas Hoelzel’s criterion [18 Theorem B] requires it to lie in the interior. We get around this by twisting the handle; see Subsection 1.2.
Attaching a two-handle to a five-manifold changes its boundary by surgery along a circle. Theorem 1.4 therefore has the following consequence, which contains the existence statement of Theorem 1.2 .
Corollary 1.5. Let W W W be a compact connected oriented smooth five-manifold that has a handle decomposition with handles of index at most two. Then ∂ W \partial W ∂ W admits a metric of positive biorthogonal curvature.
We mention some examples. Recall that the double of a compact manifold X X X with boundary is the union of two copies of X X X along their common boundary, and that the spun manifold of a closed connected three-manifold Y Y Y is the boundary of Y 0 × D 2 Y_0 \times D^2 Y 0 × D 2 , where Y 0 Y_0 Y 0 is the complement of an open ball in Y Y Y and D 2 D^2 D 2 is the closed two-dimensional disk.
Corollary 1.6. The following manifolds admit metrics of positive biorthogonal curvature.
(a) Every oriented S 2 S^2 S 2 -bundle over a closed oriented surface.
(b) The product Y × S 1 Y \times S^1 Y × S 1 , for every closed orientable three-manifold Y Y Y that admits a metric of positive scalar curvature.
(c) The double of every compact connected oriented four-manifold that has a handle decomposition with handles of index at most two.
(d) The spun manifold of every closed connected oriented three-manifold.
Part (a) is the main result of [7 ] , and part (b) with Y = S 2 × S 1 Y = S^2 \times S^1 Y = S 2 × S 1 gives a second proof for S 2 × T 2 S^2 \times T^2 S 2 × T 2 . Parts (a), (c) and (d) are special cases of Corollary 1.5.
If we allow connected sums with copies of S 2 × S 2 S^2 \times S^2 S 2 × S 2 , we can also treat manifolds that are not given to us by a surgery description. Let N N N be a closed connected oriented four-manifold with fundamental group Γ \Gamma Γ , and let c c c be a classifying map N → B Γ N \to B\Gamma N → B Γ of its universal cover, where B Γ B\Gamma B Γ denotes an Eilenberg–MacLane space of type K ( Γ , 1 ) K(\Gamma,1) K ( Γ , 1 ) .
Corollary 1.7. Suppose that N N N satisfies one of the following conditions.
(a) The universal cover of N N N is not spin, and c ∗ [ N ] = 0 c_*[N] = 0 c ∗ [ N ] = 0 in H 4 ( Γ ; Z ) H_4(\Gamma;\mathbb{Z}) H 4 ( Γ ; Z ) .
(b) N N N is spin, and for some spin structure the pair ( N , c ) (N,c) ( N , c ) is null-bordant in the spin bordism group Ω 4 S p i n ( B Γ ) \Omega^{\mathrm{Spin}}_4(B\Gamma) Ω 4 Spin ( B Γ ) .
Then the connected sum of N N N with r r r copies of S 2 × S 2 S^2 \times S^2 S 2 × S 2 admits a metric of positive biorthogonal curvature for all sufficiently large r r r .
The condition (a) of Corollary 1.7 holds, for instance, for every N N N with non-spin universal cover whose fundamental group satisfies H 4 ( Γ ; Z ) = 0 H_4(\Gamma;\mathbb{Z}) = 0 H 4 ( Γ ; Z ) = 0 ; this includes free groups, surface groups and finite cyclic groups.
Theorem 1.2 does not say that every closed four-manifold of positive scalar curvature admits positive biorthogonal curvature, and we do not know whether this is true.
Question 1.8. Is there a closed four-manifold that admits a metric of positive scalar curvature but no metric of positive biorthogonal curvature?
In the simply connected case, Bettiol’s classification shows that such a manifold cannot be detected by its homeomorphism type. Our results add two restrictions. By Theorem 1.4 , the two classes of manifolds are closed under the same surgeries, so no example can be produced, or excluded, by surgery alone. By Corollary 1.7 , in the range covered there the two conditions cannot be separated by any invariant that survives connected sums with S 2 × S 2 S^2 \times S^2 S 2 × S 2 . An example answering Question 1.8 affirmatively would therefore have to rest on an obstruction to positive biorthogonal curvature that is not an obstruction to positive scalar curvature and that is unstable in this sense. We know of no such obstruction; see Subsection 6.3 for further discussion.
Finally, we study the minimal Euler characteristic q ⊥ ( Γ ) q_{\perp}(\Gamma) q ⊥ ( Γ ) of a closed connected oriented four-manifold with fundamental group Γ \Gamma Γ that admits positive biorthogonal curvature. This is the analogue, for our curvature condition, of the Hausmann–Weinberger invariant q ( Γ ) q(\Gamma) q ( Γ ) [14 ] , the minimum over all closed oriented four-manifolds with fundamental group Γ \Gamma Γ , and of the invariant q > 0 ( Γ ) q_{>0}(\Gamma) q > 0 ( Γ ) of Mantione and Torres [21 ] , the minimum over those that admit positive scalar curvature. We show that
q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) ≤ 2 − 2 def ( Γ ) , q(\Gamma) \leq q_{>0}(\Gamma) \leq q_{\perp}(\Gamma) \leq2 - 2\operatorname{def}(\Gamma), q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) ≤ 2 − 2 def ( Γ ) ,
where def ( Γ ) \operatorname{def}(\Gamma) def ( Γ ) is the deficiency of Γ \Gamma Γ (Corollary 6.1 ). These four numbers agree if there is a field F \mathbb{F} F such that H 4 ( Γ ; F ) = 0 H_{4}(\Gamma;\mathbb{F}) = 0 H 4 ( Γ ; F ) = 0 and def ( Γ ) = b 1 ( Γ ; F ) − b 2 ( Γ ; F ) \operatorname{def}(\Gamma) = b_{1}(\Gamma;\mathbb{F}) - b_{2}(\Gamma;\mathbb{F}) def ( Γ ) = b 1 ( Γ ; F ) − b 2 ( Γ ; F ) , where b j ( Γ ; F ) = dim F H j ( Γ ; F ) b_{j}(\Gamma;\mathbb{F}) = \dim_{\mathbb{F}} H_{j}(\Gamma;\mathbb{F}) b j ( Γ ; F ) = dim F H j ( Γ ; F ) (Corollary 6.3 ). For q > 0 ( Γ ) q_{>0}(\Gamma) q > 0 ( Γ ) an equality of this kind is due to Mantione and Torres [21 Theorem B] , and the new information is that the minimum is attained by manifolds of positive biorthogonal curvature. The corollary applies to the fundamental group of a finite aspherical two-dimensional complex X X X , where the common value is 2 χ ( X ) 2\chi(X) 2 χ ( X ) (Corollary 6.4); for q ( Γ ) q(\Gamma) q ( Γ ) this is a theorem of Hillman [15 ] . It also applies to the fundamental group of a closed orientable aspherical three-manifold, where the common value is 2. For the fundamental group of a closed oriented surface Σ \Sigma Σ of positive genus we obtain q ⊥ = 2 χ ( Σ ) q_{\perp} = 2\chi(\Sigma) q ⊥ = 2 χ ( Σ ) , and the minimum is attained by both S 2 S^{2} S 2 -bundles over the surface. The invariants q q q and q ⊥ q_{\perp} q ⊥ do not agree in general, since q ( Z 4 ) = 0 q(\mathbb{Z}^{4}) = 0 q ( Z 4 ) = 0 and q ⊥ ( Z 4 ) = 6 q_{\perp}(\mathbb{Z}^{4}) = 6 q ⊥ ( Z 4 ) = 6 (Remark 6.6). We do not know whether q > 0 q_{>0} q > 0 and q ⊥ q_{\perp} q ⊥ can differ (Question 6.7 ).
Outline of the construction
In dimension four, surgery along a circle replaces S 1 × int D 3 S^{1} \times\operatorname{int} D^{3} S 1 × int D 3 by D 2 × S 2 D^{2} \times S^{2} D 2 × S 2 , where D j D^{j} D j denotes the closed j j j -dimensional disk. As explained above, the standard model S 2 × R 2 S^{2} \times\mathbb{R}^{2} S 2 × R 2 is of no use to us. The problem is not an artifact of the model. Let r r r and θ \theta θ be polar coordinates on the disk, and consider an untwisted warped product d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 g S 2 dr^{2} + f(r)^{2}d\theta^{2} + h(r)^{2}g_{S^{2}} d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 g S 2 on D 2 × S 2 D^{2} \times S^{2} D 2 × S 2 . The planes spanned by ∂ r \partial_{r} ∂ r and a direction tangent to the sphere have biorthogonal curvature − ( f h ′ ) / ( 2 f h ) -(fh')/(2fh) − ( f h ′ ) / ( 2 f h ) . The function f h ′ fh' f h ′ vanishes at the center of the disk and is positive on the collar of a round tube. Hence ( f h ′ ) (fh') ( f h ′ ) is positive somewhere, and the biorthogonal curvature of these planes is negative there.
We compensate for this by letting the sphere rotate as we go around the disk. With spherical coordinates ϑ \vartheta ϑ , ϕ \phi ϕ on S 2 S^{2} S 2 , we consider
g = d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( r ) d θ ) 2 } , (1) g = dr^{2} + f(r)^{2}d\theta^{2} + h(r)^{2}\left\{d\vartheta^{2} + \sin^{2}\vartheta\left(d\phi+ A(r)\,d\theta\right)^{2}\right\}, \tag*{(1)} g = d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( r ) d θ ) 2 } , (1)
where the function A A A is the coefficient of a rotational connection. Its curvature enters through the quantity κ = h 2 A ′ / f \kappa= h^{2}A'/f κ = h 2 A ′ / f , and it contributes the positive term κ 2 sin 2 ϑ / ( 8 h 2 ) \kappa^{2}\sin^{2}\vartheta/(8h^{2}) κ 2 sin 2 ϑ / ( 8 h 2 ) to the biorthogonal curvature of the critical planes. This term vanishes at the poles of the sphere, which are fixed by the rotation. We then use the fact that k g k_{g} k g transforms under a conformal change of the metric exactly like scalar curvature in dimension four. The conformal factor u = 1 + κ 2 cos 2 ϑ / 24 u = 1 + \kappa^{2}\cos^{2}\vartheta/24 u = 1 + κ 2 cos 2 ϑ /24 redistributes the positive term evenly and turns it into κ 2 / ( 12 h 2 ) \kappa^{2}/(12h^{2}) κ 2 / ( 12 h 2 ) , uniformly on the sphere. If h h h increases sufficiently slowly, this term dominates the negative contribution of the convexity of h h h . Once the sphere is large enough, we make h h h concave, remove the connection curvature, and match the metric exactly to the collar of a round tube. The handle obtained in this way is very long and thin, but all its parameters are explicit and finite. The precise statement is Theorem 4.1 .
To use the handle we have to bring the metric near the surgery circle into the form of a round tube. We can do this when the circle is a closed geodesic. The given metric and the round model then agree to first order along the circle, and the local flexibility theorem of Bär and Hanke [1 ] allows us to replace one by the other inside the cone of positive biorthogonal curvature. The round tube remembers the normal holonomy of the geodesic, an arbitrary rotation, and exact matching has to take it into account; a portion of the handle of constant radius adjusts the total rotation of the connection, and an integral twist of the boundary realizes either normal framing.
It remains to produce the geodesics. For Theorem 1.2 we replace each relator r j r_j r j of a presentation by the pair r j y j r_jy_j r j y j , y j y_j y j , where y j y_j y j is a new generator. The new relators are primitive elements of the enlarged free group, so that the corresponding free homotopy classes on a connected sum of copies of S 1 × S 3 S^1 \times S^3 S 1 × S 3 contain prime closed geodesics; finite perturbation lemmas of Rademacher [27 ] make these geodesics embedded and pairwise disjoint. For Theorem 1.4 we use the same device geometrically. We take the connected sum of M M M with two copies of S 1 × S 3 S^1 \times S^3 S 1 × S 3 , slide the given circle over them, and use the fact that homotopic circles in a four-manifold are isotopic.
Organization
In § 2 we collect the curvature identities that we need. In § 3 we prepare the surgery circles and their tubular neighborhoods. § 4 contains the construction of the handle and is the technical core of the paper. The main results are proved in § 5 . § 6 contains the results on Euler characteristics, the proof of Corollary 1.7 , and some open questions.
Preliminaries on biorthogonal curvature
Let ( M 4 , g ) (M^4,g) ( M 4 , g ) be an oriented Riemannian four-manifold with Levi-Civita connection ∇ \nabla ∇ . We use the conventions
R ( X , Y ) Z = ∇ X ∇ Y Z − ∇ Y ∇ X Z − ∇ [ X , Y ] Z , ⟨ X ∧ Y , Z ∧ W ⟩ = g ( X , Z ) g ( Y , W ) − g ( X , W ) g ( Y , Z ) . R(X,Y)Z = \nabla_X\nabla_YZ - \nabla_Y\nabla_XZ - \nabla_{[X,Y]}Z,
\qquad
\langle X \wedge Y, Z \wedge W\rangle= g(X,Z)g(Y,W) - g(X,W)g(Y,Z). R ( X , Y ) Z = ∇ X ∇ Y Z − ∇ Y ∇ X Z − ∇ [ X , Y ] Z , ⟨ X ∧ Y , Z ∧ W ⟩ = g ( X , Z ) g ( Y , W ) − g ( X , W ) g ( Y , Z ) .
and we define the curvature operator R \mathcal{R} R as the self-adjoint endomorphism of Λ 2 T M \Lambda^2TM Λ 2 T M with
⟨ R ( X ∧ Y ) , Z ∧ W ⟩ = g ( R ( X , Y ) W , Z ) . \langle\mathcal{R}(X \wedge Y), Z \wedge W\rangle= g(R(X,Y)W,Z). ⟨ R ( X ∧ Y ) , Z ∧ W ⟩ = g ( R ( X , Y ) W , Z ) .
We write R ( α , β ) = ⟨ R α , β ⟩ \mathcal{R}(\alpha,\beta)=\langle\mathcal{R}\alpha,\beta\rangle R ( α , β ) = ⟨ R α , β ⟩ for the associated symmetric bilinear form, and we use g g g to identify two-vectors with two-forms, keeping the same notation for the curvature operator on Λ 2 T ∗ M \Lambda^2T^*M Λ 2 T ∗ M . If X X X and Y Y Y are orthonormal, then K g ( span { X , Y } ) = R ( X ∧ Y , X ∧ Y ) K_g(\operatorname{span}\{X,Y\})=\mathcal{R}(X\wedge Y,X\wedge Y) K g ( span { X , Y }) = R ( X ∧ Y , X ∧ Y ) ; in particular, the round unit sphere has sectional curvature one. For an orthonormal frame e 1 , … , e 4 e_1,\ldots,e_4 e 1 , … , e 4 we abbreviate
R i j , k l = R ( e i ∧ e j , e k ∧ e l ) , R i j , i j = K g ( e i , e j ) \mathcal{R}_{ij,kl}=\mathcal{R}(e_i\wedge e_j,e_k\wedge e_l),
\qquad
\mathcal{R}_{ij,ij}=K_g(e_i,e_j) R ij , k l = R ( e i ∧ e j , e k ∧ e l ) , R ij , ij = K g ( e i , e j )
is the sectional curvature of the plane spanned by e i e_i e i and $e_j.
The Hodge decomposition
The Hodge star ∗ \ast ∗ is an isometric involution of Λ 2 T p M \Lambda^{2}T_pM Λ 2 T p M . Its eigenspaces Λ p ± = { α ∣ ∗ α = ± α } \Lambda_p^{\pm}=\{\alpha\mid\ast\alpha=\pm\alpha\} Λ p ± = { α ∣ ∗ α = ± α } are three-dimensional and orthogonal to each other. If e 1 , … , e 4 e_1,\ldots,e_4 e 1 , … , e 4 is an oriented orthonormal basis of T p M T_pM T p M , then
ω 1 ± = 1 2 ( e 1 ∧ e 2 ± e 3 ∧ e 4 ) , ω 2 ± = 1 2 ( e 1 ∧ e 3 ∓ e 2 ∧ e 4 ) , ω 3 ± = 1 2 ( e 1 ∧ e 4 ± e 2 ∧ e 3 ) (2) \begin{aligned}
\omega_1^{\pm}&=\frac{1}{\sqrt{2}}(e_1\wedge e_2\pm e_3\wedge e_4),\\
\omega_2^{\pm}&=\frac{1}{\sqrt{2}}(e_1\wedge e_3\mp e_2\wedge e_4),\\
\omega_3^{\pm}&=\frac{1}{\sqrt{2}}(e_1\wedge e_4\pm e_2\wedge e_3) \tag*{(2)}
\end{aligned} ω 1 ± ω 2 ± ω 3 ± = 2 1 ( e 1 ∧ e 2 ± e 3 ∧ e 4 ) , = 2 1 ( e 1 ∧ e 3 ∓ e 2 ∧ e 4 ) , = 2 1 ( e 1 ∧ e 4 ± e 2 ∧ e 3 ) (2)
are orthonormal bases of Λ p ± \Lambda_p^{\pm} Λ p ± . With respect to the decomposition Λ 2 T M = Λ + ⊕ Λ − \Lambda^2TM=\Lambda^+\oplus\Lambda^- Λ 2 T M = Λ + ⊕ Λ − the curvature operator has the block form
R = ( A + B B ∗ A − ) , A ± = W ± + s g 12 I Λ ± , (3) \mathcal{R}=\begin{pmatrix}A_+&B\\ B^*&A_-\end{pmatrix},\qquad A_{\pm}=W_{\pm}+\frac{s_g}{12}I_{\Lambda^{\pm}}, \tag*{(3)} R = ( A + B ∗ B A − ) , A ± = W ± + 12 s g I Λ ± , (3)
where W ± W_{\pm} W ± are the self-dual and the anti-self-dual Weyl curvature operators and the off-diagonal block B B B is given by the traceless Ricci tensor; see Singer and Thorpe [30 ] and [9 Section 2] . We shall only need the diagonal blocks A ± A_{\pm} A ± , because of the following formula, which reduces every curvature estimate in this paper to an estimate for two symmetric 3 × 3 3\times3 3 × 3 matrices. We write λ min \lambda_{\min} λ m i n for the smallest eigenvalue of a self-adjoint operator.
Lemma 2.1 ([9 ] ). For every oriented Riemannian four-manifold ( M , g ) (M,g) ( M , g ) and every p ∈ M p\in M p ∈ M ,
k g ( p ) = 1 2 ( λ min ( A + ( p ) ) + λ min ( A − ( p ) ) ) . (4) k_g(p)=\frac{1}{2}\left(\lambda_{\min}(A_+(p))+\lambda_{\min}(A_-(p))\right). \tag*{(4)} k g ( p ) = 2 1 ( λ m i n ( A + ( p )) + λ m i n ( A − ( p )) ) . (4)
Proof. We recall the short argument. Let σ \sigma σ be a two-plane with oriented orthonormal basis e 1 , e 2 e_1,e_2 e 1 , e 2 , complete it to an oriented orthonormal basis of T p M T_pM T p M , and put α = e 1 ∧ e 2 \alpha=e_1\wedge e_2 α = e 1 ∧ e 2 , so that ∗ α = e 3 ∧ e 4 \ast\alpha=e_3\wedge e_4 ∗ α = e 3 ∧ e 4 represents σ ⊥ \sigma^\perp σ ⊥ . Writing α = α + + α − \alpha=\alpha_++\alpha_- α = α + + α − with α ± = 1 2 ( α ± ∗ α ) ∈ Λ ± \alpha_{\pm}=\frac{1}{2}(\alpha\pm\ast\alpha)\in\Lambda^{\pm} α ± = 2 1 ( α ± ∗ α ) ∈ Λ ± , we have ∗ α = α + − α − \ast\alpha=\alpha_+-\alpha_- ∗ α = α + − α − , and therefore
K g ⊥ ( σ ) = 1 2 ( R ( α , α ) + R ( ∗ α , ∗ α ) ) = A + ( α + , α + ) + A − ( α − , α − ) , ∣ α + ∣ 2 = ∣ α − ∣ 2 = 1 2 . K_g^\perp(\sigma)=\frac{1}{2}\bigl(\mathcal{R}(\alpha,\alpha)+\mathcal{R}(\ast\alpha,\ast\alpha)\bigr)=A_+(\alpha_+,\alpha_+)+A_-(\alpha_-,\alpha_-),\qquad|\alpha_+|^2=|\alpha_-|^2=\frac{1}{2}. K g ⊥ ( σ ) = 2 1 ( R ( α , α ) + R ( ∗ α , ∗ α ) ) = A + ( α + , α + ) + A − ( α − , α − ) , ∣ α + ∣ 2 = ∣ α − ∣ 2 = 2 1 .
Conversely, every pair of vectors α ± ∈ Λ p ± \alpha_{\pm}\in\Lambda_p^{\pm} α ± ∈ Λ p ± of norm 1 / 2 1/\sqrt{2} 1/ 2 arises in this way, since α + + α − \alpha_++\alpha_- α + + α − is then a unit two-vector with α ∧ α = 0 \alpha\wedge\alpha=0 α ∧ α = 0 , hence decomposable. Minimizing over the two factors separately gives (4) . □
Conformal changes
For a function u u u on ( M , g ) (M,g) ( M , g ) we use the Laplacian Δ g u = tr g ∇ 2 u \Delta_g u=\operatorname{tr}_g\nabla^2u Δ g u = tr g ∇ 2 u . Under a conformal change of the metric, the Hessian terms in the transformation law of the sectional curvature depend on the plane, but they cancel when one averages over a plane and its orthogonal complement. As a consequence, 1 2 k g \frac{1}{2}k_g 2 1 k g transforms exactly like the scalar curvature of a four-manifold. This is the reason why the biorthogonal condition is accessible to conformal methods, and it underlies the modified Yamabe invariants of Costa [8 ] .
Lemma 2.2 (Conformal change). Let ( M 4 , g ) (M^4,g) ( M 4 , g ) be a Riemannian four-manifold and let u u u be a positive smooth function on M M M . For every two-plane σ ⊂ T p M \sigma\subset T_pM σ ⊂ T p M ,
K u 2 g ⊥ ( σ ) = u − 2 ( K g ⊥ ( σ ) − Δ g u 2 u ) . K_{u^2g}^{\perp}(\sigma)=u^{-2}\left(K_g^{\perp}(\sigma)-\frac{\Delta_g u}{2u}\right). K u 2 g ⊥ ( σ ) = u − 2 ( K g ⊥ ( σ ) − 2 u Δ g u ) .
Consequently,
k u 2 g = u − 2 ( k g − Δ g u 2 u ) . (5) k_{u^{2}g}=u^{-2}\left(k_g-\frac{\Delta_g u}{2u}\right). \tag*{(5)} k u 2 g = u − 2 ( k g − 2 u Δ g u ) . (5)
Proof. Write u = e φ u=e^{\varphi} u = e φ . For g g g -orthonormal vectors X , Y X,Y X , Y , the sectional curvature of e 2 φ g e^{2\varphi}g e 2 φ g is given by
e 2 φ K e 2 φ g ( X , Y ) = K g ( X , Y ) − ∇ 2 φ ( X , X ) − ∇ 2 φ ( Y , Y ) + ( X φ ) 2 + ( Y φ ) 2 − ∣ ∇ φ ∣ 2 ; e^{2\varphi}K_{e^{2\varphi}g}(X,Y)=K_g(X,Y)-\nabla^2\varphi(X,X)-\nabla^2\varphi(Y,Y)+(X\varphi)^2+(Y\varphi)^2-\lvert\nabla\varphi\rvert^2; e 2 φ K e 2 φ g ( X , Y ) = K g ( X , Y ) − ∇ 2 φ ( X , X ) − ∇ 2 φ ( Y , Y ) + ( X φ ) 2 + ( Y φ ) 2 − ∣ ∇ φ ∣ 2 ;
see [2 Theorem 1.159] . Orthogonal complements are the same for the two metrics. Let e 1 , … , e 4 e_1,\ldots,e_4 e 1 , … , e 4 be a g g g -orthonormal basis with σ = span { e 1 , e 2 } \sigma=\operatorname{span}\{e_1,e_2\} σ = span { e 1 , e 2 } . Adding the formulas for ( e 1 , e 2 ) (e_1,e_2) ( e 1 , e 2 ) and ( e 3 , e 4 ) (e_3,e_4) ( e 3 , e 4 ) , we obtain
e 2 φ ( K e 2 φ g ( σ ) + K e 2 φ g ( σ ⊥ ) ) = K g ( σ ) + K g ( σ ⊥ ) − Δ g φ − ∣ ∇ φ ∣ 2 , e^{2\varphi}\left(K_{e^{2\varphi}g}(\sigma)+K_{e^{2\varphi}g}(\sigma^\perp)\right)=K_g(\sigma)+K_g(\sigma^\perp)-\Delta_g\varphi-\lvert\nabla\varphi\rvert^2, e 2 φ ( K e 2 φ g ( σ ) + K e 2 φ g ( σ ⊥ ) ) = K g ( σ ) + K g ( σ ⊥ ) − Δ g φ − ∣ ∇ φ ∣ 2 ,
and Δ g φ + ∣ ∇ φ ∣ 2 = u − 1 Δ g u \Delta_g\varphi+\lvert\nabla\varphi\rvert^2=u^{-1}\Delta_g u Δ g φ + ∣ ∇ φ ∣ 2 = u − 1 Δ g u . See also [7 Lemma 2.2] . □
The cone of positive biorthogonal curvature
In order to interpolate between two metrics we shall use the fact that positive biorthogonal curvature is a convex condition on the curvature tensor, a fact that also underlies Bettiol’s connected-sum theorem [4 Proposition 3.1] . We record it together with the quantitative inequality behind it.
An algebraic curvature tensor on a Euclidean vector space is a tensor with the symmetries of a Riemannian curvature tensor, including the first Bianchi identity. For such a tensor T T T on a four-dimensional Euclidean space we write K T ( P ) K_T(P) K T ( P ) for its sectional curvature on a two-plane P P P , and we put
k ( T ) = min P K T ( P ) + K T ( P ⊥ ) 2 , k(T)=\min_{P}\frac{K_T(P)+K_T(P^\perp)}{2}, k ( T ) = P min 2 K T ( P ) + K T ( P ⊥ ) ,
where P P P runs through the Grassmannian of two-planes.
Lemma 2.3. The function k k k is concave, that is,
k ( ( 1 − a ) T 0 + a T 1 ) ≥ ( 1 − a ) k ( T 0 ) + a k ( T 1 ) for 0 ≤ a ≤ 1. k\left((1-a)T_0+aT_1\right)\ge(1-a)k(T_0)+a\,k(T_1)\qquad\text{for }0\le a\le1. k ( ( 1 − a ) T 0 + a T 1 ) ≥ ( 1 − a ) k ( T 0 ) + a k ( T 1 ) for 0 ≤ a ≤ 1. The set { T ∣ k ( T ) > 0 } \{T\mid k(T)>0\} { T ∣ k ( T ) > 0 } is an open, convex, O ( 4 ) \mathrm{O}(4) O ( 4 ) -invariant cone. On a closed four-manifold, the set of metrics of positive biorthogonal curvature is open in the C 2 C^2 C 2 topology.
Proof. For a fixed plane P P P , the quantity ( K T ( P ) + K T ( P ⊥ ) ) / 2 (K_T(P)+K_T(P^\perp))/2 ( K T ( P ) + K T ( P ⊥ )) /2 is linear in T T T , and a minimum of linear functions is concave. This gives the inequality and the convexity of the cone. Since the Grassmannian is compact, k k k is continuous, indeed locally Lipschitz, and so the cone is open; its invariance under O ( 4 ) \mathrm{O}(4) O ( 4 ) and under multiplication by positive constants is clear from the definition. Finally, the curvature tensor of a metric depends continuously on the metric and its derivatives up to order two, and on a closed manifold a metric with k g > 0 k_g>0 k g > 0 satisfies k g ≥ c k_g\ge c k g ≥ c for some constant c > 0 c>0 c > 0 . This proves the last assertion. □
Surgery circles and their tubular neighborhoods
In order to attach the handle of § 4 we need the metric near each surgery circle to be a round tube, in a sense made precise in Lemma 3.5 . We achieve this for circles that are closed geodesics. In this section we first show how to obtain closed geodesics in the free homotopy classes that we need, and that they can be made embedded and disjoint. We then replace the metric near such a geodesic by a round tube, keeping track of the normal holonomy, and finally we recall the effect of circle surgery on the fundamental group.
