betaFeedback

WorksMathematics

Rigidity of Constant-Scalar-Curvature Ricci Shrinkers with Totally Geodesic Focal Set

Abstract

Let (Mn,g,f)(M^n,g,f) be a complete connected shrinking gradient Ricci soliton with constant scalar curvature, satisfying \Ric+\nabla^2f=\frac12g. We prove that if the focal set of ff is totally geodesic, then MM is rigid. Equivalently, its universal cover is a product of a compact Einstein manifold and a Euclidean space.

Introduction

A shrinking gradient Ricci soliton is a triple (Mn,g,f)(M^n,g,f) satisfying, after normalization,

Ric⁡+∇2f=12g.(1)\operatorname{Ric}+\nabla^2 f=\frac{1}{2}g. \tag*{(1)}

Here Ric⁡\operatorname{Ric} and ∇2f\nabla^2 f denote the Ricci tensor and Hessian of the potential function ff, and RR denotes scalar curvature.

Ricci flow was introduced by Hamilton [10]. Gradient Ricci solitons generate self-similar solutions by scaling and diffeomorphisms, and shrinking solitons play a central role in the study of singularities and Perelman’s entropy [17]. Their classification is therefore a basic problem in geometric analysis. We refer to Cao [2] and Eminenti–La Nave–Mantegazza [7] for background from the geometric and elliptic viewpoints.

Classification results in low dimensions provide a starting point for this problem. In particular, Ni–Wallach [15] developed an elliptic approach to three-dimensional shrinking solitons and obtained higher-dimensional results under vanishing Weyl curvature. In general dimensions, additional geometric conditions have led to several rigidity theorems. The constant-scalar-curvature problem asks how much of this rigidity is already encoded by the scalar curvature alone.

A natural class of examples is the product of a positive Einstein manifold and a Gaussian shrinking soliton:

(Nk×Rm,gN+gRm,f),Ric⁡N=12gN,f(x,y)=∣y∣24.(N^k\times\mathbb{R}^m,g_N+g_{\mathbb{R}^m},f),\qquad\operatorname{Ric}_N=\frac{1}{2}g_N,\qquad f(x,y)=\frac{|y|^2}{4}.

Following Petersen–Wylie [18], a shrinking soliton is called rigid if it is a quotient of such a product, allowing either factor to have dimension zero. Every rigid shrinker has constant scalar curvature. The converse is the following conjecture of Cao; see, for example, [6].

Conjecture 1 (Cao’s conjecture). A complete shrinking gradient Ricci soliton has constant scalar curvature if and only if it is rigid.

Several results support this conjecture. Petersen–Wylie [18] proved that constant scalar curvature together with radial flatness, Rm⁡(⋅,∇f)∇f=0\operatorname{Rm}(\cdot,\nabla f)\nabla f=0, characterizes rigidity. Fernández-López–García-Río [8] showed that, under the constant-scalar-curvature assumption, constant rank of the Ricci tensor is sufficient. They also proved that the possible scalar curvatures are integer multiples of the soliton constant.

A related direction imposes differential conditions on curvature. Fernández-López–García-Río [9] and Munteanu–Șeșum [14] established rigidity results for shrinkers with harmonic Weyl tensor. Cao–Chen [3] classified complete Bach-flat shrinkers in dimensions at least four. Catino–Mastrolia–Monticelli [5] subsequently proved rigidity under a fourth-order divergence-free condition on the Weyl tensor. These results exhibit several curvature conditions that force an Einstein–Euclidean splitting.

In dimension four, Cheng–Zhou [6] established the remaining case of the conjecture, completing the constant-scalar-curvature classification. Ou–Qu–Wu [16] later gave another proof of the four-dimensional R=1R=1 case using the maximum principle. In dimension five, Li–Ou–Qu–Wu [12] proved rigidity when R=3/2R=3/2, identifying the shrinker as a finite quotient of R2×S3\mathbb{R}^{2}\times\mathbb{S}^{3}.

In higher dimensions, Li–Qu–Wu [13] obtained rigidity under a radial Ricci inequality outside a compact set and smooth convergence to R2×Sn−2\mathbb{R}^{2}\times\mathbb{S}^{n-2}. These results motivate the search for geometric conditions that imply rigidity in arbitrary codimension.

In this paper we prove Cao’s conjecture under the assumption that the focal set is totally geodesic, in arbitrary dimension and for every possible constant scalar curvature. Our argument applies to the general range R=k/2R=k/2 and n=k+mn=k+m, with k,m≥2k,m\ge2; the remaining cases follow from known rigidity results recalled in § 2.

To explain the geometric condition, normalize the potential by ∣∇f∣2=f\lvert\nabla f\rvert^{2}=f and let M−=f−1(0)M_{-}=f^{-1}(0) be its minimum set. For a rigid product, M−=N×{0}M_{-}=N\times\{0\} is totally geodesic. Our main result shows that this property is sufficient for rigidity. We first record the global focal structure used in the proof.

Theorem 1.1 (Automatic global focal structure). Let (Mk+m,g,f)(M^{k+m},g,f) be complete and connected, with k,m≥2k,m\ge2, and suppose

Ric⁡+∇2f=12g,R=k2.\operatorname{Ric}+\nabla^{2}f=\frac{1}{2}g,\qquad R=\frac{k}{2}.

After translating ff so that ∣∇f∣2=f\lvert\nabla f\rvert^{2}=f, the set M−=f−1(0)M_{-}=f^{-1}(0) is a compact connected smooth submanifold of dimension kk. Moreover

r:=2f=d(M−,⋅),Ric⁡♯∣M−=12ΠTM−,r:=2\sqrt{f}=d(M_{-},\cdot),\qquad\left.\operatorname{Ric}^{\sharp}\right|_{M_{-}}=\frac{1}{2}\Pi_{T M_{-}},
Exp⁡⊥:νM−⟶Mis a diffeomorphism.(2)\operatorname{Exp}^{\perp}:\nu M_{-}\longrightarrow M\quad\text{is a diffeomorphism.} \tag*{(2)}

With this focal structure in place, we can state the main rigidity theorem.

Theorem 1.2 (Arbitrary-codimension focal rigidity). Let (Mk+m,g,f)(M^{k+m},g,f) be a complete connected shrinking gradient Ricci soliton satisfying

Ric⁡+∇2f=12g,R=k2,\operatorname{Ric}+\nabla^{2}f=\frac{1}{2}g,\qquad R=\frac{k}{2},

with k,m≥2k,m\ge2. Normalize ff by ∣∇f∣2=f|\nabla f|^{2}=f and put M−=f−1(0)M_{-}=f^{-1}(0). If M−M_{-} is totally geodesic, then

(M~,g~)≅(Nk,gN)×Rm,Ric⁡gN=12gN,(3)(\widetilde{M},\widetilde{g})\cong(N^{k},g_{N})\times\mathbb{R}^{m},\qquad\operatorname{Ric}_{g_{N}}=\frac{1}{2}g_{N}, \tag*{(3)}

where NN is compact and simply connected.

Combining Theorem 1.2 with the known boundary cases gives the following form of Cao’s conjecture.

Corollary 1.1. Let (Mn,g,f)(M^{n},g,f) be a complete connected shrinking gradient Ricci soliton with constant scalar curvature, satisfying (1.1). Normalize ff by ∣∇f∣2=f|\nabla f|^{2}=f. If M−=f−1(0)M_{-}=f^{-1}(0) is totally geodesic, then MM is rigid.

The corollary imposes no restriction on the dimension or on the constant scalar curvature beyond the values allowed by the soliton equation. In the splitting, the Einstein or Euclidean factor may have dimension zero.

Remark 1.1. Total geodesy is used to eliminate the first normal derivative of the Ricci projection defect. The higher-order argument requires neither flatness of the normal connection nor triviality of the normal bundle.

