betaFeedback

WorksMathematics

Integrability of the horizontal mean curvature near characteristic points in the Heisenberg group

Abstract

We investigate the local integrability of the horizontal mean curvature near characteristic points of surfaces in the first Heisenberg group. We prove that, for every embedded surface of class C2,αC^{2,α}, with α>0α>0, the horizontal mean curvature is locally integrable with respect to (any) Riemannian surface measure, without any assumption on the structure of the characteristic set. We further establish that the horizontal unit normal belongs to BVlocBV_{\mathrm{loc}} and characterize its Wloc1,1W^{1,1}_{\mathrm{loc}} regularity in terms of the vanishing one-dimensional Hausdorff measure of the characteristic set. Finally, we answer an open question by Danielli, Garofalo and Nhieu by constructing a C2C^2 surface with a unique characteristic point whose horizontal mean curvature fails to be locally integrable, thus showing that the additional Hölder regularity cannot, in general, be replaced by C2C^2 regularity.

Introduction

The horizontal mean curvature is one of the basic geometric quantities associated with hypersurfaces in sub-Riemannian manifolds. It enters the first-variation theory for horizontal perimeter and, in particular, the study of minimal and constant-mean-curvature surfaces [9, 10, 13, 14, 16, 17, 18] and references therein. It also appears in problems of a rather different nature, including short-time heat content asymptotics [1, 15, 19, 21] and tube-volume expansions [3, 5, 7, 8]. A recurring difficulty in these questions is that the natural curvature quantities, although well defined away from characteristic points, need not remain bounded as one approaches the characteristic set. Understanding their integrability is therefore a preliminary step in extending geometric identities beyond the non-characteristic setting.

In this paper, we focus on embedded surfaces in the first Heisenberg group H\mathbb{H}. Recall that a point pp of a surface SS is characteristic if the horizontal distribution coincides with the tangent plane TpST_pS. We denote the set of such points by C(S)C(S). If ff is a local defining function near p∈Sp\in S, namely there is an open neighborhood UU of pp such that f∣Uf|_U is a submersion and S∩U=f−1(0)∩US\cap U=f^{-1}(0)\cap U, then the horizontal unit normal on U∩(S∖C(S))U\cap(S\setminus C(S)) is

ν:=∇Hf∥∇Hf∥,\nu:= \frac{\nabla_H f}{\lVert\nabla_H f\rVert},

where ∇Hf\nabla_H f denotes the horizontal gradient of ff, and its divergence defines the horizontal mean curvature H\mathcal{H}, up to a sign. The normalization is precisely where the difficulty arises: the denominator vanishes on C(S)C(S), even if ff and SS are smooth and we do not have an immediate integrable bound on the derivatives of ν\nu.

It is important here to distinguish the two measures naturally associated with SS. For a compatible Riemannian extension of the horizontal metric, the horizontal perimeter measure σH\sigma_H has, away from characteristic points, the form σH=WσR\sigma_H=W\sigma_R, where WW is the length of the horizontal projection of the Riemannian unit normal and σR\sigma_R is the induced Riemannian surface measure. The factor WW compensates for the singular normalization and thus H\mathcal{H} is locally integrable with respect to σH\sigma_H, see [11, 20]. No analogous cancellation is available when σH\sigma_H is replaced by σR\sigma_R. The latter problem is more delicate, and its answer depends both on the regularity of the surface and on how its horizontal normal vanishes.

This distinction was emphasized by Danielli, Garofalo and Nhieu [11]. They constructed a C2C^2 surface in H\mathbb{H}, with C(S)C(S) being a line and whose horizontal mean curvature is not locally integrable with respect to Riemannian surface measure, while establishing integrability in several examples with isolated characteristic points. They conjectured that, if SS is C2C^2 and C(S)C(S) consists of isolated points, H\mathcal{H} is locally integrable. Subsequent results identified geometric situations in which the answer is positive: nondegenerate characteristic points, certain mildly degenerate points, and, in particular, isolated characteristic points of real-analytic surfaces [20]. More recent work extends these finite-order conditions to the framework of 3D contact manifolds [4]; see also [5, 6] for related developments. These results show how the geometry of characteristic points may play a role in the integrability of H\mathcal{H}, but they leave open a different issue: can one obtain a general integrability theorem without prescribing the geometry, dimension, or local structure of C(S)C(S)?

The purpose of this paper is to address that issue by replacing assumptions on the characteristic set with a quantitative regularity assumption on the surface. Our first main result asserts that any positive Hölder regularity beyond C2C^2 is sufficient.

Theorem 1.1. Let 0<α≤10 < \alpha\le1, and let S⊂H1S \subset\mathbb{H}^1 be an embedded surface without boundary of class C2,αC^{2,\alpha}. Then

H∈Lloc1(S,σR),\mathcal{H} \in L^1_{\mathrm{loc}}(S,\sigma_R),

for every Riemannian surface measure σR\sigma_R induced by a smooth ambient Riemannian metric.

Note that the latter statement includes noncompact and nonorientable surfaces. In fact, the proof yields information that is strictly finer than the integrability of the horizontal mean curvature. Since H\mathcal{H} is a trace of derivatives of the horizontal unit normal, it is natural to ask whether that normal admits a weak derivative across the characteristic set. The answer involves a distinction between bounded variation and Sobolev regularity. The former holds for every C2,αC^{2,\alpha} surface, whereas the latter is governed exactly by the one-dimensional size of the characteristic set.

Theorem 1.2. Let 0<α≤10 < \alpha\le1, and let S⊂H1S \subset\mathbb{H}^1 be an embedded surface without boundary of class C2,αC^{2,\alpha}. For every p∈Sp \in S, let U⊂HU \subset\mathbb{H} be an open neighborhood of pp and f∈C2,α(U)f \in C^{2,\alpha}(U) a defining function near pp. Then,

ν∈BV⁡loc(S∩U;R3).\nu\in\operatorname{BV}_{\mathrm{loc}}(S \cap U;\mathbb{R}^3).

Moreover,

ν∈Wloc1,1(S∩U;R3)⟺H1(C(S)∩U)=0.\nu\in W^{1,1}_{\mathrm{loc}}(S \cap U;\mathbb{R}^3) \Longleftrightarrow \mathcal{H}^1(C(S) \cap U) = 0.

In the above, the functional spaces and the one-dimensional Hausdorff measure H1\mathcal{H}^1 are those defined using any smooth Riemannian metric on SS. In particular, the statement is independent of the chosen Riemannian metric.

Theorem 1.2 therefore identifies the precise obstruction to W1,1W^{1,1} regularity: a characteristic set of positive length produces a nontrivial jump in the normalized horizontal normal. In particular, the horizontal mean curvature may be integrable even though the horizontal unit normal is not a Sobolev map. This separation between integrability of a curvature quantity and weak differentiability of the normal is one of the main features of the sub-Riemannian situation considered here. Additionally, as a byproduct of the proof, we obtain that the distributional derivative of ν\nu has vanishing Cantor part.

A second important aspect is the regularity threshold. The Hölder assumption in Theorem 1.1 is sharp and cannot simply be dropped. Indeed, in Section 7 we construct an explicit C2C^2 surface with exactly one characteristic point for which

H∉Lloc1(S,σR).\mathcal{H} \notin L^1_{\mathrm{loc}}(S,\sigma_R).

Consequently, the conjecture formulated in [11] for arbitrary C2C^2 surfaces with isolated characteristic points fails in general. Combined with Theorem 1.1, this example gives a sharp distinction between the C2C^2 class and the classes C2,αC^{2,\alpha}, α>0\alpha> 0.

Acknowledgments:

T.R. gratefully acknowledges the University of Trento for its hospitality and scientific environment. The authors are members of the Istituto Nazionale di Alta Matematica (INdAM), Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA).

Preliminaries

Consider the Lie group (R3,⋆)(\mathbb{R}^{3},\star), where the group law is

(x,y,t)⋆(x′,y′,t′)=(x+x′,y+y′,t+t′+12(xy′−yx′)),(x,y,t),(x′,y′,t′)∈R3.(1)(x,y,t)\star(x',y',t')=\left(x+x',y+y',t+t'+\frac{1}{2}(xy'-yx')\right),\qquad(x,y,t),(x',y',t')\in\mathbb{R}^{3}. \tag*{(1)}

We endow this Lie group with the sub-Riemannian structure (D,g)(\mathcal{D},g), where D=span⁡{X1,X2}\mathcal{D}=\operatorname{span}\{X_{1},X_{2}\}, with X1,X2X_{1},X_{2} left-invariant vector fields defined as

X1=∂x−y2∂t,X2=∂y+x2∂t,(2)X_{1}=\partial_{x}-\frac{y}{2}\partial_{t},\qquad X_{2}=\partial_{y}+\frac{x}{2}\partial_{t}, \tag*{(2)}

and gg is obtained by declaring X1,X2X_{1},X_{2} orthonormal. The resulting sub-Riemannian manifold is the first Heisenberg group, denoted with H\mathbb{H}. We say that a curve γ:[0,T]→H\gamma:[0,T]\to\mathbb{H} is horizontal if it is absolutely continuous and for a.e. t∈[0,T]t\in[0,T],

γ˙(t)∈Dγ(t).\dot{\gamma}(t)\in\mathcal{D}_{\gamma(t)}.