Primitive relators and embedded geodesics
For n ≥ 1 n \ge1 n ≥ 1 let F n F_n F n be the free group on n n n given generators, and let F 0 F_0 F 0 be the trivial group. An element of a free group is primitive if it belongs to a free basis, and an element w w w of a group is a proper power if w = v d w = v^d w = v d for some element v v v and some integer d ≥ 2 d \ge2 d ≥ 2 . A primitive element is nontrivial and is not a proper power, because there is a homomorphism from the free group to Z \mathbb{Z} Z that sends it to 1 1 1 , whereas the image of v d v^d v d is divisible by d d d .
Free homotopy classes of oriented loops in a connected manifold correspond to conjugacy classes in its fundamental group, and reversing the orientation of a loop replaces the conjugacy class by that of the inverse. Thus a finite set of group elements that are pairwise nonconjugate, even after inverting some of them, determines free homotopy classes that are pairwise distinct up to reversal.
A length-minimizing closed geodesic in a class that is a proper power may well run several times around a shorter geodesic, and two relators of a presentation may be conjugate. The following elementary change of presentation rules out both phenomena.
Lemma 3.1. Every finitely presented group has a finite presentation whose relators are primitive elements of the free group on the generators and are pairwise nonconjugate, even after inversion.
Proof. Let Γ = ⟨ x 1 , … , x p ∣ r 1 , … , r q ⟩ \Gamma= \langle x_1, \ldots, x_p \mid r_1, \ldots, r_q \rangle Γ = ⟨ x 1 , … , x p ∣ r 1 , … , r q ⟩ be a finite presentation. We freely reduce the words r j r_j r j and discard those that become empty, so that each r j r_j r j is a nontrivial element of F p F_p F p . If q = 0 q = 0 q = 0 there is nothing to prove. Otherwise we introduce new generators y 1 , … , y q y_1, \ldots, y_q y 1 , … , y q and consider the presentation
⟨ x 1 , … , x p , y 1 , … , y q ∣ r 1 y 1 , y 1 , … , r q y q , y q ⟩ (6) \langle x_1, \ldots, x_p, y_1, \ldots, y_q \mid r_1y_1, y_1, \ldots, r_qy_q, y_q \rangle \tag*{(6)} ⟨ x 1 , … , x p , y 1 , … , y q ∣ r 1 y 1 , y 1 , … , r q y q , y q ⟩ (6)
It defines the same group Γ \Gamma Γ , since the relations y j = 1 y_j = 1 y j = 1 eliminate the new generators and turn r j y j = 1 r_jy_j = 1 r j y j = 1 into r j = 1 r_j = 1 r j = 1 . Each y j y_j y j is primitive in F p + q F_{p+q} F p + q . The endomorphism of F p + q F_{p+q} F p + q that sends y j y_j y j to r j y j r_jy_j r j y j and fixes the other generators is an automorphism, with inverse given by y j ↦ r j − 1 y j y_j \mapsto r_j^{-1}y_j y j ↦ r j − 1 y j ; hence r j y j r_jy_j r j y j is primitive as well.
To compare the relators, consider the homomorphism F p + q → Z q F_{p+q} \to\mathbb{Z}^q F p + q → Z q that sends y j y_j y j to the j j j -th standard basis vector e j e_j e j and every x i x_i x i to zero. It maps both r j y j r_jy_j r j y j and y j y_j y j to e j e_j e j . Since conjugate elements have the same image, relators with different indices are not conjugate, and no relator is conjugate to the inverse of a relator. It remains to show that r j y j r_jy_j r j y j is not conjugate to y j y_j y j . The retraction F p + q → F p F_{p+q} \to F_p F p + q → F p that kills all y i y_i y i maps these two elements to r j r_j r j and to 1 1 1 , and r j r_j r j is nontrivial in F p F_p F p .
We next represent such classes by embedded closed geodesics. Recall that a closed geodesic is prime if it is not an iterate of a shorter closed geodesic, and that two closed geodesics are geometrically distinct if their images differ.
Lemma 3.2. Let ( M , g ) (M,g) ( M , g ) be a closed connected Riemannian manifold of dimension at least three, and fix finitely many nontrivial free homotopy classes of loops in M M M that are not proper powers and are pairwise distinct up to reversal. Then there is a smooth metric, arbitrarily close to g g g in the C 2 C^2 C 2 topology, for which the given classes are represented by pairwise disjoint embedded closed geodesics. This metric can be chosen to agree with g g g outside finitely many balls of arbitrarily small radius. If M M M has dimension four and k g > 0 k_g>0 k g > 0 , the new metric can be chosen to have positive biorthogonal curvature.
Proof. Since M M M is compact, every nontrivial free homotopy class contains a closed geodesic of minimal length. Let c 1 , … , c m c_1,\ldots,c_m c 1 , … , c m be such minimizers for the given classes. Each c i c_i c i is prime, because a d d d -fold iterate with d ≥ 2 d\ge2 d ≥ 2 represents a d d d -th power. The c i c_i c i are also geometrically distinct, because two prime closed geodesics with the same image differ only by a shift or a reversal of the parameter, and they would then represent classes that agree up to reversal.
Let x x x be a double point of some c i c_i c i or a common point of two different c i c_i c i . The tangent lines at x x x of the arcs of the c i c_i c i through x x x are pairwise different, since two arcs with the same tangent line would coincide by the uniqueness of geodesics, and this would contradict the fact that the c i c_i c i are prime and geometrically distinct. Hence such points are isolated, and by compactness there are only finitely many of them; see also [27 Lemma 2] . For small η > 0 \eta>0 η > 0 , the ball of radius 2 η 2\eta 2 η around such a point x x x meets the union of the c i c_i c i in finitely many geodesic arcs through x x x only. As dim M ≥ 3 \dim M\ge3 dim M ≥ 3 , we can apply [27 Lemma 3] to these arcs. In every neighborhood of g g g in the C 2 C^2 C 2 topology it provides a metric that agrees with g g g outside a thin shell inside the ball, together with pairwise disjoint geodesic arcs of the new metric that coincide with the original arcs near the boundary of the ball. The new metric is smooth, because it is obtained from g g g by the construction of [27 Lemma 1] , which uses only smooth diffeomorphisms and smooth cutoff functions.
We choose the balls around the various points x x x to be pairwise disjoint and carry out this modification in each of them. Together with the unchanged parts of the c i c_i c i , the new arcs form closed geodesics of the new metric. These are embedded and pairwise disjoint, and each of them is freely homotopic to the geodesic c i c_i c i that it replaces, because the two differ only inside balls. Finally, if dim M = 4 \dim M=4 dim M = 4 and k g > 0 k_g>0 k g > 0 , then a sufficiently small perturbation in the C 2 C^2 C 2 topology preserves positive biorthogonal curvature by Lemma 2.3.
Remark 3.3. For a C r C^r C r -generic metric with r ≥ 2 r\ge2 r ≥ 2 , all prime closed geodesics are simple and geometrically distinct closed geodesics are disjoint [27 Theorem 1] . We only need the finite version above, in which the perturbation stays close to a given smooth metric of positive biorthogonal curvature.
Replacing a metric by one with the same first jet
We say that two metrics have the same first jet along a submanifold N N N if, in local coordinates, their coefficients and the first partial derivatives of their coefficients agree at every point of N N N . This does not depend on the choice of coordinates.
Lemma 3.4. Let N N N be a compact embedded submanifold without boundary of a smooth four-manifold M M M , and let U ⊂ M U\subset M U ⊂ M be an open neighborhood of N N N . Let g 0 g_0 g 0 and g 1 g_1 g 1 be smooth Riemannian
metrics on U U U that have the same first jet along N N N , and suppose that k g 0 > 0 k_{g_0} > 0 k g 0 > 0 on U U U and k g 1 > 0 k_{g_1} > 0 k g 1 > 0 at the points of N N N . Then for every open neighborhood V ⊂ U V \subset U V ⊂ U of N N N there is a smooth path g ( s ) g(s) g ( s ) , 0 ≤ s ≤ 1 0 \le s \le1 0 ≤ s ≤ 1 , of Riemannian metrics on U U U with the following properties.
(i) g ( 0 ) = g 0 g(0) = g_0 g ( 0 ) = g 0 , and g ( s ) = g 0 g(s) = g_0 g ( s ) = g 0 outside V V V for all s s s .
(ii) g ( 1 ) = g 1 g(1) = g_1 g ( 1 ) = g 1 on a neighborhood of N N N .
(iii) Every g ( s ) g(s) g ( s ) has positive biorthogonal curvature and has the same first jet as g 0 g_0 g 0 along N N N .
In particular, if g 0 g_0 g 0 is defined on all of M M M and k g 0 > 0 k_{g_0} > 0 k g 0 > 0 , the path extends to M M M by g 0 g_0 g 0 outside U U U .
Proof. For s ∈ [ 0 , 1 ] s \in[0,1] s ∈ [ 0 , 1 ] put g s = ( 1 − s ) g 0 + s g 1 g_s = (1-s)g_0 + sg_1 g s = ( 1 − s ) g 0 + s g 1 . In local coordinates, the components of the curvature tensor Rm g \operatorname{Rm}_g Rm g of type ( 0 , 4 ) (0,4) ( 0 , 4 ) are the sum of a linear expression in the second derivatives of the metric, with constant coefficients, and of a term that depends only on the metric and its first derivatives. Since all the metrics g s g_s g s have the same first jet along N N N , we obtain
Rm g s ∣ N = ( 1 − s ) Rm g 0 ∣ N + s Rm g 1 ∣ N . (7) \left.\operatorname{Rm}_{g_s}\right|_N = (1-s)\left.\operatorname{Rm}_{g_0}\right|_N + s\left.\operatorname{Rm}_{g_1}\right|_N. \tag*{(7)} Rm g s ∣ N = ( 1 − s ) Rm g 0 ∣ N + s Rm g 1 ∣ N . (7)
The metrics g s g_s g s induce the same inner product on T M ∣ N TM|_N T M ∣ N , so Lemma 2.3 applies and shows that k g s > 0 k_{g_s} > 0 k g s > 0 at the points of N N N for every s s s . Since N × [ 0 , 1 ] N \times[0,1] N × [ 0 , 1 ] is compact, there is an open neighborhood W W W of N N N with compact closure contained in V V V such that k g s > 0 k_{g_s} > 0 k g s > 0 on W W W for every s s s .
We now apply the local flexibility theorem of Bär and Hanke [1 Theorem 1.2] . In their notation, the ambient manifold is U U U , the closed subset is N N N , and its open neighborhood is W W W . The fiber bundle is the bundle Sym 2 T ∗ U \operatorname{Sym}^2 T^*U Sym 2 T ∗ U of symmetric bilinear forms, and the partial differential relation is the subset of its second jet bundle that consists of the jets of positive definite forms with positive biorthogonal curvature; it is open, because the curvature depends continuously on the two-jet of the metric. The section g 0 g_0 g 0 solves this relation on U U U , the path s ↦ g s ∣ W s \mapsto g_s|_W s ↦ g s ∣ W consists of solutions on W W W and starts at g 0 ∣ W g_0|_W g 0 ∣ W , and the one-jets of g s g_s g s along N N N do not depend on s s s . Since g 0 g_0 g 0 is smooth and the path s ↦ g s ∣ W s \mapsto g_s|_W s ↦ g s ∣ W is smooth, the regularity statement that is part of [1 Theorem 1.2] applies. The theorem therefore gives a smooth path g ( s ) g(s) g ( s ) of solutions on U U U with g ( 0 ) = g 0 g(0) = g_0 g ( 0 ) = g 0 that agrees with g s g_s g s on a neighborhood of N N N independent of s s s and with g 0 g_0 g 0 outside W W W . This path has all the required properties. □
Round tubes and normal holonomy
We now apply Lemma 3.4 to an embedded closed geodesic. The model metric is the product of a line with a round three-ball, written in Fermi coordinates with respect to a parallel normal frame. Such a frame need not close up after one turn around the geodesic; the resulting rotation is the normal holonomy, and it becomes part of the model.
Lemma 3.5. Let ( M 4 , g ) (M^4,g) ( M 4 , g ) be an oriented Riemannian four-manifold with k g > 0 k_g > 0 k g > 0 , let γ \gamma γ be an embedded closed geodesic of length L L L , and let K > 0 K > 0 K > 0 . Then there are a number ρ 0 > 0 \rho_0 > 0 ρ 0 > 0 with K ρ 0 < π / 2 \sqrt{K}\rho_0 < \pi/2 K ρ 0 < π /2 and a smooth path of metrics of positive biorthogonal curvature that starts at g g g , is constant outside an arbitrarily small neighborhood of γ \gamma γ , and ends at a metric that near γ \gamma γ is isometric to
d t 2 + d ρ 2 + s K ( ρ ) 2 g S 2 , s K ( ρ ) = sin ( K ρ ) K , (8) dt^2 + d\rho^2 + s_K(\rho)^2g_{S^2}, \qquad s_K(\rho) = \frac{\sin(\sqrt{K}\rho)}{\sqrt{K}}, \tag*{(8)} d t 2 + d ρ 2 + s K ( ρ ) 2 g S 2 , s K ( ρ ) = K sin ( K ρ ) , (8) on the quotient of R × B ρ 0 3 \mathbb{R} \times B^3_{\rho_0} R × B ρ 0 3 by the identification ( t + L , x ) ∼ ( t , H x ) (t+L,x) \sim(t,Hx) ( t + L , x ) ∼ ( t , H x ) . Here B ρ 0 3 ⊂ R 3 B^3_{\rho_0} \subset\mathbb{R}^3 B ρ 0 3 ⊂ R 3 is the ball of radius ρ 0 \rho_0 ρ 0 around the origin, ρ = ∣ x ∣ \rho= |x| ρ = ∣ x ∣ , the metric g S 2 g_{S^2} g S 2 is the round metric of the unit sphere,
and H ∈ S O ( 3 ) H \in\mathrm{SO}(3) H ∈ SO ( 3 ) is the holonomy of the normal bundle of γ \gamma γ with respect to g g g . All metrics of the path have the same first jet along γ \gamma γ ; in particular, γ \gamma γ remains a geodesic of length L L L with normal holonomy H H H . Finitely many pairwise disjoint embedded closed geodesics can be treated simultaneously.
Proof. Step 1. Fermi coordinates. We parametrize γ \gamma γ by arc length on R \mathbb{R} R , so that γ \gamma γ is L L L -periodic, and we choose an oriented orthonormal frame E 1 ( t ) , E 2 ( t ) , E 3 ( t ) E_1(t), E_2(t), E_3(t) E 1 ( t ) , E 2 ( t ) , E 3 ( t ) of the normal bundle that is parallel along γ \gamma γ . Then E ( t + L ) = E ( t ) H E(t+L)=E(t)H E ( t + L ) = E ( t ) H for a matrix H ∈ S O ( 3 ) H\in\mathrm{SO}(3) H ∈ SO ( 3 ) , where E ( t ) E(t) E ( t ) stands for the row ( E 1 ( t ) , E 2 ( t ) , E 3 ( t ) ) (E_1(t), E_2(t), E_3(t)) ( E 1 ( t ) , E 2 ( t ) , E 3 ( t )) . For x = ( x 1 , x 2 , x 3 ) ∈ R 3 x=(x^1,x^2,x^3)\in\mathbb{R}^3 x = ( x 1 , x 2 , x 3 ) ∈ R 3 small we put
Φ ( t , x ) = exp γ ( t ) ( ∑ a = 1 3 x a E a ( t ) ) . \Phi(t,x)=\exp_{\gamma(t)}\left(\sum_{a=1}^{3}x^aE_a(t)\right). Φ ( t , x ) = exp γ ( t ) ( a = 1 ∑ 3 x a E a ( t ) ) .
Then Φ ( t + L , x ) = Φ ( t , H x ) \Phi(t+L,x)=\Phi(t,Hx) Φ ( t + L , x ) = Φ ( t , H x ) , and since γ \gamma γ is compact and embedded, Φ \Phi Φ induces a diffeomorphism from the quotient of R × B ρ 0 3 \mathbb{R}\times B_{\rho_0}^3 R × B ρ 0 3 by ( t + L , x ) ∼ ( t , H x ) (t+L,x)\sim(t,Hx) ( t + L , x ) ∼ ( t , H x ) onto a tubular neighborhood of γ \gamma γ , provided ρ 0 \rho_0 ρ 0 is small.
Step 2. The first jet of g g g . We claim that in the coordinates ( t , x ) (t,x) ( t , x ) the coefficients of g g g satisfy
g i j ( t , 0 ) = δ i j , ∂ k g i j ( t , 0 ) = 0 (9) g_{ij}(t,0)=\delta_{ij},\qquad\partial_k g_{ij}(t,0)=0 \tag*{(9)} g ij ( t , 0 ) = δ ij , ∂ k g ij ( t , 0 ) = 0 (9)
for all indices. Along x = 0 x=0 x = 0 the coordinate vector fields are γ ˙ , E 1 , E 2 , E 3 \dot{\gamma}, E_1, E_2, E_3 γ ˙ , E 1 , E 2 , E 3 , which are orthonormal. Their covariant derivatives in the direction of ∂ t \partial_t ∂ t vanish there, because γ \gamma γ is a geodesic and the E a E_a E a are parallel; since coordinate vector fields commute, we also have ∇ ∂ a ∂ t = ∇ ∂ t ∂ a = 0 \nabla_{\partial_a}\partial_t=\nabla_{\partial_t}\partial_a=0 ∇ ∂ a ∂ t = ∇ ∂ t ∂ a = 0 along x = 0 x=0 x = 0 . Finally, for fixed t t t and v ∈ R 3 v\in\mathbb{R}^3 v ∈ R 3 the curve s ↦ Φ ( t , s v ) s\mapsto\Phi(t,sv) s ↦ Φ ( t , s v ) is a geodesic, which gives ∑ a , b v a v b ∇ ∂ a ∂ b = 0 \sum_{a,b}v^av^b\nabla_{\partial_a}\partial_b=0 ∑ a , b v a v b ∇ ∂ a ∂ b = 0 at ( t , 0 ) (t,0) ( t , 0 ) for all v v v and hence ∇ ∂ a ∂ b = 0 \nabla_{\partial_a}\partial_b=0 ∇ ∂ a ∂ b = 0 there. Thus all Christoffel symbols vanish along γ \gamma γ , and (9) follows from the compatibility of ∇ \nabla ∇ with the metric.
Step 3. The model. Let g 1 g_1 g 1 be the metric (8) . It is the product of the line with a geodesic ball in the three-sphere of curvature K K K , written in normal coordinates, and it descends to the quotient because the round ball is invariant under S O ( 3 ) \mathrm{SO}(3) SO ( 3 ) . Since s K ( ρ ) = ρ − K 6 ρ 3 + O ( ρ 5 ) s_K(\rho)=\rho-\frac{K}{6}\rho^3+O(\rho^5) s K ( ρ ) = ρ − 6 K ρ 3 + O ( ρ 5 ) , the metric g 1 g_1 g 1 is smooth along the core x = 0 x=0 x = 0 and also satisfies (9) . Hence g g g and g 1 g_1 g 1 have the same first jet along γ \gamma γ .
The biorthogonal curvature of g 1 g_1 g 1 is constant. Indeed, let e 0 = ∂ t e_0=\partial_t e 0 = ∂ t . For a two-plane P P P we have K g 1 ( P ) = K ( 1 − ∣ proj P e 0 ∣ 2 ) K_{g_1}(P)=K(1-\lvert\operatorname{proj}_P e_0\rvert^2) K g 1 ( P ) = K ( 1 − ∣ proj P e 0 ∣ 2 ) , where proj P \operatorname{proj}_P proj P is the orthogonal projection onto P P P ; this is seen by choosing an orthonormal basis of P P P whose first vector is orthogonal to e 0 e_0 e 0 . Since the squared lengths of the projections of e 0 e_0 e 0 to P P P and to P ⊥ P^\perp P ⊥ add up to one, we get
K g 1 ( P ) + K g 1 ( P ⊥ ) = K K_{g_1}(P)+K_{g_1}(P^\perp)=K K g 1 ( P ) + K g 1 ( P ⊥ ) = K
and hence
k g 1 = K 2 > 0. k_{g_1}=\frac{K}{2}>0. k g 1 = 2 K > 0.
Step 4. Conclusion. We apply Lemma 3.4 to g g g and g 1 g_1 g 1 , with N = γ N=\gamma N = γ and with V V V a small tube around γ \gamma γ . The final metric of the resulting path agrees with g 1 g_1 g 1 on a neighborhood of γ \gamma γ , and after decreasing ρ 0 \rho_0 ρ 0 we may assume that this neighborhood contains the tube of radius ρ 0 \rho_0 ρ 0 and that K ρ 0 < π / 2 \sqrt{K}\rho_0<\pi/2 K ρ 0 < π /2 . The path has all the properties stated. Indeed, since the metric and its first derivatives along γ \gamma γ do not change, neither do the Levi-Civita connection along γ \gamma γ , the length of γ \gamma γ , and the parallel transport in its normal bundle. For finitely many disjoint geodesics we use disjoint tubes.
Collar coordinates and normal framings
We now rewrite the model (8) in coordinates that are periodic along γ \gamma γ . This fixes the form of the collar to which the handle has to be matched, and our sign conventions for the holonomy.
We identify R 3 \mathbb{R}^{3} R 3 with R ⊕ C \mathbb{R} \oplus\mathbb{C} R ⊕ C and write the unit sphere as
S 2 = { ( z , w ) ∈ R ⊕ C ∣ z 2 + ∣ w ∣ 2 = 1 } . S^{2} = \{(z,w) \in\mathbb{R} \oplus\mathbb{C} \mid z^{2} + \lvert w\rvert^{2} = 1\}. S 2 = {( z , w ) ∈ R ⊕ C ∣ z 2 + ∣ w ∣ 2 = 1 } .
After a constant rotation of the parallel frame E E E , the holonomy is a rotation about the z z z -axis, and we can choose α ∈ R \alpha\in\mathbb{R} α ∈ R with
H ( z , w ) = ( z , e − 2 π i α w ) . (10) H(z,w) = (z,e^{-2\pi i\alpha}w). \tag*{(10)} H ( z , w ) = ( z , e − 2 π i α w ) . (10)
The number α \alpha α is determined up to an integer. Put F = L / ( 2 π ) F = L/(2\pi) F = L / ( 2 π ) and θ = t / F \theta= t/F θ = t / F , so that θ \theta θ has period 2 π 2\pi 2 π , and consider the map Ψ \Psi Ψ from S 1 × ( 0 , ρ 0 ) × S 2 S^{1} \times(0,\rho_{0}) \times S^{2} S 1 × ( 0 , ρ 0 ) × S 2 to the quotient ( R × B ρ 0 3 ) / ∼ (\mathbb{R} \times B^{3}_{\rho_{0}})/\sim ( R × B ρ 0 3 ) / ∼ that is given by
Ψ ( θ , ρ , ( z , w ) ) = [ t , ρ ( z , e i α θ w ) ] . \Psi(\theta,\rho,(z,w)) = [t,\rho(z,e^{i\alpha\theta}w)]. Ψ ( θ , ρ , ( z , w )) = [ t , ρ ( z , e i α θ w )] .
It is well defined, because increasing θ \theta θ by 2 π 2\pi 2 π is compensated by the identification with H H H , and it is a diffeomorphism onto the complement of the core. In these coordinates the metric (8) becomes
F 2 d θ 2 + d ρ 2 + s K ( ρ ) 2 ( d z 2 + ∣ d w + i α w d θ ∣ 2 ) . (11) F^{2}d\theta^{2} + d\rho^{2} + s_{K}(\rho)^{2}\left(dz^{2} + \lvert dw + i\alpha w\,d\theta\rvert^{2}\right). \tag*{(11)} F 2 d θ 2 + d ρ 2 + s K ( ρ ) 2 ( d z 2 + ∣ d w + i α w d θ ∣ 2 ) . (11)
In terms of the colatitude ϑ ∈ [ 0 , π ] \vartheta\in[0,\pi] ϑ ∈ [ 0 , π ] and the longitude ϕ ∈ R / 2 π Z \phi\in\mathbb{R}/2\pi\mathbb{Z} ϕ ∈ R /2 π Z , defined by z = cos ϑ z = \cos\vartheta z = cos ϑ and w = sin ϑ e i ϕ w = \sin\vartheta\,e^{i\phi} w = sin ϑ e i ϕ , the last bracket equals d ϑ 2 + sin 2 ϑ ( d ϕ + α d θ ) 2 d\vartheta^{2} + \sin^{2}\vartheta\,(d\phi+ \alpha\,d\theta)^{2} d ϑ 2 + sin 2 ϑ ( d ϕ + α d θ ) 2 . We finally use h = s K ( ρ ) h = s_{K}(\rho) h = s K ( ρ ) as a radial coordinate. Since K ρ 0 < π / 2 \sqrt{K}\rho_{0} < \pi/2 K ρ 0 < π /2 , we have d h / d ρ = cos ( K ρ ) > 0 dh/d\rho= \cos(\sqrt{K}\rho) > 0 d h / d ρ = cos ( K ρ ) > 0 on ( 0 , ρ 0 ) (0,\rho_{0}) ( 0 , ρ 0 ) , and we arrive at the following normal form.
Lemma 3.6. In the situation of Lemma Lemma 3.5 , let 2 π F 2\pi F 2 π F be the length of γ \gamma γ and choose α \alpha α as in (10) . Then, with h 1 = s K ( ρ 0 ) h_{1} = s_{K}(\rho_{0}) h 1 = s K ( ρ 0 ) , the complement of γ \gamma γ in a tubular neighborhood has coordinates ( θ , h , ( z , w ) ) ∈ S 1 × ( 0 , h 1 ) × S 2 (\theta,h,(z,w)) \in S^{1} \times(0,h_{1}) \times S^{2} ( θ , h , ( z , w )) ∈ S 1 × ( 0 , h 1 ) × S 2 in which the deformed metric is
F 2 d θ 2 + d h 2 1 − K h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + α d θ ) 2 } . (12) F^{2}d\theta^{2} + \frac{dh^{2}}{1-Kh^{2}} + h^{2}\{d\vartheta^{2} + \sin^{2}\vartheta\,(d\phi+ \alpha\,d\theta)^{2}\}. \tag*{(12)} F 2 d θ 2 + 1 − K h 2 d h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + α d θ ) 2 } . (12) The coordinate h h h tends to zero at γ \gamma γ .
The connection term α d θ \alpha\,d\theta α d θ can be removed locally by rotating the sphere, but not globally unless α \alpha α is an integer, because its holonomy around γ \gamma γ is the normal holonomy of the geodesic.
The coordinates of Lemma 3.6 also determine a normal framing of γ \gamma γ , namely the homotopy class of the trivialization of the normal bundle by the periodic frame E ( t ) E(t) E ( t ) rotated by the angle α θ \alpha\theta α θ about the first axis. The choice of α \alpha α within its class modulo integers matters here. For an integer n n n , the diffeomorphism
T n ( θ , z , w ) = ( θ , z , e i n θ w ) (13) T_{n}(\theta,z,w) = (\theta,z,e^{in\theta}w) \tag*{(13)} T n ( θ , z , w ) = ( θ , z , e in θ w ) (13)
of S 1 × S 2 S^{1} \times S^{2} S 1 × S 2 pulls the metric (12) with coefficient α \alpha α back to the same expression with coefficient α + n \alpha+n α + n , since T n ∗ ( d w + i α w d θ ) = e i n θ ( d w + i ( α + n ) w d θ ) T_{n}^{*}(dw+i\alpha w\,d\theta) = e^{in\theta}(dw+i(\alpha+n)w\,d\theta) T n ∗ ( d w + i α w d θ ) = e in θ ( d w + i ( α + n ) w d θ ) . The loop θ ↦ \theta\mapsto θ ↦ (rotation by n θ n\theta n θ ) represents the class of n n n modulo two in π 1 ( SO ( 3 ) ) ≅ Z / 2 \pi_{1}(\operatorname{SO}(3)) \cong\mathbb{Z}/2 π 1 ( SO ( 3 )) ≅ Z /2 . Hence replacing α \alpha α by α + n \alpha+n α + n does not change the holonomy, and it changes the framing if and only if n n n is odd. Normal holonomy and normal framing are thus independent data. The former is the class of α \alpha α modulo integers, and the latter is a parity.
