Home Minimal surfaces with low genus in lens spaces
Article Open Access

Minimal surfaces with low genus in lens spaces

  • Xingzhe Li , Tongrui Wang ORCID logo EMAIL logo and Xuan Yao
Published/Copyright: October 2, 2025

Abstract

Given a Riemannian RP 3 with a bumpy metric or a metric of positive Ricci curvature, we show that there either exist four distinct minimal real projective planes, or there exist one minimal real projective plane together with two distinct minimal 2-spheres. Our proof is based on a variant multiplicity one theorem for the Simon–Smith min-max theory under certain equivariant settings. In particular, we show under the positive Ricci assumption that RP 3 contains at least four distinct minimal real projective planes and four distinct minimal tori. Additionally, the number of minimal tori can be improved to five for a generic positive Ricci metric on RP 3 by the degree method. Moreover, using the same strategy, we show that, in the lens space L ( 4 m , 2 m ± 1 ) , m 1 , with a bumpy metric or a metric of positive Ricci curvature, there either exist N ( m ) distinct minimal Klein bottles, or there exist one minimal Klein bottle and three distinct minimal 2-spheres, where N ( 1 ) = 4 , N ( m ) = 2 for m 2 , and the first case happens under the positive Ricci assumption.

1 Introduction

In 1917, Birkhoff [5] proposed a min-max method and showed the existence of a closed geodesic in any Riemannian 2-sphere S 2 . Using multi-parameter families of closed curves to sweep out S 2 , Lusternik–Schnirelmann [31] (see also Grayson [18]) further showed that there are at least three closed geodesics in any Riemannian 2-sphere.

In one higher dimension, investigating minimal surfaces in S 3 is also a significant topic in differential geometry. In [28], Simon–Smith set out a min-max process to construct minimal surfaces with controls on topology, which is a variant of the min-max theory established by Almgren [1, 2] and Pitts [39] for minimal hypersurfaces (see also Schoen–Simon [41]). In particular, they proved the existence of an embedded minimal 2-sphere in any Riemannian S 3 (cf. [12]). In analogy with the result of Lusternik–Schnirelmann, S. T. Yau posed the following problem in his famous 1982 Problem Section [52].

Conjecture 1.1

Conjecture 1.1 (S. T. Yau [52])

There are four distinct embedded minimal spheres in any manifold diffeomorphic to S 3 .

Towards this conjecture, White [51] first used degree methods to show the existence of at least two embedded minimal spheres in any Riemannian S 3 with positive Ricci curvature. In particular, if the metric is sufficiently close to the round metric, White [51] showed that there are at least four embedded minimal spheres. Additionally, it was also shown in [51] that the set of bumpy metrics is generic in the Baire sense, where a metric g M on a given closed manifold 𝑀 is called bumpy if every closed embedded minimal hypersurface is non-degenerate. Using Simon–Smith min-max theory and mean curvature flow, Haslhofer–Ketover [20] applied the catenoid estimates (cf. [27]) to overcome the difficulty of multiplicities and proved the existence of at least two embedded minimal spheres in S 3 with bumpy metrics. Recently, significant progress was made by Wang–Zhou [48], which resolved Yau’s conjecture in S 3 with bumpy metrics or positive Ricci curvature via Simon–Smith min-max.

Inspired by the work of Wang–Zhou [48], we set out this paper to investigate minimal surfaces with low genus in lens spaces.

1.1 Minimal projective planes in RP 3

The motivation of Yau’s conjecture is closely connected to the topological structure of the space of embedded spheres in S 3 (cf. Hatcher’s proof of the Smale conjecture [21, Appendix (14)]). In light of the generalized Smale conjecture for lens spaces (cf. [3, 4, 26]), one would expect that the space of embedded projective planes in RP 3 can retract onto the space ℰ of great projective planes (the quotient of great spheres). Note that ℰ can be parameterized by R P 3 .[1] Therefore, except for the area minimizing projective plane (associated with H 0 ( E , Z 2 ) ), we can build three nontrivial (multi-parameter) families of great projective planes sweeping out the ambient manifold RP 3 , each of which is associated with the generator of H i ( E , Z 2 ) Z 2 , i = 1 , 2 , 3 , respectively (cf. Section 6.1). Analogous to Conjecture 1.1, it is now reasonable to ask whether there exist at least four distinct embedded minimal projective planes in ( RP 3 , g RP 3 ) ?

In this paper, we use equivariant min-max theory and a variant multiplicity one result to investigate the above question. Our first main result is the following dichotomy theorem.

Theorem 1.2

Given a Riemannian RP 3 with a bumpy metric or a metric of positive Ricci curvature, either

  1. there exist at least four distinct embedded minimal projective planes, or

  2. there exist one embedded minimal projective plane and two embedded minimal spheres.

In particular, (i) is valid when RP 3 has positive Ricci curvature.

In [20], Haslhofer–Ketover proved in ( RP 3 , g RP 3 ) with positive Ricci curvature that there are at least two embedded minimal projective planes. Our result can now improve the number of minimal projective planes in ( RP 3 , g RP 3 ) to four under the positive Ricci assumption. Indeed, a minimal sphere Σ in RP 3 will be lifted to two disjoint minimal spheres Σ + Σ in S 3 so that the antipodal map on S 3 permutes Σ ± . Combining with the embedded Frankel property in manifolds with positive Ricci curvature (i.e. any two closed embedded minimal hypersurfaces must intersect), we conclude Theorem 1.2 (ii) cannot happen if ( RP 3 , g RP 3 ) has positive Ricci curvature.

In recent years, we have witnessed the tremendous progress of Almgren–Pitts min-max theory. One peak of the development is the resolution of Yau’s conjecture [52, Problem 88] on the existence of infinitely many closed minimal surfaces due to Marques–Neves [34] and Song [43]. Meanwhile, several significant advancements [22, 36, 44] in the spatial distribution of closed minimal hypersurfaces were made by the multi-parameter min-max theory coupled with the Weyl law for the volume spectrum [29]. Moreover, after the establishment of a min-max theory for prescribed mean curvature hypersurfaces [57, 56], Zhou [54] solved the multiplicity one conjecture in the Almgren–Pitts setting. Combining with the Morse index estimates, Marques–Neves [33, 35] complete their celebrated program on the Morse theory for the area functional (see also [32]). We refer to [55] for a more detailed history.

1.2 Strategy of the proof

Although our proof shares the same spirit as Wang–Zhou’s resolution [48], one major difference is that the sweepouts are formed by 1-sided projective planes. However, Wang–Zhou’s multiplicity one theorem [48, Theorem B] relies heavily on the min-max theory [48, Theorem 2.4] for prescribing mean curvature functionals A h , (2.2), while it is inappropriate to define the A h functional or prescribed mean curvature functions for 1-sided surfaces. Hence [48, Theorem B] cannot be applied directly.

Let π : S 3 RP 3 be the double cover and g the involution map in S 3 , i.e.

π ( p ) = π ( g ( p ) ) and g ( p ) p for all p S 3 .

Heuristically, note that a projective plane RP 2 in RP 3 can be lifted to a sphere S = π 1 ( RP 2 ) which separates S 3 into two parts Ω + , Ω so that Ω ± = g ( Ω ) . Hence if 𝜈 is a unit normal along 𝑆 (with respect to the lifted metric), then d g ( ν ) = ν and the mean curvature 𝐻 of 𝑆 with respect to 𝜈 is antisymmetric H ( p ) = H ( g ( p ) ) . This inspired us to lift the sweepouts in RP 3 back into S 3 and consider the Z 2 = { id , g } equivariant min-max theory in S 3 for the prescribed mean curvature functionals A h with antisymmetric ℎ (i.e. h ( p ) = h ( g ( p ) ) ).

As one key novelty of this paper, we show the following multiplicity one theorem in a certain equivariant setting analogy to the above, which provides an alternative way to handle the multiplicities in the 1-sided Simon–Smith min-max theory.

Theorem 1.3

Let ( M , g M ) be a closed connected orientable 3-dimensional Riemannian manifold, and let 𝐺 be a finite group acting freely and effectively as isometries on 𝑀 so that 𝐺 admits an index 2 subgroup G + with coset G = G G + . Suppose Σ 0 is an orientable 𝐺-invariant genus- g 0 surface, and X G ± consists of all the 𝐺-equivariant embedding Σ = ϕ ( Σ 0 ) of Σ 0 into 𝑀 so that M Σ = Ω + Ω and G + Ω ± = Ω ± , G Ω ± = Ω . Then, given any homotopy class of smooth sweepouts formed by surfaces in X G ± , the associated min-max varifold can be chosen as a 𝐺-varifold induced by a closed embedded 𝐺-invariant minimal surface Γ with connected components { Γ j } j = 1 J and integer multiplicities { m j } j = 1 J so that

  1. if Γ j is unstable and 2-sided, then m j = 1 ;

  2. if Γ j is 1-sided, then its connected double cover is stable.

Moreover, the weighted total genus of Γ given by (4.1) is bounded by g 0 .

Remark 1.4

(1) By re-merging the components Γ j i that lie in the same 𝐺-orbit G Γ j , we can assume each Γ j is 𝐺-invariant and Γ j / G is connected. In applications, one can first take a sweepout of M / G formed by 1-sided surfaces in { Σ / G Σ X G ± } , and then pull it back to a sweepout of 𝑀 in X G ± . Clearly, the min-max 𝐺-surfaces { Γ j } j = 1 J given by Theorem 1.3 can be reduced to min-max surfaces { Γ j / G } j = 1 J in M / G with the same multiplicity. One should note that the multiplicity m j is related to the stability and the orientability of Γ j by Theorem 1.3, not to Γ j / G .

(2) For a general compact Lie group 𝐺 acting by isometries on M n + 1 with

3 codim ( G x ) 7 for all x M ,

the 𝐺-equivariant min-max theory for closed minimal 𝐺-hypersurfaces has been established by Ketover [23], Liu [30], and the second author [46, 47] in different settings. However, our scenarios cannot be covered directly by these works since we also need an equivariant min-max for the A h -functional (2.2) and both ℎ and Ω ± are G -antisymmetric.

After applying the above theorem to M = S 3 , G = { id , g } , G + = { id } , G = { g } , and X G ± = { π 1 ( P ) P is an embedded RP 2 RP 3 } , we obtain some 𝐺-invariant unions of disjoint minimal spheres { Γ j } j = 1 J with multiplicities { m j } j = 1 N satisfying (i)–(ii), which alternatively gives a multiplicity result for the min-max outputs { Γ ̃ i = π ( Γ j ) } j = 1 J in RP 3 .

Meanwhile, another difficulty arises as { Γ ̃ j } j = 1 J may also contain minimal spheres instead of minimal projective planes. Since spheres have no contribution to the total genus, the genus bound does not help to distinguish them. Nevertheless, we can show that (Proposition 5.4) every min-max minimal surface j = 1 J Γ ̃ j associated with the sweepouts formed by (1-sided) projective planes contains exactly one minimal projective plane, i.e. Γ ̃ 1 RP 2 and Γ ̃ j 2 S 2 . Hence, if RP 3 has positive Ricci curvature, we see j = 1 J Γ ̃ j = Γ ̃ 1 RP 2 (by the Frankel property) and π 1 ( Γ ̃ 1 ) is unstable in S 3 , which implies m j = 1 (Theorem 1.3) and gives four distinct minimal projective planes associated with H i ( E , Z 2 ) , i = 0 , 1 , 2 , 3 .

However, in general, we can neither eliminate minimal spheres in { Γ ̃ j } j = 1 J nor rule out the existence of minimal projective planes P RP 3 with stable π 1 ( P ) S 3 . Moreover, if we cut RP 3 along a minimal projective plane 𝑃 with stable π 1 ( P ) S 3 , we would obtain one 3-ball 𝐵 with boundary B = π 1 ( P ) . After adding a cylindrical end π 1 ( P ) × [ 0 , ) to 𝐵 and applying Song’s technique [43] as in [48, Sections 8.2, 8.4], one can only find minimal spheres (instead of minimal projective planes) confined in int ( B ) . Therefore, given the above phenomena, it is reasonable to propose the dichotomy theorem (Theorem 1.2) for the coexistence of both minimal projective planes and minimal spheres.

Indeed, we would conjecture that, under a certain metric on RP 3 , every minimal real projective plane constructed via min-max could be the same area minimizing one, but with possibly different multiplicities. A possible example is the Riemannian RP 3 so that its double cover S 3 with the pull-back metric looks like a symmetric dumbbell with a long thin neck S 2 × [ R , R ] satisfying that the center S 2 × { 0 } covers the area minimizing RP 2 , and for t 0 , S 2 × { t } has positive mean curvature pointing towards S 2 × { 0 } .

1.3 Minimal surfaces with Euler characteristic 0 in lens spaces L ( p , q )

The above idea can also be applied to investigate minimal Klein bottles and minimal tori in lens spaces L ( p , q ) S 3 / Z p , where p 2 , 1 q p 1 , and Z p = ξ p , q acts freely on S 3 by (7.1).

In particular, we have a very similar dichotomy theorem for minimal Klein bottles in L ( 4 m , 2 m ± 1 ) as shown below. Note that L ( 4 m , 2 m ± 1 ) , m 1 , are the only lens spaces that contain embedded Klein bottles (cf. [8, Corollary 6.4]).

Theorem 1.5

Let m 1 be an integer, N ( 1 ) = 4 , and N ( m ) = 2 for m 2 . Then, given a lens space L ( 4 m , 2 m ± 1 ) with a bumpy metric or a metric of positive Ricci curvature, either

  1. there exist N ( m ) embedded minimal Klein bottles, or

  2. there exist one embedded minimal Klein bottle and three embedded minimal spheres.

In particular, (i) is valid when L ( 4 m , 2 m ± 1 ) has positive Ricci curvature.

One notices that an embedded Klein bottle 𝐾 in L ( 4 m , 2 m ± 1 ) can be lifted to an embedded 𝐺-invariant torus 𝑇 in S 3 so that S 3 T has two G + -invariant components U + U with G permuting them, where

G = Z 4 m = ξ 4 m , 2 m ± 1 , G + = Z 2 m = ξ 4 m , 2 m ± 1 2 and G = G G + .

Hence, using the idea in Remark 1.4, we can lift the sweepout in L ( 4 m , 2 m ± 1 ) formed by Klein bottles to a family of Z 4 m -invariant tori in S 3 , and apply Theorem 1.3 to obtain closed embedded Z 4 m -invariant minimal surfaces j = 1 J Γ j in S 3 with the total weighted genus bounded by 1. Similarly,

j = 1 J Γ ̃ j = j = 1 J Γ j / Z 4 m

contains exactly one embedded minimal Klein bottle Γ ̃ 1 (Corollary 7.2). Then Γ ̃ 1 must have multiplicity one due to the weighted genus bound. The difficulty mainly comes from the possible existence of minimal spheres, which can be handled similarly by considering the coexistence of them.

Moreover, we also establish a new result regarding minimal tori in lens spaces.

Theorem 1.6

Let L ( p , q ) , p 2 , be a lens space with a metric of positive Ricci curvature. Then

  1. L ( 2 , 1 ) RP 3 contains at least four embedded minimal tori; in particular, for almost every (in the sense of Baire category) Riemannian metric of positive Ricci curvature on RP 3 , there are at least five embedded minimal tori;

  2. L ( p , 1 ) and L ( p , p 1 ) with p > 2 contains at least three embedded minimal tori;

  3. L ( p , q ) with q { 1 , p 1 } contains at least one embedded minimal torus.

Remark 1.7

Using the weighted genus bound [25, (2.6)] and the catenoid estimates [27], Ketover concluded under the positive Ricci assumption that there are three minimal tori in L ( p , 1 ) , p 2 (see [25, Theorem 4.9]), and two minimal Klein bottles in L ( 4 p , 2 p ± 1 ) , p > 1 (see [25, Theorem 4.8]). Theorems 1.5 and 1.6 extend these results by some new constructions and the multiplicity one theorem.

Ketover–Liokumovich [26, Theorem 1.4] showed as an equivalent formulation of the Smale conjecture in lens spaces that the space of Heegaard tori in L ( p , q ) retracts onto the space E p , q of Clifford tori. Hence the parameterization of E p , q (cf. [26, Proposition 2.3]) and the Lusternik–Schnirelmann theory support the counts of minimal tori in Theorem 1.6.

It is also noteworthy that the space of Clifford tori in S 3 and RP 3 can both be parameterized by R P 2 × R P 2 . One would then expect five distinct minimal tori in a Riemannian S 3 and RP 3 (cf. White’s conjecture in [50]). However, after White’s existence result of one minimal torus in any positive Ricci curved S 3 (see [51, Theorem 6.2]) via the degree method, there has been no substantial progress (to the best of the author’s knowledge) in the general construction of more minimal tori in a Riemannian S 3 . Nevertheless, our Theorem 1.6 (i) has fulfilled the expectation in RP 3 to a certain extent.

The main challenges in finding minimal tori lie in two aspects. Firstly, the moduli space 𝒳 of embedded tori in S 3 or RP 3 is very complicated, making it difficult to find an element

α H 1 ( X X , X ; Z 2 ) with α 5 0 H 5 ( X X , X ; Z 2 ) ,

while such an 𝛼 is crucial for applying the Lusternik–Schnirelmann theory. In Section 8.1, we address this by considering part of the boundary Y : = { great circles } X and identifying an element α H 1 ( X Y , Y ; Z 2 ) with α 4 0 (Lemma 8.2). This approach allows us to find four minimal tori in RP 3 with positive Ricci curvature and potentially five via White’s degree method. Indeed, this construction of 𝛼 is also valid in S 3 . Additionally, another challenge is to distinguish tori from 2-spheres in min-max outputs. In lens space, the positive Ricci assumption is sufficient to exclude all the minimal 2-spheres by the embedded Frankel property, which simplifies the problem considerably. In contrast, this advantage is absent in S 3 , where distinguishing between these surfaces remains challenging.

2 Preliminaries

In this section, we collect necessary preliminaries and notation in the equivariant setting.

Let ( M 3 , g M ) be an oriented connected closed Riemannian 3-dimensional manifold and let 𝐺 be a finite group acting freely and effectively as isometries on 𝑀. By [38], M 3 can be 𝐺-equivariantly isometrically embedded into some R L , namely there is an isometric embedding 𝑖 from 𝑀 into R L which is equivariant with respect to an orthogonal representation 𝜌 of 𝐺 on R L ( i g = ρ ( g ) i for all g G ). In addition, let G + be an index 2 subgroup of 𝐺 (i.e. [ G : G + ] = 2 ), and denote by G the coset of G + in 𝐺 (i.e. G = G + G ). Note that we only specify the ambient manifold M 3 and the free actions of 𝐺 to S 3 and Z p respectively in Sections 6, 7, 8.

To signify the 𝐺-invariance, we will add a superscript or a subscript “𝐺” to the usual terminologies. We also use “ G ± ” to emphasize G + -symmetry together with G -antisymmetry.

  • 𝜋: the projection π : M M / G given by p [ p ] : = { g p g G } ;

  • H m : the 𝑚-dimensional Hausdorff measure in R L ;

  • B r ( p ) : Euclidean open ball of radius r > 0 centered at 𝑝;

  • A s , r ( p ) : Euclidean open annulus B r ( p ) Clos ( B s ( p ) ) ;

  • B r G ( p ) : = B r ( G p ) = G B r ( p ) and A s , r G ( p ) : = A s , r ( G p ) = G A s , r ( p ) ;

  • h C G ± ( M ) : a smooth mean curvature prescribing function, where for k N { } ,

    (2.1) C G ± k ( M ) := { f C k ( M ) f ( p ) = f ( g p ) for all p M , g G + , f ( p ) = f ( g p ) for all p M , g G } ;

  • C ( M ) , C G ( M ) : the space of sets Ω M with finite perimeter (i.e. Caccioppoli sets, cf. [42, §40]), and the space of Ω C ( M ) with g Ω = Ω for all g G (i.e. 𝐺-invariant Caccioppoli sets);

  • C G ± ( M ) : the space of G + -invariant Caccioppoli sets Ω C ( M ) so that G permutes Ω and M Ω as two ( G + -invariant) Caccioppoli sets, i.e.

