Arxiv:2008.11888V3 [Math.DG] 11 Sep 2020 N .T a N Ym Rmvadb Lawson

Home , Richard Schoen, Scalar curvature

arXiv:2008.11888v3 [math.DG] 11 Sep 2020 n .T a n yM rmvadB Lawson. B. 1 and Theorem Gromov M. by and Yau S.-T. and imninmti fpstv clrcurvature. scalar positive of metric Riemannian a onc u ihatorus. a a an with on curvature sum scalar connect positive dimens of of metrics complete manifolds as aspherical well compact on curvature scalar cu t scalar of constant to resolution deformation Schoen’s ma conformal and concerning positive 31] lem 32, the [27, of relativity proof general Schoen–Yau’s including sequences, clrcraue nScin13blw edsusprevious that discuss expect we We below, restr 1.3 topological µ Section global yield in to curvature; surfaces scalar these here of use (called first problem curvature mean FMNFLSWT OIIESAA CURVATURE SCALAR POSITIVE WITH MANIFOLDS OF bblsfrlclgoer fsaa uvtr eg compa (e.g. curvature scalar of geometry local for -bubbles EEAIE OPBBLSADTETOPOLOGY THE AND BUBBLES SOAP GENERALIZED ebgnb ealn h olwn elkonrsl prove result well-known following the recalling by begin We nti ae,w rvd w xesoso hoe ,ruling 1, Theorem of extensions two provide we paper, this In i several had has thereof) generalizations (and result This 1 epoeteerslsb nlzn tbesltost h p the to solutions stable analyzing by results these prove We .Senhsrcnl icvrda neetn e ro of proof new interesting an discovered recently has Stern D. osntamtaReana ercwt oiiesaa curv scalar positive with metric Riemannian a admit not does ufcsta r ttoayfrprescribed-mean-curvatu for called stationary (also are that surfaces non-ne with manifolds flat conformally curvature. locally all contribu for Liouvi forthcoming Schoen–Yau holds the with that o proves combined metric this Lesourd–Unger–Yau, When complete a admit curvature. not scalar does manifold arbitrary an with Abstract. diinly eso htfor that show we Additionally, e emti oli hs eut r eeaie opbu soap generalized are results these in tool geometric key A Grc ojcue[6 8 1 7]) 31, 28, [26, Conjecture (Geroch µ bblswl n te oooia applications. topological other find will -bubbles µ epoeta for that prove We -bubbles). TSCOOHADCA LI CHAO AND CHODOSH OTIS 1. Introduction n { ∈ n 1 ≤ 4 , µ ,tecnetdsmo a of sum connected the 7, 5 bbls.Ti em ob the be to seems This -bubbles). } lsdaspherical closed a . The 1 n hswhen this trsde o admit not does -torus efunctionals re cin npositive on ictions aiescalar gative eYmb prob- Yamabe he btaymanifold rbitrary l theorem lle n io theorems). rison os4ad5 as 5, and 4 ions -manifold ature. vtr [24]. rvature positive f plctosof applications yR Schoen R. by n in by tions pratcon- mportant stermin theorem ss bbles— n n -torus u positive out se[33]). (see 3 = rescribed 2 OTISCHODOSHANDCHAOLI

1.1. Aspherical manifolds. Recall that a manifold is aspherical if it has contractible universal cover. For example, any closed manifold covered by a Cartan–Hadamard manifold (such as the torus) is aspherical. Our first main result is as follows. Theorem 2. For n ∈ {4, 5}, a smooth closed aspherical n-manifold N does not admit a smooth Riemannian metric with positive scalar curvature. Any metric of non-negative scalar curvature on N is flat. In a 1987 survey article [30], Schoen–Yau first proposed that all closed aspherical manifolds do not admit metrics of positive scalar curvature.2 They outlined a proof of this in 4-dimensions [30, Theorem 6], but many parts of the proof have not been given. Resolving these missing parts is essential to carry out their outline, as we do in this paper. Furthermore, the generalization to n = 5 is considerably more involved. See Section 1.3 for a more detailed discussion. Theorem 2 (in all dimensions n ≥ 4) was also conjectured in a slightly different form by M. Gromov (see [9, p. 113]). It has some link with the Novikov conjecture on topological invariance of certain polynomials of Pon- tryagin classes, as explained in [13, p. 25]. Furthermore, as discussed in [11, Section 16], Theorem 2 (and its conjectural higher dimensional analogue) is one of the central questions in the study of geometric and topological properties of manifolds with positive scalar curvature. Recently, J. Wang proved Theorem 2 in the special case that N is of dimension 4 with nonzero first Betti number [35, Chapter 7 (Theorem F)].3 Moreover, we note that in the first version of this article, we only considered Theorem 2 in the case of n = 4. As we were finishing writing down the generalization to n = 5, we received a preprint from Gromov containing a proof of the same generalization and in fact proving the more general fact that closed 5-manifolds that admit maps of non-zero degree to an aspherical manifold do not admit positive scalar curvature. Both approaches are relatively similar (but were obtained independently). 1.2. The Geroch conjecture with arbitrary ends. Our second main result resolves a question arising in the work of Schoen–Yau on locally conformally flat manifolds [22] (cf. [23, §6]). Theorem 3. Let n ≤ 7. For any n-manifold X, the connect sum T n#X does not admit a complete metric of positive scalar curvature. The only complete metric of non-negative scalar curvature on T n#X is flat. For example, this implies that a punctured torus does not admit a complete metric of positive scalar curvature (note however, that this particular statement follows from work of Gromov-Lawson in [8, Example 6.9] based

2S.T. Yau has informed us that they in fact conjectured this in the early 1980s. 3As pointed out in [35], there exist infinitely many aspherical 4-manifolds with zero first Betti number [20]. SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 3 on relative index theorems). When X is compact, Theorem 3 is well-known: it has recently been proven in all dimensions by Schoen–Yau [31] via an analysis of singular minimal surfaces (cf. [28, 7]). We emphasize that M. Lesourd, R. Unger, and S.-T. Yau have recently announced [18] a proof of Theorem 3 for n = 3. Their proof extends to 3 < n ≤ 7 under certain technical assumptions on the geometry and topology of the manifold at infinity. Our proof is different at a technical level (even when n = 3) as compared to their indicated strategy (which involves analyzing non-compact stable minimal hypersurfaces). One main reason to consider Theorem 3 comes from the study of the topology of locally conformally flat manifolds with nonnegative scalar curvature. In their foundational work on these manifolds [22] (cf. [23, Theorem 3.5]), Schoen–Yau have proven that the geometry of the developing map of a locally conformally flat manifold with non-negative scalar curvature is strongly influenced by the mass of the manifold obtained via a Green’s function conformal blown-up (motivated by Schoen’s solution to the Yamabe problem [24]). Such a blown-up manifold will have a distinguished asymptotically flat end, but will also have other ends with metrics that are complete but do not have any other constraints on their geometry or topology. If there were no such uncontrolled ends, a well-known argument due to Lohkamp allows one to reduce the positive mass theorem to the Geroch conjecture for T n#M (where M is compact). Such a reduction is delicate for asymptotically flat manifolds with other complete ends, since the geometry along the other ends could affect certain global arguments. However, Lesourd–Unger–Yau have recently announced [18] that, through a careful analysis of the Green’s function and modification of Lohkamp’s argument, one can reduce the study of the manifolds arising in the work of Schoen–Yau [22] to the situation in Theorem 3. As such, by combining Theorem 3 with results in [22] and [18], we have the following definitive result.

Corollary 4. Suppose that (M n, g) is a complete Riemannian manifold with n Rg ≥ 0. If Φ : M → S is a conformal map, then Φ is injective and ∂Φ(M) has zero Newtonian capacity.

As shown in [22, Theorems 4.6 and 4.7], the Liouville result in Corol- lary 4 has strong consequences for the higher homotopy groups of locally conformally ﬂat manifolds with nonnegative scalar curvature. Under additional assumptions on the geometry or dimension of M (including, e.g., that n ≥ 7), this was proven by Schoen–Yau [22, Theorem 3.1]. In fact, Corollary 4 was announced by Schoen–Yau in full generality [22, Proposition 4.4’] (cf. [23, p. 262]), but the essential ingredient along the lines of Theorem 3 has never appeared. Finally, we reiterate that Lesourd– Unger–Yau [18] have announced a proof of Corollary 4 when n = 3 (as well as for 4 ≤ n ≤ 7 with certain additional assumptions on the geometry and topology of (M, g) at inﬁnity). 4 OTISCHODOSHANDCHAOLI

We remark that our proof of Theorem 3 can be generalized to allow certain other manifolds in place of the torus, see Section 8. We expect that one may remove the dimensional restriction (following [31]) but we do not pursue this here.

1.3. Idea of the proofs of Theorem 2 and Theorem 3. We explain the idea of the proofs of the results described above. As will be seen later, both Theorem 2 and Theorem 3 have the same central difficulty: it is hard to find a (compact) stable minimal surface due to non-compactness of the ambient manifold. Thus, we will mostly focus on Theorem 2 here, and briefly indicate the strategy of Theorem 3 at the end. Our strategy of proof of Theorem 2 extends the outline of Schoen–Yau in [30] when n = 4. We now describe their outline (of Theorem 2 when n = 4) and subsequently detail our contributions as well as how to generalize the strategy to n = 5.