Remark 3.7. Let 0 < R < h 1 0<R<h_1 0 < R < h 1 and remove the open tube { h < R } \{h<R\} { h < R } around γ \gamma γ . Suppose that we are given a metric on D 2 × S 2 D^2\times S^2 D 2 × S 2 whose boundary collar is parametrized by R − < h ≤ R R_-<h\le R R − < h ≤ R and has the form (12) with coefficient α + n \alpha+n α + n instead of α \alpha α . The map T n T_n T n identifies this collar isometrically with the region { R − < h ≤ R } \{R_-<h\le R\} { R − < h ≤ R } of the tube. We can therefore glue D 2 × S 2 D^2\times S^2 D 2 × S 2 to the complement of the tube along { h = R } \{h=R\} { h = R } , and the resulting metric is smooth, because on both sides of the seam it is given by the same formula in the coordinate h h h , which increases from the handle into the exterior. The resulting manifold is the result of surgery along γ \gamma γ , with the framing determined by the coordinates if n n n is even and with the other framing if n n n is odd.
The effect of circle surgery on the fundamental group
For elements a 1 , … , a m a_1,\ldots,a_m a 1 , … , a m of a group we write ⟨ ⟨ a 1 , … , a m ⟩ ⟩ \langle\!\langle a_1,\ldots,a_m\rangle\!\rangle ⟨ ⟨ a 1 , … , a m ⟩ ⟩ for the normal subgroup that they generate. It depends only on the conjugacy classes of the a j a_j a j , so the notation makes sense for free homotopy classes of loops.
Lemma 3.8. Let M M M be a closed connected oriented smooth four-manifold, and let γ 1 , … , γ m \gamma_1,\ldots,\gamma_m γ 1 , … , γ m be pairwise disjoint embedded circles in M M M . For any choice of normal framings, the manifold M ′ M' M ′ obtained from M M M by surgery along γ 1 , … , γ m \gamma_1,\ldots,\gamma_m γ 1 , … , γ m is a closed connected oriented smooth four-manifold with
π 1 ( M ′ ) ≅ π 1 ( M ) / ⟨ ⟨ [ γ 1 ] , … , [ γ m ] ⟩ ⟩ . (14) \pi_1(M')\cong\pi_1(M)/\langle\!\langle[\gamma_1],\ldots,[\gamma_m]\rangle\!\rangle. \tag*{(14)} π 1 ( M ′ ) ≅ π 1 ( M ) / ⟨ ⟨[ γ 1 ] , … , [ γ m ]⟩ ⟩ . (14) More precisely, if W W W denotes the trace of the surgeries, a compact oriented five-manifold with boundary M ′ ⊔ − M M'\sqcup-M M ′ ⊔ − M , then the inclusion M ′ ⊂ W M'\subset W M ′ ⊂ W induces an isomorphism of fundamental groups and the inclusion M ⊂ W M\subset W M ⊂ W induces the quotient map in (14) .
Proof. The trace W W W is obtained from M × [ 0 , 1 ] M\times[0,1] M × [ 0 , 1 ] by attaching one two-handle D 2 × D 3 D^2\times D^3 D 2 × D 3 along a tubular neighborhood of each γ j × { 1 } \gamma_j\times\{1\} γ j × { 1 } . Seen from M ′ M' M ′ , it is obtained from M ′ × [ 0 , 1 ] M'\times[0,1] M ′ × [ 0 , 1 ] by attaching handles of index three. Up to homotopy, attaching a two-handle amounts to attaching a two-cell along the corresponding circle, and attaching a three-handle to attaching a three-cell. The assertions therefore follow from [13 Proposition 1.26] . □
A handle with positive biorthogonal curvature
In this section we construct the metric on D 2 × S 2 D^2\times S^2 D 2 × S 2 that replaces a round tube. Its boundary collar is the collar (12) of a closed geodesic of length 2 π F 2\pi F 2 π F whose normal holonomy is given by an arbitrary real number α \alpha α . The integer n n n in the statement is produced by the construction, but we can prescribe its parity, and hence the framing.
Theorem 4.1. Let F , K > 0 F,K>0 F , K > 0 , let R > 0 R>0 R > 0 satisfy K R 2 ≤ 1 / 100 KR^2\le1/100 K R 2 ≤ 1/100 , let α ∈ R \alpha\in\mathbb{R} α ∈ R , and let ν ∈ { 0 , 1 } \nu\in\{0,1\} ν ∈ { 0 , 1 } . Then there exist an integer n ≡ ν ( m o d 2 ) n\equiv\nu\pmod2 n ≡ ν ( mod 2 ) , a number R − ∈ ( 0 , R ) R_-\in(0,R) R − ∈ ( 0 , R ) , and a smooth Riemannian metric g ^ \hat{g} g ^ on D 2 × S 2 D^2\times S^2 D 2 × S 2 with k g ^ > 0 k_{\hat{g}}>0 k g ^ > 0 whose boundary has a collar parametrized by ( θ , h , ( z , w ) ) ∈ S 1 × ( R − , R ] × S 2 (\theta,h,(z,w))\in S^1\times(R_-,R]\times S^2 ( θ , h , ( z , w )) ∈ S 1 × ( R − , R ] × S 2 on which
g ^ = F 2 d θ 2 + d h 2 1 − K h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + ( α + n ) d θ ) 2 } . (15) \hat{g}=F^2d\theta^2+\frac{dh^2}{1-Kh^2}+h^2\left\{d\vartheta^2+\sin^2\vartheta\left(d\phi+(\alpha+n)d\theta\right)^2\right\}. \tag*{(15)} g ^ = F 2 d θ 2 + 1 − K h 2 d h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + ( α + n ) d θ ) 2 } . (15) Here θ \theta θ has period 2 π 2\pi 2 π , the boundary is { h = R } \{h=R\} { h = R } , and ( ϑ , ϕ ) (\vartheta,\phi) ( ϑ , ϕ ) are the spherical coordinates on S 2 = { ( z , w ) ∈ R ⊕ C ∣ ∣ z ∣ 2 + ∣ w ∣ 2 = 1 } S^2=\{(z,w)\in\mathbb{R}\oplus\mathbb{C}\mid|z|^2+|w|^2=1\} S 2 = {( z , w ) ∈ R ⊕ C ∣ ∣ z ∣ 2 + ∣ w ∣ 2 = 1 } given by z = cos ϑ z=\cos\vartheta z = cos ϑ and w = sin ϑ e i ϕ w=\sin\vartheta\,e^{i\phi} w = sin ϑ e i ϕ .
By Remark 3.7, the metric g ^ \hat{g} g ^ can be glued to the complement of the tube { h < R } \{h<R\} { h < R } in Lemma 3.6 by means of the map T n T_n T n of (13) , and the two values of ν \nu ν give the two normal framings. The metric that we construct is invariant under the rotations of the disk and under the rotations of the sphere about the z z z -axis, and the integer n n n depends only on F F F , K K K , R R R , α \alpha α , ν \nu ν and on the choice of a cutoff function. The rest of this section contains the proof of Theorem 4.1 . We begin with an overview.
Let r r r and θ \theta θ be polar coordinates on the disk. The simplest candidate for a metric on D 2 × S 2 D^2 \times S^2 D 2 × S 2 is the warped product
d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 g S 2 . \mathrm{d}r^2+f(r)^2\mathrm{d}\theta^2+h(r)^2g_{S^2}. d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 g S 2 .
in which f f f is the radius of the θ \theta θ -circle and h h h is the radius of the sphere. Smoothness at the center of the disk forces f ( 0 ) = 0 f(0)=0 f ( 0 ) = 0 , f ′ ( 0 ) = 1 f'(0)=1 f ′ ( 0 ) = 1 and h ′ ( 0 ) = 0 h'(0)=0 h ′ ( 0 ) = 0 . On the collar (15) , on the other hand, f = F f=F f = F and h ′ 2 = 1 − K h 2 h'^2=1-Kh^2 h ′2 = 1 − K h 2 is close to one. For the warped product this is fatal. We shall see in (22) that its biorthogonal curvature satisfies k ≤ − ( f h ′ ′ ) / ( 2 f h ) k\leq-(fh'')/(2fh) k ≤ − ( f h ′′ ) / ( 2 f h ) . Since f h ′ fh' f h ′ vanishes at r = 0 r=0 r = 0 and is positive on the collar, the right-hand side is negative somewhere. The planes responsible for this are the ones spanned by the radial direction and a direction tangent to the sphere, together with their complements, and they are the planes that are flat in the model S 2 × R 2 S^2\times\mathbb{R}^2 S 2 × R 2 . In our construction f f f will be constant outside a neighborhood of the center, and the relevant quantity is the defect
δ = 1 − h ′ 2 . \delta=1-h'^2. δ = 1 − h ′2 .
It has to decrease from 1 to a small number, and so h h h has to be convex on some interval. Our construction remedies the resulting loss of positivity by means of three devices.
A rotational connection. We let the sphere rotate about its axis as θ \theta θ varies, and we consider the metric (16) below. The function κ = h 2 A ′ / f \kappa=h^2A'/f κ = h 2 A ′ / f is a normalized curvature of this connection. By Lemma 4.3 , it adds the term κ 2 ( 1 − z 2 ) / ( 8 h 2 ) \kappa^2(1-z^2)/(8h^2) κ 2 ( 1 − z 2 ) / ( 8 h 2 ) to the biorthogonal curvature, where z = cos ϑ z=\cos\vartheta z = cos ϑ is the height function on the sphere. This term is positive except at the poles z = ± 1 z=\pm1 z = ± 1 , which are the fixed points of the rotation.
A conformal change. Since k g k_g k g transforms under conformal changes like a scalar curvature (Lemma 2.2 ), we can use a conformal factor to move positivity from the equator to the poles. The factor u = 1 + κ 2 z 2 / 24 u=1+\kappa^2z^2/24 u = 1 + κ 2 z 2 /24 is chosen in such a way that the dependence on z z z cancels in the leading terms. After the conformal change, the connection contributes κ 2 / ( 12 h 2 ) \kappa^2/(12h^2) κ 2 / ( 12 h 2 ) uniformly on the sphere (Lemma 4.4).
Slow growth. The convexity of h h h costs the term − u h ′ ′ / ( 2 h ) -u h''/(2h) − u h ′′ / ( 2 h ) , and the gain is κ 2 / ( 12 h 2 ) \kappa^2/(12h^2) κ 2 / ( 12 h 2 ) . We cannot take κ \kappa κ large, because the eigenvalue structure that we use requires, roughly, 3 κ < δ 3\kappa<\delta 3 κ < δ . We take κ = δ / 6 \kappa=\delta/6 κ = δ /6 , and positivity then holds as long as h h ′ ′ ≤ ϵ δ 2 / 2 hh''\leq\epsilon\delta^2/2 h h ′′ ≤ ϵ δ 2 /2 for a small constant ϵ \epsilon ϵ . In terms of the logarithmic radius t = log h t=\log h t = log h this means d δ / d t = − ϵ δ 2 \mathrm{d}\delta/\mathrm{d}t=-\epsilon\delta^2 d δ / d t = − ϵ δ 2 , so that 1 / δ 1/\delta 1/ δ grows only linearly in t t t , and the sphere has to grow by a factor of about exp ( 1 / ( ϵ δ ) ) \exp(1/(\epsilon\delta)) exp ( 1/ ( ϵδ )) before the defect has come down to δ \delta δ . This is why the neck of our handle is very long. It is finite, however, and no limiting argument is needed.
Once the defect is small, we let it increase again. Now h h h is concave, which is favorable, and we can switch off the connection and steer the defect to the value K h 2 Kh^2 K h 2 of the round collar. The total rotation ∫ A ′ d r \int A'\,\mathrm{d}r ∫ A ′ d r that has accumulated along the way is then corrected by inserting a cylinder of constant radius near the center, on which the connection has constant curvature; its length is chosen so that the boundary coefficient becomes α + n \alpha+n α + n with n n n of the prescribed parity.
The pieces of the handle are listed in Table 1 and sketched in Figure 1 and Figure 2 .
Figure 1. The radial profiles of the handle (schematic and not to scale). The solid curve is the radius f f f of the θ \theta θ -circle, the dashed curve is the radius h h h of the sphere. The disk closes up on the cap; the cylinder has adjustable length and is used to tune the holonomy; the sphere then grows from h 0 h_0 h 0 to R R R along the ramp, the neck and Stages I–III. In reality h 0 h_0 h 0 is smaller than R R R by many orders of magnitude, and h ′ h' h ′ is very close to one from the middle of the neck on.
Table 1. The pieces of the handle. Here t = log h t=\log h t = log h , a dot denotes d / d t d/dt d / d t , and η \eta η , χ \chi χ , ψ \psi ψ are cutoff functions; the precise definitions are given in Subsection 4.4, and the lower bounds are proved in Subsection 4.5.
Figure 2. The defect δ = 1 − h ′ 2 \delta= 1-h'^2 δ = 1 − h ′2 (solid) and the normalized connection curvature κ \kappa κ (dotted) as functions of the logarithmic radius t = log h t = \log h t = log h , on a logarithmic scale (schematic; the marks on the axis indicate that most of the neck has been omitted). The cap, the cylinder and the ramp, where h h h is equal or very close to h 0 h_0 h 0 , lie to the left of the picture.
The twisted metric and its curvature operator
Let r r r be a radial coordinate, and consider the metric
g = d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( r ) d θ ) 2 } , (16) g = dr^2 + f(r)^2d\theta^2 + h(r)^2\{d\vartheta^2 + \sin^2\vartheta\,(d\phi+ A(r)\,d\theta)^2\}, \tag*{(16)} g = d r 2 + f ( r ) 2 d θ 2 + h ( r ) 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( r ) d θ ) 2 } , (16)
where f f f and h h h are positive smooth functions and A A A is a smooth function of r > 0 r>0 r > 0 ; the coordinate θ \theta θ has period 2 π 2\pi 2 π , and ( ϑ , ϕ ) (\vartheta,\phi) ( ϑ , ϕ ) are the spherical coordinates of Theorem 4.1 . Thus r r r is the distance in the radial direction and h h h is the radius of the sphere fibers. A prime denotes the derivative with respect to r r r . We put
κ = h 2 A ′ f , z = cos ϑ , b = κ sin ϑ h , c = − κ cos ϑ h 2 , \kappa= \frac{h^2A'}{f}, \qquad z = \cos\vartheta, \qquad b = \frac{\kappa\sin\vartheta}{h}, \qquad c = -\frac{\kappa\cos\vartheta}{h^2}, κ = f h 2 A ′ , z = cos ϑ , b = h κ sin ϑ , c = − h 2 κ cos ϑ ,
U = 1 − h ′ 2 h 2 , V = − h ′ ′ h , W = − f ′ h ′ f h , X = − f ′ ′ f . (17) \mathcal{U} = \frac{1-h'^2}{h^2}, \qquad\mathcal{V} = -\frac{h''}{h}, \qquad\mathcal{W} = -\frac{f'h'}{fh}, \qquad\mathcal{X} = -\frac{f''}{f}. \tag*{(17)} U = h 2 1 − h ′2 , V = − h h ′′ , W = − f h f ′ h ′ , X = − f f ′′ . (17)
If A A A is constant, then κ = 0 \kappa=0 κ = 0 and the metric ((16) ) is locally a warped product, and the functions U , V , W , X \mathcal{U},\mathcal{V},\mathcal{W},\mathcal{X} U , V , W , X are its sectional curvatures. Here U \mathcal{U} U is the curvature of the planes tangent to the spheres, X \mathcal{X} X that of the planes tangent to the disk, and V \mathcal{V} V and W \mathcal{W} W are the curvatures of the planes spanned by a direction tangent to the sphere and by ∂ r \partial_r ∂ r or ∂ θ \partial_\theta ∂ θ , respectively. The functions b b b and c c c measure the curvature of the connection. It is more convenient to prescribe κ \kappa κ than A A A .
On the region where r > 0 r > 0 r > 0 and 0 < ϑ < π 0 < \vartheta< \pi 0 < ϑ < π we use the oriented orthonormal coframe
ω 1 = h d ϑ , ω 2 = h sin ϑ ( d ϕ + A d θ ) , ω 3 = d r , ω 4 = f d θ \omega^1 = h\,d\vartheta,\qquad\omega^2 = h\sin\vartheta\,(d\phi+ A\,d\theta),\qquad\omega^3 = dr,\qquad\omega^4 = f\,d\theta ω 1 = h d ϑ , ω 2 = h sin ϑ ( d ϕ + A d θ ) , ω 3 = d r , ω 4 = f d θ
and we write ω i j = ω i ∧ ω j \omega^{ij}=\omega^i\wedge\omega^j ω ij = ω i ∧ ω j .
Lemma 4.2. Let Ω i j = ∑ k < l R i j , k l ω k l \Omega_{ij}=\sum_{k<l}R_{ij,kl}\,\omega^{kl} Ω ij = ∑ k < l R ij , k l ω k l be the curvature forms of the metric (16) with respect to the coframe ω 1 , … , ω 4 \omega^1,\ldots,\omega^4 ω 1 , … , ω 4 , so that the coefficient of ω i j \omega^{ij} ω ij in Ω i j \Omega_{ij} Ω ij is the sectional curvature of the plane spanned by e i e_i e i and e j e_j e j . Then
Ω 12 = U ω 12 + c ω 34 − h ′ b 2 h ω 14 , Ω 13 = V ω 13 + c 2 ω 24 , Ω 14 = W ω 14 − c 2 ω 23 − h ′ b 2 h ω 12 , Ω 23 = ( V + b 2 4 ) ω 23 − c 2 ω 14 + T ω 34 , Ω 24 = ( W + b 2 4 ) ω 24 + c 2 ω 13 , Ω 34 = ( X − 3 b 2 4 ) ω 34 + c ω 12 + T ω 23 , (18) \begin{aligned}
\Omega_{12}&=\mathcal{U}\omega^{12}+c\omega^{34}-\frac{h'b}{2h}\omega^{14},&
\Omega_{13}&=\mathcal{V}\omega^{13}+\frac{c}{2}\omega^{24},\\
\Omega_{14}&=\mathcal{W}\omega^{14}-\frac{c}{2}\omega^{23}-\frac{h'b}{2h}\omega^{12},&
\Omega_{23}&=\left(\mathcal{V}+\frac{b^2}{4}\right)\omega^{23}-\frac{c}{2}\omega^{14}+\mathcal{T}\omega^{34},\\
\Omega_{24}&=\left(\mathcal{W}+\frac{b^2}{4}\right)\omega^{24}+\frac{c}{2}\omega^{13},&
\Omega_{34}&=\left(\mathcal{X}-\frac{3b^2}{4}\right)\omega^{34}+c\omega^{12}+\mathcal{T}\omega^{23}, \tag*{(18)}
\end{aligned} Ω 12 Ω 14 Ω 24 = U ω 12 + c ω 34 − 2 h h ′ b ω 14 , = W ω 14 − 2 c ω 23 − 2 h h ′ b ω 12 , = ( W + 4 b 2 ) ω 24 + 2 c ω 13 , Ω 13 Ω 23 Ω 34 = V ω 13 + 2 c ω 24 , = ( V + 4 b 2 ) ω 23 − 2 c ω 14 + T ω 34 , = ( X − 4 3 b 2 ) ω 34 + c ω 12 + T ω 23 , (18) where T = h ′ h b + b ′ 2 \mathcal{T}=\frac{h'}{h}b+\frac{b'}{2} T = h h ′ b + 2 b ′ and b ′ = ∂ b / ∂ r b'=\partial b/\partial r b ′ = ∂ b / ∂ r .
Proof. We abbreviate
σ = h ′ h , τ = f ′ f , μ = cot ϑ h . \sigma=\frac{h'}{h},\qquad\tau=\frac{f'}{f},\qquad\mu=\frac{\cot\vartheta}{h}. σ = h h ′ , τ = f f ′ , μ = h cot ϑ .
so that c = − μ b c=-\mu b c = − μ b , σ ′ + σ 2 = h ′ ′ / h \sigma'+\sigma^2=h''/h σ ′ + σ 2 = h ′′ / h and τ ′ + τ 2 = f ′ ′ / f \tau'+\tau^2=f''/f τ ′ + τ 2 = f ′′ / f . Differentiating the coframe, we find
d ω 1 = σ ω 31 , d ω 2 = σ ω 32 + μ ω 12 + b ω 34 , d ω 3 = 0 , d ω 4 = τ ω 34 . d\omega^1=\sigma\omega^{31},\qquad d\omega^2=\sigma\omega^{32}+\mu\omega^{12}+b\omega^{34},\qquad d\omega^3=0,\qquad d\omega^4=\tau\omega^{34}. d ω 1 = σ ω 31 , d ω 2 = σ ω 32 + μ ω 12 + b ω 34 , d ω 3 = 0 , d ω 4 = τ ω 34 .
and the coefficients have the differentials
d σ = σ ′ ω 3 , d τ = τ ′ ω 3 , d μ = − ( h − 2 + μ 2 ) ω 1 − σ μ ω 3 , d b = μ b ω 1 + b ′ ω 3 . d\sigma=\sigma'\omega^3,\qquad d\tau=\tau'\omega^3,\qquad d\mu=-(h^{-2}+\mu^2)\omega^1-\sigma\mu\omega^3,\qquad db=\mu b\omega^1+b'\omega^3. d σ = σ ′ ω 3 , d τ = τ ′ ω 3 , d μ = − ( h − 2 + μ 2 ) ω 1 − σ μ ω 3 , d b = μ b ω 1 + b ′ ω 3 .
Let γ i j = − γ j i \gamma_{ij}=-\gamma_{ji} γ ij = − γ j i be the connection forms, characterized by d ω i = − ∑ j γ i j ∧ ω j d\omega^i=-\sum_j\gamma_{ij}\wedge\omega^j d ω i = − ∑ j γ ij ∧ ω j . One checks directly that the structure equations are solved by
γ 12 = − μ ω 2 , γ 13 = σ ω 1 , γ 14 = 0 , \gamma_{12}=-\mu\omega^2,\qquad\gamma_{13}=\sigma\omega^1,\qquad\gamma_{14}=0, γ 12 = − μ ω 2 , γ 13 = σ ω 1 , γ 14 = 0 ,
γ 23 = σ ω 2 + b 2 ω 4 , γ 24 = − b 2 ω 3 , γ 34 = − τ ω 4 − b 2 ω 2 . \gamma_{23}=\sigma\omega^2+\frac{b}{2}\omega^4,\qquad\gamma_{24}=-\frac{b}{2}\omega^3,\qquad\gamma_{34}=-\tau\omega^4-\frac{b}{2}\omega^2. γ 23 = σ ω 2 + 2 b ω 4 , γ 24 = − 2 b ω 3 , γ 34 = − τ ω 4 − 2 b ω 2 .
The curvature forms are Ω i j = d γ i j + ∑ k γ i k ∧ γ k j \Omega_{ij}=d\gamma_{ij}+\sum_k\gamma_{ik}\wedge\gamma_{kj} Ω ij = d γ ij + ∑ k γ ik ∧ γ k j . For the round unit sphere this convention gives Ω 12 = ω 12 \Omega_{12}=\omega^{12} Ω 12 = ω 12 , in accordance with § 2 . We carry out two of the six computations in detail. First,
d γ 12 = − d μ ∧ ω 2 − μ d ω 2 = h − 2 ω 12 − μ b ω 34 , γ 13 ∧ γ 32 = − σ 2 ω 12 − σ b 2 ω 14 , d\gamma_{12}=-d\mu\wedge\omega^2-\mu\,d\omega^2=h^{-2}\omega^{12}-\mu b\omega^{34},\qquad\gamma_{13}\wedge\gamma_{32}=-\sigma^2\omega^{12}-\frac{\sigma b}{2}\omega^{14}, d γ 12 = − d μ ∧ ω 2 − μ d ω 2 = h − 2 ω 12 − μ b ω 34 , γ 13 ∧ γ 32 = − σ 2 ω 12 − 2 σ b ω 14 ,
and γ 14 = 0 \gamma_{14}=0 γ 14 = 0 , which gives the formula for Ω 12 \Omega_{12} Ω 12 because U = h − 2 − σ 2 \mathcal{U}=h^{-2}-\sigma^2 U = h − 2 − σ 2 . Second,
d γ 34 = ( X − b 2 2 ) ω 34 + c ω 12 + b ′ + σ b 2 ω 23 , γ 32 ∧ γ 24 = − b 2 4 ω 34 + σ b 2 ω 23 , d\gamma_{34}=\left(\mathcal{X}-\frac{b^2}{2}\right)\omega^{34}+c\omega^{12}+\frac{b'+\sigma b}{2}\omega^{23},\qquad\gamma_{32}\wedge\gamma_{24}=-\frac{b^2}{4}\omega^{34}+\frac{\sigma b}{2}\omega^{23}, d γ 34 = ( X − 2 b 2 ) ω 34 + c ω 12 + 2 b ′ + σ b ω 23 , γ 32 ∧ γ 24 = − 4 b 2 ω 34 + 2 σ b ω 23 ,
and γ 31 ∧ γ 14 = 0 \gamma_{31} \wedge\gamma_{14}=0 γ 31 ∧ γ 14 = 0 , which gives the formula for Ω 34 \Omega_{34} Ω 34 . The remaining four forms are obtained in the same way from
d γ 13 = V ω 13 , γ 12 ∧ γ 23 = c 2 ω 24 , d γ 14 = 0 , γ 12 ∧ γ 24 + γ 13 ∧ γ 34 = − c 2 ω 23 + W ω 14 − σ b 2 ω 12 , d γ 24 = c 2 ω 13 , γ 23 ∧ γ 34 = ( W + b 2 4 ) ω 24 . \begin{aligned}
d\gamma_{13}&=\mathcal{V}\omega^{13},&
\gamma_{12}\wedge\gamma_{23}&=\frac{c}{2}\omega^{24},\\
d\gamma_{14}&=0,&
\gamma_{12}\wedge\gamma_{24}+\gamma_{13}\wedge\gamma_{34}&=-\frac{c}{2}\omega^{23}+\mathcal{W}\omega^{14}-\frac{\sigma b}{2}\omega^{12},\\
d\gamma_{24}&=\frac{c}{2}\omega^{13},&
\gamma_{23}\wedge\gamma_{34}&=\left(\mathcal{W}+\frac{b^2}{4}\right)\omega^{24}.
\end{aligned} d γ 13 d γ 14 d γ 24 = V ω 13 , = 0 , = 2 c ω 13 , γ 12 ∧ γ 23 γ 12 ∧ γ 24 + γ 13 ∧ γ 34 γ 23 ∧ γ 34 = 2 c ω 24 , = − 2 c ω 23 + W ω 14 − 2 σ b ω 12 , = ( W + 4 b 2 ) ω 24 .
and
d γ 23 = V ω 23 + σ μ ω 12 − c 2 ω 14 + ( σ b + b ′ 2 + τ b 2 ) ω 34 , γ 21 ∧ γ 13 + γ 24 ∧ γ 43 = − σ μ ω 12 − τ b 2 ω 34 + b 2 4 ω 23 . \begin{aligned}
d\gamma_{23}&=\mathcal{V}\omega^{23}+\sigma\mu\omega^{12}-\frac{c}{2}\omega^{14}+\left(\sigma b+\frac{b'}{2}+\frac{\tau b}{2}\right)\omega^{34},\\
\gamma_{21}\wedge\gamma_{13}+\gamma_{24}\wedge\gamma_{43}&=-\sigma\mu\omega^{12}-\frac{\tau b}{2}\omega^{34}+\frac{b^2}{4}\omega^{23}.
\end{aligned} d γ 23 γ 21 ∧ γ 13 + γ 24 ∧ γ 43 = V ω 23 + σ μ ω 12 − 2 c ω 14 + ( σ b + 2 b ′ + 2 τ b ) ω 34 , = − σ μ ω 12 − 2 τ b ω 34 + 4 b 2 ω 23 .
As a check, the coefficients in (18) satisfy R i j , k l = R k l , i j R_{ij,kl}=R_{kl,ij} R ij , k l = R k l , ij and the Bianchi identity R 12 , 34 + R 13 , 42 + R 14 , 23 = c − c 2 − c 2 = 0 R_{12,34}+R_{13,42}+R_{14,23}=c-\frac{c}{2}-\frac{c}{2}=0 R 12 , 34 + R 13 , 42 + R 14 , 23 = c − 2 c − 2 c = 0 . □ \square □
We now pass to the Hodge decomposition. Recall that A + A_{+} A + and A − A_{-} A − denote the diagonal blocks of the curvature operator in (3) .