The proof compares two estimates for an averaged curvature term on the normal spheres. Their difference excludes every normal degree at least two. Real analyticity then turns this local vanishing into the identity

(Ric⁡♯)2=12Ric⁡♯,(4)(\operatorname{Ric}^{\sharp})^{2}=\frac{1}{2}\operatorname{Ric}^{\sharp}, \tag*{(4)}

which gives constant Ricci rank and hence rigidity.

The paper is organized as follows. Section 2 collects the basic soliton identities and level-set formulas. Sections 3–5 establish the focal structure, the Ricci projection defect, and its first normal derivative. Sections 6–8 develop the scalar balance and spherical estimates and exclude all higher normal coefficients. Section 9 completes the proof by analytic continuation and establishes Corollary 1.1.

Preliminaries

In this section we recall the standard soliton identities, regular level-set formulas, and known rigidity results used below. We state these facts in our notation without repeating their proofs.

Throughout the paper, RR denotes scalar curvature and Δ=tr⁡∇2\Delta=\operatorname{tr}\nabla^{2}. Our curvature convention is

Rm⁡(X,Y)Z=∇X∇YZ−∇Y∇XZ−∇[X,Y]Z.\operatorname{Rm}(X,Y)Z=\nabla_{X}\nabla_{Y}Z-\nabla_{Y}\nabla_{X}Z-\nabla_{[X,Y]}Z.

so that ⟨Rm⁡(X,Y)Y,X⟩\langle\operatorname{Rm}(X,Y)Y,X\rangle is positive on the unit sphere for orthonormal X,YX,Y. We use the weighted Laplacian Δf=Δ−∇∇f\Delta_f=\Delta-\nabla_{\nabla f}, with covariant differentiation when it acts on tensors.

The following identities are standard; see [2, 7, 18].

Lemma 2.1 (Basic soliton identities). A solution of (1.1) satisfies

R+Δf=n2.(5)R+\Delta f=\frac{n}{2}. \tag*{(5)}
dR=2Ric⁡(∇f,⋅),(6)dR=2\operatorname{Ric}(\nabla f,\cdot), \tag*{(6)}
R+∣∇f∣2−f=constant,(7)R+|\nabla f|^2-f=\text{constant}, \tag*{(7)}
ΔfR=R−2∣Ric⁡∣2.(8)\Delta_f R=R-2|\operatorname{Ric}|^2. \tag*{(8)}

Moreover, for vector fields X,Y,ZX,Y,Z,

(∇XRic⁡)(Y,Z)−(∇YRic⁡)(X,Z)=−⟨Rm⁡(X,Y)∇f,Z⟩.(9)(\nabla_X\operatorname{Ric})(Y,Z)-(\nabla_Y\operatorname{Ric})(X,Z)=-\langle\operatorname{Rm}(X,Y)\nabla f,Z\rangle. \tag*{(9)}

In particular, ∇Ric⁡\nabla\operatorname{Ric} is fully symmetric at every critical point of ff.

We now assume n=k+mn=k+m and R=k/2R=k/2, as in the main theorems. The following normalization and identities are the constant-scalar-curvature specialization of Lemma 2.1; see also [8].

Corollary 2.1. After adding a constant to ff, one has

∣∇f∣2=f≥0.(10)|\nabla f|^2=f\geq0. \tag*{(10)}

Furthermore,

Ric⁡(∇f,⋅)=0,∣Ric⁡∣2=k4,(11)\operatorname{Ric}(\nabla f,\cdot)=0,\qquad|\operatorname{Ric}|^2=\frac{k}{4}, \tag*{(11)}
Δf=m2,Δff=m2−f.(12)\Delta f=\frac{m}{2},\qquad\Delta_f f=\frac{m}{2}-f. \tag*{(12)}

For endomorphism calculations we write A=Ric⁡♯A=\operatorname{Ric}^{\sharp}, that is, g(AX,Y)=Ric⁡(X,Y)g(AX,Y)=\operatorname{Ric}(X,Y). Thus

A∇f=0,tr⁡A=k2,∣A∣2=k4.(13)A\nabla f=0,\qquad\operatorname{tr}A=\frac{k}{2},\qquad|A|^2=\frac{k}{4}. \tag*{(13)}

On {f>0}\{f>0\}, put r=2fr=2\sqrt{f}, ν=∇r\nu=\nabla r, and p=m−1p=m-1. We denote a regular level set of rr by Σr\Sigma_r and its shape operator with respect to ν\nu by SS.

We record the standard level-set formulas in the radial variable rr. They follow from the soliton, Gauss, and Riccati equations; see [1, 18, 13].

Lemma 2.2 (Regular level sets). On {f>0}\{f>0\} one has

∣∇r∣=1,Aν=0,∇νν=0,Δr=pr.(14)|\nabla r|=1,\qquad A\nu=0,\qquad\nabla_\nu\nu=0,\qquad\Delta r=\frac{p}{r}. \tag*{(14)}

On TΣrT\Sigma_r, the shape operator satisfies

S=(∇2r∣TΣr)♯=1r(Id⁡−2A),tr⁡S=pr,∣S∣2=pr2.(15)S=(\nabla^2r|_{T\Sigma_r})^\sharp=\frac{1}{r}(\operatorname{Id}-2A),\qquad\operatorname{tr}S=\frac{p}{r},\qquad|S|^2=\frac{p}{r^2}. \tag*{(15)}

Moreover, the induced Ricci endomorphism is

Ric⁡Σr♯=A−2r2∇νA+p−1r2(Id⁡−2A).(16)\operatorname{Ric}^{\sharp}_{\Sigma_r}=A-\frac{2}{r^2}\nabla_\nu A+\frac{p-1}{r^2}(\operatorname{Id}-2A). \tag*{(16)}

Remark 2.1. Tracing the Gauss equation also gives

RΣr=R−2Ric⁡(ν,ν)+(tr⁡S)2−∣S∣2=k2+p(p−1)r2.R_{\Sigma_r}=R-2\operatorname{Ric}(\nu,\nu)+(\operatorname{tr}S)^2-|S|^2=\frac{k}{2}+\frac{p(p-1)}{r^2}.

Thus the scalar curvature of the full level set is constant along each level. The scalar curvature of an individual normal sphere requires the additional fiber calculation in § 6.

We also use the properness of the potential and its compact minimum set, following Cao–Zhou [4], and the local real analyticity of the soliton metric, following Kotschwar [11].

Lemma 2.3. Under the hypotheses of Theorem 1.1, MM is noncompact, ff is proper, and M−=f−1(0)M_- = f^{-1}(0) is nonempty and compact. In addition, gg is locally real analytic in suitable coordinates.

Finally, the following known classification facts identify the range that remains to be treated by the focal-set argument.

Lemma 2.4 (Known scalar-curvature cases). Let (Mn,g,f)(M^n,g,f) be a complete shrinking gradient Ricci soliton satisfying (1.1), with constant scalar curvature RR. Then

R∈{0,12,…,n−12,n2}.R\in\left\{0,\frac{1}{2},\ldots,\frac{n-1}{2},\frac{n}{2}\right\}.

For n≥2n\ge2, the value R=1/2R=1/2 does not occur. The cases R=0R=0, R=(n−1)/2R=(n-1)/2, and R=n/2R=n/2 are rigid. In the cases R=0R=0 and R=n/2R=n/2, the universal cover is Gaussian and Einstein, respectively.

The discreteness of RR, the exclusion of R=1/2R=1/2, and the case R=(n−1)/2R=(n-1)/2 are due to Fernández–López–García–Río [8]. The extremal cases are included in Petersen–Wylie [18]. We will also use the constant-rank criterion of [8]: a complete shrinking gradient Ricci soliton with constant scalar curvature and constant Ricci rank is rigid.

The automatic global focal structure

We prove Theorem 1.1. The argument separates the local transnormal information at the minimum set from a global eikonal argument that rules out a cut locus.

Proof of Theorem 1.1. By Lemma 2.3, M−M_- is nonempty and compact. We first determine its smooth structure and the Hessian of ff along it.

