betaFeedback

WorksMathematics

Gauss–Bonnet theorem for o-minimal surfaces with isolated singularities

Abstract

We study the relationship between topology and curvature of surfaces definable in an o-minimal structure. Our main result is a Gauss–Bonnet-type formula for closed (not necessarily compact) definable surfaces with finitely many singular points, expressing the Euler characteristic in terms of the total Gaussian curvature, the local densities at the singular points, and the density at infinity. As consequences, we obtain an inequality for definable minimal surfaces with isolated singularities and, for complex algebraic curves, a formula and an inequality relating the Euler characteristic, the total curvature, the multiplicities at the singular points and the degree.

Introduction

The Gauss–Bonnet theorem is one of the fundamental results of differential geometry: it connects the topology of a compact Riemannian surface with its total Gaussian curvature. For a compact surface XX without boundary it states that

∫XK dA=2πχ(X),\int_X K\,\mathrm{d}A = 2\pi\chi(X),

where KK denotes the Gaussian curvature and χ(X)\chi(X) is the Euler characteristic of XX.

For compact singular surfaces, Grieser [16] obtained the following result.

Theorem 1.1 (Grieser [16 Theorem 1.4]). Let V⊂RNV \subset\mathbb{R}^{N} be a compact real analytic surface with isolated singular points p1,…,pRp_1,\ldots,p_R, and endow V0=V∖{p1,…,pR}V_0 = V \setminus\{p_1,\ldots,p_R\} with the Riemannian metric induced by a real analytic Riemannian metric on RN\mathbb{R}^{N}. For ε>0\varepsilon> 0 small, each connected component of

2020 Mathematics Subject Classification. 53A07; 03C64; 14P10; 14B05; 53C65.

Key words and phrases. Gauss–Bonnet formula; Definable surfaces; o-minimal structures; Local density; Density at infinity; Isolated singularities.

The second named author was partially supported by CNPq-Brazil grants 303375/2025-6, 404611/2025-7 and 405618/2023-9. The third named author was partially supported by CNPq-Brazil grants 400043/2025-4 and 405618/2023-9. This work was supported by the Serrapilheira Institute (grant number Serra – R-2110-39576). a punctured ε\varepsilon-neighbourhood of a singular point is either cone-like or horn-like [16 Theorem 1.1]; let C1,…,CSC_1,\ldots,C_S be the cone-like components (over all singular points) and let lil_i be the coefficient of rr in the asymptotic expansion [16 Theorem 1.3] of the length of the curve of points of CiC_i at distance rr from the corresponding singular point. Then KK is integrable on V0V_0 and

χ(V)=R+12π∫V0K dA−12π∑i=1Sli.\chi(V)=R+\frac{1}{2\pi}\int_{V_0}K\,\mathrm{d}A-\frac{1}{2\pi}\sum_{i=1}^{S}l_i.

Thus horn-like components give no contribution, while each singular point contributes 1−12π∑li1-\frac{1}{2\pi}\sum l_i, the sum being over the cone-like components at that point. When the metric is the Euclidean one, horn-like components have links of length o(r)o(r), and 12π∑i=1Sli=∑j=1Rθ(V,pj)\frac{1}{2\pi}\sum_{i=1}^{S}l_i=\sum_{j=1}^{R}\theta(V,p_j), where θ(V,pj)\theta(V,p_j) is the local density (see Subsection 2.3 and Proposition 4.6(b)). An essential ingredient in Grieser’s approach is the resolution of singularities, which provides a detailed description of the local geometry and the corresponding asymptotic expansions.

For noncompact surfaces, the relationship between topology and curvature is more subtle, since the geometry at infinity contributes to the global formula. For complete Riemannian surfaces of finite topological type whose total curvature exists, Cohn-Vossen [3] proved the inequality ∫MK dA≤2πχ(M)\int_M K\,\mathrm{d}A\leq2\pi\chi(M), and Shiohama [26] identified the defect 2πχ(M)−∫MK dA2\pi\chi(M)-\int_M K\,\mathrm{d}A with lim⁡t→∞L(t)/t=lim⁡t→∞2A(t)/t2\lim_{t\to\infty}L(t)/t=\lim_{t\to\infty}2A(t)/t^2, where L(t)L(t) and A(t)A(t) are the length and the area of the intrinsic metric circle and disc of radius tt. For complete submanifolds of Euclidean space with cone-like ends see Dillen and Kühnnel [5], and for closed semialgebraic sets Dutertre [11, 12] obtained Gauss–Bonnet formulas involving Lipschitz–Killing curvatures at infinity. In particular, for a smooth closed semialgebraic surface X⊂RnX\subset\mathbb{R}^n, [12 Theorem 4.3] (case d=2d=2) seems to give χ(X)=12π∫XK dA+θ(X,∞)\chi(X)=\frac{1}{2\pi}\int_X K\,\mathrm{d}A+\theta(X,\infty), after identifying the normalizations of the Lipschitz–Killing–Weyl curvatures (K0≡vol⁡(Sn−3)K_0\equiv\operatorname{vol}(S^{n-3}) and 1sn−1K2=12πK\frac{1}{s_{n-1}}K_2=\frac{1}{2\pi}K); see also [12 Theorem 3.9] for closed semialgebraic sets with singularities.

O-minimal geometry provides a natural setting for these questions. Definable sets enjoy strong finiteness properties, which make them particularly suitable for the study of singularities and of the asymptotic geometry at infinity. Throughout the paper, a set is said to be definable if it is definable in a fixed o-minimal structure on R\mathbb{R} (see Definition 2.1); we refer to [4, 8] for background. In this setting, the local density and the density at infinity provide quantitative that describe the geometry of a definable set near a point and at infinity, respectively.

In this paper we develop an elementary geometric approach to a Gauss–Bonnet formula for closed definable surfaces with isolated singularities. Our method relies on the analysis of the total geodesic curvature of the links and on its relation with the local density and with the density at infinity (see Subsection 2.3 for the definitions). In contrast with Grieser’s approach, we use neither resolution of singularities nor the theory of normal cycles and Lipschitz–Killing curvature measures; the o-minimal finiteness properties are used only through uniform bounds on the number of connected components of definable families. Moreover, our framework applies to surfaces definable in arbitrary o-minimal structures and includes the noncompact case, where the contribution of infinity is expressed in terms of the density at infinity.

For a closed set X⊂RnX \subset\mathbb{R}^{n} we denote by Sing⁡(X)=Sing⁡2(X)\operatorname{Sing}(X) = \operatorname{Sing}_{2}(X) its C2C^{2} singular set and by Reg⁡(X)=X∖Sing⁡(X)\operatorname{Reg}(X) = X \setminus\operatorname{Sing}(X) its C2C^{2} regular part (Definition 2.3); XX is pure two-dimensional if it has dimension two at each of its regular points (Definition 2.3). We write ∫XK dA:=∫Reg⁡(X)K dA\int_{X} K\,\mathrm{d}A := \int_{\operatorname{Reg}(X)} K\,\mathrm{d}A, and χ(X)\chi(X) denotes the Euler characteristic of the singular homology of XX. Our main result is the following.

Theorem 1.2. Let X⊂RnX \subset\mathbb{R}^{n} be a closed, pure two-dimensional set definable in an o-minimal structure, and assume that Sing⁡(X)={p1,…,ps}\operatorname{Sing}(X) = \{p_{1}, \ldots, p_{s}\} is finite (s≥0s \ge0). Then the homology of XX is finitely generated, the Gaussian curvature KK of Reg⁡(X)\operatorname{Reg}(X) is integrable, the local densities θ(X,pi)\theta(X,p_{i}) and the density at infinity θ(X,∞)\theta(X,\infty) exist, and

χ(X)=s+12π∫XK dA−∑i=1sθ(X,pi)+θ(X,∞).\chi(X)=s+\frac{1}{2\pi}\int_X K\,\mathrm{d}A-\sum_{i=1}^{s}\theta(X,p_i)+\theta(X,\infty).

Equivalently, χ(X)=12π∫XK dA+∑i=1s(1−θ(X,pi))+θ(X,∞)\chi(X)=\frac{1}{2\pi}\int_X K\,\mathrm{d}A+\sum_{i=1}^{s}(1-\theta(X,p_i))+\theta(X,\infty): each singular point contributes 1−θ(X,pi)1-\theta(X,p_i) and infinity contributes θ(X,∞)\theta(X,\infty). If XX is bounded (hence compact), then θ(X,∞)=0\theta(X,\infty)=0. In the compact real analytic case with the Euclidean metric, Theorem 1.2 recovers Theorem 1.1.

Example 1.3. (1) For the plane X={(x,y,z)∈R3:z=0}≅R2X=\{(x,y,z)\in\mathbb{R}^{3}:z=0\}\cong\mathbb{R}^{2}, we have χ(X)=1\chi(X)=1, K≡0K\equiv0, s=0s=0 and θ(X,∞)=1\theta(X,\infty)=1.

(2) For the circular cylinder X={(x,y,z)∈R3:x2+y2=1}X=\{(x,y,z)\in\mathbb{R}^{3}:x^{2}+y^{2}=1\}, we have χ(X)=0\chi(X)=0, s=0s=0, K≡0K\equiv0 and θ(X,∞)=0\theta(X,\infty)=0.

(3) For the upper cone X={(x,y,z)∈R3:z2=x2+y2, z≥0}X=\{(x,y,z)\in\mathbb{R}^{3}:z^{2}=x^{2}+y^{2},\ z\ge0\}, we have χ(X)=1\chi(X)=1, s=1s=1 and K≡0K\equiv0. Moreover, θ(X,0)=θ(X,∞)=1/2\theta(X,0)=\theta(X,\infty)=1/\sqrt{2}.

(4) For the double cone X={(x,y,z)∈R3:z2=x2+y2}X=\{(x,y,z)\in\mathbb{R}^{3}:z^{2}=x^{2}+y^{2}\}, we have s=1s=1, χ(X)=1\chi(X)=1 and K≡0K\equiv0. Its link consists of two circles of radius r/2r/\sqrt{2}, hence θ(X,0)=θ(X,∞)=2\theta(X,0)=\theta(X,\infty)=\sqrt{2}.

(5) For the paraboloid X={(x,y,z)∈R3:z=x2+y2}X=\{(x,y,z)\in\mathbb{R}^{3}:z=x^{2}+y^{2}\}, we have χ(X)=1\chi(X)=1, s=0s=0 and ∫XK dA=2π\int_{X}K\,\mathrm{d}A=2\pi, while θ(X,∞)=0\theta(X,\infty)=0.

(6) For the catenoid X={(x,y,z)∈R3:x2+y2=cosh⁡2z}X=\{(x,y,z)\in\mathbb{R}^{3}:x^{2}+y^{2}=\cosh^{2}z\}, which is definable in Rexp⁡\mathbb{R}_{\exp}, we have χ(X)=0\chi(X)=0, s=0s=0, ∫XK dA=−4π\int_{X}K\,\mathrm{d}A=-4\pi and θ(X,∞)=2\theta(X,\infty)=2 (each of its two ends is asymptotic to a plane).

(7) For the cuspidal surface X={(x,y,z)∈R3:x2+y2=z6, z≥0}X=\{(x,y,z)\in\mathbb{R}^{3}:x^{2}+y^{2}=z^{6},\ z\geq0\}, we have χ(X)=1\chi(X)=1, s=1s=1, ∫XK dA=−2π\int_{X}K\,\mathrm{d}A=-2\pi, θ(X,0)=0\theta(X,0)=0 and θ(X,∞)=1\theta(X,\infty)=1.

(8) If XX is a compact surface of revolution which is smooth except at a cuspidal point pp where it is locally of the form {x2+y2=z2a, z≥0}\{x^{2}+y^{2}=z^{2a},\ z\geq0\} with a>1a>1, then θ(X,p)=0\theta(X,p)=0 and the singular point contributes 1, as a smooth point would.

Remark 1.4. (1) The hypotheses closed and pure two-dimensional cannot be removed: the open unit disc X⊂R2X\subset\mathbb{R}^{2} satisfies χ(X)=1\chi(X)=1, while the right-hand side of the formula vanishes; and for X=R2×{0}∪{(0,0,1)}⊂R3X=\mathbb{R}^{2}\times\{0\}\cup\{(0,0,1)\}\subset\mathbb{R}^{3} (which is closed, with Sing⁡(X)=∅\operatorname{Sing}(X)=\varnothing but with an isolated point) we have χ(X)=2\chi(X)=2, while the right-hand side equals 1.

(2) No orientability assumption is needed.

(3) A point p∈Sing⁡(X)p\in\operatorname{Sing}(X) at which XX is a C1C^{1} submanifold satisfies θ(X,p)=1\theta(X,p)=1 and gives no contribution, as expected.