Lemma 4.3. With respect to the orthonormal bases
ω 12 ± ω 34 2 , ω 13 ∓ ω 24 2 , ω 14 ± ω 23 2 , \frac{\omega^{12}\pm\omega^{34}}{\sqrt{2}},\qquad\frac{\omega^{13}\mp\omega^{24}}{\sqrt{2}},\qquad\frac{\omega^{14}\pm\omega^{23}}{\sqrt{2}}, 2 ω 12 ± ω 34 , 2 ω 13 ∓ ω 24 , 2 ω 14 ± ω 23 , of Λ ± \Lambda^{\pm} Λ ± , the blocks A ± A_{\pm} A ± of the metric (16) are given by
A ± = ( a ± 0 D 0 β ± 0 D 0 β ± ) , a ± = U + X 2 − 3 b 2 8 ± c , β ± = V + W 2 + b 2 8 ∓ c 2 , D = κ ′ sin ϑ 4 h . (19) A_{\pm}=\begin{pmatrix}
a_{\pm} & 0 & D\\
0 & \beta_{\pm} & 0\\
D & 0 & \beta_{\pm}
\end{pmatrix},
\qquad
\begin{aligned}
a_{\pm}&=\frac{\mathcal{U}+\mathcal{X}}{2}-\frac{3b^2}{8}\pm c,\\
\beta_{\pm}&=\frac{\mathcal{V}+\mathcal{W}}{2}+\frac{b^2}{8}\mp\frac{c}{2},\\
D&=\frac{\kappa'\sin\vartheta}{4h}.
\end{aligned} \tag*{(19)} A ± = a ± 0 D 0 β ± 0 D 0 β ± , a ± β ± D = 2 U + X − 8 3 b 2 ± c , = 2 V + W + 8 b 2 ∓ 2 c , = 4 h κ ′ sin ϑ . (19) Consequently, with the gaps G ± = a ± − β ± G_{\pm}=a_{\pm}-\beta_{\pm} G ± = a ± − β ± , we have
k g = V + W 2 + κ 2 ( 1 − z 2 ) 8 h 2 − E , E = 1 2 ∑ ± ( 1 4 G ± 2 + D 2 − 1 2 G ± ) ≥ 0 , (20) k_g=\frac{\mathcal{V}+\mathcal{W}}{2}+\frac{\kappa^2(1-z^2)}{8h^2}-\mathcal{E},\qquad
\mathcal{E}=\frac{1}{2}\sum_{\pm}\left(\sqrt{\frac{1}{4}G_{\pm}^2+D^2}-\frac{1}{2}G_{\pm}\right)\geq0, \tag*{(20)} k g = 2 V + W + 8 h 2 κ 2 ( 1 − z 2 ) − E , E = 2 1 ± ∑ ( 4 1 G ± 2 + D 2 − 2 1 G ± ) ≥ 0 , (20) and if G + G_{+} G + and G − G_{-} G − are positive, then
E ≤ D 2 min { G + , G − } . (21) \mathcal{E}\leq\frac{D^2}{\min\{G_{+},G_{-}\}}. \tag*{(21)} E ≤ min { G + , G − } D 2 . (21)
Proof. The entries of A ± A_{\pm} A ± are read off from (18) . For example, the entry coupling the first and the third basis vector is
1 2 ( R 12 , 14 + R 34 , 23 ± R 12 , 23 ± R 34 , 14 ) = 1 2 ( − h ′ b 2 h + T ) = 1 4 ( b ′ + h ′ h b ) = κ ′ sin ϑ 4 h , \frac{1}{2}\left(R_{12,14}+R_{34,23}\pm R_{12,23}\pm R_{34,14}\right)=\frac{1}{2}\left(-\frac{h'b}{2h}+\mathcal{T}\right)=\frac{1}{4}\left(b'+\frac{h'}{h}b\right)=\frac{\kappa'\sin\vartheta}{4h}, 2 1 ( R 12 , 14 + R 34 , 23 ± R 12 , 23 ± R 34 , 14 ) = 2 1 ( − 2 h h ′ b + T ) = 4 1 ( b ′ + h h ′ b ) = 4 h κ ′ sin ϑ ,
where we used b = κ sin ϑ / h b=\kappa\sin\vartheta/h b = κ sin ϑ / h in the last step. In the same way one finds the diagonal entries and sees that the second basis vector is not coupled to the others. Thus A ± A_{\pm} A ± is the direct sum of the 1 × 1 1 \times1 1 × 1 matrix ( β ± ) (\beta_\pm) ( β ± ) and of the symmetric 2 × 2 2 \times2 2 × 2 matrix with diagonal entries a ± a_\pm a ± , β ± \beta_\pm β ± and off-diagonal entry D D D . The smaller eigenvalue of the latter is
a ± + β ± 2 − 1 4 G ± 2 + D 2 = β ± − E ± , E ± = 1 4 G ± 2 + D 2 − 1 2 G ± ≥ 0. \frac{a_\pm+\beta_\pm}{2}-\sqrt{\frac{1}{4}G_\pm^2+D^2}=\beta_\pm-\mathcal{E}_\pm,\qquad\mathcal{E}_\pm=\sqrt{\frac{1}{4}G_\pm^2+D^2}-\frac{1}{2}G_\pm\geq0. 2 a ± + β ± − 4 1 G ± 2 + D 2 = β ± − E ± , E ± = 4 1 G ± 2 + D 2 − 2 1 G ± ≥ 0.
Hence λ min ( A ± ) = β ± − E ± \lambda_{\min}(A_\pm)=\beta_\pm-\mathcal{E}_\pm λ m i n ( A ± ) = β ± − E ± , and (20) follows from Lemma 2.1, since 1 2 ( β + + β − ) = 1 2 ( V + W ) + 1 8 b 2 \frac{1}{2}(\beta_++\beta_-)=\frac{1}{2}(\mathcal{V}+\mathcal{W})+\frac{1}{8}b^2 2 1 ( β + + β − ) = 2 1 ( V + W ) + 8 1 b 2 and b 2 = κ 2 ( 1 − z 2 ) / h 2 b^2=\kappa^2(1-z^2)/h^2 b 2 = κ 2 ( 1 − z 2 ) / h 2 . If G ± > 0 G_\pm>0 G ± > 0 , then
E ± = 2 D 2 G ± + G ± 2 + 4 D 2 ≤ D 2 G ± , \mathcal{E}_\pm=\frac{2D^2}{G_\pm+\sqrt{G_\pm^2+4D^2}}\leq\frac{D^2}{G_\pm}, E ± = G ± + G ± 2 + 4 D 2 2 D 2 ≤ G ± D 2 ,
which gives (21) .
We note two features of these formulas. First, the coupling D D D involves only the derivative of κ \kappa κ . If κ \kappa κ is constant, the blocks A ± A_\pm A ± are diagonal, even where h h h varies. Second, as a check on the normalization, the scalar curvature is s g = 4 tr A ± = 2 U + 4 V + 4 W + 2 X − 1 2 b 2 s_g=4\operatorname{tr}A_\pm=2\mathcal{U}+4\mathcal{V}+4\mathcal{W}+2\mathcal{X}-\frac{1}{2}b^2 s g = 4 tr A ± = 2 U + 4 V + 4 W + 2 X − 2 1 b 2 .
If κ = 0 \kappa=0 κ = 0 , that is, if A A A is constant, both blocks are diagonal with entries 1 2 ( U + X ) \frac{1}{2}(\mathcal{U}+\mathcal{X}) 2 1 ( U + X ) , 1 2 ( V + W ) \frac{1}{2}(\mathcal{V}+\mathcal{W}) 2 1 ( V + W ) , 1 2 ( V + W ) \frac{1}{2}(\mathcal{V}+\mathcal{W}) 2 1 ( V + W ) , so that
k g = min { U + X 2 , V + W 2 } if κ = 0. (22) k_g=\min\left\{\frac{\mathcal{U}+\mathcal{X}}{2},\frac{\mathcal{V}+\mathcal{W}}{2}\right\}\qquad\text{if }\kappa=0. \tag*{(22)} k g = min { 2 U + X , 2 V + W } if κ = 0. (22)
Since 1 2 ( V + W ) = − ( f h ′ ) ′ / ( 2 f h ) \frac{1}{2}(\mathcal{V}+\mathcal{W})=-(fh')'/(2fh) 2 1 ( V + W ) = − ( f h ′ ) ′ / ( 2 f h ) , this gives k g ≤ − ( f h ′ ) ′ / ( 2 f h ) k_g\leq-(fh')'/(2fh) k g ≤ − ( f h ′ ) ′ / ( 2 f h ) , and where f f f is constant it gives k g ≤ − h ′ ′ / ( 2 h ) k_g\leq-h''/(2h) k g ≤ − h ′′ / ( 2 h ) . In (20) , by contrast, the term 1 2 ( V + W ) \frac{1}{2}(\mathcal{V}+\mathcal{W}) 2 1 ( V + W ) is accompanied by the nonnegative connection term, and the estimate holds for all two-planes, not only for those adapted to the product structure.
The conformal correction
The connection term in (20) vanishes at the poles z = ± 1 z=\pm1 z = ± 1 , and since E ≥ 0 \mathcal{E}\geq0 E ≥ 0 we have k g ≤ 1 2 ( V + W ) = − ( f h ′ ) ′ / ( 2 f h ) k_g\leq\frac{1}{2}(\mathcal{V}+\mathcal{W})=-(fh')'/(2fh) k g ≤ 2 1 ( V + W ) = − ( f h ′ ) ′ / ( 2 f h ) there. As we saw at the beginning of this section, the right-hand side is negative somewhere on the handle, so that the twist alone does not suffice. We correct this by a conformal change. For the metric (16) we put
B = κ 2 24 , u = 1 + B z 2 , g ^ = u 2 g . (23) \mathcal{B}=\frac{\kappa^2}{24},\qquad u=1+\mathcal{B}z^2,\qquad\widehat{g}=u^2g. \tag*{(23)} B = 24 κ 2 , u = 1 + B z 2 , g = u 2 g . (23)
and for a function φ \varphi φ of r r r we write
L r φ = φ ′ ′ + ( f ′ f + 2 h ′ h ) φ ′ . L_r\varphi=\varphi''+\left(\frac{f'}{f}+\frac{2h'}{h}\right)\varphi'. L r φ = φ ′′ + ( f f ′ + h 2 h ′ ) φ ′ .
Lemma 4.4. The metric g ^ \widehat{g} g in (23) satisfies
u 3 k ^ g = − u 2 ( h ′ ′ h + f ′ h ′ f h ) + κ 2 12 h 2 + κ 4 z 2 ( 1 − z 2 ) 192 h 2 − u E − 1 2 ( L r B ) z 2 . (24) u^3\widehat{k}_g=-\frac{u}{2}\left(\frac{h''}{h}+\frac{f'h'}{fh}\right)+\frac{\kappa^2}{12h^2}+\frac{\kappa^4z^2(1-z^2)}{192h^2}-u\mathcal{E}-\frac{1}{2}(L_r\mathcal{B})z^2. \tag*{(24)} u 3 k g = − 2 u ( h h ′′ + f h f ′ h ′ ) + 12 h 2 κ 2 + 192 h 2 κ 4 z 2 ( 1 − z 2 ) − u E − 2 1 ( L r B ) z 2 . (24)
where E \mathcal{E} E is the function in (20) .
Proof. By Lemma 2.2 ,
u 3 κ g = u κ g − 1 2 Δ g u . (25) u^{3}\kappa_{g}=u\kappa_{g}-\frac{1}{2}\Delta_{g}u. \tag*{(25)} u 3 κ g = u κ g − 2 1 Δ g u . (25)
The volume density of g g g in the coordinates ( r , θ , ϑ , ϕ ) (r,\theta,\vartheta,\phi) ( r , θ , ϑ , ϕ ) is f h 2 sin ϑ fh^{2}\sin\vartheta f h 2 sin ϑ , and for a function of r r r and ϑ \vartheta ϑ alone the gradient has no component in the directions of θ \theta θ and ϕ \phi ϕ . Therefore Δ g φ = L r φ \Delta_{g}\varphi=L_{r}\varphi Δ g φ = L r φ for a function of r r r , and the Laplacian of a function on the sphere is h − 2 h^{-2} h − 2 times its Laplacian on the unit sphere. Since Δ S 2 ( z 2 ) = 2 − 6 z 2 \Delta_{S^{2}}(z^{2})=2-6z^{2} Δ S 2 ( z 2 ) = 2 − 6 z 2 , we get
Δ g u = ( L r B ) z 2 + κ 2 24 h 2 ( 2 − 6 z 2 ) . \Delta_{g}u=(L_{r}\mathcal{B})z^{2}+\frac{\kappa^{2}}{24h^{2}}(2-6z^{2}). Δ g u = ( L r B ) z 2 + 24 h 2 κ 2 ( 2 − 6 z 2 ) .
We insert this and (20) into (25) . The terms that involve neither derivatives of h h h and f f f nor E \mathcal{E} E combine to
u κ 2 ( 1 − z 2 ) 8 h 2 − κ 2 ( 2 − 6 z 2 ) 48 h 2 = κ 2 48 h 2 ( 6 − 6 z 2 − 2 + 6 z 2 ) + κ 4 z 2 ( 1 − z 2 ) 192 h 2 = κ 2 12 h 2 + κ 4 z 2 ( 1 − z 2 ) 192 h 2 , \begin{aligned}
\frac{u\kappa^{2}(1-z^{2})}{8h^{2}}-\frac{\kappa^{2}(2-6z^{2})}{48h^{2}}=\frac{\kappa^{2}}{48h^{2}}(6-6z^{2}-2+6z^{2})+\frac{\kappa^{4}z^{2}(1-z^{2})}{192h^{2}} \\
&=\frac{\kappa^{2}}{12h^{2}}+\frac{\kappa^{4}z^{2}(1-z^{2})}{192h^{2}},
\end{aligned} 8 h 2 u κ 2 ( 1 − z 2 ) − 48 h 2 κ 2 ( 2 − 6 z 2 ) = 48 h 2 κ 2 ( 6 − 6 z 2 − 2 + 6 z 2 ) + 192 h 2 κ 4 z 2 ( 1 − z 2 ) = 12 h 2 κ 2 + 192 h 2 κ 4 z 2 ( 1 − z 2 ) ,
and (24) follows.
The cancellation of the terms ± 6 z 2 \pm6z^{2} ± 6 z 2 in this computation is the reason for the constant 24 in (23) . It is the only choice for which the leading term κ 2 / ( 12 h 2 ) \kappa^{2}/(12h^{2}) κ 2 / ( 12 h 2 ) does not depend on z z z .
The logarithmic radius
Outside the cap the function f f f will be constant and h h h will be increasing. It is then convenient to use the logarithmic radius and the defect,
t = log h , δ = 1 − h ′ 2 , (26) t=\log h,\qquad\delta=1-h'^{2}, \tag*{(26)} t = log h , δ = 1 − h ′ 2 , (26)
and to regard δ \delta δ and κ \kappa κ as functions of t t t . A dot denotes the derivative with respect to t t t .
Lemma 4.5. Suppose that f f f is constant and that 0 < h ′ < 1 0<h'<1 0 < h ′ < 1 on some interval. Then
h h ′ ′ = − δ ˙ 2 , κ ′ = 1 − δ h κ ˙ , (27) hh''=-\frac{\dot{\delta}}{2},\qquad\kappa'=\frac{\sqrt{1-\delta}}{h}\dot{\kappa}, \tag*{(27)} h h ′′ = − 2 δ ˙ , κ ′ = h 1 − δ κ ˙ , (27) the gaps of Lemma 4.3 are given by
2 h 2 G ± = δ − δ ˙ 2 − κ 2 ( 1 − z 2 ) ∓ 3 κ z , (28) 2h^{2}G_{\pm}=\delta-\frac{\dot{\delta}}{2}-\kappa^{2}(1-z^{2})\mp3\kappa z, \tag*{(28)} 2 h 2 G ± = δ − 2 δ ˙ − κ 2 ( 1 − z 2 ) ∓ 3 κ z , (28) and
h 2 L r B = ( 1 − δ ) B ¨ + ( 1 − δ − δ ˙ 2 ) B ˙ . (29) h^{2}L_{r}\mathcal{B}=(1-\delta)\ddot{\mathcal{B}}+\left(1-\delta-\frac{\dot{\delta}}{2}\right)\dot{\mathcal{B}}. \tag*{(29)} h 2 L r B = ( 1 − δ ) B ¨ + ( 1 − δ − 2 δ ˙ ) B ˙ . (29) Consequently, (24) takes the form
h 2 u 3 κ g = u δ ˙ 4 + κ 2 12 + κ 4 z 2 ( 1 − z 2 ) 192 − u h 2 E − 1 2 ( h 2 L r B ) z 2 . (30) h^{2}u^{3}\kappa_{g}=\frac{u\dot{\delta}}{4}+\frac{\kappa^{2}}{12}+\frac{\kappa^{4}z^{2}(1-z^{2})}{192}-uh^{2}\mathcal{E}-\frac{1}{2}(h^{2}L_{r}\mathcal{B})z^{2}. \tag*{(30)} h 2 u 3 κ g = 4 u δ ˙ + 12 κ 2 + 192 κ 4 z 2 ( 1 − z 2 ) − u h 2 E − 2 1 ( h 2 L r B ) z 2 . (30) If, moreover, 0 ≤ κ ≤ 3 2 0\leq\kappa\leq\frac{3}{2} 0 ≤ κ ≤ 2 3 and δ − 1 2 δ ˙ − 3 κ > 0 \delta-\frac{1}{2}\dot{\delta}-3\kappa>0 δ − 2 1 δ ˙ − 3 κ > 0 , then
2 h 2 G ± ≥ δ − δ ˙ 2 − 3 κ and E ≤ ( 1 − δ ) κ ˙ 2 8 h 2 ( δ − 1 2 δ ˙ − 3 κ ) . (31) 2h^{2}G_{\pm}\geq\delta-\frac{\dot{\delta}}{2}-3\kappa\qquad\text{and}\qquad\mathcal{E}\leq\frac{(1-\delta)\dot{\kappa}^{2}}{8h^{2}(\delta-\frac{1}{2}\dot{\delta}-3\kappa)}. \tag*{(31)} 2 h 2 G ± ≥ δ − 2 δ ˙ − 3 κ and E ≤ 8 h 2 ( δ − 2 1 δ ˙ − 3 κ ) ( 1 − δ ) κ ˙ 2 . (31)
Proof. Since d t / d r = h ′ / h dt/dr=h'/h d t / d r = h ′ / h , we have δ ˙ = − 2 h ′ h ′ ′ ⋅ h / h ′ = − 2 h h ′ ′ \dot{\delta}=-2h'h''\cdot h/h'=-2hh'' δ ˙ = − 2 h ′ h ′′ ⋅ h / h ′ = − 2 h h ′′ and κ ′ = κ h ′ / h \kappa'=\kappa h'/h κ ′ = κ h ′ / h , which is (27) . For constant f f f we have W = X = 0 \mathcal{W}=\mathcal{X}=0 W = X = 0 , and by (19)
2 h 2 G ± = h 2 ( U − V ) − h 2 b 2 ± 3 h 2 c = δ + h h ′ ′ − κ 2 ( 1 − z 2 ) ± 3 κ z 2h^2G_{\pm}=h^2(\mathcal{U}-\mathcal{V})-h^2b^2\pm3h^2c=\delta+hh''-\kappa^2(1-z^2)\pm3\kappa z 2 h 2 G ± = h 2 ( U − V ) − h 2 b 2 ± 3 h 2 c = δ + h h ′′ − κ 2 ( 1 − z 2 ) ± 3 κ z
which gives (28) . For a function φ \varphi φ of r r r the chain rule gives
φ ′ = h ′ h φ ˙ , φ ′ ′ = h 2 h 2 φ ¨ + ( h ′ ′ h − h ′ 2 h 2 ) φ ˙ , \varphi'=\frac{h'}{h}\dot{\varphi},\qquad\varphi''=\frac{h^2}{h^2}\ddot{\varphi}+\left(\frac{h''}{h}-\frac{h'^2}{h^2}\right)\dot{\varphi}, φ ′ = h h ′ φ ˙ , φ ′′ = h 2 h 2 φ ¨ + ( h h ′′ − h 2 h ′2 ) φ ˙ ,
and hence h 2 L r φ = h ′ 2 φ ¨ + ( h h ′ ′ + h ′ 2 ) φ ˙ h^2L_r\varphi=h'^2\ddot{\varphi}+(hh''+h'^2)\dot{\varphi} h 2 L r φ = h ′2 φ ¨ + ( h h ′′ + h ′2 ) φ ˙ , which is (29) . Now (30) is (24) multiplied by h 2 h^2 h 2 . For the last assertion we use the elementary fact that
max − 1 ≤ z ≤ 1 { κ 2 ( 1 − z 2 ) + 3 κ ∣ z ∣ } = 3 κ for 0 ≤ κ ≤ 3 2 (32) \max_{-1\leq z\leq1}\{\kappa^2(1-z^2)+3\kappa|z|\}=3\kappa\qquad\text{for }0\leq\kappa\leq\frac{3}{2} \tag*{(32)} − 1 ≤ z ≤ 1 max { κ 2 ( 1 − z 2 ) + 3 κ ∣ z ∣ } = 3 κ for 0 ≤ κ ≤ 2 3 (32)
indeed, the derivative of κ 2 ( 1 − x 2 ) + 3 κ x \kappa^2(1-x^2)+3\kappa x κ 2 ( 1 − x 2 ) + 3 κ x with respect to x = ∣ z ∣ x=|z| x = ∣ z ∣ is κ ( 3 − 2 κ x ) ≥ 0 \kappa(3-2\kappa x)\geq0 κ ( 3 − 2 κ x ) ≥ 0 for 0 ≤ x ≤ 1 0\leq x\leq1 0 ≤ x ≤ 1 . Together with (28) this gives the lower bound for 2 h 2 G ± 2h^2G_{\pm} 2 h 2 G ± . By (21) , (19) and (27) ,
ε ≤ D 2 min { G + , G − } ≤ ( 1 − δ ) κ ˙ 2 ( 1 − z 2 ) 16 h 4 ⋅ 2 h 2 δ − 1 2 δ ˙ − 3 κ ≤ ( 1 − δ ) κ ˙ 2 8 h 2 ( δ − 1 2 δ ˙ − 3 κ ) . \varepsilon\leq\frac{D^2}{\min\{G_+,G_-\}}\leq\frac{(1-\delta)\dot{\kappa}^2(1-z^2)}{16h^4}\cdot\frac{2h^2}{\delta-\frac{1}{2}\dot{\delta}-3\kappa}\leq\frac{(1-\delta)\dot{\kappa}^2}{8h^2\left(\delta-\frac{1}{2}\dot{\delta}-3\kappa\right)}. ε ≤ min { G + , G − } D 2 ≤ 16 h 4 ( 1 − δ ) κ ˙ 2 ( 1 − z 2 ) ⋅ δ − 2 1 δ ˙ − 3 κ 2 h 2 ≤ 8 h 2 ( δ − 2 1 δ ˙ − 3 κ ) ( 1 − δ ) κ ˙ 2 .
The identity (30) makes the strategy quantitative. With κ = δ / 6 \kappa=\delta/6 κ = δ /6 the gap condition reads 1 2 ( δ − δ ˙ ) > 0 \frac{1}{2}(\delta-\dot{\delta})>0 2 1 ( δ − δ ˙ ) > 0 , and the first two terms on the right-hand side are 1 4 u δ ˙ + 1 432 δ 2 \frac{1}{4}u\dot{\delta}+\frac{1}{432}\delta^2 4 1 u δ ˙ + 432 1 δ 2 . If δ ˙ = − ε δ 2 \dot{\delta}=-\varepsilon\delta^2 δ ˙ = − ε δ 2 with ε \varepsilon ε well below 1 108 \frac{1}{108} 108 1 , their sum is a positive multiple of δ 2 \delta^2 δ 2 , and the remaining terms turn out to be of smaller order. We shall take ε = 1 1000 \varepsilon=\frac{1}{1000} ε = 1000 1 .
The radial profiles
We now define the functions f f f , h h h and κ \kappa κ . The function A A A is then determined by
A ′ = f κ h 2 , A ( 0 ) = 0 , (33) A'=\frac{f\kappa}{h^2},\qquad A(0)=0, \tag*{(33)} A ′ = h 2 f κ , A ( 0 ) = 0 , (33)
and the metric of Theorem 4.1 will be g ^ = u 2 g \hat{g}=u^2g g ^ = u 2 g with g g g as in (16) and u u u as in (23) . The outer pieces of the handle are fixed first, and the radius h 0 h_0 h 0 of the sphere at the center is then chosen so that the neck arrives at the first outer piece with the right defect.
Constants and cutoff functions. Throughout, we fix
ε = 1 1000 , κ 0 = 1 6 , \varepsilon=\frac{1}{1000},\qquad\kappa_0=\frac{1}{6}, ε = 1000 1 , κ 0 = 6 1 ,
and a smooth nondecreasing function ζ \zeta ζ from R \mathbb{R} R to [ 0 , 1 ] [0,1] [ 0 , 1 ] that vanishes on a neighborhood of ( − ∞ , 0 ] (-\infty,0] ( − ∞ , 0 ] , is equal to one on a neighborhood of [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) , and satisfies
0 ≤ ζ ′ ≤ 2 , ∣ ζ ′ ′ ∣ ≤ 64. (34) 0\leq\zeta'\leq2,\qquad|\zeta''|\leq64. \tag*{(34)} 0 ≤ ζ ′ ≤ 2 , ∣ ζ ′′ ∣ ≤ 64. (34)
Such a function exists. Indeed, a suitable rescaling of the function exp ( − 1 / ( 1 − x 2 ) ) \exp(-1/(1-x^2)) exp ( − 1/ ( 1 − x 2 )) on ( − 1 , 1 ) (-1,1) ( − 1 , 1 ) is a nonnegative smooth function with integral one, with support in [ − 1 8 , 1 8 ] [-\frac{1}{8},\frac{1}{8}] [ − 8 1 , 8 1 ] and bounded by 16. The convolution of this function with 2 ⋅ 1 [ 1 / 4 , 3 / 4 ] 2\cdot\mathbf{1}_{[1/4,3/4]} 2 ⋅ 1 [ 1/4 , 3/4 ] has a primitive that vanishes at 0, and this primitive has all the required properties, since the absolute value of its second derivative is at most 32.
The outer parameters. The last three pieces occupy fixed intervals of the logarithmic radius. We put
t 3 = log R , t 2 = t 3 − 1 , t 1 = t 3 − 11 , t 0 = t 3 − 31 , δ 0 = K R 2 4 e − 17 . (35) t_3 = \log R,\qquad t_2 = t_3 - 1,\qquad t_1 = t_3 - 11,\qquad t_0 = t_3 - 31,\qquad\delta_0 = \frac{KR^2}{4}e^{-17}. \tag*{(35)} t 3 = log R , t 2 = t 3 − 1 , t 1 = t 3 − 11 , t 0 = t 3 − 31 , δ 0 = 4 K R 2 e − 17 . (35)
The number δ 0 \delta_0 δ 0 will be the defect at t = t 0 t=t_0 t = t 0 . On the intervals [ t 0 , t 1 ] [t_0,t_1] [ t 0 , t 1 ] and [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] , of lengths 20 and 10, the defect will satisfy δ ˙ ≤ 1 2 δ \dot{\delta}\le\frac{1}{2}\delta δ ˙ ≤ 2 1 δ , so that it grows at most by the factor e 15 e^{15} e 15 ; the remaining factor e − 2 e^{-2} e − 2 in δ 0 \delta_0 δ 0 ensures that δ ( t 2 ) ≤ 1 4 K e 2 t 2 \delta(t_2)\le\frac{1}{4}Ke^{2t_2} δ ( t 2 ) ≤ 4 1 K e 2 t 2 , which is what we need on the last interval.