The identities ∣∇f∣2=f|\nabla f|^2=f and Δf=m/2\Delta f=m/2 show that ff is isoparametric and in particular transnormal. Wang’s focal-variety theorem for a transnormal function on a complete connected manifold implies that the nonempty minimum focal variety M−M_- is a smooth submanifold [20]. For b(t)=tb(t)=t in ∣∇f∣2=b(f)|\nabla f|^2=b(f), the Hessian at the minimum variety satisfies

∇2f∣TM−=0,∇2f∣νM−=b′(0)2g=12g,∇2f(TM−,νM−)=0.(17)\left.\nabla^2 f\right|_{TM_-}=0,\qquad\left.\nabla^2 f\right|_{\nu M_-}=\frac{b'(0)}{2}g=\frac{1}{2}g,\qquad\nabla^2 f(TM_-,\nu M_-)=0. \tag*{(17)}

Taking the trace and using (2.8) yields codim⁡M−=m\operatorname{codim} M_{-}=m, hence dim⁡M−=k\dim M_{-}=k. The soliton equation then gives

Ric⁡∣TM−=12g∣TM−,Ric⁡∣νM−=0,Ric⁡(TM−,νM−)=0,(18)\left.\operatorname{Ric}\right|_{TM_{-}}=\frac{1}{2}\left.g\right|_{TM_{-}},\qquad\left.\operatorname{Ric}\right|_{\nu M_{-}}=0,\qquad\operatorname{Ric}(TM_{-},\nu M_{-})=0, \tag*{(18)}

which is exactly Ric⁡♯∣M−=12ΠTM−\left.\operatorname{Ric}^{\sharp}\right|_{M_{-}}=\frac{1}{2}\Pi_{TM_{-}}.

We now justify the global normal coordinates carefully. On M∖M−M\setminus M_{-} define

r=2f.r=2\sqrt{f}.

Then (2.6) gives

∣∇r∣2=1.(19)|\nabla r|^{2}=1. \tag*{(19)}

Differentiating (3.3) gives ∇2r(∇r,⋅)=0\nabla^{2}r(\nabla r,\cdot)=0. Hence the integral curves of ±∇r\pm\nabla r are unit-speed geodesics, and rr changes on them with speed ±1\pm1.

Fix x∈M∖M−x\in M\setminus M_{-} and let γ\gamma be the geodesic with γ(0)=x\gamma(0)=x, γ˙(0)=−∇r(x)\dot{\gamma}(0)=-\nabla r(x). For 0≤t<r(x)0\leq t<r(x),

r(γ(t))=r(x)−t>0,r(\gamma(t))=r(x)-t>0,

so the curve remains in M∖M−M\setminus M_{-}. Completeness extends γ\gamma to t=r(x)t=r(x), and continuity gives q:=γ(r(x))∈M−q:=\gamma(r(x))\in M_{-}. Thus the segment has length r(x)r(x) and

d(x,M−)≤r(x).(20)d(x,M_{-})\leq r(x). \tag*{(20)}

Conversely, if cc is any piecewise smooth curve from M−M_{-} to xx, apply ∣∇r∣=1|\nabla r|=1 to the portion where r≥εr\geq\varepsilon and let ε↓0\varepsilon\downarrow0 to obtain r(x)≤L(c)r(x)\leq L(c). Taking the infimum gives

r=d(M−,⋅).(21)r=d(M_{-},\cdot). \tag*{(21)}

It remains to show that Exp⁡⊥\operatorname{Exp}^{\perp} has neither identifications nor focal points. By the tubular-neighborhood theorem, for q∈M−q\in M_{-} and unit θ∈νqM−\theta\in\nu_qM_{-}, the normal geodesic σ(t)=exp⁡q(tθ)\sigma(t)=\exp_q(t\theta) satisfies r(σ(t))=tr(\sigma(t))=t and σ˙(t)=∇r\dot{\sigma}(t)=\nabla r for small t>0t>0. Both σ\sigma and the integral curves of ∇r\nabla r solve the geodesic equation. If there were a first finite time at which either identity failed, then at that time r=t>0r=t>0 and ∇r\nabla r would still be smooth; uniqueness of the geodesic ODE would extend the identities past that time. Therefore

r(exp⁡q(tθ))=t,ddtexp⁡q(tθ)=∇r(t>0).(22)r(\exp_q(t\theta))=t,\qquad\frac{d}{dt}\exp_q(t\theta)=\nabla r\qquad(t>0). \tag*{(22)}

By (3.5), every such normal segment minimizes the distance to M−M_{-} for every t>0t>0. Surjectivity of Exp⁡⊥\operatorname{Exp}^{\perp} follows from the backward −∇r-\nabla r geodesic constructed above; the first-variation formula shows that its terminal segment is orthogonal to M−M_{-}. For injectivity, if

exp⁡q(v)=exp⁡q′(v′)=x,\exp_q(v)=\exp_{q'}(v')=x,

then (3.6) gives ∣v∣=∣v′∣=r(x)|v|=|v'|=r(x) and both terminal velocities are ∇r(x)\nabla r(x). A geodesic is uniquely determined by terminal position and velocity, hence the two normal geodesics coincide and (q,v)=(q′,v′)(q,v)=(q',v').

Finally, dExp⁡⊥d\operatorname{Exp}^{\perp} cannot be singular at a nonzero normal vector. A singularity of the normal exponential map is a focal point of M−M_{-}; by the standard index-form characterization, a normal geodesic ceases to be minimizing beyond its first focal point. This contradicts the global minimizing property just obtained. Along the zero section, dExp⁡⊥d\operatorname{Exp}^{\perp} is the canonical isomorphism TM−⊕νM−≃TM∣M−TM_{-} \oplus\nu M_{-} \simeq TM|_{M_{-}}. Hence Exp⁡⊥\operatorname{Exp}^{\perp} is a bijective local diffeomorphism, and therefore a global diffeomorphism. The total space νM−\nu M_{-} has one connected component for each connected component of M−M_{-}. Since MM is connected and Exp⁡⊥\operatorname{Exp}^{\perp} is a diffeomorphism, M−M_{-} is connected. □

The Ricci projection defect

We next compare the Ricci endomorphism with the orthogonal projection onto the directions tangent to the minimum set. The resulting tensor has an exact square identity, which will be used in place of a curvature sign assumption. We also derive the volume identities needed when integrating over the normal spheres.

We use the endomorphism A=Ric⁡♯A = \operatorname{Ric}^{\sharp} and the identities (13) from § 2. For an orthogonal distribution, a subscript on AA denotes its compression to that distribution; it does not assert invariance under AA.

Use the normal exponential diffeomorphism to define π:M→M−\pi: M \to M_{-} and put

N=ker⁡dπ,H=N⊥.(23)\mathcal{N} = \ker d\pi,\qquad\mathcal{H} = \mathcal{N}^{\perp}. \tag*{(23)}

Both distributions extend smoothly across the zero section. On M∖M−M \setminus M_{-}, N=Rν⊕V\mathcal{N} = \mathbb{R}\nu\oplus\mathcal{V}, ν=∇r\nu= \nabla r, dim⁡V=p=m−1\dim\mathcal{V} = p = m - 1.

Define

E=A−12ΠH,v=tr⁡NA=tr⁡VA,Z=ΠHA∣N.(24)E = A - \frac{1}{2}\Pi_{\mathcal{H}},\qquad v = \operatorname{tr}_{\mathcal{N}} A = \operatorname{tr}_{\mathcal{V}} A,\qquad Z = \Pi_{\mathcal{H}} A|_{\mathcal{N}}. \tag*{(24)}

Since Aν=0A\nu= 0, the mixed block ZZ vanishes on the radial direction.

