arXiv:2103.07178v1 [math.AP] 12 Mar 2021 iiy lxnrffFnhlinequalities. Alexandroff-Fenchel bility; ueflw” rn ubrSH 1879/3-1. SCHE number ”Quermassinte Grant Project flows”; ture foundation); research German (DFG, .Itouto n results and Introduction 1. hswr a atal uddfne yte”etceForsc ”Deutsche the by funded funded partially was work This ..Ams osatcrauefunctions curvature constant inequality Almost Heintze-Karcher the results in 4.2. stability Stability sets geometric level the for of 4.1. estimates Proofs umbilicity almost of 4. Proof Notation and 3. Definitions eigenvalues Steklov for 2. Estimates problem overdetermined 1.7. Serrin’s inequalities Alexandroff-Fenchel 1.6. functions Non-convex curvature constant of 1.5. Hypersurfaces inequality Heintze-Karcher 1.4. sets The level and functions 1.3. Defining Stability vs. 1.2. Rigidity 1.1. e od n phrases. and words Key Date problem eigenvalue Steklov References order fourth source a nonlinear in Acknowledgments with Stability problem Serrin’s in 4.5. Stability inequalit Alexandroff-Fenchel non-convex in 4.4. Stability 4.3. ac 5 2021. 15, March : TBLT RMRGDT I UMBILICITY VIA RIGIDITY FROM STABILITY rtradnncne lxnrffFnhlinequalities. Alexandroff-Fenchel non-convex form bi and the space involving erator in problem Steklov theorem Examp a bubble problem, constraint. overdetermined soap geometric Alexandroff’s a pr in given stability sph stability hypersurface, for new ask a and problems of Those old ness many analysis. and of geometry treatment in application arising unified As a hypersurface. for the by allows bounded set open the for h asoffdsac oagoei peeo nebde hyp embedded an estim of an the sphere is geodesic to results a key to distance the Hausdorff of the One spaceforms. of persurfaces Abstract. L 2 nr ftetaeesHsinoeao fadfiigfunctio defining a of operator Hessian traceless the of -norm ecnie ag fgoercsaiiypolm o hy- for problems stability geometric of range a consider We opbbl hoe;Sri’ rbe;Selvpolm St problem; Steklov problem; Serrin’s theorem; bubble Soap UINSCHEUER JULIAN Contents 1 rlpeevn oa curva- local preserving gral Lpaeop- -Laplace t relating ate rclclose- erical hsresult this , hungsgemeinschaft” e include les ,Serrin’s s, ersurface ies oblems n 30 29 29 27 25 22 20 20 15 13 12 11 10 a- 7 6 4 2 2 2 J. SCHEUER
1. Introduction and results 1.1. Rigidity vs. Stability. This paper is about rigidity theorems in clas- sical hypersurface theory and their stability properties. An example is the soap bubble theorem, first proved by Alexandroff: Every constant mean cur- vature hypersurface embedded in Euclidean space, hyperbolic space or in a half-sphere of dimension n + 1, n 2, must be a round sphere, see [Ale62]. Way earlier Liebmann proved the convex≥ case [Lie00] for hypersurfaces of the Euclidean space. Even earlier Jellet [Jel53] proved the two-dimensional case for starshaped surfaces, following up the treatment of surfaces of revolution due to Delaunay [Del41]. A related famous result, which follows from Liebmann’s theorem and the Codazzi equation, is the so-called Nabelpunktsatz seemingly proven by Darboux1: For n 2, every totally umbilic closed, immersed hypersurface of the (n + 1)-dimensional≥ Euclidean space is a round sphere. Both of these theorems are instances of so-called rigidity results: A partic- ular functional defined on a subclass N of the class of closed, immersed and orientable hypersurfacesF M of a givenI Riemannian manifold N dictates their global shape. Abstractly this can be described as follows. Let = f : S R S N G { → | ⊂ } be the set of all functions defined on a subset of N. Then both of the above examples describe particular level sets of a functional : . F IN → G In Liebmann’s theorem we have = M Rn+1 : M closed,h> 0 , (M)= H. IN { ⊂ } F1 Here h is the second fundamental form of M and H its trace with respect to the induced metric. In Alexandroff’s theorem we have = M N : M closed, embedded , IN { ⊂ } where N is Euclidean, hyperbolic or spherical. In the Nabelpunktsatz we instead use (M)= A˚ 2 = A 2 1 H2, F2 | | | | − n where A is the Weingarten operator. Then the above theorems can be phrased as 1 1 (1.1) − ( f = c: c R )= M N : M = sphere = − ( f 0 ). F1 { ∈ } { ⊂ } F2 { ≡ } In the course of this paper, we will encounter many more examples of such functionals. In this abstract context, the next natural question is whether we can describe enlarged preimage sets of . (1.1) characterises the preimages of particular constant functions. We areF interested in the description of sets like 1( f c < ǫ: c R ), for some small ǫ and a norm to be F − {k − k ∈ } k · k 1I did not find a clean reference. STABILITY FROM RIGIDITY VIA UMBILICITY 3 specified. Formally, if 2B is the power set of a set B, then the associated question of stability aims to relax the constancy in the following sense: 1 Is the map − : 2G 2I continuous at all arguments, the values of which F → we can characterise? Of course, the question of continuity has to be addressed with respect to topologies to be specified within particular problems. For example, the stability question for the Nabelpunktsatz would trans- late to whether a hypersurface is close to sphere, whenever it is almost umbilical. This question has received plenty of attention, initiated by early works of Reshetnyak [Res68] and Pogorelov [Pog73] for convex surfaces with a follow up by Leichtweiß, see [Lei99]. The study of almost umbilical hyper- surfaces in the non-convex class seems pioneered by works of De Lellis and M¨uller [DLM05, DLM06], who give a W 2,2-closeness, resp. a C0-closeness in terms of an L2-pinching of A˚ 2 in Rn+1. The W 2,2-closeness result was reproved with the help of the| Willmore| flow in [KS20]. The version for hyperbolic and spherical spaces was treated in [CZ14] and some optimality results in [CJ15]. Further results of almost umbilical type in terms of other Lp-norms were deduced in [Per11, Rot13, Rot15, RS18, Sch15]. In partic- ular we will make use of the following estimate from [RS18, Thm. 1.1] for a closed, connected, oriented and immersed hypersurface M Rn+1 and p>n 2: ⊂ ≥ A˚ α (1.2) dist(M,S ) CR k kp , R ≤ H α k kp with constants α = α(n,p) and C = C(n,p, A p, M ). Here p is the Lp-norm with respect to the surface area measure.k k | | k · k For the soap bubble theorem most of the stability results available con- trol the Hausdorff-distance of the hypersurface to a sphere (or an array of spheres) in terms of various pinching assumptions on the mean curvature. In the class of convex hypersurfaces such results are pioneered by Koutroufiotis [Kou71] and Moore [Moo73], also see [Arn93] for a rather convex geometric approach. In the non-convex class the geometry of the stability problem is more complicated, since bubbling (closeness to a connected sum of spheres with small necks) can occur in case the curvature of the hypersurfaces is un- controlled. In terms of a C0-pinching condition, stability results in the Eu- clidean and hyperbolic space were proven by Ciraolo/Vezzoni [CV18, CV20], where in both cases the authors used a quantitative analysis of Alexandroff’s proof by moving planes. In [CM17] similar stability results were obtained, but with a proof that relies on the stability analysis of Ros’ proof of the soap bubble theorem via the Heintze-Karcher inequality [Ros87]. In parallel to these developments, various stability results by Magnani- ni/Poggesi appeared relaxing the C0-pinching condition to an L1-pinching condition. The key is, that the soap bubble theorem can also be proved by integral methods, see [Ros87] and also compare [Rei82]. With the help 4 J. SCHEUER of integral formulae, which are valid for solutions f of the torsion Dirichlet problem ∆¯ f =1 inΩ f =0 on ∂Ω, Magnanini/Poggesi [MP19, MP20a, MP20b] were able to prove that what they call Cauchy-Schwarz deficit ˚¯ 2f 2 := ¯ 2f 2 1 (∆¯ f)2 |∇ | |∇ | − n+1 can be controlled by the L1 deficit of the mean curvature with respect to a constant. Comparison of f with the squared distance function gave closeness of ∂Ω to a sphere after employing various estimates