betaFeedback

WorksMathematics

Remarks on thin sets in potential theory for sub-Laplacians

Abstract

In contrast to Euclidean cases, we observe that, there are sets in Heisenberg groups that are thin at the origin and block all geodesics from the origin. However, in polarizable Carnot groups, using horizontal radial curves constructed for polar coordinates by Balogh and Tyson, we show that sets, that are thin at the origin, cannot block all horizontal curves. Consequently, we are able to use the asymptotic behavior with respect to the fine topology to estimate the Hausdorff dimension of singularities of potentials for sub-Laplacians and obtain geometric applications.

Introduction

In this note we want to study thin sets in potential theory for sub-Laplacians in Carnot groups. Our main reference on potentials for sub-Laplacians in homogeneous Carnot groups is [3]. Our study results in applications in geometries modeled after those on the boundary at infinity of rank one symmetric spaces.

Let G=(RN,∘,δλ)G = (\mathbb{R}^{N}, \circ, \delta_{\lambda}) be a homogeneous Carnot group, and let Γ\Gamma be the fundamental solution to a sub-Laplacian L\mathcal{L} on GG. For a Radon measure μ\mu with the support in a domain D⊂RND \subset\mathbb{R}^{N}, the potential of μ\mu for the sub-Laplacian L\mathcal{L} is

RLμ,D(x)=∫DΓ(y−1∘x)dμ(y).R_{\mathcal{L}}^{\mu,D}(x) = \int_D \Gamma(y^{-1} \circ x)d\mu(y).

We would like to make the note that we always assume through out this paper that Radon measures we consider are such that their potentials are not identically infinite.

From [3 Definition 12.1.1], the L\mathcal{L}-fine topology is the smallest topology on RN\mathbb{R}^{N} with respect to which all potentials for sub-Laplacian L\mathcal{L} are continuous. A set EE is L\mathcal{L}-thin at a point x0x_{0} if x0x_{0} is not a limit point of EE with respect to the L\mathcal{L}-fine topology according to [3 Definition 12.2.1]. The Wiener criterion for sub-Laplacians [3 Theorem 12.4.2] is fundamental and characterizes thin sets in terms of L\mathcal{L}-capacity in Carnot groups.

2010 Mathematics Subject Classification. 53C17, 53C18, 43A80, 35J70, 35H20, 31C05, 31C15, 31B15.

Key words and phrases. potentials for sub-Laplacians L\mathcal{L}, L\mathcal{L}-thin sets, L\mathcal{L}-fine topology, geodesics in Heisenberg groups, horizontal radial curves, polarizable Carnot groups, Hausdorff dimensions, CR geometries.

We start with the study of the singular behavior of potentials for sub-Laplacians with respect to the L\mathcal{L}-fine topology. The first one is the extension of [15 Theorem 5.3 in Chapter 2]. But, to us, it is originated from [1] on subharmonic functions in 2 dimensions.

Theorem 3.1. Let RLμ,DR_{\mathcal{L}}^{\mu,D} be the potential of a Radon measure μ\mu with the support in DD for the sub-Laplacian L\mathcal{L} in a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}). Then, for any x0∈Dx_{0} \in D, there exists a set EE that is L\mathcal{L}-thin at x0x_{0} such that

lim⁡x→x0,x∉ERLμ,D(x)Γ(x0−1∘x)=μ({x0}).(1)\lim_{\substack{x\to x_{0},x\notin E}} \frac{R_{\mathcal{L}}^{\mu,D}(x)}{\Gamma(x_{0}^{-1}\circ x)}=\mu(\{x_{0}\}). \tag*{(1)}

The second one is a strong improvement of (1.1) with the help of a measure theoretic argument (see Lemma 3.1). It is the extension of [12 Theorem 1.1], which was responsible for estimating of the size of singularities in higher dimensions.

Theorem 3.2. Let RLμ,DR_{\mathcal{L}}^{\mu,D} be the potential of a Radon measure μ\mu with the support in DD for the sub-Laplacian L\mathcal{L} in a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}). Let SS be a compact subset in DD whose Hausdorff dimension is greater than ss for s∈(0,Q−2)s \in(0,Q-2), where QQ is the homogeneous dimension of G\mathbb{G}. Then there is a point x0∈Sx_{0} \in S and a set EE that is L\mathcal{L}-thin at x0x_{0} such that

RLμ,D(x)≤C(dL(x0−1∘x))2−Q+s(2)R_{\mathcal{L}}^{\mu,D}(x) \leq C\bigl(d_{\mathcal{L}}(x_{0}^{-1}\circ x)\bigr)^{2-Q+s} \tag*{(2)}

for xx close to x0x_{0} and not in EE, where dLd_{\mathcal{L}} is the L\mathcal{L}-gauge in (5).

The Hausdorff measures and the Hausdorff dimensions in Theorem 3.2 are with respect to the L\mathcal{L}-gauge dLd_{\mathcal{L}} (cf. [3 Chapter 13]). A result about potentials on Heisenberg groups similar to Theorem 3.2 was established before in [5 Lemma 1.10]. From the geodesic dynamics of Heisenberg groups Hn\mathbb{H}^{n} described in [16], one observes the following

Theorem 4.1. In Heisenberg group Hn\mathbb{H}^{n}, the set EE in (24) is ΔH\Delta_{\mathbb{H}}-thin at the origin according to Definition 2.6. However, any geodesic from the origin intersects with EE.

This is in contrast to the existence of rays escaping from a thin set in Euclidean spaces [15 Lemma 8.2 in Chapter 2] and [12 Theorem 1.2], which is the critical property of thin sets first used in [1 Theorem 1.1] in the study of the singular behavior of subharmonic functions that led to the simplified proof of Huber’s theorem on open surfaces [10] and later empowered [12, 13] to develop applications in conformal geometry in general dimensions.

On the existence of horizontal curves (not necessarily geodesics) escaping from a thin set, one realizes that polar coordinates of horizontal curves, if exist, are invaluable. Indeed the polar coordinates constructed by the horizontal flow lines (25) in [2] enable us to establish the existence of horizontal curves escaping from a thin set in polarizable Carnot groups.

Theorem 5.1. Let G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}) be a polarizable Carnot group. And let EE be a set that is thin at the origin in RN\mathbb{R}^{N}. Then there always exist horizontal radial curves which are flow lines given by (25) and do not intersect with EE, at least in a small neighborhood of the origin.

Consequently we are able to use Theorem 3.2 to estimate the Hausdorff dimension of singularities of potentials for sub-Laplacians in polarizable Carnot groups (cf. Theorem 5.2) using the same ideas in [1, 12, 13, 5]. Notice that the nilpotent part of Iwasawa decomposition of rank one simple groups is H-type (cf. [6]) and therefore polarizable due to [2 Proposition 5.6]. Thus this gives an alternative proof of the main geometric results in [14]. For background about scalar curvature and Yamabe type equations in geometries modeled after those on the boundary at infinity of rank one symmetric spaces, readers are referred to [14 Section 4] and references therein.

Theorem 5.3. ([14 Theorem 4.2]) Assume N is the nilpotent part of the Iwasawa decomposition of rank one groups: SU⁡(n+1,1)\operatorname{SU}(n+1,1), Sp⁡(n+1,1)\operatorname{Sp}(n+1,1) and F4(−20)\mathrm{F}_{4(-20)}. Let SS be a compact subset of a bounded and open subset Ω\Omega in N. And let θ=u4Q−2θ0\theta=u^{\frac{4}{Q-2}}\theta_0 be a conformal contact form on Ω∖S\Omega\setminus S, where θ0\theta_0 is the standard contact form on N and QQ is the homogeneous dimension of N. Suppose that the sub-Riemannian metric associated with θ\theta is geodesically complete near SS. Then

dim⁡HΔN(S)≤Q−22(3)\dim_{\mathcal{H}_{\Delta_N}}(S) \leq\frac{Q-2}{2} \tag*{(3)}

when the scalar curvature RθR_\theta of the conformal contact form θ\theta is nonnegative.

Theorem 5.3 in the cases of CR geometry has appeared in [5 Theorem 1.1]. The organization of this note is as follows: In Section 2, we recall definitions in potential theory for sub-Laplacians from [3]. In Section 3, we derive estimates on potentials for sub-Laplacians with respect to the fine topology similar to those on Riesz potentials in Euclidean spaces. In Section 4, we use the presentation of geodesics from the origin in Heisenberg groups in [16] and construct a set that is ΔH\Delta_H-thin at the origin and blocks all geodesics from the origin. In Section 5, we introduce horizontal radial curves for polar coordinates constructed by Balogh and Tyson in [2] and use Theorem 5.1 to derive consequences in potential theory as well as in geometry.

potentials for sub-Laplacians in homogeneous Carnot groups

In this section we mostly follow the book [3] to quickly introduce the potential theory for sub-Laplacians in homogeneous Carnot groups.

Basic notations and definitions

Definition 2.1. ([3 Definition 1.4.1]) We say a Lie group on RN\mathbb{R}^{N}, G=(RN,∘,δλ)G=(\mathbb{R}^{N},\circ,\delta_\lambda), is a homogeneous Carnot group if the following properties hold

  1. RN\mathbb{R}^{N} can be split as RN1×RN2⋯×RNr\mathbb{R}^{N_1}\times\mathbb{R}^{N_2}\cdots\times\mathbb{R}^{N_r} and the dilation

δλ(x(1),x(2),…,x(r))=(λx(1),λ2x(2),…,λrx(r)),where x(i)∈RNi,\delta_\lambda\left(x^{(1)},x^{(2)},\ldots,x^{(r)}\right)=\left(\lambda x^{(1)},\lambda^2x^{(2)},\ldots,\lambda^r x^{(r)}\right),\qquad\text{where }x^{(i)}\in\mathbb{R}^{N_i},

is an automorphism of the group GG for all λ>0\lambda>0. Hence the group with this dilation δλ\delta_\lambda is a homogeneous Lie group (cf. [3 Definition 1.3.1]).

::: (2) Let Z1,Z2,⋯ ,ZN1Z_1,Z_2,\cdots,Z_{N_1} be the left invariant vector fields on G\mathbb{G} such that

Zi(0)=∂∂xi∣0 for i=1,2,⋯ ,N1.Z_i(0)=\left.\frac{\partial}{\partial x_i}\right|_0 \text{ for } i=1,2,\cdots,N_1.

Then the rank of the Lie algebra generated by Z1,Z2,⋯ ,ZN1Z_1,Z_2,\cdots,Z_{N_1} is NN. The group together with the dilation (RN,∘,δλ)(\mathbb{R}^N,\circ,\delta_\lambda) is said to be a homogeneous Carnot group of step rr and N1N_1 generators. And the vector fields Z1,Z2,⋯ ,ZN1Z_1,Z_2,\cdots,Z_{N_1} are called the Jacobian generators.

By [3 Proposition 1.3.21], the Lebesgue measure on RN\mathbb{R}^N is the Haar measure on a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^N,\circ,\delta_\lambda). And, if we denote by ∣E∣|E| the Lebesgue measure of a measurable set E⊂RNE\subset\mathbb{R}^N, then

∣δλ(E)∣=λQ∣E∣.|\delta_\lambda(E)|=\lambda^Q|E|.

where Q=∑i=1riNiQ=\sum_{i=1}^r iN_i is called the homogeneous dimension of G\mathbb{G}. For our purpose we assume Q>2Q>2 throughout this note. Next important objects for us are sub-Laplacians.