Lemma 4.1 (Exact square). On the whole normal tube,

∣E∣2=v=∣AN∣2+2∣Z∣2+∣AH−12Id⁡H∣2.(25)|E|^2 = v = |A_{\mathcal{N}}|^2 + 2|Z|^2 + \left|A_{\mathcal{H}} - \frac{1}{2}\operatorname{Id}_{\mathcal{H}}\right|^2. \tag*{(25)}

In particular v≥0v \ge0 without any sign assumption on Ric.

Proof. Since AA is self-adjoint, its block matrix in H⊕N\mathcal{H} \oplus\mathcal{N} is

A=(AHZZ∗AN),E=(AH−12Id⁡HZZ∗AN).A = \begin{pmatrix} A_{\mathcal{H}} & Z \\ Z^{*} & A_{\mathcal{N}} \end{pmatrix},\qquad E = \begin{pmatrix} A_{\mathcal{H}} - \frac{1}{2}\operatorname{Id}_{\mathcal{H}} & Z \\ Z^{*} & A_{\mathcal{N}} \end{pmatrix}.

The two off-diagonal blocks have the same squared norm. Therefore

∣E∣2=∣AH−12Id⁡H∣2+2∣Z∣2+∣AN∣2.|E|^2 = \left|A_{\mathcal{H}} - \frac{1}{2}\operatorname{Id}_{\mathcal{H}}\right|^2 + 2|Z|^2 + |A_{\mathcal{N}}|^2.

On the other hand, ∣ΠH∣2=k|\Pi_{\mathcal{H}}|^2 = k and ⟨A,ΠH⟩=tr⁡HA\langle A,\Pi_{\mathcal{H}}\rangle= \operatorname{tr}_{\mathcal{H}} A. Hence

∣E∣2=∣A∣2−⟨A,ΠH⟩+14∣ΠH∣2|E|^2 = |A|^2 - \langle A,\Pi_{\mathcal{H}}\rangle+ \frac{1}{4}|\Pi_{\mathcal{H}}|^2
=k4−tr⁡HA+k4=tr⁡A−tr⁡HA=tr⁡NA=v.= \frac{k}{4} - \operatorname{tr}_{\mathcal{H}} A + \frac{k}{4} = \operatorname{tr} A - \operatorname{tr}_{\mathcal{H}} A = \operatorname{tr}_{\mathcal{N}} A = v.

Combining the two expressions proves the identity and v≥0v \ge0. □

Remark 4.1. The horizontal and normal distributions need not be invariant under AA; the tensor ZZ records their mixed block. All norms of endomorphisms are Hilbert–Schmidt norms. The musical isomorphisms ♯\sharp and ♭\flat raise and lower an index with the metric in use.

For x∈M−x \in M_{-}, let

Sx,r=exp⁡x(rS(νxM−))S_{x,r}=\exp_x\bigl(rS(\nu_xM_{-})\bigr)

be the actual normal sphere of radius rr, and let

hr=r−2g∣TSx,r.(26)h_r=r^{-2}g\big|_{T S_{x,r}}. \tag*{(26)}

We regard hrh_r as a metric on the fixed unit sphere S(νxM−)S(\nu_xM_{-}) by pullback under θ↦exp⁡x(rθ)\theta\mapsto\exp_x(r\theta). Write g0g_0 for its unit round metric and the normalized fiber measure as

dVhrωp=ρ(x,rθ) dω(θ),(27)\frac{dV_{h_r}}{\omega_p}=\rho(x,r\theta)\,d\omega(\theta), \tag*{(27)}

where dωd\omega is probability measure on the unit round sphere and ωp=Vol⁡(Sp,g0)\omega_p=\operatorname{Vol}(S^p,g_0).

Lemma 4.2 (Angular evolution and exact volume law). Let s=log⁡rs=\log r. Then

∂shr=−4hrAν,∂slog⁡ρ=−2v,ρ(x,0)=1.(28)\partial_s h_r=-4h_rA_\nu,\qquad\partial_s\log\rho=-2v,\qquad\rho(x,0)=1. \tag*{(28)}

Moreover

dVg=ωprp dr dVM− dω.(29)dV_g=\omega_p r^p\,dr\,dV_{M_{-}}\,d\omega. \tag*{(29)}

Proof. Let J1,J2J_1,J_2 be tangent vectors obtained by varying the angular coordinates while keeping xx fixed. They commute with ν\nu, so

∂rg(J1,J2)=g(∇J1ν,J2)+g(J1,∇J2ν)=2g(SJ1,J2).\partial_r g(J_1,J_2)=g(\nabla_{J_1}\nu,J_2)+g(J_1,\nabla_{J_2}\nu)=2g(SJ_1,J_2).

Using (15) and hr=r−2g∣νh_r=r^{-2}g|_\nu, we obtain

r∂rhr=−2r−2g∣ν+r−1∂r(g∣ν)=−2hr+2hr(Id⁡−2Aν)=−4hrAν.r\partial_rh_r=-2r^{-2}g|_\nu+r^{-1}\partial_r(g|_\nu) =-2h_r+2h_r(\operatorname{Id}-2A_\nu)=-4h_rA_\nu.

For a varying metric, ∂slog⁡dVhr=12tr⁡(hr−1∂shr)\partial_s\log dV_{h_r}=\frac{1}{2}\operatorname{tr}(h_r^{-1}\partial_sh_r). Consequently,

∂slog⁡ρ=12tr⁡(hr−1(−4hrAν))=−2tr⁡νA=−2v.\partial_s\log\rho=\frac{1}{2}\operatorname{tr}\bigl(h_r^{-1}(-4h_rA_\nu)\bigr)=-2\operatorname{tr}_\nu A=-2v.

The initial value ρ(x,0)=1\rho(x,0)=1 follows from the Euclidean leading term of the normal exponential map.

For the full Jacobian, write the pulled-back volume density locally as

dVg=J(x,r,θ) dr dVM− dVg0.dV_g=J(x,r,\theta)\,dr\,dV_{M_{-}}\,dV_{g_0}.

The variation of the level-set measure and (14) give

∂rlog⁡J=tr⁡S=pr,∂rlog⁡(Jrp)=0.\partial_r\log J=\operatorname{tr}S=\frac{p}{r},\qquad\partial_r\log\left(\frac{J}{r^p}\right)=0.

Since J/rp→1J/r^p\to1 as r↓0r\downarrow0, we have J=rpJ=r^p. Finally, dVg0=ωp dωdV_{g_0}=\omega_p\,d\omega, which proves (4.7).

The first normal jet

The square identity alone does not determine how rapidly the defect vanishes near M−M_-. In this section we compute its first normal derivative and identify the precise contribution of the second fundamental form. This is the step where the total-geodesy assumption enters the proof.

For ξ∈νxM−\xi\in\nu_xM_-, let Lξ:TxM−→TxM−L_\xi:T_xM_- \to T_xM_- be the shape operator of M−M_-,

⟨LξX,Y⟩=⟨IIM−(X,Y),ξ⟩.(30)\langle L_\xi X,Y\rangle=\langle\mathrm{II}_{M_-}(X,Y),\xi\rangle. \tag*{(30)}

Lemma 5.1 (First normal jet). At every x∈M−x \in M_- and ξ∈νxM−\xi\in\nu_xM_-,

(∇ξA)∣TxM−=12Lξ,(31)\left.(\nabla_\xi A)\right|_{T_xM_-}=\frac{1}{2}L_\xi, \tag*{(31)}

and every first derivative component of AA containing at least two normal tensor slots vanishes. Consequently,

E(x,rθ)=r2(Lθ000)+O(r2).(32)E(x,r\theta)=\frac{r}{2}\begin{pmatrix}L_\theta&0\\0&0\end{pmatrix}+O(r^2). \tag*{(32)}

Proof. At M−M_-, df=0df=0. By (9), ∇Ric⁡\nabla\operatorname{Ric} is fully symmetric there. We compute its components according to the splitting TxM=TxM−⊕νxM−T_xM=T_xM_- \oplus\nu_xM_-.