The normalized ramp. Let h ˉ \bar{h} h ˉ be the solution on [ 0 , 1 ] [0,1] [ 0 , 1 ] of the initial value problem
h ˉ s s = ϵ ζ ( s ) 2 h ˉ ( 1 − h ˉ s 2 ) 2 , h ˉ ( 0 ) = 1 , h ˉ s ( 0 ) = 0. (36) \bar{h}_{ss}=\frac{\epsilon\,\zeta(s)}{2\bar{h}}(1-\bar{h}_s^2)^2,\qquad\bar{h}(0)=1,\qquad\bar{h}_s(0)=0. \tag*{(36)} h ˉ ss = 2 h ˉ ϵ ζ ( s ) ( 1 − h ˉ s 2 ) 2 , h ˉ ( 0 ) = 1 , h ˉ s ( 0 ) = 0. (36)
The pair ( h ˉ , p ) (\bar{h},p) ( h ˉ , p ) with p = h ˉ s p=\bar{h}_s p = h ˉ s solves the first-order system h ˉ s = p \bar{h}_s=p h ˉ s = p , p s = ϵ ζ ( 1 − p 2 ) 2 / ( 2 h ˉ ) p_s=\epsilon\zeta(1-p^2)^2/(2\bar{h}) p s = ϵ ζ ( 1 − p 2 ) 2 / ( 2 h ˉ ) . The set { p = 1 } \{p=1\} { p = 1 } is invariant under this system, and since p ( 0 ) = 0 p(0)=0 p ( 0 ) = 0 , uniqueness shows that h ˉ s < 1 \bar{h}_s<1 h ˉ s < 1 as long as the solution exists. Hence 0 ≤ h ˉ s s ≤ 1 2 ϵ 0\le\bar{h}_{ss}\le\frac{1}{2}\epsilon 0 ≤ h ˉ ss ≤ 2 1 ϵ , and therefore
0 ≤ h ˉ s ( s ) ≤ ϵ s 2 , 1 ≤ h ˉ ( s ) ≤ 1 + ϵ s 2 4 , 0\le\bar{h}_s(s)\le\frac{\epsilon s}{2},\qquad1\le\bar{h}(s)\le1+\frac{\epsilon s^2}{4}, 0 ≤ h ˉ s ( s ) ≤ 2 ϵs , 1 ≤ h ˉ ( s ) ≤ 1 + 4 ϵ s 2 ,
so that the solution exists on all of [ 0 , 1 ] [0,1] [ 0 , 1 ] . We put
h ˉ ∗ = h ˉ ( 1 ) , δ ∗ = 1 − h ˉ s ( 1 ) 2 . \bar{h}_*=\bar{h}(1),\qquad\delta_*=1-\bar{h}_s(1)^2. h ˉ ∗ = h ˉ ( 1 ) , δ ∗ = 1 − h ˉ s ( 1 ) 2 .
Then 1 ≤ h ˉ ∗ ≤ 1 + 1 4 ϵ 1\le\bar{h}_*\le1+\frac{1}{4}\epsilon 1 ≤ h ˉ ∗ ≤ 1 + 4 1 ϵ and 1 − 1 4 ϵ 2 ≤ δ ∗ < 1 1-\frac{1}{4}\epsilon^2\le\delta_*<1 1 − 4 1 ϵ 2 ≤ δ ∗ < 1 ; the last inequality holds because ζ \zeta ζ is positive on an interval, so that h ˉ s ( 1 ) > 0 \bar{h}_s(1)>0 h ˉ s ( 1 ) > 0 .
The radius at the center. We define
h 0 = e t 0 h ˉ ∗ exp [ − 1 ϵ ( 1 δ 0 − 1 δ ∗ ) ] . (37) h_0=\frac{e^{t_0}}{\bar{h}_*}\exp\left[-\frac{1}{\epsilon}\left(\frac{1}{\delta_0}-\frac{1}{\delta_*}\right)\right]. \tag*{(37)} h 0 = h ˉ ∗ e t 0 exp [ − ϵ 1 ( δ 0 1 − δ ∗ 1 ) ] . (37)
Since δ 0 < δ ∗ \delta_0<\delta_* δ 0 < δ ∗ , we have 0 < h 0 h ˉ ∗ < e t 0 0<h_0\bar{h}_*<e^{t_0} 0 < h 0 h ˉ ∗ < e t 0 .
The cap and the cylinder, 0 ≤ r ≤ r 0 0\le r\le r_0 0 ≤ r ≤ r 0 . Put ϱ ( s ) = 1 − ζ ( s − 1 ) \varrho(s)=1-\zeta(s-1) ϱ ( s ) = 1 − ζ ( s − 1 ) for s ≥ 0 s\ge0 s ≥ 0 and J = ∫ 0 ∞ ϱ ( s ) d s J=\int_0^\infty\varrho(s)\,ds J = ∫ 0 ∞ ϱ ( s ) d s . The function ϱ \varrho ϱ is nonincreasing, equal to one near [ 0 , 1 ] [0,1] [ 0 , 1 ] , and it vanishes on [ 2 , ∞ ) [2,\infty) [ 2 , ∞ ) . We define
f ( r ) = F J ∫ 0 J r / F ϱ ( s ) d s , r c = 2 F J , r 0 = r c + ℓ c y l , (38) f(r)=\frac{F}{J}\int_0^{Jr/F}\varrho(s)\,ds,\qquad r_c=\frac{2F}{J},\qquad r_0=r_c+\ell_{\mathrm{cyl}}, \tag*{(38)} f ( r ) = J F ∫ 0 J r / F ϱ ( s ) d s , r c = J 2 F , r 0 = r c + ℓ cyl , (38)
where ℓ c y l ≥ 0 \ell_{\mathrm{cyl}}\ge0 ℓ cyl ≥ 0 is a parameter that will be fixed in Subsection 4.6. Thus f ( r ) = r f(r)=r f ( r ) = r near r = 0 r=0 r = 0 , f ′ ′ ≤ 0 f''\le0 f ′′ ≤ 0 , and f = F f=F f = F for r ≥ r c r\ge r_c r ≥ r c . On [ 0 , r 0 ] [0,r_0] [ 0 , r 0 ] we put
h ≡ h 0 , κ ≡ κ 0 . h\equiv h_0,\qquad\kappa\equiv\kappa_0. h ≡ h 0 , κ ≡ κ 0 .
We call [ 0 , r c ] [0,r_c] [ 0 , r c ] the cap and [ r c , r 0 ] [r_c,r_0] [ r c , r 0 ] the cylinder.
The ramp and the neck , r ≥ r 0 r \ge r_0 r ≥ r 0 and h ≤ e t 0 h \le e^{t_0} h ≤ e t 0 . From now on f = F f = F f = F . For r ≥ r 0 r \ge r_0 r ≥ r 0 we let h h h be the solution of
h ′ ′ = ϵ η ( r ) 2 h ( 1 − h ′ 2 ) 2 , h ( r 0 ) = h 0 , h ′ ( r 0 ) = 0 , where η ( r ) = ζ ( r − r 0 h 0 ) , (39) h'' = \frac{\epsilon\eta(r)}{2h}\left(1-h'^2\right)^2,\qquad h(r_0)=h_0,\qquad h'(r_0)=0,\qquad\text{where }\eta(r)=\zeta\left(\frac{r-r_0}{h_0}\right), \tag*{(39)} h ′′ = 2 h ϵη ( r ) ( 1 − h ′2 ) 2 , h ( r 0 ) = h 0 , h ′ ( r 0 ) = 0 , where η ( r ) = ζ ( h 0 r − r 0 ) , (39)
and we put
κ = δ 6 = 1 − h ′ 2 6 . \kappa=\frac{\delta}{6}=\frac{1-h'^2}{6}. κ = 6 δ = 6 1 − h ′2 .
As for (36) , we have 0 ≤ h ′ < 1 0 \le h' < 1 0 ≤ h ′ < 1 and h ′ ′ ≥ 0 h'' \ge0 h ′′ ≥ 0 , and the solution exists for all r ≥ r 0 r \ge r_0 r ≥ r 0 . On the ramp r 0 ≤ r ≤ r 1 r_0 \le r \le r_1 r 0 ≤ r ≤ r 1 , where r 1 = r 0 + h 0 r_1=r_0+h_0 r 1 = r 0 + h 0 , the substitution s = ( r − r 0 ) / h 0 s=(r-r_0)/h_0 s = ( r − r 0 ) / h 0 , h = h 0 h ˉ ( s ) h=h_0\bar h(s) h = h 0 h ˉ ( s ) transforms (39) into (36) . Hence
h ( r 1 ) = h 0 h ˉ ∗ , δ ( r 1 ) = δ ∗ , h ′ ( r 1 ) > 0. h(r_1)=h_0\bar h_*,\qquad\delta(r_1)=\delta_*,\qquad h'(r_1)>0. h ( r 1 ) = h 0 h ˉ ∗ , δ ( r 1 ) = δ ∗ , h ′ ( r 1 ) > 0.
For r ≥ r 1 r \ge r_1 r ≥ r 1 we have η ≡ 1 \eta\equiv1 η ≡ 1 and h ′ > 0 h'>0 h ′ > 0 , so that we can use the logarithmic radius. By (27) , the equation (39) becomes
δ ˙ = − ϵ δ 2 , that is, 1 δ = 1 δ ∗ + ϵ log h h 0 h ˉ ∗ . (40) \dot\delta=-\epsilon\delta^2,\qquad\text{that is,}\qquad\frac{1}{\delta}=\frac{1}{\delta_*}+\epsilon\log\frac{h}{h_0\bar h_*}. \tag*{(40)} δ ˙ = − ϵ δ 2 , that is, δ 1 = δ ∗ 1 + ϵ log h 0 h ˉ ∗ h . (40)
The neck is the region h 0 h ˉ ∗ ≤ h ≤ e t 0 h_0\bar h_* \le h \le e^{t_0} h 0 h ˉ ∗ ≤ h ≤ e t 0 . The choice (37) of h 0 h_0 h 0 means precisely that
δ = δ 0 at h = e t 0 . \delta=\delta_0\qquad\text{at }h=e^{t_0}. δ = δ 0 at h = e t 0 .
Stage I , t 0 ≤ t ≤ t 1 t_0 \le t \le t_1 t 0 ≤ t ≤ t 1 . In this stage the sign of δ ˙ \dot\delta δ ˙ is reversed. Let χ ( t ) = ζ ( ( t − t 0 ) / 20 ) \chi(t)=\zeta((t-t_0)/20) χ ( t ) = ζ (( t − t 0 ) /20 ) , and let δ \delta δ be the solution of
δ ˙ = − ϵ ( 1 − χ ) δ 2 + χ 2 δ , δ ( t 0 ) = δ 0 , and put κ = δ 6 . (41) \dot\delta=-\epsilon(1-\chi)\delta^2+\frac{\chi}{2}\delta,\qquad\delta(t_0)=\delta_0,\qquad\text{and put }\kappa=\frac{\delta}{6}. \tag*{(41)} δ ˙ = − ϵ ( 1 − χ ) δ 2 + 2 χ δ , δ ( t 0 ) = δ 0 , and put κ = 6 δ . (41)
Stage II , t 1 ≤ t ≤ t 2 t_1 \le t \le t_2 t 1 ≤ t ≤ t 2 . In this stage the connection is switched off. We let δ \delta δ continue to solve δ ˙ = 1 2 δ \dot\delta=\frac{1}{2}\delta δ ˙ = 2 1 δ , and we put
κ = δ ψ 6 , where ψ ( t ) = 1 − ζ ( t − t 1 10 ) . (42) \kappa=\frac{\delta\psi}{6},\qquad\text{where }\psi(t)=1-\zeta\left(\frac{t-t_1}{10}\right). \tag*{(42)} κ = 6 δ ψ , where ψ ( t ) = 1 − ζ ( 10 t − t 1 ) . (42)
Stage III , t 2 ≤ t ≤ t 3 t_2 \le t \le t_3 t 2 ≤ t ≤ t 3 . In the last stage the defect is steered to its value on the round collar. With δ 2 = δ ( t 2 ) \delta_2=\delta(t_2) δ 2 = δ ( t 2 ) we put κ = 0 \kappa=0 κ = 0 and
δ ( t ) = ( 1 − ζ ( t − t 2 ) ) δ 2 e ( t − t 2 ) / 2 + ζ ( t − t 2 ) K e 2 t . (43) \delta(t)=\left(1-\zeta(t-t_2)\right)\delta_2e^{(t-t_2)/2}+\zeta(t-t_2)Ke^{2t}. \tag*{(43)} δ ( t ) = ( 1 − ζ ( t − t 2 ) ) δ 2 e ( t − t 2 ) /2 + ζ ( t − t 2 ) K e 2 t . (43)
In Stages I–III the function h h h is recovered from δ \delta δ by solving
d r d t = e t 1 − δ ( t ) . (44) \frac{dr}{dt}=\frac{e^t}{\sqrt{1-\delta(t)}}. \tag*{(44)} d t d r = 1 − δ ( t ) e t . (44)
which requires δ < 1 \delta<1 δ < 1 ; this is part of the following lemma.
Lemma 4.6. The functions defined above have the following properties.
(i) On the ramp and the neck , 0 < δ ≤ 1 0<\delta\le1 0 < δ ≤ 1 , and δ ′ = − ϵ η h ′ δ 2 / h \delta'=-\epsilon\eta h'\delta^2/h δ ′ = − ϵη h ′ δ 2 / h .
(ii) In Stage I , δ > 0 \delta>0 δ > 0 and δ ˙ ≤ 1 2 δ \dot{\delta}\le\frac{1}{2}\delta δ ˙ ≤ 2 1 δ . In Stages I and II ,
0 < δ ( t ) ≤ δ 0 e 15 = K R 2 4 e − 2 < 1 100 , and in particular δ 2 ≤ 1 4 K e 2 t 2 . (45) 0<\delta(t)\le\delta_0e^{15}=\frac{KR^2}{4}e^{-2}<\frac{1}{100},\qquad\text{and in particular}\quad\delta_2\le\frac{1}{4}Ke^{2t_2}. \tag*{(45)} 0 < δ ( t ) ≤ δ 0 e 15 = 4 K R 2 e − 2 < 100 1 , and in particular δ 2 ≤ 4 1 K e 2 t 2 . (45)
(iii) In Stage III , δ ˙ > 0 \dot{\delta}>0 δ ˙ > 0 and 0 < δ ( t ) ≤ K e 2 t ≤ K R 2 ≤ 1 100 0<\delta(t)\le Ke^{2t}\le KR^2\le\frac{1}{100} 0 < δ ( t ) ≤ K e 2 t ≤ K R 2 ≤ 100 1 .
(iv) The profiles fit together smoothly, and the handle has finite length. More precisely, the coordinate r r r runs through an interval [ 0 , ℓ ] [0,\ell] [ 0 , ℓ ] , the functions f f f , h h h , κ \kappa κ and A A A are smooth on ( 0 , ℓ ] (0,\ell] ( 0 , ℓ ] , and h h h increases from h 0 h_0 h 0 to R R R on [ r 0 , ℓ ] [r_0,\ell] [ r 0 , ℓ ] .
(v) Near r = ℓ r=\ell r = ℓ we have κ = 0 \kappa=0 κ = 0 and δ = K h 2 \delta=Kh^2 δ = K h 2 , so that u = 1 u=1 u = 1 , A A A is constant, and
g ^ = g = F 2 d θ 2 + d h 2 1 − K h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( ℓ ) d θ ) 2 } . \hat{g}=g=F^2d\theta^2+\frac{dh^2}{1-Kh^2}+h^2\left\{d\vartheta^2+\sin^2\vartheta\left(d\phi+A(\ell)d\theta\right)^2\right\}. g ^ = g = F 2 d θ 2 + 1 − K h 2 d h 2 + h 2 { d ϑ 2 + sin 2 ϑ ( d ϕ + A ( ℓ ) d θ ) 2 } .
Proof. (i) Differentiating δ = 1 − h ′ 2 \delta=1-h'^2 δ = 1 − h ′2 and using (39) gives the formula for δ ′ \delta' δ ′ . We have already seen that 0 ≤ h ′ < 1 0\le h'<1 0 ≤ h ′ < 1 .
(ii) The function y = 1 / δ y=1/\delta y = 1/ δ solves the linear equation y ˙ = ϵ ( 1 − χ ) − 1 2 χ y \dot{y}=\epsilon(1-\chi)-\frac{1}{2}\chi y y ˙ = ϵ ( 1 − χ ) − 2 1 χy with y ( t 0 ) > 0 y(t_0)>0 y ( t 0 ) > 0 , and ϵ ( 1 − χ ) ≥ 0 \epsilon(1-\chi)\ge0 ϵ ( 1 − χ ) ≥ 0 ; hence y y y stays positive and finite on [ t 0 , t 1 ] [t_0,t_1] [ t 0 , t 1 ] . The inequality δ ˙ ≤ 1 2 δ \dot{\delta}\le\frac{1}{2}\delta δ ˙ ≤ 2 1 δ is clear from (41) , and it also holds in Stage II. Integrating it over [ t 0 , t 2 ] [t_0,t_2] [ t 0 , t 2 ] , an interval of length 30, gives (45) ; recall that K R 2 ≤ 1 100 KR^2\le\frac{1}{100} K R 2 ≤ 100 1 and that t 2 = t 3 − 1 t_2=t_3-1 t 2 = t 3 − 1 .
(iii) Write P ( t ) = δ 2 e ( t − t 2 ) / 2 P(t)=\delta_2e^{(t-t_2)/2} P ( t ) = δ 2 e ( t − t 2 ) /2 , Q ( t ) = K e 2 t Q(t)=Ke^{2t} Q ( t ) = K e 2 t and ξ ( t ) = ζ ( t − t 2 ) \xi(t)=\zeta(t-t_2) ξ ( t ) = ζ ( t − t 2 ) . By (ii), Q / P ≥ 4 e 3 ( t − t 2 ) / 2 > 1 Q/P\ge4e^{3(t-t_2)/2}>1 Q / P ≥ 4 e 3 ( t − t 2 ) /2 > 1 on [ t 2 , t 3 ] [t_2,t_3] [ t 2 , t 3 ] . Hence δ = ( 1 − ξ ) P + ξ Q \delta=(1-\xi)P+\xi Q δ = ( 1 − ξ ) P + ξ Q lies between P P P and Q Q Q , and
δ ˙ = 1 2 ( 1 − ξ ˙ ) P + 2 ξ Q + ξ ˙ ( Q − P ) > 0. \dot{\delta}=\frac{1}{2}(1-\dot{\xi})P+2\xi Q+\dot{\xi}(Q-P)>0. δ ˙ = 2 1 ( 1 − ξ ˙ ) P + 2 ξ Q + ξ ˙ ( Q − P ) > 0.
(iv) Each cutoff function is constant near the ends of the interval on which it is used. Near r = r 0 r=r_0 r = r 0 we have η = 0 \eta=0 η = 0 , so that h h h and κ \kappa κ are constant there. Near t = t 0 t=t_0 t = t 0 we have χ = 0 \chi=0 χ = 0 , so that (41) agrees with (40) . Near t = t 1 t=t_1 t = t 1 we have χ = 1 \chi=1 χ = 1 and ψ = 1 \psi=1 ψ = 1 , so that δ ˙ = 1 2 δ \dot{\delta}=\frac{1}{2}\delta δ ˙ = 2 1 δ and κ = 1 6 δ \kappa=\frac{1}{6}\delta κ = 6 1 δ on both sides, and near t = t 2 t=t_2 t = t 2 we have ψ = 0 \psi=0 ψ = 0 and ξ = 0 \xi=0 ξ = 0 , so that δ ˙ = 1 2 δ \dot{\delta}=\frac{1}{2}\delta δ ˙ = 2 1 δ and κ = 0 \kappa=0 κ = 0 on both sides. Thus δ \delta δ and κ \kappa κ are smooth functions of t t t on [ log ( h 0 h ˉ ∗ ) , t 3 ] [\log(h_0\bar h_*),t_3] [ log ( h 0 h ˉ ∗ ) , t 3 ] , and they are smooth functions of r r r because t t t is a smooth increasing function of r r r for r ≥ r 1 r\ge r_1 r ≥ r 1 by (44) and (39) . The neck has finite length, since δ ≤ δ ∗ < 1 \delta\le\delta_*<1 δ ≤ δ ∗ < 1 there, so that its length is
∫ h 0 h ˉ ∗ e t 0 d h 1 − δ ≤ e t 0 − h 0 h ˉ ∗ 1 − δ ∗ < ∞ . \int_{h_0\bar h_*}^{e^{t_0}}\frac{dh}{\sqrt{1-\delta}}\le\frac{e^{t_0}-h_0\bar h_*}{\sqrt{1-\delta_*}}<\infty. ∫ h 0 h ˉ ∗ e t 0 1 − δ d h ≤ 1 − δ ∗ e t 0 − h 0 h ˉ ∗ < ∞.
Stages I–III have finite length by (44) , since δ ≤ 1 100 \delta\le\frac{1}{100} δ ≤ 100 1 there. Finally, A A A is smooth by (33) .
(v) Near t = t 3 t=t_3 t = t 3 we have ξ = 1 \xi=1 ξ = 1 , hence δ = K e 2 t = K h 2 \delta=Ke^{2t}=Kh^2 δ = K e 2 t = K h 2 , that is, h ′ 2 = 1 − K h 2 h'^2=1-Kh^2 h ′2 = 1 − K h 2 and d r 2 = d h 2 / ( 1 − K h 2 ) dr^2=dh^2/(1-Kh^2) d r 2 = d h 2 / ( 1 − K h 2 ) . Since κ = 0 \kappa=0 κ = 0 there, A A A is constant and u = 1 u=1 u = 1 .
Positivity
We now prove that k ^ g > 0 \hat{k}_g>0 k ^ g > 0 on each piece of the handle. All estimates are made on the region where r > 0 r>0 r > 0 and 0 < ϑ < π 0<\vartheta<\pi 0 < ϑ < π , where the coframe of Subsection 4.1 is defined; the center of the disk and the poles of the sphere are treated at the end. Note that 0 ≤ κ ≤ 1 6 0\le\kappa\le\frac{1}{6} 0 ≤ κ ≤ 6 1 everywhere, so that
1 ≤ u ≤ 1 + κ 2 24 ≤ 865 864 . (46) 1\le u\le1+\frac{\kappa^2}{24}\le\frac{865}{864}. \tag*{(46)} 1 ≤ u ≤ 1 + 24 κ 2 ≤ 864 865 . (46)
Lemma 4.7 (Cap and cylinder). For 0 < r ≤ r 0 0 < r \le r_0 0 < r ≤ r 0 ,
u 3 k g = κ 0 2 12 h 0 2 + κ 0 4 z 2 ( 1 − z 2 ) 192 h 0 2 ≥ 1 432 h 0 2 . (47) u^3 k_{\mathfrak g} = \frac{\kappa_0^2}{12h_0^2}+\frac{\kappa_0^4z^2(1-z^2)}{192h_0^2}\ge\frac{1}{432h_0^2}. \tag*{(47)} u 3 k g = 12 h 0 2 κ 0 2 + 192 h 0 2 κ 0 4 z 2 ( 1 − z 2 ) ≥ 432 h 0 2 1 . (47) Proof. Here h h h and κ \kappa κ are constant, so that V = W = 0 \mathcal{V}=\mathcal{W}=0 V = W = 0 , D = 0 D=0 D = 0 , L r B = 0 L_r\mathcal{B}=0 L r B = 0 and U = h 0 − 2 \mathcal{U}=h_0^{-2} U = h 0 − 2 , while X = − f ′ ′ / f ≥ 0 \mathcal{X}=-f''/f\ge0 X = − f ′′ / f ≥ 0 . By (19) and (32) ,
2 h 0 2 G ± = 1 + h 0 2 X − κ 0 2 ( 1 − z 2 ) ∓ 3 κ 0 z ≥ 1 − 3 κ 0 = 1 2 . 2h_0^2G_{\pm}=1+h_0^2\mathcal{X}-\kappa_0^2(1-z^2)\mp3\kappa_0 z\ge1-3\kappa_0=\frac{1}{2}. 2 h 0 2 G ± = 1 + h 0 2 X − κ 0 2 ( 1 − z 2 ) ∓ 3 κ 0 z ≥ 1 − 3 κ 0 = 2 1 . Since the gaps are positive and D = 0 D=0 D = 0 , we have E = 0 \mathcal{E}=0 E = 0 , and (24) reduces to (47) . □
Note that the bound does not depend on the length ℓ c y l \ell_{\mathrm{cyl}} ℓ cyl of the cylinder.
Lemma 4.8 (Ramp and neck). On the ramp and on the neck,
u 3 k g > δ 2 500 h 2 . (48) u^3 k_{\mathfrak g}>\frac{\delta^2}{500h^2}. \tag*{(48)} u 3 k g > 500 h 2 δ 2 . (48) Proof. We work in the coordinate r r r , because h ′ h' h ′ vanishes at the beginning of the ramp. Recall that f = F f=F f = F , that κ = 1 6 δ \kappa=\frac{1}{6}\delta κ = 6 1 δ , hence B = δ 2 / 864 \mathcal{B}=\delta^2/864 B = δ 2 /864 , and that by (39) and Lemma 4.6(i)
h h ′ ′ = ϵ η δ 2 2 , h δ ′ = − ϵ η h ′ δ . (49) hh''=\frac{\epsilon\eta\delta^2}{2},\qquad h\delta'=-\epsilon\eta h'\delta. \tag*{(49)} h h ′′ = 2 ϵη δ 2 , h δ ′ = − ϵη h ′ δ . (49) The radial term. We claim that L r B ≤ 0 L_r\mathcal{B}\le0 L r B ≤ 0 . Differentiating B = δ 2 / 864 \mathcal{B}=\delta^2/864 B = δ 2 /864 and using (49) , we find
B ′ = δ δ ′ 432 = − ϵ η h ′ δ 3 432 h , \mathcal{B}'=\frac{\delta\delta'}{432}=-\frac{\epsilon\eta h'\delta^3}{432h}, B ′ = 432 δ δ ′ = − 432 h ϵη h ′ δ 3 , B ′ ′ = − ϵ 432 [ η ′ h ′ δ 3 h + η h ′ ′ δ 3 h + 3 η h ′ δ 2 δ ′ h − η h ′ 2 δ 3 h 2 ] . \mathcal{B}''=-\frac{\epsilon}{432}\left[\frac{\eta'h'\delta^3}{h}+\frac{\eta h''\delta^3}{h}+\frac{3\eta h'\delta^2\delta'}{h}-\frac{\eta h'^2\delta^3}{h^2}\right]. B ′′ = − 432 ϵ [ h η ′ h ′ δ 3 + h η h ′′ δ 3 + h 3 η h ′ δ 2 δ ′ − h 2 η h ′2 δ 3 ] .
and therefore, since h 2 L r B = h 2 B ′ ′ + 2 h h ′ B ′ h^2L_r\mathcal{B}=h^2\mathcal{B}''+2hh'\mathcal{B}' h 2 L r B = h 2 B ′′ + 2 h h ′ B ′ ,
h 2 L r B = − ϵ 432 [ η ′ h h ′ δ 3 + η h h ′ ′ δ 3 + 3 η h h ′ δ 2 δ ′ + η h ′ 2 δ 3 ] . h^2L_r\mathcal{B}=-\frac{\epsilon}{432}\left[\eta'hh'\delta^3+\eta hh''\delta^3+3\eta hh'\delta^2\delta'+\eta h'^2\delta^3\right]. h 2 L r B = − 432 ϵ [ η ′ h h ′ δ 3 + η h h ′′ δ 3 + 3 η h h ′ δ 2 δ ′ + η h ′2 δ 3 ] .
Substituting (49) gives
h 2 L r B = 1 864 [ − 2 ϵ η ′ h h ′ δ 3 + ϵ 2 η 2 δ 4 ( 6 h ′ 2 − δ ) − 2 ϵ η h ′ 2 δ 3 ] . (50) h^2L_r\mathcal{B}=\frac{1}{864}\left[-2\epsilon\eta'hh'\delta^3+\epsilon^2\eta^2\delta^4(6h'^2-\delta)-2\epsilon\eta h'^2\delta^3\right]. \tag*{(50)} h 2 L r B = 864 1 [ − 2 ϵ η ′ h h ′ δ 3 + ϵ 2 η 2 δ 4 ( 6 h ′2 − δ ) − 2 ϵη h ′2 δ 3 ] . (50)
The first term is nonpositive because η ′ ≥ 0 \eta'\geq0 η ′ ≥ 0 and h ′ ≥ 0 h'\geq0 h ′ ≥ 0 . With h ′ 2 = 1 − δ h'^2=1-\delta h ′2 = 1 − δ , the other two terms add up to
ϵ η δ 3 [ ϵ η δ ( 6 − 7 δ ) − 2 ( 1 − δ ) ] . \epsilon\eta\delta^3\left[\epsilon\eta\delta(6-7\delta)-2(1-\delta)\right]. ϵη δ 3 [ ϵη δ ( 6 − 7 δ ) − 2 ( 1 − δ ) ] .
This is nonpositive. Indeed, for δ ≥ 6 7 \delta\geq\frac{6}{7} δ ≥ 7 6 both summands in the bracket are nonpositive, and for δ < 6 7 \delta<\frac{6}{7} δ < 7 6 we use δ ( 6 − 7 δ ) ≤ 6 ( 1 − δ ) \delta(6-7\delta)\leq6(1-\delta) δ ( 6 − 7 δ ) ≤ 6 ( 1 − δ ) and ϵ η ≤ 1 3 \epsilon\eta\leq\frac{1}{3} ϵη ≤ 3 1 . Note that no upper bound for η ′ \eta' η ′ is needed.