For such curves we define the length as ℓ(γ)=∫0T∥γ˙(t)∥g dt\ell(\gamma)=\int_{0}^{T}\lVert\dot{\gamma}(t)\rVert_{g}\,\mathrm{d}t. The sub-Riemannian distance dSRd_{SR} between two points p,q∈Hp,q\in\mathbb{H} is defined as

dSR(p,q)=inf⁡{ℓ(γ):γ∈Cp,q}(3)d_{SR}(p,q)=\inf\{\ell(\gamma):\gamma\in\mathcal{C}_{p,q}\} \tag*{(3)}

where Cp,q\mathcal{C}_{p,q} denotes the set of horizontal curves connecting p,qp,q. By Chow-Rashevsky theorem, dSRd_{SR} is finite, continuous and induces the topology of R3\mathbb{R}^{3}.

Characteristic points

Let S⊂HS\subset\mathbb{H} be a C1C^{1} embedded hypersurface. We say that p∈Sp\in S is a characteristic point if

Dp=TpS.(4)\mathcal{D}_{p}=T_{p}S. \tag*{(4)}

We denote by C(S):={p∈S:Dp=TpS}C(S):=\{p\in S:\mathcal{D}_{p}=T_{p}S\} the set of characteristic points. Note that C(S)⊂SC(S)\subset S is a closed set. Given a surface S⊂HS\subset\mathbb{H} and a point p∈Sp\in S, we say that f∈C1(H)f\in C^{1}(\mathbb{H}) is a locally defining function for SS near pp if there exists an open neighborhood U⊂HU\subset\mathbb{H} of pp such that f∣Uf|_{U} is a submersion and S∩U=f−1(0)∩US\cap U=f^{-1}(0)\cap U. In such case, defining the horizontal gradient of f∈C1(H)f\in C^{1}(\mathbb{H}) as the horizontal (i.e. tangent to D\mathcal{D}) vector field

∇Hf:=X1fX1+X2fX2,(5)\nabla_{H}f:=X_{1}fX_{1}+X_{2}fX_{2}, \tag*{(5)}

we have that p∈C(S)p\in C(S) if and only if

∇Hf(p)=0⟺X1f(p)=X2f(p)=0.(6)\nabla_{H}f(p)=0\Longleftrightarrow X_{1}f(p)=X_{2}f(p)=0. \tag*{(6)}

In particular, outside C(S)C(S), we may define the (local) horizontal unit normal to SS as

ν(q):=∇Hf(q)∥∇Hf(q)∥,∀q∈(S∖C(S))∩U.(7)\nu(q):=\frac{\nabla_{H}f(q)}{\lVert\nabla_{H}f(q)\rVert},\qquad\forall q\in(S\setminus C(S))\cap U. \tag*{(7)}

Note that, by (6), ν\nu is well-defined outside C(S)C(S). In addition, from its definition, it follows that ν\nu is transversal to SS and g(ν,X)∣q=0g(\nu,X)|_{q}=0 with X∈Dq∩TqSX\in\mathcal{D}_{q}\cap T_{q}S, for every $q\in(S\setminus C(S))\cap U.

Perimeter measures

Let p∈Sp \in S and f∈C1(H)f \in C^{1}(\mathbb{H}) be a local defining function for SS near pp. Let ν\nu be the horizontal unit normal to SS. Then the horizontal perimeter measure σH\sigma_H on SS is the smooth measure with density on U\mathcal{U} given by ∣ıνL3∣\lvert\imath_{\nu} \mathcal{L}^{3}\rvert, where ı\imath is the contraction of forms. Analogously, to define the Riemannian perimeter measure σR\sigma_R, consider any Riemannian metric on R3\mathbb{R}^{3}, then replace ν\nu with the Riemannian unit normal, which is given by (7) defined with the Riemannian gradient. Note that σR\sigma_R coincides with the 2-dimensional Hausdorff measure on SS induced by the Riemannian structure.

By the results of [2], σH(C(S))=0\sigma_H(C(S)) = 0, and if SS is of class C2C^{2}, σR(C(S))=0\sigma_R(C(S)) = 0 for any Riemannian perimeter measure. We refer to [2, 12] for further results about the size of C(S)C(S).

Horizontal mean curvature

Let S⊂HS \subset\mathbb{H} be a C2C^{2} embedded surface. Recall that the divergence of a C1C^{1} vector field X∈Vec⁡(TH)X \in\operatorname{Vec}(T\mathbb{H}) is defined by the identity

div⁡(X) dx∧dy∧dt=LX(dx∧dy∧dt),(8)\operatorname{div}(X)\, \mathrm{d}x \wedge\mathrm{d}y \wedge\mathrm{d}t = \mathcal{L}_{X}(\mathrm{d}x \wedge\mathrm{d}y \wedge\mathrm{d}t), \tag*{(8)}

where LX\mathcal{L}_{X} denotes the Lie derivative in the direction of XX. Then, the horizontal mean curvature at q∈S∖C(S)q \in S \setminus C(S) is defined as

H(q):=div⁡(ν)∣q=div⁡(∇Hf∥∇Hf∥)∣q=X1(X1f∥∇Hf∥)∣q+X2(X2f∥∇Hf∥)∣q,(9)\mathcal{H}(q) := \left.\operatorname{div}(\nu)\right|_{q} = \left.\operatorname{div}\left(\frac{\nabla_{H}f}{\lVert\nabla_{H}f\rVert}\right)\right|_{q} = \left.X_{1}\left(\frac{X_{1}f}{\lVert\nabla_{H}f\rVert}\right)\right|_{q} + \left.X_{2}\left(\frac{X_{2}f}{\lVert\nabla_{H}f\rVert}\right)\right|_{q}, \tag*{(9)}

where f∈C2(H)f \in C^{2}(\mathbb{H}) is a local defining function for SS near qq. Note that, up to a sign, the value of H\mathcal{H} does not depend on the choice of ff. In addition, H∈C(S∖C(S))\mathcal{H} \in C(S \setminus C(S)) and locally integrable on every compact set K⊂SK \subset S satisfying K∩C(S)=∅K \cap C(S) = \varnothing. However, H\mathcal{H} is not well-defined at characteristic points and it may blow up as pp approaches C(S)C(S). In the sequel, we study the integrability of H\mathcal{H} with respect to a Riemannian perimeter measure on SS. We stress that the choice of the Riemannian metric does not play any role concerning the integrability problem.

A change of variable

Let S⊂HS \subset\mathbb{H} be a surface of class C2,αC^{2,\alpha}, where α∈(0,1]\alpha\in(0,1] and let p∈C(S)p \in C(S). Up to a left-translation, which is an isometry of H\mathbb{H}, we may assume that p=(0,0,0)p=(0,0,0). Then, a local defining function ff for SS near pp must be the tt-graph of a function g∈C2,α(R2)g \in C^{2,\alpha}(\mathbb{R}^{2}), i.e. f(x,y,t):=t−g(x,y)f(x,y,t):=t-g(x,y). Hence, there is a neighborhood U\mathcal{U} of pp such that

S∩U={(x,y,t)∈H:t=g(x,y)}∩U,g(0)=gx(0)=gy(0)=0.S \cap\mathcal{U}=\{(x,y,t)\in\mathbb{H}:t=g(x,y)\}\cap\mathcal{U}, \qquad g(0)=g_x(0)=g_y(0)=0.

Note that, since f(x,y,t)=t−g(x,y)f(x,y,t)=t-g(x,y), its horizontal gradient is

∇Hf(q)=(−gx−y2)X1(q)+(−gy+x2)X2(q),∀q=(x,y,t)∈H,(10)\nabla_{H}f(q)=\left(-g_x-\frac{y}{2}\right)X_1(q)+\left(-g_y+\frac{x}{2}\right)X_2(q), \qquad\forall q=(x,y,t)\in\mathbb{H}, \tag*{(10)}

namely X1fX_1f and X2fX_2f do not depend on tt. Thus, we can reduce the problem of integrability of H\mathcal{H} to a problem in R2\mathbb{R}^{2}. More precisely, define the vector field V∈C1,α(R2;R2)V \in C^{1,\alpha}(\mathbb{R}^{2};\mathbb{R}^{2}) by

V(x,y):=(−gx−y2,−gy+x2),∀(x,y)∈R2,(11)V(x,y):=\left(-g_x-\frac{y}{2},-g_y+\frac{x}{2}\right), \qquad\forall(x,y)\in\mathbb{R}^{2}, \tag*{(11)}

and, setting N:={(x,y)∈R2:V(x,y)≠0}\mathcal{N}:=\{(x,y)\in\mathbb{R}^{2}:V(x,y)\ne0\}, define the vector field N∈Cloc1,α(N;R2)N\in C^{1,\alpha}_{\mathrm{loc}}(\mathcal{N};\mathbb{R}^{2}) as

N(x,y):=V(x,y)∣V(x,y)∣,∀(x,y)∈N,(12)N(x,y):=\frac{V(x,y)}{\lvert V(x,y)\rvert}, \qquad\forall(x,y)\in\mathcal{N}, \tag*{(12)}

where ∣⋅∣\lvert\cdot\rvert denotes the standard Euclidean norm of R2\mathbb{R}^{2}. From (10), we see that ∣V∣=∥∇Hf∥\lvert V\rvert=\lVert\nabla_Hf\rVert and C(S)∩U={(x,y,t)∈S∩U:V(x,y)=0}C(S)\cap\mathcal{U}=\{(x,y,t)\in S\cap\mathcal{U}:V(x,y)=0\}, hence, in particular, V(0)=0V(0)=0. Moreover, the (Euclidean) differential of VV is

DV(0)=(−gxx(0)−gxy(0)−12−gxy(0)+12−gyy(0)),DV(0)=\begin{pmatrix} -g_{xx}(0) & -g_{xy}(0)-\frac{1}{2}\\ -g_{xy}(0)+\frac{1}{2} & -g_{yy}(0) \end{pmatrix},

which is always non-zero, since by the commutator relation ∂xV2−∂yV1=[X1,X2]f=1\partial_xV_2-\partial_yV_1=[X_1,X_2]f=1. This means that rk⁡(DV(0))≥1\operatorname{rk}(DV(0))\ge1.