Let X,Y∈TxM−X,Y \in T_xM_- and ξ∈νxM−\xi\in\nu_xM_-. Since Ric⁡(ξ,Y)=0\operatorname{Ric}(\xi,Y)=0 along M−M_-, differentiation in a tangential direction XX gives

0=(∇XRic⁡)(ξ,Y)+Ric⁡(∇Xξ,Y)+Ric⁡(ξ,∇XY).0=(\nabla_X\operatorname{Ric})(\xi,Y)+\operatorname{Ric}(\nabla_X\xi,Y)+\operatorname{Ric}(\xi,\nabla_XY).

The last term vanishes by (18), and the tangential component of ∇Xξ\nabla_X\xi is −LξX-L_\xi X. Hence

(∇XRic⁡)(ξ,Y)=12⟨LξX,Y⟩.(\nabla_X\operatorname{Ric})(\xi,Y)=\frac{1}{2}\langle L_\xi X,Y\rangle.

Full symmetry of ∇Ric⁡\nabla\operatorname{Ric} moves the normal derivative to the first slot and gives (5.2).

Next, the normal-normal block A∣νM−\left.A\right|_{\nu M_-} vanishes identically along M−M_-. Differentiating this identity tangentially shows that every component of ∇A\nabla A with two normal slots and one tangent slot vanishes. Along a normal geodesic γ\gamma, Aγ˙=0A\dot\gamma=0 because A∇f=0A\nabla f=0 and γ˙\dot\gamma is parallel to ∇f\nabla f off M−M_-. Since γ˙\dot\gamma is parallel, differentiation gives (∇γ˙A)γ˙=0(\nabla_{\dot\gamma}A)\dot\gamma=0. For normal vectors ξ,η\xi,\eta, this says

(∇ξRic⁡)(ξ,η)=0.(\nabla_\xi\operatorname{Ric})(\xi,\eta)=0.

Replacing ξ\xi by ξ+ζ\xi+\zeta and using the full symmetry already proved gives

2(∇ξRic⁡)(ζ,η)=0.2(\nabla_\xi\operatorname{Ric})(\zeta,\eta)=0.

Thus every all-normal component also vanishes.

Finally, use parallel transport along γ(t)=exp⁡x(tθ)\gamma(t)=\exp_x(t\theta) to compare the tensors with their values at xx. A variation of the initial normal velocity in a direction η⊥θ\eta\perp\theta gives a Jacobi field with

Jη(0)=0,Jη′(0)=η,Jη(t)=tη+O(t3).J_\eta(0)=0,\qquad J_\eta'(0)=\eta,\qquad J_\eta(t)=t\eta+O(t^3).

Together with the radial vector, these fields span N\mathcal{N}. Hence, in this parallel identification,

ΠN(t)=ΠνxM−+O(t2),ΠH(t)=ΠTxM−+O(t2).\Pi_{\mathcal{N}}(t)=\Pi_{\nu_xM_-}+O(t^2),\qquad\Pi_{\mathcal{H}}(t)=\Pi_{T_xM_-}+O(t^2).

In particular, the normal derivative of ΠH\Pi_{\mathcal H} at M−M_{-} is zero. Taylor expansion of A−12ΠHA-\frac{1}{2}\Pi_{\mathcal H} now gives (5.3).

Corollary 5.1. If M−M_{-} is totally geodesic, then

E=O(r2)(33)E=O(r^2) \tag*{(33)}

uniformly on the compact unit normal sphere bundle.

Proof. When IIM−=0\mathrm{II}_{M_{-}}=0, every LθL_{\theta} is zero, so (5.3) gives the assertion. Compactness of M−M_{-} makes the Taylor remainder uniform over its unit normal sphere bundle.

A scalar balance in arbitrary codimension

We now relate the defect to the intrinsic scalar curvature of each normal sphere. The calculation has two parts: the soliton equation determines the curvature of the level set, and the Gauss equation compares that curvature with the curvature of the fiber. We keep the resulting integrated identity in a form suitable for comparison with the spherical estimate in the next section.

Assume from now on that M−M_{-} is totally geodesic. Suppose, for contradiction, that EE is not flat to infinite normal order. Let d≥2d\geq2 be the smallest normal degree for which the Taylor coefficient is nonzero:

E(x,rθ)=rdEd(x,θ)+O(rd+1).(34)E(x,r\theta)=r^d E_d(x,\theta)+O(r^{d+1}). \tag*{(34)}

Relative to H⊕V\mathcal H\oplus\mathcal V write

Ed=(HdZdZd∗Nd).(35)E_d= \begin{pmatrix} H_d & Z_d\\ Z_d^* & N_d \end{pmatrix}. \tag*{(35)}

All expansions are taken in smooth local trivializations over M−M_{-} and are uniform, with the derivatives used below, on its compact unit normal sphere bundle. Since all lower normal coefficients vanish, the leading coefficient is unchanged by a smooth change of trivialization that is the identity at M−M_{-}. In the limiting splitting TxM−⊕νxM−T_xM_{-}\oplus\nu_xM_{-}, its components are homogeneous polynomials of degree dd in the normal variable. The identity Eν=0E\nu=0 implies Ndθ=0N_d\theta=0 and Zdθ=0Z_d\theta=0. Consequently the normal block, extended by zero in the radial direction, still has polynomial ambient components of degree dd.

Use the probability measure

dμ0=dVM−Vol⁡(M−) dωd\mu_0=\frac{dV_{M_{-}}}{\operatorname{Vol}(M_{-})}\,d\omega

on S(νM−)S(\nu M_{-}) and set

H=∫∣Hd∣2 dμ0,N=∫∣Nd∣2 dμ0,Z=∫∣Zd∣2 dμ0,(36)H=\int|H_d|^2\,d\mu_0,\qquad N=\int|N_d|^2\,d\mu_0,\qquad Z=\int|Z_d|^2\,d\mu_0, \tag*{(36)}
B=H+N+2Z>0.(37)B=H+N+2Z>0. \tag*{(37)}

Lemma 6.1 (Leading density expansion). One has

v=r2d∣Ed∣2+O(r2d+1),(38)v=r^{2d}|E_d|^2+O(r^{2d+1}), \tag*{(38)}
ρ=1−r2dd∣Ed∣2+O(r2d+1),(39)\rho=1-\frac{r^{2d}}{d}|E_d|^2+O(r^{2d+1}), \tag*{(39)}

and

tr⁡SpNd=0.(40)\operatorname{tr}_{\mathbb{S}_p} N_d = 0. \tag*{(40)}

Proof. The exact square identity and (6.1) give

v=∣rdEd+O(rd+1)∣2=r2d∣Ed∣2+O(r2d+1).v = \left|r^d E_d + O(r^{d+1})\right|^2 = r^{2d}|E_d|^2 + O(r^{2d+1}).

Since ∂rlog⁡ρ=−2v\partial_r \log\rho=-2v and ρ(x,0)=1\rho(x,0)=1, integration along a fixed ray gives

log⁡ρ(x,rθ)=−2∫0rv(x,tθ)t dt=−2∣Ed(x,θ)∣2∫0rt2d−1 dt+O(r2d+1)=−r2dd∣Ed(x,θ)∣2+O(r2d+1).\begin{aligned} \log\rho(x,r\theta)=-2\int_0^r\frac{v(x,t\theta)}{t}\,dt \\ &=-2|E_d(x,\theta)|^2\int_0^r t^{2d-1}\,dt+O(r^{2d+1}) \\ &=-\frac{r^{2d}}{d}|E_d(x,\theta)|^2+O(r^{2d+1}). \end{aligned}

Exponentiating introduces an additional error O(r4d)O(r^{4d}), which is absorbed by O(r2d+1)O(r^{2d+1}). This proves (6.6).

Finally, Aν=EνA_\nu=E_\nu, so

tr⁡νA=rdtr⁡g0Nd+O(rd+1).\operatorname{tr}_\nu A=r^d\operatorname{tr}_{g_0}N_d+O(r^{d+1}).