Suppose to the contrary, that (N 4, g) is a closed aspherical manifold with R(g) ≥ 1. Pass to a non-compact cover (N, g) which is homotopy equivalent 1 to S . Fix [σ] a generator of π1(N). Minimize area among hypersurfaces dual to [σ] to find a complete4 stable minimal hypersurface M. By Schoen– Yau’s inductive descent technique, the stable minimal hypersurface (M, g|M ) is Yamabe positive. Schoen–Yau suggest that one can derive a contradiction from this as follows. Suppose that one could find a large region Ω ⊂ M so that each component of ∂Ω (there might be many) has controlled area. One can hope that each component of ∂Ω can be filled in by some 3-manifold in N of bounded diameter. The existence of such a fill-in is expected here, as N is a covering space of a closed manifold (cf. the Lemma below Theorem 6 in [30]). Given this, it would be possible to cap off each component of ∂Ω without affecting the intersection with σ. This would yield a 3-cycle in N with non-trivial algebraic intersection with σ. This would contradict H3(N) = 0, finishing the proof.

In this paper, we obtain a complete proof of Theorem 2 (when n = 4) along the lines of the outline given above. Our main contributions are the resolutions of the following two fundamental difficulties arising in the outline above. The first, and most essential difficulty, is to actually find Ω as stated. One intuitive way to approach this would be to exploit the fact that stable minimal surfaces in a three manifold with scalar curvature ≥ 2 are spherical and have area ≤ 4π (see [2] for a related rigidity result). However, there is no reason that M will contain any minimal surfaces, much less admit an exhaustion by regions bounded by stable minimal surfaces.

4In our paper, we ﬁnd it convenient to instead work with a stable minimal hypersurface with ﬁxed boundary that is very far from σ, but this is not an important point. SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 5

Instead, we rely on generalized soap bubbles—surfaces that are stationary for the prescribed-mean-curvature functional (called here the µ-bubble after [10, 13]). The use of this functional in scalar curvature problems was first 5 considered by Gromov in [10, Section 5 6 ]. Recently, this approach was used by Gromov to obtain a list of comparison theorems in scalar curvature [12, 13], and later expanded in work of J. Zhu [39, 38].5 By applying this technique, we can localize the minimizer while still retaining the geometric properties used in the inductive descent argument. We also note that an additional issue with the outline as explained above is that the geometry of (M, g|M ) after conformally deforming to positive scalar curvature might be dramatically different near infinity from the geometry of (N, g). To handle this, we combine the µ-bubble method with the warped product descent technique from [29, 31].

Next, we briefly discuss the generalization of this argument to the case of n = 5. We argue similarly to the above sketch and produce a curve σ and 4-dimensional minimal surface M4 (we find it convenient to work in the universal cover and with a minimizing line σ, but this is not an important issue) with boundary far from σ, so that M4 has non-trivial algebraic intersection with σ. We can then use the µ-bubble argument as in n =3 to find a region in M4 whose boundary is Yamabe positive and lies far away from σ. At this point, we encounter a serious difficulty not present in the previous situation: 3-manifolds with positive scalar curvature can have arbitrarily large diameter. As such, we develop a slice-and-dice method that we use to decompose the 3-dimensional µ-bubble M3 into pieces of controlled diameter and overlap. To do so, we first slice the µ-bubble by 2-dimensional spheres chosen to simplify H2(M3) appropriately. Then, we dice the resulting manifold with boundary, using free boundary µ-bubbles (if we did not slice first, the µ-bubble argument would only produce a decomposition into regions with bounded distance to the boundary, but if there are many boundary components, such a region could have arbitrarily large diameter). Because we arrange that the free boundary µ-bubbles are of disk type, this dices M3 into connected components of controlled diameter, and so that the overlap of the sets (along their boundaries) has controlled diameter. This is sufficient to complete the fill-in argument explained for n = 4 above. We note that the slice-and-dice method can be used to show that a compact 3-manifold with positive scalar curvature can be mapped to a (1- dimensional) graph by a map with pre-images of uniformly bounded diameter, or equivalently that such a manifold has bounded first Uryson width. See also [8, 16, 14] for related results. We note that Y. Liokumovich and

5The prescribed-mean-curvature functional has recently been considered in other geometric problems as well, cf. [37, 1, 36]. 6 OTISCHODOSHANDCHAOLI

D. Maximo have recently obtained a related (but strictly more general) result along these lines; it would be interesting to compare the two approaches.

Finally we briefly motivate the proof of Theorem 3. We again seek a µ-bubble representing a suitable homology class and apply the Schoen-Yau descent argument (in fact, we argue slightly differently to this, by passing to a certain covering space so that we can consider boundaries rather than arbitrary homological relations). The key point in this argument is then to find the correct prescribing mean curvature function. A basic difficulty present here (as compared to the previous discussion) is that we are only assuming positivity of the scalar curvature rather than uniformly positive scalar curvature. The standard µ-bubble technique (like the one we use in Theorem 2) could potentially fail if—for example—the scalar curvature decayed faster than quadratically. However, because there is a sufficiently nice covering space, we can make the argument work (morally, this has to do with the fact that the torus has sufficiently complicated topology). We note that a similar argument extends Theorem 3 to manifolds in the form (M × S1)#X, where M is a Schoen–Yau–Schick manifold (see [12, Section 5]) and X is arbitrary.

The paper is organized as follows. We prove Theorem 2 in Section 2-6. In Section 2, we provide some topological preliminaries concerning aspherical n-manifolds. In Section 3 we discuss µ-bubbles. and in Section 4 we discuss the free boundary version of µ-bubbles. In Section 5 we prove diameter bounds that will apply to stable 2-dimensional µ-bubbles (with free boundary or not). Then in Section 6 we prove Theorem 2 by the slice-and-dice method mentioned above. We prove Theorem 3 in Section 7. In Section 8, we discuss an extension of Theorem 3 to Schoen–Yau–Schick manifolds.

The authors were informed by X. Zhou that recently, S.T. Yau and X. Zhou made some progress (independent of this paper) towards completing the Schoen–Yau outline of Theorem 2 in dimension n = 4.

1.4. Acknowledgements. We are grateful to Brian White useful discus- sions about isoperimetric inequalities, and to Boyu Zhang for several helpful conversations about 4-manifold topology. We would like to thank Davi Max- imo, Misha Gromov, and Fernando Cod´aMarques for their interest as well as Richard Schoen for bringing the K(π, 1) question to our attention and for his constant encouragement. We are grateful to Martin Lesourd for answering several questions concerning the work [18]. Finally, the second named author wants to thank Xin Zhou for sharing his insights on the Schoen-Yau survey paper [30] in the early stages of this work. SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 7

O.C was supported in part by a Terman Fellowship, a Sloan Fellowship, and an NSF grant DMS-1811059/2016403. C.L. was supported by NSF grant DMS-2005287.

2. Topological preliminaries In this section we collect some basic topological facts on closed aspherical manifolds. Let N be a smooth n-manifold. N is called aspherical, if its homotopy groups πj(N) is trivial for all integers j > 1. Equivalently, N is a Eilenberg–MacLane space K(π, 1), where π is the fundamental group of N. Alternatively, N is aspherical if and only if its universal cover, N˜ , is contractible [15, Theorem 4.5]. We ﬁx (N, g) a compact aspherical Riemannian n-manifold and denote by N˜ its universal cover. The following is standard. Lemma 5. N˜ is non-compact.

Proof. Because N˜ is contractible Hn(N,˜ Z2) = 0. However, any compact n-manifold has Hn(N,˜ Z2)= Z2. The next lemma is well known (and holds for the universal cover of any compact manifold). We recall the proof here for completeness.

Lemma 6. There exists a geodesic line σ : R → (N,˜ g) , i.e., d(σ(t1),σ(t2)) = |t2 − t1| for all t1,t2 ∈ R.

Proof. Fix p ∈ N˜ , and choose pi ∈ N˜ diverging. Let σi denote minimizing ′ geodesics between p and pi. Passing to a subsequence, σi converge to σ : [0, ∞) → N˜ a minimizing ray. For ti → ∞ choose deck transformations Φi ′ ′ ′ so that d(p, Φi(σ (ti))) is uniformly bounded. Then, σi(t)=Φi(σ (t + ti)) subsequentially converges to a geodesic line σ.

Choose two smooth functions ρ1, ρ2 : N → R so that

|ρ1(p) − d(p,σ([0, ∞))|≤ 1, |ρ2(p) − d(p,σ(0))|≤ 1. −1 For a large regular value of ρ1, L1 ≫ 1, set U := ρ1 ((−∞,L1)). By construction, ∂U is a smooth properly embedded hypersurface. −1 Lemma 7. For L2 suﬃciently large, σ ∩ ∂U ⊂ ρ2 ((−∞,L2]).

Proof. Observe that σ([0, ∞]) ⊂ U. Moreover, σ((−∞, −L1 − 4]) ∩ U = ∅. Indeed, if t1 < −L1 − 4 has σ(t1) ∈ U then

d(σ(t1),σ([0, ∞)) ≤ L1 + 2

As such, there is t2 ≥ 0 with

L1 + 4 ≤ |t1 − t2| = d(σ(t1),σ(t2)) ≤ L1 + 3. This is a contradiction. Thus σ ∩ ∂U is contained in a compact set. This completes the proof. 8 OTISCHODOSHANDCHAOLI

Fix a regular value L2 ≫ L1 of the function ρ2|∂R and set −1 M = (∂U) ∩ ρ2 ((−∞,L2)). Note that M is a smooth properly embedded compact oriented hypersurface with boundary. Perturb σ slightly so that it intersects M transversely (and σ ∩ ∂U = σ ∩ M) and we still have

|ρ1(p) − d(p,σ([0, ∞))|≤ 2, |ρ2(p) − d(p,σ(0))|≤ 2. We now verify several properties of M and σ.