(4) The singular Euler characteristic χ(X)\chi(X) coincides with the o-minimal Euler characteristic E(X)E(X) of [8 Ch. 4]. Indeed, for RR large, X∩B‾RX\cap\overline{B}_{R} is a deformation retract of XX (Lemma 4.3), E(X∩B‾R)=χ(X∩B‾R)E(X\cap\overline{B}_{R})=\chi(X\cap\overline{B}_{R}) by definable triangulation [8 Ch. 8], and E(X∖B‾R)=0E(X\setminus\overline{B}_{R})=0 by Hardt’s definable triviality theorem [8 Ch. 9] applied to x↦∣x∣x\mapsto|x|, since the fibres are finite unions of circles.

A surface M⊂RnM\subset\mathbb{R}^{n} of class C2C^{2} is minimal if its mean curvature vector vanishes identically. As a consequence of Theorem 1.2 we obtain the following.

Corollary 1.5. Let XX be as in Theorem 1.2 and assume that Reg⁡(X)\operatorname{Reg}(X) is a minimal surface. Then

χ(X)≤s−∑i=1sθ(X,pi)+θ(X,∞),\chi(X)\leq s-\sum_{i=1}^{s}\theta(X,p_{i})+\theta(X,\infty),

and equality holds if and only if XX is a finite union of affine 2-planes.

An interesting consequence of Theorem 1.2 concerns complex algebraic curves. By a classical result of Thie [29] (see also Kurdyka and Raby [21] for the real subanalytic setting), the local density of a complex analytic curve at a point coincides with its multiplicity. Moreover, the density at infinity of a complex algebraic set coincides with its degree; this goes back to [27]. Consequently, Theorem 1.2 yields the following formula.

Corollary 1.6. Let X⊂CnX \subset\mathbb{C}^{n} be a complex algebraic curve (i.e., a pure one-dimensional complex algebraic set) such that Sing⁡(X)={p1,…,ps}\operatorname{Sing}(X)=\{p_{1},\ldots,p_{s}\}. Then

χ(X)=s+12π∫XK dA−∑i=1sm(X,pi)+deg⁡(X).\chi(X)=s+\frac{1}{2\pi}\int_{X}K\,\mathrm{d}A-\sum_{i=1}^{s}m(X,p_{i})+\deg(X).

Moreover,

χ(X)≤deg⁡(X)−∑i=1s(m(X,pi)−1),\chi(X)\leq\deg(X)-\sum_{i=1}^{s}\left(m(X,p_{i})-1\right),

and equality holds if and only if XX is a union of deg⁡(X)\deg(X) distinct complex affine lines.

For instance, for a smooth affine curve XX the corollary gives ∫XK dA=2π(χ(X)−deg⁡X)\int_{X}K\,\mathrm{d}A=2\pi(\chi(X)-\deg X); for the hyperbola {xy=1}⊂C2\{xy=1\}\subset\mathbb{C}^{2} this gives ∫XK dA=−4π\int_{X}K\,\mathrm{d}A=-4\pi.

Remark 1.7. If Sing⁡(X)=∅\operatorname{Sing}(X)=\varnothing and XX is minimal, Theorem 1.2 reduces to

χ(X)=12π∫XK dA+θ(X,∞).\chi(X)=\frac{1}{2\pi}\int_{X}K\,\mathrm{d}A+\theta(X,\infty).

This formula is closely related to the classical Jorge–Meeks formula [20] for complete minimal surfaces of finite total curvature in R3\mathbb{R}^{3}, which states that

∫MK dA=2π(χ(M)−∑j=1kIj).\int_{M}K\,\mathrm{d}A=2\pi\left(\chi(M)-\sum_{j=1}^{k}I_{j}\right).

where IjI_{j} denotes the multiplicity of the jj-th end. Indeed, the asymptotic description of the ends given by Jorge and Meeks shows that the density at infinity is precisely the sum of these multiplicities:

θ(M,∞)=∑j=1kIj.\theta(M,\infty)=\sum_{j=1}^{k}I_{j}.

In particular, if MM is properly embedded, then each end is embedded and Ij=1I_{j}=1. Consequently, θ(M,∞)=k\theta(M,\infty)=k, and our formula agrees with the Jorge–Meeks formula.

The paper is organized as follows. In § 2 we recall the facts on o-minimal structures, densities and curvature of surfaces used in the paper. In § 3 we prove that the total absolute curvature of a definable surface is finite (Proposition 3.2). In § 4 we study the geometry of a definable surface near an isolated singular point and at infinity; the key results are Propositions 4.6 and 4.8, which compute the limits of the total geodesic curvature of the links in terms of the densities. Theorem 1.2 is proved in § 5 and the corollaries in § 6.

Preliminaries

All subsets of Rn\mathbb{R}^{n} are equipped with the metric induced by the Euclidean metric. We denote by Br(a)B_{r}(a) (resp. B‾r(a)\overline{B}_{r}(a), Sr(a)S_{r}(a)) the open ball (resp. closed ball, sphere) of Rn\mathbb{R}^{n} centred at aa with radius rr, and Br:=Br(0)B_{r}:=B_{r}(0), Sr:=Sr(0)S_{r}:=S_{r}(0). We write sm:=Hm(Sm)s_{m}:=\mathcal{H}^{m}(S^{m}) for the volume of the unit mm-sphere (with s0=2s_{0}=2) and μm\mu_{m} for the volume of the unit mm-ball. Recall that

sm+1=2πmsm−1(m≥1).(1)s_{m+1}=\frac{2\pi}{m}s_{m-1}\qquad(m\geq1). \tag*{(1)}

O-minimal structures

Definition 2.1. A structure on R\mathbb{R} is a collection S={Sn}n∈N\mathcal{S}=\{\mathcal{S}_{n}\}_{n\in\mathbb{N}}, where each Sn\mathcal{S}_{n} is a set of subsets of Rn\mathbb{R}^{n}, satisfying the following axioms:

(1) all algebraic subsets of Rn\mathbb{R}^{n} are in Sn\mathcal{S}_{n};

(2) for every nn, Sn\mathcal{S}_{n} is a Boolean subalgebra of the power set of Rn\mathbb{R}^{n};

(3) if A∈SmA\in\mathcal{S}_{m} and B∈SnB\in\mathcal{S}_{n}, then A×B∈Sm+nA\times B\in\mathcal{S}_{m+n};

(4) if π:Rn+1→Rn\pi:\mathbb{R}^{n+1}\to\mathbb{R}^{n} is the projection on the first nn coordinates and A∈Sn+1A\in\mathcal{S}_{n+1}, then π(A)∈Sn\pi(A)\in\mathcal{S}_{n}.

The structure S\mathcal{S} is said to be oo-minimal if, moreover,

(5) the elements of S1\mathcal{S}_{1} are precisely the finite unions of points and intervals.

An element of Sn\mathcal{S}_{n} is called definable (in S\mathcal{S}). A map f:A⊂Rn→Rmf:A\subset\mathbb{R}^{n}\to\mathbb{R}^{m} is definable if its graph is definable.

Remark 2.2. Classical examples of o-minimal structures are: the semialgebraic sets (by the Tarski–Seidenberg theorem); the globally subanalytic sets (by Gabrielov’s complement theorem; see [7]); and the structure Rexp⁡\mathbb{R}_{\exp} generated by the semialgebraic sets and the graph of the exponential function, whose o-minimality is due to Wilkie [30].

Throughout this paper we fix an o-minimal structure S\mathcal{S} on R\mathbb{R}, and definable means definable in S\mathcal{S}. We shall use the following standard facts.

(O1) (Monotonicity) If f:(a,b)→Rf:(a,b)\to\mathbb{R} is definable, there are a=a0<a1<⋯<ak=ba=a_0<a_1<\cdots<a_k=b such that on each (aj−1,aj)(a_{j-1},a_j) the function ff is C1C^1 and either constant or strictly monotone [8 Ch. 3, §1 and Ch. 7]. In particular, lim⁡t→a+f(t)\lim_{t\to a^+}f(t) and lim⁡t→b−f(t)\lim_{t\to b^-}f(t) exist in [−∞,+∞][-\infty,+\infty].

(O2) (Definable choice and curve selection) If A⊂Rm×RnA\subset\mathbb{R}^m\times\mathbb{R}^n is definable and π(A)\pi(A) denotes its projection on Rm\mathbb{R}^m, there is a definable map g:π(A)→Rng:\pi(A)\to\mathbb{R}^n with (y,g(y))∈A(y,g(y))\in A for all y∈π(A)y\in\pi(A) [8 Ch. 6, §1].

(O3) (Dimension) Definable sets have a well defined dimension, which coincides with the dimension as a manifold for definable C1C^1 submanifolds.

If f:A→Bf:A\to B is a definable map, then for each dd the set Bd={b∈B:dim⁡f−1(b)=d}B_d=\{b\in B:\dim f^{-1}(b)=d\} is definable and dim⁡f−1(Bd)=dim⁡Bd+d\dim f^{-1}(B_d)=\dim B_d+d [8 Ch. 4, §1]. A definable subset of a C1C^1 manifold of dimension mm whose dimension is smaller than mm has zero Hm\mathcal{H}^m-measure, since it is a finite union of C1C^1 submanifolds of dimension smaller than mm [8 Ch. 7].

(O4) (Uniform finiteness) If A⊂Rm×RnA\subset\mathbb{R}^m\times\mathbb{R}^n is definable, there is N∈NN\in\mathbb{N} such that, for every y∈Rmy\in\mathbb{R}^m, the fibre Ay={x∈Rn:(y,x)∈A}A_y=\{x\in\mathbb{R}^n:(y,x)\in A\} has at most NN connected components; in particular, if AyA_y is finite, then #Ay≤N\#A_y\leq N [8 Ch. 3, §§2–3], [4]. Moreover, every definable set has finitely many connected components, and they are definable.

(O5) (Differentiability) The partial derivatives of a definable C1C^1 map are definable; if N⊂RnN\subset\mathbb{R}^n is a definable C1C^1 submanifold, the map x↦TxNx\mapsto T_xN (equivalently, x↦Pxx\mapsto P_x, the orthogonal projection onto TxNT_xN) is definable [8 Ch. 7]. For every k≥1k\geq1 and every definable XX, the set Reg⁡k(X)\operatorname{Reg}_k(X) of Definition 2.3 is definable and dim⁡Sing⁡k(X)<dim⁡X\dim\operatorname{Sing}_k(X)<\dim X [9].

Singular set and surfaces

Definition 2.3. Let X⊂RnX\subset\mathbb{R}^n be a subset and kk a positive integer. The CkC^k singular set of XX, denoted by Sing⁡k(X)\operatorname{Sing}_k(X), is the set of points x∈Xx\in X such that U∩XU\cap X is not a CkC^k submanifold of Rn\mathbb{R}^n for any open neighbourhood UU of xx. We set Reg⁡k(X):=X∖Sing⁡k(X)\operatorname{Reg}_k(X):=X\setminus\operatorname{Sing}_k(X). If p∈Reg⁡k(X)p\in\operatorname{Reg}_k(X), there is an open neighbourhood U⊂RnU\subset\mathbb{R}^n of pp such that X∩UX\cap U is a connected CkC^k submanifold, and we set dim⁡pX:=dim⁡(X∩U)\dim_p X:=\dim(X\cap U). We say that XX is pure dd-dimensional if dim⁡pX=d\dim_p X=d for every p∈Reg⁡1(X)p\in\operatorname{Reg}_1(X).

Throughout the paper, Sing⁡(X):=Sing⁡2(X)\operatorname{Sing}(X):=\operatorname{Sing}_2(X) and Reg⁡(X):=Reg⁡2(X)\operatorname{Reg}(X):=\operatorname{Reg}_2(X).

Note that if XX is pure two-dimensional, then XX has no isolated points (an isolated point would be a point of Reg⁡1(X)\operatorname{Reg}_{1}(X) at which XX has dimension 00), so every p∈Xp \in X belongs to X∖{p}X \setminus\{p\}.

Local density and density at infinity

We introduce the local density and the density at infinity; for further details we refer to [21, 27].

Definition 2.4. Let X⊂RnX \subset\mathbb{R}^{n} be a set, d∈Nd \in\mathbb{N} and p∈Rnp \in\mathbb{R}^{n}. We say that XX has dd-dimensional density at pp if the limit

θd(X,p):=lim⁡r→0+Hd(X∩Br(p))μdrd\theta^{d}(X,p) := \lim_{r \to0^{+}} \frac{\mathcal{H}^{d}(X \cap B_{r}(p))}{\mu_{d}r^{d}}

exists, and that XX has dd-dimensional density at infinity if the limit

θd(X,∞):=lim⁡r→+∞Hd(X∩Br)μdrd\theta^{d}(X,\infty) := \lim_{r \to+\infty} \frac{\mathcal{H}^{d}(X \cap B_{r})}{\mu_{d}r^{d}}

exists. If XX is a surface, we write θ(X,p):=θ2(X,p)\theta(X,p) := \theta^{2}(X,p) and θ(X,∞):=θ2(X,∞)\theta(X,\infty) := \theta^{2}(X,\infty).