Remark 3.1. Given p=(0,0,0)∈C(S)p=(0,0,0)\in C(S), if the rank of DV(0)DV(0) is maximal, pp is a non-degenerate (and thus isolated) characteristic point. If instead the rank of DV(0)DV(0) is 1, then pp is a degenerate characteristic point. We refer to [6, 4, 5] for further details around degenerate and non-degenerate characteristic points. The integrability problem around non-degenerate characteristic points is well-understood, see e.g. [20]. Around degenerate characteristic points, only partial results were known, see e.g. [20, 4].

Using the identity ∣V∣=∥∇Hf∥|V|=\|\nabla_{H}f\|, we deduce that the horizontal unit normal to SS satisfies

ν(x,y,t)=N1(x,y)X1(x,y,t)+N2(x,y)X2(x,y,t),∀ (x,y,t)∈(S∖C(S))∩U(13)\nu(x,y,t)=N_{1}(x,y)X_{1}(x,y,t)+N_{2}(x,y)X_{2}(x,y,t),\qquad\forall\,(x,y,t)\in(S\setminus C(S))\cap U \tag*{(13)}

where N=(N1,N2)N=(N_{1},N_{2}). Thus, by its definition (9), HH is independent of tt and it holds

H(x,y,t)=Tr⁡(DN(x,y))=∂xN1(x,y)+∂yN2(x,y),∀ (x,y,t)∈(S∖C(S))∩U.(14)H(x,y,t)=\operatorname{Tr}(DN(x,y))=\partial_{x}N_{1}(x,y)+\partial_{y}N_{2}(x,y),\qquad\forall\,(x,y,t)\in(S\setminus C(S))\cap U. \tag*{(14)}

Here DNDN is the classical derivative of NN and it is well-defined whenever V≠0V\ne0. This proves the following lemma.

Lemma 3.2. Let S⊂HS\subset\mathbb{H} be a surface of class C2,αC^{2,\alpha}, where α∈(0,1]\alpha\in(0,1] and let p=(0,0,0)∈C(S)p=(0,0,0)\in C(S). Let f(x,y,t)=t−g(x,y)f(x,y,t)=t-g(x,y) be a local defining C2,αC^{2,\alpha}-function for SS on a neighborhood UU of pp. Then, there exists a constant C>0C>0 such that

∣H(x,y,t)∣≤C∣DN(x,y)∣,∀ (x,y,t)∈(S∖C(S))∩U|H(x,y,t)|\le C|DN(x,y)|,\qquad\forall\,(x,y,t)\in(S\setminus C(S))\cap U

where NN is defined in (12) and ∣DN∣|DN| denotes any (sub-multiplicative) matrix norm of DNDN.

The integrability of HH (with respect to a Riemannian perimeter measure) is therefore implied by the integrability of ∣DN∣|DN| (with respect to the standard Lebesgue measure on R2\mathbb{R}^{2}). The next key lemma says that the integrability of the latter matrix norm is controlled by the behavior of the vector field V~(u,v)=(u,b(u,v))\widetilde{V}(u,v)=(u,b(u,v)) for some function b∈C1,αb\in C^{1,\alpha}.

Lemma 3.3. Under the assumptions of Lemma 3.2, there exist an orthogonal transformation R∈O(2)R\in O(2) and a map Φ:R2→R2\Phi:\mathbb{R}^{2}\to\mathbb{R}^{2} and an open, relatively compact neighborhood of the origin U⊂R2U\subset\mathbb{R}^{2} such that

(i) Φ(0)=0\Phi(0)=0 and Φ∣U:U→Φ(U)\Phi|_{U}:U\to\Phi(U) is a C1,αC^{1,\alpha}-diffeomorphism. Moreover, defining V~(u,v):=RV(Φ−1(u,v))\widetilde{V}(u,v):=RV(\Phi^{-1}(u,v)), for every (u,v)∈Φ(U)(u,v)\in\Phi(U), we have

V~(u,v)=(u,b(u,v)),\widetilde{V}(u,v)=(u,b(u,v)),

for some b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^{2}), with b(0)=0b(0)=0;

(ii) there is a constant C>0C>0 such that

∫U∣DN∣ dx dy≤C∫Φ(U)∣DN~∣ du dv,N~:=V~∣V~∣.\int_{U}|DN|\,dx\,dy\le C\int_{\Phi(U)}|D\widetilde{N}|\,du\,dv,\qquad\widetilde{N}:=\frac{\widetilde{V}}{|\widetilde{V}|}.

Proof. (i) Since rk⁡(DV(0))≥1\operatorname{rk}(DV(0))\ge1, the image of DV(0)DV(0) is at least one-dimensional. Let w∈Im⁡(DV(0))w\in\operatorname{Im}(DV(0)), ∣w∣=1|w|=1, and choose R∈O(2)R\in O(2) such that Rw=e1Rw=e_{1}. Thus, defining Z:=RV∈C1,α(R2)Z:=RV\in C^{1,\alpha}(\mathbb{R}^{2}), the image of DZ(0)DZ(0) contains e1e_{1} and, after possibly applying the smooth diffeomorphism (x,y)↦(y,x)(x,y)\mapsto(y,x), we may assume that ∂xZ1(0)≠0\partial_{x}Z_{1}(0)\ne0. Hence, the map

Φ:R2→R2,Φ(x,y)=(Z1(x,y),y)\Phi:\mathbb{R}^{2}\to\mathbb{R}^{2},\qquad\Phi(x,y)=(Z_{1}(x,y),y)

is of C1,αC^{1,\alpha} class, fixes the origin and DΦ(0)D\Phi(0) is invertible, since

DΦ(0)=(∂xZ1(0)∂yZ1(0)01).D\Phi(0)= \begin{pmatrix} \partial_{x}Z_{1}(0) & \partial_{y}Z_{1}(0)\\ 0 & 1 \end{pmatrix}.

As a consequence, there is a relatively compact neighborhood U⊂R2U\subset\mathbb{R}^{2} of the origin such that Φ∣U\Phi|_{U} defines a C1,αC^{1,\alpha}-diffeomorphism onto its image. Hence, letting (u,v)=Φ(x,y)∈Φ(U)(u,v)=\Phi(x,y)\in\Phi(U), we explicitly compute

V~(u,v)=V~(Φ(x,y))=Z(x,y)=(Z1(x,y),Z2(x,y))=(u,b(u,v)),(15)\widetilde{V}(u,v)=\widetilde{V}(\Phi(x,y))=Z(x,y)=(Z_{1}(x,y),Z_{2}(x,y))=(u,b(u,v)), \tag*{(15)}

where b∈C1,α(R2)b \in C^{1,\alpha}(\mathbb{R}^{2}). Moreover, since p=(0,0,0)∈C(S)p=(0,0,0)\in C(S), RV(0)=Z(0)=0RV(0)=Z(0)=0, meaning that b(0)=0b(0)=0, as claimed.

(ii) Note that R∈O(2)R\in O(2), ∣Z∣=∣RV∣=∣V∣|Z|=|RV|=|V|. Hence, it holds

N~∘Φ(x,y)=Z(x,y)∣Z(x,y)∣=RN(x,y),∀(x,y)∈U\widetilde{N}\circ\Phi(x,y)=\frac{Z(x,y)}{|Z(x,y)|}=RN(x,y),\qquad\forall(x,y)\in U

or equivalently, N~=R(N∘Φ−1)\widetilde{N}=R(N\circ\Phi^{-1}) on Φ(U)\Phi(U). As a consequence, by the chain rule, we have

∣DN~∣=∣D(R(N∘Φ−1))∣=∣(DN)∘Φ−1⋅DΦ−1∣,on Φ(U).|D\widetilde{N}|=|D(R(N\circ\Phi^{-1}))|=|(DN)\circ\Phi^{-1}\cdot D\Phi^{-1}|,\qquad\text{on }\Phi(U).

Thus, after the change of variable (u,v)=Φ(x,y)(u,v)=\Phi(x,y), the integral of ∣DN∣|DN| over UU becomes

∫U∣DN∣ dx dy=∫Φ(U)∣(DN)∘Φ−1∣⋅∣det⁡(DΦ−1)∣ du dv≤∫Φ(U)∣(DN)∘Φ−1⋅DΦ−1∣⋅∣(DΦ)∘Φ−1∣⋅∣det⁡(DΦ−1)∣ du dv=∫Φ(U)∣DN~∣⋅∣(DΦ)∘Φ−1∣⋅∣det⁡(DΦ−1)∣ du dv,\begin{aligned} \int_U |DN|\,dx\,dy &=\int_{\Phi(U)} |(DN)\circ\Phi^{-1}|\cdot|\det(D\Phi^{-1})|\,du\,dv\\ &\leq\int_{\Phi(U)} |(DN)\circ\Phi^{-1}\cdot D\Phi^{-1}|\cdot|(D\Phi)\circ\Phi^{-1}|\cdot|\det(D\Phi^{-1})|\,du\,dv\\ &=\int_{\Phi(U)} |D\widetilde{N}|\cdot|(D\Phi)\circ\Phi^{-1}|\cdot|\det(D\Phi^{-1})|\,du\,dv, \end{aligned}

where the inequality follows from the sub-multiplicativity of the fixed matrix norm. Choosing C>0C>0 as an upper bound for the continuous function ∣(DΦ)∘Φ−1∣⋅∣det⁡(DΦ−1)∣|(D\Phi)\circ\Phi^{-1}|\cdot|\det(D\Phi^{-1})| on Φ(U)\Phi(U), we conclude the proof. □

Remark 3.4. Recall that, locally around p=(0,0,0)p=(0,0,0), C(S)∩U={(x,y,t)∈S∩U:V(x,y)=0}C(S)\cap U=\{(x,y,t)\in S\cap U:V(x,y)=0\}. Since Φ(0)=0\Phi(0)=0, it holds that V(x,y)=0V(x,y)=0 if and only if V~(u,v)=0\widetilde{V}(u,v)=0 with (u,v):=Φ(x,y)(u,v):=\Phi(x,y) if and only if b(0,v)=0b(0,v)=0 with (0,v)=Φ(x,y)(0,v)=\Phi(x,y). Therefore, we have

C(S)∩U={(Φ−1(0,v),t)∈S∩U:(0,v)∈Φ(U), b(0,v)=0}.C(S)\cap U=\{(\Phi^{-1}(0,v),t)\in S\cap U:(0,v)\in\Phi(U),\ b(0,v)=0\}.