Lemma 8. For L2 suﬃciently large, the curve σ has non-zero algebraic intersection with M. Proof. The curve σ(t) leaves and then re-enters U in (oppositely oriented) pairs until the smallest intersection time t, after which it never intersects ∂U. Because σ ∩ ∂U = σ ∩ M this proves the assertion.

Lemma 9. For L2 suﬃciently large, ∂M 6= ∅ and d(∂M,σ(R)) ≥ L1 − 4. −1 Proof. Note that ∂M ⊂ ρ2 (L2). First, suppose that ∂M = ∅. Then, by Lemma 8, we could conclude that [M] 6= 0 ∈ Hn−1(N˜) = 0. This is a contradiction. Suppose there is t2 ∈ R and p ∈ ∂M with d(p,σ(t2)) ≤ L1 − 3. Suppose that t2 ≥ 0. Then,

ρ1(p) ≤ d(p,σ([0, ∞))) + 2 ≤ L1 − 1. −1 This cannot hold (since ∂M ⊂ ρ1 (L1)). As such, we see that t2 < 0. Moreover, since d(∂M,σ(0)) ≥ L2 − 2, we ﬁnd

|t2| = d(σ(t2),σ(0)) ≥ d(p,σ(0)) − d(p,σ(t2)) ≥ L2 − L1 + 1.

On the other hand, we know there must be t1 ≥ 0 with d(p,σ(t1)) ≤ L1 + 2. We have

L2 − L1 + 1 ≤ |t1 − t2| = d(σ(t1),σ(t2))

≤ d(p,σ(t1)) + d(p,σ(t2)) ≤ 2L1 − 1.

This is a contradiction as long as L2 > 3L1 − 2. Proposition 10. For r > 0 there is R = R(r) with the following property. Suppose that α is a k-cycle in N˜ with α ⊂ Br(p) for some p ∈ N˜ . Then α = ∂β for β ⊂ BR(p).

Proof. Fix p0 ∈ N˜ . Since N˜ is contractible, for r > 0, Br(p0) is contractible in BR0(r)(p0) for some function R0 = R0(r) < ∞. In particular

Hk(Br(p0)) → Hk(BR0 (p0)) is the zero map, for k > 0. For any p ∈ N˜, there is a deck transformation Ψ so that d(p0, Ψ(p)) ≤ diam N. Assuch, Ψ(Br(p)) ⊂ Br+diam N (p0). There is β ∈ BR0(r+diam N)(p0) ⊂

BR0(r+diam N)+diam N (Ψ(p)) with ∂β = Ψ(α). As such −1 Ψ (β) ⊂ BR0(r+diam N)+diam N (p) SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 9 has ∂(Ψ−1(β)) = α. This completes the proof.

3. Warped µ-bubbles In this section we recall general existence and stability results for warped µ-bubbles. For n ≤ 7, consider (M, g) a Riemannian n-manifold with boundary and assume that ∂M = ∂−M ∪ ∂+M is a choice of labeling the components of ∂M so that neither of the sets ∂±M are empty. Fix a smooth function u > 0 on M and a smooth function h on M˚ with h → ±∞ on ∂±M. Choose a Caccioppoli set Ω0 with smooth boundary ∂Ω0 ⊂ M˚ and ∂+M ⊂ Ω0. Consider the following functional

n−1 n A(Ω) = u dH − (χΩ − χΩ0 )hu dH , (1) ∗ Z∂ Ω ZM for all Caccioppoli sets Ω in M with Ω∆Ω0 ⋐ M˚. We will call Ω minimizing A in this class a µ-bubble.

Remark 11. Geometrically, this is equivalent to the functional

1 n−1 ∗ 1 n ˜ 1 1 ˜ A(Ω × S )= H (∂ (Ω × S )) − (χΩ×S − χΩ0×S )h dH , (2) 1 ZΩ×S 1 1 1 for S -invariant Caccioppoli sets Ω × S inside (M1 × S , g˜), where g˜ is the warped product metric g˜ = g + u2dt2, and h˜ is deﬁned on M × S1 by h˜(x,t) = h(x). However, we ﬁnd it simplest to work with the form (1) instead of the warped product formulation.

3.1. Existence of µ-bubbles. The existence and regularity of a minimizer of A˜ among all Caccioppoli sets (without any equivarient assumptions) was claimed by Gromov in [13, Section 5.1], and was rigorously carried out by Zhu in [39, Proposition 2.1]. For the sake of completeness, we include a proof here.

Proposition 12. There exists a smooth minimizer Ω for A such that Ω∆Ω0 is compactly contained in the interior of M1.

Proof. Let Ω be a Caccioppoli set in M such that Ω∆Ω0 ⋐ M˚. By a standard approximation argument, we can assume that Ω has smooth boundary. We first show that, by adding to Ω a neighborhood of ∂+M, and subtracting τ from it a neighborhood of ∂−M, one decreases Ah. For τ > 0, denote Ω± the τ distance-τ neighborhood of ∂± in M. Choosing τ sufficiently small, Ω± has ρ a foliation {S±}ρ∈[0,τ] by smooth equidistant hypersurfaces to ∂±M. Denote ρ τ η the unit normal vector field of {S±} defined in this Ω±, pointing into M along ∂±M. Let τ > 0 be sufficiently small so that τ τ hu > div(uη) in Ω+, hu< − div(uη) in Ω−. 10 OTISCHODOSHANDCHAOLI

We compute

τ A(Ω ∪ Ω+) −A(Ω) = u dHn−1 − u dHn−1 − hu dHn ∗ τ τ Z∂ (Ω∪Ω+) Z∂Ω ZΩ+\Ω = u dHn−1 − u dHn−1 − hu dHn τ τ τ Z∂Ω+\Ω Z∂Ω∩Ω+ ZΩ+\Ω < u dHn−1 − u dHn−1 − div(uη) dHn. τ τ τ Z∂Ω+\Ω Z∂Ω∩Ω+ ZΩ+\Ω Moreover,

div(uη) dHn = (η · ν)u dHn−1 − (η · ν)u dHn−1 τ τ τ ZΩ+\Ω Z∂Ω+\Ω Z∂Ω∩Ω+ ≥ u dHn−1 − u dHn−1. τ τ Z∂Ω+\Ω Z∂Ω\Ω+ τ τ Thus A(Ω ∩ Ω+) < A(Ω). By an analogous calculation, A(Ω \ Ω−) < A(Ω). Hence it suﬃces to consider the inﬁmum of A among C, where C is the τ τ collection of Caccioppoli sets that contain Ω+ and are disjoin from Ω−. τ τ n Since |hu| < C1 in M \ (Ω+ ∪ Ω−), we conclude that A(Ω) > −C1H (M) for all Ω ∈ C. Hence I = inf{A(Ω) : Ω ∈ C} exists. Take a sequence n−1 n Ωk ∈ C with A(Ωk) → I. Then H (∂Ωk) < C(I + C1H (M)). By BV- compactness, taking a subsequence, the sets Ωk converge to a Caccioppoli set Ω. It follows that Ω is a minimizer of A, and thus has smooth boundary by standard regularity theory [34].

3.2. Stability. We now discuss the ﬁrst and second variation for a warped µ-bubble.

Lemma 13. If Ωt is a smooth 1-parameter family of regions with Ω0 = Ω and normal speed ψ at t = 0, then d A(Ω )= (Hu + h∇ u, νi− hu)ψ dHn−1 dt t M ZΣt where H is the scalar mean curvature of ∂Ωt and ν is the outwards pointing unit normal. In particular, a µ-bubble Ω satisﬁes

−1 H = −u h∇M u, νi + h along ∂Ω.

Lemma 14. Consider a µ-bubble Ω with ∂Ω=Σ. Assume that Ωt is a smooth 1-parameter family of regions with Ω0 = Ω and normal speed ψ at SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 11

d2 t = 0, then Q(ψ) := dt2 t=0(A(Ωt)) ≥ 0 where Q(ψ) satisﬁes

Q(ψ) 2 1 ˚ 2 2 2 ≤ |∇Σψ| u − 2 (RM − 1 − RΣ + |AΣ| )ψ u + (∆M u − ∆Σu)ψ Σ Z 1 −1 2 2 1 2 2 n−1 − 2 u h∇M u, νi ψ − 2 (1 + h + 2 h∇M h, νi)ψ u dH . Proof. Differentiating the first variation, we find Q(ψ)

1 2 2 2 2 = −ψu∆Σψ − 2 (RM − RΣ + |AΣ| + HΣ)ψ u + HΣh∇M u, νiψ ZΣ 2 2 2 n−1 +D u(ν,ν)ψ − h∇Σu, ∇Σψiψ − h∇M (hu),νiψ dH 2 1 ˚ 2 2 3 2 2 = |∇Σψ| u − 2 (RM − RΣ + |AΣ| )ψ u − 4 HΣψ u ZΣ 2 2 2 n−1 +(∆M u − ∆Σu)ψ −h∇M h, νi uψ − h h∇M u, νi ψ dH . Now use

1 2 2 1 −1 2 2 2 1 2 2 2 HΣψ u = 2 u h∇M u, νi ψ − h h∇M u, νi ψ + 2 h ψ u to write Q(ψ)

2 1 ˚ 2 2 2 ≤ |∇Σψ| u − 2 (RM − 1 − RΣ + |AΣ| )ψ u + (∆M u − ∆Σu)ψ Σ Z 1 −1 2 2 1 2 2 n−1 − 2 u h∇M u, νi ψ − 2 (1 + h + 2 h∇M h, νi)ψ u dH . This completes the proof.