Definition 2.2. ([3 Definition 1.5.1]) Let Z1,Z2,⋯ ,ZN1Z_1,Z_2,\cdots,Z_{N_1} be the Jacobian generators of a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^N,\circ,\delta_\lambda). The operator ΔG=∑i=1N1Zi2\Delta_{\mathbb{G}}=\sum_{i=1}^{N_1}Z_i^2 is called the canonical sub-Laplacian on G\mathbb{G}. The operator L=∑i=1N1Yi2\mathcal{L}=\sum_{i=1}^{N_1}Y_i^2, where {Y1,Y2,⋯ ,YN1}\{Y_1,Y_2,\cdots,Y_{N_1}\} is a frame of left-invariant vector fields in the horizontal distribution and generate the Lie algebra of rank NN, is said to be a sub-Laplacian on G\mathbb{G}.

Next are fundamental solutions and gauge functions.

Definition 2.3. ([3 Definition 5.3.1]) Let G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^N,\circ,\delta_\lambda) be a homogeneous Carnot group. And let L\mathcal{L} be a sub-Laplacian on G\mathbb{G}. A function Γ:RN∖{0}→R\Gamma:\mathbb{R}^N\setminus\{0\}\to\mathbb{R} is said to be a fundamental solution for L\mathcal{L} if:

  • Γ\Gamma is smooth away from the origin;

  • Γ∈Lloc1(RN)\Gamma\in L^1_{\mathrm{loc}}(\mathbb{R}^N) and Γ(x)→0\Gamma(x)\to0 as x→∞x\to\infty;

  • −LΓ=δ0-\mathcal{L}\Gamma=\delta_0, where δ0\delta_0 is the Dirac measure supported at the origin 00. More explicitly, ∫RNΓLϕ=−ϕ(0)\int_{\mathbb{R}^N}\Gamma\mathcal{L}\phi=-\phi(0), ∀ϕ∈C0∞(RN)\forall\phi\in C_0^\infty(\mathbb{R}^N).

It turns out, by [3 Theorem 5.3.2 and Proposition 5.3.10], for a sub-Laplacian L\mathcal{L}, there always exists a unique fundamental solution, which is positive with the homogeneity

Γ(δλ(x))=λ2−QΓ(x)(4)\Gamma(\delta_\lambda(x))=\lambda^{2-Q}\Gamma(x) \tag*{(4)}

(cf. [3 Proposition 5.3.12 and Proposition 5.3.13]). In fact, for a sub-Laplacian L\mathcal{L},

dL(x,y)=dL(y−1∘x)={Γ(y−1∘x)12−Qwhen x≠00when x=y(5)d_{\mathcal{L}}(x,y)=d_{\mathcal{L}}(y^{-1}\circ x)= \begin{cases} \Gamma(y^{-1}\circ x)^{\frac{1}{2-Q}} & \text{when }x\ne0\\ 0 & \text{when }x=y \end{cases} \tag*{(5)}

is an L\mathcal{L}-gauge (cf. [3 Definition 5.4.1]), which is a symmetric homogeneous norm on G\mathbb{G}. It is also good to know that, by [3 Proposition 5.3.10], L\mathcal{L}-gauges are unique up to a constant multiple.

At this point we are ready to introduce potentials of Radon measures for sub-Laplacians in homogeneous Carnot groups.

Definition 2.4. Let G=(RN,∘,δλ)\mathrm{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}) be a homogeneous Carnot group. And let L\mathcal{L} be a sub-Laplacian on G\mathrm{G} and Γ\Gamma is its fundamental solution. We define the potential for the sub-Laplacian L\mathcal{L} as

RLμ,D(x)=∫DΓ(y−1∘x)dμ(y)(6)R_{\mathcal{L}}^{\mu,D}(x)=\int_{D}\Gamma(y^{-1}\circ x)d\mu(y) \tag*{(6)}

for all x∈RNx \in\mathbb{R}^{N}, where μ\mu is a Radon measure and DD is the support of μ\mu.

It is important to know by [3 Theorem 9.4.4] that an L\mathcal{L}-superharmonic function (cf. ([3 Definition 7.2.2 and Remark 7.2.3]) is represented by a L\mathcal{L}-potential RLμ,DR_{\mathcal{L}}^{\mu,D} for a nonnegative Radon measure μ\mu up to L\mathcal{L}-harmonic functions.

L-capacity, L-thin sets, and the L-fine topology

We refer readers to [3 Chapter 11 and 12] for a more comprehensive introduction and discussion of the L\mathcal{L}-capacity, L\mathcal{L}-thinness, and the L\mathcal{L}-fine topology for a sub-Laplacian L\mathcal{L} in G=(RN,∘,δλ)\mathrm{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}). The L\mathcal{L}-fine topology is characterized as the smallest topology with respect to which all L\mathcal{L}-superharmonic functions is continuous (cf. [3 Definition 12.1.1]). L\mathcal{L}-thinness is introduced by the L\mathcal{L}-fine topology (cf. [3 Definition 12.2.1]). It is deep and powerful in the potential theory that the Wiener criterion for sub-Laplacians provides an equivalent definition of L\mathcal{L}-thinness in terms of L\mathcal{L}-capacity (cf. [3 Theorem 12.4.2]). For the introduction of exterior L\mathcal{L}-capacity for subset in RN\mathbb{R}^{N}, we refer readers to [3 Definition 11.5.4 and Definition 11.5.14]. For simplicity, we summarize it as follows.

Definition 2.5. ([3 Chapter 11]) Let G=(RN,∘,δλ)\mathrm{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}) be a homogeneous Carnot group. And let L\mathcal{L} be a sub-Laplacian on G\mathrm{G}. First, for any compact subset K⊂RNK \subset\mathbb{R}^{N}, we define the L\mathcal{L}-capacity as

cap⁡L(K)=inf⁡{μ(K):μ is a nonnegative Radon measure on K and RLμ,K≥1 on K}\operatorname{cap}_{\mathcal{L}}(K)=\inf\{\mu(K):\mu\text{ is a nonnegative Radon measure on }K\text{ and }R_{\mathcal{L}}^{\mu,K}\geq1\text{ on }K\}

or equivalently (cf. [3 Theorem 11.5.6])

cap⁡L(K)=sup⁡{μ(K):μ is a nonnegative Radon measure on K and RLμ,K≤1 in RN}\operatorname{cap}_{\mathcal{L}}(K)=\sup\{\mu(K):\mu\text{ is a nonnegative Radon measure on }K\text{ and }R_{\mathcal{L}}^{\mu,K}\leq1\text{ in }\mathbb{R}^{N}\}

Then, for any subset E⊂RNE \subset\mathbb{R}^{N}, we define the exterior L\mathcal{L}-capacity as

cap⁡L∗(E)=inf⁡U⊃E and openK⊂U and compact ⁣cap⁡L(K).\operatorname{cap}_{\mathcal{L}}^{*}(E)=\inf_{\substack{U\supset E\text{ and open}\\K\subset U\text{ and compact}}\!}\operatorname{cap}_{\mathcal{L}}(K).

To set notations for introducing L\mathcal{L}-thin sets via the exterior L\mathcal{L}-capacity, for x0∈RNx_{0}\in\mathbb{R}^{N}, we recall from [3 (12.16)],

Cn={y∈RN:αn≤Γ(y−1∘x0)≤αn+1}={y∈RN:αˉ−n−1≤dL(x0,y)≤αˉ−n}(7)C_{n}=\{y\in\mathbb{R}^{N}:\alpha^{n}\leq\Gamma(y^{-1}\circ x_{0})\leq\alpha^{n+1}\}=\{y\in\mathbb{R}^{N}:\bar{\alpha}^{-n-1}\leq d_{\mathcal{L}}(x_{0},y)\leq\bar{\alpha}^{-n}\} \tag*{(7)}

for

αˉ=α1Q−2>β,(8)\bar{\alpha}=\alpha^{\frac{1}{Q-2}}>\beta, \tag*{(8)}

where β(≥1)\beta(\ge1) is the constant in the improved pseudo-triangle inequality for the L\mathcal{L}-gauge dLd_{\mathcal{L}} of (5) (cf. [3 Proposition 5.14.1 and (12.15a)]). We remark here that αˉ\bar{\alpha} was taken as 2 and CnC_n are the standard dyadic spherical shells in Euclidean cases.

Definition 2.6. ([3 Definition 12.2.1 and Theorem 12.4.2]) Let G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^{N},\circ,\delta_{\lambda}) be a homogeneous Carnot group. And let L\mathcal{L} be a sub-Laplacian on G\mathbb{G}. Then we say a subset E⊂RNE\subset\mathbb{R}^{N} is L\mathcal{L}-thin at a point x0∈RNx_{0}\in\mathbb{R}^{N} if

∑n=1∞αncap⁡L∗(E∩Cn)=∑n=1∞αˉn(Q−2)cap⁡L∗(E∩Cn)<∞(9)\sum_{n=1}^{\infty}\alpha^{n}\operatorname{cap}^{*}_{\mathcal{L}}(E\cap C_{n})=\sum_{n=1}^{\infty}\bar{\alpha}^{n(Q-2)}\operatorname{cap}^{*}_{\mathcal{L}}(E\cap C_{n})<\infty \tag*{(9)}

where CnC_n given in (7) is the dyadic spherical shell with respect to dLd_{\mathcal{L}} and centered at x0x_0.

Asymptotic behavior of potentials with respect to the fine topology

The notion of the fine topology is such an sharp way to study the continuity of potentials due to the well-known Wiener criterion (cf. [3 Chapter 12] and [15 Chapter 2]). It turns out the fine topology is also sharp in describing the singular behavior of potentials. In Euclidean spaces, for example, we have [15 Theorem 5.3 in Chapter 2] on Riesz potentials. The following is the analog of [15 Theorem 5.3 in Chapter 2] on potentials for sub-Laplacians.