The left-hand side is v=O(r2d)v=O(r^{2d}). Dividing by rdr^d and letting r↓0r\downarrow0 proves (6.7). □

We use the level-set curvature formula (2.12) to compute the scalar curvature of a normal fiber. The difference between these two curvatures is recorded by the following geometric terms.

Inside Σr\Sigma_r, let BVB_{\mathcal V} and HVH_{\mathcal V} denote the second fundamental form and mean curvature of the integrable vertical distribution V\mathcal V. Let

M=∑a=1p∑α=1ksec⁡Σr(ea,eα)\mathcal{M}=\sum_{a=1}^{p}\sum_{\alpha=1}^{k}\sec_{\Sigma_r}(e_a,e_\alpha)

be the mixed sectional-curvature sum, and define

G=∣HV∣2−∣BV∣2−M.(41)G=|H_{\mathcal V}|^2-|B_{\mathcal V}|^2-\mathcal{M}. \tag*{(41)}

Lemma 6.2 (General-codimension fiber scalar identity). Let RhrR_{h_r} be the scalar curvature of the normalized normal-sphere metric hrh_r. Then

Rhr−p(p−1)=(r2−2(p−1))v−2vs−8∣Z∣2+r2G.(42)R_{h_r}-p(p-1)=\left(r^2-2(p-1)\right)v-2v_s-8|Z|^2+r^2G. \tag*{(42)}

Proof. We use (2.11) to derive the radial form of the level-set Ricci tensor before taking its vertical trace. Radial derivatives of endomorphisms are understood in a parallel frame.

Combine the hypersurface Gauss and Riccati equations:

Ric⁡Σr♯=A−Rm⁡(⋅,ν)ν+(tr⁡S)S−S2,\operatorname{Ric}^{\sharp}_{\Sigma_r}=A-\operatorname{Rm}(\cdot,\nu)\nu+(\operatorname{tr}S)S-S^2,
Rm⁡(⋅,ν)ν=−∇νS−S2.\operatorname{Rm}(\cdot,\nu)\nu=-\nabla_\nu S-S^2.

It follows that Ric⁡Σr♯=A+∇νS+(tr⁡S)S\operatorname{Ric}^{\sharp}_{\Sigma_r}=A+\nabla_\nu S+(\operatorname{tr}S)S. Differentiating S=r−1(Id⁡−2A)S=r^{-1}(\operatorname{Id}-2A) in a radial parallel frame gives

∇νS=−1r2(Id⁡−2A)−2r∇νA,(tr⁡S)S=pr2(Id⁡−2A).\nabla_\nu S=-\frac{1}{r^2}(\operatorname{Id}-2A)-\frac{2}{r}\nabla_\nu A,\qquad (\operatorname{tr}S)S=\frac{p}{r^2}(\operatorname{Id}-2A).

Adding the two expressions proves (16).

The vertical trace requires care because V\mathcal{V} moves with rr. Let ΠV\Pi_{\mathcal{V}} denote its orthogonal projection. A vertical Jacobi field JJ satisfies [ν,J]=0[\nu,J]=0, hence ∇rνJ=rSJ=(Id⁡−2A)J\nabla_{r\nu}J=rSJ=(\operatorname{Id}-2A)J. Its horizontal component is therefore −2ZJ-2ZJ. Differentiating the projection gives, in the order H⊕V\mathcal{H}\oplus\mathcal{V},

∇rνΠV=(0−2Z−2Z∗0).\nabla_{r\nu}\Pi_{\mathcal{V}}= \begin{pmatrix} 0 & -2Z \\ -2Z^{*} & 0 \end{pmatrix}.

The two diagonal blocks of ∇rνΠV\nabla_{r\nu}\Pi_{\mathcal{V}} vanish by differentiating ΠV2=ΠV\Pi_{\mathcal{V}}^{2}=\Pi_{\mathcal{V}}; the lower-left block is the adjoint of the upper-right block. Consequently,

tr⁡(A∇rνΠV)=−2tr⁡(ZZ∗)−2tr⁡(Z∗Z)=−4∣Z∣2.\operatorname{tr}(A\nabla_{r\nu}\Pi_{\mathcal{V}})=-2\operatorname{tr}(ZZ^{*})-2\operatorname{tr}(Z^{*}Z)=-4|Z|^{2}.

Taking the derivative of v=tr⁡(AΠV)v=\operatorname{tr}(A\Pi_{\mathcal{V}}) therefore yields

vs=tr⁡V(∇rνA)−4∣Z∣2(43)v_{s}=\operatorname{tr}_{\mathcal{V}}(\nabla_{r\nu}A)-4|Z|^{2} \tag*{(43)}

Since tr⁡V(Id⁡−2A)=p−2v\operatorname{tr}_{\mathcal{V}}(\operatorname{Id}-2A)=p-2v, the vertical trace of (16) becomes

r2tr⁡VRic⁡Σr♯=r2v−2vs−8∣Z∣2+(p−1)(p−2v)=p(p−1)+(r2−2(p−1))v−2vs−8∣Z∣2.(44)\begin{aligned} r^{2}\operatorname{tr}_{\mathcal{V}}\operatorname{Ric}^{\sharp}_{\Sigma_{r}} =r^{2}v-2v_{s}-8|Z|^{2}+(p-1)(p-2v) \\ &=p(p-1)+(r^{2}-2(p-1))v-2v_{s}-8|Z|^{2}. \tag*{(44)} \end{aligned}

Finally, choose orthonormal frames eae_{a} of V\mathcal{V} and eαe_{\alpha} of H\mathcal{H}. The vertical trace of the Ricci tensor is

tr⁡VRic⁡Σr♯=∑a,b⟨Rm⁡Σr(ea,eb)eb,ea⟩+M.\operatorname{tr}_{\mathcal{V}}\operatorname{Ric}^{\sharp}_{\Sigma_{r}} =\sum_{a,b}\langle\operatorname{Rm}_{\Sigma_{r}}(e_{a},e_{b})e_{b},e_{a}\rangle+\mathcal{M}.

The Gauss equation for Sx,r⊂ΣrS_{x,r}\subset\Sigma_{r} gives

RSx,r=∑a,b⟨Rm⁡Σr(ea,eb)eb,ea⟩+∣HV∣2−∣BV∣2.R_{S_{x,r}}=\sum_{a,b}\langle\operatorname{Rm}_{\Sigma_{r}}(e_{a},e_{b})e_{b},e_{a}\rangle+|H_{\mathcal{V}}|^{2}-|B_{\mathcal{V}}|^{2}.

Since g∣Sx,r=r2hrg|_{S_{x,r}}=r^{2}h_{r}, scalar curvature rescales by r−2r^{-2}. Thus

r−2Rhr=RSx,r=tr⁡VRic⁡Σr♯+G.r^{-2}R_{h_{r}}=R_{S_{x,r}}=\operatorname{tr}_{\mathcal{V}}\operatorname{Ric}^{\sharp}_{\Sigma_{r}}+G.

Multiplying by r2r^{2} and using (44) yields (42).

To integrate this identity, it is enough to introduce two scalar quantities:

S(r)=∫(Rhr−p(p−1))ρ dμ0,Γ(r)=∫Gρ dμ0.(45)\mathcal{S}(r)=\int(R_{h_{r}}-p(p-1))\rho\,d\mu_{0},\qquad\Gamma(r)=\int G\rho\,d\mu_{0}. \tag*{(45)}

Since ρs=−2vρ\rho_{s}=-2v\rho, differentiation under the integral gives

dds∫vρ dμ0=∫(vs−2v2)ρ dμ0.\frac{d}{ds}\int v\rho\,d\mu_{0}=\int(v_{s}-2v^{2})\rho\,d\mu_{0}.

In particular,

−2∫vsρ dμ0=−2dds∫vρ dμ0−4∫v2ρ dμ0.-2\int v_{s}\rho\,d\mu_{0}=-2\frac{d}{ds}\int v\rho\,d\mu_{0}-4\int v^{2}\rho\,d\mu_{0}.