4. Free boundary warped µ-bubbles We will need a generalization of the previous discussion to the free boundary case. For n ≤ 7, suppose that (M n, g) is a Riemannian n-manifold with co-dimension 2 corners in the sense that any point has a neighborhood diﬀeomorphic to one of the following: Rn, {x ∈ Rn : xn ≥ 0} or {x ∈ Rn : xn−1,xn ≥ 0}. Assume that M \ M˚ = ∂+M ∪ ∂−M ∪ ∂0M where ∂±M, ∂0M are all smooth submanifolds of M (possibly with boundary). We assume that ∂+M ∩ ∂−M = ∅ and ∂±M ∩ ∂0M consists of smooth co-dimension 2 closed submanifolds. Moreover, we assume that ∂±M meets ∂0M orthogonally. For an open set Ω ⊂ M we denote by ∂Ω the topological boundary in the sense that ∂Ω = Ω ∩ Ωc. Observe that with this deﬁnition, if Ω = M, ∂Ω= ∅. 12 OTISCHODOSHANDCHAOLI

∞ ∞ Fix a function u ∈ C (M) with u> 0 and h ∈ C (M \ (∂+M ∪ ∂−M)). Assume that h → ±∞ on ∂±M. Consider a Caccioppoli set Ω0 with smooth boundary satisfying ∂Ω0 ⊂ M˚ ∪ ∂0M. We assume that ∂0M satisﬁes −1 H∂0M + u h∇M u, ν∂0M i = 0 (3)

(i.e., we assume that the warped mean curvature of ∂0M vanishes; this prevents tangential contact between the free boundary µ-bubbles and ∂0M in the usual way that free boundary minimal surfaces do not make tangential contact with minimal components of the boundary). Consider the functional (as before)

n−1 n A(Ω) = u dH − (χΩ − χΩ0 )hu dH ∗ Z∂ Ω ZM for all Caccioppoli sets Ω in M with Ω∆Ω0 ⋐ M \(∂+M ∪∂−M). By similar arguments to the previous section (and using the warped mean curvature condition (3)), we can conclude the following result:

Proposition 15. There exists Ω with ∂Ω ⊂ M˚ ∪∂0M minimizing A among such regions. The boundary ∂Ω is smooth and meets ∂0M orthogonally. We have −1 H = −u h∇M u, ν∂Ωi + h along ∂Ω. Finally, if Σ is a component of ∂Ω, then for any ψ ∈ C1(Σ), we have 2 1 ˚ 2 2 2 0 ≤ |∇Σψ| u − 2 (RM − 1 − RΣ + |AΣ| )ψ u + (∆M u − ∆Σu)ψ Σ Z 1 −1 2 2 1 2 2 n−1 − 2 u h∇M u, νΣi ψ − 2 (1 + h + 2 h∇M h, νΣi)ψ u dH

2 n−2 − A∂0M (νΣ,νΣ)ψ u dH . Z∂Σ 5. Diameter bounds for certain surfaces In this section we derive diameter bounds for certain surfaces (in practice these will be stable µ-bubbles in certain 3-manifolds). Schoen–Yau [25] have proven that stable minimal surfaces in 3-manifolds of positive scalar curvature satisfy such an inequality. Moreover, they have proven that the 1-dimensional components of their minimal k-slicings satisfy a length bound [31]. It seems that the latter result contains precisely the fact needed here, but since it is slightly involved to compare the two settings, we state and prove the following result. For the amusement of the reader, we use µ-bubbles to give a slightly diﬀerent proof from the one given in [25, 31]. Lemma 16. For closed 2-dimensional Riemannian manifold (Σ2, g), suppose there is a smooth function λ> 0 so that 1 −1 2 ∆Σλ ≤−(K0 − KΣ)λ + 2 λ |∇Σλ| SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 13

2 for some K0 ∈ (0, ∞). Then diamg Σ ≤ K0 π. Proof. For contradiction, suppose there areq p,q ∈ Σ with L := d(p,q) >L := 2 π. 0 K0 There is ε> 0 and a smooth function q

ρ :Σ \ (Bε(p) ∪ Bε(q)) → (−L0/2,L0/2) with | Lip ρ| ≤ 1 − ε and ρ → −L0/2 at ∂Bε(p) and ρ → L0/2 at ∂Bε(q). We then take h(x)= −2(1 − ε) π tan( π ρ(x)). L0 L0 Observe that 1 2 K0 + 2 h(x) − |∇Σh|(x) 2 π2 2 π 2 π ≥ K0 + 2(1 − ε) 2 (tan ( ρ(x)) − sec ( ρ(x))) L0 L0 L0 2 π2 = K0 − 2(1 − ε) 2 L0 2 = K0 − (1 − ε) K0 > 0. We will use this below. Using Proposition 12 we can ﬁnd a µ-bubble Ω minimizing

1 2 A(Ω) = u dH − (χΩ − χΩ0 )hu dH ∗ Z∂ Ω ZΩ where Ω0 is some reference Caccioppoli set Bε(p) ⊂ Ω0 ⊂ Σ \ Bε(q). By −1 Lemma 13, kγ = −λ h∇Σλ, νγ i + h. As in Lemma 14, we compute

2 2 2 2 2 0 ≤ |∇γ ψ| λ − KΣψ λ − kγ ψ λ + (∆Σλ − ∆γλ)ψ Zγ 2 2 1 − h∇Σλ, νγ i ψ h − h∇Σh, νγ i ψ λ dH . 1 Take ψ = λ− 2 to ﬁnd

1 −2 2 2 −1 0 ≤ 4 λ |∇γ λ| − KΣ − kγ + λ (∆Σλ − ∆γ λ) Zγ −1 1 −λ h∇Σλ, νγ i h − h∇Σh, νγ i dH . 1 2 1 −2 2 −1 1 2 Using 2 kγ = 2 λ h∇Σλ, νγ i − λ h∇Σλ, νγi h + 2 h, we have

1 −2 2 −1 0 ≤ 4 λ |∇γ λ| − KΣ + λ (∆Σλ − ∆γλ) Zγ 1 −2 2 1 2 1 − 2 λ h∇Σλ, νγ i − 2 h − h∇Σh, νγ i dH

3 −2 2 −1 = − 4 λ |∇γλ| − KΣ + λ (∆Σλ) Zγ 1 −2 2 1 2 1 − 2 λ h∇Σλ, νγ i − 2 h − h∇Σh, νγ i dH , 14 OTISCHODOSHANDCHAOLI where we integrated by parts in the second step. Now, using the equation satisﬁed by λ, we have

3 −2 2 1 −2 2 1 −2 2 0 ≤ − 4 λ |∇γλ| + 2 λ |∇Σλ| − 2 λ h∇Σλ, νγ i Zγ 1 2 1 −K0 − 2 h − h∇Σh, νγ i dH . 2 2 2 Note that |∇Σλ| = |∇γ λ| + h∇Σλ, νγ i , so we ﬁnd

1 2 1 (K0 + 2 h + h∇Σh, νγ i)dH ≤ 0 Zγ

1 2 However, we have seen above that K0 + 2 h + h∇Σh, νγ i > 0. This is a contradiction, completing the proof.

The following is the free-boundary analogue of the previous result. We note that Carlotto–Franz [3, Proposition 1.8] have recently extended the original Schoen–Yau method [25] to the setting of stable free boundary minimal surfaces.

Lemma 17. For compact 2-dimensional Riemannian manifold (Σ2, g) with boundary, suppose there is a smooth function λ> 0 so that

1 −1 2 ∆Σλ ≤−(K0 − KΣ)λ + 2 λ |∇Σλ| in Σ for some K0 ∈ (0, ∞). Suppose also that h∇Σλ, ηi = −k∂Σλ along ∂Σ for η the outwards pointing unit normal. Then, diam Σ ≤ 2 π. g K0 q Proof. If the diameter bound fails, we can argue precisely as in Lemma 16 to ﬁnd a free boundary µ-bubble Ω. Choose some curve γ in ∂Ω. If γ is a closed loop, the calculations in Lemma 16 yield a contradiction. As such, we assume that γ : [a, b] → Σ is an arc with ∂γ ⊂ ∂Σ. Using Proposition 15, the calculation above carries over to yield

1 −2 2 −1 0 ≤ 4 λ |∇γλ| − KΣ + λ (∆Σλ − ∆γλ) Zγ 1 −2 2 1 2 1 − 2 λ h∇Σλ, νγi − 2 h −h∇Σh, νγ i dH

− (k∂Σ(γ(b)) − k∂Σ(γ(a))). SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 15

At this point, we integrate by parts and pick up boundary terms

3 −2 2 −1 0 ≤ − 4 λ |∇γ λ| − KΣ + λ (∆Σλ) Zγ 1 −2 2 1 2 1 − 2 λ h∇Σλ, νγ i − 2 h − h∇Σh, νγ i dH −1 −1 − (λ(b) h∇Σλ, ηi (γ(b)) − λ(a) h∇Σλ, ηi (γ(a)))

− (k∂Σ(γ(b)) − k∂Σ(γ(a)))

3 −2 2 1 −2 2 1 −2 2 = − 4 λ |∇γλ| + 2 λ |∇Σλ| − 2 λ h∇Σλ, νγi Zγ 1 2 1 −K0 − 2 h − h∇Σh, νγ i dH . The proof is now completed as before.