The gaps and the error term. By (19) , (32) and (49) ,
2 h 2 G ± = δ + h h ′ ′ − κ 2 ( 1 − z 2 ) + 3 κ z ≥ δ + ϵ η δ 2 2 − 3 κ ≥ δ 2 . 2h^2G_{\pm}=\delta+hh''-\kappa^2(1-z^2)+3\kappa z\geq\delta+\frac{\epsilon\eta\delta^2}{2}-3\kappa\geq\frac{\delta}{2}. 2 h 2 G ± = δ + h h ′′ − κ 2 ( 1 − z 2 ) + 3 κ z ≥ δ + 2 ϵη δ 2 − 3 κ ≥ 2 δ .
Since κ ′ = 1 6 δ ′ \kappa'=\frac{1}{6}\delta' κ ′ = 6 1 δ ′ , we obtain from (21) , (19) and (49)
E ≤ D 2 min { G + , G − } ≤ ϵ 2 η 2 h ′ 2 δ 4 ( 1 − z 2 ) 576 h 4 ⋅ 4 h 2 δ ≤ ϵ 2 η 2 δ 3 144 h 2 . \mathcal{E}\leq\frac{D^2}{\min\{G_+,G_-\}}\leq\frac{\epsilon^2\eta^2h'^2\delta^4(1-z^2)}{576h^4}\cdot\frac{4h^2}{\delta}\leq\frac{\epsilon^2\eta^2\delta^3}{144h^2}. E ≤ min { G + , G − } D 2 ≤ 576 h 4 ϵ 2 η 2 h ′2 δ 4 ( 1 − z 2 ) ⋅ δ 4 h 2 ≤ 144 h 2 ϵ 2 η 2 δ 3 .
Conclusion. We insert these estimates into (24) , discard the nonnegative terms κ 4 z 2 ( 1 − z 2 ) / ( 192 h 2 ) \kappa^4z^2(1-z^2)/(192h^2) κ 4 z 2 ( 1 − z 2 ) / ( 192 h 2 ) and − 1 2 ( L r B ) z 2 -\frac{1}{2}(L_r\mathcal{B})z^2 − 2 1 ( L r B ) z 2 , and use (46) . This gives
u 3 k g ≥ − u ϵ η δ 2 4 h 2 + δ 2 432 h 2 − u ϵ 2 η 2 δ 3 144 h 2 ≥ δ 2 h 2 [ 1 432 − 865 864 ( ϵ 4 + ϵ 2 144 ) ] > δ 2 500 h 2 . u^3k_g\geq-\frac{u\epsilon\eta\delta^2}{4h^2}+\frac{\delta^2}{432h^2}-\frac{u\epsilon^2\eta^2\delta^3}{144h^2}\geq\frac{\delta^2}{h^2}\left[\frac{1}{432}-\frac{865}{864}\left(\frac{\epsilon}{4}+\frac{\epsilon^2}{144}\right)\right]>\frac{\delta^2}{500h^2}. u 3 k g ≥ − 4 h 2 u ϵη δ 2 + 432 h 2 δ 2 − 144 h 2 u ϵ 2 η 2 δ 3 ≥ h 2 δ 2 [ 432 1 − 864 865 ( 4 ϵ + 144 ϵ 2 ) ] > 500 h 2 δ 2 .
The last inequality holds because the bracket is larger than 0.00206. □
Lemma 4.9 (Stage I). In Stage I,
u 3 k ˙ g > δ 2 1000 h 2 . (51) u^3\dot{k}_g>\frac{\delta^2}{1000h^2}. \tag*{(51)} u 3 k ˙ g > 1000 h 2 δ 2 . (51)
Proof. Put λ = δ ˙ / δ \lambda=\dot{\delta}/\delta λ = δ ˙ / δ . By (41) , λ = − ϵ ( 1 − χ ) δ + 1 2 χ \lambda=-\epsilon(1-\chi)\delta+\frac{1}{2}\chi λ = − ϵ ( 1 − χ ) δ + 2 1 χ , and hence
− ϵ δ ≤ λ ≤ 1 2 , λ ˙ = χ ˙ ( 1 2 + ϵ δ ) − ϵ ( 1 − χ ) λ δ ≤ 1 16 . (52) -\epsilon\delta\leq\lambda\leq\frac{1}{2},\qquad\dot{\lambda}=\dot{\chi}\left(\frac{1}{2}+\epsilon\delta\right)-\epsilon(1-\chi)\lambda\delta\leq\frac{1}{16}. \tag*{(52)} − ϵδ ≤ λ ≤ 2 1 , λ ˙ = χ ˙ ( 2 1 + ϵδ ) − ϵ ( 1 − χ ) λ δ ≤ 16 1 . (52)
For the last inequality we use that 0 ≤ χ ˙ ≤ 1 10 0\leq\dot{\chi}\leq\frac{1}{10} 0 ≤ χ ˙ ≤ 10 1 by (34) and that δ ≤ 1 100 \delta\leq\frac{1}{100} δ ≤ 100 1 by Lemma 4.6(ii); the term − ϵ ( 1 − χ ) λ δ -\epsilon(1-\chi)\lambda\delta − ϵ ( 1 − χ ) λ δ is positive only if λ < 0 \lambda<0 λ < 0 , and then it is at most ϵ 2 δ 2 \epsilon^2\delta^2 ϵ 2 δ 2 .
We apply Lemma 4.5. Since κ = 1 6 δ \kappa=\frac{1}{6}\delta κ = 6 1 δ ,
δ − δ ˙ 2 − 3 κ = δ ( 1 − λ ) 2 ≥ δ 4 , \delta-\frac{\dot{\delta}}{2}-3\kappa=\frac{\delta(1-\lambda)}{2}\geq\frac{\delta}{4}, δ − 2 δ ˙ − 3 κ = 2 δ ( 1 − λ ) ≥ 4 δ ,
so that (31) gives h 2 E ≤ κ ˙ 2 / ( 2 δ ) h^2\mathcal{E}\leq\dot{\kappa}^2/(2\delta) h 2 E ≤ κ ˙ 2 / ( 2 δ ) . As κ ˙ = 1 6 λ δ \dot{\kappa}=\frac{1}{6}\lambda\delta κ ˙ = 6 1 λ δ , we obtain
h 2 E ≤ λ 2 δ 72 . h^2\mathcal{E}\leq\frac{\lambda^2\delta}{72}. h 2 E ≤ 72 λ 2 δ .
For B = δ 2 / 864 \mathcal{B}=\delta^2/864 B = δ 2 /864 we have B ˙ = λ δ 2 / 432 \dot{\mathcal{B}}=\lambda\delta^2/432 B ˙ = λ δ 2 /432 and B ¨ = δ 2 ( λ ˙ + 2 λ 2 ) / 432 \ddot{\mathcal{B}}=\delta^2(\dot{\lambda}+2\lambda^2)/432 B ¨ = δ 2 ( λ ˙ + 2 λ 2 ) /432 , and (29) becomes
h 2 L r B = δ 2 432 [ ( 1 − δ ) ( λ ˙ + 2 λ 2 + λ ) − λ 2 δ 2 ] ≤ δ 2 432 max { 0 , λ ˙ + 2 λ 2 + λ } . (53) h^2L_r\mathcal{B}=\frac{\delta^2}{432}\left[(1-\delta)(\dot{\lambda}+2\lambda^2+\lambda)-\frac{\lambda^2\delta}{2}\right]\leq\frac{\delta^2}{432}\max\{0,\dot{\lambda}+2\lambda^2+\lambda\}. \tag*{(53)} h 2 L r B = 432 δ 2 [ ( 1 − δ ) ( λ ˙ + 2 λ 2 + λ ) − 2 λ 2 δ ] ≤ 432 δ 2 max { 0 , λ ˙ + 2 λ 2 + λ } . (53)
Inserting these estimates into (30) and discarding the quartic term, we obtain
h 2 u 3 κ g ≥ u λ δ 4 + δ 2 432 − u λ 2 δ 72 − δ 2 864 max { 0 , λ ˙ + 2 λ 2 + λ } . (54) h^2u^3\kappa_g\geq\frac{u\lambda\delta}{4}+\frac{\delta^2}{432}-\frac{u\lambda^2\delta}{72}-\frac{\delta^2}{864}\max\{0,\dot{\lambda}+2\lambda^2+\lambda\}. \tag*{(54)} h 2 u 3 κ g ≥ 4 u λ δ + 432 δ 2 − 72 u λ 2 δ − 864 δ 2 max { 0 , λ ˙ + 2 λ 2 + λ } . (54)
We distinguish two cases. If 0 ≤ λ ≤ 1 2 0\leq\lambda\leq\frac{1}{2} 0 ≤ λ ≤ 2 1 , then
u λ δ 4 − u λ 2 δ 72 = u λ δ ( 1 4 − λ 72 ) ≥ 0 \frac{u\lambda\delta}{4}-\frac{u\lambda^2\delta}{72}=u\lambda\delta\left(\frac{1}{4}-\frac{\lambda}{72}\right)\geq0 4 u λ δ − 72 u λ 2 δ = u λ δ ( 4 1 − 72 λ ) ≥ 0
and
λ ˙ + 2 λ 2 + λ ≤ 1 16 + 1 2 + 1 2 = 17 16 , \dot{\lambda}+2\lambda^2+\lambda\leq\frac{1}{16}+\frac{1}{2}+\frac{1}{2}=\frac{17}{16}, λ ˙ + 2 λ 2 + λ ≤ 16 1 + 2 1 + 2 1 = 16 17 ,
so that (54) gives
h 2 u 3 κ g ≥ δ 2 ( 1 432 − 17 13824 ) = 5 δ 2 4608 > δ 2 1000 . h^2u^3\kappa_g\geq\delta^2\left(\frac{1}{432}-\frac{17}{13824}\right)=\frac{5\delta^2}{4608}>\frac{\delta^2}{1000}. h 2 u 3 κ g ≥ δ 2 ( 432 1 − 13824 17 ) = 4608 5 δ 2 > 1000 δ 2 .
If − ϵ δ ≤ λ < 0 -\epsilon\delta\leq\lambda<0 − ϵδ ≤ λ < 0 , then 2 λ 2 + λ ≤ 0 2\lambda^2+\lambda\leq0 2 λ 2 + λ ≤ 0 , so that the maximum in (54) is at most 1 16 \frac{1}{16} 16 1 . Using ∣ λ ∣ ≤ ϵ δ |\lambda|\leq\epsilon\delta ∣ λ ∣ ≤ ϵδ , δ ≤ 1 \delta\leq1 δ ≤ 1 and (46) , we obtain
h 2 u 3 κ g ≥ δ 2 [ 1 432 − 865 864 ( ϵ 4 + ϵ 2 72 ) − 1 13824 ] > δ 2 1000 , h^2u^3\kappa_g\geq\delta^2\left[\frac{1}{432}-\frac{865}{864}\left(\frac{\epsilon}{4}+\frac{\epsilon^2}{72}\right)-\frac{1}{13824}\right]>\frac{\delta^2}{1000}, h 2 u 3 κ g ≥ δ 2 [ 432 1 − 864 865 ( 4 ϵ + 72 ϵ 2 ) − 13824 1 ] > 1000 δ 2 ,
since the bracket is larger than 0.0019. □
Lemma 4.10 (Stage II). In Stage II,
u 3 κ g > δ 12 h 2 . (55) u^3\kappa_g>\frac{\delta}{12h^2}. \tag*{(55)} u 3 κ g > 12 h 2 δ . (55)
Proof. By (34) , the function ψ \psi ψ in (42) satisfies 0 ≤ ψ ≤ 1 0\leq\psi\leq1 0 ≤ ψ ≤ 1 , ∣ ψ ˙ ∣ ≤ 1 5 |\dot{\psi}|\leq\frac{1}{5} ∣ ψ ˙ ∣ ≤ 5 1 and ∣ ψ ¨ ∣ ≤ 64 100 |\ddot{\psi}|\leq\frac{64}{100} ∣ ψ ¨ ∣ ≤ 100 64 ; we shall only use that ∣ ψ ˙ ∣ , ∣ ψ ¨ ∣ ≤ 1 |\dot{\psi}|,|\ddot{\psi}|\leq1 ∣ ψ ˙ ∣ , ∣ ψ ¨ ∣ ≤ 1 . From δ ˙ = 1 2 δ \dot{\delta}=\frac{1}{2}\delta δ ˙ = 2 1 δ and κ = 1 6 δ ψ \kappa=\frac{1}{6}\delta\psi κ = 6 1 δ ψ we get
κ ˙ = δ 6 ( ψ ˙ + ψ 2 ) , κ ¨ = δ 6 ( ψ ¨ + ψ ˙ + ψ 4 ) , \dot{\kappa}=\frac{\delta}{6}\left(\dot{\psi}+\frac{\psi}{2}\right),\qquad
\ddot{\kappa}=\frac{\delta}{6}\left(\ddot{\psi}+\dot{\psi}+\frac{\psi}{4}\right), κ ˙ = 6 δ ( ψ ˙ + 2 ψ ) , κ ¨ = 6 δ ( ψ ¨ + ψ ˙ + 4 ψ ) ,
hence
∣ κ ˙ ∣ ≤ δ 4 , ∣ κ ¨ ∣ ≤ 3 δ 8 . |\dot{\kappa}|\leq\frac{\delta}{4},\qquad|\ddot{\kappa}|\leq\frac{3\delta}{8}. ∣ κ ˙ ∣ ≤ 4 δ , ∣ κ ¨ ∣ ≤ 8 3 δ .
Since δ − 1 2 δ ˙ − 3 κ = 3 4 δ − 1 2 δ ψ ≥ 1 4 δ \delta-\frac{1}{2}\dot{\delta}-3\kappa=\frac{3}{4}\delta-\frac{1}{2}\delta\psi\geq\frac{1}{4}\delta δ − 2 1 δ ˙ − 3 κ = 4 3 δ − 2 1 δ ψ ≥ 4 1 δ , Lemma 4.5 applies, and (31) gives
h 2 E ≤ κ ˙ 2 2 δ ≤ δ 32 . h^2\mathcal{E}\leq\frac{\dot{\kappa}^2}{2\delta}\leq\frac{\delta}{32}. h 2 E ≤ 2 δ κ ˙ 2 ≤ 32 δ .
For B = κ 2 / 24 \mathcal{B}=\kappa^2/24 B = κ 2 /24 the bounds above imply
∣ B ˙ ∣ = ∣ κ κ ˙ ∣ 12 ≤ δ 2 288 , ∣ B ¨ ∣ ≤ κ 2 + ∣ κ κ ¨ ∣ 12 ≤ δ 2 96 . |\dot{\mathcal{B}}|=\frac{|\kappa\dot{\kappa}|}{12}\leq\frac{\delta^2}{288},\qquad
|\ddot{\mathcal{B}}|\leq\frac{\kappa^2+|\kappa\ddot{\kappa}|}{12}\leq\frac{\delta^2}{96}. ∣ B ˙ ∣ = 12 ∣ κ κ ˙ ∣ ≤ 288 δ 2 , ∣ B ¨ ∣ ≤ 12 κ 2 + ∣ κ κ ¨ ∣ ≤ 96 δ 2 .
The coefficients in (29) are 1 − δ 1-\delta 1 − δ and 1 − 5 4 δ 1-\frac{5}{4}\delta 1 − 4 5 δ , and both lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] because δ ≤ 1 100 \delta\leq\frac{1}{100} δ ≤ 100 1 by (45) . Hence ∣ h 2 L r B ∣ ≤ 1 96 δ 2 + 1 288 δ 2 = 1 72 δ 2 |h^2L_r\mathcal{B}|\leq\frac{1}{96}\delta^2+\frac{1}{288}\delta^2=\frac{1}{72}\delta^2 ∣ h 2 L r B ∣ ≤ 96 1 δ 2 + 288 1 δ 2 = 72 1 δ 2 . Since 1 4 u δ ˙ = 1 8 u δ \frac{1}{4}u\dot{\delta}=\frac{1}{8}u\delta 4 1 u δ ˙ = 8 1 u δ , the identity (30) yields
h 2 u 3 κ g ≥ u δ 8 − u δ 32 − δ 2 144 = 3 u δ 32 − δ 2 144 > δ 12 , h^2u^3\kappa_g\geq\frac{u\delta}{8}-\frac{u\delta}{32}-\frac{\delta^2}{144}=\frac{3u\delta}{32}-\frac{\delta^2}{144}>\frac{\delta}{12}, h 2 u 3 κ g ≥ 8 u δ − 32 u δ − 144 δ 2 = 32 3 u δ − 144 δ 2 > 12 δ ,
where we used u ≥ 1 u\geq1 u ≥ 1 and δ ≤ 1 100 \delta\leq\frac{1}{100} δ ≤ 100 1 . □
Lemma 4.11 (Stage III). In Stage III we have g ^ = g \hat{g}=g g ^ = g and
k g = 1 h 2 min { δ 2 , δ ˙ 4 } > 0. (56) k_g=\frac{1}{h^2}\min\left\{\frac{\delta}{2},\frac{\dot{\delta}}{4}\right\}>0. \tag*{(56)} k g = h 2 1 min { 2 δ , 4 δ ˙ } > 0. (56) Near t = t 3 t=t_3 t = t 3 this gives k g = 1 2 K k_g=\frac{1}{2}K k g = 2 1 K .
Proof. Here κ = 0 \kappa=0 κ = 0 , so that u = 1 u=1 u = 1 and g g g is a warped product with constant f f f . By (22) and (27) , k g = min { 1 2 U , 1 2 V } k_g=\min\{\frac{1}{2}\mathcal{U},\frac{1}{2}\mathcal{V}\} k g = min { 2 1 U , 2 1 V } with U = δ / h 2 \mathcal{U}=\delta/h^2 U = δ / h 2 and V = δ ˙ / ( 2 h 2 ) \mathcal{V}=\dot{\delta}/(2h^2) V = δ ˙ / ( 2 h 2 ) . Both are positive by Lemma 4.6 (iii). Near t 3 t_3 t 3 we have δ = K h 2 \delta=Kh^2 δ = K h 2 and δ ˙ = 2 K h 2 \dot{\delta}=2Kh^2 δ ˙ = 2 K h 2 . □
Proof of Theorem 4.1
Proof of Theorem 4.1. It remains to fix the length ℓ c y l \ell_{\mathrm{cyl}} ℓ cyl of the cylinder, to verify that g ^ \hat{g} g ^ is a smooth metric on D 2 × S 2 D^2\times S^2 D 2 × S 2 , and to collect the results.
Holonomy and framing. By Lemma 4.6 (v), the coefficient of d θ d\theta d θ in the boundary collar is the total rotation A ( ℓ ) = ∫ 0 ℓ f κ h − 2 d r A(\ell)=\int_0^\ell f\kappa h^{-2}\,dr A ( ℓ ) = ∫ 0 ℓ f κ h − 2 d r . Let A 0 A_0 A 0 be its value for ℓ c y l = 0 \ell_{\mathrm{cyl}}=0 ℓ cyl = 0 . For general ℓ c y l ≥ 0 \ell_{\mathrm{cyl}}\geq0 ℓ cyl ≥ 0 the cylinder contributes F ℓ c y l κ 0 / h 0 2 = F ℓ c y l / ( 6 h 0 2 ) F\ell_{\mathrm{cyl}}\kappa_0/h_0^2=F\ell_{\mathrm{cyl}}/(6h_0^2) F ℓ cyl κ 0 / h 0 2 = F ℓ cyl / ( 6 h 0 2 ) to this integral, while all the later pieces are merely translated in r r r , so that their contribution does not change. Thus A ( ℓ ) = A 0 + F ℓ c y l / ( 6 h 0 2 ) A(\ell)=A_0+F\ell_{\mathrm{cyl}}/(6h_0^2) A ( ℓ ) = A 0 + F ℓ cyl / ( 6 h 0 2 ) .
We now choose
n = ν + 2 ⌊ A 0 − α − ν 2 ⌋ , ℓ c y l = 6 h 0 2 F ( α + n − A 0 ) . (57) n=\nu+2\left\lfloor\frac{A_0-\alpha-\nu}{2}\right\rfloor,\qquad
\ell_{\mathrm{cyl}}=\frac{6h_0^2}{F}(\alpha+n-A_0). \tag*{(57)} n = ν + 2 ⌊ 2 A 0 − α − ν ⌋ , ℓ cyl = F 6 h 0 2 ( α + n − A 0 ) . (57)
Then n ≡ ν ( m o d 2 ) n\equiv\nu\pmod2 n ≡ ν ( mod 2 ) . Moreover, 0 ≤ α + n − A 0 < 2 0\leq\alpha+n-A_0<2 0 ≤ α + n − A 0 < 2 , so that ℓ c y l ≥ 0 \ell_{\mathrm{cyl}}\geq0 ℓ cyl ≥ 0 , and A ( ℓ ) = α + n A(\ell)=\alpha+n A ( ℓ ) = α + n . By Lemma 4.6 (v), the metric near r = ℓ r=\ell r = ℓ is (15) . We define R − R_- R − as a radius such that this holds for R − < h ≤ R R_-<h\leq R R − < h ≤ R .
Smoothness at the center of the disk and at the poles. Near r = 0 r=0 r = 0 we have
f = r , h = h 0 , κ = 1 6 , A = r 2 12 h 0 2 . f=r,\qquad h=h_0,\qquad\kappa=\frac{1}{6},\qquad A=\frac{r^2}{12h_0^2}. f = r , h = h 0 , κ = 6 1 , A = 12 h 0 2 r 2 .
In the Cartesian coordinates x = r cos θ x=r\cos\theta x = r cos θ , y = r sin θ y=r\sin\theta y = r sin θ on the disk,
d r 2 + f 2 d θ 2 = d x 2 + d y 2 , A d θ = x d y − y d x 12 h 0 2 , dr^2+f^2d\theta^2=dx^2+dy^2,\qquad
A\,d\theta=\frac{x\,dy-y\,dx}{12h_0^2}, d r 2 + f 2 d θ 2 = d x 2 + d y 2 , A d θ = 12 h 0 2 x d y − y d x ,
and the latter is a smooth one-form. In terms of the coordinates ( z , w ) (z,w) ( z , w ) on the sphere, the fiber part of (16) is
h 2 ( d z 2 + ∣ d w + i w A d θ ∣ 2 ) on z 2 + ∣ w ∣ 2 = 1 , h^2\left(dz^2+\lvert dw+iwA\,d\theta\rvert^2\right)\quad\text{on }z^2+\lvert w\rvert^2=1, h 2 ( d z 2 + ∣ d w + i w A d θ ∣ 2 ) on z 2 + ∣ w ∣ 2 = 1 ,
which is smooth at the poles w = 0 w=0 w = 0 , and also where a pole meets the center of the disk. Hence g g g is a smooth metric on D 2 × S 2 D^2\times S^2 D 2 × S 2 . The conformal factor u = 1 + κ 2 z 2 / 24 u=1+\kappa^2z^2/24 u = 1 + κ 2 z 2 /24 is smooth and positive, because κ \kappa κ is constant near r = 0 r=0 r = 0 and smooth for r > 0 r>0 r > 0 . Therefore g ^ = u 2 g \hat{g}=u^2g g ^ = u 2 g is a smooth Riemannian metric on D 2 × S 2 D^2\times S^2 D 2 × S 2 .
Positivity. By Lemmas 4.7 –4.11 , k g ^ > 0 k_{\hat{g}}>0 k g ^ > 0 on the dense open set where r > 0 r>0 r > 0 and 0 < ϑ < π 0<\vartheta<\pi 0 < ϑ < π . In fact, the lower bounds proved there are bounded below by a positive constant, because h ≤ R h\leq R h ≤ R and u ≤ 865 864 u\leq\frac{865}{864} u ≤ 864 865 , and because the continuous functions δ \delta δ on [ r 0 , ℓ ] [r_0,\ell] [ r 0 , ℓ ] and δ ˙ \dot{\delta} δ ˙ on [ t 2 , t 3 ] [t_2,t_3] [ t 2 , t 3 ] are positive on compact intervals. Since k g ^ k_{\hat{g}} k g ^ is continuous on D 2 × S 2 D^2\times S^2 D 2 × S 2 , it is positive everywhere. This completes the proof of Theorem 4.1 . □
Remark 4.12. We have made no attempt to optimize the construction, and the handle is extremely thin. With K R 2 = 1 100 K R^{2} = \frac{1}{100} K R 2 = 100 1 , the defect at the end of the neck is δ 0 ≈ 10 − 10 \delta_{0} \approx10^{-10} δ 0 ≈ 1 0 − 10 , and (37) gives h 0 ≈ R exp ( − 10 13 ) h_{0} \approx R \exp(-10^{13}) h 0 ≈ R exp ( − 1 0 13 ) . Accordingly, the total rotation A 0 A_{0} A 0 of the connection and the integer n n n in (57) are very large. This is harmless, since only the parity of n n n has a topological meaning and we do not claim any effective bound on the geometry of the handle.
Proofs of the main results
Surgery along closed geodesics
We first combine the results of Sections 3 and 4.
Proposition 5.1. Let ( M , g ) (M,g) ( M , g ) be a closed oriented Riemannian four-manifold with k g > 0 k_{g} > 0 k g > 0 , and let γ 1 , … , γ m \gamma_{1},\ldots,\gamma_{m} γ 1 , … , γ m be pairwise disjoint embedded closed geodesics of g g g . Then, for any choice of normal framings, the manifold obtained from M M M by surgery along γ 1 , … , γ m \gamma_{1},\ldots,\gamma_{m} γ 1 , … , γ m admits a metric of positive biorthogonal curvature. This metric can be chosen to agree with g g g outside any given neighborhood of the geodesics.
Proof. We choose pairwise disjoint neighborhoods of the geodesics, as small as we wish, and apply Lemma 3.5 with K = 1 K=1 K = 1 . This deforms g g g inside these neighborhoods, through metrics of positive biorthogonal curvature, to a metric that is a round tube near each γ j \gamma_{j} γ j . Let 2 π F j 2\pi F_{j} 2 π F j be the length of γ j \gamma_{j} γ j . By Lemma 3.6, we can choose α j ∈ R \alpha_{j} \in\mathbb{R} α j ∈ R and coordinates ( θ , h , ( z , w ) ) (\theta,h,(z,w)) ( θ , h , ( z , w )) near γ j \gamma_{j} γ j in which the new metric is (12) with F = F j F=F_{j} F = F j , K = 1 K=1 K = 1 and α = α j \alpha=\alpha_{j} α = α j , and we choose α j \alpha_{j} α j , within its class modulo integers, such that these coordinates induce the given framing of γ j \gamma_{j} γ j . We then fix R j > 0 R_{j}>0 R j > 0 with R j 2 ≤ 1 100 R_{j}^{2}\leq\frac{1}{100} R j 2 ≤ 100 1 so small that the level set { h = R j } \{h=R_{j}\} { h = R j } lies in the domain of these coordinates.
Theorem 4.1, applied with ( F , K , R , α , ν ) = ( F j , 1 , R j , α j , 0 ) (F,K,R,\alpha,\nu)=(F_{j},1,R_{j},\alpha_{j},0) ( F , K , R , α , ν ) = ( F j , 1 , R j , α j , 0 ) , provides a metric of positive biorthogonal curvature on D 2 × S 2 D^{2}\times S^{2} D 2 × S 2 whose boundary collar is (15) with an even integer n j n_{j} n j . We remove the open tubes { h < R j } \{h<R_{j}\} { h < R j } and glue in these handles by means of the maps T n j T_{n_{j}} T n j , as described in Remark 3.7. The metric is smooth across the seams because it is given by the same expression on both sides, and it has positive biorthogonal curvature everywhere. Since n j n_{j} n j is even, the resulting manifold is the result of surgery along the γ j \gamma_{j} γ j with the given framings. □ \square □
Proof of Theorem 1.2
Proof of Theorem 1.2. Step 1. Necessity, and the choice of a presentation. The fundamental group of a closed smooth manifold is finitely presented, because the manifold has a finite triangulation.
Conversely, let Γ \Gamma Γ be a finitely presented group. By Lemma 3.1 and its proof, Γ \Gamma Γ has a presentation with N N N generators and m m m relators that are primitive in the free group F N F_{N} F N and pairwise nonconjugate, even after inversion; if we start from a presentation with p p p generators and q q q relators, none of which is trivial in F p F_{p} F p , then N = p + q N=p+q N = p + q and m = 2 q m=2q m = 2 q .