Hence, in the coordinates (u,v)(u,v), the characteristic set is described by the zeros of the C1,αC^{1,\alpha}-function h(v):=b(0,v)h(v):=b(0,v).

Local integrability of ∣DN~∣|D\widetilde{N}|

In light of Lemma 3.3, we may assume that the vector field VV, locally around the origin, has the form:

V(x,y)=(x,b(x,y)),∀(x,y)∈U,(16)V(x,y)=(x,b(x,y)),\qquad\forall(x,y)\in U, \tag*{(16)}

where b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^{2}), with b(0)=0b(0)=0.

Theorem 4.1. Let U⊂R2U\subset\mathbb{R}^{2} be open, 0∈U0\in U, and let V∈C1,α(U;R2)V\in C^{1,\alpha}(U;\mathbb{R}^{2}), 0<α≤10<\alpha\leq1, defined as in {(15)}. Then, denoting by N:=V∣V∣N:=\frac{V}{|V|}, there are constants C,r∗>0C,r_{*}>0 such that

∫Qr∣DN∣ dz<Cr,∀0<r<r∗,(17)\int_{Q_r}|DN|\,dz<Cr,\qquad\forall0<r<r_{*}, \tag*{(17)}

where Qr=(−r,r)2⊂R2Q_r=(-r,r)^2\subset\mathbb{R}^{2} is the standard square of edge length 2r2r.

Proof. Throughout the proof, C>0C>0 denotes a constant, independent of rr, whose value may change line by line.

For a vector field VV as in {(15)}, ∣V(x,y)∣2=x2+b(x,y)2|V(x,y)|^2=x^2+b(x,y)^2. Hence, we write

N(x,y)=V(x,y)∣V(x,y)∣=(x,b(x,y))(x2+b(x,y)2)1/2,∀(x,y)∈U∖{V=0}.N(x,y)=\frac{V(x,y)}{|V(x,y)|}=\frac{(x,b(x,y))}{(x^2+b(x,y)^2)^{1/2}},\qquad\forall(x,y)\in U\setminus\{V=0\}.

Differentiating NN, at points where V≠0V\ne0, we obtain:

∂xN=(b(b−xbx),−x(b−xbx))(x2+b2)3/2,∂yN=(−xbby,x2by)(x2+b2)3/2.\partial_xN=\frac{(b(b-xb_x),-x(b-xb_x))}{(x^2+b^2)^{3/2}},\qquad\partial_yN=\frac{(-xbb_y,x^2b_y)}{(x^2+b^2)^{3/2}}.

Consequently,

∣∂xN~∣=∣b−xbx∣x2+b2,∣∂yN~∣=∣xby∣x2+b2.(18)\left|\partial_x \widetilde N\right|=\frac{\left|b-xb_x\right|}{x^2+b^2},\qquad\left|\partial_y \widetilde N\right|=\frac{\left|xb_y\right|}{x^2+b^2}. \tag*{(18)}

We set h(y):=b(0,y)h(y):=b(0,y) for every y∈Ry\in\mathbb{R}. Since b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^2), h∈C1,α(R)h\in C^{1,\alpha}(\mathbb{R}). We claim that ∣V(x,y)∣2\lvert V(x,y)\rvert^2 is controlled by x2+h(y)2x^2+h(y)^2. Indeed, since bxb_x is bounded, we have

∣b(x,y)−h(y)∣≤∫0x∣bu(s,y)∣ ds≤C∣x∣.\left|b(x,y)-h(y)\right|\leq\int_0^x\left|b_u(s,y)\right|\,ds\leq C|x|.

Combining the latter and the triangle inequality, we then obtain

1C(x2+h(y)2)≤x2+b(x,y)2≤C(x2+h(y)2),\frac{1}{C}\left(x^2+h(y)^2\right)\leq x^2+b(x,y)^2\leq C\left(x^2+h(y)^2\right),

as claimed. For the numerators involved in (18), we exploit in a crucial way the Hölder regularity of bb. More precisely, we estimate

∣by(x,y)∣≤∣h′(y)∣+∣by(x,y)−by(0,y)∣≤∣h′(y)∣+C∣x∣α.\left|b_y(x,y)\right|\leq\left|h'(y)\right|+\left|b_y(x,y)-b_y(0,y)\right|\leq\left|h'(y)\right|+C|x|^\alpha.

Additionally, using the identity

b(x,y)−xbx(x,y)=h(y)+∫0x[bx(s,y)−bx(x,y)] ds,b(x,y)-xb_x(x,y)=h(y)+\int_0^x\left[b_x(s,y)-b_x(x,y)\right]\,ds,

we deduce that

∣b−xbx∣≤∣h(y)∣+C∣x∣1+α.\left|b-xb_x\right|\leq|h(y)|+C|x|^{1+\alpha}.

All in all, combining the previous bounds, we deduce that, for every (x,y)∈U∖{V=0}(x,y)\in U\setminus\{V=0\},

∣DN~(x,y)∣≤C∣h(y)∣+∣x∣1+α+∣x∣∣h′(y)∣x2+h(y)2.(19)\left|D\widetilde N(x,y)\right|\leq C\frac{|h(y)|+|x|^{1+\alpha}+|x||h'(y)|}{x^2+h(y)^2}. \tag*{(19)}

We study the integrability on QrQ_r of the right-hand side of (19). Firstly, observe that

∫Qr∣h(y)∣x2+h(y)2 dx dy=∫−rr∫−rr∣h(y)∣x2+h(y)2 dx dy=∫(−r,r)∩{h≠0}2arctan⁡r∣h(y)∣ dy≤2πr.\int_{Q_r}\frac{|h(y)|}{x^2+h(y)^2}\,dx\,dy=\int_{-r}^r\int_{-r}^r\frac{|h(y)|}{x^2+h(y)^2}\,dx\,dy=\int_{(-r,r)\cap\{h\ne0\}}2\arctan\frac{r}{|h(y)|}\,dy\leq2\pi r.

Secondly, recalling that α>0\alpha>0, we have:

∫Qr∣x∣1+αx2+h(y)2 dx dy≤2r∫−rr∣x∣−1+α dx=4αr1+α.\int_{Q_r}\frac{|x|^{1+\alpha}}{x^2+h(y)^2}\,dx\,dy\leq2r\int_{-r}^r|x|^{-1+\alpha}\,dx=\frac{4}{\alpha}r^{1+\alpha}.

Finally, for the last term of (19), we observe that

∫Qr∣x∣∣h′(y)∣x2+h(y)2 dx dy=∫−rr∣h′(y)∣log⁡(1+r2h(y)2) dy.(20)\int_{Q_r}\frac{|x||h'(y)|}{x^2+h(y)^2}\,dx\,dy=\int_{-r}^r|h'(y)|\log\left(1+\frac{r^2}{h(y)^2}\right)\,dy. \tag*{(20)}

with the convention that the integrand of the right-hand side is 0 on the set {h=h′=0}\{h=h'=0\}, so that it is defined almost everywhere. The term in (20) is bounded by CrCr, thanks to Lemma 4.2, proved below, which holds since h∈C1,α([−r,r])h\in C^{1,\alpha}([-r,r]). This concludes the proof. □

Lemma 4.2. Let h∈C1,α([−r,r])h\in C^{1,\alpha}([-r,r]) for some α∈(0,1]\alpha\in(0,1]. Then, there exists C>0C>0 such that

∫−rr∣h′(y)∣log⁡(1+r2h(y)2) dy≤Cr.\int_{-r}^r|h'(y)|\log\left(1+\frac{r^2}{h(y)^2}\right)\,dy\leq Cr.

Proof. We define the function Fr:R∖{0}→RF_r:\mathbb{R}\setminus\{0\}\to\mathbb{R} by

Fr(t):=log⁡(1+r2t2).(21)F_r(t):=\log\left(1+\frac{r^2}{t^2}\right). \tag*{(21)}

Observe that Fr∈C1(R∖{0})F_r\in C^1(\mathbb{R}\setminus\{0\}) and, for every p∈[1,+∞)p\in[1,+\infty), its LpL^p norm on (−R,R)(-R,R) satisfies

∥Fr∥Lp(−R,R)≤Cp,R.(22)\|F_r\|_{L^p(-R,R)}\leq C_{p,R}. \tag*{(22)}

for a constant Cp,R>0C_{p,R}>0, depending only on pp and R>0R>0. Using the function FrF_r, we write

∫−rr∣h′(y)∣log⁡(1+r2h(y)2) dy=∫−rr∣h′(y)∣(Fr∘h)(y) dy.\int_{-r}^{r} |h'(y)|\log\left(1+\frac{r^2}{h(y)^2}\right)\,dy=\int_{-r}^{r}|h'(y)|(F_r\circ h)(y)\,dy.