6. Proof of Theorem 2 For n = 4, 5, consider (N n, g) a smooth closed aspherical n-manifold with a Riemannian metric of non-negative scalar curvature. Running the Ricci flow for a short time yields a metric with positive scalar curvature unless g is Ricci flat. However, if g is Ricci flat, the Cheeger–Gromoll splitting theorem [4, Theorem 3] implies that the universal cover (N,˜ g˜) splits isometrically as (N˜ ′, g˜′) × Rk for N˜ ′ compact. Because N is aspherical, N˜ ′ is a point, so (N,˜ g˜) is flat Rn. As such, it suffices to prove that N does not admit a metric of positive scalar curvature. We will assume that n = 5, since if (N 4, g) is a compact aspherical 4-manifold with positive scalar curvature, then N × S1 is acyclic (cf. [15, Proposition 4.2]) and has positive scalar curvature when equipped with the product metric. To summarize the above discussion, we can assume that (N 5, g) is a compact aspherical Riemannian 5-manifold with scalar curvature Rg ≥ 5. We write N˜ for the universal cover of N. By the results in Section 2, there is a geodesic line σ in the universal cover N˜ so that for any L> 0 there is a compact two-sided hypersurface with boundary Mˆ 4 in N˜ with d(∂Mˆ4,σ(R)) ≥ 3L and so that Mˆ 4 has non-zero algebraic intersection with σ. We will take L sufficiently large below.

6.1. The σ-transversal 4-dimensional area-minimizer. Find a smooth two-sided compact area-minimizing hypersurface M4 homologous to Mˆ 4 relative to ∂M4 = ∂Mˆ 4. Stability of M4 and Rg ≥ 5 implies that

2 1 2 2 4 1 (|∇M4 ψ| − 2 (5 − RM4 + |A| )ψ dH ≥ 0, ∀ψ ∈ C0 (M). ZM4 ∞ ˚ As such, there is u4 ∈ C0 (M4), u4 > 0 on M4, with 1 ∆M4 u4 ≤− 2 (5 − RM4 )u4. (4) 16 OTISCHODOSHANDCHAOLI

Our goal is to show that for L sufficiently large, we can find Ω4 ⊂ M4 so that ∂Ω4 can be filled in by 4-chains that avoid σ(R). Then, combined with Ω4 this will yield a 4-cycle with non-trivial algebraic intersection with σ, contradicting H4(N˜) = 0.

6.2. The 3-dimensional µ-bubble. Pick ρ4 : N˜ → R a smoothing of d(·,σ(R)) so that | Lip ρ4| ≤ 2, ρ4 ≥ L + 4π +1 on ∂M4 and ρ4 = 0 on M4 ∩ σ(R). Fix ε> 0 small so that L + 4π + ε and L − ε are regular values of ρ4 and then deﬁne

ρ4(x) − L − 2π ρ˜4(x)= 2 , h4(x)= − tan (˜ρ4(x)) 4+ π ε ′ π π for x ∈ M4 := {ρ˜4(x) ∈ (− 2 , 2 )}. Observe that

2 1+ h4(x) − 2|∇h4|

2 2 1 ≥ 1 + tan (˜ρ4(x)) − 2 | Lip ρ4(x)| 2 ≥ 0 (5) 4+ π ε cos (˜ρ4(x)) ′ −1 on M4. Choosing a4 ∼ 0 a regular value ofρ ˜4, we can set Ω0 =ρ ˜ ((−∞, a4)) and use Proposition 12 to minimize

3 4 A4(Ω) := u4 dH − (χΩ − χΩ0 )h4u4 dH Z∂Ω ZM4 ⋐ ′ among Caccioppoli sets Ω with Ω∆Ω0 M4. Denote this minimizer by Ω4 and let M3 := ∂Ω4. Note that M3 is a cycle and 1 d(M3,σ(R)) ≥ 2 (L − ε). Using (4), (5) and Lemma 14 we see that

2 1 2 2 3 |∇M3 ψ| u4 − 2 (4 − RM3 )ψ u4 − (∆M3 u4)ψ dH ≥ 0 ZM3 1 for all ψ ∈ C (M3). In particular, there is u3 with

1 divM3 (u4∇M3 u3) ≤− 2 (4 − RM3 )u3u4 − (∆M3 u4)u3. This implies that

∆M3 (u3u4) = divM3 (u3∇M3 u4)+divM3 (u4∇M3 u3) (6) 1 ≤− 2 (4 − RM3 )u3u4 + h∇M3 u3, ∇M3 u4i .

As mentioned above, our goal is now to show that each component of M3 can be ﬁlled by a chain with uniformly bounded diameter as L →∞. Note that M3 is Yamabe positive but it might hold that the diameter of each component of M3 is unbounded as L →∞. SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 17

6.3. A slice and dice procedure for M3. We can consider each component of M3 separately, so it suﬃces to assume that M3 is connected. It will be important later to observe that M3 is orientable, by construction. 2 Set λ3 := u3u4 > 0. For Σ ⊂ M3, we deﬁne

Aˆ3(Σ) = λ3. ZΣ As one would expect, we can derive the following result for stable critical points of Aˆ3. Lemma 18. Suppose that Σ is a connected (two-sided) stable critical point 2 of Aˆ3. Then Σ is a topological sphere with H (Σ) ≤ 2π and diam Σ ≤ π. −1 Proof. Lemma 13 implies that H = −λ3 h∇M3 λ3,νi. Similarly, Lemma 14 combined with (6) yields

2 2 2 0 ≤ |∇Σψ| λ3 − (2 − KΣ)ψ λ3 − (∆Σλ3)ψ ZΣ 2 1 −1 2 2 2 + h∇M3 u3, ∇M3 u4i ψ − 2 λ3 h∇M3 λ3,νi ψ dH . Note that 2 2 2 2 2λ3 h∇M3 u3, ∇M3 u4i−h∇M3 λ3,νi ≤ |∇M3 λ3| − h∇M3 λ3,νi = |∇Σλ3| This yields

2 2 2 1 −1 2 2 2 0 ≤ |∇Σψ| λ3 − (2 − KΣ)ψ λ3 − (∆Σλ3)ψ + 2 λ3 |∇Σλ3| ψ dH ZΣ (7) 1 − 2 Take ψ = λ3 to ﬁnd

3 −2 2 −1 2 0 ≤ 4 λ3 |∇Σλ3| − (2 − KΣ) − λ3 (∆Σλ3) dH . ZΣ Integrating by parts on the last term, this implies

2 2 2H (Σ) ≤ KΣ dH = 2πχ(Σ). ZΣ As such, Σ is a topological sphere and H2(Σ) ≤ 2π. Returning to (7), we ﬁnd a smooth function u2 > 0 on Σ with 1 −1 2 divΣ(λ3∇Σu2) ≤−(2 − KΣ)u2λ3 − (∆Σλ3)u2 + 2 λ3 |∇Σλ3| u2. Set λ2 = u2λ3 = u2u3u4. We compute 1 −1 2 ∆Σλ2 ≤−(2 − KΣ)λ2 + h∇Σu2, ∇Σλ3i + 2 λ3 |∇Σλ3| . Now, we observe that −1 2 −1 2 λ2 |∇Σλ2| ≥ 2 h∇Σu2, ∇Σλ3i + λ3 |∇Σλ3| , so 1 −1 2 ∆Σλ2 ≤−(2 − KΣ)λ2 + 2 λ2 |∇Σλ2| . As such, the diameter bounds follow from Lemma 16. 18 OTISCHODOSHANDCHAOLI

We ﬁrst use Aˆ3-minimization to slice M3 into a manifold with simple second homology. This will allow us to use free boundary µ-bubbles to dice the resulting manifold into pieces that can be ﬁlled in a bounded radius.

Lemma 19. There are Σˆ 1,..., Σˆ k ⊂ M3 pairwise disjoint two-sided stable ˆ ˆ k ˆ critical points of A3 so that the manifold with boundary M3 := M3 \(∪j=1Σj) is connected and has H2(∂Mˆ 3) → H2(Mˆ 3) surjective. Proof. We proceed inductively. Assume that for k ≥ 0 we have chosen

Σˆ 1,..., Σˆ k ⊂ M3 pairwise disjoint two-sided stable critical points of Aˆ3. Suppose that for ˆ (k) k ˆ ˆ (k) ˆ (k) M3 := M3 \ (∪j=1Σj), the map H2(∂M3 ) → H2(M3 ) is not surjective. Consider α not in the image. Minimize Aˆ3 area in the homology class. ˆ (k) ˆ Because each component of ∂M3 is smooth and stationary for A3, we can find a representative of α ′ ′ α = [Σ1]+ · · · + [Σℓ] ′ ′ with the connected surfaces Σ1,..., Σℓ pairwise disjoint two-sided stable ′ ˆ (k) critical points. Moreover, each Σj is either contained in the interior of M3 ˆ (k) or coincides with a component of ∂M3 . By choice of α, there must be some ′ ˆ (k) j with Σj in the interior of M3 , not separating, and not in the image of ˆ (k) ˆ ′ H2(∂M3 ). Define Σk+1 =Σj. It suffices to show that this process terminates. If not, we obtain a sequence Σˆ 1,..., Σˆ k,... of stable critical points of Aˆ3. By Lemma 18, each Σˆ k has bounded area. Moreover, by a standard blow-up argument using [6, 5, 19], the Σˆ k have uniform curvature bounds. As such, passing to a ˆ ˆ subsequence, the Σk1 , Σk2 ,... are converging as one-sheeted graphs to some Σˆ ∞. However, this clearly contradicts the construction of the Σˆ j above, ˆ ˆ ˆ (kj+1) since it shows that Σkj+1 is homologous to Σkj in M3 , for j large. This completes the proof.