    C G ± ( M ) := { Ω C ( M ) g Ω = Ω for all g G + , g Ω = M Ω in C ( M ) for all g G } ;

  • V ( M ) , V G ( M ) : the space of 2-varifolds in M 3 , and the space of V V ( M ) with g # V = V for all g G (i.e. 𝐺-invariant 2-varifolds);

  • X ( M ) , X G ( M ) : the space of smooth vector fields in 𝑀, and the space of X X ( M ) with d g X = X for all g G ;

  • Diff 0 ( M ) : the connected component of the diffeomorphism group of 𝑀 containing the identity.

For any Ω C ( M ) , denote by Ω the (reduced) boundary of [ [ Ω ] ] as a 3-dimensional mod 2 flat chain, by | Ω | the induced integral varifold, and by ν Ω the outward pointing unit normal of Ω (see [42, §14]). In addition, denote by Ω and V the induced Radon measure in 𝑀 associated with Ω and V V ( M ) .

Remark 2.1

For any Ω C G ± ( M 3 ) , the 2-dimensional mod 2 flat chain Ω and the varifold | Ω | are both 𝐺-invariant. Besides, the outward unit normal ν Ω is G + -invariant so that d g ( ν Ω ) = ν Ω for any g G . As ℎ is G + -invariant with h g = h for any g G , we conclude that h ν Ω is 𝐺-invariant.

Let U M be an open set. Then we denote by I s ( U ) the set of isotopies of 𝑀 supported in 𝑈. Additionally, after replacing 𝑀 by 𝑈, one can similarly define the localized notions C ( U ) , V ( U ) , X ( U ) , and C G ( U ) , C G ± ( U ) , V G ( U ) , X G ( U ) provided that 𝑈 is 𝐺-invariant.

Note that a map F : M M is said to be 𝐺-equivariant if F ( g p ) = g F ( p ) for all p M , g G . We can now introduce the following notation:

  • Diff 0 G ( M ) : the connected component of the 𝐺-equivariant diffeomorphism group of 𝑀 containing the identity;

  • I s G ( U ) : the set of 𝐺-equivariant isotopies of 𝑀 supported in an open 𝐺-set 𝑈.

Furthermore, for any 𝐺-invariant subset A M with connected components { A i } i = 1 I , we say that 𝐴 is 𝐺-connected if, for any i , j { 1 , , I } , there is g i , j G with g i , j A j = A i . Finally, given a (𝐺-)connected open set U M , we say that a set of (𝐺-)connected C 1 -embedded (𝐺-)surfaces { Γ i } i = 1 l U with Γ i U = is ordered, denoted by Γ 1 Γ l , if, for each 𝑖, Γ i separates 𝑈 into two (𝐺-)connected components U ± i (i.e. U Γ i = U + i U i ) so that Γ 1 , , Γ i 1 Clos ( U i ) and Γ i + 1 , , Γ l Clos ( U + i ) .

Since the group actions in this paper are free and effective, we can make the following definition.

Definition 2.2

A 𝐺-invariant open set U M is said to be appropriately small if each connected component of 𝑈 is diffeomorphic to π ( U ) = U / G .

Namely, an appropriately small open 𝐺-set 𝑈 shall have # G connected components { U i } i = 1 # G with 𝐺 permuting them. For instance, any open 𝐺-set U B r G ( p ) is appropriately small provided r > 0 is smaller than the injectivity radius of G p .

2.1 A h -functional and V C G ± -space

For any pairs

( V , Ω ) V G ( M ) × C G ± ( M ) V ( M ) × C ( M )

and h C G ± ( M ) , define the prescribing mean curvature functional by

(2.2) A h ( V , Ω ) = V ( M ) Ω h d H 3 .

Note that F # ( V , Ω ) : = ( F # V , F ( Ω ) ) is still a pair in V G ( M ) × C G ± ( M ) for any 𝐺-equivariant diffeomorphism 𝐹 of 𝑀. Hence we have the following definitions generalizing [48, Definition 1.1].

Definition 2.3

A pair ( V , Ω ) V G ( M ) × C G ± ( M ) is said to be A h -stationary (resp. ( G , A h ) -stationary) in a (resp. 𝐺-invariant) open set U M if, for any X X ( U ) (resp. X X G ( U ) ) with generated diffeomorphisms { ϕ t } ,

(2.3) δ A V , Ω h ( X ) := d d t | t = 0 A h ( ϕ # t ( V , Ω ) ) = G 2 ( M ) div S X ( x ) d V ( x , S ) Ω X , ν Ω h d μ Ω = 0 .

In addition, ( V , Ω ) is said to be A h -stable (resp. ( G , A h ) -stable) in 𝑈 if

δ 2 A V , Ω h ( X , X ) : = d 2 d t 2 | t = 0 A h ( ϕ # t ( V , Ω ) ) 0

for any X X ( U ) (resp. X X G ( U ) ).

In the following lemma, we shall prove that ( G , A h ) -stationarity implies A h -stationarity. Note that the proof also works for general (non-free non-effective) 𝐺-actions.

Lemma 2.4

Given any 𝐺-invariant open set U M , a pair

( V , Ω ) V G ( M ) × C G ± ( M )

is ( G , A h ) -stationary in 𝑈 if and only if ( V , Ω ) is A h -stationary in 𝑈.

Proof

It is sufficient to show that, for any X X ( U ) , there exists X G X G ( U ) so that δ A V , Ω h ( X ) = δ A V , Ω h ( X G ) . Indeed, define X G : = ( g G d g 1 ( X ) ) / # G to be the averaged 𝐺-invariant vector field compactly supported in 𝑈. Since Ω and h ν Ω are both 𝐺-invariant (Remark 2.1), we have Ω X , h ν Ω d μ Ω = Ω X G , h ν Ω d μ Ω . In addition, note that the diffeomorphisms generated by 𝑋 and d g 1 ( X ) are { ϕ t } and { g 1 ϕ t g } respectively. Hence the 𝐺-invariance of 𝑉 implies ϕ # t V ( M ) = ( g 1 ϕ t g ) # V ( M ) and div S X d V = div S ( d g 1 ( X ) ) d V = div S X G d V . ∎

Remark 2.5

In particular, if the open 𝐺-set 𝑈 is appropriately small (Definition 2.2), then we also have the equivalence between the 𝐺-stability and stability in 𝑈. Indeed, since the stability of ( V , Ω ) in 𝑈 is equivalent to the stability in every component U i of 𝑈, i = 1 , , # G , it follows from (2.6) and the proof of Lemma 2.18 that

# G δ 2 A V , Ω h ( X , X ) = δ 2 A V , Ω h ( X G , X G ) ,

where X X ( U i ) and X G = g G d g 1 ( X ) X G ( U ) .

Motivated by Almgren’s VZ-space [2], we make the following definition for the weak topological completion of the diagonal pairs

{ ( | Ω | , Ω ) Ω C G ± ( M ) } V G ( M ) × C G ± ( M ) .

Definition 2.6

The V C -space V C ( M ) (resp. V C G ± -space V C G ± ( M ) ) is the space of all pairs ( V , Ω ) in V ( M ) × C ( M ) (resp. V G ( M ) × C G ± ( M ) ) so that V = lim k | Ω k | and Ω = lim k Ω k for some sequence { Ω k } k N in C ( M ) (resp. C G ± ( M ) ).

For any ( V , Ω ) , ( V , Ω ) V C ( M ) , define the ℱ-distance between them by

F ( ( V , Ω ) , ( V , Ω ) ) : = F ( V , V ) + F ( Ω , Ω ) ,

where 𝐅 is the varifolds 𝐅-metric [39, Section 2.1] and ℱ is the flat metric [42, §31].

Note that V G ( M ) and C G ± ( M ) are closed sub-spaces of V ( M ) and C ( M ) respectively. One verifies that V C G ± ( M ) is also a closed sub-space of V C ( M ) . Hence the ℱ-distance can be induced to V C G ± ( M ) .

Moreover, combining Lemma 2.4 with [48, Lemmas 1.4–1.7], we have the following results.

Lemma 2.7

Set c : = sup p M | h ( p ) | . Then, for any C > 0 , ( V , Ω ) V C G ± ( M ) , and 𝐺-invariant open set U M , the following statements are valid:

  1. spt ( Ω ) spt ( V ) and Ω V ;

  2. if ( V , Ω ) is ( G , A h ) -stationary in 𝑈, then 𝑉 has 𝑐-bounded first variation in 𝑈, i.e.

    | δ V ( X ) | : = | G 2 ( M ) div S X ( x ) d V ( S , x ) | c M | X | d V for all X X ( U ) ;

  3. A C : = { ( V , Ω ) V C G ± ( M ) V ( M ) C } is a compact metric space with respect to the ℱ-distance;

  4. A 0 C : = { ( V , Ω ) A C ( V , Ω ) is ( G , A h ) -stationary } is a compact subset of A C with respect to the ℱ-distance.

Proof

Since V C G ± ( M ) is a closed sub-space of V C ( M ) , (i) and (iii) follow directly from [48, Lemmas 1.4, 1.6]. Combining Lemma 2.4 and [48, Lemmas 1.5, 1.7], we conclude (ii) and (iv). ∎

2.2 C 1 , 1 almost embedded ( G ± , h ) -surfaces

In this subsection, we consider the A h -functional for C 1 , 1 (almost embedded) G ± -boundaries.

Definition 2.8

Let U M be an open subset. A C 1 , 1 immersed surface ϕ : Σ U with ϕ ( Σ ) U = is said to be a C 1 , 1 almost embedded surface in 𝑈 if, at any non-embedded point p ϕ ( Σ ) , there is a neighborhood W U of 𝑝 so that

  • Σ ϕ 1 ( W ) is a disjoint union of connected components i = 1 l Γ i ;

  • ϕ ( Γ i ) W is a C 1 , 1 embedding for each i = 1 , , l ;

  • for each 𝑖, any other component ϕ ( Γ j ) ( j i ) lies on one side of ϕ ( Γ i ) , i.e.

    ϕ ( Γ j ) ϕ ( Γ i ) or ϕ ( Γ i ) ϕ ( Γ j ) .

For simplicity, we will denote ϕ ( Σ ) and ϕ ( Γ i ) by Σ and Γ i respectively in appropriate context. The subset of non-embedded points in Σ, denoted by S ( Σ ) , is called the touching set, and R ( Σ ) : = Σ S ( Σ ) is the regular set.

Definition 2.9

Definition 2.9 ( C 1 , 1 G ± -boundary)

Let U M be a 𝐺-invariant open subset. Then a 𝐺-equivariant C 1 , 1 almost embedded surface ϕ : Σ U is said to be a C 1 , 1 (almost embedded) G ± -boundary in 𝑈 if Σ is oriented and

ϕ # ( [ [ Σ ] ] ) = Ω

as 2-currents with Z 2 -coefficients in 𝑈 for some Ω C G ± ( U ) .

Remark 2.10

Let ( Σ , Ω ) be a C 1 , 1 G ± -boundary in 𝑈. Then both of R ( Σ ) and S ( Σ ) are 𝐺-invariant. Additionally, it follows from [48, Lemma 1.11] that Σ (as an immersed surface) admits a unit normal ν Σ so that

  • ν Σ = ν Ω along spt ( Ω ) provided Ω or 𝑈;

  • if Σ decomposes into ordered 𝐺-connected sheets Γ 1 Γ l in a 𝐺-connected open set W U , then ν Σ must alternate orientations along { Γ i } i = 1 l .

For any C 1 , 1 G ± -boundary ( Σ , Ω ) (in 𝑀), define its A h -functional by

A h ( Σ , Ω ) : = H 2 ( Σ ) Ω h d H 3

It then follows from (2.3) and [48, Lemma 1.12] that, for any X X ( M ) ,

(2.4) δ A Σ , Ω h ( X ) = Σ div Σ X d H 2 Ω X , ν Ω h d μ Ω = Σ div Σ X X , ν Σ h d H 2 ,

where ν Σ is given in the above remark.

Definition 2.11

Definition 2.11 ( C 1 , 1 ( G ± , h ) -boundary)

A C 1 , 1 G ± -boundary ( Σ , Ω ) in a 𝐺-invariant open set 𝑈 is said to be A h -stationary (resp. ( G , A h ) -stationary) in 𝑈 if, for any X X ( U ) (resp. X X G ( U ) ), δ A Σ , Ω h ( X ) = 0 . In particular, we say that ( Σ , Ω ) is a C 1 , 1 (almost embedded) ( G ± , h ) -boundary in 𝑈 if it is ( G , A h ) -stationary in 𝑈.

Similar to Lemma 2.4, we have the following result.

Lemma 2.12

Given any 𝐺-invariant open set U M , a C 1 , 1 G ± -boundary ( Σ , Ω ) is ( G , A h ) -stationary in 𝑈 if and only if ( Σ , Ω ) is A h -stationary in 𝑈.

Proof

For any X X ( U ) , let X G : = ( g G d g 1 ( X ) ) / # G X G ( U ) . Since Σ is 𝐺-invariant and Ω C G ± ( M ) , we can combine (2.4) with the proof of Lemma 2.4 to conclude δ A Σ , Ω h ( X ) = δ A Σ , Ω h ( X G ) and get the desired result. ∎

By the above lemma, ( Σ , Ω ) is a C 1 , 1 ( G ± , h ) -boundary if and only if it is a C 1 , 1 G ± -boundary (cf. Definition 2.9) and a C 1 , 1 ℎ-boundary (in the sense of [48, Definition 1.13]). In particular, combining the first variation formula (2.4) with the standard elliptic regularity theory, we know that the regular set R ( Σ ) of a C 1 , 1 ( G ± , h ) -boundary ( Σ , Ω ) is smoothly embedded of prescribed mean curvature

H | R ( Σ ) = h | R ( Σ )

with respect to the unit normal ν Σ = ν Ω .

2.3 Strong A h -stationarity

In [48], Wang–Zhou introduced the strong A h -stationarity which plays an important role in the regularity theory of PMC min-max and their multiplicity one theorem.

Definition 2.13

Definition 2.13 (Strong A h -stationarity)

A C 1 , 1 ( G ± , h ) -boundary ( Σ , Ω ) is said to be strongly A h -stationary in an open set 𝑈 if the following holds.

For any p S ( Σ ) U , there exist a small neighborhood W U of 𝑝 and a decomposition i = 1 l Γ i of Σ W by l : = Θ 2 ( Σ , p ) 2 connected disks with a natural ordering Γ 1 Γ l . Let W 1 , W l be the bottom and top components of W Σ . Then, for i { 1 , l } and all X X ( W ) pointing into W i along Γ i ,

(2.5) δ A Γ i , W i h ( X ) 0 when W i Ω , δ A Γ i , W W i h ( X ) 0 when W i Ω = .

Remark 2.14

One can also say ( Σ , Ω ) is strongly ( G , A h ) -stationary in 𝑈 if Σ has a local decomposition by 𝐺-connected ordered sheets Γ 1 Γ l in a 𝐺-neighborhood W U of G p S ( Σ ) so that (2.5) is valid for all 𝐺-invariant X X G ( W ) pointing away from all other sheets along Γ i ( i { 1 , l } ). Nevertheless, since the strong A h -stationarity is a local property, a C 1 , 1 ( G ± , h ) -boundary ( Σ , Ω ) is strongly A h -stationary in 𝑈 if and only if it is strongly ( G , A h ) -stationary in 𝑈.

In particular, the results in [48, Section 1.3] remain valid for strongly A h -stationary (or strongly ( G ± , A h ) -stationary) C 1 , 1 ( G ± , h ) -boundaries. The mean curvature 𝐻 of Σ with respect to 𝜈 still satisfies

H ( p ) = { h ( p ) when p R ( Σ ) U , 0 for H 2 -a.e. p S ( Σ ) U .

Moreover, for the generalized mean curvature H l of Γ l defined with respect to ν Σ from Remark 2.10, the following holds: given that Ω does not contain the region above the top sheet Γ l , we know that H l = h | Γ l in a neighborhood where h > 0 and that H l h | Γ l in a neighborhood where h < 0 . We have a corresponding statement for Γ 1 by flipping the order.

2.4 𝐺-stability and compactness

Since we only consider 𝐺-equivariant deformations of G ± -boundaries, we will extend Wang–Zhou’s compactness theorem for stable ℎ-boundaries to an equivariant version.

Definition 2.15

Definition 2.15 (𝐺-stable C 1 , 1 ( G ± , h ) -boundary)

Let U M be a 𝐺-invariant open set, and let ( Σ , Ω ) be a C 1 , 1 ( G ± , h ) -boundary in 𝑈. Then ( Σ , Ω ) is said to be stable (resp. 𝐺-stable) in 𝑈 if, for any flow { ϕ t } generated by X X ( U ) (resp. X X G ( U ) ),

δ 2 A Σ , Ω h ( X ) : = d 2 d t 2 | t = 0 A h ( ϕ # t ( Σ , Ω ) ) 0 .

If in addition ( Σ , Ω ) is strongly A h -stationary, then this is equivalent to the following stability inequality for all X X ( U ) (resp. X X G ( U ) ):

Σ | X | 2 Ric M ( X , X ) | A Σ | 2 | X | 2 d H 2 Ω X , h X , ν Ω d H 2 ,

where denotes the normal part with respect to Σ, and A Σ denotes the second fundamental form of Σ (as an immersion).

Note that if the touching set S ( Σ ) is empty (or consider Σ instead of ϕ ( Σ ) ), the first eigenfunction φ 1 of the Jacobi operator L Σ h : = Σ ( Ric M ( ν Σ , ν Σ ) + | A Σ | 2 + ν Σ h ) is 𝐺-invariant since ν Σ h is 𝐺-invariant and 𝐺-actions are isometries. Hence the (intrinsic) 𝐺-stability is equivalent to the (intrinsic) stability provided S ( Σ ) = . However, S ( Σ ) in general even if ( Σ , Ω ) solves the isotopy minimizing problem [48, Theorem 1.25]. Nevertheless, as the group actions we considered here are free and effective, we still have the following lemma.

Lemma 2.16

Let ( Σ , Ω ) be a C 1 , 1 ( G ± , h ) -boundary in an appropriately small open 𝐺-subset 𝑈. Then ( Σ , Ω ) is stable in 𝑈 if and only if ( Σ , Ω ) is 𝐺-stable in 𝑈.

Proof

The “only if” part is direct. For the “if” part, we only need to show ( Σ , Ω ) is stable in each connected component { U i } i = 1 k , k : = # G , of 𝑈. Indeed, given X X ( U 1 ) (with X U j = 0 for j 1 ), we can define X G X G ( U ) by X G g U 1 : = d g ( X ) since 𝑈 is appropriately small and the 𝐺-action is free and effective. Then we notice h and ν Ω will keep or change the sign simultaneously under the push forward of g G + or g G . Therefore, δ 2 A Σ , Ω h ( X ) = δ 2 A Σ , Ω h ( d g ( X ) ) , and thus 0 δ 2 A Σ , Ω h ( X G ) = k δ 2 A Σ , Ω h ( X ) by the appropriate smallness of 𝑈. ∎

Combining Lemma 2.16 with [48, Proposition 1.24], we conclude the following compactness result.

Proposition 2.17

Suppose that U M is an appropriately small open 𝐺-subset and that h j , h C G ± 2 ( M ) (cf. (2.1)) satisfies lim j h j h C 2 = 0 . Let { ( Σ j , Ω j ) } j N be a sequence of 𝐺-stable C 1 , 1 ( G ± , h j ) -boundaries in 𝑈 so that H 2 ( Σ j ) Λ for some Λ > 0 . Then there is a stable C 1 , 1 ( G ± , h ) -boundary ( Σ , Ω ) in 𝑈 so that

  1. Σ j converges to Σ in 𝑈 as varifolds and also in the sense of C loc 1 , α for all α ( 0 , 1 ) ;

  2. Ω j converge to Ω as currents in C G ± ( U ) C ( U ) .