Theorem 3.1. Let RLμ,D\mathcal{R}_{\mathcal{L}}^{\mu,D} be the potential of a Radon measure μ\mu with the support in DD for the sub-Laplacian L\mathcal{L} in a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^{N},\circ,\delta_{\lambda}). Then, for any x0∈Dx_{0}\in D, there exists a set EE that is L\mathcal{L}-thin at x0x_{0} such that

lim⁡x→x0x∉ERLμ,D(x)Γ(x0−1∘x)=μ({x0}).(10)\lim_{\substack{x\to x_{0}\\x\notin E}}\frac{\mathcal{R}_{\mathcal{L}}^{\mu,D}(x)}{\Gamma(x_{0}^{-1}\circ x)}=\mu(\{x_{0}\}). \tag*{(10)}

Proof. The proof in principle goes with little changes from that of [15 Theorem 5.3 in Chapter 2]. But we want to take the opportunity and present the proof to show how Definition 2.6 is used beautifully. To deal with dLd_{\mathcal{L}}, we will need to use the reverse improved pseudo-triangle inequality, for example, [3 (12.15b)], that is,

dL(y−1∘x)≥1β∣dL(x)−dL(y)∣.(11)d_{\mathcal{L}}(y^{-1}\circ x)\ge\frac{1}{\beta}\lvert d_{\mathcal{L}}(x)-d_{\mathcal{L}}(y)\rvert. \tag*{(11)}

For simplicity we may let x0=0x_{0}=0 and BL(0,r)=BL(r)={x∈RN:dL(x)<r}B_{\mathcal{L}}(0,r)=B_{\mathcal{L}}(r)=\{x\in\mathbb{R}^{N}:d_{\mathcal{L}}(x)<r\} with respect to dLd_{\mathcal{L}}. For xx near 0, we may assume x∈Cix\in C_i for some ii. Then we write

RLμ,D(x)=(∫D∖BL(αˉ−i0)+∫BL(αˉ−i0)∖Cˉi+∫Cˉi)Γ(y−1∘x)dμ(y)=I+II+III.\mathcal{R}_{\mathcal{L}}^{\mu,D}(x)=\left(\int_{D\setminus B_{\mathcal{L}}(\bar{\alpha}^{-i_0})}+\int_{B_{\mathcal{L}}(\bar{\alpha}^{-i_0})\setminus\bar{C}_i}+\int_{\bar{C}_i}\right)\Gamma(y^{-1}\circ x)d\mu(y)=I+II+III.

where Cˉi=Ci−1∪Ci∪Ci+1\bar{C}_i=C_{i-1}\cup C_i\cup C_{i+1}. Hence we have, for i0i_0 to be chosen to be large but fixed,

I≤Cμ(D)αˉ(Q−2)(i0−i)Γ(x)=o(1)Γ(x)(12)I\le C\mu(D)\bar{\alpha}^{(Q-2)(i_0-i)}\Gamma(x)=o(1)\Gamma(x) \tag*{(12)}

as x→0x \to0 or equivalently i→∞i \to\infty. We write

II=(∫BL(αˉ−i−2)A+∫BL(αˉ−i0)∖BL(αˉ−i+1))Γ(y−1∘x)dμ(y)=II1+II2II = \left(\int_{B_{\mathcal{L}}(\bar{\alpha}^{-i-2})}^{\phantom{A}} + \int_{B_{\mathcal{L}}(\bar{\alpha}^{-i_0})\setminus B_{\mathcal{L}}(\bar{\alpha}^{-i+1})}\right)\Gamma(y^{-1}\circ x)d\mu(y)=II_1+II_2

where

II1=(μ({0})+o(1))Γ(x) as x→0II_1=(\mu(\{0\})+o(1))\Gamma(x)\text{ as }x\to0

and

II2≤C(μ(BL(αˉ−i0))−μ({0}))Γ(x)=ϵΓ(x)II_2\leq C\bigl(\mu(B_{\mathcal{L}}(\bar{\alpha}^{-i_0}))-\mu(\{0\})\bigr)\Gamma(x)=\epsilon\Gamma(x)

for any ϵ>0\epsilon>0 as i0i_0 chosen to be sufficiently large. Next, for IIIIII, we set

Eiλi={x∈Cˉi:∫CˉiΓ(y−1∘x)dμ(y)≥λiαi}.E_i^{\lambda_i}=\left\{x\in\bar{C}_i:\int_{\bar{C}_i}\Gamma(y^{-1}\circ x)d\mu(y)\geq\lambda_i\alpha^i\right\}.

for some λi→0\lambda_i\to0 to be chosen. Notice that

∫CˉiΓ(y−1∘x)dμ(y)Γ(x)≤Cα−i∫CˉiΓ(y−1∘x)dμ(y)≤Cλi→0\frac{\int_{\bar{C}_i}\Gamma(y^{-1}\circ x)d\mu(y)}{\Gamma(x)} \leq C\alpha^{-i}\int_{\bar{C}_i}\Gamma(y^{-1}\circ x)d\mu(y)\leq C\lambda_i\to0

when x∉Eiλix\notin E_i^{\lambda_i}. Moreover, by Definition 2.5, capL∗(Eiλi)≤1λiα−iμ(Cˉi)cap_{\mathcal{L}}^*(E_i^{\lambda_i})\leq\frac{1}{\lambda_i}\alpha^{-i}\mu(\bar{C}_i). Therefore, if set E=⋃iEiλiE=\bigcup_i E_i^{\lambda_i}, then

∑nαncapL∗(E∩Cn)≤∑nαncapL∗(Enλn)≤∑n1λnμ(Cˉn).\sum_n\alpha^n cap_{\mathcal{L}}^*(E\cap C_n)\leq\sum_n\alpha^n cap_{\mathcal{L}}^*(E_n^{\lambda_n})\leq\sum_n\frac{1}{\lambda_n}\mu(\bar{C}_n).

Finally, ∑nαncapL∗(E∩Cn)<∞\sum_n\alpha^n cap_{\mathcal{L}}^*(E\cap C_n)<\infty for some sequence λn→0\lambda_n\to0 when ∑nμ(Cˉn)≤3μ(D)<∞\sum_n\mu(\bar{C}_n)\leq3\mu(D)<\infty, thanks to [4 (5) Page 40] (see also [7]). Thus EE is L\mathcal{L}-thin at the origin by Definition 2.6. □

(10) can be strengthened if one may choose the point x0x_0 in a given set of Hausdorff dimension ss. This improvement is the key to estimate Hausdorff dimensions of singularities. Namely, we have the following extension of [12 Theorem 1.1].

Theorem 3.2. Let RLμ,DR_{\mathcal{L}}^{\mu,D} be the potential of a Radon measure μ\mu supported in DD for the sub-Laplacian L\mathcal{L} in a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^N,\circ,\delta_\lambda). Let SS be a compact subset in DD whose Hausdorff dimension dim⁡HL\dim_{\mathcal{H}}\mathcal{L} is greater than ss for s∈(0,Q−2)s\in(0,Q-2). Then there is a point x0∈Sx_0\in S and a set EE that is L\mathcal{L}-thin at x0x_0 such that

RLμ,D(x)≤C(dL(x0−1∘x))2−Q+s(13)R_{\mathcal{L}}^{\mu,D}(x)\leq C\bigl(d_{\mathcal{L}}(x_0^{-1}\circ x)\bigr)^{2-Q+s} \tag*{(13)}

for xx close to x0x_0 and not in $E.

The proof of Theorem 3.2 in principle goes again with little changes from that of [12 Theorem 1.1]. The only issue is that we need an extension of [11 Proposition 1.4] to cover Carnot-Caratheodory spaces.

Lemma 3.1. Let dLd_{\mathcal{L}} be the L\mathcal{L}-gauge in (5) associated with a sub-Laplacian L\mathcal{L} in a homogeneous Carnot group G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N},\circ,\delta_{\lambda}). And let μ\mu be a nonnegative Radon measure on RN\mathbb{R}^{N} and

Gs∞={x∈RN:lim sup⁡r→0r−sμ(BL(x,r))=∞}.G_s^{\infty} = \left\{x\in\mathbb{R}^{N}:\limsup_{r\to0}r^{-s}\mu(B_{\mathcal{L}}(x,r))=\infty\right\}.

Then

HLs(Gs∞)=0\mathcal{H}_{\mathcal{L}}^{s}(G_s^{\infty})=0

where BL(x,r)B_{\mathcal{L}}(x,r) is the ball with respect to dLd_{\mathcal{L}} and HLs\mathcal{H}_{\mathcal{L}}^{s} is the Hausdorff measure of dimension ss that is associated with dLd_{\mathcal{L}} (cf. [3 Chapter 13]).

Proof. The proof goes similar to the proof of [11 (iii) of Proposition 1.4]. One only needs to make sure that [11 Lemma 2.3] is applicable to Carnot-Caratheodory spaces, which is known, for example, thanks to [8 (1) of Theorem 2.10.18]. □

Some version of Theorem 3.2 for Heisenberg groups appeared earlier as [5 Lemma 1.10]. The Radon measures μ\mu in Theorem 3.2 and Lemma 3.1 in this paper need not to be absolutely continuous with respect to the Lebesgue measure in RN\mathbb{R}^{N}.

Geodesics in Heisenberg groups and counter-examples

In this section we construct a counter-example in Heisenberg groups to the property of thin sets observed in [1 Theorem 1.1], [15 Lemma 8.2 in Chapter 2], and [12 Theorem 1.2] in Euclidean cases.

Heisenberg groups

The Heisenberg group, as a homogeneous Carnot group, is

Hn={(x,y,z):x,y∈Rn and z∈R}=R2n+1\mathbb{H}^{n}=\{(x,y,z):x,y\in\mathbb{R}^{n}\text{ and }z\in\mathbb{R}\}=\mathbb{R}^{2n+1}

with the multiplication

(x,y,z)∘(x′,y′,z′)=(x+x′,y+y′,z+z′+12(x⋅y′−x′⋅y))(14)(x,y,z)\circ(x',y',z')=\left(x+x',y+y',z+z'+\frac{1}{2}(x\cdot y'-x'\cdot y)\right) \tag*{(14)}

and the dilation δλ(x,y,z)=(λx,λy,λ2z)\delta_{\lambda}(x,y,z)=(\lambda x,\lambda y,\lambda^{2}z). The Jacobian generators in the Heisenberg group Hn\mathbb{H}^{n} are

Xi=∂xi−yi2∂zandYi=∂yi+xi2∂zX_i=\partial_{x_i}-\frac{y_i}{2}\partial_z\quad\text{and}\quad Y_i=\partial_{y_i}+\frac{x_i}{2}\partial_z

for i=1,2,…,ni=1,2,\ldots,n. The canonical sub-Laplacian is

ΔH=∑i=1nXi2+Yi2=ΔX+ΔY+∣x∣2+∣y∣24∂z2+(−y⋅∂x+x⋅∂y)∂z(15)\Delta_{\mathbb{H}}=\sum_{i=1}^{n}X_i^{2}+Y_i^{2}=\Delta_X+\Delta_Y+\frac{|x|^{2}+|y|^{2}}{4}\partial_z^{2}+(-y\cdot\partial_x+x\cdot\partial_y)\partial_z \tag*{(15)}

and the fundamental solution to the canonical sub-Laplacian ΔH\Delta_{\mathbb{H}} is

ΓΔH(0,(x,y,z))=[dΔH(x,y,z)]2−Q,(16)\Gamma_{\Delta_{\mathbb{H}}}(0,(x,y,z))=[d_{\Delta_{\mathbb{H}}}(x,y,z)]^{2-Q}, \tag*{(16)}

where

dΔH(x,y,z)=dΔH(0,(x,y,z))=cn((∣x∣2+∣y∣2)2+16z2)14(17)d_{\Delta_{\mathbb{H}}}(x,y,z)=d_{\Delta_{\mathbb{H}}}(0,(x,y,z))=c_n\left((|x|^{2}+|y|^{2})^{2}+16z^{2}\right)^{\frac{1}{4}} \tag*{(17)}

is the ΔH\Delta_{\mathrm{H}}-gauge for some dimensional constant cnc_n (cf. [9]) and Q=2n+2Q = 2n+2 is the homogeneous dimension of Hn\mathbb{H}^n. The ΔH\Delta_{\mathrm{H}}-gauge dΔHd_{\Delta_{\mathrm{H}}} satisfies the scaling property

dΔH(δλ(x,y,z))=λdΔH(x,y,z)for λ>0.(18)d_{\Delta_{\mathrm{H}}}\left(\delta_\lambda(x,y,z)\right)=\lambda d_{\Delta_{\mathrm{H}}}(x,y,z) \quad\text{for } \lambda>0. \tag*{(18)}

Finally the ΔH\Delta_{\mathrm{H}}-gauge dΔHd_{\Delta_{\mathrm{H}}} is equivalent to the Carnot-Caratheodory distance, which is the geodesic distance of the sub-Riemannian structure on Hn\mathbb{H}^n.