Step 2. Geodesics that represent the relators. If N = 0 N=0 N = 0 , then Γ \Gamma Γ is trivial and the round four-sphere has the required properties. Suppose that N ≥ 1 N\geq1 N ≥ 1 , and let
M 0 = # N ( S 1 × S 3 ) M_{0}=\mathbin{\#}^{N}(S^{1}\times S^{3}) M 0 = # N ( S 1 × S 3 )
be the connected sum of N N N copies of S 1 × S 3 S^1 \times S^3 S 1 × S 3 , so that π 1 ( M 0 ) ≅ F N \pi_1(M_0) \cong F_N π 1 ( M 0 ) ≅ F N . The product of a circle and a round three-sphere has constant biorthogonal curvature 1 2 \frac{1}{2} 2 1 , as we saw in the proof of Lemma 3.5 , and Bettiol’s connected-sum theorem [4 Proposition 3.1] gives a metric g 0 g_0 g 0 on M 0 M_0 M 0 with k g 0 > 0 k_{g_0} > 0 k g 0 > 0 . If m = 0 m = 0 m = 0 , we are done.
Otherwise we apply Lemma 3.2 to the free homotopy classes of the relators. Its hypotheses are satisfied, since a primitive element of a free group is nontrivial and not a proper power, and the relators are pairwise nonconjugate even after inversion. We obtain a metric of positive biorthogonal curvature on M 0 M_0 M 0 and pairwise disjoint embedded closed geodesics γ 1 , … , γ m \gamma_1,\ldots,\gamma_m γ 1 , … , γ m of this metric that represent the relators.
Step 3. Surgery. Let M M M be the manifold obtained from M 0 M_0 M 0 by surgery along γ 1 , … , γ m \gamma_1,\ldots,\gamma_m γ 1 , … , γ m , with arbitrary framings. It is a closed connected oriented smooth four-manifold, it admits a metric of positive biorthogonal curvature by Proposition 5.1, and by Lemma 3.8
π 1 ( M ) ≅ F N / ⟨ ⟨ [ γ 1 ] , … , [ γ m ] ⟩ ⟩ ≅ Γ . \pi_1(M) \cong F_N/\langle\langle[\gamma_1],\ldots,[\gamma_m]\rangle\rangle\cong\Gamma. π 1 ( M ) ≅ F N / ⟨⟨[ γ 1 ] , … , [ γ m ]⟩⟩ ≅ Γ.
This proves the theorem.
Proof of Corollary 1.3
Proof of Corollary 1.3. Let p p p and q q q be as in the corollary. If p + q = 0 p+q=0 p + q = 0 , the round sphere satisfies all the assertions. Otherwise, let M M M be the manifold constructed above; it is obtained from M 0 = # p + q ( S 1 × S 3 ) M_0=\#^{p+q}(S^1 \times S^3) M 0 = # p + q ( S 1 × S 3 ) by surgery along 2 q 2q 2 q circles, and we are free to choose the framings. Since χ ( M 0 ) = 2 − 2 ( p + q ) \chi(M_0)=2-2(p+q) χ ( M 0 ) = 2 − 2 ( p + q ) and since replacing S 1 × D 3 S^1 \times D^3 S 1 × D 3 by D 2 × S 2 D^2 \times S^2 D 2 × S 2 increases the Euler characteristic by two,
χ ( M ) = 2 − 2 ( p + q ) + 4 q = 2 ( 1 − p + q ) . (58) \chi(M)=2-2(p+q)+4q=2(1-p+q). \tag*{(58)} χ ( M ) = 2 − 2 ( p + q ) + 4 q = 2 ( 1 − p + q ) . (58)
To make M M M stably parallelizable, we regard M 0 M_0 M 0 as the boundary of the five-dimensional handlebody V V V with one zero-handle and p + q p+q p + q one-handles. Since V V V is homotopy equivalent to a graph, H 2 ( V ; Z / 2 ) = 0 H^2(V;\mathbb{Z}/2)=0 H 2 ( V ; Z /2 ) = 0 , and V V V is spin; we fix a spin structure on V V V . The surgeries on M 0 = ∂ V M_0=\partial V M 0 = ∂ V are the effect on the boundary of attaching five-dimensional two-handles D 2 × D 3 D^2 \times D^3 D 2 × D 3 to V V V along the circles γ j \gamma_j γ j . The two framings of γ j \gamma_j γ j differ by the generator of π 1 ( SO ( 3 ) ) ≅ Z / 2 \pi_1(\operatorname{SO}(3)) \cong\mathbb{Z}/2 π 1 ( SO ( 3 )) ≅ Z /2 , and the spin structure of V V V extends over the corresponding two-handle for exactly one of them [23 pp. 199–202] . We choose these framings. The resulting five-manifold W W W is compact and spin, and ∂ W = M \partial W=M ∂ W = M . Moreover, W W W is homotopy equivalent to a two-dimensional complex, because it is built from handles of index at most two. An oriented vector bundle of rank five over a two-dimensional complex is trivial if its second Stiefel–Whitney class vanishes. Indeed, B S O ( 5 ) BSO(5) B S O ( 5 ) is simply connected and π 2 ( B S O ( 5 ) ) ≅ Z / 2 \pi_2(BSO(5)) \cong\mathbb{Z}/2 π 2 ( B S O ( 5 )) ≅ Z /2 , so that the only obstruction to a null-homotopy of the classifying map lies in the second cohomology group with coefficients in Z / 2 \mathbb{Z}/2 Z /2 , and this obstruction is the second Stiefel–Whitney class. Hence the tangent bundle of W W W is trivial. Its restriction to M M M is the sum of the tangent bundle of M M M and a trivial line bundle, and so M M M is stably parallelizable. Finally, Poincaré duality gives χ ( M ) = 2 − 2 b 1 ( M ) + b 2 ( M ) \chi(M)=2-2b_1(M)+b_2(M) χ ( M ) = 2 − 2 b 1 ( M ) + b 2 ( M ) , and b 1 ( M ) = b 1 ( Γ ) b_1(M)=b_1(\Gamma) b 1 ( M ) = b 1 ( Γ ) because H 1 ( M ; Q ) H_1(M;\mathbb{Q}) H 1 ( M ; Q ) is the rationalized abelianization of Γ \Gamma Γ . Together with (58) this yields b 2 ( M ) = 2 ( q − p + b 1 ( Γ ) ) b_2(M)=2(q-p+b_1(\Gamma)) b 2 ( M ) = 2 ( q − p + b 1 ( Γ )) .
For later use we record what the proof gives.
Remark 5.2. The manifold M M M of Corollary 1.3 is the boundary of a compact parallelizable five-manifold W W W such that the inclusion M ⊂ W M \subset W M ⊂ W induces an isomorphism of fundamental groups. Indeed, W W W is the union of V V V and the trace of the surgeries, glued along M 0 M_0 M 0 ; the inclusion M 0 ⊂ V M_0 \subset V M 0 ⊂ V induces an isomorphism of fundamental groups, and so the claim follows from Lemma 3.8 and van Kampen’s theorem. By construction, W W W has a handle decomposition with one zero-handle, p + q p+q p + q one-handles and 2 q 2q 2 q two-handles.
Remark 5.3. The auxiliary generators y j y_j y j are needed to produce the geodesics, but they do not affect the manifold. To see this, we slide, for each j j j , the two-handle of W W W that is attached along a circle representing r j y j r_j y_j r j y j over the two-handle that is attached along a circle representing y j y_j y j . For a suitable choice of the slides, the new attaching circles represent r 1 , … , r q r_1,\ldots,r_q r 1 , … , r q . By Lemma 5.5 below we may assume that these q q q circles lie in the part of M 0 M_0 M 0 that comes from the first p p p summands S 1 × S 3 S^1 \times S^3 S 1 × S 3 , and that the attaching circle representing y j y_j y j is the circle S 1 × { x } S^1 \times\{x\} S 1 × { x } in the summand that corresponds to y j y_j y j . For each j j j , the one-handle that corresponds to y j y_j y j and the two-handle attached along S 1 × { x } S^1 \times\{x\} S 1 × { x } then cancel. Hence W W W is diffeomorphic to a parallelizable handlebody with one zero-handle, p p p one-handles, and q q q two-handles attached along circles that represent r 1 , … , r q r_1,\ldots,r_q r 1 , … , r q . A regular neighborhood of the presentation complex of ⟨ x 1 , … , x p ∣ r 1 , … , r q ⟩ \langle x_1,\ldots,x_p \mid r_1,\ldots,r_q\rangle ⟨ x 1 , … , x p ∣ r 1 , … , r q ⟩ , embedded in R 5 \mathbb{R}^5 R 5 , is a parallelizable handlebody of the same form. Thus the manifolds of Corollary 1.3 are of the kind used by Carr [5 ] , who proved that a regular neighborhood of a subcomplex of codimension at least three in a Riemannian manifold can be chosen in such a way that its boundary has positive scalar curvature. For positive scalar curvature, statements corresponding to Corollary 1.3 and Remark 5.2 are recorded in [21 Theorem 6] .
Remark 5.4. Corollary 1.3 applies to groups with torsion as well. For example, the presentation ⟨ x ∣ x a ⟩ \langle x \mid x^a\rangle ⟨ x ∣ x a ⟩ of the cyclic group of order a ≥ 2 a \ge2 a ≥ 2 has p = q = 1 p=q=1 p = q = 1 and gives a spin manifold with χ = 2 \chi=2 χ = 2 and b 2 = 0 b_2=0 b 2 = 0 , that is, a rational homology four-sphere. For the standard presentation of Z d \mathbb{Z}^d Z d with d d d generators and ( d 2 ) \binom{d}{2} ( 2 d ) commutator relators we obtain
χ ( M ) = ( d − 1 ) ( d − 2 ) , b 2 ( M ) = d ( d − 1 ) . \chi(M)=(d-1)(d-2), \qquad b_2(M)=d(d-1). χ ( M ) = ( d − 1 ) ( d − 2 ) , b 2 ( M ) = d ( d − 1 ) .
In particular, the manifold that realizes Z 4 \mathbb{Z}^4 Z 4 has χ = 6 \chi=6 χ = 6 and b 2 = 12 b_2=12 b 2 = 12 ; it is not the four-torus, which admits no metric of positive scalar curvature [28 ] . Theorem 1.2 prescribes the fundamental group, not the manifold. We shall see in Remark 6.6 that the manifold constructed here has the smallest Euler characteristic among all closed oriented four-manifolds of positive scalar curvature with fundamental group Z d \mathbb{Z}^d Z d .
Proof of Theorem 1.4
The proof of Theorem 1.2 uses surgery along closed geodesics only. In order to pass to arbitrary circles we use two facts from differential topology. The first is that in dimension four circles cannot be knotted or linked.
Lemma 5.5. Let M M M be a closed connected smooth manifold of dimension at least four. Let c 1 , … , c m c_1,\ldots,c_m c 1 , … , c m and c 1 ′ , … , c m ′ c'_1,\ldots,c'_m c 1 ′ , … , c m ′ be two families of pairwise disjoint embedded circles in M M M such that c j c_j c j is freely homotopic to c j ′ c'_j c j ′ for every j j j , for suitable orientations. Then there is a diffeomorphism of M M M , isotopic to the identity, that maps c j c_j c j onto c j ′ c'_j c j ′ for every j j j .
Proof. The two families define embeddings of the disjoint union of m m m circles into M M M , and these are homotopic by assumption. Since dim M ≥ 2 ⋅ 1 + 2 \dim M \ge2 \cdot1 + 2 dim M ≥ 2 ⋅ 1 + 2 , homotopic embeddings are isotopic by Whitney’s general position argument [32 ] ; see also [12 ] . The reason is that during a generic one-parameter family of maps from a one-dimensional manifold to a manifold of dimension at least four, no double points and no singular points occur. By the isotopy extension theorem [17 Chapter 8, Theorem 1.3] , the isotopy is induced by an isotopy of M M M . □
The second fact is that surgery along the circle factor undoes a connected sum with S 1 × S 3 S^{1} \times S^{3} S 1 × S 3 .
Lemma 5.6. Surgery on S 1 × S 3 S^{1} \times S^{3} S 1 × S 3 along the circle S 1 × { x } S^{1} \times\{x\} S 1 × { x } , with either normal framing, yields S 4 S^{4} S 4 .
Proof. Write S 3 S^{3} S 3 as the union of two balls D + 3 D^{3}_{+} D + 3 and D − 3 D^{3}_{-} D − 3 with x x x the center of D + 3 D^{3}_{+} D + 3 . Then S 1 × D + 3 S^{1} \times D^{3}_{+} S 1 × D + 3 is a tubular neighborhood of the circle, and the result of the surgery is D 2 × S 2 ∪ φ S 1 × D − 3 D^{2} \times S^{2} \mathop{\cup}_{\varphi} S^{1} \times D^{3}_{-} D 2 × S 2 ∪ φ S 1 × D − 3 , where the gluing diffeomorphism φ \varphi φ of S 1 × S 2 S^{1} \times S^{2} S 1 × S 2 is the identity for one framing and the map T 1 T_{1} T 1 of (13) for the other. Since T 1 T_{1} T 1 extends to a diffeomorphism of S 1 × D − 3 S^{1} \times D^{3}_{-} S 1 × D − 3 , given by the same formula, both manifolds are diffeomorphic to D 2 × S 2 ∪ i d S 1 × D 3 = ∂ ( D 2 × D 3 ) = S 4 D^{2} \times S^{2} \mathop{\cup}_{\mathrm{id}} S^{1} \times D^{3} = \partial(D^{2} \times D^{3}) = S^{4} D 2 × S 2 ∪ id S 1 × D 3 = ∂ ( D 2 × D 3 ) = S 4 . □
Proof of Theorem 1.4. Surgery along finitely many pairwise disjoint circles can be performed one circle at a time, and the result of each step is again a closed connected oriented manifold. It therefore suffices to consider a single circle. Let M M M be as in the theorem, let c ⊂ M c \subset M c ⊂ M be an embedded circle, and let M ′ M' M ′ be the result of surgery along c c c with one of its two framings. We write Γ = π 1 ( M ) \Gamma= \pi_{1}(M) Γ = π 1 ( M ) , and we let r ∈ Γ r \in\Gamma r ∈ Γ be an element in the conjugacy class determined by c c c , for some orientation of c c c .
Step 1. Stabilization. Let
M 2 = M # ( S 1 × S 3 ) # ( S 1 × S 3 ) , M_{2} = M \mathbin{\#} (S^{1} \times S^{3}) \mathbin{\#} (S^{1} \times S^{3}), M 2 = M # ( S 1 × S 3 ) # ( S 1 × S 3 ) ,
where the connected sums are formed in two balls that are disjoint from c c c . Then π 1 ( M 2 ) ≅ Γ ∗ F 2 \pi_{1}(M_{2}) \cong\Gamma* F_{2} π 1 ( M 2 ) ≅ Γ ∗ F 2 , where the free group F 2 F_{2} F 2 is generated by the classes y 1 , y 2 y_{1}, y_{2} y 1 , y 2 of the circles a i = S 1 × { x i } a_{i} = S^{1} \times\{x_{i}\} a i = S 1 × { x i } in the two new summands. Let c ′ ′ ⊂ M 2 c'' \subset M_{2} c ′′ ⊂ M 2 be an embedded circle, disjoint from a 1 a_{1} a 1 and a 2 a_{2} a 2 , that represents r y 1 y 2 r y_{1} y_{2} r y 1 y 2 ; for instance, we can take a connected sum of c c c with parallel copies of a 1 a_{1} a 1 and a 2 a_{2} a 2 along two bands.
We claim that M ′ M' M ′ is obtained from M 2 M_{2} M 2 by surgery along a 1 a_{1} a 1 , a 2 a_{2} a 2 and c ′ ′ c'' c ′′ , for arbitrary framings of a 1 a_{1} a 1 and a 2 a_{2} a 2 and a suitable framing of c ′ ′ c'' c ′′ . Indeed, by Lemma 5.6 , surgery along a 1 a_{1} a 1 and a 2 a_{2} a 2 turns M 2 M_{2} M 2 into M # S 4 # S 4 = M M \mathbin{\#} S^{4} \mathbin{\#} S^{4} = M M # S 4 # S 4 = M , and this identification is the identity outside the two balls. By Lemma 3.8 , it induces on fundamental groups the homomorphism Γ ∗ F 2 → Γ \Gamma* F_{2} \to\Gamma Γ ∗ F 2 → Γ that is the identity on Γ \Gamma Γ and kills y 1 y_{1} y 1 and y 2 y_{2} y 2 . Therefore c ′ ′ c'' c ′′ , regarded as a circle in M M M , is freely homotopic to c c c . By Lemma 5.5 , there is a diffeomorphism of M M M that maps c ′ ′ c'' c ′′ onto c c c ; it maps the two framings of c ′ ′ c'' c ′′ to the two framings of c c c . This proves the claim.
Step 2. Geodesics. The manifold M 2 M_{2} M 2 admits a metric of positive biorthogonal curvature by Bettiol’s connected-sum theorem. Consider the homomorphism Γ ∗ F 2 → Z 2 \Gamma* F_{2} \to\mathbb{Z}^{2} Γ ∗ F 2 → Z 2 that is trivial on Γ \Gamma Γ and sends y 1 , y 2 y_{1}, y_{2} y 1 , y 2 to the standard basis vectors e 1 , e 2 e_{1}, e_{2} e 1 , e 2 . It maps y 1 , y 2 y_{1}, y_{2} y 1 , y 2 and r y 1 y 2 r y_{1} y_{2} r y 1 y 2 to e 1 , e 2 e_{1}, e_{2} e 1 , e 2 and e 1 + e 2 e_{1} + e_{2} e 1 + e 2 . None of these vectors is divisible by an integer d ≥ 2 d \ge2 d ≥ 2 , they are pairwise different, and none of them is the negative of another. Hence the three elements are nontrivial, they are not proper powers, and they are pairwise nonconjugate, even after inversion. Lemma 3.2 therefore provides a metric of positive biorthogonal curvature on M 2 M_{2} M 2 together with pairwise disjoint embedded closed geodesics γ 1 , γ 2 , γ 0 \gamma_{1}, \gamma_{2}, \gamma_{0} γ 1 , γ 2 , γ 0 in the free homotopy classes of a 1 , a 2 , c ′ ′ a_{1}, a_{2}, c'' a 1 , a 2 , c ′′ .
Step 3. Conclusion. By Lemma 5.5 , there is a diffeomorphism of M 2 M_{2} M 2 that maps a 1 , a 2 , c ′ ′ a_{1}, a_{2}, c'' a 1 , a 2 , c ′′ onto γ 1 , γ 2 , γ 0 \gamma_{1}, \gamma_{2}, \gamma_{0} γ 1 , γ 2 , γ 0 . Hence, by Step 1, M ′ M' M ′ is diffeomorphic to a manifold obtained from M 2 M_{2} M 2 by surgery along the geodesics γ 1 \gamma_1 γ 1 , γ 2 \gamma_2 γ 2 , γ 0 \gamma_0 γ 0 , for suitable framings, and this manifold admits a metric of positive biorthogonal curvature by Proposition 5.1. □
Remark 5.7. The stabilization in Step 1 is needed only if the class of c c c is a proper power. Suppose first that this class is nontrivial and not a proper power. Then Lemma 3.2 provides a metric of positive biorthogonal curvature on M M M and an embedded closed geodesic γ \gamma γ of this metric that is freely homotopic to c c c . By Lemma 5.5 there is a diffeomorphism φ \varphi φ of M M M with φ ( γ ) = c \varphi(\gamma)=c φ ( γ ) = c . Transporting the metric with φ \varphi φ , we obtain a metric of positive biorthogonal curvature for which c c c itself is a closed geodesic, and Proposition 5.1 applies directly. Suppose now that c c c is null-homotopic. Then c c c is isotopic to a circle in a coordinate ball, and the result of the surgery is the connected sum of M M M with S 2 × S 2 S^2 \times S^2 S 2 × S 2 or with the nontrivial S 2 S^2 S 2 -bundle over S 2 S^2 S 2 , depending on the framing. Both admit positive biorthogonal curvature by the results of Bettiol [3 , 4 ] .
Remark 5.8. Theorem 1.2 also follows from Theorem 1.4 , by performing the surgeries along the relators one after another; the passage from r r r to r y 1 y 2 r y_1 y_2 r y 1 y 2 , y 1 y_1 y 1 , y 2 y_2 y 2 in Step 1 is the geometric counterpart of Lemma 3.1 . We have given the direct proof because it treats all relators simultaneously and exhibits the manifold as an explicit boundary, as needed in Corollary 1.3 and Remark 5.2 .
Proof of Corollary 1.5
Proof of Corollary 1.5. By the standard rearrangement and cancellation of handles we may assume that the handles of W W W are attached in the order of their indices and, since W W W is connected, that there is exactly one zero-handle. Let W 1 ⊂ W W_1 \subset W W 1 ⊂ W be the union of the zero-handle and the one-handles. Since W W W is oriented, ∂ W 1 \partial W_1 ∂ W 1 is the connected sum of N N N copies of S 1 × S 3 S^1 \times S^3 S 1 × S 3 , where N N N is the number of one-handles, to be read as S 4 S^4 S 4 if N = 0 N=0 N = 0 . This manifold admits a metric of positive biorthogonal curvature, as we saw in the proof of Theorem 1.2 . The two-handles are attached to W 1 W_1 W 1 along pairwise disjoint embedded circles in ∂ W 1 \partial W_1 ∂ W 1 , and attaching a two-handle D 2 × D 3 D^2 \times D^3 D 2 × D 3 changes the boundary by surgery along its attaching circle. Hence ∂ W \partial W ∂ W is obtained from ∂ W 1 \partial W_1 ∂ W 1 by surgery along finitely many pairwise disjoint embedded circles, and the assertion follows from Theorem 1.4 . □
Proof of Corollary 1.6
Proof of Corollary 1.6. Part (a). Let Σ \Sigma Σ be a closed oriented surface. Smale [31 ] proved that every oriented S 2 S^2 S 2 -bundle over Σ \Sigma Σ is the sphere bundle of an oriented vector bundle of rank three. Hence it is the boundary of the associated disk bundle W → Σ W \to\Sigma W → Σ . Choose a handle decomposition of Σ \Sigma Σ . Since the bundle is trivial over each handle, the part of W W W over a handle of index j j j is a five-dimensional handle of index j j j . Thus W W W is a compact connected oriented five-manifold that has a handle decomposition with handles of index at most two, and ∂ W \partial W ∂ W admits a metric of positive biorthogonal curvature by Corollary 1.5. For the trivial bundle over the sphere this gives a new proof that S 2 × S 2 S^2 \times S^2 S 2 × S 2 admits positive biorthogonal curvature [3 ] .
Part (b). By Perelman’s work [26 ] , a closed orientable three-manifold of positive scalar curvature is a connected sum of spherical space forms S 3 / G S^3/G S 3 / G and of copies of S 2 × S 1 S^2 \times S^1 S 2 × S 1 . In other words, it is obtained from a disjoint union of spherical space forms by finitely many surgeries of index zero that preserve orientability. We argue by induction on the number of these surgeries. For a spherical space form, the product of the round metric on S 3 / G S^3/G S 3 / G with a circle is locally isometric to S 3 × R S^3 \times\mathbb{R} S 3 × R and has biorthogonal curvature 1 2 \frac{1}{2} 2 1 .
For the inductive step, let Y ′ Y' Y ′ be obtained from Y Y Y by a surgery of index zero, where Y Y Y is a possibly disconnected closed oriented three-manifold, and suppose that each component of Y × S 1 Y \times S^1 Y × S 1 admits positive biorthogonal curvature. The trace of the surgery is Y × [ 0 , 1 ] Y \times[0,1] Y × [ 0 , 1 ] with a one-handle D 1 × D 3 D^1 \times D^3 D 1 × D 3 attached, and its product with S 1 S^1 S 1 is a cobordism from Y × S 1 Y \times S^1 Y × S 1 to Y ′ × S 1 Y' \times S^1 Y ′ × S 1 that is obtained by attaching D 1 × D 3 × S 1 D^1 \times D^3 \times S^1 D 1 × D 3 × S 1 . Writing S 1 S^1 S 1 as the union of two arcs I 1 I_1 I 1 and I 2 I_2 I 2 , we see that D 1 × D 3 × I 1 D^1 \times D^3 \times I_1 D 1 × D 3 × I 1 is a five-dimensional one-handle and that D 1 × D 3 × I 2 = ( D 1 × I 2 ) × D 3 D^1 \times D^3 \times I_2 = (D^1 \times I_2) \times D^3 D 1 × D 3 × I 2 = ( D 1 × I 2 ) × D 3 is a two-handle. Hence Y ′ × S 1 Y' \times S^1 Y ′ × S 1 is obtained from Y × S 1 Y \times S^1 Y × S 1 by a surgery of index zero followed by surgery along an embedded circle. The first of these replaces Y × S 1 Y \times S^1 Y × S 1 by a connected sum of two of its components, or by the connected sum of one component with S 1 × S 3 S^1 \times S^3 S 1 × S 3 , and preserves positive biorthogonal curvature by Bettiol's theorem; the second preserves it by Theorem 1.4 .
Parts (c) and (d). Let X X X be a compact connected oriented four-manifold that has a handle decomposition with handles of index at most two. Then X × [ 0 , 1 ] X \times[0,1] X × [ 0 , 1 ] has a handle decomposition with handles of the same indices, and its boundary, after the corners have been rounded, is the double of X X X . Let Y Y Y be a closed connected oriented three-manifold, and let Y 0 Y_0 Y 0 be the complement of an open ball in Y Y Y . A handle decomposition of Y Y Y with exactly one three-handle restricts to a handle decomposition of Y 0 Y_0 Y 0 with handles of index at most two, and Y 0 × D 2 Y_0 \times D^2 Y 0 × D 2 has a handle decomposition with handles of the same indices. Its boundary is the spun manifold of Y Y Y . In both cases Corollary 1.5 applies. □
Euler characteristics, stabilization, and open questions
The minimal Euler characteristic
Let Γ \Gamma Γ be a finitely presented group. Following Hausmann and Weinberger [14 ] , we write q ( Γ ) q(\Gamma) q ( Γ ) for the minimum of the Euler characteristics of closed connected oriented smooth four-manifolds with fundamental group Γ \Gamma Γ . Following Mantione and Torres [21 ] , we write q > 0 ( Γ ) q_{>0}(\Gamma) q > 0 ( Γ ) for the minimum over those manifolds that admit a metric of positive scalar curvature, and we let q ⊥ ( Γ ) q_{\perp}(\Gamma) q ⊥ ( Γ ) be the minimum over those that admit a metric of positive biorthogonal curvature. By Theorem 1.2 such manifolds exist, and their Euler characteristics are bounded below because χ ( M ) = 2 − 2 b 1 ( Γ ) + b 2 ( M ) \chi(M) = 2 - 2b_1(\Gamma) + b_2(M) χ ( M ) = 2 − 2 b 1 ( Γ ) + b 2 ( M ) ; hence the three minima exist. Since positive biorthogonal curvature implies positive scalar curvature,
q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) . (59) q(\Gamma) \le q_{>0}(\Gamma) \le q_{\perp}(\Gamma). \tag*{(59)} q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) . (59)
For a field F \mathbb{F} F we put b j ( ⋅ ; F ) = dim F H j ( ⋅ ; F ) b_j(\,\cdot\,;\mathbb{F}) = \dim_{\mathbb{F}} H_j(\,\cdot\,;\mathbb{F}) b j ( ⋅ ; F ) = dim F H j ( ⋅ ; F ) , both for spaces and for groups, so that b j ( ⋅ ) = b j ( ⋅ ; Q ) b_j(\,\cdot\,) = b_j(\,\cdot\,;\mathbb{Q}) b j ( ⋅ ) = b j ( ⋅ ; Q ) . Recall that the deficiency def ( Γ ) \operatorname{def}(\Gamma) def ( Γ ) is the maximum of p − q p-q p − q over all finite presentations of Γ \Gamma Γ with p p p generators and q q q relators. It exists because
p − q ≤ b 1 ( Γ ; F ) − b 2 ( Γ ; F ) (60) p-q \le b_1(\Gamma;\mathbb{F}) - b_2(\Gamma;\mathbb{F}) \tag*{(60)} p − q ≤ b 1 ( Γ ; F ) − b 2 ( Γ ; F ) (60)
for every such presentation and every field F \mathbb{F} F . Indeed, the presentation complex C C C satisfies 1 − p + q = χ ( C ) = 1 − b 1 ( C ; F ) + b 2 ( C ; F ) 1-p+q = \chi(C) = 1-b_1(C;\mathbb{F}) + b_2(C;\mathbb{F}) 1 − p + q = χ ( C ) = 1 − b 1 ( C ; F ) + b 2 ( C ; F ) . Moreover, H 1 ( C ; F ) H_1(C;\mathbb{F}) H 1 ( C ; F ) is isomorphic to H 1 ( Γ ; F ) H_1(\Gamma;\mathbb{F}) H 1 ( Γ ; F ) , and H 2 ( C ; F ) H_2(C;\mathbb{F}) H 2 ( C ; F ) maps onto H 2 ( Γ ; F ) H_2(\Gamma;\mathbb{F}) H 2 ( Γ ; F ) , because a space of type K ( Γ , 1 ) K(\Gamma,1) K ( Γ , 1 ) can be obtained from C C C by attaching cells of dimension at least three.