Substituting this into the integral of (6.9) yields the exact balance

S(r)=(r2−2(p−1))∫vρ dμ0−2dds∫vρ dμ0−4∫v2ρ dμ0−8∫∣Z∣2ρ dμ0+r2Γ(r).(46)\mathcal{S}(r) = \left(r^2 - 2(p-1)\right)\int v\rho\,d\mu_0 - 2\frac{d}{ds}\int v\rho\,d\mu_0 - 4\int v^2\rho\,d\mu_0 - 8\int|Z|^2\rho\,d\mu_0 + r^2\Gamma(r). \tag*{(46)}

The weight ρ\rho is the actual fiber-volume density. It is different from the unweighted measure in the full level-set volume law, a distinction that will be essential in Section 8.

A sharp spherical quadratic estimate

The preceding balance contains the scalar curvature of a metric close to the round sphere. We now estimate its first nonzero variation. The estimate depends only on the degree of the polynomial leading term; no decomposition into tensor spherical harmonics is needed.

For a metric hh on Sp\mathbb{S}^p, let

V(h)=Vol⁡(Sp,h)ωp,I(h)=1ωp∫SpRh dVh.V(h) = \frac{\operatorname{Vol}(\mathbb{S}^p,h)}{\omega_p}, \qquad I(h) = \frac{1}{\omega_p}\int_{\mathbb{S}^p} R_h\,dV_h.

For p≥2p \ge2, define

Yp(h)=V(h)−(p−2)/pI(h).(47)Y_p(h) = V(h)^{-(p-2)/p}I(h). \tag*{(47)}

Then Yp(g0)=p(p−1)Y_p(g_0)=p(p-1). For p=1p=1, set Y1≡0Y_1\equiv0.

Lemma 7.1 (Spherical Einstein–Hilbert estimate). Let KK be a smooth trace-free symmetric two-tensor tangent to the unit sphere Sp\mathbb{S}^p. Assume that, after extending KK by zero on the radial direction, every ambient Cartesian component is the restriction of a polynomial of degree at most dd. If

hε=g0+εK+O(ε2)h_\varepsilon= g_0 + \varepsilon K + O(\varepsilon^2)

and

Yp(hε)=p(p−1)+ε2y(K)+O(ε3),Y_p(h_\varepsilon) = p(p-1) + \varepsilon^2 y(K) + O(\varepsilon^3),

then

y(K)≥−14d(d+p−1)∣K∣L2(Sp)2.(48)y(K) \ge-\frac{1}{4}d(d+p-1)|K|_{L^2(\mathbb{S}^p)}^2. \tag*{(48)}

The L2L^2 norm is normalized by ωp−1\omega_p^{-1}.

Proof. For p=1p=1 the assertion is the definition, and for p=2p=2 the Einstein–Hilbert functional is topological, so y(K)=0y(K)=0. Assume p≥3p\ge3.

We give the second variation directly. Put (div⁡K)j=∇iKij(\operatorname{div}K)_j=\nabla^iK_{ij} and let ∇∗∇\nabla^*\nabla be the positive rough Laplacian. The first variation of the volume-normalized total scalar curvature vanishes at the round Einstein metric. Therefore its second variation depends only on KK, not on the second derivative of the chosen path.

The volume and total-scalar-curvature variations are

DV∣h(K)=12ωp∫tr⁡hK dVh,DI∣h(K)=−1ωp∫⟨Ric⁡h−12Rhh,K⟩h dVh.DV|_h(K) = \frac{1}{2\omega_p}\int\operatorname{tr}_h K\,dV_h, \qquad DI|_h(K) = -\frac{1}{\omega_p}\int\left\langle\operatorname{Ric}_h-\frac{1}{2}R_hh,K\right\rangle_h\,dV_h.

In the second formula the divergence terms have zero integral on the closed sphere. Differentiating Yp=V−(p−2)/pIY_p=V^{-(p-2)/p}I therefore gives

DYp∣h(K)=−V(h)−(p−2)/p1ωp∫⟨Ric⁡h−12Rhh+p−22pI(h)V(h)h,K⟩h dVh.DY_p|_h(K)=-V(h)^{-(p-2)/p}\frac{1}{\omega_p}\int\left\langle\operatorname{Ric}_h-\frac{1}{2}R_hh+\frac{p-2}{2p}\frac{I(h)}{V(h)}h,K\right\rangle_h\,dV_h.

At g0g_0, the tensor in brackets is zero. Since tr⁡g0K=0\operatorname{tr}_{g_0}K=0, differentiating this formula gives

D2Yp∣g0(K,K)=−∫⟨Ric⁡′(K)−(p−1)K,K⟩ dω.D^2Y_p|_{g_0}(K,K)=-\int\left\langle\operatorname{Ric}'(K)-(p-1)K,K\right\rangle\,d\omega.

Terms proportional to g0g_0 have zero pairing with KK. To compute Ric⁡′(K)\operatorname{Ric}'(K), the variation of the connection in local coordinates is

(Γijℓ)′=12gℓa(∇iKja+∇jKia−∇aKij).(\Gamma^\ell_{ij})'=\frac{1}{2}g^{\ell a}\left(\nabla_iK_{ja}+\nabla_jK_{ia}-\nabla_aK_{ij}\right).

Differentiating curvature and contracting gives

Ric⁡′(K)ij=∇ℓ(Γijℓ)′−∇j(Γiℓℓ)′.\operatorname{Ric}'(K)_{ij}=\nabla_\ell(\Gamma^\ell_{ij})'-\nabla_j(\Gamma^\ell_{i\ell})'.

Substitution of the connection variation, commutation of covariant derivatives, and tr⁡K=0\operatorname{tr}K=0 give the standard formula [1]

Ric⁡′(K)=12(ΔLK+L(div⁡K)♯g0).\operatorname{Ric}'(K)=\frac{1}{2}\left(\Delta_LK+\mathcal{L}_{(\operatorname{div}K)^\sharp}g_0\right).

where

ΔLK=∇∗∇K+Ric⁡∘K+K∘Ric⁡−2R˚K.\Delta_LK=\nabla^*\nabla K+\operatorname{Ric}\circ K+K\circ\operatorname{Ric}-2\mathring{R}K.

On the unit sphere R˚K=(tr⁡K)g0−K=−K\mathring{R}K=(\operatorname{tr}K)g_0-K=-K, so

ΔLK−2(p−1)K=∇∗∇K+2K.(49)\Delta_LK-2(p-1)K=\nabla^*\nabla K+2K. \tag*{(49)}

Integration by parts gives

∫⟨L(div⁡K)♯g0,K⟩ dω=−2∫∣div⁡K∣2 dω.\int\left\langle\mathcal{L}_{(\operatorname{div}K)^\sharp}g_0,K\right\rangle\,d\omega=-2\int|\operatorname{div}K|^2\,d\omega.

As y(K)=12D2Yp∣g0(K,K)y(K)=\frac{1}{2}D^2Y_p|_{g_0}(K,K), we conclude that

y(K)=−14∫(∣∇K∣2+2∣K∣2) dω+12∫∣div⁡K∣2 dω≥−14∫(∣∇K∣2+2∣K∣2) dω.(50)\begin{aligned} y(K)=-\frac{1}{4}\int\left(|\nabla K|^2+2|K|^2\right)\,d\omega+\frac{1}{2}\int|\operatorname{div}K|^2\,d\omega\\ &\geq-\frac{1}{4}\int\left(|\nabla K|^2+2|K|^2\right)\,d\omega. \tag*{(50)} \end{aligned}

To compare the right-hand side with scalar spherical harmonics, regard KK as an ambient symmetric bilinear form with K(θ,⋅)=0K(\theta,\cdot)=0. Let DD be the Euclidean connection and e1,…,epe_1,\ldots,e_p an orthonormal tangent frame. The Gauss formula gives

∑a=1p∣DeaK∣2=∣∇K∣2+2∣K∣2.(51)\sum_{a=1}^{p}|D_{e_a}K|^2=|\nabla K|^2+2|K|^2. \tag*{(51)}

Indeed, the tangent-tangent part of DKDK is ∇K\nabla K, while the two terms in which one tensor slot differentiates into the radial direction are both −K(ea,⋅)-K(e_a,\cdot), producing the factor 2∣K∣22|K|^2.