Now, if the set Z:={y∈(−r,r):h′(y)=0}\mathcal{Z}:=\{y\in(-r,r):h'(y)=0\} is empty, then the change of variable t=h(y)t=h(y) is well-defined on (−r,r)(-r,r) and gives

∫−rr∣h′(y)∣Fr(h(y)) dy=∫min⁡{h(r),h(−r)}max⁡{h(r),h(−r)}Fr(t) dt≤Cr,\int_{-r}^{r}|h'(y)|F_r(h(y))\,dy=\int_{\min\{h(r),h(-r)\}}^{\max\{h(r),h(-r)\}}F_r(t)\,dt\leq C_r,

by a direct computation, also using the fact that h(0)=0h(0)=0. If instead Z≠∅\mathcal{Z}\neq\varnothing, since h′h' is continuous, the complement of Z\mathcal{Z} is open and thus it can be written as a countable disjoint union of open intervals, namely

(−r,r)∖Z=⋃j∈NIj,Ij:=(aj,bj),(-r,r)\setminus\mathcal{Z}=\bigcup_{j\in\mathbb{N}}I_j,\qquad I_j:=(a_j,b_j),

with at least one between aj,bja_j,b_j belonging to Z\mathcal{Z}. If h′(aj)=0h'(a_j)=0, by Hölder continuity of h′h', we deduce that

∣h(bj)−h(aj)∣=∣∫ajbjh′(x) dx∣=∫ajbj∣h′(x)−h′(aj)∣ dx≤H∣bj−aj∣1+α,(23)|h(b_j)-h(a_j)|=\left|\int_{a_j}^{b_j}h'(x)\,dx\right|=\int_{a_j}^{b_j}|h'(x)-h'(a_j)|\,dx\leq H|b_j-a_j|^{1+\alpha}, \tag*{(23)}

where H>0H>0 is the Hölder seminorm of h′h' on [−r,r][-r,r]. The same estimate holds if h′(bj)=0h'(b_j)=0. Therefore, on the interval IjI_j, since h∣Ijh|_{I_j} is strictly monotone, we can perform the change of variable t=h(x)t=h(x) and obtain:

∫Ij∣h′(x)∣Fr(h(x)) dx=∫min⁡{h(aj),h(bj)}max⁡{h(aj),h(bj)}Fr(t) dt,(24)\int_{I_j}|h'(x)|F_r(h(x))\,dx=\int_{\min\{h(a_j),h(b_j)\}}^{\max\{h(a_j),h(b_j)\}}F_r(t)\,dt, \tag*{(24)}

The right-hand side of (23) can be estimated using Hölder inequality:

∫min⁡{h(aj),h(bj)}max⁡{h(aj),h(bj)}Fr(t) dt≤∥Fr∥Lp([−R,R])∣h(bj)−h(aj)∣1−1p≤Cp,RH1−1p∣bj−aj∣(1+α)(1−1p),\int_{\min\{h(a_j),h(b_j)\}}^{\max\{h(a_j),h(b_j)\}}F_r(t)\,dt\leq\|F_r\|_{L^p([-R,R])}|h(b_j)-h(a_j)|^{1-\frac{1}{p}}\leq C_{p,R}H^{1-\frac{1}{p}}|b_j-a_j|^{(1+\alpha)(1-\frac{1}{p})},

where, in the last inequality, we have used (22), together with (21), and R>0R>0 is a bound for the L∞L^\infty norm of hh. Since the latter inequality holds for every p∈[1,∞)p\in[1,\infty), we may choose pp such that

1−1p=11+α,1-\frac{1}{p}=\frac{1}{1+\alpha},

so that we obtain

∫Ij∣h′(x)∣Fr(h(x)) dx≤C∣bj−aj∣,\int_{I_j}|h'(x)|F_r(h(x))\,dx\leq C|b_j-a_j|,

where C>0C>0 is independent of ε\varepsilon and jj. Since the latter estimate holds for every j∈Nj\in\mathbb{N}, we conclude that

∫−rr∣h′(x)∣Fr(h(x)) dx=∑j∈N∫Ij∣h′(x)∣Fr(h(x)) dx≤C∑j∈N∣bj−aj∣≤2Cr.\int_{-r}^{r}|h'(x)|F_r(h(x))\,dx=\sum_{j\in\mathbb{N}}\int_{I_j}|h'(x)|F_r(h(x))\,dx\leq C\sum_{j\in\mathbb{N}}|b_j-a_j|\leq2Cr.

This concludes the proof. □

Proof of Theorem 1.1

Recalling that H∈C(S∖C(S))\mathcal{H}\in C(S\setminus C(S)), we are left to prove the local integrability of H\mathcal{H} around characteristic points. So, fix p=(0,0,0)∈C(S)p=(0,0,0)\in C(S), U⊂SU\subset S open neighborhood of pp and assume that S∩US\cap U is the graph of a function g∈C2,α(R2)g\in C^{2,\alpha}(\mathbb{R}^2). Observe that any Riemannian measure on SS is mutually absolutely continuous with the Lebesgue measure in coordinates (x,y)(x,y), namely

σR∣U=ρ(x,y) dx dy,\sigma_R|_U=\rho(x,y)\,dx\,dy,

where ρ,ρ−1∈L∞(R2)\rho,\rho^{-1} \in L^\infty(\mathbb{R}^2). By Lemma 3.2, it is enough to prove the integrability of ∣DN∣|DN| with respect to the Lebesgue measure on a sufficiently small ball centered at the origin Br⊂R2B_r \subset\mathbb{R}^2, where NN is defined in (12). By the change of coordinates of Lemma 3.3, it holds

∫Br∣DN∣ dx dy≤C∫Φ(Br)∣DN~∣ du dv,N~:=V~∣V~∣.\int_{B_r} |DN|\,dx\,dy \leq C \int_{\Phi(B_r)} |\widetilde{D N}|\,du\,dv,\qquad\widetilde{N}:=\frac{\widetilde{V}}{|\widetilde{V}|}.

where V~(u,v)=(u,b(u,v))\widetilde{V}(u,v)=(u,b(u,v)) for some b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^2). Since Φ\Phi is a C1,αC^{1,\alpha}-diffeomorphism, Φ(Br)\Phi(B_r) is an open set and Φ(Br)⊂BLr⊂QLr\Phi(B_r)\subset B_{Lr}\subset Q_{Lr}, where L>0L>0 is the Lipschitz constant of Φ\Phi. Hence, applying Theorem 4.1, we deduce that, up to possibly shrinking rr and updating the constant,

∫Br∣DN∣ dx dy≤Cr.\int_{B_r} |DN|\,dx\,dy \leq Cr.

This proves that HH is locally integrable close to p∈C(S)p\in C(S).

Proof of Theorem 1.2

We first show that ν∈BVloc(S;R3)\nu\in BV_{\mathrm{loc}}(S;\mathbb{R}^3). As before, it is enough to check this property near characteristic points. Fix p=(0,0,0)∈C(S)p=(0,0,0)\in C(S), U⊂SU\subset S open neighborhood of pp and assume that S∩US\cap U is the graph of a function g∈C2,α(R2)g\in C^{2,\alpha}(\mathbb{R}^2). According to (13), up to restricting UU, ν∈BVloc(U;R3)\nu\in BV_{\mathrm{loc}}(U;\mathbb{R}^3) if and only if N∈BVloc(Br;R2)N\in BV_{\mathrm{loc}}(B_r;\mathbb{R}^2), where NN is defined in (12) and Br={∣z∣<r}⊂R2B_r=\{|z|<r\}\subset\mathbb{R}^2.

As a first step, we assume that V(x,y)=(x,b(x,y))V(x,y)=(x,b(x,y)) with b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^2). Let h(y):=b(0,y)h(y):=b(0,y) for every y∈Ry\in\mathbb{R}, E={y∣h(y)=0}E=\{y\mid h(y)=0\} and a(y):=bx(0,y)a(y):=b_x(0,y). Recall that ∣V(x,y)∣=0|V(x,y)|=0 if and only if x=0x=0 and y∈Ey\in E. Hence, outside of the set {0}×E\{0\}\times E, N∈C1,αN\in C^{1,\alpha}, its classical derivative DNDN is well-defined and, by Theorem 4.1, DN∈Lloc1(Br;R2,2)DN\in L^1_{\mathrm{loc}}(B_r;\mathbb{R}^{2,2}). On the set {0}×E\{0\}\times E, the differential of NN has a jump part. Indeed, by the fundamental theorem of calculus, for every (x,y)∈R2(x,y)\in\mathbb{R}^2, we have

∣b(x,y)−h(y)−xa(y)∣=∣∫0x(bx(s,y)−bx(0,y)) ds∣≤C∣x∣α+1,|b(x,y)-h(y)-xa(y)|=\left|\int_0^x\bigl(b_x(s,y)-b_x(0,y)\bigr)\,ds\right|\leq C|x|^{\alpha+1},

where, in the last step, we have used that b∈C1,α(R2)b\in C^{1,\alpha}(\mathbb{R}^2). As a consequence, we find that

b(x,y)=h(y)+xa(y)+O(∣x∣1+α),as x→0, y∈R.b(x,y)=h(y)+xa(y)+O\left(|x|^{1+\alpha}\right),\qquad\text{as }x\to0,\ y\in\mathbb{R}.