Since Br−∣a∣⊂Br(a)⊂Br+∣a∣B_{r-|a|} \subset B_{r}(a) \subset B_{r+|a|}, the density at infinity does not depend on the choice of the centre of the balls. If XX is a C1C^{1} surface near p∈Xp \in X, then θ(X,p)=1\theta(X,p)=1, since near pp it is the graph over p+TpXp+T_{p}X of a C1C^{1} map with vanishing derivative at pp.

Curvature of surfaces

Let M⊂RnM \subset\mathbb{R}^{n} be a C2C^{2} surface (two-dimensional submanifold). We denote by DD the Euclidean connection, by ∇\nabla the Levi-Civita connection of MM, by ∇⊥\nabla^{\perp} the normal connection and by NxMN_{x}M the normal space of MM at xx. For tangent vector fields VV, WW and normal vector fields ν\nu we have the Gauss and Weingarten formulas (see [22 Ch. 8])

DVW=∇VW+II(V,W),DVν=−AνV+∇V⊥ν.D_{V}W=\nabla_{V}W+\mathrm{II}(V,W), \qquad D_{V}\nu=-A_{\nu}V+\nabla^{\perp}_{V}\nu.

where II\mathrm{II} is the second fundamental form (symmetric, bilinear and continuous since MM is C2C^{2}) and the shape operator Aν:TxM→TxMA_{\nu}:T_{x}M \to T_{x}M is the self-adjoint operator given by ⟨Aνv,w⟩=⟨II(v,w),ν⟩\langle A_{\nu}v,w\rangle=\langle\mathrm{II}(v,w),\nu\rangle; it depends linearly on ν∈NxM\nu\in N_{x}M. The Gaussian curvature of MM is defined by the Gauss equation: if e1,e2e_{1},e_{2} is an orthonormal basis of TxMT_{x}M and ν1,…,νn−2\nu_{1},\ldots,\nu_{n-2} is an orthonormal basis of NxMN_{x}M, then

K(x)=⟨II(e1,e1),II(e2,e2)⟩−∣II(e1,e2)∣2=∑α=1n−2det⁡Aνα.(2)K(x)=\langle\mathrm{II}(e_{1},e_{1}),\mathrm{II}(e_{2},e_{2})\rangle-\lvert\mathrm{II}(e_{1},e_{2})\rvert^{2}=\sum_{\alpha=1}^{n-2}\det A_{\nu_{\alpha}}. \tag*{(2)}

If MM is of class C3C^{3}, this is the intrinsic Gaussian curvature of the induced metric, by Gauss’ Theorema Egregium [22 Ch. 8].

Let Γ⊂M\Gamma\subset M be a C2C^2 curve with unit tangent vector TT (defined up to sign) and arc length ss. Its curvature vector κ⃗:=dTds\vec{\kappa} := \frac{dT}{ds} does not depend on the orientation. A unit conormal along Γ\Gamma is a continuous unit vector field η\eta tangent to MM and orthogonal to Γ\Gamma. The geodesic curvature of Γ\Gamma with respect to η\eta is

kgη:=⟨κ⃗,η⟩=⟨∇TT,η⟩,k_{\mathrm{g}}^\eta:= \langle\vec{\kappa},\eta\rangle= \langle\nabla_T T,\eta\rangle,

where the second equality follows from the Gauss formula, since II(T,T)⊥TxM\mathrm{II}(T,T) \perp T_xM. Clearly kg−η=−kgηk_{\mathrm{g}}^{-\eta}=-k_{\mathrm{g}}^\eta and ∣kgη∣≤∣κ⃗∣\lvert k_{\mathrm{g}}^\eta\rvert\le\lvert\vec{\kappa}\rvert.

Theorem 2.5 (Gauss–Bonnet). Let S⊂RnS \subset\mathbb{R}^n be a compact C2C^2 surface with C2C^2 boundary, and let ηin\eta_{\mathrm{in}} be the inward pointing unit conormal along ∂S\partial S. Then

∫SK dA+∫∂Skgηin ds=2πχ(S).\int_S K\,\mathrm{d}A+\int_{\partial S} k_{\mathrm{g}}^{\eta_{\mathrm{in}}}\,\mathrm{d}s=2\pi\chi(S).

For smooth oriented compact surfaces with boundary this is classical, see e.g. [22 Ch. 9]. The general case reduces to this one as follows. Let SS be as in the theorem. By [19 Ch. 2], SS has a C∞C^\infty structure compatible with its C2C^2 structure, and the inclusion ι:S→Rn\iota:S\to\mathbb{R}^n can be approximated in the C2C^2 topology by C∞C^\infty embeddings ιk:S→Rn\iota_k:S\to\mathbb{R}^n. The Gaussian curvature, the curvature vector of the boundary, the inward conormal, and the area and length elements of ιk(S)\iota_k(S) depend continuously on the 2-jet of ιk\iota_k; hence they converge uniformly to those of SS, and it suffices to prove the formula for the smooth surfaces ιk(S)\iota_k(S). If such a surface is not orientable, we apply the oriented case to its orientation double cover, endowed with the pulled back metric: its Euler characteristic and both integrals are twice those of the surface.

Area formula

We shall use the area formula in the following form (see [13 3.2.3 and 3.2.5], applied in charts and summed over a countable Borel partition): if F:N→N′F:N\to N' is a locally Lipschitz map between C1C^1 Riemannian manifolds of the same dimension mm (second countable), with Jacobian JFJF, and A⊂NA\subset N is a Borel set, then

∫AJF dHm=∫N′#(A∩F−1(y)) dHm(y).(3)\int_A JF\,\mathrm{d}\mathcal{H}^m=\int_{N'}\#(A\cap F^{-1}(y))\,\mathrm{d}\mathcal{H}^m(y). \tag*{(3)}

On a C1C^1 submanifold of Rn\mathbb{R}^n, Hm\mathcal{H}^m coincides with the Riemannian volume [13 §3.2].

Finiteness of the total curvature

The finiteness of the total absolute curvature of definable submanifolds is well known; see, e.g., Fu [14] for compact subanalytic sets, Bröcker and Kuppe [1] for compact definable sets, and Dutertre [10] and Grandjean [15, Proposition 3.1] for (possibly noncompact) definable hypersurfaces. Since we need it for noncompact surfaces of arbitrary codimension, we include the short proof, which combines the classical argument of Chern and Lashof [2] with the uniform finiteness property (O4). Dutertre pointed out to us that the proofs in [10, 15] should work for the regular part of any definable set.

We first recall an elementary identity.

Lemma 3.1. For q≥1q \ge1 and α,β∈{1,…,q}\alpha,\beta\in\{1,\ldots,q\},

∫Sq−1tαtβ dσ(t)=sq−1qδαβ.\int_{\mathbb{S}^{q-1}} t_\alpha t_\beta\,\mathrm{d}\sigma(t) = \frac{s_{q-1}}{q}\delta_{\alpha\beta}.

Proof. If α≠β\alpha\ne\beta, the reflection t↦(t1,…,−tα,…,tq)t \mapsto(t_1,\ldots,-t_\alpha,\ldots,t_q) preserves σ\sigma and changes the sign of tαtβt_\alpha t_\beta, so the integral vanishes. The integrals of t12,…,tq2t_1^2,\ldots,t_q^2 coincide by symmetry, and their sum is ∫Sq−11 dσ=sq−1\int_{\mathbb{S}^{q-1}} 1\,\mathrm{d}\sigma=s_{q-1}. □

Proposition 3.2. Let X⊂RnX \subset\mathbb{R}^n be a definable set and let MM be the union of the two-dimensional connected components of Reg⁡2(X)\operatorname{Reg}_2(X). Then there is an integer N0N_0 such that

∫M∣K∣ dA≤2πN0<∞.\int_M |K|\,\mathrm{d}A \le2\pi N_0 < \infty.

In particular, K∈L1(M)K \in L^1(M).

Proof. MM is a definable C2C^2 surface by (O5). If n=2n=2, then MM is open in R2\mathbb{R}^2 and K≡0K \equiv0; so assume n≥3n \ge3 and put q=n−2q=n-2.

A pointwise inequality. Fix x∈Mx \in M and an orthonormal basis ν1,…,νq\nu_1,\ldots,\nu_q of NxMN_xM. Every unit normal vector is ν=∑αtανα\nu=\sum_\alpha t_\alpha\nu_\alpha with t∈Sq−1t \in\mathbb{S}^{q-1}, and Aν=∑αtαAναA_\nu=\sum_\alpha t_\alpha A_{\nu_\alpha}. Writing the matrix of AναA_{\nu_\alpha} in an orthonormal basis of TxMT_xM as

(aαbαbαcα),\begin{pmatrix} a_\alpha& b_\alpha\\ b_\alpha& c_\alpha \end{pmatrix},

we get

det⁡Aν=(∑αtαaα)(∑βtβcβ)−(∑αtαbα)2=∑α,βtαtβ(aαcβ−bαbβ).\det A_\nu = \left(\sum_\alpha t_\alpha a_\alpha\right) \left(\sum_\beta t_\beta c_\beta\right) - \left(\sum_\alpha t_\alpha b_\alpha\right)^2 = \sum_{\alpha,\beta}t_\alpha t_\beta\left(a_\alpha c_\beta-b_\alpha b_\beta\right).

Since t↦∑αtαναt \mapsto\sum_\alpha t_\alpha\nu_\alpha is an isometry from Sq−1\mathbb{S}^{q-1} onto the unit sphere S(NxM)S(N_xM) of NxMN_xM, Lemma 3.1 and (2) give

∫S(NxM)det⁡Aν dσx(ν)=sq−1q∑α(aαcα−bα2)=sq−1qK(x).\int_{S(N_xM)}\det A_\nu\,\mathrm{d}\sigma_x(\nu) = \frac{s_{q-1}}{q}\sum_\alpha\left(a_\alpha c_\alpha-b_\alpha^2\right) = \frac{s_{q-1}}{q}K(x).

Hence

∣K(x)∣≤qsq−1∫S(NxM)∣det⁡Aν∣ dσx(ν).(4)|K(x)| \le\frac{q}{s_{q-1}}\int_{S(N_xM)}|\det A_\nu|\,d\sigma_x(\nu). \tag*{(4)}

The generalized Gauss map. Let W⊂MW \subset M be an open set on which there is a C1C^1 orthonormal frame ν1,…,νq\nu_1,\ldots,\nu_q of the normal bundle (such frames exist locally, since NMNM is a C1C^1 vector bundle). Consider the C1C^1 map

G:W×Sq−1→Sn−1,G(x,t)=∑α=1qtανα(x),G: W \times S^{q-1} \to S^{n-1}, \qquad G(x,t)=\sum_{\alpha=1}^{q}t_\alpha\nu_\alpha(x),

and endow W×Sq−1W \times S^{q-1} with the product metric. Fix (x,t)(x,t) and put ν=G(x,t)\nu=G(x,t). For v∈TxMv\in T_xM and w∈TtSq−1=t⊥w\in T_tS^{q-1}=t^\perp, the Weingarten formula gives

dG(x,t)(v,w)=∑αtαDvνα+∑αwανα(x)=−Aνv+B(v)+ι(w),dG_{(x,t)}(v,w)=\sum_\alpha t_\alpha D_v\nu_\alpha+\sum_\alpha w_\alpha\nu_\alpha(x)=-A_\nu v+B(v)+\iota(w),

where B(v)=∑αtα∇v⊥ναB(v)=\sum_\alpha t_\alpha\nabla^\perp_v\nu_\alpha and ι(w)=∑αwανα(x)\iota(w)=\sum_\alpha w_\alpha\nu_\alpha(x). Differentiating ⟨να,νβ⟩=δαβ\langle\nu_\alpha,\nu_\beta\rangle=\delta_{\alpha\beta} we see that the matrix (⟨∇v⊥να,νβ⟩)α,β\left(\langle\nabla^\perp_v\nu_\alpha,\nu_\beta\rangle\right)_{\alpha,\beta} is skew-symmetric, hence