Corollary 6.1. Every finitely presented group Γ \Gamma Γ satisfies
2 − 2 b 1 ( Γ ) + b 2 ( Γ ) ≤ q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) ≤ 2 − 2 def ( Γ ) . 2 - 2b_1(\Gamma) + b_2(\Gamma) \le q(\Gamma) \le q_{>0}(\Gamma) \le q_{\perp}(\Gamma) \le2 - 2\operatorname{def}(\Gamma). 2 − 2 b 1 ( Γ ) + b 2 ( Γ ) ≤ q ( Γ ) ≤ q > 0 ( Γ ) ≤ q ⊥ ( Γ ) ≤ 2 − 2 def ( Γ ) . All these inequalities are equalities for the free group F d F_d F d of rank d ≥ 0 d \ge0 d ≥ 0 , where the common value is 2 − 2 d 2 - 2d 2 − 2 d , and for nontrivial finite cyclic groups, where it is 2 2 2 .
Proof. The first inequality is due to Hausmann and Weinberger [14 ] . It holds because χ ( M ) = 2 − 2 b 1 ( Γ ) + b 2 ( M ) \chi(M) = 2 - 2b_1(\Gamma) + b_2(M) χ ( M ) = 2 − 2 b 1 ( Γ ) + b 2 ( M ) for every closed connected oriented four-manifold M M M with fundamental group Γ \Gamma Γ , and because H 2 ( M ; Q ) H_2(M;\mathbb{Q}) H 2 ( M ; Q ) maps onto H 2 ( Γ ; Q ) H_2(\Gamma;\mathbb{Q}) H 2 ( Γ ; Q ) , a space of type K ( Γ , 1 ) K(\Gamma,1) K ( Γ , 1 ) being obtained from M M M by attaching cells of dimension at least three. The next two inequalities are (59) . For the last one we choose a presentation that realizes the deficiency. None of its relators is trivial in the free group on the generators, since omitting such a relator would increase p − q p-q p − q . Corollary 1.3 therefore gives a manifold of positive biorthogonal curvature with χ = 2 − 2 def ( Γ ) \chi= 2 - 2\operatorname{def}(\Gamma) χ = 2 − 2 def ( Γ ) . For F d F_d F d , the presentation with d d d generators and no relators shows that def ( F d ) = d = b 1 ( F d ) \operatorname{def}(F_d) = d = b_1(F_d) def ( F d ) = d = b 1 ( F d ) , and b 2 ( F d ) = 0 b_2(F_d) = 0 b 2 ( F d ) = 0 . For a nontrivial finite cyclic group, the presentation with one generator and one relator shows that def ≥ 0 \operatorname{def} \ge0 def ≥ 0 , and b 1 = b 2 = 0 b_1 = b_2 = 0 b 1 = b 2 = 0 . In both cases the two outer bounds agree. □
The inequality q ( Γ ) ≤ 2 − 2 def ( Γ ) q(\Gamma) \le2 - 2\operatorname{def}(\Gamma) q ( Γ ) ≤ 2 − 2 def ( Γ ) is also due to Hausmann and Weinberger [14 ] , and the inequality q > 0 ( Γ ) ≤ 2 − 2 def ( Γ ) q_{>0}(\Gamma) \le2 - 2\operatorname{def}(\Gamma) q > 0 ( Γ ) ≤ 2 − 2 def ( Γ ) follows from the construction of Carr; see [21 Theorem 6] . What Corollary 6.1 adds is that the manifolds which realize this upper bound admit positive biorthogonal curvature.
The lower bound can be improved for manifolds whose fundamental class vanishes in the homology of the group. The following observation is known; see [16 Lemma 3.11] for groups with H 4 ( Γ ; F ) = 0 H_4(\Gamma;\mathbb{F}) = 0 H 4 ( Γ ; F ) = 0 , and [19 ] and [21 Proposition 3] for F = Q \mathbb{F} = \mathbb{Q} F = Q .
Lemma 6.2. Let M M M be a closed connected oriented four-manifold with fundamental group Γ \Gamma Γ , let c c c be a classifying map M → B Γ M \to B\Gamma M → B Γ of its universal cover, and let F \mathbb{F} F be a field. If c ∗ [ M ] = 0 c_*[M] = 0 c ∗ [ M ] = 0 in H 4 ( Γ ; F ) H_4(\Gamma;\mathbb{F}) H 4 ( Γ ; F ) , then b 2 ( M ; F ) ≥ 2 b 2 ( Γ ; F ) b_2(M;\mathbb{F}) \ge2b_2(\Gamma;\mathbb{F}) b 2 ( M ; F ) ≥ 2 b 2 ( Γ ; F ) , and hence
χ ( M ) ≥ 2 ( 1 − b 1 ( Γ ; F ) + b 2 ( Γ ; F ) ) . \chi(M) \ge2(1 - b_1(\Gamma;\mathbb{F}) + b_2(\Gamma;\mathbb{F})). χ ( M ) ≥ 2 ( 1 − b 1 ( Γ ; F ) + b 2 ( Γ ; F )) .
Proof. We may assume that B Γ B\Gamma B Γ is obtained from M M M by attaching cells of dimension at least three and that c c c is the inclusion. Then the homomorphism c ∗ c^* c ∗ from H 2 ( Γ ; F ) H^2(\Gamma;\mathbb{F}) H 2 ( Γ ; F ) to H 2 ( M ; F ) H^2(M;\mathbb{F}) H 2 ( M ; F ) is injective, and its image has dimension b 2 ( Γ ; F ) b_2(\Gamma;\mathbb{F}) b 2 ( Γ ; F ) . For a , b ∈ H 2 ( Γ ; F ) a,b \in H^2(\Gamma;\mathbb{F}) a , b ∈ H 2 ( Γ ; F ) we have
⟨ c ∗ a ∪ c ∗ b , [ M ] ⟩ = ⟨ a ∪ b , c ∗ [ M ] ⟩ = 0 , \langle c^*a \cup c^*b,[M]\rangle= \langle a \cup b,c_*[M]\rangle= 0, ⟨ c ∗ a ∪ c ∗ b , [ M ]⟩ = ⟨ a ∪ b , c ∗ [ M ]⟩ = 0 ,
so that the cup product pairing of H 2 ( M ; F ) H^2(M;\mathbb{F}) H 2 ( M ; F ) vanishes on the image of c ∗ c^* c ∗ . This pairing is nondegenerate by Poincaré duality, and therefore a subspace on which it vanishes has at most half the dimension of H 2 ( M ; F ) H^2(M;\mathbb{F}) H 2 ( M ; F ) . This proves the first assertion. The second one follows, because χ ( M ) = 2 − 2 b 1 ( M ; F ) + b 2 ( M ; F ) \chi(M) = 2 - 2b_1(M;\mathbb{F}) + b_2(M;\mathbb{F}) χ ( M ) = 2 − 2 b 1 ( M ; F ) + b 2 ( M ; F ) and b 1 ( M ; F ) = b 1 ( Γ ; F ) b_1(M;\mathbb{F}) = b_1(\Gamma;\mathbb{F}) b 1 ( M ; F ) = b 1 ( Γ ; F ) . □
Corollary 6.3. Let Γ \Gamma Γ be a finitely presented group, and suppose that there is a field F \mathbb{F} F such that H 4 ( Γ ; F ) = 0 H_4(\Gamma;\mathbb{F}) = 0 H 4 ( Γ ; F ) = 0 and def ( Γ ) = b 1 ( Γ ; F ) − b 2 ( Γ ; F ) \operatorname{def}(\Gamma) = b_1(\Gamma;\mathbb{F}) - b_2(\Gamma;\mathbb{F}) def ( Γ ) = b 1 ( Γ ; F ) − b 2 ( Γ ; F ) . Then
q ( Γ ) = q > 0 ( Γ ) = q ⊥ ( Γ ) = 2 − 2 def ( Γ ) . q(\Gamma) = q_{>0}(\Gamma) = q_{\perp}(\Gamma) = 2 - 2\operatorname{def}(\Gamma). q ( Γ ) = q > 0 ( Γ ) = q ⊥ ( Γ ) = 2 − 2 def ( Γ ) .
Proof. By Lemma 6.2 , every closed connected oriented four-manifold with fundamental group Γ \Gamma Γ has Euler characteristic at least 2 ( 1 − b 1 ( Γ ; F ) + b 2 ( Γ ; F ) ) = 2 − 2 def ( Γ ) 2(1 - b_1(\Gamma;\mathbb{F}) + b_2(\Gamma;\mathbb{F})) = 2 - 2\operatorname{def}(\Gamma) 2 ( 1 − b 1 ( Γ ; F ) + b 2 ( Γ ; F )) = 2 − 2 def ( Γ ) . The assertion now follows from Corollary 6.1 . □
For q > 0 ( Γ ) q_{>0}(\Gamma) q > 0 ( Γ ) and F = Q \mathbb{F} = \mathbb{Q} F = Q , an equality of this kind was proved by Mantione and Torres [21 Theorem B] . For groups that admit a finite aspherical two-dimensional complex we recover a theorem of Hillman [15 ] , who determined q ( Γ ) q(\Gamma) q ( Γ ) for these groups. In both cases we see that the minimum is attained by manifolds of positive biorthogonal curvature. Recall that a connected CW complex is aspherical if its universal cover is contractible.
Corollary 6.4. Let X X X be a finite connected aspherical two-dimensional CW complex with fundamental group Γ \Gamma Γ . Then
q ( Γ ) = q > 0 ( Γ ) = q ⊥ ( Γ ) = 2 χ ( X ) . q(\Gamma) = q_{>0}(\Gamma) = q_{\perp}(\Gamma) = 2\chi(X). q ( Γ ) = q > 0 ( Γ ) = q ⊥ ( Γ ) = 2 χ ( X ) . In particular, for a closed oriented surface Σ \Sigma Σ of positive genus,
q ⊥ ( π 1 ( Σ ) ) = 2 χ ( Σ ) , q_{\perp}(\pi_1(\Sigma)) = 2\chi(\Sigma), q ⊥ ( π 1 ( Σ )) = 2 χ ( Σ ) , and the minimum is attained by both oriented S 2 S^2 S 2 -bundles over $\Sigma.
Proof. Since X X X is a space of type K ( Γ , 1 ) K(\Gamma,1) K ( Γ , 1 ) of dimension two, we have H 4 ( Γ ; Q ) = 0 H_4(\Gamma;\mathbb{Q}) = 0 H 4 ( Γ ; Q ) = 0 and χ ( X ) = 1 − b 1 ( Γ ) + b 2 ( Γ ) \chi(X) = 1 - b_1(\Gamma) + b_2(\Gamma) χ ( X ) = 1 − b 1 ( Γ ) + b 2 ( Γ ) . Collapsing a maximal tree in the one-skeleton of X X X , we obtain a presentation of Γ \Gamma Γ with p p p generators and q q q relators such that χ ( X ) = 1 − p + q \chi(X) = 1 - p + q χ ( X ) = 1 − p + q . Hence def ( Γ ) ≥ p − q = b 1 ( Γ ) − b 2 ( Γ ) \operatorname{def}(\Gamma) \ge p - q = b_1(\Gamma) - b_2(\Gamma) def ( Γ ) ≥ p − q = b 1 ( Γ ) − b 2 ( Γ ) , and equality holds by (60) . Corollary 6.3 now gives the first assertion, since 2 − 2 def ( Γ ) = 2 χ ( X ) 2 - 2\operatorname{def}(\Gamma) = 2\chi(X) 2 − 2 def ( Γ ) = 2 χ ( X ) . A closed oriented surface Σ \Sigma Σ of positive genus is aspherical. An S 2 S^2 S 2 -bundle over Σ \Sigma Σ has Euler characteristic χ ( S 2 ) χ ( Σ ) = 2 χ ( Σ ) \chi(S^2)\chi(\Sigma) = 2\chi(\Sigma) χ ( S 2 ) χ ( Σ ) = 2 χ ( Σ ) and admits a metric of positive biorthogonal curvature by Corollary 1.6 (a). □
Remark 6.5. Corollary 6.3 also applies to groups of cohomological dimension three. Let Y Y Y be a closed connected orientable aspherical three-manifold with fundamental group Γ \Gamma Γ . Then H 4 ( Γ ; Q ) = 0 H_4(\Gamma;\mathbb{Q}) = 0 H 4 ( Γ ; Q ) = 0 , and b 1 ( Γ ) = b 2 ( Γ ) b_1(\Gamma) = b_2(\Gamma) b 1 ( Γ ) = b 2 ( Γ ) by Poincaré duality. A Heegaard splitting of Y Y Y gives a presentation of Γ \Gamma Γ with as many relators as generators, so that def ( Γ ) ≥ 0 \operatorname{def}(\Gamma) \ge0 def ( Γ ) ≥ 0 , and (60) shows that def ( Γ ) = 0 \operatorname{def}(\Gamma) = 0 def ( Γ ) = 0 . Hence q ( Γ ) = q ⊥ ( Γ ) = 2 q(\Gamma) = q_{\perp}(\Gamma) = 2 q ( Γ ) = q ⊥ ( Γ ) = 2 . The minimum is attained by the spun manifold of Y Y Y , which has fundamental group Γ \Gamma Γ and Euler characteristic 2 2 2 and admits positive biorthogonal curvature by Corollary 1.6 (d). For Γ = Z 3 \Gamma= \mathbb{Z}^3 Γ = Z 3 it is also attained by the manifold of Remark 5.4.
Remark 6.6. The invariants q q q and q > 0 q_{>0} q > 0 do not agree in general. The four-torus and Corollary 6.1 show that q ( Z 4 ) = 0 q(\mathbb{Z}^4) = 0 q ( Z 4 ) = 0 , whereas
q > 0 ( Z d ) = q ⊥ ( Z d ) = ( d − 1 ) ( d − 2 ) for every d ≥ 1 , q_{>0}(\mathbb{Z}^d) = q_{\perp}(\mathbb{Z}^d) = (d-1)(d-2) \qquad\text{for every } d \ge1, q > 0 ( Z d ) = q ⊥ ( Z d ) = ( d − 1 ) ( d − 2 ) for every d ≥ 1 ,
so that q > 0 ( Z 4 ) = q ⊥ ( Z 4 ) = 6 q_{>0}(\mathbb{Z}^4) = q_{\perp}(\mathbb{Z}^4) = 6 q > 0 ( Z 4 ) = q ⊥ ( Z 4 ) = 6 . Indeed, let M M M be a closed connected oriented four-manifold with fundamental group Z d \mathbb{Z}^d Z d that admits a metric of positive scalar curvature, and let c c c be a classifying map M → T d M \to T^d M → T d . The classes of the four-dimensional coordinate tori form a basis of H 4 ( T d ; Q ) H_4(T^d;\mathbb{Q}) H 4 ( T d ; Q ) . If c ∗ [ M ] c_*[M] c ∗ [ M ] were nonzero in this group, then the composition of c c c with the projection of T d T^d T d onto one of these coordinate tori would therefore have nonzero degree. This is impossible, because a closed oriented four-manifold that admits a map of nonzero degree to the four-torus carries no metric of positive scalar curvature [28 ] . Hence Lemma 6.2 applies with F = Q \mathbb{F} = \mathbb{Q} F = Q and gives χ ( M ) ≥ 2 ( 1 − d + ( d 2 ) ) = ( d − 1 ) ( d − 2 ) \chi(M) \ge2(1-d+\binom{d}{2}) = (d-1)(d-2) χ ( M ) ≥ 2 ( 1 − d + ( 2 d ) ) = ( d − 1 ) ( d − 2 ) . On the other hand, the manifold of Remark 5.4 that realizes Z d \mathbb{Z}^d Z d has this Euler characteristic and admits positive biorthogonal curvature. For q > 0 q_{>0} q > 0 , results of this kind are due to Mantione and Torres [21 Theorem B] , and the numbers q ( Z d ) q(\mathbb{Z}^d) q ( Z d ) were determined by Kirk and Livingston [19 ] .
We do not know whether q > 0 q_{>0} q > 0 and q ⊥ q_{\perp} q ⊥ can differ.
Question 6.7. Is q ⊥ ( Γ ) = q > 0 ( Γ ) q_{\perp}(\Gamma)=q_{>0}(\Gamma) q ⊥ ( Γ ) = q > 0 ( Γ ) for every finitely presented group Γ \Gamma Γ ?
A negative answer would also answer Question 1.8 , since a manifold of positive scalar curvature that realizes q > 0 ( Γ ) q_{>0}(\Gamma) q > 0 ( Γ ) admits no metric of positive biorthogonal curvature if q > 0 ( Γ ) < q ⊥ ( Γ ) q_{>0}(\Gamma)<q_{\perp}(\Gamma) q > 0 ( Γ ) < q ⊥ ( Γ ) .
Positive biorthogonal curvature after stabilization
We now prove Corollary 1.7 . The proof combines Theorem 1.2 with Kreck’s stable classification of four-manifolds [20 Theorem C] . It states that two closed four-manifolds with the same Euler characteristic and the same normal 1-type that admit bordant normal 1-smoothings become diffeomorphic after connected sum with r r r copies of S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 , for some r r r . We recall from [20 Proposition 2] what this means for a closed connected oriented four-manifold N N N with fundamental group Γ \Gamma Γ . If the universal cover of N N N is not spin, the normal 1-type of N N N is B Γ × B S O B\Gamma\times BSO B Γ × B S O ; a normal 1-smoothing is an orientation together with a map N → B Γ N\to B\Gamma N → B Γ that induces an isomorphism of fundamental groups, and two such smoothings are bordant if they represent the same element of the oriented bordism group Ω 4 S O ( B Γ ) \Omega^{SO}_{4}(B\Gamma) Ω 4 S O ( B Γ ) . If N N N is spin, the same holds with B S p i n BSpin B S p in , spin structures, and Ω 4 S p i n ( B Γ ) \Omega^{Spin}_{4}(B\Gamma) Ω 4 S p in ( B Γ ) .
Proof of Corollary 1.7. Let M Γ M_{\Gamma} M Γ be a manifold as in Corollary 1.3, for some finite presentation of Γ \Gamma Γ , and let W W W be the parallelizable five-manifold of Remark 5.2. Since the inclusion induces an isomorphism π 1 ( M Γ ) ≅ π 1 ( W ) \pi_{1}(M_{\Gamma})\cong\pi_{1}(W) π 1 ( M Γ ) ≅ π 1 ( W ) , a classifying map c Γ c_{\Gamma} c Γ from M Γ M_{\Gamma} M Γ to B Γ B\Gamma B Γ extends to W W W . Hence ( M Γ , c Γ ) (M_{\Gamma},c_{\Gamma}) ( M Γ , c Γ ) is null-bordant in Ω 4 S O ( B Γ ) \Omega^{SO}_{4}(B\Gamma) Ω 4 S O ( B Γ ) , and also in Ω 4 S p i n ( B Γ ) \Omega^{Spin}_{4}(B\Gamma) Ω 4 S p in ( B Γ ) for the spin structure induced from W W W .
Case (b). By assumption, ( N , c ) (N,c) ( N , c ) and ( M Γ , c Γ ) (M_{\Gamma},c_{\Gamma}) ( M Γ , c Γ ) are bordant normal 1-smoothings in B Γ × B S p i n B\Gamma\times BSpin B Γ × B S p in . Both manifolds have signature zero, so that their Euler characteristics have the same parity. Connected sum with S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 raises the Euler characteristic by two and changes neither the normal 1-type nor the bordism class, because S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 bounds the spin manifold S 2 × D 3 S^{2}\times D^{3} S 2 × D 3 . After adding copies of S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 to one of the two manifolds we may therefore assume that their Euler characteristics agree, and Kreck’s theorem shows that
N # r ( S 2 × S 2 ) ≅ M Γ # r ′ ( S 2 × S 2 ) N\mathbin{\#}r\left(S^{2}\times S^{2}\right)\cong M_{\Gamma}\mathbin{\#}r'\left(S^{2}\times S^{2}\right) N # r ( S 2 × S 2 ) ≅ M Γ # r ′ ( S 2 × S 2 )
for some r , r ′ ≥ 0 r,r'\geq0 r , r ′ ≥ 0 . The manifold on the right admits a metric of positive biorthogonal curvature, because M Γ M_{\Gamma} M Γ and S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 do and the condition is preserved under connected sums [4 Proposition 3.1] . Adding further copies of S 2 × S 2 S^{2}\times S^{2} S 2 × S 2 on both sides, we obtain the assertion for all larger values of r r r .
Case (a). Let P P P be a connected sum of a a a copies of the complex projective plane and b b b copies of the complex projective plane with the opposite orientation, where a , b ≥ 1 a,b\geq1 a , b ≥ 1 and a − b = sign ( N ) a-b=\operatorname{sign}(N) a − b = sign ( N ) , and put M 1 = M Γ # P M_{1}=M_{\Gamma}\mathbin{\#}P M 1 = M Γ # P . Since the Fubini–Study metric has positive sectional curvature, M 1 M_{1} M 1 admits a metric of positive biorthogonal curvature. Its fundamental group is Γ \Gamma Γ , and its universal cover is not spin, because it contains a punctured copy of P P P and hence an embedded sphere with odd self-intersection number. Let c 1 c_{1} c 1 be the map from M 1 M_{1} M 1 to B Γ B\Gamma B Γ that is c Γ c_{\Gamma} c Γ on the first summand and constant on P P P ; it induces an isomorphism of fundamental groups, and ( c 1 ) ∗ [ M 1 ] = 0 (c_{1})_{*}[M_{1}]=0 ( c 1 ) ∗ [ M 1 ] = 0 .
Since Ω j S O = 0 \Omega^{SO}_{j}=0 Ω j S O = 0 for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 and Ω 4 S O ≅ Z \Omega^{SO}_{4}\cong\mathbb{Z} Ω 4 S O ≅ Z via the signature, the Atiyah–Hirzebruch spectral sequence gives an isomorphism
Ω 4 S O ( B Γ ) ≅ Z ⊕ H 4 ( Γ ; Z ) , [ M , f ] ⟼ ( sign ( M ) , f ∗ [ M ] ) . \Omega^{SO}_{4}(B\Gamma)\cong\mathbb{Z}\oplus H_{4}(\Gamma;\mathbb{Z}),\qquad[M,f]\longmapsto\left(\operatorname{sign}(M),f_{*}[M]\right). Ω 4 S O ( B Γ ) ≅ Z ⊕ H 4 ( Γ ; Z ) , [ M , f ] ⟼ ( sign ( M ) , f ∗ [ M ] ) .
By assumption, ( N , c ) (N,c) ( N , c ) and ( M 1 , c 1 ) (M_1,c_1) ( M 1 , c 1 ) have the same image ( sign ( N ) , 0 ) (\operatorname{sign}(N),0) ( sign ( N ) , 0 ) . Hence they are bordant normal 1-smoothings in B Γ × B S O B\Gamma\times BSO B Γ × B S O . Since N N N and M 1 M_1 M 1 have the same signature, their Euler characteristics have the same parity, and we conclude as in case (b). □
Remark 6.8. By the same argument, the conclusion of Corollary 1.7 holds whenever a normal 1-smoothing of N N N is bordant to a normal 1-smoothing of a manifold of positive biorthogonal curvature. This includes the third case of Kreck’s classification, in which N N N is not spin but its universal cover is spin. In this case there is a unique class w ∈ H 2 ( Γ ; Z / 2 ) w \in H^2(\Gamma;\mathbb{Z}/2) w ∈ H 2 ( Γ ; Z /2 ) with c ∗ w = w 2 ( N ) c^*w = w_2(N) c ∗ w = w 2 ( N ) , and the normal 1-type of N N N is determined by the pair ( Γ , w ) (\Gamma,w) ( Γ , w ) . In the proof of Corollary 1.3 we can choose the framings of the two-handles in such a way that the second Stiefel–Whitney class of W W W is the image of w w w under the homomorphism induced by a classifying map W → B Γ W \to B\Gamma W → B Γ , because changing the framing of a two-handle changes w 2 ( W ) w_2(W) w 2 ( W ) by the class of the corresponding two-cell, and these classes generate H 2 ( W ; Z / 2 ) H^2(W;\mathbb{Z}/2) H 2 ( W ; Z /2 ) . The boundary of the resulting manifold W W W admits positive biorthogonal curvature by Corollary 1.5, it has the same normal 1-type as N N N , and it bounds over this normal 1-type. Hence the connected sum of N N N with sufficiently many copies of S 2 × S 2 S^2 \times S^2 S 2 × S 2 admits positive biorthogonal curvature if some normal 1-smoothing of N N N is null-bordant. Since positive biorthogonal curvature implies positive scalar curvature, the manifolds in Corollary 1.7 also admit positive scalar curvature after stabilization.
Open questions
Positive scalar versus positive biorthogonal curvature. Question 1.8 asks for an obstruction to positive biorthogonal curvature that is not an obstruction to positive scalar curvature. The Bochner technique does not provide one. By Lemma 2.1, the condition k g > 0 k_g > 0 k g > 0 means that λ min ( A + ) + λ min ( A − ) > 0 \lambda_{\min}(A_+) + \lambda_{\min}(A_-) > 0 λ m i n ( A + ) + λ m i n ( A − ) > 0 , so that at each point at least one of the operators A + A_+ A + and A − A_- A − is positive definite. Where A + A_+ A + is positive definite, the curvature term 1 3 s g − 2 W + \frac{1}{3}s_g - 2W_+ 3 1 s g − 2 W + in the Weitzenböck formula for self-dual harmonic two-forms is positive definite as well. However, which of the two operators is positive may depend on the point, and no vanishing theorem results. Indeed, S 2 × S 2 S^2 \times S^2 S 2 × S 2 admits positive biorthogonal curvature, and both its self-dual and its anti-self-dual harmonic forms are nontrivial. Likewise, by Theorem 1.2, the minimal surface arguments that restrict the fundamental groups of manifolds of positive isotropic curvature have no analogue for biorthogonal curvature. By Theorem 1.4 and Corollary 1.7, whatever distinguishes the two conditions is invisible to surgery of codimension at least three and, in the range of Corollary 1.7, to connected sums with S 2 × S 2 S^2 \times S^2 S 2 × S 2 .
Relative surgery. Proposition 5.1 is a relative statement, in the sense that the metric is changed only near the geodesics. Theorem 1.4 holds for arbitrary embedded circles, but only up to diffeomorphism, because its proof moves the circle to a geodesic of a different metric on a stabilization of the manifold. We do not know whether, for a given metric of positive biorthogonal curvature and an arbitrary embedded circle, the surgery can be performed by a modification of the metric inside a prescribed neighborhood of the circle. The handle of Theorem 4.1 allows for both framings and for arbitrary normal holonomy; what is missing is a deformation of the given metric to a round tube near a circle that is not a geodesic.
Nonorientable manifolds. Orientability of the manifold enters our arguments only through the triviality of the normal bundle of the surgery circle. The constructions of Sections 3 and 4 take place in a tubular neighborhood of the circle, and Proposition 5.1 remains valid, with the same proof, for closed geodesics with trivial normal bundle in a closed nonorientable four-manifold. of positive biorthogonal curvature. We expect that Theorems 1.2 and 1.4 have nonorientable versions as well, in which the orientation character of the manifold is prescribed, as in the result of Mantione and Torres for positive scalar curvature [21 Theorem 7] . We do not pursue this here.
Acknowledgements
This work was partially supported by NSFC (Grant Nos. 12571025, 12131012 and 12301032), NSF of Jiangsu (Grant No. BK20230803), the Fundamental Research Funds for the Central Universities (Grant No. 4007012402), and the Zhishan Scholars Program of Southeast University (Grant No. 2242024RCB0039).
Use of generative AI. ChatGPT was used for exploratory computations, proof development, and editorial assistance.
Zhiqi Chen
School of Mathematics and Statistics, Guangdong University of Technology, Guangzhou 510520, P.R. China
E-mail: chenzhiqi@nankai.edu.cn
Hui Zhang
School of Mathematics, Southeast University, Nanjing 211189, P.R. China
E-mail: huizhang@mail.nankai.edu.cn