Therefore, for every y∈Ey\in E, we compute

N+(y):=lim⁡x↓0N(x,y)=lim⁡x↓0(x,b(x,y))(x2+b(x,y)2)1/2=lim⁡x↓0(1,b(x,y)/x)(1+(b(x,y)/x)2)1/2=(1,a(y))(1+a(y)2)1/2,N^+(y):=\lim_{x\downarrow0}N(x,y)=\lim_{x\downarrow0}\frac{(x,b(x,y))}{(x^2+b(x,y)^2)^{1/2}}=\lim_{x\downarrow0}\frac{(1,b(x,y)/x)}{(1+(b(x,y)/x)^2)^{1/2}}=\frac{(1,a(y))}{(1+a(y)^2)^{1/2}},

and similarly

N−(y):=lim⁡x↑0N(x,y)=−(1,a(y))(1+a(y)2)1/2.N^-(y):=\lim_{x\uparrow0}N(x,y)=-\frac{(1,a(y))}{(1+a(y)^2)^{1/2}}.

While, for y∈R∖Ey\in\mathbb{R}\setminus E, since h(y)≠0h(y)\neq0, N(x,y)N(x,y) admits a limit as x→0x\to0, so that

N+(y)=N−(y):=lim⁡x→0N(x,y)=(0,sgn⁡h(y)).N^+(y)=N^-(y):=\lim_{x\to0}N(x,y)=(0,\operatorname{sgn}h(y)).

Define Br±:=Br∩{±x>0}B_r^\pm:=B_r\cap\{\pm x>0\}. Then, given φ∈Cc∞(Br;R2)\varphi\in C^\infty_c(B_r;\mathbb{R}^2), integration by parts shows that

∫Br±⟨N,∂xφ⟩ dx dy=∓∫−rr⟨N±,φ⟩ dy−∫Br±⟨∂xN,φ⟩ dx dy.\int_{B_r^\pm}\langle N,\partial_x\varphi\rangle\,dx\,dy =\mp\int_{-r}^r\langle N^\pm,\varphi\rangle\,dy-\int_{B_r^\pm}\langle\partial_xN,\varphi\rangle\,dx\,dy.

Similarly, by integration by parts, we see that ∂yN\partial_yN does not carry a singular part, indeed

∫Br⟨N,∂yφ⟩ dx dy=−∫Br⟨∂yN,φ⟩ dx dy.\int_{B_r}\langle N,\partial_y\varphi\rangle\,dx\,dy =-\int_{B_r}\langle\partial_yN,\varphi\rangle\,dx\,dy.

In conclusion, using that N+N^+ and N−N^- coincide on R∖E\mathbb{R}\setminus E, the distributional derivative [DN][DN] of NN is given by

[DN]=DN\mathcal{L}^2+\bigl((N^+-N^-)\otimes e_1\bigr)\mathcal{H}^1\mathbin{\vrule height 1.4ex depth -0.3ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}\_{\{0\}\times E},\qquad\text{on }B_r. \tag*{(25)}

Since DNDN is locally integrable and ∣N+−N−∣=2|N^{+}-N^{-}|=2, we finally conclude that N∈BVloc(Br;R2)N \in BV_{\mathrm{loc}}(B_r;\mathbb{R}^{2}).

The second step consists in proving that NN is of locally bounded variation for a general vector field VV. We use the change of coordinates of Lemma 3.3 and, up to restricting rr, write

V(x,y)=RTV~(Φ(x,y)),∀ (x,y)∈Br,V(x,y)=R^{T}\widetilde{V}(\Phi(x,y)),\qquad\forall\,(x,y)\in B_r,

where R∈O(2)R\in O(2), Φ∣Br\Phi|_{B_r} is the C1,αC^{1,\alpha}-diffeomorphism given by Lemma 3.3 and V~(u,v)=(u,b(u,v))\widetilde{V}(u,v)=(u,b(u,v)), for some b∈C1,αb\in C^{1,\alpha}. Then, on BrB_r, it holds

N=RTN~∘Φ,N~:=V~∣V~∣.N=R^{T}\widetilde{N}\circ\Phi,\qquad\widetilde{N}:=\frac{\widetilde{V}}{|\widetilde{V}|}.

Since Φ(Br)⊂BLr\Phi(B_r)\subset B_{Lr}, where L>0L>0 is the Lipschitz constant of Φ\Phi, by the previous step, N~∈BVloc(Φ(Br);R2)\widetilde{N}\in BV_{\mathrm{loc}}(\Phi(B_r);\mathbb{R}^{2}). Therefore, as NN is the composition of a function of locally bounded variation with a C1,αC^{1,\alpha}-diffeomorphism (hence bi-Lipschitz), we conclude that N∈BVloc(Br;R2)N\in BV_{\mathrm{loc}}(B_r;\mathbb{R}^{2}). Furthermore, (24) implies that

[DN]=DN⌞L2+JH1⌞Σ,(26)[DN]=DN\mathbin{\mathop{\llcorner}} \mathcal{L}^{2}+J\mathcal{H}^{1}\mathbin{\mathop{\llcorner}}\Sigma, \tag*{(26)}

where Σ:=Φ−1({0}×E)\Sigma:=\Phi^{-1}(\{0\}\times E) and J∈Lloc1(Σ;R2,2)J\in L^{1}_{\mathrm{loc}}(\Sigma;\mathbb{R}^{2,2}) is a suitable matrix-valued jump part.

We are now in position to prove also the second statement of the theorem. Assume that H1(C(S))=0\mathcal{H}^{1}(C(S))=0. By Remark 3.4, C(S)∩U={(x,y,t)∈S∩U:(x,y)∈Φ−1({0}×E)}C(S)\cap U=\{(x,y,t)\in S\cap U:(x,y)\in\Phi^{-1}(\{0\}\times E)\}. Thus, H1(C(S))=0\mathcal{H}^{1}(C(S))=0 implies that H1(Φ−1({0}×E))=0\mathcal{H}^{1}(\Phi^{-1}(\{0\}\times E))=0. Therefore, exploiting (25), we deduce that [DN]=DN⌞L2[DN]=DN\mathbin{\mathop{\llcorner}}\mathcal{L}^{2} and, thus, N∈Wloc1,1(Br;R2)N\in W^{1,1}_{\mathrm{loc}}(B_r;\mathbb{R}^{2}). As in the first part of the proof, thanks to (13), the latter statement is equivalent to ν∈Wloc1,1(U;R3)\nu\in W^{1,1}_{\mathrm{loc}}(U;\mathbb{R}^{3}), as claimed. Conversely, if ν∈Wloc1,1(U;R3)\nu\in W^{1,1}_{\mathrm{loc}}(U;\mathbb{R}^{3}), then, after the change of coordinate of Lemma 3.3, N~∈Wloc1,1(Φ(Br);R2)\widetilde{N}\in W^{1,1}_{\mathrm{loc}}(\Phi(B_r);\mathbb{R}^{2}). However, from (24), we see that the jump part of [DN~][D\widetilde{N}] always satisfy the identity ∣N~+−N~−∣=2|\widetilde{N}^{+}-\widetilde{N}^{-}|=2. Hence, to avoid contradictions, H1(Σ)=0\mathcal{H}^{1}(\Sigma)=0, or equivalently H1(C(S))=0\mathcal{H}^{1}(C(S))=0.

Remark 6.1. Theorem 1.2 is sharp in the sense that there are examples where H1(C(S))>0\mathcal{H}^{1}(C(S))>0 and ν\nu is not Wloc1,1W^{1,1}_{\mathrm{loc}}. Let f(x,y,t):=t−g(x,y)f(x,y,t):=t-g(x,y), where g∈C∞(R2)g\in C^{\infty}(\mathbb{R}^{2}) is defined by

g(x,y):=−x22−xy2,∀ (x,y)∈R2,g(x,y):=-\frac{x^{2}}{2}-\frac{xy}{2},\qquad\forall\,(x,y)\in\mathbb{R}^{2},

and consider the smooth surface

S={(x,y,t)∈H:f(x,y,t)=0}={(x,y,t)∈H:t=g(x,y)}.S=\{(x,y,t)\in\mathbb{H}:f(x,y,t)=0\}=\{(x,y,t)\in\mathbb{H}:t=g(x,y)\}.

Then ∇Hf=x(X1+X2)\nabla_{H}f=x(X_{1}+X_{2}), ∥∇Hf∥=2∣x∣\|\nabla_{H}f\|=\sqrt{2}|x| and C(S)={(0,y,0):y∈R}C(S)=\{(0,y,0):y\in\mathbb{R}\}. In particular, C(S)C(S) is a smooth curve of positive H1\mathcal{H}^{1}-measure. A horizontal unit normal, as a function on SS, is

ν(x,y)=sgn⁡(x)2(X1+X2)=sgn⁡(x)2(1,1,x−y2),x≠0.\nu(x,y)=\frac{\operatorname{sgn}(x)}{\sqrt{2}}(X_{1}+X_{2})=\frac{\operatorname{sgn}(x)}{\sqrt{2}}\left(1,1,\frac{x-y}{2}\right),\qquad x\ne0.

A direct computation shows that the matrix-valued distributional derivative of ν\nu is

Dν=sgn⁡(x)2(000012−12)+2(1,1,−y2)⊗e1H1⌞{x=0}.D\nu=\frac{\operatorname{sgn}(x)}{\sqrt{2}} \begin{pmatrix} 0 & 0\\ 0 & 0\\ \frac{1}{2} & -\frac{1}{2} \end{pmatrix} +\sqrt{2}\left(1,1,-\frac{y}{2}\right)\otimes e_{1}\mathcal{H}^{1}\mathbin{\mathop{\llcorner}}\{x=0\}.