Geodesics in Heisenberg groups

The description of geodesics in Heisenberg groups here follows [16 Section 1.5 and 1.6]. Let us start with momentum functions

PXi=pxi−yi2pzandPYi=pyi+xi2pz.P_{X_i}=p_{x_i}-\frac{y_i}{2}p_z \quad\text{and} \quad P_{Y_i}=p_{y_i}+\frac{x_i}{2}p_z.

on the cotangent bundle

T∗Hn=R2n+1×R2n+1={(x1,⋯ ,xn,y1,⋯ ,yn,z,px1,⋯ ,pxn,py1,⋯ ,pyn,pz)}.T^*\mathbb{H}^n=\mathbb{R}^{2n+1}\times\mathbb{R}^{2n+1}=\left\{(x_1,\cdots,x_n,y_1,\cdots,y_n,z,p_{x_1},\cdots,p_{x_n},p_{y_1},\cdots,p_{y_n},p_z)\right\}.

Then the Hamiltonian for the sub-Riemannian structure associated with the Jacobian generators is

H=12∑i=1n(PXi2+PYi2)=12∑i=1n(pxi2+pyi2−yipxipz+xipyipz+xi2+yi24pz2).H=\frac{1}{2}\sum_{i=1}^{n}\left(P_{X_i}^2+P_{Y_i}^2\right)=\frac{1}{2}\sum_{i=1}^{n}\left(p_{x_i}^2+p_{y_i}^2-y_ip_{x_i}p_z+x_ip_{y_i}p_z+\frac{x_i^2+y_i^2}{4}p_z^2\right).

Therefore the Hamiltonian dynamics on the cotangent bundle is

x˙i=PXi,y˙i=PYi,z˙=∑i=1n(−yi2PXi+xi2PYi),p˙xi=−pz2PYi,p˙yi=pz2PXi,p˙z=0.\begin{aligned} \dot{x}_i&=P_{X_i},\quad\dot{y}_i=P_{Y_i},\quad\dot{z}=\sum_{i=1}^{n}\left(-\frac{y_i}{2}P_{X_i}+\frac{x_i}{2}P_{Y_i}\right),\\ \dot{p}_{x_i}&=-\frac{p_z}{2}P_{Y_i},\quad\dot{p}_{y_i}=\frac{p_z}{2}P_{X_i},\quad\dot{p}_z=0. \end{aligned}

This describes geodesics of the sub-Riemannian structure on Hn\mathbb{H}^n.

Normal geodesics from the origin

For geodesics from the origin with initial momentum (px,py,0)(p_x,p_y,0), we have

x(t)=pxt,y(t)=pyt,z(t)=0,x(t)=p_xt,\quad y(t)=p_yt,\quad z(t)=0,

which is a straight ray from the origin in XY-hyperplane. For geodesics from the origin with initial momentum (px,py,pz)(p_x,p_y,p_z), where pz≠0p_z\ne0, we have

x(t)=pxpzsin⁡pzt+pypzcos⁡pzt−pypz,y(t)=−pxpzcos⁡pzt+pypzsin⁡pzt+pxpz,\begin{aligned} x(t)&=\frac{p_x}{p_z}\sin p_zt+\frac{p_y}{p_z}\cos p_zt-\frac{p_y}{p_z},\\ y(t)&=-\frac{p_x}{p_z}\cos p_zt+\frac{p_y}{p_z}\sin p_zt+\frac{p_x}{p_z}, \end{aligned}

which is a circle in XY-hyperplane centered at (−pypz,pxpz)\left(-\frac{p_y}{p_z},\frac{p_x}{p_z}\right) and with the radius ρ=∣px∣2+∣py∣2∣pz∣\rho=\frac{\sqrt{|p_x|^2+|p_y|^2}}{|p_z|} passing through the origin, and

z(t)=∣px∣2+∣py∣22pz2(pzt−sin⁡pzt).z(t)=\frac{|p_x|^2+|p_y|^2}{2p_z^2}(p_zt-\sin p_zt).

To summarize, the geodesics from the origin are called normal geodesics and minimal, when they only possibly meet the zz-axis the second time at the end. They are all horizontal lifts of circles in XYXY-hyperplane passing the origin, including the straight lines in XYXY-hyperplane passing the origin, which are considered to be circles of an infinite radius. The geodesic length of the horizontal lift of a circle of a finite radius is the circumference of the circle, and only the straight lines in XYXY-hyperplane can go an infinite length.

4.2.2. The scaling property of geodesics from the origin. As in Euclidean spaces, the dilation takes normal geodesics to normal geodesics and scales the length. However, in contrast to Euclidean cases, the dilation does not simply extend a normal geodesic from the origin, instead, it takes it to another completely distinct normal geodesic from the origin, unless the geodesic is a straight line passing the origin inside XYXY-hyperplane. This distinction is significant for geometry as well as analysis on homogeneous Carnot groups.

The anisotropic nature of Heisenberg groups Hn

In analysis to study potentials for sub-Laplacians on Hn\mathbb{H}^n, we want to consider level sets of the ΔH\Delta_{\mathbb{H}}-gauge.

SrΔH={(x,y,z):dΔH(x,y,z)=r}.S_r^{\Delta_{\mathbb{H}}}=\{(x,y,z):d_{\Delta_{\mathbb{H}}}(x,y,z)=r\}.

Let GrΔHG_r^{\Delta_{\mathbb{H}}} be the collection of normal geodesics from the origin that reach to SrΔHS_r^{\Delta_{\mathbb{H}}}, excluding those that stop at zz-axis before they reach to SrΔHS_r^{\Delta_{\mathbb{H}}}. Then, we consider the set

Dr1,r2ΔH={(x,y,z)∈Sr2ΔH:(x,y,z) on the trajectory of γ∈Gr1ΔH} for r2<r1.(19)D_{r_1,r_2}^{\Delta_{\mathbb{H}}}=\{(x,y,z)\in S_{r_2}^{\Delta_{\mathbb{H}}}:(x,y,z)\text{ on the trajectory of }\gamma\in G_{r_1}^{\Delta_{\mathbb{H}}}\}\text{ for }r_2<r_1. \tag*{(19)}

It is easily seen that Gr1ΔH⊈Gr2ΔHG_{r_1}^{\Delta_{\mathbb{H}}}\not\subseteq G_{r_2}^{\Delta_{\mathbb{H}}} for r2<r1r_2<r_1 and therefore

Dr1,r2ΔH⊈Dr1,r2′ΔH for r2<r1<r1′.(20)D_{r_1,r_2}^{\Delta_{\mathbb{H}}}\not\subseteq D_{r_1,r_2'}^{\Delta_{\mathbb{H}}}\text{ for }r_2<r_1<r_1'. \tag*{(20)}

In fact, by the summary in § 4.2.1, we have

Dr,1ΔH→S1ΔH∩{z=0} as r→∞,(21)D_{r,1}^{\Delta_{\mathbb{H}}}\to S_1^{\Delta_{\mathbb{H}}}\cap\{z=0\}\text{ as }r\to\infty, \tag*{(21)}

where S1ΔH∩{z=0}S_1^{\Delta_{\mathbb{H}}}\cap\{z=0\} is nothing but the round sphere S2n−1(1/cn)S^{2n-1}(1/c_n) of radius 1/cn1/c_n in the XYXY-hyperplane by (17). Finally we want to mention that

Dλr1,λr2ΔH=δλ(Dr1,r2ΔH) for λ>0,D_{\lambda r_1,\lambda r_2}^{\Delta_{\mathbb{H}}}=\delta_\lambda(D_{r_1,r_2}^{\Delta_{\mathbb{H}}})\text{ for }\lambda>0,

due to the scaling property of both ΔH\Delta_{\mathbb{H}} gauge (18) and normal geodesics (cf. Section 4.2.2).

Geodesically unavoidable H-thin sets

Let cap⁡ΔH∗\operatorname{cap}_{\Delta_{\mathbb{H}}}^{*} be the exterior capacity defined in Definition 2.5 associated with the canonical sub-Laplacian ΔH\Delta_{\mathbb{H}} in Heisenberg group Hn\mathbb{H}^n. First, in the light of the scaling property (4) of fundamental solution ΓΔH\Gamma_{\Delta_{\mathbb{H}}}, we have

cap⁡ΔH∗(δλ(E))=λQ−2cap⁡ΔH∗(E).(22)\operatorname{cap}_{\Delta_{\mathbb{H}}}^{*}(\delta_\lambda(E))=\lambda^{Q-2}\operatorname{cap}_{\Delta_{\mathbb{H}}}^{*}(E). \tag*{(22)}

Lemma 4.1. Let S2n−1(1/cn)=S1ΔH∩{z=0}S^{2n-1}(1/c_n)=S_1^{\Delta_{\mathbb{H}}}\cap\{z=0\}. Then

cap⁡ΔH∗(S1ΔH∩{z=0})=0.\operatorname{cap}_{\Delta_{\mathbb{H}}}^{*}(S_1^{\Delta_{\mathbb{H}}}\cap\{z=0\})=0.

Proof. For simplicity, we want to show cap⁡ΔH∗(S2n−1)=0\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(\mathbb{S}^{2n-1})=0 instead for the unit round sphere S2n−1\mathbb{S}^{2n-1} in XYXY-hyperplane of Hn\mathbb{H}^{n} due to (22). According to [3 Lemma 11.5.11], it suffices show that S2n−1\mathbb{S}^{2n-1} is ΔH\Delta_{\mathrm H}-polar. Let μ0(A)=m0(A∩S2n−1)\mu_{0}(A)=m_{0}(A\cap\mathbb{S}^{2n-1}) be the Radon measure on Hn\mathbb{H}^{n}, where m0m_{0} is the standard Lebesgue measure on S2n−1\mathbb{S}^{2n-1}. Then we calculate the potential

RΔHμ0,R2n+1(x,y,0)=∫S2n−1ΓΔH((x,y,0)−1∘u) dm0(u)=∞(23)R_{\Delta_{\mathrm H}}^{\mu_{0},\mathbb{R}^{2n+1}}(x,y,0)=\int_{\mathbb{S}^{2n-1}}\Gamma_{\Delta_{\mathrm H}}\left((x,y,0)^{-1}\circ u\right)\,dm_{0}(u)=\infty \tag*{(23)}

for (x,y,0)∈S2n−1⊂Hn(x,y,0)\in\mathbb{S}^{2n-1}\subset\mathbb{H}^{n}. In fact, due to the symmetry, one only needs to calculate the potential RΔHμ0,R2n+1R_{\Delta_{\mathrm H}}^{\mu_{0},\mathbb{R}^{2n+1}} at x0=(1,0,⋯ ,0)x_{0}=(1,0,\cdots,0) and y0=(0,0,⋯ ,0)y_{0}=(0,0,\cdots,0). Notice that

ΓΔH((x0,y0,0)−1∘u)=c−2n2−n((1−ux1)2+uy12)−n2\Gamma_{\Delta_{\mathrm H}}\left((x_{0},y_{0},0)^{-1}\circ u\right)=c^{-2n}2^{-n}\left((1-u_{x_{1}})^{2}+u_{y_{1}}^{2}\right)^{-\frac{n}{2}}

for u=(ux,uy,0)∈S2n−1⊂Hnu=(u_{x},u_{y},0)\in\mathbb{S}^{2n-1}\subset\mathbb{H}^{n}, by (14), (16), and (17). To complete (23), one uses the spherical coordinates on S2n−1\mathbb{S}^{2n-1}, in which