Moreover,
  1. if Σ is smooth, then Σ j converges to Σ in 𝑈 in the C loc 1 , 1 topology;

  2. if ( Σ j , Ω j ) is strongly A h j -stationary in 𝑈, then ( Σ , Ω ) is strongly A h -stationary in 𝑈.

Proof

By Lemma 2.16, we can apply the compactness result [48, Proposition 1.24] and the C 1 , 1 -regularity result [49, Theorem 1.1] to ( Σ j , Ω j ) and obtain C 1 , 1 ℎ-boundary ( Σ , Ω ) satisfying (1)–(2) and (i)–(ii). Since 𝐺 acts by isometries, we get the 𝐺-invariance of Σ and Ω C G ± ( U ) . ∎

2.5 𝐺-isotopy minimizing problem

Let

0 < r 0 < min { ρ 0 , inf p M inj ( G p ) }

be a sufficiently small constant, where ρ 0 = ρ 0 ( M , g M , sup | h | ) > 0 is given in [40, §14] and inj ( G p ) is the injectivity radius of the normal exponential map exp G p . Note that B r 0 G ( p ) is appropriately small for all p M and inf p M inj ( G p ) > 0 since 𝐺 acts freely by isometries.

For any open 𝐺-subset U B r 0 G ( p ) , suppose that R C G ± ( U ) has a smoothly embedded 𝐺-invariant boundary Σ : = R U in 𝑈. Then we say that a pair ( V , Ω ) V C ( U ) (resp. ( V , Ω ) V C G ± ( U ) ) is an isotopy minimizer (resp. 𝐺-isotopy minimizer) of ( Σ , R ) in 𝑈 if there exists a sequence { ϕ k } k N in I s ( U ) (resp. I s G ( U ) ) so that

  • lim k A h ( ϕ k ( Σ , R ) ) = inf { A h ( ϕ ( Σ , R ) ) ϕ I s ( U ) ( resp. ϕ I s G ( U ) ) } ;

  • lim k F ( ( V , Ω ) , ϕ k ( Σ , R ) ) = 0 ,

where we used the notation ϕ ( Σ , R ) : = ( ϕ ( 1 , Σ ) , ϕ ( 1 , R ) ) for simplicity.

Similar to Lemma 2.16, we also have the following equivalence between isotopy minimizers and 𝐺-isotopy minimizers in appropriately small open 𝐺-sets by the definitions of h C G ± ( U ) and C G ± ( U ) .

Lemma 2.18

Let U M be an appropriately small open 𝐺-subset and let ( Σ , R ) be given as above. Then a 𝐺-isotopy minimizer ( V , Ω ) V C G ± ( U ) of ( Σ , R ) in 𝑈 is also an isotopy minimizer of ( Σ , R ) in 𝑈.

Proof

Denote by { U i } i = 1 # G the connected components of 𝑈, and fix any component, e.g. U 1 . Then, for any ( V , Ω ) V C G ± ( U ) , we have

(2.6) A h ( V U i , Ω U i ) = { A h ( V U 1 , Ω U 1 ) if U i G + U 1 , A h ( V U 1 , Ω U 1 ) + U 1 h d H 3 if U i G U 1 .

Therefore, for any 𝐺-isotopy minimizing sequence { ϕ k ( Σ , R ) } k N ,

A h ( ϕ k ( Σ , R ) ) = i = 1 # G A h ( ϕ k ( Σ U i , R U i ) ) = ( # G ) A h ( ϕ k ( Σ U 1 , R U 1 ) ) + # G 2 U 1 h d H 3 .

Since any isotopy ϕ I s ( U 1 ) can be extended to a 𝐺-isotopy ϕ G I s G ( U ) by taking

ϕ G U i = g ϕ g 1

for U i = g U 1 , it follows from the above equality that ϕ k ( Σ U 1 , R U 1 ) is also an isotopy minimizing sequence of ( Σ , R ) in U 1 , which implies ( V U 1 , Ω U 1 ) is also an isotopy minimizer in U 1 . ∎

By the choice of r 0 , any open 𝐺-set U B r 0 G ( p ) is appropriately small. Hence we have the following regularity theorem for 𝐺-isotopy minimizers in an appropriately and sufficiently small open 𝐺-set.

Theorem 2.19

Let U B r 0 G ( p ) and ( Σ , R ) be as above. Suppose ( V , Ω ) C G ± ( U ) is a 𝐺-isotopy minimizer of ( Σ , R ) in 𝑈. Then ( V , Ω ) is a strongly A h -stationary and stable C 1 , 1 ( G ± , h ) -boundary in 𝑈.

Proof

This result follows directly from Lemma 2.18 and [48, Theorem 1.25]. ∎

3 Min-max theory for C 1 , 1 ( G ± , h ) -boundaries

In this section, we follow the approach in [48] to set up the relative 𝐺-equivariant min-max problem for the A h -functional and prove the main regularity results for ( G , A h ) -min-max pairs.

3.1 𝐺-equivariant min-max problem

Fix a 𝐺-connected closed surface Σ 0 of genus g 0 . A 𝐺-equivariant embedding ϕ : Σ 0 M is said to be G ± -separating if

M ϕ ( Σ 0 ) = Ω + Ω ,

where Ω + , Ω are two nonempty domains sharing a common boundary ϕ ( Σ 0 ) so that

G + Ω ± = Ω ± and G Ω ± = Ω .

For convenience, write Σ = ϕ ( Σ 0 ) with the orientation induced by the outer normal 𝜈 of Ω, where Ω is an arbitrary choice of { Ω + , Ω } . We then denote

E G ± : = { ( Σ , Ω ) Σ is a G -equivariant G ± -separating embedding of Σ 0 in M }

endowed with the oriented smooth topology in the usual sense.

Let 𝑋 be a finite-dimensional cubical complex, and let Z X be a subcomplex. Let Φ 0 : X E G ± be a continuous map. Denote by Π the set of all continuous maps Φ : X E G ± which are homotopic to Φ 0 relative to Φ 0 | Z : Z E G ± . We refer to such a Φ as an ( X , Z ) -sweepout, or simply a sweepout. Definitions 2.1–2.3 in [48] are directly transferred to our situation as follows.

Definition 3.1

Given ( X , Z ) and Φ 0 as above, we call Π the ( X , Z ) -homotopy class of Φ 0 . The ℎ-width of Π is defined by

L h = L h ( Π ) = inf Φ Π sup x X A h ( Φ ( x ) ) .

A sequence { Φ i } i N Π is called a minimizing sequence if

L h ( Φ i ) : = sup x X A h ( Φ i ( x ) ) L h when i .

A subsequence { Φ i j ( x j ) x j X } j N is called a min-max (sub)sequence if

A h ( Φ i j ( x j ) ) L h when j .

The critical set of a minimizing sequence { Φ i } is defined by

C ( { Φ i } ) = { ( V , Ω ) V C G ± ( M ) there exists a min-max subsequence { Φ i j ( x j ) } such that F ( Φ i j ( x j ) , ( V , Ω ) ) 0 as j } .

We have the following min-max theorem, and the proof will be given later.

Theorem 3.2

Theorem 3.2 (PMC min-max theorem)

With all notions as above, suppose

(3.1) L h ( Π ) > max { max x Z A h ( Φ 0 ( x ) ) , 0 } .

Then there exist a minimizing sequence { Φ i } Π and a strongly A h -stationary, C 1 , 1 ( G ± , h ) -boundary ( Σ , Ω ) lying in the critical set C ( { Φ i } ) such that A h ( Σ , Ω ) = L h ( Π ) .

3.2 Tightening

We follow the pull-tight process in [48, Section 2.2] with some alterations.

Theorem 3.3

Theorem 3.3 (Pull-tight)

Let Π be an ( X , Z ) -homotopy class generated by some continuous Φ 0 : X E G ± relative to Φ 0 | Z . For a minimizing sequence { Φ i } i N Π associated with A h , there exists another minimizing sequence { Φ i } i N Π such that C ( { Φ i } ) C ( { Φ i } ) and every element ( V , Ω ) C ( { Φ i } ) is either ( G , A h ) -stationary (and thus A h -stationary by Lemma 2.4) or belongs to B = Φ 0 ( Z ) E G ± .

Proof

Given C : = L h + sup M | h ( p ) | Vol ( M ) + 1 , let A C and A 0 C be defined as in Lemma 2.7. Set B : = Φ 0 ( Z ) and A 0 = A 0 C B . For any X X ( M ) , we choose

X G = ( Σ g G d g 1 ( X ) ) / # G ,

which satisfies (Lemma 2.4)

δ A V , Ω h ( X ) = δ A V , Ω h ( X G ) for every ( V , Ω ) V C G ± ( M ) .

Then, combining with the constructions in [48, Section 2.2, Steps 1, 2], we obtain a continuous map X G : A C X G ( M ) (under the C 1 -topology on X G ( M ) ) so that X G A 0 = 0 and X G ( A C A 0 ) is continuous under the smooth topology on X G ( M ) . Additionally, by [48, Section 2.2, Step 3], we also have two continuous functions T , L : ( 0 , ) ( 0 , ) with T ( x ) , L ( x ) 0 as x 0 so that, for each ( V , Ω ) A C with γ = F ( ( V , Ω ) , A 0 ) , the homotopy map H : I × A C A C defined by

( t , ( V , Ω ) ) ( V T ( γ ) t , Ω T ( γ ) t ) : = Φ V , Ω G ( T ( γ ) t ) # ( V , Ω )

is continuous in the ℱ-metric satisfying

  • H ( t , ( V , Ω ) ) = ( V , Ω ) if ( V , Ω ) A 0 C B ,

  • A h ( V 1 , Ω 1 ) A h ( V , Ω ) L ( γ ) ,

where { Φ V , Ω G ( t , ) } t 0 Diff 0 G ( M ) is the flow associated with X G ( V , Ω ) .

Without loss of generality, we may assume that Φ i ( x ) A C for all i N and x X . Let X i G ( x ) = X G ( Φ i ( x ) ) for x X , which is continuous under the C 1 -topology on X G ( M ) with X i G Z = 0 . For each i N , define H i : X I s G ( M ) by H i ( x ) = H ( , Φ i ( x ) ) . By smoothing out X i G to some X ̃ i G : X X G ( M ) , which is continuous under the smooth topology with X ̃ i G ( x ) = 0 for any x Z and X i G X ̃ i G C 1 1 / i , we define H ̃ i : X I s G ( M ) using X ̃ i G rather than X i G . Denoting Φ i ( x ) = H ̃ i ( 1 , Φ i ( x ) ) , we obtain Φ i Π and the same estimate [48, (2.9)], i.e.

A h ( Φ i ( x ) ) A h ( Φ i ( x ) ) L ( F ( Φ i ( x ) , A 0 ) ) + C i ,

where C > 0 is a universal constant.

Given a min-max sequence { Φ i j ( x j ) } , by the above estimate and the fact that { Φ i } is a minimizing sequence, we know that { Φ i j ( x j ) } is also a min-max sequence and

F ( Φ i j ( x j ) , A 0 ) 0 as j .

Hence

lim j F ( Φ i j ( x j ) , Φ i j ( x j ) ) = 0 and C ( { Φ i } ) C ( { Φ i } ) .

Since F ( Φ i j ( x j ) , A 0 ) 0 as j , each element in C ( { Φ i } ) is either ( G , A h ) -stationary or belongs to B = Φ 0 ( Z ) E G ± . ∎

3.3 Almost minimizing

We now adapt the almost minimizing property to the ( G , A h ) -functional using 𝐺-equivariant embedded surfaces.

Definition 3.4

Given ϵ , δ > 0 , a 𝐺-invariant open set U M , and ( Σ , Ω ) E G ± , we say that ( Σ , Ω ) is 𝐺-equivariantly ( A h , ϵ , δ ) -almost minimizing in 𝑈 if there does not exist any isotopy ψ I s G ( U ) such that

  • A h ( ψ ( t , Σ , Ω ) ) A h ( Σ , Ω ) + δ for all t [ 0 , 1 ] ;

  • A h ( ψ ( 1 , Σ , Ω ) ) A h ( Σ , Ω ) ϵ .

Definition 3.5

Definition 3.5 ( ( G , A h ) -almost minimizing pairs)

Given a 𝐺-invariant open subset U M , a pair ( V , Ω ) V C G ± ( M ) , and a sequence { ( Σ j , Ω j ) } j N E G ± , we say that ( V , Ω ) is ( G , A h ) -almost minimizing with respect to { ( Σ j , Ω j ) } in 𝑈 if there exist ϵ j 0 and δ j 0 such that

  • ( Σ j , Ω j ) ( V , Ω ) in the ℱ-metric as j ;

  • ( Σ j , Ω j ) is 𝐺-equivariantly ( A h , ϵ j , δ j ) -almost minimizing in 𝑈.

Sometimes, we also say that ( V , Ω ) is ( G , A h ) -almost minimizing in 𝑈 without referring to { ( Σ j , Ω j ) } .

Moreover, we say that ( V 0 , Ω 0 ) is ( G , A h ) -almost minimizing in small 𝐺-annuli with respect to { ( Σ j , Ω j ) } if, for any p M , there exists r am ( G p ) > 0 such that ( V 0 , Ω 0 ) is ( G , A h ) -almost minimizing with respect to { ( Σ j , Ω j ) } in every A s , r G ( p ) A 0 , r am ( G p ) G ( p ) .

The lemma below emerges as a consequence of being ( G , A h ) -almost minimizing. The proof imitates the argument in [48, Lemma 3.3], which is skipped here.

Lemma 3.6

Let ( V , Ω ) V C G ± ( M ) be ( G , A h ) -almost minimizing in a 𝐺-invariant open set U M ; then

  1. ( V , Ω ) is ( G , A h ) -stationary in 𝑈;

  2. ( V , Ω ) is ( G , A h ) -stable in 𝑈.

In particular, if 𝑈 is appropriately small, then ( V , Ω ) is A h -stationary and A h -stable in 𝑈 by Lemma 2.4 and Remark 2.5.

We also need the following 𝐺-equivariant notions generalized from [13, 48].

Definition 3.7

Given L N and p M , let

C G = { A s 1 , r 1 G ( p ) , , A s L , r L G ( p ) }

be a collection of 𝐺-annuli centered at G p . We say that C G is 𝐿-admissible if 2 r j + 1 < s j for all j = 1 , , L 1 , and B r 1 G ( p ) is appropriately small.

In addition, a pair ( V , Ω ) V C G ± ( M ) is said to be ( G , A h ) -almost minimizing in C G with respect to a sequence { ( Σ j , Ω j ) } E G ± if there exist ϵ j 0 and δ j 0 so that

  • ( Σ j , Ω j ) ( V , Ω ) in the ℱ-metric as j ;

  • for each 𝑗, ( Σ j , Ω j ) is 𝐺-equivariantly ( A h , ϵ j , δ j ) -almost minimizing in at least one 𝐺-annulus in C G .

Comparing with [48, Definition 3.9], our 𝐿-admissible collection C G of 𝐺-annuli requires r 1 is less than the injectivity radius of G p so that every An G C G is appropriately small.

Now, our goal is to find a ( G , A h ) -min-max-pair ( V , Ω ) which is ( G , A h ) -almost minimizing in small 𝐺-annuli. Consider the setup in Section 3.1 where the nontriviality condition (3.1) is met. If { Φ i } i N Π is a pull-tight minimizing sequence obtained by Theorem 3.3, then every ( V , Ω ) C ( { Φ i } ) is A h -stationary.

Theorem 3.8

Theorem 3.8 (Existence of ( G , A h ) -almost minimizing pairs)

Let Π be an ( X , Z ) -homotopy class generated by some continuous Φ 0 : X E G ± relative to Φ 0 | Z so that (3.1) holds. Then there exists a min-max subsequence { ( Σ j , Ω j ) = Φ i j ( x j ) } j N E G ± converging to an A h -stationary pair ( V 0 , Ω 0 ) C ( { Φ i } ) so that ( V 0 , Ω 0 ) is ( G , A h ) -almost minimizing in every 𝐿-admissible collection of 𝐺-annuli with respect to { ( Σ j , Ω j ) } , where L = L ( m ) is an integer depending only on m : = dim ( X ) .

In addition, up to a subsequence of { ( Σ j , Ω j ) } , ( V 0 , Ω 0 ) is ( G , A h ) -almost minimizing in small 𝐺-annuli with respect to { ( Σ j , Ω j ) } .

Proof

The proof is essentially the same as that of [13, Appendix] (for the area functional) and [48, Section 3.3] (for the A h -functional). Namely, if there is no such min-max sequence, then one can apply a combinatorial argument of Almgren–Pitts [39] to find several isotopies supported in many disjoint annuli so that the A h -functional of a certain sweepout will be pulled down to strictly below L h ( Π ) via these isotopies, which contradicts the definition of L h ( Π ) . Since the proof is combinatorial, the argument would also carry over with our 𝐺-equivariant objects. ∎

An immediate consequence of Lemma 3.6 and Theorem 3.8 is the following.

Corollary 3.9

The limit A h -stationary pair ( V 0 , Ω 0 ) C ( { Φ i } ) satisfies

(R) for every L ( m ) -admissible collection C G of G -annuli , ( V 0 , Ω 0 ) is A h -stable in at least one G -annulus in C G .

3.4 Regularity of min-max pairs: Part I

In this subsection, we will introduce the notions concerning ( G ± , A h ) -replacements, and show the regularity for the pairs

( V , Ω ) V C G ± ( M )

with a certain ( G ± , A h ) -replacement chain property.

Definition 3.10

Given an open 𝐺-subset U M and ( V , Ω ) V C G ± ( M ) , a pair ( V , Ω ) V C G ± ( M ) is said to be a ( G ± , A h ) -replacement of ( V , Ω ) in 𝑈 if

  1. ( V , Ω ) = ( V , Ω ) in M Clos ( U ) ;

  2. A h ( V , Ω ) = A h ( V , Ω ) ;

  3. ( V , Ω ) is a strongly A h -stationary and stable C 1 , 1 ( G ± , h ) -boundary in 𝑈.

Definition 3.11

As above, ( V , Ω ) is said to have (weak) good ( G ± , A h ) -replacement property in 𝑈 if, for any p U , there exists r G p > 0 such that ( V , Ω ) has an ( G ± , A h ) -replacement ( V , Ω ) in any open 𝐺-annulus An G A 0 , r G p G ( p ) .

Proposition 3.12

Proposition 3.12 (Classification of tangent cones)

Let ( V , Ω ) V C G ± ( M ) be A h -stationary in an open 𝐺-set 𝑈 and have (weak) good ( G ± , A h ) -replacement property in 𝑈. Then 𝑉 is integer rectifiable in 𝑈, and for any p spt ( V ) U , every tangent varifold of 𝑉 at 𝑝 is an integer multiple of a plane in T p M .

Proof

The proof can be taken almost verbatim from [12, Lemma 6.4] and [40, Lemma 20.2]. ∎

Definition 3.13

Definition 3.13 ( ( G ± , A h ) -replacement chain property)

Let ( V , Ω ) V C G ± ( M ) and let U M be an open 𝐺-set. Then ( V , Ω ) is said to have the ( G ± , A h ) -replacement chain property in 𝑈 if the following statement holds. For any sequence B 1 G , , B k G U of open 𝐺-subsets, there exist a sequence

( V , Ω ) = ( V 0 , Ω 0 ) , ( V 1 , Ω 1 ) , , ( V k , Ω k ) V C G ± ( M )

satisfying that

  1. ( V j , Ω j ) is an ( G ± , A h ) -replacement of ( V j 1 , Ω j 1 ) in B j G for j = 1 , , k ;

  2. ( V j , Ω j ) is A h -stationary and stable in 𝑈;

  3. for another sequence

    B 1 G , , B k G , B ̃ k + 1 G , , B ̃ G U

    of open 𝐺-subsets, the sequence of ( G ± , A h ) -replacements ( V ̃ j , Ω ̃ j ) can be chosen so that ( V ̃ j , Ω ̃ j ) = ( V j , Ω j ) for j = 1 , , k .