⟨B(v),ν⟩=∑α,βtαtβ⟨∇v⊥να,νβ⟩=0,\langle B(v),\nu\rangle=\sum_{\alpha,\beta}t_\alpha t_\beta\langle\nabla^\perp_v\nu_\alpha,\nu_\beta\rangle=0,

i.e., B(v)∈NxM∩ν⊥B(v)\in N_xM\cap\nu^\perp. Moreover, ι\iota is a linear isometry from t⊥t^\perp onto NxM∩ν⊥N_xM\cap\nu^\perp. With respect to the orthogonal decompositions

T(x,t)(W×Sq−1)=TxM⊕t⊥andTνSn−1=ν⊥=TxM⊕(NxM∩ν⊥),T_{(x,t)}(W\times S^{q-1})=T_xM\oplus t^\perp \qquad\text{and}\qquad T_\nu S^{n-1}=\nu^\perp=T_xM\oplus(N_xM\cap\nu^\perp),

the matrix of dG(x,t)dG_{(x,t)} is block lower triangular,

(−Aν0Bι),\begin{pmatrix} -A_\nu& 0\\ B & \iota \end{pmatrix},

so that JG(x,t)=∣det⁡Aν∣∣det⁡ι∣=∣det⁡Aν∣JG(x,t)=|\det A_\nu||\det\iota|=|\det A_\nu|. On the other hand, for u∈Sn−1u\in S^{n-1} we have G(x,t)=uG(x,t)=u if and only if u⊥TxMu\perp T_xM and tα=⟨u,να(x)⟩t_\alpha=\langle u,\nu_\alpha(x)\rangle for all α\alpha; hence #((Y×Sq−1)∩G−1(u))=#{x∈Y:u⊥TxM}\#((Y\times S^{q-1})\cap G^{-1}(u))=\#\{x\in Y:u\perp T_xM\} for every Y⊂WY\subset W. Cover MM by countably many such open sets WkW_k and choose a Borel partition M=⨆kYkM=\bigsqcup_kY_k with Yk⊂WkY_k\subset W_k. Applying (3) to each Yk×Sq−1Y_k\times S^{q-1}, using Fubini’s theorem and summing (monotone convergence), we obtain

∫M∫S(NxM)∣det⁡Aν∣ dσx(ν) dA(x)=∫Sn−1N(u) dHn−1(u),(5)\int_M\int_{S(N_xM)}|\det A_\nu|\,d\sigma_x(\nu)\,dA(x)=\int_{S^{n-1}}\mathcal{N}(u)\,d\mathcal{H}^{n-1}(u), \tag*{(5)}

where N(u):=#{x∈M:u⊥TxM}\mathcal{N}(u):=\#\{x\in M:u\perp T_xM\} is the number of critical points of the height function x↦⟨x,u⟩x\mapsto\langle x,u\rangle on MM.

O-minimal bound. The set A={(u,x)∈Sn−1×M:u⊥TxM}\mathcal{A}=\{(u,x)\in S^{n-1}\times M:u\perp T_xM\} is definable by (O5), and it is a C1C^1 manifold of dimension $2+(q-1)=n-1$ (it is diffeomorphic to the unit normal bundle of MM). Let $E={u\in S^{n-1}: dim⁡Au≥1}\dim\mathcal{A}_u \ge1\}. By (O3), for each d≥1d \ge1 the set Ed={u:dim⁡Au=d}E_d = \{u : \dim\mathcal{A}_u = d\} is definable and dim⁡Ed+d=dim⁡(A∩(Ed×M))≤n−1\dim E_d + d = \dim(\mathcal{A} \cap(E_d \times M)) \le n - 1, so dim⁡E≤n−2\dim E \le n - 2 and Hn−1(E)=0\mathcal{H}^{n-1}(E) = 0. For u∉Eu \notin E the fibre Au\mathcal{A}_u is finite, and by (O4) there is N0N_0 with N(u)=#Au≤N0\mathcal{N}(u) = \#\mathcal{A}_u \le N_0 for all u∉Eu \notin E.

Combining (4), (5) and (1) (with m=qm = q) we conclude

∫M∣K∣ dA≤qsq−1∫Sn−1N dHn−1≤q sn−1sq−1N0=2πN0.\int_M |K|\,dA \le\frac{q}{s_{q-1}}\int_{S^{n-1}}\mathcal{N}\,d\mathcal{H}^{n-1} \le\frac{q\,s_{n-1}}{s_{q-1}}N_0 = 2\pi N_0.

Remark 3.3. The proof gives the Chern–Lashof type inequality ∫M∣K∣ dA≤2πsn−1∫Sn−1N(u) dHn−1(u)\int_M |K|\,dA \le\frac{2\pi}{s_{n-1}}\int_{S^{n-1}}\mathcal{N}(u)\,d\mathcal{H}^{n-1}(u), i.e., 12π∫M∣K∣ dA\frac{1}{2\pi}\int_M |K|\,dA is bounded by the average number of critical points of the height functions on MM.

Local geometry near an isolated singular point and at infinity

Radial tangency

For a∈Rna \in\mathbb{R}^n we write ρa(x)=∣x−a∣\rho_a(x) = |x-a| and ρ=ρ0\rho= \rho_0. If N⊂RnN \subset\mathbb{R}^n is a C1C^1 submanifold, PxP_x denotes the orthogonal projection onto TxNT_xN and ∇Nρa(x)=Px(x−a∣x−a∣)\nabla_N\rho_a(x) = P_x\left(\frac{x-a}{|x-a|}\right) is the gradient of ρa∣N\rho_a|_N at x≠ax \ne a. Note that ∣∇Nρa(x)∣=max⁡{⟨x−a∣x−a∣,v⟩:v∈TxN, ∣v∣=1}|\nabla_N\rho_a(x)| = \max\{\langle\frac{x-a}{|x-a|},v\rangle: v \in T_xN,\ |v| = 1\} is the cosine of the angle between x−ax-a and TxNT_xN.

The following lemma is well known; for a∈N‾a \in\overline{N} it expresses the Whitney (b)-regularity of the pair (N,{a})(N,\{a\}), which holds in o-minimal structures (see Loi [23]); see also [6 proof of Proposition 2.1]. We include a short proof.

Lemma 4.1. Let N⊂RnN \subset\mathbb{R}^n be a definable C1C^1 submanifold of positive dimension.

(i) If a∈N‾a \in\overline{N}, then ∣∇Nρa(x)∣→1|\nabla_N\rho_a(x)| \to1 as x→ax \to a, x∈N∖{a}x \in N \setminus\{a\}.

(ii) If NN is unbounded, then ∣∇Nρ(x)∣→1|\nabla_N\rho(x)| \to1 as ∣x∣→∞|x| \to\infty, x∈Nx \in N.

Proof. (i) Suppose the contrary. Then there is δ∈(0,1)\delta\in(0,1) such that a∈Z‾a \in\overline{Z}, where Z={x∈N∖{a}:∣∇Nρa(x)∣≤1−δ}Z = \{x \in N \setminus\{a\} : |\nabla_N\rho_a(x)| \le1-\delta\}. The set ZZ is definable by (O5), so ρa(Z)⊂(0,∞)\rho_a(Z) \subset(0,\infty) is a definable set with 00 in its closure; by o-minimality it contains an interval (0,ϵ1)(0,\epsilon_1). By (O2) there is a definable map γ:(0,ϵ1)→Z\gamma: (0,\epsilon_1) \to Z with ∣γ(t)−a∣=t|\gamma(t)-a| = t, and by (O1), after shrinking ϵ1\epsilon_1, γ\gamma is C1C^1 and each coordinate γj′(t)\gamma_j'(t) has a limit Lj∈[−∞,+∞]L_j \in[-\infty,+\infty] as t→0+t \to0^+. Since γj(t)−aj→0\gamma_j(t)-a_j \to0, l’Hôpital’s rule gives lim⁡t→0+(γj(t)−aj)/t=Lj\lim_{t\to0^+}(\gamma_j(t)-a_j)/t = L_j; as ∣γj(t)−aj∣≤t|\gamma_j(t)-a_j| \le t, LjL_j is finite. Thus γ′(t)→L\gamma'(t) \to L and (γ(t)−a)/t→L(\gamma(t)-a)/t \to L, with ∣L∣=1|L|=1. Since γ′(t)∈Tγ(t)N\gamma'(t) \in T_{\gamma(t)}N and γ′(t)≠0\gamma'(t) \ne0 for tt small,

∣∇Nρa(γ(t))∣≥⟨γ(t)−at,γ′(t)∣γ′(t)∣⟩⟶⟨L,L⟩=1,|\nabla_N\rho_a(\gamma(t))| \ge\left\langle\frac{\gamma(t)-a}{t},\frac{\gamma'(t)}{|\gamma'(t)|}\right\rangle\longrightarrow\langle L,L\rangle= 1,

which contradicts γ(t)∈Z\gamma(t) \in Z.

(ii) The proof is the same: if the claim fails, the corresponding set ZZ is unbounded, ρ(Z)\rho(Z) contains an interval (t1,+∞)(t_1,+\infty), and we get a definable C1C^1 curve γ:(t1,∞)→Z\gamma:(t_1,\infty)\to Z with ∣γ(t)∣=t|\gamma(t)|=t and γ′(t)→L∈[−∞,∞]n\gamma'(t)\to L\in[-\infty,\infty]^n; l’Hôpital’s rule (for the quotient γj(t)/t\gamma_j(t)/t, whose denominator tends to +∞+\infty) gives γ(t)/t→L\gamma(t)/t\to L, so ∣L∣=1|L|=1, and we reach the same contradiction. □

Structure of the links

Setting 4.2. Let X⊂RnX\subset\mathbb{R}^n be a closed definable set. We consider two situations.

(L) (Local case) p∈Xp\in X satisfies p∈X∖{p}‾p\in\overline{X\setminus\{p\}}, and there is ε0>0\varepsilon_0>0 such that X∩Bε0(p)∖{p}X\cap B_{\varepsilon_0}(p)\setminus\{p\} is a two-dimensional C2C^2 submanifold of Rn\mathbb{R}^n. We put ϱ=ρp\varrho=\rho_p and J=(0,ε0)J=(0,\varepsilon_0).

(I) (Case at infinity) XX is unbounded and there is R0>0R_0>0 such that X∖BR0‾X\setminus\overline{B_{R_0}} is a two-dimensional C2C^2 submanifold of Rn\mathbb{R}^n. We put ϱ=ρ\varrho=\rho and J=(R0,+∞)J=(R_0,+\infty).

In both cases, for an interval J′⊂JJ'\subset J we write UJ′=X∩ϱ−1(J′)U_{J'}=X\cap\varrho^{-1}(J'), which is an open subset of a C2C^2 surface; for t∈Jt\in J we write

Lt=X∩ϱ−1(t)andℓ(t)=H1(Lt).L_t=X\cap\varrho^{-1}(t)\qquad\text{and}\qquad\ell(t)=\mathcal{H}^1(L_t).

We write ∇ϱ\nabla\varrho for the gradient of ϱ∣UJ\varrho|_{U_J}, u=∇ϱ/∣∇ϱ∣u=\nabla\varrho/|\nabla\varrho| wherever ∇ϱ≠0\nabla\varrho\ne0, and V=∇ϱ/∣∇ϱ∣2V=\nabla\varrho/|\nabla\varrho|^2. Since UJU_J is C2C^2, x↦Pxx\mapsto P_x is C1C^1, so ∇ϱ\nabla\varrho, uu and VV are C1C^1 on their domains.

The vector field uu is the unit conormal along the curves LtL_t pointing towards increasing values of ϱ\varrho, i.e., away from pp in case (L) and away from the origin in case (I).

Lemma 4.3. In Setting 4.2, there is an interval J1⊂JJ_1\subset J, of the form J1=(0,ε)J_1=(0,\varepsilon) in case (L) and J1=(R1,+∞)J_1=(R_1,+\infty) in case (I), such that:

(a) ∣∇ϱ∣≥12|\nabla\varrho|\ge\frac{1}{2} on UJ1U_{J_1}, and δ(t):=sup⁡x∈Lt∣∣∇ϱ(x)∣−1−1∣→0\delta(t):=\sup_{x\in L_t}\bigl||\nabla\varrho(x)|^{-1}-1\bigr|\to0 as t→0+t\to0^+ in case (L), and as t→+∞t\to+\infty in case (I);

(b) for every t∈J1t\in J_1, LtL_t is a nonempty compact C2C^2 curve, i.e., a finite disjoint union of C2C^2 circles;

(c) for every t0∈J1t_0\in J_1, there is a C1C^1 diffeomorphism h:J1×Lt0→UJ1h:J_1\times L_{t_0}\to U_{J_1} such that ϱ(h(t,q))=t\varrho(h(t,q))=t, h(t0,q)=qh(t_0,q)=q and ∂th=V∘h\partial_t h=V\circ h; in particular, each h(t,⋅):Lt0→Lth(t,\cdot):L_{t_0}\to L_t is a C1C^1 diffeomorphism, all the curves LtL_t, t∈J1t\in J_1, have the same number bb of connected components, and U(a,a′)U_{(a,a')} is diffeomorphic to (a,a′)×Lt0(a,a')\times L_{t_0} for $(a,a')\subset J_1;

(d) in case (L), for every r∈(0,ε]r \in(0,\varepsilon], X∩Br(p)X \cap B_r(p) is homeomorphic to the open cone Cone⁡(Lt0)=([0,r)×Lt0)/({0}×Lt0)\operatorname{Cone}(L_{t_0}) = ([0,r) \times L_{t_0})/(\{0\} \times L_{t_0}); in particular, it is contractible.

Proof. (a) Apply Lemma 4.1 to the definable C1C^1 submanifold N=UJN = U_J: in case (L), p∈N‾p \in\overline{N} because p∈X∖{p}p \in X \setminus\{p\}; in case (I), NN is unbounded because XX is.

(b) By (a), ϱ∣UJ1\varrho|_{U_{J_1}} is a C2C^2 submersion, so each LtL_t, t∈J1t \in J_1, is a C2C^2 curve. It is compact because XX is closed and LtL_t is bounded. The map ϱ∣UJ1:UJ1→J1\varrho|_{U_{J_1}}: U_{J_1} \to J_1 is open (it is a submersion) and closed (if ϱ(xk)→t∈J1\varrho(x_k) \to t \in J_1, then the xkx_k eventually lie in the compact set X∩ϱ−1([t−τ,t+τ])⊂UJ1X \cap\varrho^{-1}([t-\tau,t+\tau]) \subset U_{J_1} for some small τ>0\tau> 0, so a subsequence converges to some x∈Xx \in X with ϱ(x)=t\varrho(x) = t); since UJ1≠∅U_{J_1} \ne\varnothing and J1J_1 is connected, ϱ(UJ1)=J1\varrho(U_{J_1}) = J_1. Finally, a compact one-dimensional manifold is a finite disjoint union of circles (see, e.g., [24 Appendix]).

(c) Let Φ\Phi be the maximal flow of the C1C^1 vector field VV on UJ1U_{J_1}; it is of class C1C^1 [17 Ch. V]. Since dϱ(V)=⟨∇ϱ,V⟩=1d\varrho(V) = \langle\nabla\varrho,V\rangle= 1, we have ϱ(Φs(x))=ϱ(x)+s\varrho(\Phi_s(x)) = \varrho(x) + s while Φs(x)\Phi_s(x) is defined (see the figure 1). We claim that Φs(x)\Phi_s(x) is defined for every ss with ϱ(x)+s∈J1\varrho(x)+s \in J_1. Indeed, if the maximal interval of definition were (α,ω)(\alpha,\omega) with ϱ(x)+ω∈J1\varrho(x)+\omega\in J_1, then Φs(x)\Phi_s(x), s∈[0,ω)s \in[0,\omega), would stay in the compact set X∩ϱ−1([ϱ(x),ϱ(x)+ω])⊂UJ1X \cap\varrho^{-1}([\varrho(x),\varrho(x)+\omega]) \subset U_{J_1}, on which VV is bounded by (a); hence it could be extended beyond ω\omega [17 Ch. II, §3], a contradiction; the same argument applies to α\alpha. Then h(t,q)=Φt−t0(q)h(t,q) = \Phi_{t-t_0}(q) has the required properties, and its inverse x↦(ϱ(x),Φt0−ϱ(x)(x))x \mapsto(\varrho(x),\Phi_{t_0-\varrho(x)}(x)) is C1C^1.

Figure 1. Schematic representation of the flow of VV, which transports points between the level curves of ϱ\varrho.

(d) Define h^:[0,r)×Lt0→X∩Br(p)\hat{h}: [0,r) \times L_{t_0} \to X \cap B_r(p) by h^(0,q)=p\hat{h}(0,q) = p and h^=h\hat{h} = h on (0,r)×Lt0(0,r) \times L_{t_0}. Since ∣h^(t,q)−p∣=t|\hat{h}(t,q)-p| = t, h^\hat{h} is continuous, and it induces a continuous bijection Cone⁡(Lt0)→X∩Br(p)\operatorname{Cone}(L_{t_0}) \to X \cap B_r(p). For every r′<rr' < r, its restriction to the compact image of [0,r′]×Lt0[0,r'] \times L_{t_0} is a homeomorphism onto X∩Br′(p)X \cap B_{r'}(p) (a continuous bijection from a compact space onto a Hausdorff space); hence the bijection is a homeomorphism. □

Lemma 4.4. In Setting 4.2, let J1J_1 be as in Lemma 4.3, and for x∈UJ1x \in U_{J_1} let κ⃗(x)\vec{\kappa}(x) be the curvature vector at xx of the curve Lϱ(x)L_{\varrho(x)}. Then κ⃗:UJ1→Rn\vec{\kappa}: U_{J_1} \to\mathbb{R}^n is continuous and definable, and there is a constant C>0C > 0 such that

∫Lt∣κ⃗∣ ds≤Cfor every t∈J1.\int_{L_t} |\vec{\kappa}| \,\mathrm{d}s \le C \qquad\text{for every } t \in J_1.

In particular, ∫Lt∣κgu∣ ds≤C\int_{L_t} |\kappa_g^u| \,\mathrm{d}s \le C for every t∈J1t \in J_1. Moreover, in case (L), ℓ(t)≤C1t\ell(t) \le C_1t for every t∈J1t \in J_1, where C1=4(b+3C/π)C_1 = 4(b+3C/\pi).

Proof. For x∈UJ1x \in U_{J_1}, the tangent line of Lϱ(x)L_{\varrho(x)} at xx is TxUJ1∩∇ϱ(x)⊥T_xU_{J_1} \cap\nabla\varrho(x)^\perp, and the orthogonal projection onto it is Qx=Px−u(x)⊗u(x)Q_x = P_x-u(x)\otimes u(x), where (a⊗a)(w)=⟨a,w⟩a(a\otimes a)(w)=\langle a,w\rangle a. Thus x↦Qxx\mapsto Q_x is C1C^1 and definable. Near each point there is a C1C^1 unit vector field TT spanning these tangent lines, unique up to sign, and κ⃗=dT(T)\vec{\kappa}=dT(T). Since Q=T⊗TQ=T\otimes T, we have dQ(v)=dT(v)⊗T+T⊗dT(v)dQ(v)=dT(v)\otimes T+T\otimes dT(v), and using ⟨dT(v),T⟩=0\langle dT(v),T\rangle=0 we get

κ⃗(x)=dQx(T(x))T(x).\vec{\kappa}(x)=dQ_x(T(x))T(x).

The right-hand side is an even function of T(x)T(x); hence κ⃗\vec{\kappa} is a well defined continuous map, and it is definable, its graph being {(x,w):∃v∈Im⁡Qx, ∣v∣=1, w=dQx(v)v}\{(x,w):\exists v\in\operatorname{Im}Q_x,\ |v|=1,\ w=dQ_x(v)v\}, where dQdQ is definable by (O5).

Fix j∈{1,…,n}j\in\{1,\ldots,n\} and let Zj={x∈UJ1:κj(x)=0}Z_j=\{x\in U_{J_1}:\kappa_j(x)=0\}, where κj\kappa_j is the jj-th coordinate of κ⃗\vec{\kappa}. The family {Lt∖Zj}t∈J1\{L_t\setminus Z_j\}_{t\in J_1} is definable, so by (O4) there is NjN_j such that Lt∖ZjL_t\setminus Z_j has at most NjN_j connected components for every tt.

Let II be one of them; it is a connected open subset of some component Σ\Sigma of LtL_t. Let σ\sigma be an arc length parametrization of Σ\Sigma, so that σ′′=κ⃗∘σ\sigma''=\vec{\kappa}\circ\sigma. On σ−1(I)\sigma^{-1}(I) the continuous function σj′=κj∘σ\sigma_j'=\kappa_j\circ\sigma does not vanish, so it has constant sign. If I=ΣI=\Sigma, then ∫Σκj ds=∫σj′′=0\int_\Sigma\kappa_j\,\mathrm{d}s=\int\sigma_j''=0 by periodicity, which is impossible. Hence I=σ((c,c′))I=\sigma((c,c')) is an open arc and

∫I∣κj∣ ds=∣∫cc′σj′′∣=∣σj′(c′)−σj′(c)∣≤2.\int_I|\kappa_j|\,\mathrm{d}s=\left|\int_c^{c'}\sigma_j''\right|=|\sigma_j'(c')-\sigma_j'(c)|\le2.

Since κj=0\kappa_j=0 on ZjZ_j, we get ∫Lt∣κj∣ ds≤2Nj\int_{L_t}|\kappa_j|\,\mathrm{d}s\le2N_j, and therefore ∫Lt∣κ⃗∣ ds≤∑j∫Lt∣κj∣ ds≤C:=2∑jNj\int_{L_t}|\vec{\kappa}|\,\mathrm{d}s\le\sum_j\int_{L_t}|\kappa_j|\,\mathrm{d}s\le C:=2\sum_jN_j. The second assertion follows from ∣κgu∣≤∣κ⃗∣|\kappa_g^u|\le|\vec{\kappa}|.

Assume now we are in case (L), and let Σ\Sigma be a component of LtL_t, of length λ\lambda and total curvature cΣ=∫Σ∣κ⃗∣ dsc_\Sigma=\int_\Sigma|\vec{\kappa}|\,\mathrm{d}s, and σ:[0,λ]→Σ\sigma:[0,\lambda]\to\Sigma an arc length parametrization. Choose 0=τ0<τ1<⋯<τm=λ0=\tau_0<\tau_1<\cdots<\tau_m=\lambda with m≤1+3cΣ/πm\le1+3c_\Sigma/\pi and ∫τk−1τk∣σ′′∣≤π/3\int_{\tau_{k-1}}^{\tau_k}|\sigma''|\le\pi/3 for every kk (this is possible since τ↦∫0τ∣σ′′∣\tau\mapsto\int_0^\tau|\sigma''| is continuous and nondecreasing). For τ∈[τk−1,τk]\tau\in[\tau_{k-1},\tau_k], the spherical distance between σ′(τ)\sigma'(\tau) and σ′(τk−1)\sigma'(\tau_{k-1}) is at most the length ∫τk−1τ∣σ′′∣≤π/3\int_{\tau_{k-1}}^\tau\lvert\sigma''\rvert\le\pi/3 of the curve σ′\sigma' on Sn−1S^{n-1}, so ⟨σ′(τ),σ′(τk−1)⟩≥12\langle\sigma'(\tau),\sigma'(\tau_{k-1})\rangle\ge\frac{1}{2}. Since Σ⊂St⁡(p)\Sigma\subset\operatorname{St}(p) has diameter at most 2t2t,

2t≥⟨σ(τk)−σ(τk−1),σ′(τk−1)⟩=∫τk−1τk⟨σ′(τ),σ′(τk−1)⟩ dτ≥τk−τk−12.2t \ge\langle\sigma(\tau_k)-\sigma(\tau_{k-1}),\sigma'(\tau_{k-1})\rangle= \int_{\tau_{k-1}}^{\tau_k}\langle\sigma'(\tau),\sigma'(\tau_{k-1})\rangle\,\mathrm{d}\tau\ge\frac{\tau_k-\tau_{k-1}}{2}.

Thus λ≤4tm≤4t(1+3cΣ/π)\lambda\le4tm \le4t(1+3c_{\Sigma}/\pi). Summing over the bb components of LtL_t we get ℓ(t)≤4t(b+3C/π)\ell(t) \le4t(b+3C/\pi).

Lemma 4.5. In Setting 4.2, let J1J_1, δ\delta and CC be as in Lemmas 4.3 and 4.4, and put

I(t):=∫Ltkgu ds=∫Lt⟨κ⃗,u⟩ ds,t∈J1.I(t):=\int_{L_t} k_g^u\,\mathrm{d}s=\int_{L_t}\langle\vec{\kappa},u\rangle\,\mathrm{d}s,\qquad t\in J_1.

Then:

(i) (First variation) ℓ\ell is of class C1C^1 on J1J_1 and ℓ′(t)=−∫Lt⟨κ⃗,u⟩∣∇ϱ∣ ds\ell'(t)=-\int_{L_t}\frac{\langle\vec{\kappa},u\rangle}{\lvert\nabla\varrho\rvert}\,\mathrm{d}s; in particular, ∣ℓ′(t)+I(t)∣≤Cδ(t)\lvert\ell'(t)+I(t)\rvert\le C\delta(t).

(ii) (Coarea) For a<a′a<a' in J1J_1, H2(U(a,a′))=∫aa′(∫Ltds∣∇ϱ∣)dt\mathcal{H}^2(U_{(a,a')})=\int_a^{a'}\left(\int_{L_t}\frac{\mathrm{d}s}{\lvert\nabla\varrho\rvert}\right)\mathrm{d}t; in particular, ∣H2(U(a,a′))−∫aa′ℓ(t) dt∣≤∫aa′δ(t)ℓ(t) dt\left\lvert\mathcal{H}^2(U_{(a,a')})-\int_a^{a'}\ell(t)\,\mathrm{d}t\right\rvert\le\int_a^{a'}\delta(t)\ell(t)\,\mathrm{d}t.

(iii) (Gauss–Bonnet on annuli) For a<a′a<a' in J1J_1, I(a′)−I(a)=∫U(a,a′)K dAI(a')-I(a)=\int_{U_{(a,a')}}K\,\mathrm{d}A.

Proof. Fix t0∈J1t_0\in J_1 and let hh be as in Lemma 4.3(c). Let Σ1,…,Σb\Sigma_1,\ldots,\Sigma_b be the components of Lt0L_{t_0}, let ci:R/λiZ→Σic_i:\mathbb{R}/\lambda_i\mathbb{Z}\to\Sigma_i be arc length parametrizations, and put Hi(t,ϑ)=h(t,ci(ϑ))H_i(t,\vartheta)=h(t,c_i(\vartheta)). Then ℓ(t)=∑i∫0λi∣∂ϑHi(t,ϑ)∣ dϑ\ell(t)=\sum_i\int_0^{\lambda_i}\lvert\partial_\vartheta H_i(t,\vartheta)\rvert\,\mathrm{d}\vartheta, and ∂ϑHi≠0\partial_\vartheta H_i\ne0.

(i) The map HiH_i is C1C^1 and ∂tHi=V∘Hi\partial_tH_i=V\circ H_i is C1C^1; hence ∂ϑ∂tHi\partial_\vartheta\partial_tH_i exists and is continuous, and by Schwarz’s theorem [25 Theorem 9.41] also ∂t∂ϑHi\partial_t\partial_\vartheta H_i exists and equals ∂ϑ∂tHi\partial_\vartheta\partial_tH_i. Differentiating under the integral sign,

ℓ′(t)=∑i∫0λi⟨∂ϑ(V∘Hi),T∘Hi⟩ dϑ,\ell'(t)=\sum_i\int_0^{\lambda_i}\langle\partial_\vartheta(V\circ H_i),T\circ H_i\rangle\,\mathrm{d}\vartheta,

where T∘Hi=∂ϑHi/∣∂ϑHi∣T\circ H_i=\partial_\vartheta H_i/\lvert\partial_\vartheta H_i\rvert, and ℓ′\ell' is continuous. For fixed tt, ϑ↦T(Hi(t,ϑ))\vartheta\mapsto T(H_i(t,\vartheta)) is C1C^1, because TT is a C1C^1 unit tangent field along the C2C^2 curve LtL_t and Hi(t,⋅)H_i(t,\cdot) is a C1C^1 regular parametrization of a component of LtL_t; moreover ∂ϑ(T∘Hi)=∣∂ϑHi∣ κ⃗∘Hi\partial_\vartheta(T\circ H_i)=\lvert\partial_\vartheta H_i\rvert\,\vec{\kappa}\circ H_i. Integrating by parts on the circle R/λiZ\mathbb{R}/\lambda_i\mathbb{Z},

ℓ′(t)=−∑i∫0λi⟨V∘Hi,κ⃗∘Hi⟩∣∂ϑHi∣ dϑ=−∫Lt⟨V,κ⃗⟩ ds=−∫Lt⟨u,κ⃗⟩∣∇ϱ∣ ds,\ell'(t)=-\sum_i\int_0^{\lambda_i}\langle V\circ H_i,\vec{\kappa}\circ H_i\rangle\lvert\partial_\vartheta H_i\rvert\,\mathrm{d}\vartheta=-\int_{L_t}\langle V,\vec{\kappa}\rangle\,\mathrm{d}s=-\int_{L_t}\frac{\langle u,\vec{\kappa}\rangle}{\lvert\nabla\varrho\rvert}\,\mathrm{d}s,

since V=u/∣∇ϱ∣V=u/\lvert\nabla\varrho\rvert. Consequently

∣ℓ′(t)+I(t)∣=∣∫Lt⟨u,κ⃗⟩(1−∣∇ϱ∣−1) ds∣≤δ(t)∫Lt∣κ⃗∣ ds≤Cδ(t).\lvert\ell'(t)+I(t)\rvert=\left\lvert\int_{L_t}\langle u,\vec{\kappa}\rangle(1-\lvert\nabla\varrho\rvert^{-1})\,\mathrm{d}s\right\rvert\le\delta(t)\int_{L_t}\lvert\vec{\kappa}\rvert\,\mathrm{d}s\le C\delta(t).

(ii) Endow (a,a′)×Lt0(a,a')\times L_{t_0} with the product metric. At (t,q)(t,q) we have dh(∂t)=Vdh(\partial_t)=V, which is orthogonal to dh(TqLt0)=Th(t,q)Ltdh(T_qL_{t_0})=T_{h(t,q)}L_t; hence Jh(t,q)=Jh(t,q)= ∣V(h(t,q))∣ J(h(t,⋅))(q)|V(h(t,q))|\,J(h(t,\cdot))(q). By (3) applied to hh, Fubini’s theorem, and (3) applied to h(t,⋅):Lt0→Lth(t,\cdot):L_{t_0}\to L_t,

H2(U(a,a′))=∫aa′∫Lt0∣V(h(t,q))∣ J(h(t,⋅))(q) dH1(q) dt=∫aa′∫Lt∣V∣ ds dt.\mathcal{H}^{2}(U_{(a,a')})=\int_{a}^{a'}\int_{L_{t_0}}|V(h(t,q))|\,J(h(t,\cdot))(q)\,d\mathcal{H}^{1}(q)\,dt=\int_{a}^{a'}\int_{L_t}|V|\,ds\,dt.

and ∣V∣=∣∇ϱ∣−1|V|=|\nabla\varrho|^{-1}. The last assertion follows from ∣∣∇ϱ∣−1−1∣≤δ(t)\left||\nabla\varrho|^{-1}-1\right|\leq\delta(t) on LtL_t.

(iii) The set S=X∩ϱ−1([a,a′])S=X\cap\varrho^{-1}([a,a']) is a compact C2C^2 surface with boundary La⊔La′L_a\sqcup L_{a'} (since ϱ∣UJ1\varrho|_{U_{J_1}} is a C2C^2 submersion), which is diffeomorphic to [a,a′]×Lt0[a,a']\times L_{t_0} by Lemma 4.3(c), i.e., to a disjoint union of bb annuli. Hence χ(S)=0\chi(S)=0. The inward unit conormal of SS is uu along LaL_a and −u-u along La′L_{a'}. By Theorem 2.5,

0=∫SK dA+∫Lakgu ds+∫La′kg−u ds=∫U(a,a′)K dA+I(a)−I(a′).0=\int_S K\,dA+\int_{L_a}k_g^u\,ds+\int_{L_{a'}}k_g^{-u}\,ds=\int_{U_{(a,a')}}K\,dA+I(a)-I(a').

where we used that La∪La′L_a\cup L_{a'} has zero area. □

Total geodesic curvature of the links and densities

The following proposition is motivated by Grieser’s approach in [16]: we analyse the geodesic curvature of the links, but the asymptotic expansions coming from resolution of singularities are replaced by o-minimal finiteness arguments.

Proposition 4.6. In Setting 4.2, case (L), let Lr=X∩Sr(p)L_r=X\cap S_r(p) and let uu be the unit conormal along LrL_r pointing away from pp. Then:

(a) the limit Θ:=−12πlim⁡r→0+∫Lrkgu ds\Theta:=-\frac{1}{2\pi}\lim_{r\to0+}\int_{L_r}k_g^u\,ds exists;

(b) lim⁡r→0+H1(Lr)2πr=Θ\lim_{r\to0+}\frac{\mathcal{H}^1(L_r)}{2\pi r}=\Theta; in particular, Θ≥0\Theta\geq0;

(c) the density θ(X,p)\theta(X,p) exists and θ(X,p)=Θ\theta(X,p)=\Theta; thus

lim⁡r→0+∫Lrkgu ds=−2πθ(X,p);\lim_{r\to0+}\int_{L_r}k_g^u\,ds=-2\pi\theta(X,p);

(d) for every r∈J1r\in J_1,

∫Lrkgu ds=−2πθ(X,p)+∫X∩Br(p)K dA,\int_{L_r}k_g^u\,ds=-2\pi\theta(X,p)+\int_{X\cap B_r(p)}K\,dA,

and

H2(X∩Br(p))≤C1r2.\mathcal{H}^2(X\cap B_r(p))\leq C_1r^2.

Proof. Let J1=(0,ε)J_1=(0,\varepsilon) be as in Lemma 4.3. By Proposition 3.2 (applied to XX), KK is integrable on UJ1U_{J_1}.

By Lemma 4.5(iii), for 0<a<a′<ε0<a<a'<\varepsilon we have ∣I(a′)−I(a)∣≤∫U(a,a′)∣K∣ dA|I(a')-I(a)|\leq\int_{U_{(a,a')}}|K|\,dA, which tends to 00 as a,a′→0+a,a'\to0^+ by the integrability of KK. By the Cauchy criterion, I0:=lim⁡a→0+I(a)I_0:=\lim_{a\to0+}I(a) exists, which proves (a) with Θ=−I0/(2π)\Theta=-I_0/(2\pi).

Letting a→0+a\to0^+ in Lemma 4.5(iii) we also get

I(r)=I0+∫U(0,r)K dA(r∈J1).(6)I(r)=I_0+\int_{U_{(0,r)}}K\,dA\qquad(r\in J_1). \tag*{(6)}

By Lemma 4.5(i) and Lemma 4.3(a), ∣ℓ′(t)+I(t)∣≤Cδ(t)→0|\ell'(t)+I(t)|\leq C\delta(t)\to0, hence

lim⁡t→0+ℓ′(t)=−I0=2πΘ.\lim_{t\to0+}\ell'(t)=-I_0=2\pi\Theta.

By Lemma 4.4, 0≤ℓ(t)≤C1t0 \leq\ell(t) \leq C_1t, so ℓ(t)→0\ell(t) \to0 as t→0+t \to0^{+}. Since ℓ′\ell' is bounded near 00 (it has a limit), we get ℓ(t)=∫0tℓ′(τ) dτ\ell(t)=\int_0^t\ell'(\tau)\,\mathrm{d}\tau, and therefore ℓ(t)/t→2πΘ\ell(t)/t \to2\pi\Theta. This proves (b).

Since X∩Br(p)={p}∪U(0,r)X\cap B_r(p)=\{p\}\cup U_{(0,r)}, letting a→0+a\to0^{+} in Lemma 4.5(ii) (monotone convergence) gives

∣H2(X∩Br(p))−∫0rℓ(t) dt∣≤∫0rδ(t)ℓ(t) dt≤(sup⁡(0,r)δ)C1r22=o(r2).\left|\mathcal{H}^2(X\cap B_r(p))-\int_0^r\ell(t)\,\mathrm{d}t\right|\leq\int_0^r\delta(t)\ell(t)\,\mathrm{d}t\leq\left(\sup_{(0,r)}\delta\right)\frac{C_1r^2}{2}=o(r^2).

By (b), ∫0rℓ(t) dt=πΘr2+o(r2)\int_0^r\ell(t)\,\mathrm{d}t=\pi\Theta r^2+o(r^2). Hence H2(X∩Br(p))/(πr2)→Θ\mathcal{H}^2(X\cap B_r(p))/(\pi r^2)\to\Theta, which proves (c). Finally, (d) follows from (6) and I0=−2πθ(X,p)I_0=-2\pi\theta(X,p), together with H2(X∩Br(p))≤∫0r2ℓ(t) dt≤C1r2\mathcal{H}^2(X\cap B_r(p))\leq\int_0^r2\ell(t)\,\mathrm{d}t\leq C_1r^2 (note ∣∇ϱ∣−1≤2|\nabla\varrho|^{-1}\leq2). □

Remark 4.7. (1) Applying Proposition 4.6 to the closure of each connected component XjX^j of UJ1U_{J_1}, we see that the density of XX at pp is the sum of the densities of its local “branches” Xj‾\overline{X^j}, and that lim⁡r→0+∫Lr∩Xjkgu ds=−2πθ(Xj‾,p)\lim_{r\to0^+}\int_{L_r\cap X^j}k_g^u\,\mathrm{d}s=-2\pi\theta(\overline{X^j},p).

(2) Part (b) is a special case of [6 Proposition 2.1], where the density of a definable set of any dimension is expressed by the volumes of its links. The existence of the density of subanalytic sets is due to Kurdyka and Raby [21] (for complex analytic sets, to Thie [29]); for sets definable in an arbitrary o-minimal structure, we refer to Sampaio and Silva [27] and Dinh and Nguyen [6].

The behaviour at infinity is analysed in exactly the same way; the only difference is the sign of the conormal that points into the compact part.

Proposition 4.8. In Setting 4.2, case (I), let LR=X∩SRL_R=X\cap S_R and let −u-u be the unit conormal along LRL_R pointing towards the origin. Then:

(a) the limit Θ∞:=12πlim⁡R→+∞∫LRkg−u ds\Theta_\infty:=\frac{1}{2\pi}\lim_{R\to+\infty}\int_{L_R}k_g^{-u}\,\mathrm{d}s exists;

(b) lim⁡R→+∞H1(LR)2πR=Θ∞\lim_{R\to+\infty}\frac{\mathcal{H}^1(L_R)}{2\pi R}=\Theta_\infty; in particular, Θ∞≥0\Theta_\infty\geq0;

(c) if H2(X∩BR)<∞\mathcal{H}^2(X\cap B_R)<\infty for some R∈J1R\in J_1, then the density at infinity θ(X,∞)\theta(X,\infty) exists and equals Θ∞\Theta_\infty; thus

lim⁡R→+∞∫LRkg−u ds=2πθ(X,∞);\lim_{R\to+\infty}\int_{L_R}k_g^{-u}\,\mathrm{d}s=2\pi\theta(X,\infty);

(d) for every R∈J1R\in J_1,

∫LRkg−u ds=2πΘ∞+∫X∖BRK dA.\int_{L_R}k_g^{-u}\,\mathrm{d}s=2\pi\Theta_\infty+\int_{X\setminus B_R}K\,\mathrm{d}A.

Proof. Let J1=(R1,+∞)J_1=(R_1,+\infty) be as in Lemma 4.3; KK is integrable on UJ1U_{J_1} by Proposition 3.2. Note that ∫LRkg−u ds=−I(R)\int_{L_R}k_g^{-u}\,\mathrm{d}s=-I(R).

As in the local case, Lemma 4.5(iii) and the integrability of KK show that I(R)I(R) satisfies the Cauchy criterion as R→+∞R\to+\infty; let I∞I_\infty be its limit and Θ∞=−I∞/(2π)\Theta_\infty=-I_\infty/(2\pi). Letting a′→+∞a' \to+\infty in Lemma 4.5(iii) we get I∞−I(R)=∫U(R,∞)K dAI_\infty-I(R)=\int_{U(R,\infty)}K\,dA, which gives (a) and (d).

By Lemma 4.5(i), ℓ′(R)→−I∞=2πΘ∞\ell'(R)\to-I_\infty=2\pi\Theta_\infty as R→+∞R\to+\infty. Fixing R2∈J1R_2\in J_1, from ℓ(R)=ℓ(R2)+∫R2Rℓ′(τ) dτ\ell(R)=\ell(R_2)+\int_{R_2}^{R}\ell'(\tau)\,d\tau we obtain ℓ(R)/R→2πΘ∞\ell(R)/R\to2\pi\Theta_\infty, which proves (b). In particular ℓ(t)=O(t)\ell(t)=O(t) on [R2,∞)[R_2,\infty).

Assume that H2(X∩BR)<∞\mathcal{H}^2(X\cap B_R)<\infty for some R∈J1R\in J_1; by Lemma 4.5(ii) this then holds for every R∈J1R\in J_1. For R>R2R>R_2, since H2(LR2)=0\mathcal{H}^2(L_{R_2})=0,

H2(X∩BR)=H2(X∩BR2)+H2(U(R2,R)).\mathcal{H}^2(X\cap B_R)=\mathcal{H}^2(X\cap B_{R_2})+\mathcal{H}^2(U_{(R_2,R)}).

and by Lemma 4.5(ii), (b) and δ(t)→0\delta(t)\to0,

H2(U(R2,R))=∫R2Rℓ(t) dt+O(∫R2Rδ(t)t dt)=πΘ∞R2+o(R2).\mathcal{H}^2(U_{(R_2,R)})=\int_{R_2}^{R}\ell(t)\,dt+O\left(\int_{R_2}^{R}\delta(t)t\,dt\right)=\pi\Theta_\infty R^2+o(R^2).

Hence H2(X∩BR)/(πR2)→Θ∞\mathcal{H}^2(X\cap B_R)/(\pi R^2)\to\Theta_\infty, which proves (c). □

Remark 4.9. Proposition 4.8 can also be deduced from Proposition 4.6 via the inversion ι(x)=x/∣x∣2\iota(x)=x/|x|^2: the set X~=ι(X∖{0})\widetilde{X}=\iota(X\setminus\{0\}) satisfies Setting 4.2(L) at 00, ι\iota maps LRL_R onto X~∩S1/R\widetilde{X}\cap S_{1/R} by a homothety of ratio R−2R^{-2}, so that H1(X~∩S1/R)/(2πR−1)=H1(LR)/(2πR)\mathcal{H}^1(\widetilde{X}\cap S_{1/R})/(2\pi R^{-1})=\mathcal{H}^1(L_R)/(2\pi R) and therefore θ(X,0)=θ(X,∞)\theta(X,0)=\theta(X,\infty) by Proposition 4.6(b) and Proposition 4.8(b),(c); and, since ι\iota is conformal with factor ∣x∣−2|x|^{-2}, the transformation rule of the geodesic curvature under conformal changes of metric gives

∫LRkg−u~ ds=∫X~∩S1/Rkg~u~ ds~+2R∫LR∣∇ρ∣ ds,\int_{L_R}k_g^{-\widetilde{u}}\,ds=\int_{\widetilde{X}\cap S_{1/R}}k_{\widetilde{g}}^{\widetilde{u}}\,d\widetilde{s}+\frac{2}{R}\int_{L_R}|\nabla\rho|\,ds,

where u~\widetilde{u} is the conormal pointing away from 00. The direct proof above avoids these computations.

Proof of Theorem 1.2

We divide the proof into four steps. First, we choose suitable radii. Next, we compute the Euler characteristic of Xr,RX_{r,R}. We then apply the Gauss–Bonnet theorem to Xr,RX_{r,R}. Finally, we combine the previous results to obtain the desired formula.

Let XX be as in Theorem 1.2 and M=Reg⁡(X)M=\operatorname{Reg}(X), which is a two-dimensional C2C^2 surface since XX is pure two-dimensional. By Proposition 3.2, K∈L1(M)K\in L^1(M).

Choice of the radii. Let ε0∈(0,1]\varepsilon_0\in(0,1] be smaller than 12∣pi−pj∣\frac{1}{2}|p_i-p_j| for all i≠ji\ne j. Since XX has no isolated points, each pip_i satisfies Setting 4.2(L) with this ε0\varepsilon_0. Let ε>0\varepsilon>0 be such that the conclusions of Lemmas 4.3, 4.4 and 4.5 and of Proposition 4.6 hold on (0,ε)(0,\varepsilon) for every pip_i. If XX is unbounded, XX satisfies Setting 4.2(I) with R0=1+max⁡i∣pi∣R_0=1+\max_i |p_i|, and we let R1R_1 be such that the same results and Proposition 4.8 hold on (R1,+∞)(R_1,+\infty). Fix r∈(0,ε)r\in(0,\varepsilon) and R>R1+εR>R_1+\varepsilon; if XX is bounded, fix instead RR with X⊂BR/2X\subset B_{R/2}. Put

Xr,R={x∈X:∣x−pi∣≥r for all i, ∣x∣≤R}.X_{r,R}=\{x\in X:|x-p_i|\ge r\text{ for all }i,\ |x|\le R\}.

Then Xr,R⊂MX_{r,R}\subset M, and Xr,RX_{r,R} is a compact C2C^2 surface with boundary

∂Xr,R=⨆i=1sLr(pi)⊔LR,Lr(pi)=X∩Sr(pi),LR=X∩SR,\partial X_{r,R}=\bigsqcup_{i=1}^{s}L_r(p_i)\sqcup L_R,\qquad L_r(p_i)=X\cap S_r(p_i),\qquad L_R=X\cap S_R,

because the corresponding distance functions are C2C^2 submersions near these curves (Lemma 4.3(a)); if XX is bounded, LR=∅L_R=\varnothing.

The Euler characteristic. Consider the open subsets of XX

Ω1={x∈X:∣x−pi∣>r/2 for all i, ∣x∣<2R},\Omega_1=\{x\in X:|x-p_i|>r/2\text{ for all }i,\ |x|<2R\},
Ω2=X∩(⋃i=1sBr(pi)∪{∣x∣>R}).\Omega_2=X\cap\left(\bigcup_{i=1}^{s}B_r(p_i)\cup\{|x|>R\}\right).

Then X=Ω1∪Ω2X=\Omega_1\cup\Omega_2 and, by Lemma 4.3(c),

Ω1∩Ω2=(⨆i=1s(X∩{r/2<ρpi<r}))⊔(X∩{R<ρ<2R})\Omega_1\cap\Omega_2=\left(\bigsqcup_{i=1}^{s}(X\cap\{r/2<\rho_{p_i}<r\})\right)\sqcup(X\cap\{R<\rho<2R\})

is diffeomorphic to a disjoint union of products of open intervals with finite unions of circles, so χ(Ω1∩Ω2)=0\chi(\Omega_1\cap\Omega_2)=0. By Lemma 4.3(d), each X∩Br(pi)X\cap B_r(p_i) is contractible, and by Lemma 4.3(c), X∩{∣x∣>R}X\cap\{|x|>R\} is diffeomorphic to (R,∞)×LR(R,\infty)\times L_R; hence χ(Ω2)=s\chi(\Omega_2)=s. Using the product structures given by Lemma 4.3(c) on the collars X∩{r/2<ρpi≤r}X\cap\{r/2<\rho_{p_i}\le r\} and X∩{R≤ρ<2R}X\cap\{R\le\rho<2R\}, we see that Ω1\Omega_1 deformation retracts onto Xr,RX_{r,R}, so χ(Ω1)=χ(Xr,R)\chi(\Omega_1)=\chi(X_{r,R}). All these spaces have finitely generated homology: Xr,RX_{r,R} is a compact definable set, hence definably triangulable [8 Ch. 8], and the others are homotopy equivalent to points or to finite unions of circles. By the Mayer–Vietoris sequence for the open cover {Ω1,Ω2}\{\Omega_1,\Omega_2\} (see [18 §2.2]), the homology of XX is finitely generated and

χ(X)=χ(Ω1)+χ(Ω2)−χ(Ω1∩Ω2)=χ(Xr,R)+s.(7)\chi(X)=\chi(\Omega_1)+\chi(\Omega_2)-\chi(\Omega_1\cap\Omega_2)=\chi(X_{r,R})+s. \tag*{(7)}

Gauss–Bonnet on Xr,RX_{r,R}. By Theorem 2.5,

2πχ(Xr,R)=∫Xr,RK dA+∑i=1s∫Lr(pi)kgη ds+∫LRkgη ds.(8)2\pi\chi(X_{r,R})=\int_{X_{r,R}}K\,\mathrm{d}A+\sum_{i=1}^{s}\int_{L_r(p_i)}k_g^{\eta}\,\mathrm{d}s+\int_{L_R}k_g^{\eta}\,\mathrm{d}s. \tag*{(8)}

Along Lr(pi)L_r(p_i) the inward unit conormal of Xr,RX_{r,R} points away from pip_i, and along LRL_R it points towards the origin. Hence Propositions 4.6(d) and 4.8(d) give

∫Lr(pi)kgin ds=−2πθ(X,pi)+∫X∩Br(pi)K dA(i=1,…,s),\int_{L_{r}(p_i)} k_g^{\mathrm{in}}\,\mathrm{d}s=-2\pi\theta(X,p_i)+\int_{X\cap B_r(p_i)}K\,\mathrm{d}A \qquad(i=1,\ldots,s),
∫LRkgin ds=2πΘ∞+∫X∖BR‾K dA.\int_{L_R} k_g^{\mathrm{in}}\,\mathrm{d}s=2\pi\Theta_\infty+\int_{X\setminus\overline{B_R}}K\,\mathrm{d}A.

Moreover, H2(X∩BR)<∞\mathcal{H}^2(X\cap B_R)<\infty: indeed, X∩BRX\cap B_R is the union of the sets X∩Br(pi)X\cap B_r(p_i), which have finite area by Proposition 4.6(d), and of the compact C2C^2 surface Xr,RX_{r,R}. Hence, by Proposition 4.8(c), θ(X,∞)\theta(X,\infty) exists and equals Θ∞\Theta_\infty. If XX is bounded, θ(X,∞)=0\theta(X,\infty)=0 and the term corresponding to LRL_R is absent.

Conclusion. Since MM is, up to the null set ⋃iLr(pi)∪LR\bigcup_i L_r(p_i)\cup L_R, the disjoint union of Xr,R∩MX_{r,R}\cap M, of the sets X∩Br(pi)∖{pi}X\cap B_r(p_i)\setminus\{p_i\} and of X∖BR‾X\setminus\overline{B_R}, substituting into (8) we get

2πχ(Xr,R)=∫XK dA−2π∑i=1sθ(X,pi)+2πθ(X,∞).2\pi\chi(X_{r,R})=\int_XK\,\mathrm{d}A-2\pi\sum_{i=1}^{s}\theta(X,p_i)+2\pi\theta(X,\infty).

Together with (7) this gives

χ(X)=s+12π∫XK dA−∑i=1sθ(X,pi)+θ(X,∞),\chi(X)=s+\frac{1}{2\pi}\int_XK\,\mathrm{d}A-\sum_{i=1}^{s}\theta(X,p_i)+\theta(X,\infty),

which completes the proof of Theorem 1.2. □

Remark 5.1. The proof shows slightly more: for every rr and RR as in Step 1,

χ(Xr,R)=12π∫XK dA−∑iθ(X,pi)+θ(X,∞);\chi(X_{r,R})=\frac{1}{2\pi}\int_XK\,\mathrm{d}A-\sum_i\theta(X,p_i)+\theta(X,\infty);

no limit r→0+r\to0^+, R→+∞R\to+\infty is needed, since the curvature of the removed pieces is accounted for exactly by Propositions 4.6(d) and 4.8(d).

Proofs of the corollaries

Proof of Corollary 1.5. Let M=Reg⁡(X)M=\operatorname{Reg}(X) and x∈Mx\in M, and let e1,e2e_1,e_2 be an orthonormal basis of TxMT_xM. Minimality means II(e1,e1)+II(e2,e2)=0\mathrm{II}(e_1,e_1)+\mathrm{II}(e_2,e_2)=0, so by (2)

K(x)=−∣II(e1,e1)∣2−∣II(e1,e2)∣2=−12∣IIx∣2≤0.K(x)=-\lvert\mathrm{II}(e_1,e_1)\rvert^2-\lvert\mathrm{II}(e_1,e_2)\rvert^2=-\frac{1}{2}\lvert\mathrm{II}_x\rvert^2\leq0.

Hence ∫XK dA≤0\int_XK\,\mathrm{d}A\leq0, and the inequality follows from Theorem 1.2. If equality holds, then II≡0\mathrm{II}\equiv0 on MM. Thus the Gauss map x↦TxMx\mapsto T_xM is locally constant (its derivative is given by II\mathrm{II}), and each connected component MαM_\alpha of MM (there are finitely many by (O4)) is an open subset of an affine 22-plane PαP_\alpha. Since XX is closed and the components of MM are open and closed in MM, the set Fα=Mα‾∖MαF_\alpha=\overline{M_\alpha}\setminus M_\alpha is contained in the finite set Sing⁡(X)\operatorname{Sing}(X), hence Fα⊂PαF_\alpha\subset P_\alpha is finite. Now MαM_\alpha is open and closed in the connected set Pα∖FαP_\alpha\setminus F_\alpha, so Mα=Pα∖FαM_{\alpha}=P_{\alpha}\setminus F_{\alpha} and Mα‾=Pα\overline{M_{\alpha}}=P_{\alpha}. Since X=M‾X=\overline{M} (no point of XX is isolated), X=⋃αPαX=\bigcup_{\alpha}P_{\alpha}. Conversely, if XX is a finite union of affine planes, then K≡0K\equiv0 and equality holds by Theorem 1.2. □\square

Proof of Corollary 1.6. A complex algebraic curve X⊂Cn=R2nX\subset\mathbb{C}^{n}=\mathbb{R}^{2n} is a closed semialgebraic set. Its complex singular set Sing⁡C(X)\operatorname{Sing}^{\mathbb{C}}(X) is finite, and X∖Sing⁡C(X)X\setminus\operatorname{Sing}^{\mathbb{C}}(X) is a real analytic surface, so XX is pure two-dimensional and Sing⁡2(X)⊂Sing⁡C(X)\operatorname{Sing}_{2}(X)\subset\operatorname{Sing}^{\mathbb{C}}(X). Conversely, if p∈Xp\in X is a point at which XX is a C1C^{1} submanifold, then θ(X,p)=1\theta(X,p)=1 (Subsection 2.3), hence m(X,p)=1m(X,p)=1 by Thie’s theorem [29], so pp is a smooth point of XX. Therefore Sing⁡2(X)=Sing⁡1(X)=Sing⁡C(X)={p1,…,ps}\operatorname{Sing}_{2}(X)=\operatorname{Sing}_{1}(X)=\operatorname{Sing}^{\mathbb{C}}(X)=\{p_{1},\ldots,p_{s}\}. By [29] we have θ(X,pi)=m(X,pi)\theta(X,p_{i})=m(X,p_{i}), and θ(X,∞)=deg⁡(X)\theta(X,\infty)=\deg(X) by [28] (see also [27]). The formula follows from Theorem 1.2.

The regular part of a complex curve is a minimal surface (the second fundamental form satisfies II(Jv,Jw)=−II(v,w)\mathrm{II}(Jv,Jw)=-\mathrm{II}(v,w), where JJ is the complex structure, so that II(e1,e1)+II(Je1,Je1)=0\mathrm{II}(e_{1},e_{1})+\mathrm{II}(Je_{1},Je_{1})=0). Hence Corollary 1.5 gives the inequality, with equality if and only if XX is a finite union of affine real 22-planes. Each of these planes is the closure of a connected component of Reg⁡(X)\operatorname{Reg}(X), hence it is a complex curve, i.e., a complex affine line; and a union of dd distinct complex lines has degree dd. □\square

Acknowledgments: During the preparation of this work, the authors used the AI assistants Claude, Gemini, and ChatGPT to refine arguments, locate references, and draft and revise the manuscript. All AI-generated suggestions were independently verified by the authors, who take full responsibility for the content and accuracy of the final manuscript.

Lucas Holanda Pinheiro: Departamento de Matemática, Universidade Federal do Ceará, Av. Humberto Monte, s/n Campus do Pici - Bloco 914, 60455-760, Fortaleza-CE, Brazil. E-mail: [email protected]

José Edson Sampaio: Departamento de Matemática, Universidade Federal do Ceará, Av. Humberto Monte, s/n Campus do Pici - Bloco 914, 60455-760, Fortaleza-CE, Brazil. E-mail: [email protected]

Euripedes Carvalho da Silva: Departamento de Matemática, Instituto Federal de Educação, Ciência e Tecnologia do Ceará, Av. Parque Central, 1315, Distrito Industrial I, 61939-140, Maracanaú-CE, Brazil. And Departamento de Matemática, Universidade Federal do Ceará, Rua Campus do Pici, s/n, Bloco 914, Pici, 60440-900, Fortaleza-CE, Brazil. E-mail: [email protected]

References

  1. [1]L. Bröcker and M. Kuppe, Integral geometry of tame sets. Geom. Dedicata 82 (2000), 285–323.DOI
  2. [2]S.-S. Chern and R. K. Lashof, On the total curvature of immersed manifolds. Amer. J. Math. 79 (1957), 306–318.
  3. [3]S. Cohn-Vossen, Kürzeste Wege und Totalkrümmung auf Flächen. Compositio Math. 2 (1935), 69–133.
  4. [4]M. Coste, An introduction to o-minimal geometry. Dip. Mat. Univ. Pisa, Dottorato di Ricerca in Matematica, Istituti Editoriali e Poligrafici Internazionali, Pisa, 2000.
  5. [5]F. Dillen and W. Kühnеl, Total curvature of complete submanifolds of Euclidean space. Tohoku Math. J. (2) 57 (2005), no. 2, 171–200.DOI
  6. [6]S. T. Dinh and N. Nguyen, On the density at infinity of definable functions. Preprint, arXiv:2310.06324 (2023).arxiv.org/abs/2310.06324
  7. [7]L. van den Dries, A generalization of the Tarski–Seidenberg theorem, and some non-definability results. Bull. Amer. Math. Soc. (N.S.) 15 (1986), 189–193.DOI
  8. [8]L. van den Dries, Tame topology and o-minimal structures. London Mathematical Society Lecture Note Series, vol. 248, Cambridge University Press, Cambridge, 1998.
  9. [9]L. van den Dries and C. Miller, Geometric categories and o-minimal structures. Duke Math. J. 84 (1996), no. 2, 497–540.DOI
  10. [10]N. Dutertre, Geometrical and topological properties of real polynomial fibres. Geom. Dedicata 105 (2004), 43–59.DOI
  11. [11]N. Dutertre, A Gauss–Bonnet formula for closed semi-algebraic sets. Adv. Geom. 8 (2008), no. 1, 33–51.DOI
  12. [12]N. Dutertre, Euler characteristic and Lipschitz–Killing curvatures of closed semi-algebraic sets. Geom. Dedicata 158 (2012), 167–189.arxiv.org/abs/1002.0767
  13. [13]H. Federer, Geometric Measure Theory. Die Grundlehren der mathematischen Wissenschaften, vol. 153, Springer-Verlag, New York, 1969.
  14. [14]J. H. G. Fu, Curvature measures of subanalytic sets. Amer. J. Math. 116 (1994), no. 4, 819–880.DOI
  15. [15]V. Grandjean, On the total curvatures of a tame function. Bull. Braz. Math. Soc. (N.S.) 39 (2008), no. 4, 515–535.
  16. [16]D. Grieser, Local geometry of singular real analytic surfaces. Trans. Amer. Math. Soc. 355 (2003), no. 4, 1559–1577.arxiv.org/abs/math/9901062
  17. [17]P. Hartman, Ordinary Differential Equations. 2nd ed., Birkhäuser, Boston, 1982 (reprinted: Classics in Applied Mathematics 38, SIAM, 2002).
  18. [18]A. Hatcher, Algebraic Topology. Cambridge University Press, Cambridge, 2002.
  19. [19]M. W. Hirsch, Differential Topology. Graduate Texts in Mathematics, vol. 33, Springer-Verlag, New York, 1976.DOI
  20. [20]L. P. Jorge and W. H. Meeks III, The topology of complete minimal surfaces of finite total Gaussian curvature. Topology 22 (1983), no. 2, 203–221.DOI
  21. [21]K. Kurdyka and G. Raby, Densité des ensembles sous-analytiques. Ann. Inst. Fourier (Grenoble) 39 (1989), no. 3, 753–771.DOI
  22. [22]J. M. Lee, Introduction to Riemannian Manifolds. 2nd ed., Graduate Texts in Mathematics, vol. 176, Springer, Cham, 2018.DOI
  23. [23]T. L. Loi, Verdier and strict Thom stratifications in o-minimal structures. Illinois J. Math. 42 (1998), no. 2, 347–356.
  24. [24]J. W. Milnor, Topology from the Differentiable Viewpoint. The University Press of Virginia, Charlottesville, 1965.
  25. [25]W. Rudin, Principles of Mathematical Analysis. 3rd ed., McGraw-Hill, New York, 1976.
  26. [26]K. Shiohama, Total curvatures and minimal areas of complete surfaces. Proc. Amer. Math. Soc. 94 (1985), no. 2, 310–316.
  27. [27]J. E. Sampaio and E. C. da Silva, On the Kurdyka–Raby formula at infinity and the Moser’s Bernstein theorem. Ann. Sc. Norm. Super. Pisa Cl. Sci. (2026). Preprint version: arXiv:2312.01141.
  28. [28]W. Stoll, The growth of the area of a transcendental analytic set. I, II. Math. Ann. 156 (1964), 47–78 and 144–170.DOI
  29. [29]P. R. Thie, The Lelong number of a point of a complex analytic set. Math. Ann. 172 (1967), 269–312.DOI
  30. [30]A. J. Wilkie, Model completeness results for expansions of the ordered field of real numbers by restricted Pffafian functions and the exponential function. J. Amer. Math. Soc. 9 (1996), no. 4, 1051–1094.

Paper details

vlnc:W_3avnwxsv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

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

  • Mathematics

Claude, Gemini (multiple versions), ChatGPT

View statement
Looking for related papers…
Contents