  • ux1=cos⁡θu_{x_{1}}=\cos\theta, uy1=sin⁡θu_{y_{1}}=\sin\theta, where θ∈[0,2π]\theta\in[0,2\pi], and RΔHμ0,R2n+1(x0,y0,0)≥c∫0δθ−1 dθ=∞R_{\Delta_{\mathrm H}}^{\mu_{0},\mathbb{R}^{2n+1}}(x_{0},y_{0},0)\geq c\int_{0}^{\delta}\theta^{-1}\,d\theta=\infty for some δ>0\delta>0 and c>0c>0, when n=1n=1;

  • ux1=cos⁡θ1u_{x_{1}}=\cos\theta_{1}, uy1=sin⁡θ1sin⁡θ2u_{y_{1}}=\sin\theta_{1}\sin\theta_{2}, where θ1∈[0,π]\theta_{1}\in[0,\pi] and θ2∈[−π2,π2]\theta_{2}\in[-\frac{\pi}{2},\frac{\pi}{2}], and

RΔHμ0,R2n+1(x0,y0,0)≥c∫0δθ1n−2∫0θ1(θ12+θ22)−n2 dθ2 dθ1≥2−n2c∫0δθ1−1 dθ1=∞R_{\Delta_{\mathrm H}}^{\mu_{0},\mathbb{R}^{2n+1}}(x_{0},y_{0},0)\geq c\int_{0}^{\delta}\theta_{1}^{n-2}\int_{0}^{\theta_{1}}(\theta_{1}^{2}+\theta_{2}^{2})^{-\frac{n}{2}}\,d\theta_{2}\,d\theta_{1}\geq2^{-\frac{n}{2}}c\int_{0}^{\delta}\theta_{1}^{-1}\,d\theta_{1}=\infty

for some δ>0\delta>0 and c>0c>0, when n≥2n\geq2.

Consequently, by (11.19) in [3 Lemma 11.5.10] as well as (20) and (21), we have

Lemma 4.2. lim⁡r→∞cap⁡ΔH∗(Dr,1ΔH)=0\lim_{r\to\infty}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(D^{\Delta_{\mathrm H}}_{r,1})=0.

Therefore we may select a sequence ki→∞k_{i}\to\infty such that ∑i=1∞cap⁡ΔH∗(Dαˉ−ki,1ΔH)<∞\sum_{i=1}^{\infty}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(D^{\Delta_{\mathrm H}}_{\bar{\alpha}^{-k_{i},1}})<\infty and set

E=⋃i=1∞Dαˉ−ki,αˉ−2kiΔH(24)E=\bigcup_{i=1}^{\infty}D^{\Delta_{\mathrm H}}_{\bar{\alpha}^{-k_{i},\bar{\alpha}^{-2k_{i}}}} \tag*{(24)}

Proposition 4.1. For the set EE given in (24), ∑n=1∞αncap⁡ΔH∗(E∩Cn)<∞\sum_{n=1}^{\infty}\alpha^{n}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(E\cap\mathcal{C}_{n})<\infty. In other words, the set EE in (24) is ΔH\Delta_{\mathrm H}-thin at the origin according to Definition 2.6.

Proof. First, from (24),

\sum_{n=1}^{\infty}\alpha^{n}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(E\cap\mathcal{C}_{n})=\sum_{i=1}^{\infty}\alpha^{2k_{i}}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}(E\cap\mathcal{C}_{2k_{i}})=\sum_{i=1}^{\infty}\alpha^{2k_{i}}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}\left(D^{\Delta_{\mathrm H}}_{\bar{\alpha}^{-k_{i},\bar{\alpha}^{-2k_{i}}}\right).}

Then, by the scaling property (22) and (8),

\sum_{i=1}^{\infty}\alpha^{2k_{i}}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}\left(D^{\Delta_{\mathrm H}}_{\bar{\alpha}^{-k_{i},\bar{\alpha}^{-2k_{i}}}\right)\leq3\sum_{i=1}^{\infty}\operatorname{cap}^{*}_{\Delta_{\mathrm H}}\left(D^{\Delta_{\mathrm H}}_{\bar{\alpha}^{-k_{i},1}}\right)<\infty.}

Consequently,

Theorem 4.1. In Heisenberg group Hn\mathbb{H}^{n} the set EE in (24) is ΔH\Delta_{\mathbb{H}}-thin at the origin according to Definition 2.6. However, any geodesic from the origin intersects with EE.

Proof. This is simply because that a geodesic is always normal initially and that a normal geodesic always intersects with EE. In fact, due to the definition (19), any normal geodesic that reaches to the level set of Sαˉ−kiΔHS^{\Delta_{\mathbb{H}}}_{\bar{\alpha}-k_i} intersects with Dαˉ−ki,αˉ−2kiD_{\bar{\alpha}-k_i,\bar{\alpha}-2k_i} as ki→∞k_i \to\infty. □

On thin sets in polarizable groups and geometric applications

It is worth pointing out that, for a version of the existence of rays escaping from a thin set in Carnot groups to utilize the geometric completeness, one may use horizontal curves which need not be geodesics. Also a measurement for a set to be of small size in S∖ZS \setminus\mathcal{Z} may simply be a measure instead of the capacity. With those in mind, we will employ the foliation of horizontal radial curves for the polar coordinates in polarizable Carnot groups constructed in [2] to establish Theorem 5.1. G\mathbb{G} is assumed to be polarizable (cf. [2 Definition 2.12]) throughout this section.

Polarizable groups and polar coordinates

5.1. Polarizable groups and polar coordinates. In this subsection we follow [2] to introduce polar coordinates using the foliation of horizontal radial curves in polarizable Carnot groups. To introduce flow lines as the horizontal radial curves, we let

S={x∈RN:dΔG(x)=1} and Z={0}∪{x∈RN:∇0dΔG(x)=0},S = \{x \in\mathbb{R}^{N} : d_{\Delta_{\mathbb{G}}}(x) = 1\} \text{ and } \mathcal{Z} = \{0\} \cup\{x \in\mathbb{R}^{N} : \nabla_{0}d_{\Delta_{\mathbb{G}}}(x) = 0\},

where ∇0dΔG\nabla_{0}d_{\Delta_{\mathbb{G}}} is the horizontal gradient of the ΔG\Delta_{\mathbb{G}}-gauge dΔGd_{\Delta_{\mathbb{G}}}. And in RN∖Z\mathbb{R}^{N} \setminus\mathcal{Z}, one considers the flow lines ϕ(s,g):RN∖Z→RN∖Z\phi(s,g) : \mathbb{R}^{N} \setminus\mathcal{Z} \to\mathbb{R}^{N} \setminus\mathcal{Z} that satisfy

{∂∂sϕ(s,g)=dΔG(ϕ(s,g))s∇0dΔG(ϕ(s,g))∥∇0dΔG(ϕ(s,g))∥02,ϕ(1,g)=g(25)\begin{cases} \frac{\partial}{\partial s}\phi(s,g) = \frac{d_{\Delta_{\mathbb{G}}}(\phi(s,g))}{s} \frac{\nabla_{0}d_{\Delta_{\mathbb{G}}}(\phi(s,g))}{\lVert\nabla_{0}d_{\Delta_{\mathbb{G}}}(\phi(s,g))\rVert_{0}^{2}},\\ \phi(1,g) = g \tag*{(25)} \end{cases}

where ∥⋅∥0\lVert\cdot\rVert_{0} is the norm on the horizontal space with respect to which the Jacobian generators form the orthonormal basis.

Lemma 5.1. ([2 Lemma 3.3]) Let G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N}, \circ, \delta_{\lambda}) be polarizable. Then the flow lines ϕ(s,g)\phi(s,g) satisfies

  • dΔG(ϕ(s,g))=sdΔG(g)d_{\Delta_{\mathbb{G}}}(\phi(s,g)) = s d_{\Delta_{\mathbb{G}}}(g) for s>0s > 0 and g∈RN∖Zg \in\mathbb{R}^{N} \setminus\mathcal{Z};

  • the velocity ∂∂sϕ(s,g)\frac{\partial}{\partial s}\phi(s,g) is horizontal and the speed ∥∂∂sϕ(s,g)∥0=dΔG(g)∥∇0dΔG(g)∥0\left\lVert\frac{\partial}{\partial s}\phi(s,g)\right\rVert_{0} = \frac{d_{\Delta_{\mathbb{G}}}(g)}{\lVert\nabla_{0}d_{\Delta_{\mathbb{G}}}(g)\rVert_{0}} is independent of ss.

Lemma 5.2. ([2 Theorem 3.7]) Let G=(RN,∘,δλ)\mathbb{G} = (\mathbb{R}^{N}, \circ, \delta_{\lambda}) be polarizable. There exists a unique Radon measure σ\sigma on S∖ZS \setminus\mathcal{Z} such that for all u∈L1(RN)u \in L^{1}(\mathbb{R}^{N})

∫RNu(x) dx=∫0∞∫S∖Zu(ϕ(s,v))sQ−1 dσ(v) ds.(26)\int_{\mathbb{R}^{N}} u(x)\,dx = \int_{0}^{\infty} \int_{S \setminus\mathcal{Z}} u(\phi(s,v))s^{Q-1}\,d\sigma(v)\,ds. \tag*{(26)}

Thanks to [2 Proposition 5.6], H-type groups (cf. [2 Definition 5.1]) are polarizable. And, according to [6], H-type groups include those that are the nilpotent part of Iwasawa decomposition of rank one groups: SU(n+1,1)SU(n+1,1), Sp(n+1,1)Sp(n+1,1) and F4(−20)\mathbb{F}_{4(-20)}, which are the relevant groups for our geometric applications.

The flow lines ϕ(s,⋅)\phi(s,\cdot) as horizontal radial curves

To fix the upper bound for the speed of flow lines ϕ(s,⋅)\phi(s,\cdot) in the light of Lemma 5.1, let us introduce

Sa={g∈S:∥∇0dΔG(g)∥0>a}andGa={ϕ(s,g):g∈Sa and s∈(0,∞)}.(27)S_a=\{g\in S:\|\nabla_0d_{\Delta_G}(g)\|_0>a\}\quad\text{and}\quad G_a=\{\phi(s,g):g\in S_a\text{ and }s\in(0,\infty)\}. \tag*{(27)}

Then

Lemma 5.3. Let G\mathbb{G} be a polarizable group and g∈Sag\in S_a for a>0a>0. Then

dΔG(ϕ(s,g)−1∘ϕ(r,g))≤Ca∣r−s∣(28)d_{\Delta_G}\left(\phi(s,g)^{-1}\circ\phi(r,g)\right)\leq C_a|r-s| \tag*{(28)}

for some constant CaC_a, which depends only on the number aa.

Proof. It suffices to prove (28) for the Carnot-Caratheodory norm dcd_c instead of the gauge dΔGd_{\Delta_G}. Because they are equivalent on any Carnot group. To finish the proof, one may use the speed upper bound of the horizontal radial curve ϕ(t,g)\phi(t,g) connecting ϕ(r,g)\phi(r,g) and ϕ(s,g)\phi(s,g). □

Let

P(x)=P(ϕ(s,g))=g:RN∖Z→S∖Z\mathcal{P}(x)=\mathcal{P}(\phi(s,g))=g:\mathbb{R}^{N}\setminus\mathcal{Z}\to S\setminus\mathcal{Z}

be the projection based on the polar coordinates by horizontal radial curves x=ϕ(s,g)x=\phi(s,g) for g∈S∖Zg\in S\setminus\mathcal{Z}. The following is the key lemma of this note that facilitates the existence of horizontal radial curves escaping from a thin set. From now on we will always work with CnC_n that are centered at the origin.