Choose Mˆ 3 as in the previous lemma. We recall that for Ω ⊂ Mˆ 3, we write ∂Ω for the topological boundary. Suppose that Ω ⊂ Mˆ 3 is a connected region with Ω 6= M and ∂Ω consisting of smooth surfaces with boundary. ′ Then, Ω is a manifold with corners. Write ∂ Mˆ 3 for the portion of the boundary of Mˆ 3 coinciding with the surfaces ∂Ω.

Lemma 20. A connected component of Mˆ 3 \ Ω contains exactly one com- ′ ponent of ∂ Mˆ 3.

Proof. Clearly a component of Mˆ 3 \ Ω contains at least one component of ′ ∂ Ω. If some component of Mˆ 3 \ Ω contained two (or more) components of ∂′Ω, we can ﬁnd a loop σ and component Σ′ ⊂ ∂′Ω so that σ intersects Σ′ SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 19 transversely in precisely one point. This implies that [σ] is not torsion in H1(Mˆ 3). On the other hand, we note that the long exact sequence in homology for a pair yields

H2(∂Mˆ3) → H2(Mˆ3) → H2(Mˆ 3, ∂Mˆ3) → H1(∂Mˆ3) = 0.

The final term vanishes since ∂Mˆ 3 consists of spheres, by Lemma 18. Thus we conclude that H2(Mˆ 3, ∂M3) = 0. Lefschetz duality (cf. [15, Theorem 1 3.43]) implies that H (Mˆ 3) = 0, so the universal coefficients theorem implies that H1(Mˆ 3) is torsion. This is a contradiction, finishing the proof.

We now show how to dice Mˆ 3 by free boundary µ-bubbles of controlled diameter and boundary behavior. Fix p in the interior of Mˆ 3. For technical reasons, we start by choosing Ω1 = Bε(p), chosen so that Bε(p) is contained in the interior of Mˆ 3. Assume that we have chosen regions

Ω1 ⊂ Ω2 ⊂ . . . Ωk with the following properties:

(1) ∂Ωj+1 ∩ ∂Ωj = ∅, (2) each component of Ωj+1 \ Ωj has diameter at most 10π (3) any component Υ ⊂ ∂Ωj has diam Υ ≤ π, and (4) each component of ∂Ωj is either a topological sphere or a topological disk with boundary in ∂Mˆ 3.

Assume that Ωk 6= Mˆ 3 and that there is some p ∈ Mˆ 3 with d(p, Ωk) > 4π (otherwise we set Ωk+1 = Mˆ 3. We now choose a µ-bubble based on a smoothing ρ of d(·, Ωk) with

| Lip ρ| < 2 and ρ|Ωk = 0 and then taking 1 π h(x)= − tan( 4 ρ(x) − 2 ) Observe that 1+ h(x)2 − 2|∇h|2 ≥ 0, by a similar calculation to the one used above. By Proposition 15, we can ﬁnd a free boundary µ-bubble Ωk+1 minimizing

2 3 A3(Ω) = λ3 dH − (χΩ − χΩ0 )λ3h dH . ∗ Z∂ Ω ZΩ Choose a component Σ of ∂Ωk+1. Clearly condition (1) is satisﬁed since h →∞ on ∂Ωk by construction.

Lemma 21. Ωk+1 satisﬁes property (2).

Proof. By Lemma 20, if p,q are in the same connected component of Ωk+1 \ Ωk, then they are connected there to a unique component of ∂Ωk. Call this component Σk. By construction d(p, Σk), d(q, Σk) ≤ 4π. Moreover, diam Σk ≤ π. Thus, d(p,q) ≤ 9π. This completes the proof. 20 OTISCHODOSHANDCHAOLI

We now verify that Ωk+1 satisﬁes conditions (3) and (4). Take a component Σ of ∂Ωk+1. By Proposition 15, Σ has (possibly empty) free boundary at ∂Mˆ 3. If Σ has no boundary, then Σ satisﬁes (6), and thus by the same proof as in Lemma 18, Σ is a topological sphere with diam Σ ≤ π. If Σ does have boundary, we can conclude that

2 1 2 2 0 ≤ |∇Σψ| λ3 − 2 (RMˆ 3 − 1 − 2KΣ)ψ λ3 + (∆Mˆ 3 λ3 − ∆Σλ3)ψ ZΣ 2 − 1 λ−1 ∇ λ ,ν ψ2 dH2 2 3 Mˆ 3 3 Σ D E 2 1 − A∂Mˆ 3 (νΣ,νΣ)ψ λ3 dH . Z∂Σ Using (6) (recall that λ3 = u3u4), we ﬁnd

2 2 2 0 ≤ |∇Σψ| λ3 − (1 − KΣ)ψ λ3 − (∆Σλ3)ψ ZΣ 2 2 1 −1 2 2 + ∇Mˆ 3 u3, ∇Mˆ 3 u4 ψ − 2 λ3 ∇Mˆ 3 λ3,νΣ ψ dH D E D E − A (ν ,ν )ψ2λ dH1. ∂Mˆ 3 Σ Σ 3 ∂Σ Z 1 − 2 Taking ﬁrst ψ = λ3 , we conclude that

1 −2 2 −1 0 ≤ 4 λ3 |∇Σλ3| − (1 − KΣ) − λ3 (∆Σλ3) ZΣ 2 +λ−1 ∇ u , ∇ u − 1 λ−2 ∇ λ ,ν dH2 3 Mˆ 3 3 Mˆ 3 4 2 3 Mˆ 3 3 Σ D E D E − A (ν ,ν ) dH1 ∂Mˆ 3 Σ Σ Z∂Σ 2 = K − 1 − 3 λ−2|∇ λ |2 + λ−1 ∇ u , ∇ u − 1 λ−2 ∇ λ ,ν dH2 Σ 4 3 Σ 3 3 Mˆ 3 3 Mˆ 3 4 2 3 Mˆ 3 3 Σ Σ Z D E D E − (λ−1 h∇ λ , ηi + A (ν ,ν )) dH1 3 Σ 3 ∂Mˆ 3 Σ Σ Z∂Σ 2 = K − 1 − 3 λ−2|∇ λ |2 + λ−1 ∇ u , ∇ u − 1 λ−2 ∇ λ ,ν dH2 Σ 4 3 Σ 3 3 Mˆ 3 3 Mˆ 3 4 2 3 Mˆ 3 3 Σ Σ Z D E D E − (−λ−1 ∇ λ ,ν + H − A (τ ,τ )) dH1 3 Σ 3 ∂Mˆ 3 ∂Mˆ 3 ∂Mˆ 3 ∂Σ ∂Σ Z∂Σ D E 2 3 −2 2 −1 1 −2 2 = KΣ − 1 − 4 λ3 |∇Σλ3| + λ3 ∇Mˆ 3 u3, ∇Mˆ 3 u4 − 2 λ3 ∇Mˆ 3 λ3,νΣ dH Σ Z D E D E + A (τ ,τ ) dH1. ∂Mˆ 3 ∂Σ ∂Σ Z∂Σ Here τ∂Σ is the unit tangential vector along ∂Σ. Since Σ meets ∂Mˆ 3 orthog- onall, A (τ ,τ )= k . Also, as in Lemma 18, the terms involving λ ∂Mˆ 3 ∂Σ ∂Σ ∂Σ 3 SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 21 are non-positive. Therefore we ﬁnd

2 1 H (Σ) ≤ KΣ + k∂Σ dH = 2πχ(Σ). ZΣ Z∂Σ This implies that Σ is a topological disk. Finally, as in Lemma 18 we ﬁnd u2 satisfying 1 −1 2 divΣ(λ3∇Σu2) ≤−(2 − KΣ)u2λ3 − (∆Σλ3)u2 + 2 λ3 |∇Σλ3| u2 in Σ

(h∇Σu2, ηi = A∂Mˆ 3 (νΣ,νΣ)u2 on ∂Σ

Thus, we have that λ2 = u2λ3 = u2u3u4 satisﬁes 1 −1 2 ∆Σλ2 ≤−(2 − KΣ)λ2 + 2 λ2 |∇Σλ2| . in Σ and

h∇Σλ2, ηi = A∂Mˆ 3 (νΣ,νΣ)λ2 + h∇Σλ3, ηi u2 = H λ − λ−1 ∇ λ ,ν λ − k λ ∂Mˆ 3 2 3 Σ 3 ∂Mˆ 3 2 ∂Σ 2

= −k∂Σλ2. D E Thus, Lemma 17 implies that diam Σ ≤ π.