Note that the ( G ± , A h ) -replacement chain property of ( V , Ω ) V C G ± ( M ) implies that ( V , Ω ) is A h -stationary and stable in 𝑈, and has the (weak) good ( G ± , A h ) -replacement property in 𝑈, which further indicates the rectifiability of 𝑉 by Proposition 3.12. Additionally, if ( V , Ω ) is a ( G ± , A h ) -replacement of ( V , Ω ) in B G U , then ( V , Ω ) not only is an A h -replacement of ( V , Ω ) in B G U in the sense of [48, Definition 3.4] but also has certain symmetries, i.e. ( V , Ω ) V C G ± ( M ) V C ( M ) . Hence the above definitions concerning ( G ± , A h ) -replacements are stronger than [48, Definitions 3.4, 3.5, 3.6]. We then have the following regularity theorem by [48, Theorem 4.4].

Theorem 3.14

Theorem 3.14 (First regularity)

Let ( V , Ω ) V C G ± ( M ) satisfy the ( G ± , A h ) -replacement chain property in a given open 𝐺-set U M . Then ( V , Ω ) is induced by a strongly A h -stationary and stable C 1 , 1 ( G ± , h ) -boundary in 𝑈.

Proof

By the above definitions, ( V , Ω ) V C G ± ( M ) also has the replacement chain property in 𝑈 in the sense of [48, Definition 3.6]. Hence the desired regularity result follows from [48, Theorem 4.4] and the fact that ( V , Ω ) V C G ± ( M ) . ∎

3.5 Regularity of min-max pairs: Part II

In this subsection, we prove the regularity of the ( G , A h ) -almost minimizing pairs ( V , Ω ) V C G ± ( M ) in an appropriately small open 𝐺-set 𝑈 by constructing ( G ± , A h ) -replacement chains. Throughout this subsection, we always assume that

U M is an appropriately small (cf. Definition  2.2 ) open G -set

with connected components { U i } i = 1 # G .

To begin with, consider a constrained A h -minimizing problem. For any 𝐺-equivariant G ± -separating embedded surface ( Σ , Ω ) E G ± and δ > 0 , take

I s δ G , h ( U ) := { ψ I s G ( U ) A h ( ψ ( t , ( Σ , Ω ) ) ) A h ( Σ , Ω ) + δ } , I s δ h ( U i ) := { ψ I s ( U i ) A h ( ψ ( t , ( Σ , Ω ) ) ) A h ( Σ , Ω ) + δ } .

Then a sequence { ( Σ k , Ω k ) } k N E G ± is said to be minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) if there exists a sequence { ψ k } k N I s δ G , h ( U ) with ( Σ k , Ω k ) = ψ k ( 1 , ( Σ , Ω ) ) so that

A h ( Σ , Ω ) A h ( Σ k , Ω k ) m δ G : = inf { A h ( ψ ( 1 , ( Σ , Ω ) ) ) ψ I s δ G , h ( U ) } as k

Similarly, if { ψ ̂ k } k N I s δ h ( U i ) with ( Σ ̂ k , Ω ̂ k ) = ψ ̂ k ( 1 , ( Σ , Ω ) ) so that

A h ( Σ , Ω ) A h ( Σ ̂ k , Ω ̂ k ) m δ : = inf { A h ( ψ ̂ ( 1 , ( Σ , Ω ) ) ) ψ ̂ I s δ h ( U i ) } as k ,

then we say that { ( Σ ̂ k , Ω ̂ k ) } k N is minimizing in problem ( Σ , Ω , I s δ h ( U i ) ) .

Using the arguments in Lemma 2.18, we also have the following result indicating the equivalence between minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) and minimizing in problem ( Σ , Ω , I s δ / # G h ( U i ) ) .

Lemma 3.15

Given

i { 1 , , # G } , { ψ k } k N I s δ G , h ( U ) and ( Σ k , Ω k ) = ψ k ( 1 , ( Σ , Ω ) ) ,

define

ψ ̂ k : = { ψ k in U i , id in M U i , ( Σ ̂ k , Ω ̂ k ) : = ψ ̂ k ( 1 , ( Σ , Ω ) ) = { ( Σ k , Ω k ) in U i , ( Σ , Ω ) in M U i .

Then ( Σ k , Ω k ) is minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) if and only if ( Σ ̂ k , Ω ̂ k ) is minimizing in problem ( Σ , Ω , I s δ / # G h ( U i ) ) .

Proof

By (2.6) and the above definitions, we have

(3.2) A h ( ψ k ( t , ( Σ , Ω ) ) ) = # G A h ( ψ ̂ k ( t , ( Σ U i , Ω U i ) ) ) + # G 2 U i h + A h ( Σ U , Ω U ) = # G A h ( ψ ̂ k ( t , ( Σ , Ω ) ) ) + C 0 ,

where C 0 = C 0 ( U , Σ , Ω , h , # G ) is a constant (independent of ψ k ). In particular,

A h ( Σ , Ω ) = # G A h ( Σ U i , Ω U i ) + # G 2 U i h + A h ( Σ U , Ω U ) = # G A h ( Σ , Ω ) + C 0 .

Thus { ψ ̂ k } I s δ / # G h ( U i ) , and every ϕ ̂ I s δ / # G h ( U i ) can be recovered to ϕ I s δ G , h ( U ) by taking

ϕ U j = g ϕ ̂ g 1 for U j = g U i .

Moreover, the above formulae also imply that m δ G = # G m δ / # G + C 0 . Therefore,

A h ( Σ , Ω ) A h ( Σ k , Ω k ) m δ G

if and only if A h ( Σ , Ω ) A h ( Σ ̂ k , Ω ̂ k ) m δ / # G by the above equalities again. ∎

In the following lemma, we show that any isotopy in a small enough open 𝐺-set which does not increase A h ( Σ k , Ω k ) can also be replaced by an isotopy in I s δ G , h ( U ) .

Lemma 3.16

Let { ( Σ k , Ω k ) } k N be minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) . Given any 𝐺-subset U U , there exist ρ 0 > 0 , k 0 1 so that, for any k k 0 and B 2 ρ G ( x ) U with ρ < ρ 0 , if φ I s G ( B ρ G ( x ) ) with

A h ( φ ( 1 , ( Σ k , Ω k ) ) ) A h ( Σ k , Ω k ) ,

then there is Φ I s G ( B 2 ρ G ( x ) ) with

Φ ( 1 , ) = φ ( 1 , ) and A h ( Φ ( t , ( Σ k , Ω k ) ) ) A h ( Σ k , Ω k ) + δ for all t [ 0 , 1 ] .

Moreover, ρ 0 depends on H 2 ( Σ ) , h L , U , M , δ , but does not depend on { ( Σ k , Ω k ) } k N .

Proof

Let { ψ ̂ k } and { ( Σ ̂ k , Ω ̂ k ) } be given as in Lemma 3.15 with respect to some fixed 𝑖. Hence { ( Σ ̂ k , Ω ̂ k ) } is minimizing in problem ( Σ , Ω , I s δ / # G h ( U i ) ) so that [48, Lemma 4.5] is applicable in U i : = U U i , which gives us the desired ρ 0 > 0 and k 0 1 .

Indeed, take any φ I s G ( B ρ G ( x ) ) so that

A h ( φ ( 1 , ( Σ k , Ω k ) ) ) A h ( Σ k , Ω k ) ,

where k k 0 , ρ < ρ 0 , and x U i with B 2 ρ G ( x ) U . Let φ ̂ = φ in B ρ ( x ) = U i B ρ G ( x ) and φ ̂ = id outside B ρ ( x ) . Then φ ̂ I s ( B ρ ( x ) ) and A h ( φ ̂ ( 1 , ( Σ k , Ω k ) ) ) A h ( Σ k , Ω k ) by (3.2). Thus, by [48, Lemma 4.5], we have an isotopy Φ ̂ I s ( B 2 ρ ( x ) ) with Φ ̂ ( 1 , ) = φ ̂ ( 1 , ) and

A h ( Φ ̂ ( t , ( Σ k , Ω k ) ) ) A h ( Σ k , Ω k ) + δ / # G for all t [ 0 , 1 ] .

After recovering Φ ̂ to Φ I s G ( B 2 ρ G ( x ) ) by taking Φ U j = g Φ ̂ g 1 for U j = g U i , we can use (3.2) again to show that Φ is the desired isotopy. ∎

Combining Lemma 3.16, 2.18 and Proposition 2.17, we have the following regularity result for the minimizers in problem ( Σ , Ω , I s δ G , h ( U ) ) .

Proposition 3.17

Suppose ( Σ , Ω ) E G ± is 𝐺-equivariantly ( A h , ϵ , δ ) -almost minimizing in an appropriately small open 𝐺-set 𝑈. Let { ψ k } I s δ G , h ( U ) so that

{ ( Σ k , Ω k ) = ψ k ( 1 , ( Σ , Ω ) ) }

is minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) . Then, up to a subsequence, ( Σ k , Ω k ) converges to some ( V ̂ , Ω ̂ ) V C G ± ( M ) so that

  1. A h ( Σ , Ω ) ϵ A h ( V ̂ , Ω ̂ ) A h ( Σ , Ω ) ;

  2. ( V ̂ , Ω ̂ ) U is a strongly A h -stationary and stable C 1 , 1 ( G ± , h ) -boundary in 𝑈.

Proof

Statement (i) follows directly from Definition 3.4. Similar to the proof of Lemma 3.6 (cf. [48, Lemma 3.3]), one can show that ( V ̂ , Ω ̂ ) U is strongly ( G , A h ) -stationary and 𝐺-stable in 𝑈 because { ( Σ k , Ω k ) } is minimizing in problem ( Σ , Ω , I s δ G , h ( U ) ) . Then, since 𝑈 is appropriately small, we have ( V ̂ , Ω ̂ ) U is strongly A h -stationary and stable in 𝑈 by Lemma 2.4 and Remark 2.5

Next, for any p U , take r 0 ( 0 , min { r 1 , ρ 0 } ) , where r 1 = dist M ( p , U ) / 4 and ρ 0 is given by Lemma 3.15 for U = B r 1 G ( p ) . We claim that ( V ̂ , Ω ̂ ) has the ( G ± , A h ) -replacement chain property in B r 0 G ( p ) , which indicates the regularity of ( V ̂ , Ω ̂ ) U by Theorem 3.14 and the arbitrariness of p U . Indeed, since 𝑈 is appropriately small, the proof of [48, Proposition 4.6] would carry over with our 𝐺-equivariant objects by using Theorem 2.19, Proposition 2.17 and Lemma 3.16 in place of [48, Theorem 1.25, Proposition 1.24, Lemma 4.5] respectively. ∎

Now, we can show the regularity for ( G , A h ) -almost minimizing pairs.

Theorem 3.18

Theorem 3.18 (Regularity of ( G ± , A h ) -almost minimizing pairs)

Given any appropriately small open 𝐺-set U M , let ( V , Ω ) V C G ± ( M ) be ( G ± , A h ) -almost minimizing with respect to { ( Σ j , Ω j ) } j N E G ± in 𝑈. Then ( V , Ω ) U is induced by a strongly A h -stationary and stable C 1 , 1 ( G ± , h ) -boundary.

Proof

Since 𝑈 is appropriately small, it follows from Lemma 3.6 that ( V , Ω ) U is strongly A h -stationary and stable.

Fix any open 𝐺-subsets

B 1 G U U .

Note that ( Σ j , Ω j ) is 𝐺-equivariantly ( A h , ϵ j , δ j ) -almost minimizing in 𝑈 for some ϵ j 0 and δ j 0 as j . Hence it follows that, for each j N , we can take a minimizing sequence { ( Σ j , l 1 , Ω j , l 1 ) } l N for problem ( Σ j , Ω j , I s δ j G , h ( B 1 G ) ) , and apply Proposition 3.17 to see

lim l ( Σ j , l 1 , Ω j , l 1 ) = ( V j 1 , Ω j 1 ) V C G ± ( M )

(up to a subsequence) so that

  1. A h ( Σ j , Ω j ) ϵ j A h ( V j 1 , Ω j 1 ) A h ( Σ j , Ω j ) ;

  2. ( V j 1 , Ω j 1 ) = ( Σ j , Ω j ) in M Clos ( B 1 G ) ;

  3. ( V j 1 , Ω j 1 ) is a strongly A h -stationary stable C 1 , 1 ( G ± , h ) -boundary in B 1 G .

Up to a subsequence, ( V j 1 , Ω j 1 ) converges to ( V 1 , Ω 1 ) V C G ± ( M ) . Combining Proposition 2.17 with (1)–(3), we see ( V 1 , Ω 1 ) is a ( G ± , A h ) -replacement of ( V , Ω ) in B 1 G . Moreover, ( Σ j , l 1 , Ω j , l 1 ) is also 𝐺-equivariantly ( A h , ϵ j , δ j ) -almost minimizing in 𝑈 since

( Σ j , l 1 , Ω j , l 1 ) = ψ l 1 ( 1 , ( Σ j , Ω j ) ) for some ψ l 1 I s δ j G , h ( B 1 G )

and A h ( Σ j , l 1 , Ω j , l 1 ) A h ( Σ j , Ω j ) .

Next, for another open 𝐺-set B 2 G U , the above arguments can be applied in B 2 G to a diagonal sequence { ( Σ j , l ( j ) 1 , Ω j , l ( j ) 1 ) } j N that converges to ( V 1 , Ω 1 ) . Then we obtain a ( G ± , A h ) -replacement

( V 2 , Ω 2 ) = lim ( Σ j , l ( j ) 2 , Ω j , l ( j ) 2 )

of ( V 1 , Ω 1 ) in B 2 G so that each ( Σ j , l ( j ) 2 , Ω j , l ( j ) 2 ) is 𝐺-equivariantly ( A h , ϵ j , δ j ) -almost minimizing in 𝑈. Hence ( V , Ω ) has ( G ± , A h ) -replacement chain property in U by repeating this procedure, which indicates the regularity of ( V , Ω ) by Theorem 3.14. ∎

3.6 Proof of Theorem 3.2

By Theorem 3.8, there exists an A h -stationary pair

( V 0 , Ω 0 ) C ( { Φ i } )

that is ( G , A h ) -almost minimizing in small 𝐺-annuli with respect to a min-max subsequence { ( Σ j , Ω j ) = Φ i j ( x j ) } j N E G ± . Let { B r i G ( p i ) } i = 1 m be a finite set of appropriately small open 𝐺-sets covering 𝑀 with radius

r i = 1 2 min { r am ( G p i ) , inj ( G p i ) } ,

where r am ( G p i ) is given by Definition 3.5. After applying Theorem 3.18 to ( V 0 , Ω 0 ) in any open 𝐺-set U A 0 , r i G ( p i ) , we see ( V 0 , Ω 0 ) = ( Σ 0 , Ω 0 ) is a strongly A h -stationary C 1 , 1 ( G ± , h ) -boundary in M { G p 1 , , G p m } , also A h -stable in any U A 0 , r i G ( p i ) . Finally, using Propositions 2.17, 3.12 in place of [48, Propositions 1.24, 4.1], the arguments in [48, Section 4.4] can be taken almost verbatim to show the regularity of ( V 0 , Ω 0 ) extends across each G p i .

4 Compactness for min-max ( G ± , h ) -boundaries

4.1 Strong convergence and weighted genus bound

Given an h C G ± ( M ) and a sequence of positive numbers ϵ k 0 as k , we denote A ϵ k h simply by A k . With notation from Section 3.1, consider the equivariant min-max problem associated with Π for each A k , k N . By assuming that the nontriviality condition (3.1) is met for all 𝑘, we apply Theorem 3.8 to the A k -functional for each 𝑘. This yields a ( G , A k ) -min-max pair

( V k , Ω k ) V C G ± ( M )

and an associated min-max sequence { ( Σ k , j , Ω k , j ) } j N E G ± such that ( V k , Ω k ) is A k -stationary and ( G , A k ) -almost minimizing in small 𝐺-annuli with respect to { ( Σ k , j , Ω k , j ) } . By Theorem 3.2, ( V k , Ω k ) is a strongly A k -stationary, C 1 , 1 ( G ± , ϵ k h ) -boundary ( Σ k , Ω k ) with A k ( Σ k , Ω k ) = L ϵ k h ( Π ) .

In this part, our goal is to show the smooth regularity of a subsequential varifold limit V of { Σ k } and upgrade the convergence to C loc 1 , 1 . Moreover, for specially chosen ℎ, we prove the weighted genus bound (see (4.1)).

To begin with, by Corollary 3.9, we note that V satisfies

(R′) for every L ( m ) -admissible collection C G of G -annuli , V is stable (for area) in at least one G -annulus in C G .

Arguing similarly to the proof of [48, Proposition 5.1], we have the following proposition, which is crucial for the removable singularity step.

Proposition 4.1

There exists a subsequence of { ( Σ k , Ω k ) } k N such that

(S) given any p M , there exists r G p > 0 such that , for each A s , r G ( p ) with 0 < s < r < r G p , ( Σ k , Ω k ) is A k -stable in A s , r G ( p ) for all sufficiently large k .

Theorem 4.2

spt V is a closed embedded 𝐺-invariant minimal surface Σ . Moreover, there exists a finite set of points Y M such that, up to a subsequence, { Σ k } k N converges in C loc 1 , 1 to Σ in any compact subset of M Y .

Proof

The theorem is readily verified by combining Proposition 2.17, property (S), and the standard removable singularity theorem (see [41]). ∎

Although V may not be 𝐺-equivariantly A 0 -almost minimizing in small 𝐺-annuli, we may choose a special h C G ± ( M ) so that the limit minimal surface still has the total genus less than g 0 , the genus of elements in E G ± .

Theorem 4.3

Theorem 4.3 (Genus bound)

Let ( M , g M ) be a closed orientable 3-dimensional Riemannian manifold and let 𝐺 be a finite group acting freely and effectively as isometries on 𝑀 so that 𝐺 admits an index 2 subgroup G + with coset G = G G + . Consider V as above. Suppose that there are finitely many pairwise disjoint appropriately small open 𝐺-balls

B 1 G , , B α G M

such that

  1. π ( spt V B j G ) is an embedded disk for j = 1 , , α ;

  2. h 0 in a small neighborhood of spt V j B j G .

Assume that V = i = 1 N m i [ Γ i ] , where { Γ i } i = 1 N is a pairwise disjoint collection of connected, closed, embedded, minimal surfaces. Denote by I O { 1 , , N } (resp. I U ) the collection of 𝑖 such that Γ i is orientable (resp. non-orientable). Then we have

(4.1) i I O m i g ( Γ i ) + 1 2 i I U m i ( g ( Γ i ) 1 ) g 0 ,

where g 0 and g ( Γ i ) are the genus of Σ 0 and Γ i respectively.