Lemma 5.4. Suppose that G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^{N},\circ,\delta_\lambda) is polarizable. Let FF be an open set and a>0a>0 be appropriately small. Then

cap⁡ΔG∗(F∩Cn∩Ga)≥Cαˉ−(Q−2)nσ(P(F∩Cn∩Ga))Q−2Q−1,(29)\operatorname{cap}^{*}_{\Delta_G}(F\cap C_n\cap G_a)\geq C\bar{\alpha}^{-(Q-2)n}\sigma\left(\mathcal{P}(F\cap C_n\cap G_a)\right)^{\frac{Q-2}{Q-1}}, \tag*{(29)}

where the measure σ\sigma is the one in Lemma 5.2 and CC depends on aa.

Before we prove this key lemma, we first use it to prove the existence of horizontal radial curves escaping from a thin set in polarizable Carnot groups.

Theorem 5.1. Let G=(RN,∘,δλ)\mathbb{G}=(\mathbb{R}^{N},\circ,\delta_\lambda) be polarizable. And let EE be a set that is thin at the origin in G\mathbb{G}. Then there always exists a flow line ϕ(s,g)\phi(s,g) given by (25) that does not intersect with EE, at least in a small neighborhood of the origin.

Proof. According to Definition 2.6, we consider, for a thin set EE,

∞>∑n=1∞αˉn(Q−2)cap⁡ΔG∗(E∩Cn)≥∑n=1∞αˉn(Q−2)cap⁡ΔG∗(E∩Cn∩Ga)≥12∑n:cap⁡ΔG∗(E∩Cn∩Ga)>0αˉn(Q−2)cap⁡ΔG∗(F∩Cn∩Ga)≥C∑n:cap⁡ΔG∗(E∩Cn∩Ga)>0(σ(P(F∩Cn∩Ga)))Q−2Q−1≥C∑n:cap⁡ΔG∗(E∩Cn∩Ga)>0σ(P(F∩Cn∩Ga))≥Cσ(⋃n:cap⁡ΔG∗(E∩Cn∩Ga)>0P(F∩Cn∩Ga)).(30)\begin{aligned} \infty&> \sum_{n=1}^{\infty} \bar{\alpha}^{n(Q-2)}\operatorname{cap}_{\Delta_G}^{*}(E \cap\mathcal{C}_n) \ge\sum_{n=1}^{\infty} \bar{\alpha}^{n(Q-2)}\operatorname{cap}_{\Delta_G}^{*}(E \cap\mathcal{C}_n \cap\mathcal{G}_a) \\ &\ge\frac{1}{2}\sum_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)>0}} \bar{\alpha}^{n(Q-2)}\operatorname{cap}_{\Delta_G}^{*}(F \cap\mathcal{C}_n \cap\mathcal{G}_a) \\ &\ge C\sum_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)>0}} \left(\sigma\left(\mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a)\right)\right)^{\frac{Q-2}{Q-1}} \\ &\ge C\sum_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)>0}} \sigma\left(\mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a)\right) \\ &\ge C\sigma\left(\bigcup_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)>0}} \mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a)\right). \tag*{(30)} \end{aligned}

and

1=∑n=1∞2−n≥∑n:cap⁡ΔG∗(E∩Cn∩Ga)=0αˉn(Q−2)cap⁡ΔG∗(F∩Cn∩Ga)≥σ(⋃n:cap⁡ΔG∗(E∩Cn∩Ga)=0P(F∩Cn∩Ga)).(31)\begin{aligned} 1=\sum_{n=1}^{\infty}2^{-n} &\ge\sum_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)=0}} \bar{\alpha}^{n(Q-2)}\operatorname{cap}_{\Delta_G}^{*}(F \cap\mathcal{C}_n \cap\mathcal{G}_a) \\ &\ge\sigma\left(\bigcup_{\substack{n:\operatorname{cap}_{\Delta_G}^{*}(E\cap\mathcal{C}_n\cap\mathcal{G}_a)=0}} \mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a)\right). \tag*{(31)} \end{aligned}

for som appropriate open set F⊃EF \supset E due to the definition of the exterior capacity (cf. Definition 2.5). This implies Sa∖⋃n=k∞P(F∩Cn∩Ga)S_a \setminus\bigcup_{n=k}^{\infty}\mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a) is not empty when kk is sufficiently large. Because

σ(⋂k=1∞⋃n=k∞P(F∩Cn∩Ga))=0<σ(Sa),(32)\sigma\left(\bigcap_{k=1}^{\infty}\bigcup_{n=k}^{\infty}\mathcal{P}(F \cap\mathcal{C}_n \cap\mathcal{G}_a)\right)=0<\sigma(S_a), \tag*{(32)}

which is like [15 Lemma 8.2 in Chapter 2] and suffices for our purpose. σ(Sa)\sigma(S_a) is positive when aa is sufficiently small (see [2 Proposition 2.18 (i), Lemma 3.3, and Theorem 3.7]). □

We now prove Lemma 5.4 in the following subsections. The idea is to construct a competitive Radon measure supported in F∩C0∩GaF \cap\mathcal{C}_0 \cap\mathcal{G}_a for open sets FF. Then Lemma 5.4 in general follows in the light of the scaling property of the homogeneously induced measure on ϕ(s,S∖Z)\phi(s, S \setminus\mathcal{Z}):

σs(ϕ(s,A))=sQ−1σ(A)(33)\sigma_s(\phi(s,A))=s^{Q-1}\sigma(A) \tag*{(33)}

A competitive Radon measure

For a set F∩C0∩GaF \cap\mathcal{C}_0 \cap\mathcal{G}_a and a Radon measure μ\mu supported in F∩C0∩GaF \cap\mathcal{C}_0 \cap\mathcal{G}_a is said to be competitive if it satisfies

(1) For a positive constant AA

σ(P(F∩C0∩Ga))≥μ(F∩C0∩Ga)≥Aσ(P(F∩C0∩Ga)).(34)\sigma(\mathcal{P}(F \cap\mathcal{C}_0 \cap\mathcal{G}_a)) \ge\mu(F \cap\mathcal{C}_0 \cap\mathcal{G}_a) \ge A\sigma(\mathcal{P}(F \cap\mathcal{C}_0 \cap\mathcal{G}_a)). \tag*{(34)}

(2) For δ0>0\delta_0 > 0 there exists a constant BB

μ(BΔG(p,ρ)∩F∩C0∩Ga)≤BρQ−1(35)\mu\left(B_{\Delta_G}(p,\rho) \cap F \cap C_0 \cap G_a\right) \le B\rho^{Q-1} \tag*{(35)}

when dΔG(p)=r≥δ0d_{\Delta_G}(p)=r\ge\delta_0, where BΔG(p,ρ)={g∈G:dΔG(p−1∘g)<ρ}B_{\Delta_G}(p,\rho)=\{g\in G:d_{\Delta_G}(p^{-1}\circ g)<\rho\}.

A competitive Radon measure for spherical wedge shells

It is easy to have a competitive measure for a spherical wedge shell with respect to the flow lines ϕ(s,⋅)\phi(s,\cdot). Here a spherical wedge shell with respect to the flow lines ϕ(s,⋅)\phi(s,\cdot) is

Wr1,r2A={ϕ(s,g)∈G:g∈A⊂Sa and s∈(r1,r2]}=⋃s∈(r1,r2]ϕ(s,A),(36)W^A_{r_1,r_2}=\{\phi(s,g)\in G:g\in A\subset S_a\text{ and }s\in(r_1,r_2]\}=\bigcup_{s\in(r_1,r_2]}\phi(s,A), \tag*{(36)}

where ϕ(s,A)={ϕ(s,g):g∈A⊂Sa}\phi(s,A)=\{\phi(s,g):g\in A\subset S_a\} is a cap in GaG_a for a measurable set AA and s>0s>0. Clearly we have P(Wr1,r2A)=A\mathcal{P}(W^A_{r_1,r_2})=A. A candidate for competitive measures is simply the restriction

μr,A(E)=r1−Qσr(E∩ϕ(r,A))=σ(P(E∩ϕ(r,A)))\mu_{r,A}(E)=r^{1-Q}\sigma_r(E\cap\phi(r,A))=\sigma(\mathcal{P}(E\cap\phi(r,A)))

for a fixed r∈(r1,r2]⊂(αˉ−1,1]r\in(r_1,r_2]\subset(\bar{\alpha}^{-1},1], which clearly is a Radon measure supported on the cap ϕ(r,A)\phi(r,A) in Wr1,r2AW^A_{r_1,r_2} and satisfies (34). In order to verify (35) for μr,A\mu_{r,A}, we recall the improved pseudo-triangle inequality [3 (12.15a)]

dΔG(u−1∘w)=dΔG((u−1∘v)∘(v−1∘w))≤βdΔG(u−1∘v)+dΔG(v−1∘w)(37)d_{\Delta_G}(u^{-1}\circ w)=d_{\Delta_G}((u^{-1}\circ v)\circ(v^{-1}\circ w))\le\beta d_{\Delta_G}(u^{-1}\circ v)+d_{\Delta_G}(v^{-1}\circ w) \tag*{(37)}

for some β≥1\beta\ge1. Then we observe

Lemma 5.5. Let Σp,ρ=P(BΔG(p,ρ)∩Ga)⊂Sa\Sigma_{p,\rho}=\mathcal{P}(B_{\Delta_G}(p,\rho)\cap G_a)\subset S_a for dΔG(p)=r>0d_{\Delta_G}(p)=r>0. Then

ϕ(t,Σp,ρ)⊂BΔG(p,(1+2Ca)βρ)for t∈(r,r+ρ).(38)\phi(t,\Sigma_{p,\rho})\subset B_{\Delta_G}(p,(1+2C_a)\beta\rho)\quad\text{for }t\in(r,r+\rho). \tag*{(38)}

Consequently we have

σ(Σp,ρ)≤ωG((1+2Ca)β)QrQ−1ρQ−1,(39)\sigma(\Sigma_{p,\rho})\le\frac{\omega_G((1+2C_a)\beta)^Q}{r^{Q-1}}\rho^{Q-1}, \tag*{(39)}

where ωG\omega_G is the Euclidean volume of BΔG(0,1)B_{\Delta_G}(0,1) in GG and CaC_a is from (28).

Proof. From (37), for any ϕ(s,g)∈BΔG(p,ρ)∩Ga\phi(s,g)\in B_{\Delta_G}(p,\rho)\cap G_a and g∈Σp,ρ⊂Sag\in\Sigma_{p,\rho}\subset S_a, one has

dΔG(p−1∘ϕ(r,g))≤βdΔG(p−1∘ϕ(s,g))+dΔG(ϕ(s,g)−1∘ϕ(r,g))≤βρ+Ca∣r−s∣,d_{\Delta_G}(p^{-1}\circ\phi(r,g))\le\beta d_{\Delta_G}(p^{-1}\circ\phi(s,g))+d_{\Delta_G}(\phi(s,g)^{-1}\circ\phi(r,g))\le\beta\rho+C_a|r-s|,

where CaC_a is from (28). On the other hand, from (11), one also has

dΔG(p−1∘ϕ(s,g))≥1β∣r−s∣d_{\Delta_G}(p^{-1}\circ\phi(s,g))\ge\frac{1}{\beta}|r-s|

which implies ∣r−s∣≤βρ|r-s|\le\beta\rho if ϕ(s,g)∈BΔG(p,ρ)\phi(s,g)\in B_{\Delta_G}(p,\rho). Therefore

dΔG(p−1∘ϕ(r,g))≤(1+Ca)βρfor all g∈Σp,ρ.d_{\Delta_G}(p^{-1}\circ\phi(r,g))\le(1+C_a)\beta\rho\quad\text{for all }g\in\Sigma_{p,\rho}.

Applying (37) again, one gets, for g∈Σp,ρg\in\Sigma_{p,\rho},

dΔG(p−1∘ϕ(t,g))≤βCa∣t−r∣+(1+Ca)βρ<(1+2Ca)βρwhen t∈(r,r+ρ).d_{\Delta_G}(p^{-1}\circ\phi(t,g))\le\beta C_a|t-r|+(1+C_a)\beta\rho<(1+2C_a)\beta\rho\quad\text{when }t\in(r,r+\rho).

Therefore (38) is verified. To complete the proof, we invoke Lemma 5.2 and the fact that the Euclidean measure is the Haar measure for GG. Therefore

ωG((1+2Ca)β)QρQ=∫BΔG(p,(1+2Ca)βρ)dx≥∫rr+ρ∫Σp,ρsQ−1 dσds≥rQ−1ρσ(Σp,ρ).\omega_G((1+2C_a)\beta)^Q\rho^Q=\int_{B_{\Delta_G}(p,(1+2C_a)\beta\rho)}dx\ge\int_r^{r+\rho}\int_{\Sigma_{p,\rho}}s^{Q-1}\,d\sigma ds\ge r^{Q-1}\rho\sigma(\Sigma_{p,\rho}).

Consequently (35) for μr,A\mu_{r,A} holds from Lemme 5.5.

Corollary 5.1. Let dΔG(p)=r≥δ0>0d_{\Delta_G}(p)=r\ge\delta_0>0 and Wr1,r2A⊂C0∩GaW^A_{r_1,r_2}\subset C_0\cap G_a. Then

μr,A(BΔG(p,ρ)∩Wr1,r2A)≤σ(P(BΔG(p,ρ)∩Ga))≤CρQ−1.\mu_{r,A}(B_{\Delta_G}(p,\rho)\cap W^A_{r_1,r_2})\le\sigma(\mathcal{P}(B_{\Delta_G}(p,\rho)\cap G_a))\le C\rho^{Q-1}.

Proof. Under the assumption that dΔG(p)=r≥δ0>0d_{\Delta_G}(p)=r\ge\delta_0>0, (39) implies

μr,A(Wr1,r2A∩BΔG(p,ρ))≤σ(A∩Σp,ρ)≤σ(Σp,ρ)≤CρQ−1(40)\mu_{r,A}(W^A_{r_1,r_2}\cap B_{\Delta_G}(p,\rho))\le\sigma(A\cap\Sigma_{p,\rho})\le\sigma(\Sigma_{p,\rho})\le C\rho^{Q-1} \tag*{(40)}

for some constant CC determined by δ0\delta_0 and aa.

A competitive measure for sets F∩C0∩GaF\cap C_0\cap G_a

To construct a competitive measure for a set F∩C0∩GaF\cap C_0\cap G_a, where FF is open, we want to pick up adequately many spherical wedge shells inside F∩C0∩GaF\cap C_0\cap G_a such that the union of their projections approaches the projection of AA in measure in S∖ZS\setminus\mathcal{Z}. The strategy is to use the grids of thinner and thinner spherical shells to divide AA.

For natural numbers k=1,2,⋯k=1,2,\cdots and l=0,1,⋯ ,2k−1l=0,1,\cdots,2^k-1, let slk=αˉ−1+2−kls_l^k=\bar{\alpha}^{-1+2^{-k}l} and let

Slk={ϕ(s,g):s∈(slk,sl+1k] and g∈S∖Z}={x∈RN∖Z:dΔG(x)∈(slk,sl+1k]}(41)S_l^k=\{\phi(s,g):s\in(s_l^k,s_{l+1}^k]\text{ and }g\in S\setminus\mathcal{Z}\}=\{x\in\mathbb{R}^N\setminus\mathcal{Z}:d_{\Delta_G}(x)\in(s_l^k,s_{l+1}^k]\} \tag*{(41)}

be the spherical shells with respect to the gauge dΔGd_{\Delta_G}. For an open set A⊂C0∩GaA\subset C_0\cap G_a, we denote

Alk=A∩SlkandAt={ϕ(s,g)∈A:s=t and g∈Sa}.A_l^k=A\cap S_l^k\quad\text{and}\quad A_t=\{\phi(s,g)\in A:s=t\text{ and }g\in S_a\}.

Then we consider

A‾lk={ϕ(s,g):g∈⋂t∈(slk,sl+1k]P(At) and s∈(slk,sl+1k]},\underline{A}_l^k=\{\phi(s,g):g\in\bigcap_{t\in(s_l^k,s_{l+1}^k]}\mathcal{P}(A_t)\text{ and }s\in(s_l^k,s_{l+1}^k]\},

which is the largest spherical wedge shell inside AlkA_l^k of the radius between slks_l^k and sl+1ks_{l+1}^k at kkth level with respect to the flow lines ϕ(s,⋅)\phi(s,\cdot). Hence A‾k=⋃l=02k−1A‾lk⊂A\underline{A}^k=\bigcup_{l=0}^{2^k-1}\underline{A}_l^k\subset A, where A‾k\underline{A}^k is a union of spherical wedge shells inside AA at kkth level and monotone increasing as kk increases. To avoid overlaps or multiplicities, for each kk, we let

B‾lk={ϕ(s,g):g∈(⋂t∈(slk,sl+1k]P(At))∖⋃m=0l−1P(A‾mk) and s∈(slk,sl+1k]}⊂Alk\underline{B}_l^k=\left\{\phi(s,g):g\in\left(\bigcap_{t\in(s_l^k,s_{l+1}^k]}\mathcal{P}(A_t)\right)\setminus\bigcup_{m=0}^{l-1}\mathcal{P}(\underline{A}_m^k)\text{ and }s\in(s_l^k,s_{l+1}^k]\right\}\subset A_l^k

inductively from l=0l=0 to l=2k−1l=2^k-1 (BlkB_l^k may be empty). And we let Bk=⋃l=02k−1Blk⊂Ak⊂AB^k=\bigcup_{l=0}^{2^k-1}B_l^k\subset A^k\subset A. This makes

⋃l=02k−1P(Alk‾)=P(Ak)=P(Bk‾)=⋃l=02k−1P(Blk‾)\bigcup_{l=0}^{2^k-1}\mathcal{P}(\underline{A_l^k})=\mathcal{P}(A^k)=\mathcal{P}(\underline{B^k})=\bigcup_{l=0}^{2^k-1}\mathcal{P}(\underline{B_l^k})

is a union of disjoint subsets in S∖Z\mathbb{S}\setminus\mathcal{Z} and {Blk}l=02k−1\{B_l^k\}_{l=0}^{2^k-1} is a collection of disjoint spherical wedge shells in AA for each kk.

Lemma 5.6. Suppose that A=F∩C0∩GaA=F\cap C_0\cap\mathcal{G}_a, where FF is open. Then {Ak}\{A^k\} is an increasing sequence of unions of spherical wedge shells inside AA as kk increases and

⋃k=1∞Ak=A.(42)\bigcup_{k=1}^{\infty}A^k=A. \tag*{(42)}

Hence σ(P(Ak‾))→σ(P(A))\sigma(\mathcal{P}(\underline{A^k}))\to\sigma(\mathcal{P}(A)) as k→∞k\to\infty, where σ\sigma is the Radon measure on S∖Z\mathbb{S}\setminus\mathcal{Z} in (26). Meanwhile, for each kk, we have

⋃l=02k−1P(Blk‾)=P(Bk‾)=P(Ak‾).(43)\bigcup_{l=0}^{2^k-1}\mathcal{P}(\underline{B_l^k})=\mathcal{P}(\underline{B^k})=\mathcal{P}(\underline{A^k}). \tag*{(43)}

Thus ∑l=02k−1σ(P(Blk‾))=σ(P(Bk‾))=σ(P(Ak‾))→σ(P(A))\sum_{l=0}^{2^k-1}\sigma(\mathcal{P}(\underline{B_l^k}))=\sigma(\mathcal{P}(\underline{B^k}))=\sigma(\mathcal{P}(\underline{A^k}))\to\sigma(\mathcal{P}(A)) as k→∞k\to\infty.

Proof. It suffices to see (42) holds because that AA is open in C0∩GaC_0\cap\mathcal{G}_a with respect to the induced topology on C0∩GaC_0\cap\mathcal{G}_a. □

For each spherical wedge shell Blk‾\underline{B_l^k}, we pick the Radon measure μrlk,P(Blk‾)\mu_{r_l^k,\mathcal{P}(\underline{B_l^k})} for some rlk∈(slk,sl+1k]r_l^k\in(s_l^k,s_{l+1}^k] as in § 5.3. Let

μk=∑l=02k−1μrlk,P(Blk‾).(44)\mu_k=\sum_{l=0}^{2^k-1}\mu_{r_l^k,\mathcal{P}(\underline{B_l^k})}. \tag*{(44)}

Then

μk(A)=∑l=02k−1σ(P(Blk‾))=σ(P(Bk‾))=σ(P(Ak‾))→σ(P(A)) as k→∞.(45)\mu_k(A)=\sum_{l=0}^{2^k-1}\sigma(\mathcal{P}(\underline{B_l^k}))=\sigma(\mathcal{P}(\underline{B^k}))=\sigma(\mathcal{P}(\underline{A^k}))\to\sigma(\mathcal{P}(A))\text{ as }k\to\infty. \tag*{(45)}

To summarize we have

Lemma 5.7. Suppose that A=F∩C0∩GaA=F\cap C_0\cap\mathcal{G}_a, where FF is open. Then the Radon measure μk\mu_k given by (44) supported in AA satisfies both (34) and (35) when kk is sufficiently large.

Proof. (34) holds for μk\mu_k for sufficiently large kk by (45). And (35) holds for μk\mu_k because of Lemma 5.5 and the last inequality in (40) in the proof of Corollary 5.1. Because Wr1,r2AW^A_{r_1,r_2} and AA are not really essential in (40). □

The estimate of capacity - the proof of Lemma 5.4

To estimate the capacity cap⁡ΔG∗(A)\operatorname{cap}_{\Delta_G}^{*}(A), where A=F∩C0∩GaA = F \cap C_0 \cap\mathbb{G}_a and FF is open, we consider the potential of the measure μk\mu_k, for sufficiently large kk, constructed in Section 5.2.3. To invoke Definition 2.5, let KK be any compact subset of AA such that

σ(S∖Z)≥μk(A)≥μk(K)≥12μk(A)≥Cσ(P(A)).(46)\sigma(S \setminus\mathcal{Z}) \ge\mu_k(A) \ge\mu_k(K) \ge\frac{1}{2}\mu_k(A) \ge C\sigma(\mathcal{P}(A)). \tag*{(46)}

Now, let us calculate the potential for the Radon measure μk\mu_k restricted to KK. We claim

RΔGμk,K(p)≤Cμk(K)1Q−1(47)R_{\Delta_G}^{\mu_k,K}(p) \le C\mu_k(K)^{\frac{1}{Q-1}} \tag*{(47)}

for constant CC and all p∈RNp \in\mathbb{R}^{N}. For p∈{p∈RN:inf⁡q∈KdΔG(p−1∘q)≥μk(K)1Q−1}p \in\{p \in\mathbb{R}^{N} : \inf_{q \in K} d_{\Delta_G}(p^{-1} \circ q) \ge\mu_k(K)^{\frac{1}{Q-1}}\}, it is easily calculated that

RΔGμk,K(p)=∫K1dΔG(p−1∘q)Q−2 dμk(q)≤μk(K)1Q−1.R_{\Delta_G}^{\mu_k,K}(p) = \int_K \frac{1}{d_{\Delta_G}(p^{-1} \circ q)^{Q-2}} \,d\mu_k(q) \le\mu_k(K)^{\frac{1}{Q-1}}.

Otherwise, we use (35) to calculate

RΔGμk,K(p)=∫K1dΔG(p−1∘q)Q−2 dμk(q)=(Q−2)∫0∞μk(Bt(p)∩K)tQ−2dtt≤(Q−2)(B∫0μk(K)1Q−1dt+μk(K)∫μk(K)1Q−1∞t1−Qdt)≤(B(Q−2)+1)μk(K)1Q−1,\begin{aligned} R_{\Delta_G}^{\mu_k,K}(p) &= \int_K \frac{1}{d_{\Delta_G}(p^{-1} \circ q)^{Q-2}} \,d\mu_k(q) \\ &= (Q-2)\int_0^\infty\frac{\mu_k(B_t(p)\cap K)}{t^{Q-2}}\frac{dt}{t} \\ &\le(Q-2)\left(B\int_0^{\mu_k(K)^{\frac{1}{Q-1}}}dt+\mu_k(K)\int_{\mu_k(K)^{\frac{1}{Q-1}}}^{\infty}t^{1-Q}dt\right) \\ &\le(B(Q-2)+1)\mu_k(K)^{\frac{1}{Q-1}}, \end{aligned}

provided that

μk(K)1Q−1<12α‾−1.(48)\mu_k(K)^{\frac{1}{Q-1}} < \frac{1}{2}\overline{\alpha}^{-1}. \tag*{(48)}

Because, for some q∈K⊂C0q \in K \subset C_0, dΔG(p−1∘q)<μk(K)1Q−1d_{\Delta_G}(p^{-1} \circ q)<\mu_k(K)^{\frac{1}{Q-1}}, and therefore, by (11),

dΔG(p)≥1β∣dΔG(q)−dΔG(p−1∘q)∣≥12βα‾−1>0.d_{\Delta_G}(p) \ge\frac{1}{\beta}\lvert d_{\Delta_G}(q)-d_{\Delta_G}(p^{-1}\circ q)\rvert\ge\frac{1}{2\beta}\overline{\alpha}^{-1}>0.

This makes (35) is applicable. If (48) is reversed, then we repeat the above calculations after substituting μk(K)1Q−1\mu_k(K)^{\frac{1}{Q-1}} with 12α‾−1\frac{1}{2}\overline{\alpha}^{-1} and arrive at

RΔGμk,K(p)≤Cμk(K)1Q−1+Cα‾Q−2μk(K)R_{\Delta_G}^{\mu_k,K}(p) \le C\mu_k(K)^{\frac{1}{Q-1}}+C\overline{\alpha}^{Q-2}\mu_k(K)

which implies (47) since μk(K)≤μk(A)≤σ(S∖Z)<∞\mu_k(K)\le\mu_k(A)\le\sigma(S\setminus\mathcal{Z})<\infty.

Thus, for the measure μ~k\widetilde{\mu}_k that is the measure μk\mu_k rescaled by the right hand side of (47), we have

RΔGμ~k,K(p)≤1for all p∈RNR_{\Delta_G}^{\widetilde{\mu}_k,K}(p)\le1\quad\text{for all }p\in\mathbb{R}^{N}

and

cap⁡ΔG(K)≥μ~k(K)=Cμk(K)Q−2Q−1≥Cσ(P(A))Q−2Q−1.\operatorname{cap}_{\Delta_G}(K)\ge\widetilde{\mu}_k(K)=C\mu_k(K)^{\frac{Q-2}{Q-1}}\ge C\sigma(\mathcal{P}(A))^{\frac{Q-2}{Q-1}}.

in the light of Definition 2.5 and (46). The proof of Lemma 5.4 is complete.

Geometric applications

In this subsection our goal is to derive [14 Theorem 3.4] and then the geometric applications [14 Theorem 4.2] based on Theorem 3.2 and Theorem 5.1 by the same ideas as in [1, 12, 13, 5], which would not work without Theorem 5.1.

Theorem 5.2. ([14 Theorem 3.4]) Let G=(RN,∘,δλ)G = (\mathbb{R}^{N}, \circ, \delta_{\lambda}) be a polarizable Carnot group. And let RΔGμ,DR_{\Delta_G}^{\mu,D} be the potential of a nonnegative Radon measure μ\mu supported in D⊂RND \subset\mathbb{R}^{N} for the sub-Laplacian ΔG\Delta_G in GG. Suppose that SS is a compact subset in DD and that

lim⁡t→0RΔGμ,D(γ(t))=∞and∫0δ(RΔGμ,D(γ(t)))2Q−2 dt=∞,(49)\lim_{t \to0} R_{\Delta_G}^{\mu,D}(\gamma(t)) = \infty\quad\text{and} \quad\int_{0}^{\delta} \left(R_{\Delta_G}^{\mu,D}(\gamma(t))\right)^{\frac{2}{Q-2}} \, dt = \infty, \tag*{(49)}

for any curve γ(t):[0,δ]→RN\gamma(t) : [0,\delta] \to\mathbb{R}^{N} with γ(0)∈S\gamma(0) \in S and γ∈C[0,δ]∩C1(0,δ]\gamma\in C[0,\delta] \cap C^{1}(0,\delta] that is horizontal with respect to the sub-Riemannian structure of GG and some δ>0\delta> 0, where QQ is the homogeneous dimension of GG. Then

dim⁡HΔG(S)≤Q−22.(50)\dim_{\mathcal{H}_{\Delta_G}}(S) \le\frac{Q-2}{2}. \tag*{(50)}

Proof. Our proof here is different from that of [14 Theorem 3.4] and relies on Theorem 3.2 and Theorem 5.1 in this paper. By Theorem 5.1, one is able to pick up a horizontal radial curve that escapes from the thin set EE in Theorem 3.2 and make use of (13) to reach to a contradiction to (49) if assuming otherwise dim⁡HΔN(S)>Q−22\dim_{\mathcal{H}_{\Delta_N}}(S) > \frac{Q-2}{2}. □

Consequently, due to the fact that the nilpotent part of the Iwasawa decomposition of rank one simple groups is H-type (cf. [6]) and therefore polarizable (cf. [2 Proposition 5.6]), we then have

Theorem 5.3. ([14 Theorem 4.2]) Assume NN is the nilpotent part of the Iwasawa decomposition of rank one groups: SU(n+1,1)SU(n+1,1), Sp(n+1,1)Sp(n+1,1) and F4(−20)F_{4(-20)}. Let SS be a compact subset of a bounded and open subset Ω\Omega in NN. And let θ=u4Q−2θ0\theta= u^{\frac{4}{Q-2}}\theta_{0} be a conformal contact form on Ω∖S\Omega\setminus S, where θ0\theta_{0} is the standard contact form on NN and QQ is the homogeneous dimension of NN. Suppose that the sub-Riemannian metric associated with θ\theta is geodesically complete near SS. Then

dim⁡HΔN(S)≤Q−22(51)\dim_{\mathcal{H}_{\Delta_N}}(S) \le\frac{Q-2}{2} \tag*{(51)}

when the scalar curvature RθR_{\theta} of the conformal contact form θ\theta is nonnegative.

Proof. The proof uses the same idea from that in [12, 13, 5]. The argument to make use of potentials and Theorem 5.2 for singular solutions to scalar curvature equations is the same as that presented in [14 Section 4.2]. (49) is implied by the geodesic completeness of the sub-Riemannian structure associated with θ=u4Q−2θ0\theta= u^{\frac{4}{Q-2}}\theta_{0}. □

With some efforts as in [12], Theorem 5.3 holds even when the scalar curvature is not pointwisely nonnegative analogous to that in [12 Theorem 1.3]. A version of Theorem 5.3 for CR geometry has appeared in [5 Theorem 1.1].

References

References

  1. [1]M. Arsove and A. Huber, Local behavior of subharmonic functions, Indiana Univ. Math. J. 22 (1973), 1191 - 1199.
  2. [2]Z. M. Balogh and J. T. Tyson, Polar coordinates in Carnot groups, Math. Z. 241, (2002), 697 - 730.DOI
  3. [3]A. Bonfiglioli, E. Lanconelli, F. Uguzzoni, Stratified Lie Groups and Potential Theory for their Sub-Laplacians, Springer Monographs in Mathematics, Springer, Berlin, 2007.DOI
  4. [4]T. Bromwich, Introduction to the theory of infinite series, The Macmillan, New York, 1908.
  5. [5]S. Chanillo and P. Yang, The singular CR Yamabe problem and Hausdorff dimension, Commun. Contemp. Math. 80 (2026) No 5 Paper No. 2540005, 20 pp. https://arxiv.org/abs/2504.06888.
  6. [6]M. Cowling, A.H. Dooley, A. Korányi, and F. Ricci, H-type groups and Iwasawa decompositions, Adv. in Math. 87 (1991),1 - 41.DOI
  7. [7]P. du Bois-Reymond, Eine neue Theorie der Convergenz und Divergenz von Reihen mit positiven Gliedern, Journal für die reine und angewandte Mathematik 76 (1873), 61 - 91.
  8. [8]H. Federer, Geometric Measure Theory, Classics in Mathematics, Reprint of the 1969 ed - Berlin; Heidelberg; New York; Barcelona; Budapest; Hong Kong; London; Milan; Santa Clara; Singapore; Tokyo; Springer 1996.
  9. [9]G. Folland, A fundamental solution for a subelliptic operator, Bulleton of AMS, 79 (1973) no. 2, 373 - 376.DOI
  10. [10]A. Huber, On subharmonic functions and differential geometry in the large, Comment. Math. Helv. 32 (1957), 13 - 72.
  11. [11]B. Kpata, On a decomposition of non-negative Radon measures. Arch. Math. (Brno) 55 (2019), no. 4, 203 - 210.DOI
  12. [12]S. Ma and J. Qing, On linear potentials and applications in conformal geometry, Anal. PDE 18 (2025), no. 3, 773 - 803.arxiv.org/abs/2209.02823
  13. [13]S. Ma and J. Qing, Potential theory and applications in conformal geometry, Note Mat 45 (2025) suppl. 1, 147 - 167.DOI
  14. [14]S. Ma and J. Qing, On potentials for sub-Laplacians and geometric applications, preprint 2025. https://arxiv.org/pdf/2512.18221.
  15. [15]Y. Mizuata, Potential theory in Euclidean spaces, GAKUTO International Series. Mathematical Sciences and Applications, 6. Gakkōtosho Co., Ltd., Tokyo, 1996. viii+341 pp. ISBN: 4-7625-0415-7
  16. [16]R. Montgomery, A Tour of Subriemannian Geometries, Their Geodesics and Applications, Math. Surveys Monogr., 91 American Mathematical Society, Providence, RI, 2002, xx+259 pp.

Paper details

vlnc:W_bnvzj2mx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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