6.4. Filling M3. We now show that M3 can be filled in N˜ by a 4-chain of bounded diameter (as L → ∞). As explained above, this will give the desired contradiction, proving Theorem 2. We first summarize the slice and dice procedure used above as it applies to M3. There exists a set of disjoint embedded spheres Σˆ 1,..., Σˆ k ⊂ M3 with diam Σˆ j ≤ π. Moreover, there is a set Υˆ 1,..., Υˆ ℓ ⊂ M of embedded disks ˆ k ˆ with diam Υj ≤ π and ∂Υj contained in ∪i=1Σi and so that the interiors of Υˆ j are pairwise disjoint with each other as well as with each Σˆ j. Finally, each connected component U1,U2,...,Um of k ˆ ℓ ˆ M3 \ ((∪j=1Σj) ∪ (∪j=1Υj)) has diameter bounded by 10π. Observe that any component of ∂Uj is a topological sphere (it will be smooth except it could have a right angle corner along one or two closed curves arising from points where the free boundary disks intersect the original spheres) of extrinsic diameter at most 3π. Write these spheres as 1 n(j) Γj ,..., Γj . By Proposition 10, there is R > 0 (independent of L) so i î that we can fill the Γj by Γj with extrinsic diameter at most R. Then, ˆ1 ˆn(j) Uj + Γj + · · · + Γj is a cycle with extrinsic diameter at most 2R + 10π. As such, there is R>ˆ 0 (independent of L) so that by Proposition 10 we can find a 4-chain Uˆj of extrinsic diameter at most Rˆ with ˆ ˆ1 ˆn(j) ∂Uj = Uj + Γj + · · · + Γj . 22 OTISCHODOSHANDCHAOLI

Note that m m n(j) ˆ i M3 − ∂Uj = − Γj. j j i X=1 X=1 X=1 i i Note that for each Γj, there is an index u(j, i) ∈ {1,...,k} so that Γj ˆ intersects the sphere Σu(j,i) but not any of the other spheres. As such, we i group the Γj by u(j, i) = 1, 2,...,k. Note that for a ∈ {1, 2,...,k},

i Γj i,j:uX(i,j)=a is thus a cycle of diameter at most 7π. As such, there is R>˜ 0 (independent of L) so that by Proposition 10 yet again, there is a 3-chain Ξa with i ∂Ξa = − Γj i,j:uX(i,j)=a and extrinsic diameter at most R˜. Thus, we have written

m k M = ∂ Uˆj + Ξa 3   j a=1 X=1 X where each term in the sum has diameter uniformly bounded as L → ∞. This completes the proof.

7. Proof of Theorem 3 Let n ≥ 3. For X a n-manifold (compact or non-compact), suppose that g is a complete metric on M = T n#X with non-negative scalar curvature. By a result of Kazdan [17], either g is Ricci flat or M admits a complete metric of positive scalar curvature. However, a complete Ricci flat metric on T n#X is easily seen to be flat (this follows from the splitting theorem; for example, see [4, Theorem 4]). As such, it suffices to consider the case of positive scalar curvature. We will pass to an appropriate covering space and apply the µ-bubble technique on the cover, after carefully choosing a weight function h. Fix ε> 0 small and define n n Ξ := {~x = (x1, · · · ,xn) ∈ R : |~x − ~k| > ε,~k ∈ Z }/ ∼ where (x1, · · · ,xn) ∼ (x1 + k1, · · · ,xn + kn) for k1, · · · , kn ∈ Z. By assump- tion, there is a map Ψ : Ξ → M so that Ψ is a diffeomorphism onto its image. By scaling, we can assume that Rg > 1 on Ψ(Ξ). Observe that M is (topologically) covered by Mˆ = (T n−1 × R)#ZX (unwind one of the S1 factors in T n). Define n n Ξ=ˆ {~x = (x1, · · · ,xn) ∈ R : |~x − ~k| > ε,~k ∈ Z }/ ∼ (8) SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 23 where (x1, · · · ,xn) ∼ (x1 + k1, · · · ,xn−1 + kn−1,xn) and note that the map Ψ lifts to Ψˆ : Ξˆ → Xˆ, a diffeomorphism onto its image Mˆ 0. It is useful to write

Mˆ = Mˆ 0 ∪ ∪k∈ZX˚k where each X˚k is (topologically X \ B for an n-balls B in X) attached to Mˆ 0 := Ψ(ˆ Ξ)ˆ along small spheres centered at (0, 0, k). We now define a function ρ0 : Mˆ → R as follows. Define Ξˆ2ε as in (8) but with 2ε in the place of ε. On Ψ(Ξˆ2ε), we take ρ0(x1, · · · ,xn)= xn. On the annuli Ψ(ˆ Ξˆ2ε \ Ξ)ˆ centered at (0, · · · , 0, k), interpolate between xn and 1 1 k + 2 (we can do this with uniformly C -norm independent of k). Then, on X˚k define k + 1 + dist (p, ∂X˚ ) k ≥ 0; ρ (p)= 2 g k 0 1 ˚ (k + 2 − distg(p, ∂Xk) k< 0. 1 We now define ρ1 to be a smoothing of ρ0. We can assume that ρ1 ≡ k + 2 in a small neighborhood of ∂X˚k. Since Mˆ is a covering space of M, there is L> 0 so that | Lip(ρ1)|g < L. πL 3 We may take L larger if necessary to assume that 2 = J + 4 for some J ∈ N. ˆ ˆ πL We now define a function h ∈ C(M, [−∞, ∞]) as follows. On M0∩{− 2 ≤ πL ρ1 ≤ 2 }, we define 1 h(p)= − tan( L ρ1(p)).

On the rest of Mˆ 0 we set h = ±∞ such that it is continuous to [−∞, ∞]. We now deﬁne h on X˚k. When |k| > J, set h = −∞ on X˚k. Now assume k ≤ J. For 0 ≤ k ≤ J and 2L p ∈ X˚ ∩ ρ < k + 1 + , k 1 2 −1 1 ( tan(L (k + 2 ))) or −J ≤ k< 0 and 2L p ∈ X˚ ∩ ρ > k + 1 + , k 1 2 −1 1 ( tan(L (k + 2 ))) we set 2L h(p)= 1 2L . ρ1(p) − (k + ) − 1 2 L−1 k tan( ( + 2 )) Otherwise we set h(p)= ±∞ such that h is continuous. We make several observations. First of all, since −J ≤ k ≤ J, we see that π −1 1 π − 2

Moreover, for p ∈ ∂X˚k, we have that −1 1 −1 h(p)= − tan(L (k + 2 )) = − tan(L ρ1(p)), and thus h is Lipchitz across ∂X˚k. Finally, if 0 ≤ k ≤ J, p ∈ X˚k and 2L ρ (p) ր k + 1 + , 1 2 −1 1 tan(L (k + 2 )) we have that h(p) → −∞. Similarly, if −J ≤ k ≤ 0, p ∈ X˚k and 2L ρ (p) ց k + 1 + , 1 2 −1 1 tan(L (k + 2 )) h1(p) →∞. Thus h is continuous. Note that {|h| < ∞} is compact. This is because this region is compact in M0, only ﬁnitely many ends X˚k are included in this set, and in each X˚k, the region where {|h| < ∞} is bounded.

Lemma 22. We can smooth h slightly to ﬁnd a function h ∈ C∞(Mˆ ) satisfying 2 Rgˆ + h − 2|∇h| > 0 (9) on {|h| < ∞}.

Proof. The function h constructed above is smooth away from ∂Xk (and Lipschitz there). As such, if we prove (9) for function h considered above, then we can easily ﬁnd a smooth function satisfying (9). We ﬁrst check (9) on Mˆ 0. There, Rgˆ > 1. As such, we have that 2 2 −1 −2 −1 Rgˆ + h − 2|∇h| > 1 + tan (L ρ1(p)) − 2sin (L ρ1(p)) > 0.

On the other hand, on X˚k (we assume that k ≥ 0 as the k < 0 case is similar), we only know that Rgˆ > 0. Nevertheless, we compute 2 2 2 4L − 4L Rgˆ + h − 2|∇h| > 2 = 0. 1 2L ρ1(p) − (k + ) − 1 2 tan(L−1(k+ )) 2 This completes the proof.

We can thus consider µ-bubbles with respect to the function h we have just defined. We fix ˆ ˆ 1 ˚ Ω0 := Ψ(Ξ ∩ {xn < − 2 }) ∪ (∪k<0Xk) ∩ {|h| < ∞}. We can minimize A among all Cacioppoli sets Ω such that Ω∆Ω0 is compactly contained in {|h| < ∞} by the argument given in Proposition 12 (with u = 1). Denote by Ω the connected component of the minimizer containing {ρ1 = −J}. Since n ≤ 7, each component of ∂Ω is compact and SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 25 regular. By the stability inequality for A from Lemma 14 (with u = 1) and (9), we see that Σ = ∂Ω satisfies

2 1 2 n−1 |∇ϕ| + 2 RΣϕ dH > 0 (10) ZΣ for all ϕ ∈ C∞(Σ). We can find a compact region Mˆ ′ ⊂ Mˆ with smooth boundary so that ′ ′ ∂Ω ⊂ Mˆ . Furthermore, we can arrange that ∂Mˆ ∩Mˆ 0 = Ψ(ˆ {z = ±(J+1)}). Note that the other boundary components of Mˆ ′ thus lie completely in some X˚k. In particular, ∂Mˆ \ Mˆ 0 bounds some compact manifold with boundary. Cap these components off and then glue the {z = J +1} and {z = −(J +1)} tori to each other. We thus find a manifold M˜ diffeomorphic to T n#X˜ for X˜ closed and Σn−1 ⊂ M˜ a hypersurface that is homologous to [T n−1 × {∗}] ∈ Hn−1(M˜ ) that satisfies (10). (We have not constructed a metric on M˜ , but this does not matter in the remaining part of the argument, all we need is the topology of M˜ and Σ as well as the fact that Σ satisfies (10).) We claim that this leads to a contradiction following the argument in [28, 31]. Indeed, on the one hand (10) implies that (each component of) Σ has positive first eigenvalue of the conformal Laplacian, and thus admits a metric of positive scalar curvature. On the other hand, we can pull back the 1-forms dx1,...,dxn−1 along the map π : (T n#X˜) → T n to find 1-forms ω1,...,ωn−1 so that ω1 ∧···∧ ωn−1 6= 0. ZΣ (this follows from the fact that Σ is homologous to T n−1 × {∗}). The proof can now be completed using the inductive method of [28, 31].

8. Schoen-Yau-Schick manifolds In this section, we briefly indicate an extension of Theorem 3 to manifolds in the form of (M ×S1)#X, where M is a Schoen–Yau–Schick manifold (abbreviated as SYS manifold, following the definition of Gromov [12, Sec- tion 5]). These manifolds was first considered in the celebrated work [28] of Schoen-Yau where the inductive descent argument was introduced. In [28], these manifolds are said to be of class Cn. We recall the definition here. Definition 23 ([28][21][12][31]). Let n ≥ 2. A compact orientable n-manifold M is called Schoen–Yau–Schick, if there exist n−2 integer homology classes 1 h1, · · · , hn−2 ∈ H (M) such that σ = h1 ⌢ · · · ⌢ hn−2 ⌢ [M] ∈ H2(M, Z) is non-spherical. That is, σ is not contained in the image of the Hurewicz homomorphism π2(M) → H2(M). For example, the torus is an SYS manifold. Using minimal surface and induction descent argument, Schoen–Yau in [28] proved that SYS manifolds of dimension at most 7 does not admit positive scalar curvature metrics. In 26 OTISCHODOSHANDCHAOLI

[21], Schick constructed an SYS manifold which is a counterexample to the unstable Gromov-Lawson-Rosenberg conjecture. Let n ≤ 6, and M n be an SYS manifold. By passing to the cover M × R and constructing the same functions ρ0, ρ1 and h as in Section 6, we can extend Theorem 3 to the following.

Theorem 24. Let 2 ≤ n ≤ 6, M n be an Schoen–Yau–Schick manifold. For any (n + 1)-manifold X, the connected sum (M × S1)#X does not admit a complete metric of positive scalar curvature.

References [1] Costante Bellettini and Neshan Wickramasekera, Stable prescribed-mean- curvature integral varifolds of codimension 1: regularity and compactness, https://arxiv.org/abs/1902.09669 (2019). [2] Hubert Bray, Simon Brendle, and Andre Neves, Rigidity of area-minimizing two-spheres in three-manifolds, Comm. Anal. Geom. 18 (2010), no. 4, 821–830. MR 2765731 [3] Alessandro Carlotto and Giada Franz, Inequivalent complexity criteria for free boundary minimal surfaces, Adv. Math. 373 (2020), 107322. MR 4129480 [4] Jeff Cheeger and Detlef Gromoll, The splitting theorem for manifolds of nonnegative Ricci curvature, J. Differential Geometry 6 (1971/72), 119–128. MR 303460 [5] M. do Carmo and C. K. Peng, Stable complete minimal surfaces in R3 are planes, Bull. Amer. Math. Soc. (N.S.) 1 (1979), no. 6, 903–906. MR 546314 (80j:53012) [6] Doris Fischer-Colbrie and Richard Schoen, The structure of complete stable minimal surfaces in 3-manifolds of nonnegative scalar curvature, Comm. Pure Appl. Math. 33 (1980), no. 2, 199–211. MR 562550 (81i:53044) [7] Mikhael Gromov and H. Blaine Lawson, Jr., Spin and scalar curvature in the presence of a fundamental group. I, Ann. of Math. (2) 111 (1980), no. 2, 209–230. MR 569070 [8] , Positive scalar curvature and the Dirac operator on complete Riemann- ian manifolds, Inst. Hautes Etudes´ Sci. Publ. Math. (1983), no. 58, 83–196 (1984). MR 720933 [9] Misha Gromov, Large Riemannian manifolds, Curvature and topology of Riemannian manifolds (Katata, 1985), Lecture Notes in Math., vol. 1201, Springer, Berlin, 1986, pp. 108–121. MR 859578 [10] , Positive curvature, macroscopic dimension, spectral gaps and higher signa- tures, Functional analysis on the eve of the 21st century, Vol. II (New Brunswick, NJ, 1993), Progr. Math., vol. 132, Birkhäuser Boston, Boston, MA, 1996, pp. 1–213. MR 1389019 [11] , 101 questions, problems and conjectures around scalar curvature, 2017. [12] , Metric inequalities with scalar curvature, Geom. Funct. Anal. 28 (2018), no. 3, 645–726. MR 3816521 [13] , Four lectures on scalar curvature, 2019. [14] Larry Guth, Volumes of balls in large Riemannian manifolds, Ann. of Math. (2) 173 (2011), no. 1, 51–76. MR 2753599 [15] Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002. MR 1867354 [16] Mikhail Katz, The first diameter of 3-manifolds of positive scalar curvature, Proc. Amer. Math. Soc. 104 (1988), no. 2, 591–595. MR 962834 [17] Jerry L. Kazdan, Deformation to positive scalar curvature on complete manifolds, Math. Ann. 261 (1982), no. 2, 227–234. MR 675736 SOAP BUBBLES AND POSITIVE SCALAR CURVATURE 27

[18] Martin Lesourd, Ryan Unger, and S.-T. Yau, R> 0 for open manifolds and the positive mass conjecture with bends, https://www.youtube.com/watch?v=c4EG87V8tWI (2020). [19] Aleksei V. Pogorelov, On the stability of minimal surfaces, Soviet Math. Dokl. 24 (1981), 274–276. [20] John G. Ratcliffe and Steven T. Tschantz, Some examples of aspherical 4-manifolds that are homology 4-spheres, Topology 44 (2005), no. 2, 341–350. MR 2114711 [21] Thomas Schick, A counterexample to the (unstable) Gromov-Lawson-Rosenberg conjecture, Topology 37 (1998), no. 6, 1165–1168. MR 1632971 [22] R. Schoen and S.-T. Yau, Conformally flat manifolds, Kleinian groups and scalar curvature, Invent. Math. 92 (1988), no. 1, 47–71. MR 931204 [23] , Lectures on differential geometry, Conference Proceedings and Lecture Notes in Geometry and Topology, I, International Press, Cambridge, MA, 1994, Lecture notes prepared by Wei Yue Ding, Kung Ching Chang [Gong Qing Zhang], Jia Qing Zhong and Yi Chao Xu, Translated from the Chinese by Ding and S. Y. Cheng, With a preface translated from the Chinese by Kaising Tso. MR 1333601 [24] Richard Schoen, Conformal deformation of a Riemannian metric to constant scalar curvature, J. Differential Geom. 20 (1984), no. 2, 479–495. MR 788292 [25] Richard Schoen and S. T. Yau, The existence of a black hole due to condensation of matter, Comm. Math. Phys. 90 (1983), no. 4, 575–579. MR 719436 [26] Richard Schoen and Shing-Tung Yau, Existence of incompressible minimal surfaces and the topology of three-dimensional manifolds with nonnegative scalar curvature, Ann. of Math. (2) 110 (1979), no. 1, 127–142. MR 541332 [27] , On the proof of the positive mass conjecture in general relativity, Comm. Math. Phys. 65 (1979), no. 1, 45–76. MR 526976 [28] , On the structure of manifolds with positive scalar curvature, Manuscripta Math. 28 (1979), no. 1-3, 159–183. MR 535700 [29] , Complete three-dimensional manifolds with positive Ricci curvature and scalar curvature, Seminar on Differential Geometry, Ann. of Math. Stud., vol. 102, Princeton Univ. Press, Princeton, N.J., 1982, pp. 209–228. MR 645740 [30] , The structure of manifolds with positive scalar curvature, Directions in partial differential equations, Elsevier, 1987, pp. 235–242. [31] , Positive scalar curvature and minimal hypersurface singularities, https://arxiv.org/abs/1704.05490 (2017). [32] Richard M. Schoen, Variational theory for the total scalar curvature functional for Riemannian metrics and related topics, Topics in calculus of variations (Montecatini Terme, 1987), Lecture Notes in Math., vol. 1365, Springer, Berlin, 1989, pp. 120–154. MR 994021 [33] Daniel Stern, Scalar curvature and harmonic maps to S1, https://arxiv.org/abs/1908.09754 (2019). [34] Italo Tamanini, Regularity results for almost minimal oriented hypersurfaces in Rn, Quaderni del Dipartimento di Matematica dell’ Università di Lecce 1 (1984), 1–92. [35] Jian Wang, Contractible 3-manifolds and positive scalar curvature, Ph.D. thesis, Uni- versitéGrenoble Alpes, 2019. [36] Xin Zhou, On the multiplicity one conjecture in min-max theory, https://arxiv.org/abs/1901.01173 (2019). [37] Xin Zhou and Jonathan Zhu, Existence of hypersurfaces with prescribed mean curvature I—generic min-max, Camb. J. Math. 8 (2020), no. 2, 311–362. MR 4091027 [38] Jintian Zhu, Rigidity results for complete manifolds with nonnegative scalar curvature, https://arxiv.org/abs/2008.07028 (2020). [39] , Width estimate and doubly warped product, to appear in Trans. Amer. Math. Soc., https://arxiv.org/abs/2003.01315 (2020). 28 OTISCHODOSHANDCHAOLI

Department of Mathematics, Stanford University, Building 380, Stanford, CA 94305, USA E-mail address: [email protected]

Department of Mathematics, Princeton University, Fine Hall, 304 Wash- ington Road, Princeton, NJ 08540, USA E-mail address: [email protected]