Since this is a locally finite Radon measure, with a nonzero singular part,

ν∉Wloc1,1(S;R3).\nu\notin W^{1,1}_{\mathrm{loc}}(S;\mathbb{R}^{3}).

A C2 counterexample to the local integrability

In this section, we show that the additional Hölder regularity in Theorem 1.1 cannot, in general, be replaced by mere C2C^{2} regularity. More precisely, we construct a C2C^{2} surface whose characteristic set consists of a single isolated point and whose horizontal mean curvature fails to be locally integrable.

Proposition 7.1. Let S={(x,y,t)∈H:t=g(x,y)}⊂HS=\{(x,y,t)\in\mathbb{H}:t=g(x,y)\}\subset\mathbb{H} be the graph of the function

g(x,y):=−x22−xy2−∫0yh(s) ds(27)g(x,y):=-\frac{x^{2}}{2}-\frac{xy}{2}-\int_{0}^{y}h(s)\,\mathrm{d}s \tag*{(27)}

where

h(y):={e−1/y2(2+sin⁡(y4e1/y2)),y≠0,0,y=0.(28)h(y):= \begin{cases} e^{-1/y^{2}}\left(2+\sin\left(y^{4}e^{1/y^{2}}\right)\right), & y\ne0,\\ 0, & y=0. \end{cases} \tag*{(28)}

Then, the surface SS is of class C2C^{2}, the origin is its only characteristic point and

H∉Lloc1(S,σR),(29)\mathcal{H}\notin L^{1}_{\mathrm{loc}}(S,\sigma_{R}), \tag*{(29)}

where σR\sigma_{R} is any Riemannian perimeter measure.

Proof. Firstly, we claim that SS is a surface of class C2C^{2}. Observe that h(y)→0h(y)\to0 as y→0y\to0, so that hh is continuous. Furthermore, h(y)/y→0h(y)/y\to0 as y→0y\to0, and h′(y)→0h'(y)\to0 as y→0y\to0 since, for every y≠0y\ne0, it holds

h′(y)=2e−1/y2y3(2+sin⁡(y4e1/y2))+(4y3−2y)cos⁡(y4e1/y2).(30)h'(y)=\frac{2e^{-1/y^{2}}}{y^{3}}\left(2+\sin\left(y^{4}e^{1/y^{2}}\right)\right)+(4y^{3}-2y)\cos\left(y^{4}e^{1/y^{2}}\right). \tag*{(30)}

Thus, h∈C1(R)h\in C^{1}(\mathbb{R}) and h′(0)=0h'(0)=0. This implies that g∈C2(R2)g\in C^{2}(\mathbb{R}^{2}) and proves the claim. Secondly, we show that C(S)={(0,0,0)}C(S)=\{(0,0,0)\}. Let f(x,y,t):=t−g(x,y)f(x,y,t):=t-g(x,y), then

X1f(x,y,t)=x,X2f(x,y,t)=x+h(y).X_{1}f(x,y,t)=x,\qquad X_{2}f(x,y,t)=x+h(y).

Recalling (6), we have that

p∈C(S)⟺(x,x+h(y))=(0,0)⟺x=0, h(y)=0.p\in C(S)\quad\Longleftrightarrow\quad(x,x+h(y))=(0,0)\quad\Longleftrightarrow\quad x=0,\ h(y)=0.

Now, h(0)=0h(0)=0, so that (0,0,0)∈C(S)(0,0,0)\in C(S). On the other hand, h(y)>0h(y)>0 for every y≠0y\ne0, since 2+sin⁡(y4e1/y2)≥12+\sin\left(y^{4}e^{1/y^{2}}\right)\ge1 for every y≠0y\ne0.

To conclude the proof, we are left to prove (28) (with respect to the Lebesgue measure dx dy\mathrm{d}x\,\mathrm{d}y). Since SS is the tt-graph of (26), identity (14) holds and thus

H(x,y)=∂x(xW)+∂y(x+h(y)W),W:=(x2+(x+h(y))2)1/2.\mathcal{H}(x,y)=\partial_{x}\left(\frac{x}{W}\right)+\partial_{y}\left(\frac{x+h(y)}{W}\right),\qquad W:=\left(x^{2}+(x+h(y))^{2}\right)^{1/2}.

Computing explicitly the derivatives yields

H(x,y)=h′(y)x2+h(y)x+h(y)2(x2+(x+h(y))2)3/2.(31)\mathcal{H}(x,y)=\frac{h'(y)x^{2}+h(y)x+h(y)^{2}}{\left(x^{2}+(x+h(y))^{2}\right)^{3/2}}. \tag*{(31)}

Fix 0<r<1/40<r<1/4 so small that 0<h(y)<r/40<h(y)<r/4 whenever 0<∣y∣<r0<|y|<r. For such yy, set

Ir(y):=∫−rrx2W(x,y)3 dx,Jr(y):=∫−rr∣h(y)x+h(y)2∣W(x,y)3 dx.I_{r}(y):=\int_{-r}^{r}\frac{x^{2}}{W(x,y)^{3}}\,\mathrm{d}x,\qquad J_{r}(y):=\int_{-r}^{r}\frac{\left|h(y)x+h(y)^{2}\right|}{W(x,y)^{3}}\,\mathrm{d}x.

The identity (30) and the reverse triangle inequality imply the lower bound

∫−rr∣H(x,y)∣ dx≥∣h′(y)∣Ir(y)−Jr(y).(32)\int_{-r}^{r}|\mathcal{H}(x,y)|\,\mathrm{d}x\ge|h'(y)|I_{r}(y)-J_{r}(y). \tag*{(32)}

On the one hand, the term Jr(y)J_{r}(y) is uniformly bounded for every 0<∣y∣<r0<|y|<r, indeed, performing the change of variable x=h(y)sx=h(y)s gives

Jr(y)=∫−rr∣h(y)x+h(y)2∣(x2+(x+h(y))2)3/2 dx≤∫−∞+∞∣s+1∣(s2+(s+1)2)3/2 ds<+∞.(33)J_{r}(y)=\int_{-r}^{r}\frac{|h(y)x+h(y)^{2}|}{\left(x^{2}+(x+h(y))^{2}\right)^{3/2}}\,\mathrm{d}x\le\int_{-\infty}^{+\infty}\frac{|s+1|}{\left(s^{2}+(s+1)^{2}\right)^{3/2}}\,\mathrm{d}s<+\infty. \tag*{(33)}

On the other hand, for the term Ir(y)I_{r}(y), we observe that, for 2h(y)≤x≤r/22h(y)\le x\le r/2, W(x,y)W(x,y) is comparable to xx, with absolute constants. Therefore

Ir(y)≥c∫2h(y)r/2dxx=clog⁡r4h(y).(34)I_{r}(y)\ge c\int_{2h(y)}^{r/2}\frac{\mathrm{d}x}{x}=c\log\frac{r}{4h(y)}. \tag*{(34)}

Therefore, combining the lower bound (32) with (33) and (34), we see that H\mathcal{H} is not locally integrable around the origin if

∫0r∣h′(y)∣log⁡rh(y) dy=∞.(35)\int_{0}^{r} |h'(y)|\log\frac{r}{h(y)}\,\mathrm{d}y=\infty. \tag*{(35)}

We now prove (35). Since ∣sin⁡θ∣≤1|\sin\theta|\le1, it holds e−1/y2≤h(y)≤3e−1/y2e^{-1/y^2}\le h(y)\le3e^{-1/y^2}, which in turn implies that, for 0<y<r0<y<r sufficiently small,

cy2≤log⁡rh(y)≤Cy2,(36)\frac{c}{y^2}\le\log\frac{r}{h(y)}\le\frac{C}{y^2}, \tag*{(36)}

for constants 0<c<C0<c<C. Denoting by ϕ(y):=y4e1/y2\phi(y):=y^4e^{1/y^2}, for every y≠0y\ne0, by (30), we have

∣h′(y)∣≥2y∣cos⁡ϕ(y)∣−4y3−6e−1/y2y3.|h'(y)|\ge2y|\cos\phi(y)|-4y^3-\frac{6e^{-1/y^2}}{y^3}.

Combining this with (36), we obtain

∣h′(y)∣log⁡rh(y)≥c∣cos⁡ϕ(y)∣y−C(y+e−1/y2y5).(37)|h'(y)|\log\frac{r}{h(y)}\ge c\frac{|\cos\phi(y)|}{y}-C\left(y+\frac{e^{-1/y^2}}{y^5}\right). \tag*{(37)}

Since y↦y+y−5e−1/y2y\mapsto y+y^{-5}e^{-1/y^2} is locally integrable on R\mathbb{R}, (35) follows if we prove that

∫0r∣cos⁡ϕ(y)∣y dy=∞.(38)\int_{0}^{r}\frac{|\cos\phi(y)|}{y}\,\mathrm{d}y=\infty. \tag*{(38)}

Using ∣cos⁡θ∣≥cos⁡2θ=(1+cos⁡2θ)/2|\cos\theta|\ge\cos^2\theta=(1+\cos2\theta)/2, for every δ∈(0,r)\delta\in(0,r), we get

∫δr∣cos⁡ϕ(y)∣y dy≥12∫δr1+cos⁡(2ϕ(y))y dy≥12log⁡rδ+12∫δrcos⁡(2ϕ(y))y dy.(39)\int_{\delta}^{r}\frac{|\cos\phi(y)|}{y}\,\mathrm{d}y\ge\frac{1}{2}\int_{\delta}^{r}\frac{1+\cos(2\phi(y))}{y}\,\mathrm{d}y\ge\frac{1}{2}\log\frac{r}{\delta}+\frac{1}{2}\int_{\delta}^{r}\frac{\cos(2\phi(y))}{y}\,\mathrm{d}y. \tag*{(39)}

For 0<y<r<1/40<y<r<1/4, ϕ′(y)=e1/y2(4y3−2y)≠0\phi'(y)=e^{1/y^2}(4y^3-2y)\ne0, so that we may write

cos⁡(2ϕ(y))y=q(y)ddysin⁡(2ϕ(y)),q(y):=12yϕ′(y)=−e−1/y24y2(2y2−1).\frac{\cos(2\phi(y))}{y}=q(y)\frac{\mathrm{d}}{\mathrm{d}y}\sin(2\phi(y)),\qquad q(y):=\frac{1}{2y\phi'(y)}=-\frac{e^{-1/y^2}}{4y^2(2y^2-1)}.