Proof

Let { γ i } i = 1 k be a collection of simple closed curves contained in i = 1 N Γ i . By assumption (1), we can perturb { γ i } i = 1 k in the same isotopy class so that i γ i does not intersect j B j G . Hence, by assumption (2), ϵ k h 0 in a neighborhood of i γ i . Additionally, one notices that [48, Proposition 5.3] can be easily generalized to our 𝐺-equivariant setting. Hence, by possibly perturbing { γ i } and shrinking r G p > 0 , we can assume that ( Σ k , j ( k ) , Ω k , j ( k ) ) is 𝐺-equivariantly ( A 0 , ϵ k , δ k ) -almost minimizing in B r G p G ( p ) for any p i γ i . To prove the curve lifting lemma, we follow the strategy in [24, Proposition 2.2] (see also [14]). For simplicity, we only consider the setting of two appropriately small open 𝐺-balls B 1 G and B 2 G intersecting along a curve 𝛾. Suppose further that the 𝐺-equivariantly A 0 -almost minimizing property holds in B 1 G B 2 G . By taking successive ( G ± , A 0 ) -replacements of ( Σ k , j ( k ) , Ω k , j ( k ) ) on B 1 G and B 2 G , we obtain a new 𝐺-invariant surface W k (arising topologically from Σ k , j ( k ) after finitely many 𝐺-equivariant surgeries), which has the same limit as Σ k , j ( k ) . Combining Schoen’s estimates, a no-folding property, and an integrated Gauss–Bonnet argument, we show the graphical smooth convergence of W k away from finitely many points. Hence we can lift 𝛾 with the correct multiplicity by perturbing it slightly to avoid those singularities. ∎

4.2 Existence of supersolution

For a sequence of strongly A ϵ k h -stationary C 1 , 1 ϵ k h -boundaries { ( Σ k , Ω k ) } converging as varifolds to a closed 2-sided minimal surface Σ, Wang–Zhou [48] proved in the non-equivariant setting that Σ admits a nonnegative weak supersolution to a variant of the Jacobi equation provided that h Σ changes sign and the convergence is C loc 1 , 1 away from a finite set 𝒴 with multiplicity m 2 . Noting 𝐺 is finite and h C G ± ( M ) must change sign unless h 0 , we immediately have the following generalization.

Proposition 4.4

Let { ( Σ k , Ω k ) } k N be a sequence of strongly A ϵ k h -stationary C 1 , 1 ( G ± , ϵ k h ) -boundaries with lim k ϵ k = 0 so that Σ k converges as varifolds to a closed embedded 2-sided minimal 𝐺-surface Σ with multiplicity m 2 . Suppose the convergence is also C loc 1 , 1 away from a finite 𝐺-set 𝒴. Then Σ admits a nonnegative 𝐺-invariant function φ W 1 , 2 ( Σ ) satisfying φ L 2 ( Σ ) = 1 and

(4.2) Σ φ , f ( Ric ( ν , ν ) + | A Σ | 2 ) φ f d H 2 Σ 2 c h f d H 2 for all f C 1 ( Σ ) and f 0 ,

where c 0 is a constant so that c = 0 if m 3 is odd.

Proof

As we explained before, [48, Proposition 6.4] can be applied directly to get a nonnegative function φ W 1 , 2 ( Σ ) satisfying (4.2). To show 𝜑 is 𝐺-invariant, we recall the constructions in [48, Section 6]. Take any open 𝐺-set U Σ Y and a unit normal 𝜈 of Σ. Then, for sufficiently large 𝑘, Σ k admits a decomposition by ordered 𝑚-sheets Γ k 1 Γ k m inside a thickened 𝐺-neighborhood U δ = U × ( δ , δ ) , and each sheet Γ k ι ( 1 ι m ) is a normal graph of some function u k ι C 1 , 1 ( U ) such that u k 1 u k m , and u k ι 0 in C 1 , 1 ( U ) as k . Although a single sheet Γ k i may not be 𝐺-invariant, we notice Γ k m Γ k 1 is 𝐺-invariant, and thus | φ k | = φ k : = u k m u k 1 is 𝐺-invariant. Since φ U is the C 1 , α limit of φ k / φ k L 2 ( U ) , we know that 𝜑 is 𝐺-invariant. ∎

5 Multiplicity one for generalized Simon–Smith min-max theory

In this section, we will generalize two multiplicity one theorems in [48] to the equivariant setting, namely the relative min-max in the space of oriented 𝐺-equivariant G ± -separating surfaces in Section 5.1 and the classical min-max in the space of unoriented 𝐺-equivariant G ± -separating surfaces in Section 5.2.

5.1 Multiplicity one for relative equivariant min-max

To start with, we have the following compactness theorem, which admits a proof similar to [48].

Theorem 5.1

Let L N and let C > 0 be a constant. Consider { Σ k } a sequence of closed embedded 𝐺-invariant minimal surfaces satisfying sup k H 2 ( Σ k ) C and property (R′). Then Σ k converges subsequentially as varifolds to a closed embedded 𝐺-invariant minimal surface Σ possibly with integer multiplicities. Moreover, Σ is degenerate if Σ k Σ for infinitely many 𝑘.

Recall that the space E G ± of 𝐺-equivariant G ± -separating surfaces of genus g 0 is defined in Section 3. By choosing an appropriate prescribing function h C G ± ( M ) , we obtain the first multiplicity one type result as follows.

Theorem 5.2

Let ( M , g M ) be a closed connected orientable 3-dimensional Riemannian manifold, and let 𝐺 be a finite group acting freely and effectively as isometries on 𝑀 so that 𝐺 admits an index 2 subgroup G + with coset G = G G + . Suppose that 𝑋 is a finite-dimensional cubical complex with Z X a subcomplex. Consider Φ 0 : X E G ± a continuous map and Π the ( X , Z ) -homotopy class of Φ 0 . Assume that

L ( Π ) > max x Z H 2 ( Φ 0 ( x ) ) .

Then there exists a closed embedded 𝐺-invariant minimal surface Γ with connected components { Γ j } j = 1 J and integer multiplicities { m j } j = 1 J so that

  1. if Γ j is unstable and 2-sided, then m j = 1 ;

  2. if Γ j is 1-sided, then its connected double cover is stable.

Moreover, the weighted total genus of Γ (4.1) is bounded by g 0 .

Proof

It is sufficient to verify the theorem when g M is 𝐺-bumpy, i.e. any finite cover of a closed embedded 𝐺-invariant minimal hypersurface in ( M , g M ) is non-degenerate. Given a constant C > 0 (e.g. C : = L ( Π ) + 1 ), set M ( C ) the space of all closed embedded 𝐺-invariant minimal surfaces Γ satisfying H 2 ( Γ ) C and property (R′). By Theorem 5.1, it follows that M ( C ) = { S 1 , , S α } is a finite set. Take p 1 , , p α in 𝑀 so that p i S j if and only if j = i . Let r > 0 be small enough such that

  • B r G ( p 1 ) , , B r G ( p α ) are pairwise disjoint appropriately small open 𝐺-balls;

  • B r G ( p i ) intersects S j if and only if j = i ;

  • π ( B r G ( p i ) S i ) is an embedded disk for all i = 1 , , α .

Now, choose h C G ± ( M ) with h ( M ) [ 1 , 1 ] satisfying that, for all i = 1 , , α ,
  1. h = 0 outside i B r G ( p i ) ;

  2. h > 0 in some component of B r / 2 G ( p i ) (then h < 0 in another component of B r / 2 G ( p i ) );

  3. if S i is 2-sided, then S i h ϕ i d H 2 = 0 , where ϕ i is the first eigenfunction of the Jacobi operator on S i ;

  4. if S i is 1-sided, then S ̃ i h ϕ i d H 2 = 0 , where ϕ i is the first eigenfunction of the Jacobi operator on S ̃ i and S ̃ i is the connected double cover of S i .

Choose ϵ k 0 so that the nontriviality condition (3.1) with ℎ replaced by ϵ k h is met for sufficiently large 𝑘. By combining Theorem 3.2, Theorem 4.2 and arguing similarly to [48], we obtain a closed embedded 𝐺-invariant minimal surface Γ = j = 1 J Γ j , where each Γ j is 𝐺-invariant and belongs to M ( C ) . Assuming Γ j = S i j for i j { 1 , , α } , the sign of ℎ must change on each Γ j . If Γ j is 2-sided with m j 2 , it follows from Proposition 4.4 (nonnegativity and nontriviality of 𝜑), the positivity of ϕ i j and our choice of ℎ that the first eigenvalue satisfies λ 1 ( Γ j ) 0 . Hence any connected component Γ j is stable if Γ j is 2-sided with m j 2 , which shows the first item. The same argument applied to the double cover of 1-sided 𝐺-invariant Γ Γ proves the second item.

By the choice of ℎ, we have h = 0 outside B r G ( p i ) . Since π ( B r G ( p i ) Γ j ) is a disk, it follows from Theorem 4.3 that Γ admits the genus bound. ∎

5.2 Multiplicity one for classical equivariant min-max

In this part, we show an equivariant version of a multiplicity one type result in analogy to [48, Theorem 7.3]. This relies on a version of Simon–Smith min-max theory for unoriented 𝐺-equivariant G ± -separating surfaces.

Consider ( M , g M ) and 𝐺 as in Theorem 5.2. Fix Σ 0 a 𝐺-connected closed surface of genus g 0 . We equip

X G ± ( Σ 0 ) : = { ϕ ( Σ 0 ) ϕ : Σ 0 M is a G -equivariant G ± -separating embedding }

with the unoriented smooth topology. Let 𝑋 be a finite-dimensional cubical complex. Given a fixed continuous map Φ 0 : X X G ± , we denote by Π the collection of all continuous Φ : X X G ± which is homotopic to Φ 0 . Such a Φ is called a sweepout by Σ 0 , or simply a sweepout. Define

L ( Π ) : = inf Φ Π sup x X H 2 ( Φ ( x ) ) .

In our setting, the space E G ± defined in Section 3 double covers X G ± . Let λ ̄ H 1 ( X G ± ; Z 2 ) be the generator dual to the nontrivial element of π 1 ( X G ± ) corresponding to the projection π : E G ± X G ± . Note that a loop 𝜙 in X G ± forms a sweepout in the sense of Almgren–Pitts if and only if λ ̄ [ ϕ ] 0 .

Theorem 5.3

Consider the above setup and let Π be a homotopy class of sweepouts by Σ 0 with L ( Π ) > 0 . Then there exists a closed embedded 𝐺-invariant minimal surface Γ with 𝐺-connected components { Γ j } j = 1 J and integer multiplicities { m j } j = 1 J so that

L ( Π ) = H 2 ( Γ ) = j = 1 J m j H 2 ( Γ j )

and

  1. if Γ j is unstable and 2-sided, then m j = 1 ;

  2. if Γ j is 1-sided, then its connected double cover is stable.

Moreover, the genus bound (4.1) holds for Γ with g 0 = g ( Σ 0 ) .

Proof

We follow the proof of [48, Theorem 7.3] (also see [54, Theorem 5.2]) with some alterations. Similar to before, we only need to check the theorem for 𝐺-bumpy metric g M . The existence of a closed embedded 𝐺-invariant minimal surface Σ satisfying H 2 ( Σ ) = L ( Π ) and property (R′) is guaranteed by Theorem 3.3, Theorem 3.8 and Theorem 3.2.

Let 𝒮 be the collection of all stationary 2-varifolds with mass lying in

[ L ( Π ) 1 , L ( Π ) + 1 ] ,

whose support is a closed embedded 𝐺-invariant minimal surface satisfying (R′). Note that 𝒮 is a finite set by bumpiness. Given a small ϵ ̄ > 0 , set

Z i = { x X F ( Φ i ( x ) , S ) ϵ ̄ } and Y i = X Z i ̄ .

Since each Y i is topologically trivial, by adapting a continuous version of Pitts’ combinatorial argument to { Φ i } , we can find another minimizing sequence, still denoted by { Φ i } , such that L ( { Φ i | Z i } ) < L ( Π ) for sufficiently large 𝑖. Lifting the maps Φ i : Y i X G ± to their double covers Φ ̃ i : Y i E G ± , we have, for 𝑖 large enough,

sup x Y i H 2 ( Φ ̃ i ( x ) ) sup x Z i H 2 ( Φ ̃ i ( x ) ) < L ( Π ) .

Denote Π ̃ i the ( Y i , Y i ) -homotopy class associated with Φ ̃ i | Y i in E G ± . By employing a contradiction argument as in [54, Lemma 5.8], we obtain

L ( Π ̃ i ) L ( Π ) > sup x Y i H 2 ( Φ ̃ i ( x ) ) .

Then the proof is completed by applying Theorem 5.2 to Π ̃ i and letting i . ∎

Proposition 5.4

In the above theorem, there exists a subset I { 1 , , J } so that Γ = j I Γ j is a G ± -separating 𝐺-surface.

Proof

Extracting a diagonal subsequence, one easily obtains a sequence Ω k C G ± ( M ) with Ω k X G ± so that | Ω k | V = j = 1 J m j | Γ j | V G ( M ) as k in the varifolds sense, where V is the min-max varifold. It also follows from the compactness theorem that Ω k converges (up to a subsequence) to some Ω C G ± ( M ) . Hence Ω V and Ω is a 𝐺-invariant integral current supported in j = 1 J Γ j . As an elementary fact, Ω is also a 𝐺-invariant integral 𝑛-cycle in j = 1 J Γ j (cf. [53, Appendix 8]), which implies Ω = j = 1 J k j [ [ Γ j ] ] for some k j { 0 , 1 } by the constancy theorem [42, Theorem 26.27]. Since Vol ( Ω ) = Vol ( M ) / 2 , we see that Γ : = { Γ j k j = 1 } is the smooth boundary of Ω C G ± ( M ) , which is G ± -separating. ∎

6 Minimal RP 2 in RP 3

In this section, let M : = S 3 R 4 be the unit 3-sphere, and G : = Z 2 acts on 𝑀 by the identity map G + = { [ 0 ] } and the antipodal map G = { [ 1 ] } . Consider the space

X G ± : = { ϕ ( S 2 ) ϕ : S 2 S 3 a G -equivariant G ± -separating smooth embedding }

endowed with the unoriented smooth topology, where G = Z 2 is given as above.

6.1 Sweepouts formed by real projective planes

We now describe three classes of sweepouts that detect three nontrivial cohomology classes in H k ( X G ± ; Z 2 ) , k { 1 , 2 , 3 } .

To begin with, use ( a 1 , a 2 , a 3 , a 4 ) and [ a 1 , a 2 , a 3 , a 4 ] with i = 1 4 a i 2 = 1 to denote the points in S 3 and R P 3 respectively. Define then

G ̃ ( ( a 1 , a 2 , a 3 , a 4 ) ) := ( { a 1 x 1 + a 2 x 2 + a 3 x 3 + a 4 x 4 < 0 } S 3 ) , G ( [ a 1 , a 2 , a 3 , a 4 ] ) := { a 1 x 1 + a 2 x 2 + a 3 x 3 + a 4 x 4 = 0 } S 3 ,

where x 1 , , x 4 are the coordinate functions in R 4 . Note that G ̃ ( S 3 ) S 3 is the space of oriented great spheres and is a double cover of G ( R P 3 ) R P 3 , the space of unoriented great spheres. Since G ( R P 3 ) X G ± , we can define our three maps

Φ 1 : R P 1 X G ± , [ a 1 , a 2 ] G ( [ a 1 , a 2 , 0 , 0 ] ) , Φ 2 : R P 2 X G ± , [ a 1 , a 2 , a 3 ] G ( [ a 1 , a 2 , a 3 , 0 ] ) , Φ 3 : R P 3 X G ± , [ a 1 , a 2 , a 3 , a 4 ] G ( [ a 1 , a 2 , a 3 , a 4 ] ) .

One notices that, for any k { 1 , 2 , 3 } , Φ k is a 𝑘-sweepout of M = S 3 in the sense of Almgren–Pitts. Indeed, consider the closed curve

γ : S 1 = [ 0 , 2 π ] / { 0 2 π } dmn ( Φ k ) = R P k

given by γ ( e i θ ) = [ cos ( θ / 2 ) , sin ( θ / 2 ) , 0 , 0 ] , which is a generator of π 1 ( R P k ) . Then the curve ϕ k : = Φ k γ in G ( R P 3 ) can be lifted to the curve

ϕ ̃ k : θ [ 0 , 2 π ] G ̃ ( ( cos ( θ / 2 ) , sin ( θ / 2 ) , 0 , 0 ) )

in the double cover G ̃ ( S 3 ) satisfying that ϕ ̃ k ( 0 ) is ϕ ̃ k ( 2 π ) with the opposite orientation. Hence ϕ k is a sweepout in the sense of Almgren–Pitts. Combining with the fact that the generator λ H 1 ( R P k ; Z 2 ) satisfies λ ( γ ) = 1 and λ k 0 , we conclude Φ k is a 𝑘-sweepout in the sense of Almgren–Pitts (cf. [34, Definition 4.1]).

Next, denote by ι : X G ± Z 2 ( S 3 ; Z 2 ) the natural inclusion map into the space of mod-2 integral 2-cycles, and by λ ̄ the generator of H 1 ( Z 2 ( S 3 ; Z 2 ) ; Z 2 ) . Then the above result indicates ( ι Φ k ) ( λ ̄ k ) 0 H k ( R P k ; Z 2 ) (cf. [34, Definition 4.1]). In particular, we have

α : = ι ( λ ̄ ) H 1 ( X G ± ; Z 2 )

satisfies α k 0 H k ( X G ± ; Z 2 ) for each k { 1 , 2 , 3 } .

Finally, define P k , k { 1 , 2 , 3 } , to be the collection of continuous maps Φ from any cubical complex 𝑋 to X G ± that detects α k H k ( X G ± ; Z 2 ) , i.e.

P k : = { Φ : X X G ± Φ ( α k ) 0 H k ( X ; Z 2 ) } .

Clearly, Φ k P k for all k = 1 , 2 , 3 . Recall that

L ( P k ) : = inf Φ P k sup x dmn ( Φ ) H 2 ( Φ ( x ) ) .

6.2 Proof of Theorem 1.2

We have the following direct corollary by applying Theorem 5.3 and Proposition 5.4 to any homotopy class in P k .

Corollary 6.1

Suppose Π P k , k { 1 , 2 , 3 } , is a homotopy class of sweepouts in X G ± . Then L ( Π ) > 0 and the 𝐺-connected min-max minimal 𝐺-surfaces { Γ j } j = 1 J associated to Π given by Theorem 5.3 satisfy that

  1. there is one 𝐺-component, Γ 1 , that is a G ± -separating minimal 2-sphere, i.e. Γ 1 X G ± ;

  2. every other 𝐺-component Γ j , j 2 , is a 𝐺-invariant disjoint union of two minimal 2-spheres.

Proof

Firstly, L ( Π ) > 0 since Π P k are 𝑘-sweepouts in the sense of Almgren–Pitts. By Theorem 5.3, { Γ j } j = 1 J satisfies the genus bound (4.1), which implies each Γ j is either a G ± -separating minimal sphere in X G ± or a 𝐺-invariant disjoint union of two minimal spheres. Combining Proposition 5.4 with the fact that every two elements in X G ± have nonempty intersections, there is exactly one element of { Γ j } j = 1 J that is in X G ± . ∎

Fix any Riemannian metric g RP 3 on RP 3 . By [6, Proposition 2.3], there exists an embedded area minimizing RP 2 denoted as Σ 0 , i.e.

H 2 ( Σ 0 ) = inf { H 2 ( Σ ) Σ is an embedded RP 2 RP 3 } .

Let M = S 3 with the pull-back metric g M , G = Z 2 , and X G ± be defined as above with respect to ( RP 3 , g RP 3 ) . Then Σ is an embedded minimal projective plane in ( RP 3 , g RP 3 ) if and only if π 1 ( Σ ) X G ± is a minimal 2-sphere in ( M , g M ) .

Theorem 6.2

Using the above notation, suppose X G ± contains only finitely many elements that are minimal in ( M = S 3 , g M ) . Then

0 < 2 M 0 < L ( P 1 ) < L ( P 2 ) < L ( P 3 ) ,

where M 0 : = H 2 ( Γ 0 ) .

Proof

The fact that 0 < L ( P k ) < L ( P k + 1 ) for k = 1 , 2 follows from the standard Lusternik–Schnirelmann theory (see [48, Lemma 8.3] or [34, Theorem 6.1]). Indeed, one can take 𝒮 to be the collection of 𝐺-invariant integral varifolds with mass bounded by L ( P k + 1 ) whose support is an element in X G ± . Then, using α H 1 ( X G ± ; Z 2 ) and X G ± in place of λ ̄ H 1 ( Z 2 ( S 3 ; Z 2 ) ; Z 2 ) and Z 2 ( S 3 ; Z 2 ) , the proof of [34, Theorem 6.1] would carry over, leading to a contradiction to Corollary 6.1 (i). Our goal is now to prove 2 M 0 < L ( P 1 ) .

Suppose by contradiction that 2 M 0 = L ( P 1 ) . By the finiteness assumption, the union

S : = { Γ X G ± H 2 ( Γ ) = 2 M 0 }

is a closed set. Fix a Z 2 -invariant open ball B S 3 B r ( S ) . We have the following claim.

Claim 1

For any δ > 0 , there is Φ δ P 1 so that sup x dmn ( Φ δ ) H 2 ( Φ δ ( x ) B ) < δ .

Proof of Claim 1

Suppose by contradiction that, for some δ 0 > 0 , every Φ P 1 has x dmn ( Φ ) with H 2 ( Φ ( x ) B ) δ 0 . Let { Φ i } be a minimizing sequence in P 1 , i.e.

lim i sup x dmn ( Φ i ) H 2 ( Φ i ( x ) ) = L ( P 1 ) = 2 M 0 .

For each 𝑖, we may pick some x i dmn ( Φ i ) so that H 2 ( Φ i ( x i ) B ) δ 0 by assumptions. Based on the fact that

2 M 0 H 2 ( Φ i ( x i ) ) sup x dmn ( Φ i ) H 2 ( Φ i ( x ) ) 2 M 0 as i ,

we have Φ i ( x i ) / Z 2 RP 3 converges (up to a subsequence) as varifolds to an area minimizing projective plane (cf. [6, Proposition 2.3]), and thus the varifolds limit V of Φ i ( x i ) satisfies spt V S . This forces

lim i H 2 ( Φ i ( x i ) B ) = 0 ,

which contradicts the choice of x i . ∎

Let B r ( q ) B be a small ball with 0 < r < r 0 and δ ( 0 , α 0 r 2 ) , where α 0 , r 0 > 0 are given by [34, Proposition 8.2] with respect to Z 2 ( S 3 ; Z 2 ) . Consider Φ δ in Claim 1 with respect to this given 𝛿. Then, by definitions, ι Φ δ : dmn ( Φ δ ) Z 2 ( S 3 ; Z 2 ) is a 1-sweepout in the sense of Almgren–Pitts. However, since M ( ι Φ δ ( x ) B r ( q ) ) M ( ι Φ δ ( x ) B ) < δ < α 0 r 2 for all x dmn ( Φ δ ) , we obtain a contradiction to [34, Proposition 8.2]. ∎

At the end of this section, we prove Theorem 1.2.

Proof of Theorem 1.2

Firstly, we have a minimizing RP 2 embedded in ( RP 3 , g RP 3 ) denoted as Σ 0 (cf. [6, Proposition 2.3]).

Case I: g RP 3 is a metric with positive Ricci curvature. In this case, g M is a Z 2 -invariant metric on M = S 3 with positive Ricci curvature. Hence

(6.1) every Z 2 -connected minimal 2 -sphere in ( M , g M ) is ( Z 2 ) ± -separating ,

i.e. there is no minimal 2-sphere in ( RP 3 , g RP 3 ) . Otherwise, there will be a pair of minimal S 2 M , denoted as Γ = { Γ + , Γ } , such that [ 1 ] Γ ± = Γ and Γ + Γ = , where [ 1 ] Z 2 acting on M = S 3 by the antipodal map. This violates the embedded Frankel property [15].

Next, without loss of generality, we assume

(6.2) ( M = S 3 , g M ) contains finitely many minimal 2 -spheres in X Z 2 , ± ,

whose quotients in RP 3 are minimal projective planes. By Theorem 6.2 and (6.2), we have

(6.3) 0 < 2 M 0 < L ( P 1 ) < L ( P 2 ) < L ( P 3 ) .

Additionally, although each P k , k { 1 , 2 , 3 } , may contain many different homotopy classes, it follows from the finiteness assumptions (6.2) and Corollary 6.1 that the min-max values of the homotopy classes in P k must be stabilized to L ( P k ) . Moreover, for each k { 1 , 2 , 3 } , L ( P k ) is realized by the area of an embedded connected minimal 2-sphere Γ k X G ± with integer multiplicity m k Z + , i.e. L ( P k ) = m k H 2 ( Γ k ) .

Since manifolds with positive Ricci curvature contain no 2-sided stable minimal hypersurface, we see Γ i must be unstable, and thus m i = 1 by Theorem 5.3. We can then conclude that Σ 0 , π ( Γ 1 ) , π ( Γ 2 ) , π ( Γ 3 ) are distinct embedded minimal real projective planes in ( RP 3 , g RP 3 ) .

Case II: g RP 3 is a bumpy metric. In this case, every embedded Z 2 -invariant minimal surface in ( M = S 3 , g M ) is non-degenerate. Without loss of generality, we also assume that

(6.4) ( M = S 3 , g M ) contains finitely many Z 2 -invariant minimal 2 -spheres ,

i.e. ( RP 3 , g RP 3 ) has finitely many embedded minimal RP 2 and finitely many embedded minimal S 2 . In particular, (6.3) is still valid by Theorem 6.2.

For each k { 1 , 2 , 3 } , it follows from (6.4) and Theorem 5.3 that the min-max values of homotopy classes in P k must be stabilized to L ( P k ) , and there are disjoint embedded Z 2 -invariant Z 2 -connected minimal 2-spheres { Γ k , j } j = 1 J k and integer multiplicities { m k , j } j = 1 J k so that

(6.5) L ( P k ) = j = 1 J k m k , j H 2 ( Γ k , j ) .

Also, by Corollary 6.1, we can assume Γ k , 1 X G ± , and Γ k , j with j 2 is a Z 2 -invariant disjoint union of two minimal 2-spheres. By Theorem 5.2, Γ k , 1 is either unstable with multiplicity one or stable.

Sub-case II (a): Γ k , 1 is unstable with multiplicity one for all k { 1 , 2 , 3 } . Suppose Theorem 1.2 (i) fails, i.e. the number of distinct minimal embedded real projective planes in ( RP 3 , g RP 3 ) is strictly less than 4. Clearly, this assumption and (6.3), (6.5) indicate there must be some k 0 { 1 , 2 , 3 } with J k 0 2 . Thus Γ k 0 , 2 is a Z 2 -invariant union of two minimal spheres Γ ± with [ 1 ] Γ ± = Γ and Γ ± lies in the different components of M Γ k 0 , 1 .

If Γ + is stable, and thus strictly stable due to the bumpiness of the metric, then by [48, Proposition 8.8] (using Song’s strategy [43]), we can find another two minimal spheres in ( M , g M ) lying in a fundamental domain of RP 3 (i.e. in a component of M Γ k 0 , 1 containing Γ + ). Together, we obtain three minimal 2-spheres and one area minimizing real projective plane Σ 0 in ( RP 3 , g RP 3 ) .

If Γ + is unstable, then since Γ k 0 , 1 is also assumed to be unstable, there exists an (isotopic area minimizing) stable minimal sphere Γ ̃ + lying between Γ + and Γ k 0 , 1 . Using [48, Proposition 8.8], we obtain at least one more minimal sphere in a fundamental domain of RP 3 (i.e. in the component of M Γ ̃ + containing Γ + ) which is different from Γ + , Γ ̃ + . Therefore, we still have one area minimizing minimal real projective plane Σ 0 and three minimal spheres in ( RP 3 , g RP 3 ) .

Sub-case II (b): Γ k , 1 is stable for some k { 1 , 2 , 3 } . Since Γ k , 1 is stable (and thus strictly stable due to the bumpiness) for some k { 1 , 2 , 3 } , we can apply [48, Proposition 8.8] again to find another two distinct minimal spheres in the fundamental domain of ( RP 3 , g RP 3 ) (i.e. in a component of M Γ k , 1 ). In conclusion, we can find one area minimizing real projective plane Σ 0 and two distinct minimal spheres in ( RP 3 , g RP 3 ) . ∎

7 Minimal Klein bottles in L ( 4 m , 2 m ± 1 )

Consider the unit 3-sphere S 3 = { ( z , w ) C 2 | z | 2 + | w | 2 = 1 } . For any two coprime integers p 1 and q [ 1 , p ) , we have the following cyclic Z p -action on S 3 generated by ξ p , q :

(7.1) [ 1 ] ( z , w ) = ξ p , q ( z , w ) : = ( e 2 π i / p z , e 2 π q i / p w ) .

Then the lens space L ( p , q ) is defined to be S 3 / Z p .

By [8, Corollary 6.4], the Klein bottle embeds into L ( 4 m , 2 m ± 1 ) only. Hence, in this section, we always set

(7.2) M : = S 3 , G : = Z p = ξ p , q , G + = Z p / 2 : = ξ p , q 2 , G : = G G + ,

where p = 4 m , q = 2 m ± 1 and m 1 . In addition, consider the space

X G ± = { ϕ ( T 2 ) ϕ : T 2 S 3 a G -equivariant G ± -separating smooth embedding } ,

endowed with the smooth unoriented topology. One can verify that, for any 𝐺-surface Σ M , Σ / G is a Klein bottle in L ( 4 m , 2 m ± 1 ) if and only if Σ X G ± .

7.1 Sweepouts formed by Klein bottles

We now describe the sweepouts that detect the nontrivial cohomology classes of H k ( X G ± ; Z 2 ) , where k = 3 for m = 1 , and k = 1 for m > 1 . Firstly, we mention the following results given by Ketover [25, Proposition 4.2].

Proposition 7.1

Proposition 7.1 ([25])

Let L ( p , q ) denote the lens space endowed with the round metric, p 2 . Then L ( p , q ) admits an embedded Klein bottle if and only if p = 4 m and q = 2 m ± 1 for m 1 . If m > 1 , then L ( 4 m , 2 m ± 1 ) admits an S 1 -family of minimal Klein bottles. If m = 1 , then L ( 4 , 1 ) admits an S 1 × R P 2 -family of minimal Klein bottles.

For the sake of completeness, we provide a relatively detailed explanation following the constructions in [25, Section 4] (see also [45]). Firstly, one can identify S 3 and S 2 with the group of unit quaternions and pure unit quaternions (without real part) respectively, i.e.

S 3 : = { a + b i + c j + d k | a | 2 + | b | 2 + | c | 2 + | d | 2 = 1 } , S 2 : = { b i + c j + d k S 3 } .

Note that any orientation preserving isometry f Isom + ( S 3 ) = S O ( 4 ) can be represented by f ( q ) = q 1 q q 2 1 for some q 1 , q 2 S 3 . Hence an oriented 2-plane in R 4 spanned by two orthonormal vectors u , v S 3 can be written as u , v = ( q 1 , q 2 ) 1 , i : = q 1 q 2 1 , q 1 i q 2 1 for some q 1 , q 2 S 3 . Denote by G ̃ 2 ( R 4 ) the Grassmannian manifold consisting of all oriented 2-dimensional subspaces of R 4 . Then the space of oriented geodesics in round S 3 is homeomorphic to G ̃ 2 ( R 4 ) S 2 × S 2 , where the homeomorphism is specified by the map

P : G ̃ 2 ( R 4 ) S 2 × S 2 , P ( ( q 1 , q 2 ) 1 , i ) = ( q 1 i q 1 1 , q 2 i q 2 1 ) .

Note that ( a , b ) , ( a , b ) S 2 × S 2 correspond to the same geodesic with opposite orientations.

Next, for a S 2 , B S 2 , denote

τ ( a , B ) : = { x S 3 x P 1 ( a , b ) S 3 , b B } .

This is known to be a Clifford torus if a S 2 and 𝐵 is a great circle of S 2 (cf. [25, (4.16)]). Given b S 2 , denote by

E ( b ) : = ( { x S 2 b 1 x 1 + b 2 x 2 + b 3 x 3 < 0 } ) S 2

an oriented equator. Since τ ( a , E ( b ) ) = τ ( a , E ( b ) ) is τ ( a , E ( b ) ) = τ ( a , E ( b ) ) with the opposite orientation, we see the space of unoriented Clifford tori is homeomorphic to R P 2 × R P 2 . In addition, by [45, Section 5.1], we know that ξ p , q can be represented by ξ p , q ( x ) = e i π ( q + 1 ) / p x e i π ( q 1 ) / p using quaternions. Moreover, given a S 2 and a great circle E S 2 ,

(7.3) ξ p , q ( τ ( a , E ) ) = τ ( η ̂ p , q 1 ( a ) , η ̂ p , q 2 ( E ) ) ,

where η ̂ p , q 1 , η ̂ p , q 2 is the rotation (in the { j , k } -plane) on S 2 by the angle of 2 π ( q + 1 ) / p and 2 π ( q 1 ) / p respectively with fixed points { ± i } .

Case A: p = 4 and q = 1 (or q = 3 ). In this case, one verifies that ξ 4 , 1 is orientation preserving, and ξ 4 , 1 ( τ ( cos ( θ ) j + sin ( θ ) k , E ( b ) ) ) = τ ( cos ( θ ) j sin ( θ ) k , E ( b ) ) is τ ( cos ( θ ) j + sin ( θ ) k , E ( b ) ) with the opposite orientation for any θ [ 0 , 2 π ] , b S 2 , which implies the support of τ ( cos ( θ ) j + sin ( θ ) k , E ( b ) ) is an element in X G ± . Hence we have a family of Klein bottles in L ( 4 , 1 ) L ( 4 , 3 ) parameterized by G : R P 1 × R P 2 X G ± ,

G ( [ a 1 , a 2 ] , [ b 1 , b 2 , b 3 ] ) : = spt ( τ ( a 1 j + a 2 k , E ( b 1 i + b 2 j + b 3 k ) ) ) .

We can now define three maps into X G ± as follows:

Φ 1 : R P 1 X G ± , [ b 1 , b 2 ] G ( [ 1 , 0 ] , [ b 1 , b 2 , 0 ] ) , Φ 2 : R P 2 X G ± , [ b 1 , b 2 , b 3 ] G ( [ 1 , 0 ] , [ b 1 , b 2 , b 3 ] ) , Φ 3 : R P 1 × R P 2 X G ± , ( [ a 1 , a 2 ] , [ b 1 , b 2 , b 3 ] ) G ( [ a 1 , a 2 ] , [ b 1 , b 2 , b 3 ] ) .

One notices that Φ k , k { 1 , 2 , 3 } , is a 𝑘-sweepout in the sense of Almgren–Pitts (cf. [34, Definition 4.1]). Indeed, denote by α l the generator of H 1 ( R P l ; Z 2 ) , l { 1 , 2 } , by λ ̄ the generator of H 1 ( Z 2 ( M ; Z 2 ) ; Z 2 ) and by ι : X G ± Z 2 ( M ; Z 2 ) the natural inclusion. Then the closed curves γ 1 ( e i θ ) = ( [ cos ( θ / 2 ) , sin ( θ / 2 ) ] , [ 1 , 0 , 0 ] ) and γ 2 ( e i θ ) = ( [ 1 , 0 ] , [ cos ( θ / 2 ) , sin ( θ / 2 ) , 0 ] ) in R P 1 × R P 2 satisfy that

  • ι G γ 1 : S 1 Z 2 ( M ; Z 2 ) , ι G γ 2 : S 1 Z 2 ( M ; Z 2 ) are sweepouts in the sense of Almgren–Pitts, since τ ( 1 j , E ( 1 j ) ) and τ ( 1 j , E ( 1 j ) ) respectively are τ ( 1 j , E ( 1 j ) ) with the opposite orientation;

  • α 1 α 2 ( γ 1 ) = α 1 α 2 ( γ 2 ) = 1 , α 1 α 2 ( γ 1 + γ 2 ) = 0 ,

where 0 α 1 α 2 H 1 ( R P 1 × R P 2 ; Z 2 ) . Combining the first bullet with the fact that

α 2 2 0 H 2 ( R P 2 ; Z 2 ) ,

we conclude Φ 1 and Φ 2 are 1-sweepout and 2-sweepout in the sense of Almgren–Pitts respectively. Moreover, we also have G ι ( λ ̄ ) = α 1 α 2 H 1 ( R P 1 × R P 2 ; Z 2 ) by the above two bullets. Together with ( α 1 α 2 ) 3 0 H 3 ( R P 1 × R P 2 ; Z 2 ) , we conclude that Φ 3 is a 3-sweepout in the sense of Almgren–Pitts.

Furthermore, since 0 ( α 1 α 2 ) 3 = G ι ( λ ̄ 3 ) H 3 ( R P 1 × R P 2 ; Z 2 ) , we know that

α : = ι ( λ ̄ ) H 1 ( X G ± ; Z 2 )

satisfies α k 0 H k ( X G ± ; Z 2 ) for every k { 1 , 2 , 3 } . Thus we can define P k , k { 1 , 2 , 3 } , to be the collection of continuous maps Φ from any cubical complex 𝑋 to X G ± that detects α k H k ( X G ± ; Z 2 ) , i.e.

(7.4) P k : = { Φ : X X G ± Φ ( α k ) 0 H k ( X ; Z 2 ) } .

Clearly, the above Φ k P k for all k = 1 , 2 , 3 .

Case B: p = 4 m and q = 2 m ± 1 with m 2 . In this case, one verifies that

ξ p , q ( τ ( cos ( θ ) j + sin ( θ ) k , E ( 1 j ) ) )

is τ ( cos ( θ ) j + sin ( θ ) k , E ( 1 j ) ) with the opposite orientation for any θ [ 0 , 2 π ] . Hence we have a family of Klein bottles in L ( 4 m , 2 m ± 1 ) ( m 2 ) parameterized by G : R P 1 X G ± ,

G ( [ a 1 , a 2 ] ) : = spt ( τ ( a 1 j + a 2 k , E ( 1 j ) ) ) .

We can define the map

Φ 1 = G : R P 1 X G ± ,

which is a sweepout in the sense of Almgren–Pitts since τ ( 1 j , E ( 1 j ) ) is τ ( 1 j , E ( 1 j ) ) with the opposite orientation. Hence G ι ( λ ̄ ) 0 H 1 ( R P 1 ; Z 2 ) ,

α : = ι ( λ ̄ ) 0 H 1 ( X G ± ; Z 2 ) ,

and we can similarly define

(7.5) P 1 : = { Φ : X X G ± Φ ( α ) 0 H 1 ( X ; Z 2 ) }

with Φ 1 P 1 .

7.2 Proof of Theorem 1.5

Using Theorem 5.3 and Proposition 5.4, we have the following corollary, which gives the existence of min-max and minimizing minimal Klein bottles.

Corollary 7.2

Suppose P k is given by (7.4) or (7.5), and Π P k is a homotopy class of sweepouts in X G ± . Then L ( Π ) > 0 , and the 𝐺-connected min-max minimal 𝐺-surfaces { Γ j } j = 1 J with integer multiplicities { m j } j = 1 J given by Theorem 5.3 satisfy that

  1. there is exactly one 𝐺-component, say Γ 1 , that is a G ± -separating minimal torus with multiplicity one, i.e. Γ 1 X G ± and m 1 = 1 ;

  2. any other 𝐺-component Γ j , j 2 , is a 𝐺-invariant disjoint union of # G = 4 m minimal 2-spheres in 𝑀.

Moreover, suppose { Σ k } k N X G ± is an area minimizing sequence in X G ± satisfying

lim k H 2 ( Σ k ) = inf { H 2 ( Γ ) Γ X G ± } ;

then, up to a subsequence, Σ k converges as varifolds to an area minimizing minimal torus Σ 0 X G ± with multiplicity one.

Proof

Firstly, L ( Π ) > 0 since Π P k are 𝑘-sweepouts in the sense of Almgren–Pitts. By Theorem 5.3, { Γ j } j = 1 J satisfies the genus bound (4.1), which implies each Γ j is either

  1. a G ± -separating minimal torus in X G ± with multiplicity one, i.e. π ( Γ j ) is a minimal Klein bottle in L ( 4 m , 2 m ± 1 ) ; or

  2. a connected 𝐺-invariant minimal torus not in X G ± with multiplicity one, i.e. π ( Γ j ) is a minimal torus in L ( 4 m , 2 m ± 1 ) ; or

  3. a 𝐺-invariant disjoint union of minimal spheres, i.e. π ( Γ j ) is a minimal sphere in lens spaces L ( 4 m , 2 m ± 1 ) . (Note that there is no embedded RP 2 in L ( 4 m , 2 m ± 1 ) ; see Geiges–Thies [16].)

Combining Proposition 5.4 with the fact that every two elements in X G ± have nonempty intersections, we know that there is exactly one element of { Γ j } j = 1 J that is in X G ± , which gives (i). Then, by (4.1), every other 𝐺-component Γ j , j 2 , is in case (c). Note that there is no embedded RP 2 in L ( 4 m , 2 m ± 1 ) , and the only nontrivial finite effective free action on S 2 is Z 2 with the quotient homeomorphic to RP 2 . Hence Γ j , j 2 , has # G = 4 m components with 𝐺 permuting them.

Suppose { Σ k } k N X G ± is an area minimizing sequence in X G ± . It then follows from the 𝐺-invariance of Σ k and [37, Theorem 1] that Σ k converges (up to a subsequence) as varifolds to a disjoint union of embedded 𝐺-connected minimal 𝐺-surfaces { Γ j } j = 1 J with integer multiplicities { m j } j = 1 J so that the genus bound (4.1) is also valid. Next, combining the proof of Proposition 5.4 and the above arguments, we know that there is exactly one 𝐺-component, say Γ 1 , satisfying Γ 1 X G ± and m j = 1 . Finally, we notice J = 1 since

inf { H 2 ( Γ ) Γ X G ± } H 2 ( Γ 1 ) lim k H 2 ( Σ k ) .

Next, we can use the Lusternik–Schnirelmann theory to show the following result.

Theorem 7.3

Given any Riemannian metric g L on L ( 4 m , 2 m ± 1 ) with m 1 , let M , G , G ± be given by (7.2), and let g M be the 𝐺-invariant pull-back metric on 𝑀. Suppose X G ± contains only finitely many elements that are minimal in ( M , g M ) . Then

0 < 4 M 0 < L ( P 1 ) < L ( P 2 ) < L ( P 3 ) if m = 1 , and 0 < 4 m M 0 < L ( P 1 ) if m 2 ,

where 4 m M 0 = inf { H 2 ( Γ ) Γ X G ± } , and P k is defined in (7.4), (7.5).

Proof

Using Corollary 7.2 in place of Corollary 6.1 and [6, Proposition 2.3], the proof in Theorem 6.2 can be taken almost verbatim to show the desired results. ∎

At the end of this subsection, we prove Theorem 1.5.

Proof of Theorem 1.5

Firstly, given any Riemannian metric g L on L ( 4 m , 2 m ± 1 ) , we have an embedded area minimizing Klein bottle denoted as Σ 0 by the second part of Corollary 7.2. Let M = S 3 , G = Z 4 m and G ± G be defined as in (7.2). Endow 𝑀 with the pull-back metric g M so that ( M , g M ) is locally isometric to ( L ( 4 m , 2 m ± 1 ) , g L ) .

Case I: g L is a metric with positive Ricci curvature. In this case, g M is a 𝐺-invariant Riemannian metric on M = S 3 with positive Ricci curvature. Hence, similar to (6.1), the embedded Frankel property indicates there is no embedded 𝐺-invariant union of minimal 2-spheres in ( M , g M ) , i.e. there is no minimal 2-sphere in ( L ( 4 m , 2 m ± 1 ) , g L ) .

Without loss of generality, we assume that ( L ( 4 m , 2 m ± 1 ) , g L ) contains only finitely many distinct minimal embedded Klein bottles, i.e.

(7.6) ( M , g M ) contains finitely many minimal tori in X G ± .

Sub-case I (A): m = 1 . Combining (7.6) with Theorem 7.3, { P k } k = 1 3 in (7.4) satisfies

(7.7) 0 < 4 H 2 ( Σ 0 ) < L ( P 1 ) < L ( P 2 ) < L ( P 3 ) .

For each k { 1 , 2 , 3 } , although P k may contain many different homotopy classes, we see from (7.6) and Corollary 7.2 (i)–(ii) that the min-max values of the homotopy classes in P k must be stabilized to L ( P k ) , and moreover, L ( P k ) is realized by the area of an embedded 𝐺-connected minimal torus Γ k X G ± with multiplicity one, i.e. L ( P k ) = H 2 ( Γ k ) .

Therefore, we have found four distinct embedded minimal Klein bottles in ( L ( 4 , 1 ) , g L ) .

Sub-case I (B): m 2 . The proof in Sub-case I (A) would carry over for Σ 0 and P 1 in (7.5) to find two distinct embedded minimal Klein bottles in ( L ( 4 m , 2 m ± 1 ) , g L ) .

Case II: g L is a bumpy metric. In this case, every embedded 𝐺-invariant minimal surface in ( M , g M ) is non-degenerate. Without loss of generality, we also assume that ( M , g M ) contains finitely many 𝐺-invariant minimal tori in X G ± and finitely many 𝐺-invariant union of minimal 2-spheres.

Sub-case II (A): m = 1 . Combining the constructions of { P k } k = 1 3 in (7.4) and the above finiteness assumptions with Theorem 7.3, we know that (7.7) is still valid here. In addition, for each k { 1 , 2 , 3 } , using the arguments in Sub-case I (A) and Corollary 7.2 (i)–(ii), there are disjoint embedded 𝐺-connected minimal 𝐺-surfaces { Γ k , j } j = 1 J k and integer multiplicities { m k , j } j = 1 J k so that

L ( P k ) = j = 1 J k m k , j H 2 ( Γ k , j ) ,

where Γ k , 1 X G ± has multiplicity m k , 1 = 1 , and Γ k , j with j { 2 , , J k } is a 𝐺-invariant disjoint union of 4 m minimal 2-spheres.

Suppose Theorem 1.5 (i) fails, i.e. the number of distinct embedded minimal Klein bottles in ( L ( 4 , 1 ) , g L ) is strictly less than 4. Clearly, this assumption and (7.7) indicate the existence of k 0 { 1 , 2 , 3 } with J k 0 2 . Hence we have a minimal Klein bottle K = π ( Γ k 0 , 1 ) and a minimal 2-sphere S = π ( Γ k 0 , 2 ) in ( L ( 4 , 1 ) , g L ) with K S = .

By cutting L ( 4 , 1 ) along 𝐾, we obtain a compact manifold L ̃ with a connected boundary L ̃ diffeomorphic to the oriented double cover K ̃ of 𝐾. Then, by isotopic area minimizing 𝑆 within the region in L ̃ enclosed between L ̃ = K ̃ and 𝑆, we have a stable (and thus strictly stable) minimal 2-sphere S 0 L ̃ L ̃ so that L ̃ S 0 has a component 𝑁 with N L ̃ = . After applying [48, Proposition 8.8] (using Song’s strategy [43]) to 𝑁, we can find either

  • at least one more minimal sphere in 𝑁 that is different from 𝑆 if S S 0 ; or

  • at least two more minimal spheres in 𝑁 if S = S 0 .

Thus we obtain three minimal 2-spheres and an area minimizing Klein bottle in ( L ( 4 , 1 ) , g L ) .

Sub-case II (B): m 2 . The proof in Sub-case II (A) would carry over for Σ 0 and P 1 in (7.5) to find either two distinct embedded minimal Klein bottles or one area minimizing Klein bottle together with three embedded minimal spheres in ( L ( 4 m , 2 m ± 1 ) , g L ) . ∎

8 Minimal torus in lens spaces

Note that Theorem 1.6 (ii) has been proven by Ketover in [25, Theorem 4.8]. Hence, in this section, we only consider minimal tori in L ( 2 , 1 ) = RP 3 and L ( p , q ) with q { 1 , p 1 } . In particular, we always set

M : = S 3 , G : = Z p = ξ p , q , G + = Z p / 2 : = ξ p , q 2 , G : = G G + ,

where ξ p , q is given by (7.1), and either ( p , q ) = ( 2 , 1 ) or q { 1 , p 1 } . In addition, we consider the following spaces:

  • X : = { ϕ ( T 2 ) ϕ : T 2 M / G a separating embedding } ,

  • Y : = { π ( S 1 ) S 1 is a G -invariant equator (great circle) in S 3 } ,

  • X ̄ : = X Y ,

which are endowed with the smooth unoriented topology. Note that ϕ ( T 2 ) is separating if ( M / G ) ϕ ( T 2 ) has two connected components U 1 , U 2 . Moreover, denote by

ι : X ̄ Z 2 ( M / G ; Z 2 )

the natural inclusion, and by λ ̄ the generator of H 1 ( Z 2 ( M / G ; Z 2 ) ; Z 2 ) .

8.1 Sweepouts formed by tori

To begin with, we have a family of T 2 in 𝒳 by Ketover’s construction [25, Proposition 4.2].

Proposition 8.1

Proposition 8.1 (Ketover [25])

Let L ( p , q ) denote the lens space endowed with the round metric, p 2 . Then we have

  1. RP 3 admits a family of Clifford tori parameterized by R P 2 × R P 2 ;

  2. L ( p , q ) with q { 1 , p 1 } admits exactly one Clifford torus.

For the sake of completeness, we adopt the notation as in Section 7 and explain the constructions of sweepouts formed by tori in RP 3 or L ( p , q ) with q { 1 , p 1 } .

Firstly, for any a , b S 2 , recall that E ( b ) = { x S 2 x b < 0 } is an oriented equator and τ ( a , E ( b ) ) is an oriented Clifford torus in S 3 . Then, for any t [ 1 , 1 ] , we define

E t ( b ) : = { x S 2 x b : = x 1 b 1 + x 2 b 2 + x 3 b 3 < t }

to be an oriented circle parallel to E ( b ) with E 1 ( b ) = { b } and E 1 ( b ) = { b } . It is known that τ ( a , E t ( b ) ) is an oriented flat torus in S 3 parallel to the Clifford torus τ ( a , E 0 ( b ) ) (cf. [25, page 22]), which degenerates to an oriented great circle in S 3 at t = ± 1 .

Case A: ( p , q ) = ( 2 , 1 ) and L ( p , q ) = RP 3 . Using the notation in Section 7 and (7.3), ξ 2 , 1 is orientation preserving and the oriented (possibly degenerated) torus ξ 2 , 1 ( τ ( a , E t ( b ) ) ) in S 3 is the same as τ ( a , E t ( b ) ) for all a , b S 2 , t [ 1 , 1 ] , which implies the quotient of spt ( τ ( a , E t ( b ) ) ) is an element in X ̄ . Therefore, we have a continuous family of oriented tori in RP 3 parameterized by G ̃ : [ 1 , 1 ] × S 2 × S 2 X ̄ (with oriented smooth topology),

G ̃ ( t , a , b ) = ( τ ( a , E t ( b ) ) ) / G for all ( t , a , b ) [ 1 , 1 ] × S 2 × S 2 .

Note that the oriented torus G ̃ ( t , a , b ) = G ̃ ( t , a , b ) is the oriented torus

G ̃ ( t , a , b ) = G ̃ ( t , a , b )

with the opposite orientation for t ( 1 , 1 ) . Hence G ̃ induces a continuous map into ( X ̄ , Y ) with unoriented smooth topology,

G : ( X ̄ , Y ) ( X ̄ , Y ) , G ( [ t , a , b ] ) = spt ( G ̃ ( t , a , b ) ) ,

where [ t , a , b ] X ̄ ,

X ̄ : = ( [ 1 , 1 ] × S 2 × S 2 ) / Z 2 Z 2 , Y : = X ̄ , X : = X ̄ Y ,

and Z 2 Z 2 acts on [ 1 , 1 ] × S 2 × S 2 by ( 0 , 0 ) = id ,

(8.1) ( 1 , 0 ) ( t , a , b ) = ( t , a , b ) , ( 0 , 1 ) ( t , a , b ) = ( t , a , b ) , ( 1 , 1 ) ( t , a , b ) = ( t , a , b ) .

In addition, one notices that the parameter space X ̄ retracts onto

X 4 : = { [ 0 , a , b ] X a , b S 2 } R P 2 × R P 2 ,

the parameter space of Clifford tori. Then, by Poincaré–Lefschetz duality,

(8.2) H 1 ( X ̄ , Y ; Z 2 ) H 4 ( X ̄ ; Z 2 ) H 4 ( X 4 ; Z 2 ) Z 2

has a generator λ H 1 ( X ̄ , Y ; Z 2 ) , and for any fixed a , b S 2 , the curve γ ( t ) : = [ t , a , b ] X ̄ satisfies that

ι G γ : ( [ 1 , 1 ] , { ± 1 } ) ( Z 2 ( M ; Z 2 ) , { 0 } )

is a sweepout in the sense of Almgren–Pitts, which implies G ι ( λ ̄ ) = λ H 1 ( X ̄ , Y ; Z 2 ) .

Lemma 8.2

λ 4 = G ι ( λ ̄ 4 ) 0 H 4 ( X ̄ , Y ; Z 2 ) and λ 5 = 0 .

The proof of Lemma 8.2 is left to Appendix A. Now, we know that

ι G : ( X ̄ , Y ) ( Z 2 ( M / G ; Z 2 ) , { 0 } )

is a 4-sweepout in the sense of Almgren–Pitts as λ 4 = ( ι G ) ( λ ̄ 4 ) 0 . In particular,

α = ι ( λ ̄ ) H 1 ( X ̄ , Y ; Z 2 )

satisfies α 4 0 H 4 ( X ̄ , Y ; Z 2 ) . Hence we can define the collection of continuous maps

(8.3) P k : = { Φ : ( X , Z ) ( X ̄ , Y ) Φ ( α k ) 0 H k ( X , Z ; Z 2 ) } ,

where k { 1 , 2 , 3 , 4 } and Z X are any two cubical complexes.

Case B: q { 1 , p 1 } . In this case, we fix a 0 = b 0 = ( 1 , 0 , 0 ) S 3 . Similarly, by (7.3), the oriented (and possibly degenerated) torus ξ p , q ( τ ( a 0 , E t ( b 0 ) ) ) in S 3 is the same as τ ( a 0 , E t ( b 0 ) ) for all t [ 1 , 1 ] , which implies the quotient of spt ( τ ( a 0 , E t ( b 0 ) ) ) is an element in X ̄ . Therefore, we have a continuous family of unoriented tori in RP 3 parameterized by

G : ( [ 1 , 1 ] , { ± 1 } ) ( X ̄ , Y ) , G ( t ) : = spt ( τ ( a 0 , E t ( b 0 ) ) / G ) for all t [ 1 , 1 ] .

Note that ι G is a sweepout in the sense of Almgren–Pitts. Hence

α : = ι ( λ ̄ ) 0 H 1 ( X ̄ , Y ; Z 2 ) ,

and we can define the collection of continuous maps

(8.4) P 1 : = { Φ : ( X , Z ) ( X ̄ , Y ) Φ ( α ) 0 H 1 ( X , Z ; Z 2 ) } ,

where Z X are any two cubical complexes.

8.2 Proof of Theorem 1.6

Since T 2 is orientable, we can apply [48, Theorem B] to obtain the following corollary.

Corollary 8.3

Let g L be a Riemannian metric on L ( p , q ) with positive Ricci curvature, where ( p , q ) = ( 2 , 1 ) or q { 1 , p 1 } . Suppose P k is given by (8.3) or (8.4), and Π P k is a homotopy class of sweepouts in X ̄ . Then

L ( Π ) = H 2 ( Γ ) > 0

for a smooth connected embedded minimal torus Γ in ( L ( p , q ) , g L ) .

Proof

Firstly, we know that L ( Π ) > 0 since Π is a family of sweepouts in the sense of Almgren–Pitts. Then, combining [48, Theorem B] with the embedded Frankel property, there exists a connected smooth embedded minimal surface Γ in ( L ( p , q ) , g L ) so that

L ( Π ) = m H 2 ( Γ ) for some integer m Z + ,

and (by the genus bound) Γ is either a minimal

  1. torus with multiplicity one, or

  2. 2-sphere, or

  3. real projective plane with a stable connected double cover (only if L ( p , q ) RP 3 ), or

  4. Klein bottle with a stable connected double cover (only if ( p , q ) = ( 4 m , 2 m ± 1 ) ).

Note that the pull-back metric g M = π g L on M = S 3 also has positive Ricci curvature. Hence (c), (d) cannot happen. Additionally, if Γ is in case (b), then π 1 ( Γ ) is a # G = p union of minimal spheres in S 3 , since 2-spheres only have trivial finite covers. This contradicts the embedded Frankel property in ( M , g M ) . Therefore, Γ has to be in case (a). ∎

Recall that L ( P k ) : = inf Φ P k sup x dmn ( Φ ) H 2 ( Φ ( x ) ) .

Theorem 8.4

Let g L be a Riemannian metric on L ( p , q ) with positive Ricci curvature, where ( p , q ) = ( 2 , 1 ) or q { 1 , p 1 } . Suppose 𝒳 contains only finitely many elements that are minimal in ( L ( p , q ) , g L ) . Then

0 < L ( P 1 ) < L ( P 2 ) < L ( P 3 ) < L ( P 4 )

when ( p , q ) = ( 2 , 1 ) , and 0 < L ( P 1 ) when q { 1 , p 1 } .

Proof

This result follows from the Lusternik–Schnirelmann theory. One can easily use Corollary 8.3 and [20, Lemma 5.3] to extend the proof of [34, Theorem 6.1] and [20, Theorem 5.2]. ∎

We mention that, for a general Riemannian metric g L on L ( p , q ) , the above result is still valid if ( L ( p , q ) , g L ) contains only finitely many minimal 2-spheres, minimal real projective planes, minimal Klein bottles and minimal tori.

Finally, we prove Theorem 1.6.

Proof of Theorem 1.6

By combining Corollary 8.3 and Theorem 8.4, we find four embedded minimal tori in L ( 2 , 1 ) = RP 3 and one embedded minimal torus in L ( p , q ) with q { 1 , p 1 } . It remains to show the second part of (i), namely to find one more embedded minimal torus in RP 3 under the additional assumption of bumpiness.

To begin with, we recall the setup of White’s manifold structure theorem [51]. Let M , N be compact smooth Riemannian manifolds such that dim N < dim M . Given an open subset Γ of C q Riemannian metrics on 𝑀, consider the space ℳ of ordered pairs ( γ , [ N ] ) , where γ Γ and [ N ] denotes the diffeomorphism classes of minimal embeddings of a Riemannian manifold 𝑁 into 𝑀 with respect to 𝛾. It was shown that ℳ is a separable C 2 Banach manifold and the projection map

Π : M Γ , ( γ , [ N ] ) γ

is smooth Fredholm with Fredholm index 0.

Now, let 𝑁 be an unoriented torus, M = RP 3 , and let Γ be the set of all C 4 Riemannian metrics on RP 3 with positive Ricci curvature. Since Γ is connected [19] and Π : M Γ is proper [11], we can calculate the mapping degree 𝑑 of Π using [51, Theorem 5.1]. Denote by γ 0 the round metric on RP 3 . By Proposition 8.1 (1), the set of all Clifford tori in ( RP 3 , γ 0 ) is diffeomorphic to R P 2 × R P 2 . As a direct corollary of Lawson’s conjecture [9, 10], every embedded minimal torus in ( RP 3 , γ 0 ) is congruent to the Clifford torus. Hence Σ = Π 1 ( γ 0 ) is diffeomorphic to R P 2 × R P 2 with χ ( Σ ) = 1 . Together with the fact that every Clifford torus in ( RP 3 , γ 0 ) has nullity 4 and Morse index 1, we see d = ( 1 ) index ( Σ ) χ ( Σ ) = 1 . For generic metrics 𝛾 of positive Ricci curvature on RP 3 , we further have

S Π 1 ( γ ) ( 1 ) index ( S ) = d = 1 ,

which implies that the number of elements in Π 1 ( γ ) is odd. Given that we already found four embedded minimal tori in RP 3 , there must be at least a fifth one. This completes the proof. ∎

Award Identifier / Grant number: 25ZR1402252

Award Identifier / Grant number: 12501076

Award Identifier / Grant number: DMS-2243149

Funding statement: Tongrui Wang is supported by the Natural Science Foundation of Shanghai 25ZR1402252 and the National Natural Science Foundation of China 12501076. Xingzhe Li and Xuan Yao are supported by NSF grant DMS-2243149.

A Proof of Lemma 8.2

Proof of Lemma 8.2

Using the notation in Section 8.1, Case A, for any 𝑘-dimensional closed submanifold 𝑁 in X ̄ , we denote by [ N ] H k ( X ̄ ; Z 2 ) the images of the fundamental classes of 𝑁 under the inclusion into X ̄ , and denote by [ N ] H 5 k ( X ̄ , Y ; Z 2 ) its Poincaré–Lefschetz dual. It then follows from the Poincaré–Lefschetz duality (8.2) that the generator λ H 1 ( X ̄ , Y ; Z 2 ) coincides with [ X 4 ] , where X 4 : = { [ 0 , a , b ] X ̄ } R P 2 × R P 2 is the parameter space of Clifford tori. We now use the transversal intersection product to visualize the cup product λ k , k = 2 , 3 , 4 , 5 .

Firstly, after regarding a = ( a 1 , a 2 , a 3 ) , b = ( b 1 , b 2 , b 3 ) S 2 as vectors in R 3 , we can define

X 4 := { [ a b , a , b ] X ̄ a , b S 2 } , X 3 := X 4 X 4 = { [ 0 , a , b ] X ̄ a , b S 2 with a b = 0 } ,

where a b = i = 1 3 a i b i is the inner product in R 3 . Since { ( a b , a , b ) a , b S 2 } is a 4-submanifold of [ 1 , 1 ] × S 2 × S 2 invariant under the ( Z 2 Z 2 ) -action in (8.1), we know X 4 is a closed 4-submanifold of X ̄ and is homotopic to X 4 through ( s , [ t , a , b ] ) [ s t , a , b ] , s [ 0 , 1 ] . In addition, for any [ 0 , a , b ] X 3 = X 4 X 4 , we have a b , and thus there exists a curve σ S 2 with σ ( 0 ) = a and σ ( 0 ) = b . Hence σ ̃ ( s ) = [ σ ( s ) b , σ ( s ) , b ] is a curve in X 4 so that σ ̃ ( 0 ) has a component ( σ b ) ( 0 ) t = t , which implies X 4 is transversal to X 4 along their intersection X 3 . Therefore, by [17, Theorem 3.6] (see also [7, VI, Theorem 11.9]), we have λ 2 = [ X 4 ] [ X 4 ] = [ X 4 X 4 ] = [ X 3 ] H 2 ( X ̄ , Y ; Z 2 ) .

Next, set e 1 : = ( 1 , 0 , 0 ) , e 2 : = ( 0 , 1 , 0 ) , e 3 : = ( 0 , 0 , 1 ) . For any a , b S 2 with a b = 0 , we can take the cross product vector v = ( v 1 , v 2 , v 3 ) : = a × b in R 3 , i.e.

v 1 ( a , b ) : = a 1 b 3 a 3 b 2 , v 2 ( a , b ) : = a 3 b 1 a 1 b 3 , v 3 ( a , b ) = a 1 b 2 a 2 b 1 .

Then { a , b , v } forms an orthonormal basis of R 3 . Define then

X 3 := { [ v 1 ( a , b ) , a , b ] X ̄ a , b S 2 with a b = 0 } , X 2 := X 4 X 3 = { [ 0 , a , b ] X ̄ a , b S 2 with a b = 0 , v 1 ( a , b ) = 0 } .

Since { ( v 1 ( a , b ) , a , b ) a , b S 2 , a b = 0 } is a closed 3-submanifold of [ 1 , 1 ] × S 2 × S 2 invariant under the ( Z 2 Z 2 ) -action in (8.1), it is clear that X 3 is a closed 3-submanifold of X ̄ and is homotopic to X 3 . Additionally, for any a , b S 2 with a b = 0 and v 1 ( a , b ) = 0 , we can rotate 𝑎, 𝑏 and v = a × b together around the ( v × e 1 ) -axis to obtain curves a ( θ ) , b ( θ ) , v ( θ ) = a ( θ ) × b ( θ ) S 2 so that

a ( 0 ) = a , b ( 0 ) = b , v ( 0 ) = v , and v 1 ( θ ) = v 1 ( a ( θ ) , b ( θ ) ) = sin ( θ ) .

Hence it follows that σ ̃ ( θ ) = [ v 1 ( θ ) , a ( θ ) , b ( θ ) ] is a curve in X 3 so that σ ̃ ( 0 ) has a component v 1 ( 0 ) t = t , which implies X 3 is transversal to X 4 along their intersection X 2 . Therefore, we have

λ 3 = [ X 4 ] [ X 3 ] = [ X 4 X 3 ] = [ X 2 ] H 3 ( X ̄ , Y ; Z 2 ) .

Moreover, define

X 2 := { [ v 2 ( a , b ) , a , b ] X ̄ a , b S 2 with a b = 0 , v 1 ( a , b ) = 0 } , X 1 := X 4 X 2 = { [ 0 , a , b ] X ̄ a , b S 2 with a b = 0 , v 1 ( a , b ) = v 2 ( a , b ) = 0 } .

Similarly, one notices that X 2 is a closed 2-submanifold of X ̄ and is homotopic to X 2 . Given any a , b S 2 with a b = 0 and v 1 ( a , b ) = v 2 ( a , b ) = 0 , one can use the rotation around the e 1 -axis ( ± e 1 = v × e 2 ) to show that X 2 is transversal to X 4 along their intersection X 1 . Hence λ 4 = [ X 4 ] [ X 2 ] = [ X 4 X 2 ] = [ X 1 ] H 4 ( X ̄ , Y ; Z 2 ) .

Finally, we notice that X 1 is a nontrivial closed curve in X 4 X given by

θ [ 0 , ( cos ( θ / 2 ) , sin ( θ / 2 ) , 0 ) , ( sin ( θ / 2 ) , cos ( θ / 2 ) , 0 ) ] for all θ [ 0 , 2 π ] .

Hence [ X 1 ] 0 H 1 ( X 4 ; Z 2 ) H 1 ( X ̄ ; Z 2 ) , and thus λ 4 = [ X 1 ] 0 in H 4 ( X ̄ , Y ; Z 2 ) . Moreover, one verifies that X 1 : = { [ a 1 b 1 , a , b ] [ 0 , a , b ] X 1 } is homotopic to X 1 and transversal to X 4 , and X 1 X 4 has two points, which implies

λ 5 = [ X 1 X 4 ] = 0 H 5 ( X ̄ , Y ; Z 2 ) .

Acknowledgements

The authors would like to thank Professor Gang Tian for his interest in this work, and Professor Xin Zhou for the guidance and discussions. They also thank Professor Zhichao Wang, Professor Jintian Zhu, Alex Xu and Nick Sun for helpful discussions. Xingzhe Li and Xuan Yao are grateful to Tongrui Wang and Westlake University for hosting them for a visit during which many of these ideas were conceived. Part of this work was done when Xingzhe Li and Xuan Yao visited Princeton University and we would also like to thank Princeton University for its hospitality. We also thank the anonymous reviewers for helpful comments.

References

[1] F. J. Almgren, Jr., The homotopy groups of the integral cycle groups, Topology 1 (1962), 257–299. 10.1016/0040-9383(62)90016-2Search in Google Scholar

[2] F. J. Almgren, Jr., The theory of varifolds, Mimeographed Notes, (1965). Search in Google Scholar

[3] R. H. Bamler and B. Kleiner, Ricci flow and contractibility of spaces of metrics, preprint (2019), https://arxiv.org/abs/1909.08710. Search in Google Scholar

[4] R. H. Bamler and B. Kleiner, Ricci flow and diffeomorphism groups of 3-manifolds, J. Amer. Math. Soc. 36 (2023), no. 2, 563–589. 10.1090/jams/1003Search in Google Scholar

[5] G. D. Birkhoff, Dynamical systems with two degrees of freedom, Trans. Amer. Math. Soc. 18 (1917), no. 2, 199–300.10.1090/S0002-9947-1917-1501070-3Search in Google Scholar

[6] H. Bray, S. Brendle, M. Eichmair and A. Neves, Area-minimizing projective planes in 3-manifolds, Comm. Pure Appl. Math. 63 (2010), no. 9, 1237–1247. 10.1002/cpa.20319Search in Google Scholar

[7] G. E. Bredon, Topology and geometry, Grad. Texts in Math. 139, Springer, New York 1993. 10.1007/978-1-4757-6848-0Search in Google Scholar

[8] G. E. Bredon and J. W. Wood, Non-orientable surfaces in orientable 3-manifolds, Invent. Math. 7 (1969), 83–110. 10.1007/BF01389793Search in Google Scholar

[9] S. Brendle, Embedded minimal tori in S 3 and the Lawson conjecture, Acta Math. 211 (2013), no. 2, 177–190. 10.1007/s11511-013-0101-2Search in Google Scholar

[10] S. Brendle, Minimal surfaces in S 3 : A survey of recent results, Bull. Math. Sci. 3 (2013), no. 1, 133–171. 10.1007/s13373-013-0034-2Search in Google Scholar

[11] H. I. Choi and R. Schoen, The space of minimal embeddings of a surface into a three-dimensional manifold of positive Ricci curvature, Invent. Math. 81 (1985), no. 3, 387–394. 10.1007/BF01388577Search in Google Scholar

[12] T. H. Colding and C. De Lellis, The min-max construction of minimal surfaces, Surveys in differential geometry, Vol. VIII, Surv. Differ. Geom. 8, International Press, Somerville (2003), 75–107. 10.4310/SDG.2003.v8.n1.a3Search in Google Scholar

[13] T. H. Colding, D. Gabai and D. Ketover, On the classification of Heegaard splittings, Duke Math. J. 167 (2018), no. 15, 2833–2856. 10.1215/00127094-2018-0023Search in Google Scholar

[14] C. De Lellis and F. Pellandini, Genus bounds for minimal surfaces arising from min-max constructions, J. reine angew. Math. 644 (2010), 47–99. 10.1515/crelle.2010.052Search in Google Scholar

[15] T. Frankel, On the fundamental group of a compact minimal submanifold, Ann. of Math. (2) 83 (1966), 68–73. 10.2307/1970471Search in Google Scholar

[16] H. Geiges and N. Thies, Klein bottles in lens spaces, Involve 16 (2023), no. 4, 621–636. 10.2140/involve.2023.16.621Search in Google Scholar

[17] R. M. Goresky, Whitney stratified chains and cochains, Trans. Amer. Math. Soc. 267 (1981), no. 1, 175–196. 10.1090/S0002-9947-1981-0621981-XSearch in Google Scholar

[18] M. A. Grayson, Shortening embedded curves, Ann. of Math. (2) 129 (1989), no. 1, 71–111. 10.2307/1971486Search in Google Scholar

[19] R. S. Hamilton, Three-manifolds with positive Ricci curvature, J. Differential Geom. 17 (1982), no. 2, 255–306. 10.4310/jdg/1214436922Search in Google Scholar

[20] R. Haslhofer and D. Ketover, Minimal 2-spheres in 3-spheres, Duke Math. J. 168 (2019), no. 10, 1929–1975. 10.1215/00127094-2019-0009Search in Google Scholar

[21] A. E. Hatcher, A proof of the Smale conjecture, Diff ( S 3 ) O ( 4 ) , Ann. of Math. (2) 117 (1983), no. 3, 553–607. 10.2307/2007035Search in Google Scholar

[22] K. Irie, K. Irie, F. C. Marques and A. Neves, Density of minimal hypersurfaces for generic metrics, Ann. of Math. (2) 187 (2018), no. 3, 963–972. 10.4007/annals.2018.187.3.8Search in Google Scholar

[23] D. Ketover, Equivariant min-max theory, preprint (2016), https://arxiv.org/abs/1612.08692. Search in Google Scholar

[24] D. Ketover, Genus bounds for min-max minimal surfaces, J. Differential Geom. 112 (2019), no. 3, 555–590. 10.4310/jdg/1563242473Search in Google Scholar

[25] D. Ketover, Flipping heegaard splittings and minimal surfaces, preprint (2022), https://arxiv.org/abs/2211.03745. Search in Google Scholar

[26] D. Ketover and Y. Liokumovich, The Smale conjecture and min-max theory, Invent. Math. 241 (2025), no. 1, 1–34. 10.1007/s00222-025-01334-zSearch in Google Scholar

[27] D. Ketover, F. C. Marques and A. Neves, The catenoid estimate and its geometric applications, J. Differential Geom. 115 (2020), no. 1, 1–26. 10.4310/jdg/1586224840Search in Google Scholar

[28] F. R. Smith, On the existence of embedded minimal 2-spheres in the 3-sphere, endowed with an arbitrary Riemannian metric, Ph.D. Thesis, Australian National University, 1982. 10.1017/S0004972700026216Search in Google Scholar

[29] Y. Liokumovich, F. C. Marques and A. Neves, Weyl law for the volume spectrum, Ann. of Math. (2) 187 (2018), no. 3, 933–961. 10.4007/annals.2018.187.3.7Search in Google Scholar

[30] Z. Liu, The existence of embedded 𝐺-invariant minimal hypersurface, Calc. Var. Partial Differential Equations 60 (2021), no. 1, Paper No. 36. 10.1007/s00526-020-01804-7Search in Google Scholar

[31] L. Lyusternik and L. Šnirelman, Topological methods in variational problems and their application to the differential geometry of surfaces, Uspehi Matem. Nauk (N. S.) 2 (1947), no. 1(17), 166–217. Search in Google Scholar

[32] F. C. Marques, R. Montezuma and A. Neves, Morse inequalities for the area functional, J. Differential Geom. 124 (2023), no. 1, 81–111. 10.4310/jdg/1685121320Search in Google Scholar

[33] F. C. Marques and A. Neves, Morse index and multiplicity of min-max minimal hypersurfaces, Camb. J. Math. 4 (2016), no. 4, 463–511. 10.4310/CJM.2016.v4.n4.a2Search in Google Scholar

[34] F. C. Marques and A. Neves, Existence of infinitely many minimal hypersurfaces in positive Ricci curvature, Invent. Math. 209 (2017), no. 2, 577–616. 10.1007/s00222-017-0716-6Search in Google Scholar

[35] F. C. Marques and A. Neves, Morse index of multiplicity one min-max minimal hypersurfaces, Adv. Math. 378 (2021), Article ID 107527. 10.1016/j.aim.2020.107527Search in Google Scholar

[36] F. C. Marques, A. Neves and A. Song, Equidistribution of minimal hypersurfaces for generic metrics, Invent. Math. 216 (2019), no. 2, 421–443. 10.1007/s00222-018-00850-5Search in Google Scholar

[37] W. Meeks, III, L. Simon and S. T. Yau, Embedded minimal surfaces, exotic spheres, and manifolds with positive Ricci curvature, Ann. of Math. (2) 116 (1982), no. 3, 621–659. 10.2307/2007026Search in Google Scholar

[38] J. D. Moore and R. Schlafly, On equivariant isometric embeddings, Math. Z. 173 (1980), no. 2, 119–133. 10.1007/BF01159954Search in Google Scholar

[39] J. T. Pitts, Existence and regularity of minimal surfaces on Riemannian manifolds, Math. Notes 27, Princeton University, Princeton 1981. 10.1515/9781400856459Search in Google Scholar

[40] L. Sarnataro and D. Stryker, Optimal regularity for minimizers of the prescribed mean curvature functional over isotopies, Camb. J. Math. 13 (2025), no. 3, 609–706. 10.4310/CJM.250715011408Search in Google Scholar

[41] R. Schoen and L. Simon, Regularity of stable minimal hypersurfaces, Comm. Pure Appl. Math. 34 (1981), no. 6, 741–797. 10.1002/cpa.3160340603Search in Google Scholar

[42] L. Simon, Lectures on geometric measure theory, Australian National University, Canberra 1983. Search in Google Scholar

[43] A. Song, Existence of infinitely many minimal hypersurfaces in closed manifolds, Ann. of Math. (2) 197 (2023), no. 3, 859–895. 10.4007/annals.2023.197.3.1Search in Google Scholar

[44] A. Song and X. Zhou, Generic scarring for minimal hypersurfaces along stable hypersurfaces, Geom. Funct. Anal. 31 (2021), no. 4, 948–980. 10.1007/s00039-021-00571-7Search in Google Scholar

[45] C. Tucker, Geodesic fibrations of elliptic 3-manifolds, Ph.D. Thesis, The University of Oklahoma, 2013. Search in Google Scholar

[46] T. Wang, Min-max theory for 𝐺-invariant minimal hypersurfaces, J. Geom. Anal. 32 (2022), no. 9, Paper No. 233. 10.1007/s12220-022-00966-4Search in Google Scholar

[47] T. Wang, Min-max theory for free boundary 𝐺-invariant minimal hypersurfaces, Adv. Math. 425 (2023), Article ID 109087. 10.1016/j.aim.2023.109087Search in Google Scholar

[48] Z. Wang and X. Zhou, Existence of four minimal spheres in S 3 with a bumpy metric, preprint (2023), https://arxiv.org/abs/2305.08755. Search in Google Scholar

[49] Z. Wang and X. Zhou, Improved C 1 , 1 regularity for multiple membranes problem, Peking Math. J. (2025), 10.1007/s42543-025-00097-z. 10.1007/s42543-025-00097-zSearch in Google Scholar

[50] B. White, Every three-sphere of positive Ricci curvature contains a minimal embedded torus, Bull. Amer. Math. Soc. (N. S.) 21 (1989), no. 1, 71–75. 10.1090/S0273-0979-1989-15765-0Search in Google Scholar

[51] B. White, The space of minimal submanifolds for varying Riemannian metrics, Indiana Univ. Math. J. 40 (1991), no. 1, 161–200. 10.1512/iumj.1991.40.40008Search in Google Scholar

[52] S. T. Yau, Problem section, Seminar on differential geometry, Ann. of Math. Stud. 102, Princeton University, Princeton (1982), 669–706. 10.1515/9781400881918-035Search in Google Scholar

[53] X. Zhou, Min-max minimal hypersurface in ( M n + 1 , g ) with Ric > 0 and 2 n 6 , J. Differential Geom. 100 (2015), no. 1, 129–160. 10.4310/jdg/1427202766Search in Google Scholar

[54] X. Zhou, On the multiplicity one conjecture in min-max theory, Ann. of Math. (2) 192 (2020), no. 3, 767–820. 10.4007/annals.2020.192.3.3Search in Google Scholar

[55] X. Zhou, Mean curvature and variational theory, ICM—International Congress of Mathematicians. Vol. 4. Sections 5–8, EMS Press, Berlin (2023), 2696–2717. 10.4171/icm2022/27Search in Google Scholar

[56] X. Zhou and J. J. Zhu, Min-max theory for constant mean curvature hypersurfaces, Invent. Math. 218 (2019), no. 2, 441–490. 10.1007/s00222-019-00886-1Search in Google Scholar

[57] X. Zhou and J. J. Zhu, Existence of hypersurfaces with prescribed mean curvature I—generic min-max, Camb. J. Math. 8 (2020), no. 2, 311–362. 10.4310/CJM.2020.v8.n2.a2Search in Google Scholar

Received: 2024-09-26
Revised: 2025-08-27
Published Online: 2025-10-02

© 2025 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 20.10.2025 from https://www.degruyterbrill.com/document/doi/10.1515/crelle-2025-0061/html
Scroll to top button