for harmonic functions. The key idea of the present paper is that it is possible to control the distance to a sphere by an L2-norm of the Cauchy-Schwarz deficit, without relying on f being a solution to a PDE. This flexibility will allow us to treat a broader range of stability results, such as an extension of the results by Magnanini/Poggesi to spaceforms of constant curvature, stability results for the deficit in the Heintze-Karcher inequality in spaceforms, a double- stability result for Serrin’s overdetermined problem with nonlinear source term, as well as a stability result in bi-Laplace Steklov eigenvalue problems. Everything comes with a price: The cost of this flexibility is that in par- ticular geometric problems, some constants may be improved with methods tailor-made for the problem at hand. Let’s continue this introduction with a precise description of the results and more background information on the specific geometric problems. 1.2. Defining functions and level sets. A key theorem in our approach, which can be viewed as a generalisation and quantitative version of a result due to Reilly [Rei80, Lemma 3], that a function f on a domain Ω Rn+1, which satisfies for some constant L ⊂ ¯ 2 f = Lg,¯ f ∂Ω = 0, ∇ | is the squared distance function to a suitable point up to normalisation constants, is as follows: 1.1. Theorem. Let n 2 and let (N n+1, g¯) be a smooth conformally flat Riemannian domain, i.e.≥ N Rn+1 is open and ⊂ g¯ = e2ψ , h· ·i 2 with ψ C∞(N). Let Ω ⋐ N be an open set with C -boundary and let f C2(∈Ω)¯ satisfy ∈ ¯ (1.3) f ∂Ω = 0, f ∂Ω = 0. | ∇ | 6 Then for every connected component M of ∂Ω there exist constants α = α(n) and 1 2 1 M n ¯ f ,Ω C = C n, ψ ,Ω, M n ¯ ψ ,Ω, | | k∇ k∞ , k k∞ | | k∇ k∞ min ¯ f M |∇ | ! STABILITY FROM RIGIDITY VIA UMBILICITY 5 such that α 1 1 2 n+1 dist(M,S) C M n ˚¯ f ≤ | | min ¯ f n+1 ˆ |∇ | M |∇ | Ω for a totally umbilical hypersurface S N. ⊂ Here ¯ is the Levi-Civita connection ofg ¯, M is the surface area of M, ∇ | | ,Ω the sup-norm and dist the Hausdorff distance of sets. Ω ⋐ N means k · k∞ that Ω¯ is compact and Ck(Ω)¯ is the usual space of k-times differentiable functions in Ω where all derivatives are continuous up to the boundary. A function with the properties (1.3) is called defining function for Ω in the sequel. In many applications one has an estimate on the L2-norm of the traceless Hessian, hence we also state the following corollary. 1.2. Corollary. Under the assumptions of Theorem 1.1, for every connected component M of ∂Ω there exist constants α = α(n) and 1 2 1 M n ¯ f ,Ω C = C n, ψ ,Ω, M n ¯ ψ ,Ω, | | k∇ k∞ , k k∞ | | k∇ k∞ min ¯ f M |∇ | ! such that − − α 1 α(n 1) 1 2 2 dist(M,S) C M n ˚¯ f ≤ | | min ¯ f 2 ˆ |∇ | M |∇ | Ω for a totally umbilical hypersurface S. Theorem 1.1 will be proven with the help of an almost umbilical type theorem [RS18] in Section 3. The key feature of Corollary 1.2 is the following procedure: Whenever a rigidity result is proven by discarding an L2-norm of the Cauchy-Schwarz deficit of a defining function for Ω, you get an associated stability result. We will see plenty of examples throughout the following discussion and there are many more slumbering in the literature. 1.3. Remark. (i) Note that results like the one above can not hold without some dependence of the constant C on the Hessian of f, due to the possibility of bubbling, see for example [CV18, Rem. 5.2] for further details. Hence all of the stability results in this paper will come with some sort of curvature dependence. This also means that the choice of the Ln+1-norm in Theorem 1.1 is not essential and can be replaced by any Lp norm of the traceless Hessian. For scaling reasons the formu- lation in terms of the Ln+1-norm gives the most aesthetic form. (ii) The proof of this theorem crucially depends on the almost-umbilicity type estimate (1.2), which holds for a large class of hypersurfaces. In case that more properties are being given, such as convexity or mean- convexity, one can say more about the value of α, see for example [Lei99] for estimates in the convex class. 6 J. SCHEUER
(iii) Note that in the Euclidean space one may scale the problem to M = min ¯ f = 1 | | M |∇ | whenever convenient. 1.3. The Heintze-Karcher inequality. The following inequality, known today as the Heintze-Karcher inequality, was proved by Ros [Ros87] upon inspiration by related inequalities of Heintze/Karcher [HK78]. It goes as follows. For a closed and mean convex hypersurface M = ∂Ω Rn+1 we have ⊂ 1 (n + 1) Ω = u, ˆM H1 ≥ | | ˆM with equality precisely when M is a round sphere. Here u is the support function u = x,ν h i and nH1 = H. In particular this trivially yields a new proof of Alexandroff’s theorem, after invoking the formula
M = uH1. | | ˆM As to the soap bubble theorem, there was of course nothing new to be discovered from that method, however, Ros [Ros87] used this method to prove that constant Hk hypersurfaces must be spheres. We will recall that proof in Section 4 in the course of proving the stability result. In fact, Ros proved this inequality in ambient spaces of non-negative Ricci curvature, however the equality case forces the ambient space to be Euclidean. For constant curvature spaceforms, refined versions of the Heintze-Karcher in- equality have emerged during the recent years. Using flow methods, Brendle [Bre13] proved that for closed and mean-convex hypersurfaces M = ∂Ω of the half-sphere or the hyperbolic space there holds ϑ ′ u, ˆM H1 ≥ ˆM with equality precisely on geodesic balls. Here
u =g ¯(ϑ∂r,ν) and the metric of the ambient space is written in polar coordinates as g¯ = dr2 + ϑ2(r)σ, with the round metric σ. This result was, among others, recovered by Qiu/Xia [QX15] via elliptic methods, i.e. by employing a new Reilly type formula, see (4.2) for further details. Similarly as above, those inequalities can be used to reprove Alexandroff’s theorem in spaceforms. But also on other occasions the Heintze-Karcher inequality has proven useful, for ex- ample for the deduction of isoperimetric type inequalities in Riemannian manifolds, see for example [SX19]. STABILITY FROM RIGIDITY VIA UMBILICITY 7
Following the previous discussion, it becomes an interesting question whether we have an associated stability result. Namely, does almost-equality of the Heintze-Karcher inequality imply any type of closeness to the sphere? For the Euclidean case there are the recent works of Magnanini/Poggesi [MP19, MP20a]. They prove that for a C2-hypersurface M = ∂Ω there holds τn 1 2 dist(M,S) C (n + 1) Ω , ≤ ˆ H − | | M 1 for some τn and with a constant C depending on n, a lower inball radius bound for Ω at M and on the diameter of Ω. In Section 4 we will prove the following theorem for hypersurfaces of non-Euclidean spaceforms MK of n+1 sectional curvature K, where M1 = S+ is the open northern hemisphere. The proof applies to K = 0 too, but that result would not be new. In the following, if a constant depends on maxM r, in case K = 1 this means the constant may blow up at the equator r = π/2 . { } n 1.4. Theorem. Let n 2, β > 0 and M MK , K = 1, be a closed, embedded and strictly mean≥ convex C2,β-hypersurface⊂ with uniform± interior sphere condition with radius ρ. Then there exist constants α = α(n) and 1 C = C(n, β, ρ− , M , max r M , A β,M ), such that | | | | | α 1−α(n−1) ϑ′ dist(M,S) C M n u , ≤ | | ˆ H − ˆ M 1 M for a suitable geodesic sphere S. Here A is the H¨older norm of the Weingarten operator on M. | |β,M 1.4. Hypersurfaces of constant curvature functions. In [Ale62], Alex- androff did not only prove the soap bubble theorem. The robustness of the moving plane method allowed to treat much more general elliptic equations, also see [Vos56] and [Hsi56] as well as [Bre13, DM19, KLP18, Mon99, WX14] fur further developments. His theorem says that every embedded hypersur- face with
(1.4) F (κ1,...,κn) = const is a sphere. Here F is a strictly monotone function of the principal curvatures κi of the hypersurface. Particular important examples are the higher order mean curvatures 1 Hk = n κi1 ...κik ,H0 := 1, k 1 i1<