Therefore, an integration by parts gives

∫δrcos⁡(2ϕ(y))y dy=[q(y)sin⁡(2ϕ(y))]δr−∫δrq′(y)sin⁡(2ϕ(y)) dy.\int_{\delta}^{r}\frac{\cos(2\phi(y))}{y}\,\mathrm{d}y=[q(y)\sin(2\phi(y))]_{\delta}^{r}-\int_{\delta}^{r}q'(y)\sin(2\phi(y))\,\mathrm{d}y.

Since, both q(y)q(y) and q′(y)q'(y) are absolutely integrable near zero, and q(y)→0q(y)\to0 as y→0y\to0, we find that the latter integral is bounded uniformly as δ→0+\delta\to0^+. Consequently, the logarithmic term in the right-hand side of (39) positively diverges as δ→0+\delta\to0^+, while the remaining integral stays bounded. This finally proves (38), and thus (35). □

Remark 7.2. In the previous example, the vector field VV as defined in (11) already has the form V(x,y)=(x,b(x,y))V(x,y)=(x,b(x,y)), with b(x,y)=x+h(y)b(x,y)=x+h(y). Then, as displayed by Theorem 4.1, what controls the local integrability of H\mathcal{H} is the function h(y)=b(0,y)h(y)=b(0,y), and in particular the integral

∫0r∣h′(y)∣∣log⁡(h(y))∣ dy.\int_{0}^{r}|h'(y)||\log(h(y))|\,\mathrm{d}y.

Then, the choice of hh is motivated by Lemma 4.2: in order to obtain (35) (and find a function falling outside the hypothesis of that lemma), one needs to choose a function whose first derivative has fast oscillations, and is not Hölder, close to zero. We verify that indeed the function gg is C2C^2, but it is not C2,αC^{2,\alpha} near the origin for any α>0\alpha>0. For y>0y>0 small, ϕ(y)\phi(y) is strictly decreasing and unbounded as y→0+y\to0^+. Hence, for every j∈Nj\in\mathbb{N}, we choose yj,zj→0y_j,z_j\to0 such that ϕ(yj)=2πj\phi(y_j)=2\pi j and ϕ(zj)=(2j+1)π\phi(z_j)=(2j+1)\pi. The mean value theorem applied to ϕ\phi, together with ∣ϕ′(y)∣≍ye1/y2|\phi'(y)|\asymp ye^{1/y^2}, gives

∣yj−zj∣≍e−1/yj2yj,∣h′(yj)−h′(zj)∣≥cyj.|y_j-z_j|\asymp\frac{e^{-1/y_j^2}}{y_j},\qquad|h'(y_j)-h'(z_j)|\ge cy_j.

As a consequence, since gyy(x,y)=−h′(y)g_{yy}(x,y)=-h'(y), for every α>0\alpha>0, it holds

∣gyy(0,yj)−gyy(0,zj)∣∣yj−zj∣α=∣h′(yj)−h′(zj)∣∣yj−zj∣α≳yjαeα/yj2∣h′(yj)−h′(zj)∣≥cyj1+αeα/yj2→j→∞∞.\frac{\left|g_{yy}(0,y_j)-g_{yy}(0,z_j)\right|}{\left|y_j-z_j\right|^\alpha} = \frac{\left|h'(y_j)-h'(z_j)\right|}{\left|y_j-z_j\right|^\alpha} \gtrsim y_j^\alpha e^{\alpha/y_j^2}\left|h'(y_j)-h'(z_j)\right| \ge c y_j^{1+\alpha}e^{\alpha/y_j^2} \xrightarrow{j\to\infty}\infty.

AI statement: The authors used ChatGPT 6 during the preparation of this manuscript for language editing, revision and to develop the counterexample of Section 7. The authors verified all mathematical content and references and assume responsibility for the final manuscript.

Dipartimento di Matematica, Università di Trento, Via Sommarive, 14, 38123 Povo TN, Italy Email address: [email protected]

Department of Information Engineering, Computer Science and Mathematics, Università degli Studi dell’Aquila, Via Vetoio 1, 67100 L’Aquila, Italy Email address: [email protected]

References

  1. [1]A. Agrachev, L. Rizzi, and T. Rossi. Relative heat content asymptotics for sub-Riemannian manifolds. Anal. PDE, 17(9):2997–3037, 2024.
  2. [2]Z. M. Balogh. Size of characteristic sets and functions with prescribed gradient. J. Reine Angew. Math., 564:63–83, 2003.DOI
  3. [3]Z. M. Balogh, F. Ferrari, B. Franchi, E. Vecchi, and K. Wildrick. Steiner’s formula in the Heisenberg group. Nonlinear Anal., 126:201–217, 2015.DOI
  4. [4]D. Barilari, E. Bellini, and A. Pinamonti. Curvature measures and the sub-Riemannian Gauss-Bonnet theorem, 2025. ArXiv:2509.26460. https://arxiv.org/abs/2509.26460.
  5. [5]D. Barilari, E. Bellini, and T. Rossi. Curvature distributions and the Steiner formula for characteristic surfaces in the Heisenberg group, 2026. ArXiv:2610.26706. https://arxiv.org/abs/2610.26706.
  6. [6]D. Barilari, U. Boscain, and D. Cannarsa. On the induced geometry on surfaces in 3D contact sub-Riemannian manifolds. ESAIM Control Optim. Calc. Var., 28:Paper No. 9, 28, 2022.DOI
  7. [7]D. Barilari and T. Bossio. Steiner and tube formulae in 3D contact sub-Riemannian geometry. Commun. Contemp. Math., 26(7):Paper No. 2350034, 40, 2024.DOI
  8. [8]T. Bossio, L. Rizzi, and T. Rossi. Titles in sub-Riemannian geometry and a Weyl’s invariance result for curves in the Heisenberg groups. Ann. Sc. Norm. Super. Pisa Cl. Sci., 2026. To appear.arxiv.org/abs/2408.16838
  9. [9]J.-H. Cheng, J.-F. Hwang, A. Malchiodi, and P. Yang. Minimal surfaces in pseudohermitian geometry. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5), 4(1):129–177, 2005.arxiv.org/abs/math/0401136
  10. [10]D. Danielli, N. Garofalo, and D. M. Nhieu. Sub-Riemannian calculus on hypersurfaces in Carnot groups. Adv. Math., 215(1):292–378, 2007.DOI
  11. [11]D. Danielli, N. Garofalo, and D. M. Nhieu. Integrability of the sub-Riemannian mean curvature of surfaces in the Heisenberg group. Proc. Amer. Math. Soc., 140(3):811–821, 2012.DOI
  12. [12]M. Derridj. Sur un théorème de traces. Ann. Inst. Fourier (Grenoble), 22(2):73–83, 1972.DOI
  13. [13]G. Giovannardi, A. Pinamonti, J. Pozuelo, and S. Verzellesi. The prescribed mean curvature equation for t-graphs in the sub-Finsler Heisenberg group ℕℝ. Adv. Math., 451:109788, 2024.DOI
  14. [14]G. Giovannardi, A. Pinamonti, and S. Verzellesi. Curvature estimates for minimal hypersurfaces in the Heisenberg group, 2025. ArXiv:2409.20359. https://arxiv.org/abs/2409.20359.
  15. [15]M. Gordinia, L. Luo, A. Pinamonti, and R. Sarkar. Heat content and spectrum for subordinated sub-Laplacians, 2026. ArXiv:2609.04045. https://arxiv.org/abs/2609.04045.
  16. [16]R. K. Hladky and S. D. Pauls. Constant mean curvature surfaces in sub-Riemannian geometry. J. Differential Geom., 79(1):111–139, 2008.DOI
  17. [17]F. Montefalcone. Stable H-minimal hypersurfaces. J. Geom. Anal., 25(2):820–870, 2015.
  18. [18]S. D. Pauls. Minimal surfaces in the Heisenberg group. Geom. Dedicata, 104:201–231, 2004.DOI
  19. [19]L. Rizzi and T. Rossi. Heat content asymptotics for sub-Riemannian manifolds. J. Math. Pures Appl., 148:267–307, 2021.DOI
  20. [20]T. Rossi. Integrability of the sub-Riemannian mean curvature at degenerate characteristic points in the Heisenberg group. Adv. Calc. Var., 16(1):99–110, 2023.
  21. [21]J. Tyson and J. Wang. Heat content and horizontal mean curvature on the Heisenberg group. Comm. Partial Differential Equations, 43(3):467–505, 2018.DOI

Paper details

vlnc:W_wrvfagjw
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents