betaFeedback

WorksMathematics

Nonexistence of cusp singularities for gravity waves of extreme form

Abstract

In this paper we revisit the geometric approach developed by Vărvărucă and Weiss for the Stokes conjecture for gravity water waves near stagnation points, in both the irrotational and rotational settings. They introduced monotonicity formulas and frequency formulas to classify all possible singularities of solutions at stagnation points. Two questions were left open: 1. whether cusp singularities can be excluded in higher dimensions n≥3n\geq 3 [Acta Math. 206 (2011)]; 2. whether cusp singularities for rotational waves can be excluded under the Rayleigh-Taylor condition [Ann. Inst. H. Poincaré C Anal. Non Linéaire 29 (2012)]. We answer both questions in the affirmative. Our main results show that cusp singularities do not exist in higher dimensions and cannot occur for rotational waves even without assuming the Rayleigh-Taylor condition. Our argument combines a delicate analysis of the weighted volume arising from the corresponding free boundary problem with a weighted isoperimetric inequality.

Introduction

The investigation of singularities in gravity water waves lies at the intersection of partial differential equations and fluid mechanics. Mathematically, the physical problem can be translated into the study of the one-phase semilinear Bernoulli-type problem

{Δu=−f(u)in Ω∩{u>0},∣∇u∣2=xnon Ω∩∂{u>0}(1)\begin{cases} \Delta u=-f(u) & \text{in } \Omega\cap\{u>0\},\\ |\nabla u|^2=x_n & \text{on } \Omega\cap\partial\{u>0\} \tag*{(1)} \end{cases}

where Ω⊂Rn\Omega\subset\mathbb{R}^{n} (n≥2n \ge2) is a bounded open set meeting the hyperplane {xn=0}\{x_n=0\} and ff is a real-valued continuous function. In two dimensions, the unknown uu in (1) is the stream function and ff the vorticity-strength function. When f=0f=0, the flow is irrotational; we refer the reader to the introductions of [11, 12] for a precise derivation of (1) from the Euler equations in two dimensions.

From the perspective of water waves, we are restricted to n=2n=2 and are particularly interested in solutions of (1) with singularities at the stagnation points, at which the relative fluid velocity (uy,−ux)(u_y,-u_x) vanishes. Here subscripts denote partial derivatives. It follows from the second equation of (1) that such points lie on the intersection ∂{u>0}∩{x2=0}\partial\{u>0\} \cap\{x_2=0\}. On the other hand, from the

Mathematics Subject Classification 2020: 35R35, 35B65, 76B15.

Keywords: Free boundary problems, Stokes conjecture, vorticity, cusp singularities, Weiss weighted density.

Corresponding author: Yang Pu, [email protected]. point of view of free boundary problems, the Bernoulli free boundary condition ∣∇u∣2=xn|\nabla u|^2=x_n leads us to consider the natural rescaling

ur(x):=u(x0+rx)r3/2for every x0∈∂{u>0}∩{xn=0}.(2)u_r(x):=\frac{u(x^0+rx)}{r^{3/2}} \qquad\text{for every } x^0\in\partial\{u>0\}\cap\{x_n=0\}. \tag*{(2)}

This scaling preserves the Bernoulli free boundary condition.

The singular behavior of the free surface near stagnation points is closely related to the conjecture proposed by G. Stokes [9], which predicts a symmetric corner of angle 2π/32\pi/3 at the crest of the wave. When f=0f=0, this conjecture was first proved independently in [2, 8], under structural assumptions including the isolation of stagnation points and the symmetry and monotonicity of the free surface. The rotational case was proved in [10] under the same structural assumptions, together with sign conditions on the vorticity function ff.

In 2011–2012, Vărvărucă and Weiss [11, 12] studied the problem from a geometric point of view, without relying on the structural assumptions that were required in previous works. They established monotonicity formulas and analyzed the rescaling (2). Moreover, under the local growth assumption

∣∇u∣2≤C0xn+locally in Ω∩{u>0},C0>0,(3)|\nabla u|^2\le C_0x_n^+ \quad\text{locally in }\Omega\cap\{u>0\},\qquad C_0>0, \tag*{(3)}

they proved that ur→u0u_r\to u_0 strongly in Wloc1,2(Rn)W_{\mathrm{loc}}^{1,2}(\mathbb{R}^n) and locally uniformly in Rn\mathbb{R}^n along a subsequence as r→0+r\to0^+. If u0≢0u_0\not\equiv0, then u0u_0 is a homogeneous function of degree 3/23/2 and {u0>0}\{u_0>0\} is an open sector of angle 120∘120^\circ in two dimensions. On the other hand, if u0≡0u_0\equiv0, they introduced the Weiss weighted density, defined as the limit

Mx0(0+)=lim⁡r→0+r−n−1∫Br(x0)xn+χ{u>0} dx.(4)M_{x^0}(0^+)=\lim_{r\to0^+}r^{-n-1}\int_{B_r(x^0)}x_n^+\chi_{\{u>0\}}\,dx. \tag*{(4)}

Here χ{u>0}\chi_{\{u>0\}} denotes the characteristic function of the set {u>0}\{u>0\}. A stagnation point x0∈Ω∩∂{u>0}∩{xn=0}x^0\in\Omega\cap\partial\{u>0\}\cap\{x_n=0\} is a cusp point if Mx0(0+)=0M_{x^0}(0^+)=0, meaning that the Weiss weighted density at x0x^0 vanishes, see Figure 1.

Figure 1. The cusp figure.

On the other hand, it is called a horizontally flat singularity if

Mx0(0+)=∫B1xn+ dx.M_{x^0}(0^+)=\int_{B_1}x_n^+\,dx.

To give a rigorous mathematical proof of the Stokes conjecture, one needs to exclude these two singularities. The nonexistence of horizontally flat singularities has been established for the irrotational case when the zero set {u=0}\{u=0\} has only finitely many components, and has been proved for the rotational case when the vorticity satisfies ∣f(z)∣≤Cz|f(z)| \le Cz [12]; it has also been proved in [5] under some weaker structural assumptions on ff. This topic is beyond the scope of the present paper, and we focus only on cusp singularities.

The exclusion of cusp singularities for the irrotational case (f=0f=0 in (1)) relies on two key ingredients [11 Lemma 4.4]: the strong Bernstein estimate ∣∇u∣2≤x2+|\nabla u|^2 \le x_2^+, that is, C0=1C_0=1 in (3), and the smoothness of the topological free boundary ∂{u>0}∩Ω∩{x2>τ}\partial\{u>0\}\cap\Omega\cap\{x_2>\tau\} for every τ>0\tau>0.1 The key inequality in this method is Δu≥x2+H1⌊∂red{u>0}\Delta u \ge\sqrt{x_2^+}\mathcal{H}^1\lfloor\partial_{\mathrm{red}}\{u>0\},2 and the geometric property is that, for cusp points, a non-empty portion of ∂red{u>0}\partial_{\mathrm{red}}\{u>0\} touches the stagnation point x0x^0. Then the authors asked the following question [11 Page 380]:

In higher dimensions it is not so clear whether cusps can be excluded. Of course that does not happen for Lipschitz free boundaries.

The nonexistence of cusp singularities for Lipschitz free boundaries was proved in [11 Lemma 4.6] under the growth condition (3). Recently, McCurdy [7 Theorem 1.2] excluded cusp singularities in two dimensions under the weaker growth condition (3) by introducing a height function for the free boundary near the cusp singularity. However, the exclusion of cusp singularities in higher dimensions remained open prior to the present work. Let us stress that the primary aim of this paper is to rule out cusp singularities in higher dimensions, thereby answering the open question raised in [11 Page 380].

The Stokes conjecture for waves with non-zero vorticity was considered by Vărvărucă and Weiss in [12]. There they extended their geometric approach from [11] to the semilinear Bernoulli-type problem (1) (f≢0f\not\equiv0 in (1)). Their analysis showed that the limiting problem reduces to the irrotational case. However, the analysis of the cusp singularities introduces new difficulties due to the non-zero right-hand side, which causes additional challenges compared to the irrotational case. Even under the strong Bernstein estimate ∣∇u∣2≤x2+|\nabla u|^2\le x_2^+, the method of [11 Lemma 4.4] breaks down here. Consequently, they conjectured that the cusp alternative is also not possible under the strong Rayleigh-Taylor condition [12 Page 872].

Unfortunately, we do not have the Rayleigh-Taylor condition

∣∇u∣2+2F(u)−xn+≤0,whereF(s)=∫0sf(t) dt.(5)|\nabla u|^2+2F(u)-x_n^+\le0,\qquad\text{where}\quad F(s)=\int_0^s f(t)\,\mathrm{d}t. \tag*{(5)}

in the case with non-zero vorticity, and the method of [11] breaks down here. Still we conjecture that the cusps in case (iii) are not possible when assuming the Rayleigh-Taylor condition.

Let us remark that there are two sources of difficulty for this rotational problem. One difficulty is that the combination of vorticity and gravity makes it hard to obtain the estimate (5). On top of that, we cannot follow the argument of [11 Lemma 4.4] to exclude the cusp singularity even if we assume (5). Consequently, the second goal of this paper is to exclude cusp singularities in the two-dimensional case, settling this open question.

In this paper, we prove that cusp singularities cannot occur by showing that the Weiss weighted density (4) has a positive lower bound:

Mx0(0+)≥c(n,C0)>0M_{x^{0}}(0^{+}) \ge c(n,C_{0}) > 0

for every stagnation point x0∈∂{u>0}∩Ω∩{xn=0}x^{0} \in\partial\{u>0\} \cap\Omega\cap\{x_{n}=0\}, where c(n,C0)c(n,C_{0}) is a constant depending only on nn and C0C_{0}. The lower bound is established for variational solutions, which cover the solution class of [11, 12, 7], and holds in dimensions n≥2n \ge2. When the flow is rotational, we assume only that the vorticity function ff is continuous. Moreover, our results do not rely on any structural assumptions including the isolation of stagnation points and the monotonicity and symmetry of the free boundary. Furthermore, our results show that the Rayleigh-Taylor condition is in fact not necessary for the nonexistence of cusp singularities.

The paper is organized as follows. In § 2, we state the main results. In § 3, we fix the notation and collect the Weiss-type monotonicity formula and the weighted isoperimetric inequality. In § 4, we give the proof of Theorem I.

Main results

Before we state the main results, we recall the notion of variational solutions of (1). Assume that the nonlinearity ff in (1) is a continuous function with primitive F(s)=∫0sf(t) dtF(s) = \int_{0}^{s} f(t)\,\mathrm{d}t.

Definition 2.1 (Variational solutions, see Definition 3.1 in [12]). We define u∈Wloc1,2(Ω)u \in W_{\mathrm{loc}}^{1,2}(\Omega) to be a variational solution of (1) if u∈C0(Ω)∩C0(Ω∩{u>0})u \in C^{0}(\Omega) \cap C^{0}(\Omega\cap\{u>0\}), u≥0u \ge0 in Ω\Omega and u≡0u \equiv0 in Ω∩{xn≤0}\Omega\cap\{x_{n} \le0\}, and the first variation with respect to domain variations of the functional

J(v):=∫Ω(∣∇v∣2−2F(v)+xnχ{v>0}) dxJ(v) := \int_{\Omega} \left(|\nabla v|^{2} - 2F(v) + x_{n}\chi_{\{v>0\}}\right)\,\mathrm{d}x

vanishes at v=uv=u, i.e.,

0=∫Ω((∣∇u∣2−2F(u))div⁡ϕ−2∇uDϕ∇u) dx+∫Ω(xndiv⁡ϕ+ϕn)χ{u>0} dx(6)0 = \int_{\Omega} \left((|\nabla u|^{2}-2F(u))\operatorname{div}\phi- 2\nabla u D\phi\nabla u\right)\,\mathrm{d}x + \int_{\Omega}(x_{n}\operatorname{div}\phi+\phi_{n})\chi_{\{u>0\}}\,\mathrm{d}x \tag*{(6)}

for every ϕ∈Cc1(Ω;Rn)\phi\in C_{c}^{1}(\Omega;\mathbb{R}^{n}).

Remark 2.2. Since our results are completely local, we do not specify any boundary conditions on ∂Ω\partial\Omega.

Remark 2.3. If f≡0f \equiv0, then the notion of variational solutions coincides with the one in [11 Definition 3.1] for the irrotational case.

We write x=(x′,xn)x=(x',x_{n}), xn+=max⁡{xn,0}x_{n}^{+}=\max\{x_{n},0\}, and Br(x0)={x:∣x−x0∣<r}B_{r}(x^{0})=\{x:|x-x^{0}|<r\} for the open ball of radius rr centered at x0x^{0}. Throughout the paper, we assume that the growth condition (3) holds. Let BR(x0)B_{R}(x^{0}) be a ball with xn0=0x_{n}^{0}=0 on which (3) holds, and define

M0=max⁡{f+(s):0≤s≤23C0R3/2}(7)M_{0}=\max\left\{f^{+}(s):0\le s\le\frac{2}{3}\sqrt{C_{0}}R^{3/2}\right\} \tag*{(7)}

where f+(s)=max⁡{f(s),0}f^{+}(s)=\max\{f(s),0\} denotes the positive part of $f.

Remark 2.4. The growth condition (3) implies that uu is locally Lipschitz in BR(x0)B_R(x^0). At almost every point of its zero set, uu is differentiable and attains its minimum, so

∇u=0almost everywhere on {u=0}∩BR(x0).(8)\nabla u = 0 \quad\text{almost everywhere on } \{u = 0\} \cap B_R(x^0). \tag*{(8)}

Remark 2.5. Let uu be a variational solution of (1). Since u∈C2(BR(x0)∩{u>0})u \in C^2(B_R(x^0)\cap\{u>0\}), the domain-stationarity identity (6) holds for test functions supported in the positive phase, so we can deduce that (∣∇u∣2−2F(u))I−2∇u⊗∇u(|\nabla u|^2 - 2F(u))I - 2\nabla u \otimes\nabla u is divergence free in the positive phase. Here ∇u⊗∇u\nabla u \otimes\nabla u is the symmetric matrix with entries (∇u⊗∇u)ij=(∂iu)(∂ju)(\nabla u \otimes\nabla u)_{ij} = (\partial_i u)(\partial_j u). Its divergence equals −2(Δu+f(u))∇u-2(\Delta u + f(u))\nabla u, hence

(Δu+f(u))∇u=0in BR(x0)∩{u>0}.(9)(\Delta u + f(u))\nabla u = 0 \quad\text{in } B_R(x^0) \cap\{u > 0\}. \tag*{(9)}

If Δu+f(u)\Delta u + f(u) were nonzero at some positive point, this identity would make uu constant on the connected component of BR(x0)∩{u>0}∩{Δu+f(u)≠0}B_R(x^0) \cap\{u > 0\} \cap\{\Delta u + f(u) \ne0\} containing it. Write u=a>0u = a > 0 on this component. Then Δu=0\Delta u = 0 and Δu+f(u)=f(a)≠0\Delta u+f(u) = f(a) \ne0 there. At every limit point of the component inside BR(x0)B_R(x^0), continuity of uu gives u=a>0u = a > 0, and continuity of its second derivatives in the positive phase gives Δu=0\Delta u = 0. Such a point therefore still belongs to {u>0}∩{Δu+f(u)≠0}\{u > 0\} \cap\{\Delta u + f(u) \ne0\} and to the same connected component. That component would then be both open and closed in BR(x0)B_R(x^0), and since BR(x0)B_R(x^0) is connected it would equal BR(x0)B_R(x^0), contradicting u=0u = 0 in the lower half-ball. Therefore

Δu=−f(u)in BR(x0)∩{u>0}.(10)\Delta u = -f(u) \quad\text{in } B_R(x^0) \cap\{u > 0\}. \tag*{(10)}

We now define the set of stagnation points and the set of cusp singularities.

Definition 2.6 (Stagnation points). Let uu be a variational solution of (1). We call Su:={x∈Ω∩∂{u>0}:xn=0}S^u := \{x \in\Omega\cap\partial\{u > 0\} : x_n = 0\} the set of stagnation points.

Remark 2.7. It should be noted that Definition 2.6 introduces a generalized notion of stagnation points, which corresponds to the physical concept of stagnation points only when n=2n = 2.

The cusp singularities form the subset of the stagnation points at which the Weiss weighted density vanishes.

Definition 2.8 (Cusp singularities). Let uu be a variational solution of (1), and let x0∈Sux^0 \in S^u be a stagnation point. We call Cu={x0∈Su:Mx0(0+)=0}C^u = \{x^0 \in S^u : M_{x^0}(0+) = 0\} the set of cusp singularities. Here Mx0(0+)M_{x^0}(0+) denotes the Weiss weighted density at x0x^0 and is defined as the limit (4).

Remark 2.9. The existence and finiteness of the Weiss weighted density are proved in [12 Lemma 4.4] for the rotational case and in [11 Lemma 4.2] for the irrotational case, both under the growth condition (3). Moreover, if x0∈Cux^0 \in C^u is a cusp singularity, then

ur→u0≡0as r→0+,u_r \to u_0 \equiv0 \quad\text{as } r \to0^+,

strongly in Wloc1,2(Rn)W_{\mathrm{loc}}^{1,2}(\mathbb{R}^n) and locally uniformly in Rn\mathbb{R}^n, where uru_r is the natural scaling defined in (2).

We now state our main results.

Theorem I. Let n≥2n \ge2, let uu be a variational solution of (1) and BR(x0)⋐ΩB_R(x^0) \Subset\Omega be a ball with x0∈Sux^0 \in S^u. Assume that the growth condition (3) holds a.e. in BR(x0)B_R(x^0). Then there exists a constant c(n,C0)>0c(n,C_0)>0 depending only on nn and C0C_0 such that

r−n−1∫Br(x0)xn+χ{u>0} dx≥c(n,C0)(1−23C0M0r)2n+1(11)r^{-n-1}\int_{B_r(x^0)}x_n^+\chi_{\{u>0\}}\,dx \ge c(n,C_0)\left(1-\frac{2}{3}\sqrt{C_0}M_0\sqrt{r}\right)^{2n+1} \tag*{(11)}

for every 0<r<R0<r<R such that C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1, with M0M_0 defined in (7). Consequently,

Cu=∅.C^u=\varnothing.

Remark 2.10. It follows from the definition of the Weiss weighted density in (4) and the estimate (11) that

Mx0(0+)≥c(n,C0)>0.M_{x^0}(0+) \ge c(n,C_0)>0.

In addition, the asymptotic lower bound is independent of the vorticity. When f≤0f\le0 on the relevant interval, M0=0M_0=0 and (11) holds for every 0<r<R0<r<R.

As a direct corollary of Theorem I, we obtain the following vanishing result for cusp singularities.

Corollary 2.11. Let n≥2n\ge2, let uu be a variational solution of (1), and let f∈C([0,∞))f\in C([0,\infty)). Assume that the growth condition (3) holds near x0∈Ω∩{xn=0}x^0\in\Omega\cap\{x_n=0\}. Then Mx0(0+)=0M_{x^0}(0+)=0 implies that u≡0u\equiv0 in some open nn-dimensional ball containing x0x^0.

Proof. Since xn0=0x_n^0=0, we have u(x0)=0u(x^0)=0. If uu did not vanish in any neighborhood of x0x^0, every ball centered at x0x^0 would contain a point of {u>0}\{u>0\}. Continuity would then give x0∈∂{u>0}x^0\in\partial\{u>0\}. Theorem I would imply Mx0(0+)≥c(n,C0)>0M_{x^0}(0+)\ge c(n,C_0)>0, contradicting the hypothesis. Hence uu vanishes in some ball centered at x0x^0. □

Remark 2.12. The main novelty of our approach is a new isoperimetric argument that excludes cusp singularities in arbitrary dimensions and for general continuous vorticity, without the Rayleigh–Taylor condition. Indeed, through Theorem I and Corollary 2.11, we answer two open questions raised in [11 Page 380] and [12 Remark 4.7], respectively:

  • The nonexistence of cusp singularities in higher dimensions when f≡0f\equiv0;

  • The nonexistence of cusp singularities for general continuous ff without assuming the Rayleigh–Taylor condition.

Remark 2.13. The continuity of ff is not necessary; it can be weakened to f∈C((0,∞))∩L1((0,∞))f\in C((0,\infty))\cap L^1((0,\infty)) with f+f^+ bounded near 00.

Remark 2.14. In the irrotational case the pressure inside the fluid is superharmonic, which yields the Rayleigh–Taylor condition ∣∇u∣2≤xn+|\nabla u|^2\le x_n^+ on smooth parts of the free boundary. For general vorticity, however, the pressure need not be superharmonic, and our argument uses neither this property nor the Rayleigh–Taylor condition. Our approach combines the domain-variation identity (6) with a weighted isoperimetric inequality to derive directly a positive lower bound for the Weiss weighted density, a quantity that provides an important geometric characterization of cusp singularities.

Preliminaries

In the rest of the paper, we translate a fixed center x0x^0 to zero and write BrB_r for Br(0)B_r(0). The outward unit normal to ∂Br\partial B_r is ν=x/r\nu=x/r, and ωn=∣B1∣\omega_n=|B_1|.

For a measurable set AA, χA\chi_A is its characteristic function, ∇χA\nabla\chi_A its distributional derivative, and ∣∇χA∣|\nabla\chi_A| the total variation of that derivative. Here Hn−1\mathcal{H}^{n-1} is the (n−1)(n-1)-dimensional Hausdorff measure. For a smooth set with outward normal νA\nu_A, the convention is

\nabla\chi_A=-\nu_A\mathcal{H}^{n-1}\mathbin{\vrule height 1.4ex depth -0.2ex width 0.07ex\vrule height 0.07ex depth -0.02ex width 0.7ex}\partial A.

We use the standard local definition of BVBV and the conventions of [6, 1]. Every perimeter statement below is made in {xn>0}\{x_n>0\}. After the weighted perimeter has been shown to be finite, we use

V(r)=∫Brxn+χ{u>0} dx,P(r)=∫Br∩{xn>0}xn d∣∇χ{u>0}∣.(12)\begin{aligned} V(r)&=\int_{B_r}\sqrt{x_n^+}\chi_{\{u>0\}}\,dx,\\ P(r)&=\int_{B_r\cap\{x_n>0\}}\sqrt{x_n}\,d|\nabla\chi_{\{u>0\}}|. \tag*{(12)} \end{aligned}

The growth condition (3) makes uu locally Lipschitz in BR(x0)B_R(x^0). Since uu is nonnegative, ∇u=0\nabla u=0 almost everywhere on {u=0}\{u=0\}. After translating x0x^0 to zero, the vertical segment joining (x′,0)(x',0) to (x′,xn)∈BR(x',x_n)\in B_R stays in the ball. Thus, we have

u(x′,xn)=∫0xn∂nu(x′,s) ds≤C0∫0xns ds.u(x',x_n)=\int_0^{x_n}\partial_nu(x',s)\,ds\le\sqrt{C_0}\int_0^{x_n}\sqrt{s}\,ds.

Continuity extends the result to every point, so

0≤u(x)≤23C0(xn+)3/2x∈BR.(13)0\le u(x)\le\frac{2}{3}\sqrt{C_0}(x_n^+)^{3/2}\qquad x\in B_R. \tag*{(13)}

We now record the monotonicity formula for variational solutions uu of (1).

Proposition 3.1 (Weiss formula and weighted density). Let n≥2n\ge2, let uu be a variational solution of (1), and let f∈C([0,∞))f\in C([0,\infty)). Assume that the growth condition (3) holds in BR(x0)⋐ΩB_R(x^0)\Subset\Omega with xn0=0x_n^0=0. Define

Mx0(r)=r−n−1∫Br(x0)(∣∇u∣2−uf(u)+xn+χ{u>0}) dx−32r−n−2∫∂Br(x0)u2 dHn−1,0<r<R.(14)M_{x^0}(r)=r^{-n-1}\int_{B_r(x^0)}\left(|\nabla u|^2-uf(u)+x_n^+\chi_{\{u>0\}}\right)\,dx-\frac{3}{2}r^{-n-2}\int_{\partial B_r(x^0)}u^2\,d\mathcal{H}^{n-1},\qquad0<r<R. \tag*{(14)}

Then Mx0M_{x^0} is locally absolutely continuous on (0,R)(0,R) and, after translating the center to zero,

M′(r)=2r−n−1∫∂Br(∇u⋅ν−3u2r)2 dHn−1+r−n−2K(r)(15)M'(r)=2r^{-n-1}\int_{\partial B_r}\left(\nabla u\cdot\nu-\frac{3u}{2r}\right)^2\,d\mathcal{H}^{n-1}+r^{-n-2}K(r) \tag*{(15)}

for a.e. r∈(0,R)r\in(0,R), where

K(r)=r∫∂Br(2F(u)−uf(u)) dHn−1+∫Br((n−2)uf(u)−2nF(u)) dx.(16)K(r)=r\int_{\partial B_r}\left(2F(u)-uf(u)\right)\,d\mathcal{H}^{n-1}+\int_{B_r}\left((n-2)uf(u)-2nF(u)\right)\,dx. \tag*{(16)}

Moreover, we have

(i) K(r)K(r) satisfies ∣r−n−2K(r)∣≤Cr−1/2|r^{-n-2}K(r)|\le Cr^{-1/2} for small r>0r>0.

(ii) The limit Mx0(0+)M_{x^0}(0+) exists, is finite, and satisfies (4).

(iii) Every sequence rj→0+r_j \to0^+ has a subsequence along which urju_{r_j} converges locally uniformly and strongly in Wloc1,2(Rn)W_{\mathrm{loc}}^{1,2}(\mathbb{R}^n) to a 3/23/2-homogeneous function, harmonic in its positive phase and zero on {xn≤0}\{x_n \le0\}.

Proof. The formulas (15) and (16) were proved in [12 Theorem 3.4], the error estimate for K(r)K(r) is [12 Remark 4.2], and the assertions about the limit and blow-ups are [12 Lemma 4.4(i)–(iv)]. □

Remark 3.2. The results in Proposition 3.1 hold for nn-dimensional variational solutions under (3).

For variational solutions, the convergence ur→u0≡0u_r \to u_0 \equiv0 does not determine the limiting behavior of χ{ur>0}\chi_{\{u_r>0\}}. The Weiss weighted density (4) retains information about the rescaled positive phases. We then introduce the sharp isoperimetric inequality for sets in the upper half-space with monomial weight xn1/2x_n^{1/2}. The following lemma is proved in [3 Theorem 1.4], and the version for measurable sets is contained in [4 Theorem 1.3].

Lemma 3.3. Let A⊂{xn>0}A \subset\{x_n > 0\} be bounded and measurable with finite weighted perimeter. Then

∫{xn>0}xn d∣∇χA∣≥(n+12)(∫B1∩{xn>0}xn dx)22n+1(∫Axn dx)2n−12n+1.(17)\int_{\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}|\nabla\chi_A| \geq\left(n+\frac{1}{2}\right) \left(\int_{B_1\cap\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}} \left(\int_A \sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2n-1}{2n+1}}. \tag*{(17)}

In addition to the weighted isoperimetric inequality, we will use the following comparison between the ordinary volume and the weighted perimeter. It will allow us to control the volume term arising from the vorticity and absorb it at sufficiently small scales.

If A⊂Br∩{xn>0}A \subset B_r \cap\{x_n > 0\} has finite perimeter and A‾⊂{xn>0}\overline{A} \subset\{x_n > 0\}, then

∣A∣≤r∫{xn>0}xn d∣∇χA∣.(18)|A| \leq\sqrt{r}\int_{\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}|\nabla\chi_A|. \tag*{(18)}

To prove this, choose

η∈Cc∞({xn>0}),η≡1in a neighborhood of A‾,\eta\in C_c^\infty(\{x_n>0\}), \qquad\eta\equiv1 \quad\text{in a neighborhood of } \overline{A},

and set

ϕ(x)=η(x)xnen,\phi(x)=\eta(x)x_n e_n,

where ene_n denotes the nn-th coordinate unit vector. Then

div⁡ϕ=1in a neighborhood of A‾.\operatorname{div}\phi=1 \quad\text{in a neighborhood of } \overline{A}.

The Gauss–Green formula therefore gives

∣A∣=−∫{xn>0}ϕ⋅d(∇χA)=−∫{xn>0}ηxn d(∂nχA)=−∫{xn>0}xn d(∂nχA).\begin{aligned} |A|&=-\int_{\{x_n>0\}}\phi\cdot\mathrm{d}(\nabla\chi_A)\\ &=-\int_{\{x_n>0\}}\eta x_n\,\mathrm{d}(\partial_n\chi_A)\\ &=-\int_{\{x_n>0\}}x_n\,\mathrm{d}(\partial_n\chi_A). \end{aligned}

Here the last equality follows from η=1\eta=1 on supp⁡∣∇χA∣⊂A‾\operatorname{supp}|\nabla\chi_A|\subset\overline{A}. Since A⊂Br∩{xn>0}A\subset B_r\cap\{x_n>0\} has finite perimeter and 0<xn≤r0<x_n\le r on this support, we obtain

∣A∣≤∫{xn>0}xn d∣∇χA∣≤r∫{xn>0}xn d∣∇χA∣.|A|\le\int_{\{x_n>0\}}x_n\,\mathrm{d}|\nabla\chi_A|\le\sqrt{r}\int_{\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}|\nabla\chi_A|.

We shall also need some preliminary estimates that recover the boundary measure and the weighted perimeter from the domain-variation identity (6) satisfied by the variational solution uu.

Lemma 3.4. Let n≥2n\ge2, let f∈C([0,∞))f\in C([0,\infty)), and let uu be a variational solution of (1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R=B_R(0)\Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR|\nabla u|^2\le C_0x_n^+ \quad\text{a.e. in }B_R

for some C0>0C_0>0. Set

M0=max⁡{f+(s):0≤s≤23C0R3/2}.M_0=\max\left\{f^+(s):0\le s\le\frac{2}{3}\sqrt{C_0}R^{3/2}\right\}.

Then

μ=Δu+f(u)χ{u>0} dx\mu=\Delta u+f(u)\chi_{\{u>0\}}\,\mathrm{d}x

is a nonnegative Radon measure on BRB_R, supported in BR∩∂{u>0}B_R\cap\partial\{u>0\}, and

μ(Br)≤C(n)C0 rn−1/2+C(n)M0rn\mu(B_r)\le C(n)\sqrt{C_0}\,r^{n-1/2}+C(n)M_0r^n

for 0<2r<R0<2r<R.

Proof. Define the distribution

μ=Δu+f(u)χ{u>0} dxin BR.(19)\mu=\Delta u+f(u)\chi_{\{u>0\}}\,\mathrm{d}x\quad\text{in }B_R. \tag*{(19)}

The vorticity term f(u)χ{u>0}f(u)\chi_{\{u>0\}} is locally integrable because it is bounded. We show that μ\mu is nonnegative by a cutoff at positive values of uu.

Choose a smooth nondecreasing function η:R→[0,1]\eta:\mathbb{R}\to[0,1] such that

η(s)=0(s≤1),η(s)=1(s≥2).\eta(s)=0\quad(s\le1),\qquad\eta(s)=1\quad(s\ge2).

If ζ∈Cc1(BR)\zeta\in C^1_c(B_R) and ε>0\varepsilon>0, the function η(u/ε)ζ\eta(u/\varepsilon)\zeta is supported in a compact subset of the positive phase. It is therefore an admissible test in (10). Multiplication by this test and integration by parts yield

1ε∫BRζη′(u/ε)∣∇u∣2 dx=−∫BRη(u/ε)∇u⋅∇ζ dx+∫BRζη(u/ε)f(u) dx.(20)\frac{1}{\varepsilon}\int_{B_R}\zeta\eta'(u/\varepsilon)|\nabla u|^2\,\mathrm{d}x =-\int_{B_R}\eta(u/\varepsilon)\nabla u\cdot\nabla\zeta\,\mathrm{d}x +\int_{B_R}\zeta\eta(u/\varepsilon)f(u)\,\mathrm{d}x. \tag*{(20)}

As ε→0+\varepsilon\to0^+, the cutoff converges pointwise to χ{u>0}\chi_{\{u>0\}}. Boundedness of f(u)f(u) permits dominated convergence in the vorticity term. The gradient term converges as well, and (8) removes the characteristic function from its limit. Thus the right-hand side tends to

−∫BR∇u⋅∇ζ dx+∫BRζf(u)χ{u>0} dx=⟨μ,ζ⟩.-\int_{B_R}\nabla u\cdot\nabla\zeta\,\mathrm{d}x+\int_{B_R}\zeta f(u)\chi_{\{u>0\}}\,\mathrm{d}x=\langle\mu,\zeta\rangle.

For ζ≥0\zeta\ge0, the left-hand side of (20) is nonnegative. Hence μ\mu is a positive distribution, and so it is a nonnegative Radon measure. It vanishes in the positive phase by (10). It also vanishes in the interior of the zero set. Therefore

supp⁡μ⊂BR∩∂{u>0},(21)\operatorname{supp}\mu\subset B_R\cap\partial\{u>0\}, \tag*{(21)}

where the support is understood relative to BRB_R.

The same calculation provides approximating measures

με=1εη′(u/ε)∣∇u∣2 dx.(22)\mu_\varepsilon= \frac{1}{\varepsilon}\eta'(u/\varepsilon)|\nabla u|^2\,dx. \tag*{(22)}

Equation (20) proves convergence against smooth compactly supported tests. To use continuous tests below, fix a compact set K⋐BRK \Subset B_R and a nonnegative smooth function equal to one near KK. Applying (20) to this function bounds με(K)\mu_\varepsilon(K) uniformly in ε\varepsilon. Approximation of a continuous compactly supported function by smooth functions then gives

με⇀μlocally as Radon measures in BR.(23)\mu_\varepsilon\rightharpoonup\mu\quad\text{locally as Radon measures in } B_R. \tag*{(23)}

More explicitly, choose a larger compact set inside BRB_R and smooth approximations to a given continuous test, all supported in that larger set. The error in integration against με\mu_\varepsilon is at most the uniform approximation error times the uniformly bounded mass of the larger set. The same estimate holds for μ\mu. First let ε→0+\varepsilon\to0^+ for each smooth approximation, using (20), and then let the approximation error tend to zero. This proves (23) with the full class of continuous tests that will occur in the perimeter estimate.

The measure also satisfies the following scale bound. For 2r<R2r<R, choose ζ=1\zeta=1 on BrB_r, supported in B2rB_{2r}, with 0≤ζ≤10\leq\zeta\leq1 and ∣∇ζ∣≤C/r|\nabla\zeta|\leq C/r. Positivity, the distributional definition, and the upper bound for ff give

μ(Br)≤∫ζ dμ=−∫∇u⋅∇ζ dx+∫ζf(u)χ{u>0} dx≤Cr∫B2rC0xn+ dx+M0∣B2r∣≤C(n)C0 rn−1/2+C(n)M0rn.\begin{aligned} \mu(B_r) &\leq\int\zeta\,d\mu\\ &= -\int\nabla u\cdot\nabla\zeta\,dx+\int\zeta f(u)\chi_{\{u>0\}}\,dx \\ &\leq\frac{C}{r}\int_{B_{2r}}\sqrt{C_0x_n^+}\,dx+M_0|B_{2r}| \\ &\leq C(n)\sqrt{C_0}\,r^{n-1/2}+C(n)M_0r^n. \end{aligned}

Lemma 3.5. Let n≥2n\geq2, let f∈C([0,∞))f\in C([0,\infty)), and let uu be a variational solution of (1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R=B_R(0)\Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR|\nabla u|^2\leq C_0x_n^+ \quad\text{a.e. in } B_R

for some C0>0C_0>0. Let

μ=Δu+f(u)χ{u>0} dx\mu=\Delta u+f(u)\chi_{\{u>0\}}\,dx

be the nonnegative Radon measure furnished by Lemma 3.4. Then

χ{u>0}∈BVloc(BR∩{xn>0}),\chi_{\{u>0\}}\in BV_{\mathrm{loc}}(B_R\cap\{x_n>0\}),

and

xn ∣∇χ{u>0}∣≤C0 μin BR∩{xn>0}.(24)\sqrt{x_n}\,|\nabla\chi_{\{u>0\}}|\leq\sqrt{C_0}\,\mu\quad\text{in } B_R\cap\{x_n>0\}. \tag*{(24)}

Proof. The argument proves local finiteness of the distributional derivative and (24) in one step.

Take ϕ∈Cc1(BR∩{xn>0};Rn)\phi\in C_c^1(B_R\cap\{x_n>0\};\mathbb{R}^n). By the interior equation, the divergence of the tensor appearing in the calculation leading to (9) is zero in the positive phase. We can test this identity against η(u/ε)ϕ\eta(u/\varepsilon)\phi. The support of this test is separated from the zero set by the positive level ε\varepsilon, so the computation takes place entirely in the classical region of the solution. The product rule gives

D(η(u/ε)ϕ)=η(u/ε)Dϕ+1εη′(u/ε)ϕ⊗∇uD\left(\eta(u/\varepsilon)\phi\right)=\eta(u/\varepsilon)D\phi+\frac{1}{\varepsilon}\eta'(u/\varepsilon)\phi\otimes\nabla u

and, in particular,

div⁡(η(u/ε)ϕ)=η(u/ε)div⁡ϕ+1εη′(u/ε)∇u⋅ϕ.\operatorname{div}\left(\eta(u/\varepsilon)\phi\right)=\eta(u/\varepsilon)\operatorname{div}\phi+\frac{1}{\varepsilon}\eta'(u/\varepsilon)\nabla u\cdot\phi.

The contraction of the second term with ∇u⊗∇u\nabla u\otimes\nabla u is

1εη′(u/ε)∣∇u∣2∇u⋅ϕ.\frac{1}{\varepsilon}\eta'(u/\varepsilon)|\nabla u|^2\nabla u\cdot\phi.

Thus the terms containing η′\eta' in the divergence-free identity have coefficient (∣∇u∣2−2F(u))−2∣∇u∣2=−∣∇u∣2−2F(u)(|\nabla u|^2-2F(u))-2|\nabla u|^2=-|\nabla u|^2-2F(u). Moving them to the other side yields

∫BRη(u/ε)((∣∇u∣2−2F(u))div⁡ϕ−2∇uDϕ∇u) dx=1ε∫BRη′(u/ε)(∣∇u∣2+2F(u))∇u⋅ϕ dx=∫BR∇u⋅ϕ dμε+2ε∫BRη′(u/ε)F(u)∇u⋅ϕ dx.(25)\begin{aligned} \int_{B_R}\eta(u/\varepsilon)\left((|\nabla u|^2-2F(u))\operatorname{div}\phi-2\nabla uD\phi\nabla u\right)\,dx &=\frac{1}{\varepsilon}\int_{B_R}\eta'(u/\varepsilon)(|\nabla u|^2+2F(u))\nabla u\cdot\phi\,dx \\ &=\int_{B_R}\nabla u\cdot\phi\,d\mu_\varepsilon+\frac{2}{\varepsilon}\int_{B_R}\eta'(u/\varepsilon)F(u)\nabla u\cdot\phi\,dx. \tag*{(25)} \end{aligned}

The sign and coefficient of the potential term follow from the identity

((∣∇u∣2−2F(u))I−2∇u⊗∇u)∇u=−(∣∇u∣2+2F(u))∇u.\left((|\nabla u|^2-2F(u))I-2\nabla u\otimes\nabla u\right)\nabla u=-(|\nabla u|^2+2F(u))\nabla u.

Continuity of ff bounds the potential term. On the range of uu, we have ∣F(s)∣≤Cs|F(s)|\le C_s. Since η′\eta' is supported in [1,2][1,2], the last term of (25) satisfies

∣2ε∫BRη′(u/ε)F(u)∇u⋅ϕ dx∣≤C∫{0<u<2ε}∩supp⁡ϕ∣∇u∣∣ϕ∣ dx⟶0.(26)\left|\frac{2}{\varepsilon}\int_{B_R}\eta'(u/\varepsilon)F(u)\nabla u\cdot\phi\,dx\right| \le C\int_{\{0<u<2\varepsilon\}\cap\operatorname{supp}\phi}|\nabla u||\phi|\,dx \longrightarrow0. \tag*{(26)}

Indeed, the integrand is bounded by an integrable function independent of ε\varepsilon, and the sets {0<u<2ε}\{0<u<2\varepsilon\} decrease to the empty set.

For the other term, the gradient estimate gives

∣∫BR∇u⋅ϕ dμε∣≤C0∫BRxn∣ϕ∣ dμε.\left|\int_{B_R}\nabla u\cdot\phi\,d\mu_\varepsilon\right| \le\sqrt{C_0}\int_{B_R}\sqrt{x_n}|\phi|\,d\mu_\varepsilon.

The function xn∣ϕ∣\sqrt{x_n}|\phi|, extended by zero outside the support of ϕ\phi, is continuous and compactly supported in BRB_R. The right-hand side consequently converges by (23). On the left of (25), dominated convergence applies because uu is locally Lipschitz, F(u)F(u) is bounded, and both ∣∇u∣2|\nabla u|^2 and F(u)F(u) vanish almost everywhere on the zero set. We obtain

∣∫BR((∣∇u∣2−2F(u))div⁡ϕ−2∇uDϕ∇u) dx∣≤C0∫BRxn∣ϕ∣ dμ.(27)\left|\int_{B_R}\left((|\nabla u|^2-2F(u))\operatorname{div}\phi-2\nabla uD\phi\nabla u\right)\,dx\right| \le\sqrt{C_0}\int_{B_R}\sqrt{x_n}|\phi|\,d\mu. \tag*{(27)}

Use the domain-variation identity (6) to replace the integral on the left of (27). Since

xndiv⁡ϕ+ϕn=div⁡(xnϕ),x_n\operatorname{div}\phi+\phi_n=\operatorname{div}(x_n\phi),

we find

∣∫BRχ{u>0}div⁡(xnϕ) dx∣≤C0∫BRxn∣ϕ∣ dμ.(28)\left|\int_{B_R}\chi_{\{u>0\}}\operatorname{div}(x_n\phi)\,dx\right| \le\sqrt{C_0}\int_{B_R}\sqrt{x_n}|\phi|\,d\mu. \tag*{(28)}

For any ζ∈Cc1(BR∩{xn>0};Rn)\zeta\in C_c^1(B_R \cap\{x_n>0\};\mathbb{R}^n), the vector field ϕ=ζ/xn\phi=\zeta/x_n is admissible. Thus

∣∫BRχ{u>0}div⁡ζ dx∣≤C0∫BR∣ζ∣xn dμ.(29)\left|\int_{B_R}\chi_{\{u>0\}}\operatorname{div}\zeta\,dx\right|\leq\sqrt{C_0}\int_{B_R}\frac{|\zeta|}{\sqrt{x_n}}\,d\mu. \tag*{(29)}

On each compact subset of BR∩{xn>0}B_R\cap\{x_n>0\}, the measure on the right is finite. To apply the dual characterization of BVBV, restrict to an open set compactly contained in this region and take the supremum of the left-hand side of (29) over vector fields supported there with ∣ζ∣≤1|\zeta|\leq1. The right-hand side is bounded by the mass of C0xn−1/2μ\sqrt{C_0}x_n^{-1/2}\mu on that open set, so the defining total-variation supremum is finite.

Repeating this on all relatively compact open subsets, and using regularity of the resulting Radon measures, yields

χ{u>0}∈BVloc(BR∩{xn>0}),∣∇χ{u>0}∣≤C0xnμ.\chi_{\{u>0\}}\in BV_{\mathrm{loc}}(B_R\cap\{x_n>0\}),\qquad|\nabla\chi_{\{u>0\}}|\leq\frac{\sqrt{C_0}}{\sqrt{x_n}}\mu.

Multiplication by xn\sqrt{x_n} proves (24). By exhaustion of Br∩{xn>0}B_r\cap\{x_n>0\} and monotone convergence,

P(r)≤C0μ(Br)<∞(0<r<R).(30)P(r)\leq\sqrt{C_0}\mu(B_r)<\infty\qquad(0<r<R). \tag*{(30)}

The measure μ\mu is finite on Br‾\overline{B_r} due to the fact that this set is compactly contained in BRB_R.

Proof of Theorem I

In this section, we prove Theorem I. The argument combines the weighted perimeter and spherical-flux estimates with a truncation of the positive phase and the sharp xn1/2x_n^{1/2}-weighted isoperimetric inequality. After absorbing the vorticity error, this yields a differential inequality for the weighted phase volume, whose integration gives the desired positive lower bound for the Weiss weighted density. The following lemma relates the weighted volume and the interior perimeter of the positive phase in BrB_r, which are exactly the quantities defined in (12).

Lemma 4.1. Let n≥2n\geq2, let f∈C([0,∞))f\in C([0,\infty)), and let uu be a variational solution of {(xrefext}(1.1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R=B_R(0)\Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR|\nabla u|^2\leq C_0x_n^+\quad\text{a.e. in }B_R

for some C0>0C_0>0. Set

M0=max⁡{f+(s):0≤s≤23C0R3/2}.M_0=\max\left\{f^+(s):0\leq s\leq\frac{2}{3}\sqrt{C_0}R^{3/2}\right\}.

Let

μ=Δu+f(u)χ{u>0} dx,\mu=\Delta u+f(u)\chi_{\{u>0\}}\,dx,

and let VV and PP be defined by (12). For almost every r∈(0,R)r\in(0,R), μ(Br)\mu(B_r) equals the flux of uu across ∂Br\partial B_r plus the vorticity term, and

P(r)≤C0V′(r)+C0M0∣Br∩{u>0}∣.P(r)\leq C_0V'(r)+\sqrt{C_0}M_0|B_r\cap\{u>0\}|.

Proof. The weighted volume VV is absolutely continuous on compact intervals of radii, and the coarea formula gives

V′(r)=∫∂Brxn+χ{u>0} dHn−1(31)V'(r)=\int_{\partial B_r}\sqrt{x_n^+}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1} \tag*{(31)}

for almost every r∈(0,R)r\in(0,R).

We now express the mass of μ\mu in a ball through its spherical flux.

Fix 0<r<R0<r<R and 0<h<R−r0<h<R-r. Define the radial cut-off function ηh\eta_h by

ηh(x)={1,∣x∣≤r,r+h−∣x∣h,r<∣x∣<r+h,0,∣x∣≥r+h.\eta_h(x)= \begin{cases} 1, & |x|\le r,\\ \dfrac{r+h-|x|}{h}, & r<|x|<r+h,\\ 0, & |x|\ge r+h. \end{cases}

By smooth approximation, ηh\eta_h is admissible in the distributional identity (19) for μ\mu. Testing with ηh\eta_h yields

∫BRηh dμ=1h∫Br+h∖Br∇u⋅x∣x∣ dx+∫BRηhf(u)χ{u>0} dx.(32)\int_{B_R}\eta_h\,\mathrm{d}\mu = \frac{1}{h}\int_{B_{r+h}\setminus B_r}\nabla u\cdot\frac{x}{|x|}\,\mathrm{d}x + \int_{B_R}\eta_h f(u)\chi_{\{u>0\}}\,\mathrm{d}x. \tag*{(32)}

We pass to the limit as h→0+h\to0^+, beginning with the left-hand side. Dominated convergence implies that

lim⁡h→0+∫BRηh dμ=μ(Br‾)=μ(Br)\lim_{h\to0^+}\int_{B_R}\eta_h\,\mathrm{d}\mu = \mu(\overline{B_r})=\mu(B_r)

whenever μ(∂Br)=0\mu(\partial B_r)=0, which holds for almost every r∈(0,R)r\in(0,R).

For the first term on the right-hand side, the coarea formula gives

1h∫Br+h∖Br∇u⋅x∣x∣ dx=1h∫rr+h∫∂Bs∇u⋅ν dHn−1 ds.\frac{1}{h}\int_{B_{r+h}\setminus B_r}\nabla u\cdot\frac{x}{|x|}\,\mathrm{d}x = \frac{1}{h}\int_r^{r+h}\int_{\partial B_s}\nabla u\cdot\nu\,\mathrm{d}\mathcal{H}^{n-1}\,\mathrm{d}s.

The spherical flux is locally integrable in ss. The Lebesgue differentiation theorem therefore shows that this expression converges to

∫∂Br∇u⋅ν dHn−1\int_{\partial B_r}\nabla u\cdot\nu\,\mathrm{d}\mathcal{H}^{n-1}

for almost every r∈(0,R)r\in(0,R).

For the remaining term, the boundedness of f(u)f(u) in BRB_R and dominated convergence give

lim⁡h→0+∫BRηhf(u)χ{u>0} dx=∫Brf(u)χ{u>0} dx.\lim_{h\to0^+}\int_{B_R}\eta_hf(u)\chi_{\{u>0\}}\,\mathrm{d}x = \int_{B_r}f(u)\chi_{\{u>0\}}\,\mathrm{d}x.

It follows that

μ(Br)=∫∂Br∇u⋅ν dHn−1+∫Brf(u)χ{u>0} dx(33)\mu(B_r)=\int_{\partial B_r}\nabla u\cdot\nu\,\mathrm{d}\mathcal{H}^{n-1} + \int_{B_r}f(u)\chi_{\{u>0\}}\,\mathrm{d}x \tag*{(33)}

for almost every r∈(0,R)r\in(0,R).

By Fubini’s theorem in polar coordinates and (8), the growth bound implies that, for almost every rr,

∫∂Br∇u⋅ν dHn−1≤∫∂Br∣∇u∣χ{u>0} dHn−1≤C0∫∂Brxn+χ{u>0} dHn−1=C0V′(r).\int_{\partial B_r}\nabla u\cdot\nu\,\mathrm{d}\mathcal{H}^{n-1} \le \int_{\partial B_r}|\nabla u|\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1} \le \sqrt{C_0}\int_{\partial B_r}\sqrt{x_n^+}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1} = \sqrt{C_0}V'(r).

Moreover, by the definition of M0M_0,

∫Brf(u)χ{u>0} dx≤M0∣Br∩{u>0}∣.\int_{B_r}f(u)\chi_{\{u>0\}}\,\mathrm{d}x \le M_0|B_r\cap\{u>0\}|.

Inserting these estimates into (33) and using (30), we conclude that

P(r)≤C0V′(r)+C0M0∣Br∩{u>0}∣(34)P(r) \le C_0 V'(r) + \sqrt{C_0 M_0}|B_r \cap\{u > 0\}| \tag*{(34)}

Proposition 4.2. Let n≥2n \ge2, let f∈C([0,∞))f \in C([0,\infty)), and let uu be a variational solution of (1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R = B_R(0) \Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR|\nabla u|^2 \le C_0 x_n^+ \quad\text{a.e. in } B_R

for some C0>0C_0 > 0. Let VV and PP be defined by (12). For almost every r∈(0,R)r \in(0,R),

∣Br∩{u>0}∣≤r [P(r)+V′(r)],|B_r \cap\{u > 0\}| \le\sqrt{r}\,[P(r) + V'(r)],

and

P(r)+V′(r)≥(n+12)(∫B1∩{xn>0}xn dx)22n+1V(r)2n−12n+1.P(r) + V'(r) \ge\left(n+\frac{1}{2}\right)\left(\int_{B_1\cap\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}} V(r)^{\frac{2n-1}{2n+1}}.

Proof. We first describe the idea. The positive phase {u>0}\{u > 0\} may fail to have finite perimeter down to the plane {xn=0}\{x_n = 0\}, so the isoperimetric inequality (17) and the volume comparison (18) cannot be applied directly. Instead, we truncate the positive phase at a positive height, apply the two geometric inequalities to the truncated sets, and then let the height tend to zero. The technical work is to choose admissible radii and heights for which the boundary traces are well defined, to decompose the weighted perimeter into its interior, spherical and horizontal contributions, and to show that the horizontal contribution disappears in the limit.

We divide the proof into six steps.

Step 1: Selection of an admissible radius. Choose 0<r<R0 < r < R for which (31), (33), and (34) hold and μ(∂Br)=0\mu(\partial B_r) = 0. We also require the interior BVBV trace of χ{u>0}\chi_{\{u>0\}} on the sphere, at positive height, to agree Hn−1\mathcal{H}^{n-1}-almost everywhere with its restriction to that sphere. This additional requirement holds for almost every radius.

Exhaust BR∩{xn>0}B_R \cap\{x_n > 0\} by a countable collection of relatively compact regions on which χ{u>0}\chi_{\{u>0\}} has finite variation, and apply the slicing and trace theorems for BVBV functions on each region [6]. The coarea formula identifies the slice with the chosen representative of χ{u>0}\chi_{\{u>0\}} for almost every radius. Intersecting these countably many full-measure sets with those described above still leaves a set of full measure in (0,R)(0,R).

For this choice of rr, (24) and μ(∂Br)=0\mu(\partial B_r) = 0 also imply that the original perimeter measure gives no mass to ∂Br∩{xn>0}\partial B_r \cap\{x_n > 0\}. We keep this radius fixed throughout.

Step 2: Finite perimeter and selection of admissible heights. For 0<ε<r0 < \varepsilon< r, set

Aε=Br∩{u>0}∩{xn>ε}.A_\varepsilon= B_r \cap\{u > 0\} \cap\{x_n > \varepsilon\}.

The closed cap Br∩{xn>ε}‾\overline{B_r \cap\{x_n > \varepsilon\}} has a neighborhood compactly contained in BR∩{xn>0}B_R \cap\{x_n > 0\}. On that neighborhood, χ{u>0}\chi_{\{u>0\}} has finite variation by (24), since the weight is bounded below by a positive constant there. The cap is a bounded Lipschitz domain. The BV restriction and zero-extension formula, including its boundary trace term, therefore gives

χAε∈BV(Rn)(0<ε<r).\chi_{A_\varepsilon} \in BV(\mathbb{R}^{n}) \qquad(0 < \varepsilon< r).

In fact, the interior variation is finite and the boundary trace lies between 0 and 1, so its integral over the two faces of the cap is finite. Thus finite perimeter does not require any special choice of ε\varepsilon. Also, Aε⊂{xn≥ε}A_\varepsilon\subset\{x_n \ge\varepsilon\}, so both geometric inequalities apply to AεA_\varepsilon.

Good heights are needed to express the boundary traces using χ{u>0}\chi_{\{u>0\}} itself. For the fixed radius above, apply the horizontal slicing theorem on a countable exhaustion away from xn=0x_n=0. This gives a full-measure set of heights at which the trace from above on Br∩{xn=ε}B_r \cap\{x_n=\varepsilon\} agrees with the slice of χ{u>0}\chi_{\{u>0\}} almost everywhere.

We may also exclude the heights whose planes carry original perimeter mass. On each region of finite variation there are at most countably many such heights. Choose from the remaining full-measure set a sequence

0<εj+1<12εj<εj<r,εj⟶0.0 < \varepsilon_{j+1} < \frac{1}{2}\varepsilon_j < \varepsilon_j < r,\qquad\varepsilon_j \longrightarrow0.

In the next two steps, ε\varepsilon denotes one of these good heights.

Step 3: Decomposition of the distributional derivative. Apply the BV Gauss–Green formula to χ{u>0}\chi_{\{u>0\}} in the cap Br∩{xn>ε}B_r\cap\{x_n>\varepsilon\}. The outward normal is ν=x/r\nu=x/r on its spherical face and −en-e_n on its horizontal face. By the trace identifications just established, for every ϕ∈Cc1(Rn;Rn)\phi\in C^1_c(\mathbb{R}^n;\mathbb{R}^n) the formula reads

−∫Aεdiv⁡ϕ dx=∫Br∩{xn>ε}ϕ⋅d(∇χ{u>0})−∫∂Br∩{xn>ε}χ{u>0}ϕ⋅ν dHn−1+∫Br∩{xn=ε}χ{u>0}ϕn dHn−1.(35)-\int_{A_\varepsilon} \operatorname{div}\phi\,\mathrm{d}x = \int_{B_r\cap\{x_n>\varepsilon\}} \phi\cdot\mathrm{d}(\nabla\chi_{\{u>0\}}) -\int_{\partial B_r\cap\{x_n>\varepsilon\}} \chi_{\{u>0\}}\phi\cdot\nu\,\mathrm{d}\mathcal{H}^{n-1} +\int_{B_r\cap\{x_n=\varepsilon\}} \chi_{\{u>0\}}\phi_n\,\mathrm{d}\mathcal{H}^{n-1}. \tag*{(35)}

The left-hand side is the action of ∇χAε\nabla\chi_{A_\varepsilon} on ϕ\phi, so (35) is an identity of vector-valued Radon measures whose right-hand side splits into the derivative on the open cap and two surface measures. The signs come from the outward normals, and the two faces meet in a set of zero Hn−1\mathcal{H}^{n-1} measure.

Step 4: Weighted perimeter decomposition and estimates. The three measures in (35) are mutually singular: the first is restricted to the open cap, while the other two lie on its distinct boundary faces, up to their null intersection. Consequently their total variations add, without cancellation.

Since the traces are nonnegative, multiplying by xn\sqrt{x_n} and integrating gives the exact identity

∫{xn>0}xn d∣∇χAε∣=∫Br∩{xn>ε}xn d∣∇χ{u>0}∣+∫∂Br∩{xn>ε}xnχ{u>0} dHn−1+ε∫Br∩{xn=ε}χ{u>0} dHn−1.\int_{\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}|\nabla\chi_{A_\varepsilon}| = \int_{B_r\cap\{x_n>\varepsilon\}} \sqrt{x_n}\,\mathrm{d}|\nabla\chi_{\{u>0\}}| +\int_{\partial B_r\cap\{x_n>\varepsilon\}} \sqrt{x_n}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1} +\sqrt{\varepsilon}\int_{B_r\cap\{x_n=\varepsilon\}} \chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1}.

Each term now has a direct bound. The first is at most P(r)P(r) by (12). The second is at most V′(r)V'(r) by (31), since its integrand is nonnegative and its domain is a subset of the sphere.

For the third, use 0≤χ{u>0}≤10 \leq\chi_{\{u>0\}} \leq1. We obtain

∫{xn>0}xn d∣∇χAε∣≤P(r)+∫∂Br∩{xn>ε}xnχ{u>0} dHn−1+εHn−1(Br∩{xn=ε})≤P(r)+V′(r)+ε ωn−1rn−1.(36)\begin{aligned} \int_{\{x_n>0\}} \sqrt{x_n}\,\mathrm{d}|\nabla\chi_{A_\varepsilon}| \leq P(r) + \int_{\partial B_r\cap\{x_n>\varepsilon\}} \sqrt{x_n}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1} + \sqrt{\varepsilon}\mathcal{H}^{n-1}(B_r\cap\{x_n=\varepsilon\}) \\ &\leq P(r) + V'(r) + \sqrt{\varepsilon}\,\omega_{n-1}r^{n-1}. \tag*{(36)} \end{aligned}

The last inequality uses

Hn−1(Br∩{xn=ε})=ωn−1(r2−ε2)(n−1)/2≤ωn−1rn−1.\mathcal{H}^{n-1}(B_r\cap\{x_n=\varepsilon\})=\omega_{n-1}(r^2-\varepsilon^2)^{(n-1)/2}\leq\omega_{n-1}r^{n-1}.

The unweighted area of the horizontal cut need not tend to zero. Its weighted contribution nevertheless tends to zero because the weight is the constant ε\sqrt{\varepsilon} on that cut. The bound is uniform over all good heights. Figure 2 illustrates the truncation in the planar case.

Figure 2. The truncated phase Aε=Br∩{u>0}∩{x2>ε}A_\varepsilon=B_r\cap\{u>0\}\cap\{x_2>\varepsilon\}.

Step 5: Passage to the limit in the truncation parameter. Because u=0u=0 on {xn≤0}\{x_n\leq0\} and εj→0+\varepsilon_j\to0^+, we have

Aεj→Br∩{u>0}.A_{\varepsilon_j}\to B_r\cap\{u>0\}.

The weighted and unweighted volumes therefore converge by monotone convergence. More explicitly, the omitted part is contained in the strip Br∩{0<xn≤εj}B_r\cap\{0<x_n\leq\varepsilon_j\}. Horizontal slicing and the bound on the area of a section of BrB_r give

0≤V(r)−∫Aεjxn dx≤ωn−1rn−1∫0εjt dt=23ωn−1rn−1εj3/2⟶0,\begin{aligned} 0\leq V(r)-\int_{A_{\varepsilon_j}}\sqrt{x_n}\,\mathrm{d}x\leq\omega_{n-1}r^{n-1}\int_0^{\varepsilon_j}\sqrt{t}\,\mathrm{d}t \\ &=\frac{2}{3}\omega_{n-1}r^{n-1}\varepsilon_j^{3/2}\longrightarrow0, \end{aligned}

and

0≤∣Br∩{u>0}∣−∣Aεj∣≤ωn−1rn−1εj⟶0.0\leq|B_r\cap\{u>0\}|-|A_{\varepsilon_j}|\leq\omega_{n-1}r^{n-1}\varepsilon_j\longrightarrow0.

The exact perimeter identity in Step 4 also permits passing to the limit term by term. Monotone convergence with respect to the weighted variation measure gives

∫Br∩{xn>εj}xn d∣∇χ{u>0}∣→P(r)<∞.\int_{B_r\cap\{x_n>\varepsilon_j\}}\sqrt{x_n}\,\mathrm{d}\lvert\nabla\chi_{\{u>0\}}\rvert\to P(r)<\infty.

On the fixed sphere, monotone convergence and (31) give

∫∂Br∩{xn>εj}xnχ{u>0} dHn−1→V′(r).\int_{\partial B_r\cap\{x_n>\varepsilon_j\}}\sqrt{x_n}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1}\to V'(r).

Finally, the horizontal term satisfies

0≤εj∫Br∩{xn=εj}χ{u>0} dHn−1≤ωn−1rn−1εj⟶0.0\leq\sqrt{\varepsilon_j}\int_{B_r\cap\{x_n=\varepsilon_j\}}\chi_{\{u>0\}}\,\mathrm{d}\mathcal{H}^{n-1}\leq\omega_{n-1}r^{n-1}\sqrt{\varepsilon_j}\longrightarrow0.

Adding these three limits yields

lim⁡j→∞∫{xn>0}xn d∣∇χAεj∣=P(r)+V′(r).\lim_{j\to\infty}\int_{\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}\lvert\nabla\chi_{A_{\varepsilon_j}}\rvert=P(r)+V'(r).

Step 6: Application of the geometric inequalities. For each jj, AεjA_{\varepsilon_j} is a set of finite perimeter whose closure stays at positive height. Apply (17) to this set and then use (36) to obtain

(n+12)(∫B1∩{xn>0}xn dx)22n+1(∫Aεjxn dx)2n−12n+1≤∫{xn>0}xn d∣∇χAεj∣≤P(r)+V′(r)+εj ωn−1rn−1.\begin{aligned} \left(n+\frac{1}{2}\right)\left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}}\left(\int_{A_{\varepsilon_j}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2n-1}{2n+1}} \leq\int_{\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}\lvert\nabla\chi_{A_{\varepsilon_j}}\rvert\\ &\leq P(r)+V'(r)+\sqrt{\varepsilon_j}\,\omega_{n-1}r^{n-1}. \end{aligned}

All terms except the explicit error have finite limits by Step 5. The error tends to zero. Since the exponent (2n−1)/(2n+1)(2n-1)/(2n+1) is positive, passage to the limit gives

P(r)+V′(r)≥(n+12)(∫B1∩{xn>0}xn dx)22n+1V(r)2n−12n+1.(37)P(r)+V'(r)\geq\left(n+\frac{1}{2}\right)\left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}}V(r)^{\frac{2n-1}{2n+1}}. \tag*{(37)}

Similarly, apply the volume comparison (18) to each truncated set before taking a limit:

∣Aεj∣≤r∫{xn>0}xn d∣∇χAεj∣≤r(P(r)+V′(r)+εj ωn−1rn−1).\begin{aligned} \lvert A_{\varepsilon_j}\rvert\leq\sqrt{r}\int_{\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}\lvert\nabla\chi_{A_{\varepsilon_j}}\rvert\\ &\leq\sqrt{r}\left(P(r)+V'(r)+\sqrt{\varepsilon_j}\,\omega_{n-1}r^{n-1}\right). \end{aligned}

The unweighted volume convergence from Step 5 and the vanishing of the last error yield

∣Br∩{u>0}∣≤r[P(r)+V′(r)].(38)\lvert B_r\cap\{u>0\}\rvert\leq\sqrt{r}\left[P(r)+V'(r)\right]. \tag*{(38)}

The radius was chosen from a fixed full-measure set, so (37) and (38) hold for almost every r<Rr<R. □

Lemma 4.3. Let n≥2n\geq2, let f∈C([0,∞))f\in C([0,\infty)), and let uu be a variational solution of (1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R=B_R(0)\Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR\lvert\nabla u\rvert^2\leq C_0x_n^+ \qquad\text{a.e. in }B_R

:::`

for some C0>0C_0>0. Assume in addition that 0∈Su0\in S^u. Set

M0=max⁡{f+(s):0≤s≤23C0R3/2}.M_0=\max\left\{f^+(s):0\le s\le\frac{2}{3}\sqrt{C_0}R^{3/2}\right\}.

Let VV and PP be defined by (12). For almost every r∈(0,R)r\in(0,R),

(1−C0M0r)[P(r)+V′(r)]≤(1+C0)V′(r),\left(1-\sqrt{C_0}M_0\sqrt{r}\right)[P(r)+V'(r)]\le(1+C_0)V'(r),

and

V(r)≥c1rn+1/2V(r)\ge c_1r^{n+1/2}

for every 0<r<R0<r<R with C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1, where c1=c1(n,C0)>0c_1=c_1(n,C_0)>0.

Proof. Insert (38) into the flux estimate (34). For almost every radius r∈(0,R)r\in(0,R),

P(r)≤C0V′(r)+C0M0r[P(r)+V′(r)].(39)P(r)\le C_0V'(r)+\sqrt{C_0}M_0\sqrt{r}[P(r)+V'(r)]. \tag*{(39)}

Adding V′(r)V'(r) and moving the last term to the left, we obtain

(1−C0M0r)[P(r)+V′(r)]≤(1+C0)V′(r).(40)\left(1-\sqrt{C_0}M_0\sqrt{r}\right)[P(r)+V'(r)]\le(1+C_0)V'(r). \tag*{(40)}

If C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1, the coefficient on the left is positive. We may therefore substitute the lower bound (37) into (40) and deduce

V′(r)≥n+1/21+C0(1−C0M0r)(∫B1∩{xn>0}xn dx)22n+1V(r)2n−12n+1.(41)V'(r)\ge\frac{n+1/2}{1+C_0}\left(1-\sqrt{C_0}M_0\sqrt{r}\right) \left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}} V(r)^{\frac{2n-1}{2n+1}}. \tag*{(41)}

Since 0∈∂{u>0}0\in\partial\{u>0\}, each ball BrB_r contains a point where uu is positive. Such a point has positive height. Continuity supplies a smaller ball around it contained in Br∩{u>0}∩{xn>0}B_r\cap\{u>0\}\cap\{x_n>0\}, and therefore

V(r)>0(0<r<R).(42)V(r)>0\qquad(0<r<R). \tag*{(42)}

Dividing by a power of VV requires (42). Fix 0<ρ<r<R0<\rho<r<R with C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1. On [ρ,r][\rho,r], the function VV is absolutely continuous and bounded below by the positive number V(ρ)V(\rho). The ordinary chain rule consequently applies and gives

ddsV(s)22n+1=22n+1V(s)−2n−12n+1V′(s)≥1−C0M0s1+C0(∫B1∩{xn>0}xn dx)22n+1\begin{aligned} \frac{\mathrm{d}}{\mathrm{d}s}V(s)^{\frac{2}{2n+1}} =\frac{2}{2n+1}V(s)^{-\frac{2n-1}{2n+1}}V'(s) \\ &\ge\frac{1-\sqrt{C_0}M_0\sqrt{s}}{1+C_0} \left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}} \end{aligned}

for almost every s∈(ρ,r)s\in(\rho,r). In the last step, the factor n+1/2n+1/2 in (41) cancels with 2/(2n+1)2/(2n+1).

Integrating from ρ\rho to rr yields

V(r)22n+1−V(ρ)22n+1≥11+C0(∫B1∩{xn>0}xn dx)22n+1(r−ρ−23C0M0(r3/2−ρ3/2)).(43)V(r)^{\frac{2}{2n+1}}-V(\rho)^{\frac{2}{2n+1}} \ge\frac{1}{1+C_0} \left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,\mathrm{d}x\right)^{\frac{2}{2n+1}} \left(r-\rho-\frac{2}{3}\sqrt{C_0}M_0(r^{3/2}-\rho^{3/2})\right). \tag*{(43)}

On the other hand,

0≤V(ρ)≤ρ∣Bρ∩{xn>0}∣=ωn2ρn+1/2⟶0.0 \le V(\rho) \le\sqrt{\rho}\lvert B_\rho\cap\{x_n>0\}\rvert= \frac{\omega_n}{2}\rho^{n+1/2} \longrightarrow0.

Letting ρ→0+\rho\to0^+ in (43), we obtain

V(r)22n+1≥r1+C0(∫B1∩{xn>0}xn dx)22n+1(1−23C0M0r).(44)V(r)^{\frac{2}{2n+1}} \ge\frac{r}{1+C_0}\left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,dx\right)^{\frac{2}{2n+1}}\left(1-\frac{2}{3}\sqrt{C_0M_0\sqrt{r}}\right). \tag*{(44)}

The last factor is positive under the radius restriction. Raising both sides to the power (2n+1)/2(2n+1)/2 gives

V(r)≥rn+1/2(1+C0)n+1/2∫B1∩{xn>0}xn dx(1−23C0M0r)n+1/2.(45)V(r) \ge\frac{r^{n+1/2}}{(1+C_0)^{n+1/2}}\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,dx\left(1-\frac{2}{3}\sqrt{C_0M_0\sqrt{r}}\right)^{n+1/2}. \tag*{(45)}

Since C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1, the last factor in parentheses is greater than 1/31/3. Thus the asserted lower bound holds with

c1(n,C0)=1[3(1+C0)]n+1/2∫B1∩{xn>0}xn dx>0.c_1(n,C_0)=\frac{1}{[3(1+C_0)]^{n+1/2}}\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,dx>0.

The differential inequality holds for almost every radius, and VV is absolutely continuous, so its integrated form is valid for every 0<r<R0<r<R satisfying C0M0r<1\sqrt{C_0}M_0\sqrt{r}<1. When M0=0M_0=0, the coefficient in (40) equals one and the argument applies on all of (0,R)(0,R). This includes the case f≤0f\le0. For general continuous ff, the coefficient approaches one as the radius decreases.

Lemma 4.4. Let n≥2n\ge2, let f∈C([0,∞))f\in C([0,\infty)), and let uu be a variational solution of (1) in Ω\Omega. Assume that BR=BR(0)⋐ΩB_R=B_R(0)\Subset\Omega and

∣∇u∣2≤C0xn+a.e. in BR|\nabla u|^2\le C_0x_n^+\quad\text{a.e. in }B_R

for some C0>0C_0>0. Assume in addition that 0∈Su0\in S^u. Set

M0=max⁡{f+(s):0≤s≤23C0R3/2}.M_0=\max\left\{f^+(s):0\le s\le\frac{2}{3}\sqrt{C_0}R^{3/2}\right\}.

For every 0<r<R0<r<R with C0M0r<1\sqrt{C_0M_0\sqrt{r}}<1,

r−n−1∫Brxn+χ{u>0} dx≥c(n,C0)(1−23C0M0r)2n+1,r^{-n-1}\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx\ge c(n,C_0)\left(1-\frac{2}{3}\sqrt{C_0M_0\sqrt{r}}\right)^{2n+1},

where

c(n,C0)=2ωn(1+C0)2n+1(∫B1∩{xn>0}xn dx)2>0.c(n,C_0)=\frac{2}{\omega_n(1+C_0)^{2n+1}}\left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,dx\right)^2>0.

Proof. The estimate obtained so far concerns the weight xn\sqrt{x_n}, whereas the Weiss weighted density uses xnx_n. Cauchy–Schwarz on the positive phase gives

V(r)2=(∫Br∩{u>0}xn dx)2≤∣Br∩{u>0}∣∫Brxn+χ{u>0} dx≤ωn2rn∫Brxn+χ{u>0} dx.\begin{aligned} V(r)^2=\left(\int_{B_r\cap\{u>0\}}\sqrt{x_n}\,dx\right)^2 \\ &\le|B_r\cap\{u>0\}|\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx \\ &\le\frac{\omega_n}{2}r^n\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx. \end{aligned}

The factor 1/21/2 is available because the positive phase lies in the upper half-ball. Rearranging and using (45), we find

r−n−1∫Brxn+χ{u>0} dx≥2V(r)2ωnr2n+1≥2ωn(1+C0)2n+1(∫B1∩{xn>0}xn dx)2×(1−23C0M0r)2n+1.\begin{aligned} r^{-n-1}\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx\ge\frac{2V(r)^2}{\omega_n r^{2n+1}} \\ &\ge\frac{2}{\omega_n(1+C_0)^{2n+1}}\left(\int_{B_1\cap\{x_n>0\}}\sqrt{x_n}\,dx\right)^2 \times\left(1-\frac{2}{3}\sqrt{C_0}M_0\sqrt{r}\right)^{2n+1}. \end{aligned}

This is (11). Letting r→0+r\to0^+ yields the positive lower bound

lim inf⁡r→0+r−n−1∫Brxn+χ{u>0} dx≥c(n,C0)>0.(46)\liminf_{r\to0^+}r^{-n-1}\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx\ge c(n,C_0)>0. \tag*{(46)}

For continuous vorticity, Proposition 3.1 gives the existence of the limiting Weiss weighted density and the identity (4). Consequently,

M(0+)=lim⁡r→0+r−n−1∫Brxn+χ{u>0} dx≥c(n,C0)>0.M(0+)=\lim_{r\to0^+}r^{-n-1}\int_{B_r}x_n^+\chi_{\{u>0\}}\,dx\ge c(n,C_0)>0.

The origin is therefore not a cusp point. Undoing the horizontal translation proves the conclusion at x0x^0 and completes the proof of Theorem I. □

Acknowledgement. L. Du is supported by National Natural Science Foundation of China under Grants 12125102, 12526202, 12671245, 12631009. Y. Pu is supported by National Natural Science Foundation of China under Grant 12571226. C. Yang is supported by National Natural Science Foundation of China under Grant 12601435.

AI declarations. OpenAI’s GPT was used only for grammar, spelling and citation checks, with all edits reviewed to safeguard mathematical reasoning.

Data availability. No data was used in this research.

References

Lili Du

Department of Mathematics, Sichuan University 610065, Chengdu, China [email protected]

Yang Pu

School of Mathematics, Southwestern University of Finance and Economics 610031, Chengdu, China [email protected]

Chunlei Yang

School of Mathematical Sciences, Shenzhen University 518060, Shenzhen, China [email protected]

References

  1. [1]L. Ambrosio, N. Fusco, and D. Pallara, Functions of Bounded Variation and Free Discontinuity Problems, Oxford Mathematical Monographs, Clarendon Press, Oxford, 2000.
  2. [2]C. J. Amick, L. E. Fraenkel, and J. F. Toland, On the Stokes conjecture for the wave of extreme form, Acta Math. 148 (1982), 193–214.DOI
  3. [3]X. Cabré and X. Ros-Oton, Sobolev and isoperimetric inequalities with monomial weights, J. Differential Equations 255 (2013), no. 11, 4312–4336.DOI
  4. [4]X. Cabré, X. Ros-Oton, and J. Serra, Sharp isoperimetric inequalities via the ABP method, J. Eur. Math. Soc. 18 (2016), no. 12, 2971–2998.arxiv.org/abs/1304.1724
  5. [5]L. Du and C. Yang, Full-density degenerate stagnation points for water waves with general vorticity, arXiv:2608.04280, 2026.arxiv.org/abs/2608.04280
  6. [6]L. C. Evans and R. F. Gariepy, Measure Theory and Fine Properties of Functions, revised ed., Textbooks in Mathematics, CRC Press, Boca Raton, 2015.
  7. [7]S. McCurdy, Non-existence of cusps for a free-boundary problem for water waves, Interfaces Free Bound. 27 (2025), no. 1, 1–11.arxiv.org/abs/2208.01873
  8. [8]P. I. Plotnikov, A proof of the Stokes conjecture in the theory of surface waves, Stud. Appl. Math. 108 (2002), no. 2, 217–244.DOI
  9. [9]G. G. Stokes, Considerations relative to the greatest height of oscillatory irrotational waves which can be propagated without change of form, Mathematical and Physical Papers I, Cambridge University Press, Cambridge, 1880, pp. 225–228.
  10. [10]E. Vărvăruță, On the existence of extreme waves and the Stokes conjecture with vorticity, J. Differential Equations 246 (2009), no. 10, 4043–4076.arxiv.org/abs/0707.2224
  11. [11]E. Vărvăruță and G. S. Weiss, A geometric approach to generalized Stokes conjectures, Acta Math. 206 (2011), no. 2, 363–403.arxiv.org/abs/0908.1031
  12. [12]E. Vărvăruță and G. S. Weiss, The Stokes conjecture for waves with vorticity, Ann. Inst. H. Poincaré Anal. Non Linéaire 29 (2012), no. 6, 861–885.DOI

Paper details

vlnc:W_y8t3sdbq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

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

  • Mathematics

GPT (multiple versions)

View statement
Looking for related papers…
Contents