Choose a fixed ambient orthonormal basis EiE_i and write Kij=K(Ei,Ej)K_{ij}=K(E_i,E_j). Decompose each component into scalar spherical harmonics:

Kij=∑ℓ=0d(Kij)ℓ,−Δg0(Kij)ℓ=ℓ(ℓ+p−1)(Kij)ℓ.K_{ij}=\sum_{\ell=0}^{d}(K_{ij})_\ell,\qquad-\Delta_{g_0}(K_{ij})_\ell=\ell(\ell+p-1)(K_{ij})_\ell.

The summands are mutually orthogonal in L2L^2. Integration by parts therefore yields

∫Sp∣∇Kij∣2 dω=∑ℓ=0dℓ(ℓ+p−1)∫Sp∣(Kij)ℓ∣2 dω≤d(d+p−1)∑ℓ=0d∫Sp∣(Kij)ℓ∣2 dω=d(d+p−1)∫SpKij2 dω.\begin{aligned} \int_{\mathbb{S}^p}|\nabla K_{ij}|^2\,d\omega=\sum_{\ell=0}^{d}\ell(\ell+p-1)\int_{\mathbb{S}^p}|(K_{ij})_\ell|^2\,d\omega\\ &\le d(d+p-1)\sum_{\ell=0}^{d}\int_{\mathbb{S}^p}|(K_{ij})_\ell|^2\,d\omega\\ &=d(d+p-1)\int_{\mathbb{S}^p}K_{ij}^2\,d\omega. \end{aligned}

Summing over i,ji,j gives

∑i,j∫Sp∣∇Kij∣2 dω≤d(d+p−1)∑i,j∫SpKij2 dω.\sum_{i,j}\int_{\mathbb{S}^p}|\nabla K_{ij}|^2\,d\omega\le d(d+p-1)\sum_{i,j}\int_{\mathbb{S}^p}K_{ij}^2\,d\omega.

The left side is exactly the integral of ∑a∣DeaK∣2\sum_a|D_{e_a}K|^2, while the right side is d(d+p−1)∣K∣L22d(d+p-1)|K|_{L^2}^2. Combining (50)–(51) proves (48). □

Exclusion of every higher normal jet

We are now ready to exclude a nonzero leading coefficient of degree d≥2d\ge2. First we use the spherical estimate to bound the scalar term in (46). We then obtain the opposite bound for Γ(r)\Gamma(r) from a weighted curvature identity on the level set. The final comparison is strictly positive in every normal dimension.

Proposition 8.1 (Higher-jet exclusion). If the first normal jet of EE vanishes along M−M_-, then EE has no nonzero normal Taylor coefficient of any finite order d≥2d\ge2.

By (34), AV=rdNd+O(rd+1)A_V=r^dN_d+O(r^{d+1}). Using hr→g0h_r\to g_0 in the angular evolution equation, we first obtain hr−g0=O(rd)h_r-g_0=O(r^d). Substituting this back gives

∂rhr=−4rd−1Nd♭+O(rd).\partial_r h_r=-4r^{d-1}N_d^\flat+O(r^d).

Integration from 00 to rr yields

hr=g0+rdKd+O(rd+1),Kd=−4dNd♭.(52)h_r=g_0+r^dK_d+O(r^{d+1}),\qquad K_d=-\frac{4}{d}N_d^\flat. \tag*{(52)}

By (40), KdK_d is trace-free. The round metric is a critical point of YpY_p, so the terms of orders d+1,…,2d−1d+1,\ldots,2d-1 in hrh_r contribute no linear term to Yp(hr)Y_p(h_r). Its coefficient of r2dr^{2d} is consequently exactly the quadratic variation in KdK_d. The ambient polynomial property established after (35) applies to KdK_d. Since

∫∣Kd∣2 dμ0=16d2N,\int|K_d|^2\,d\mu_0=\frac{16}{d^2}\mathbb{N},

Lemma 7.1, applied fiberwise and averaged over M−M_{-}, gives

yd≥−4(d+p−1)dN.(53)y_d \geq-\frac{4(d+p-1)}{d}\mathbb{N}. \tag*{(53)}

where ydy_d is the averaged coefficient of r2dr^{2d} in Yp(hr)Y_p(h_r).

Lemma 8.1 (Expansion of the fiber scalar defect). One has

S(r)=r2d(2(p−1)dB+yd)+O(r2d+1).(54)S(r)=r^{2d}\left(\frac{2(p-1)}{d}\mathrm{B}+y_d\right)+O(r^{2d+1}). \tag*{(54)}

Proof. For p=1p=1, both the intrinsic scalar curvature of the circle and Y1Y_1 vanish, so the assertion is immediate. Suppose p≥2p\geq2. For fixed x∈M−x\in M_{-}, let

Vx(r)=∫Spρ(x,rθ) dω.V_x(r)=\int_{S_p}\rho(x,r\theta)\,d\omega.

By (39),

Vx(r)=1−axr2d+O(r2d+1),ax=1d∫Sp∣Ed(x,θ)∣2 dω.V_x(r)=1-a_xr^{2d}+O(r^{2d+1}),\qquad a_x=\frac{1}{d}\int_{S_p}|E_d(x,\theta)|^2\,d\omega.

Write Yp(hr)=p(p−1)+yxr2d+O(r2d+1)Y_p(h_r)=p(p-1)+y_xr^{2d}+O(r^{2d+1}). Since

∫SpRhrρ dω=Vx(r)(p−2)/pYp(hr),\int_{S_p}R_{h_r}\rho\,d\omega=V_x(r)^{(p-2)/p}Y_p(h_r),

expand Vx(p−2)/p=1−p−2paxr2d+O(r2d+1)V_x^{(p-2)/p}=1-\frac{p-2}{p}a_xr^{2d}+O(r^{2d+1}) and subtract p(p−1)Vx(r)p(p-1)V_x(r). The constant terms cancel, and the coefficient of r2dr^{2d} is

yx−p−2pp(p−1)ax+p(p−1)ax=yx+(p−(p−2))(p−1)ax=yx+2(p−1)ax.\begin{aligned} y_x-\frac{p-2}{p}p(p-1)a_x+p(p-1)a_x =y_x+\bigl(p-(p-2)\bigr)(p-1)a_x \\ &=y_x+2(p-1)a_x. \end{aligned}

Averaging over x∈M−x\in M_{-} gives (54). □

The density expansion also gives

∫vρ dμ0=r2dB+O(r2d+1),dds∫vρ dμ0=2dr2dB+O(r2d+1),\int v\rho\,d\mu_0=r^{2d}\mathrm{B}+O(r^{2d+1}),\qquad \frac{d}{ds}\int v\rho\,d\mu_0=2dr^{2d}\mathrm{B}+O(r^{2d+1}),
∫∣Z∣2ρ dμ0=r2dZ+O(r2d+1),∫v2ρ dμ0=O(r4d).(55)\int|Z|^2\rho\,d\mu_0=r^{2d}\mathrm{Z}+O(r^{2d+1}),\qquad \int v^2\rho\,d\mu_0=O(r^{4d}). \tag*{(55)}

Substitute these expansions and (8.3) into (46). After division by r2dr^{2d}, the terms r2∫vρ dμ0r^2\int v\rho\,d\mu_0 and ∫v2ρ dμ0\int v^2\rho\,d\mu_0 tend to zero. The remaining terms give

r−(2d−2)Γ(r)=(4d+2(p−1)+2(p−1)d)B+8Z+yd+O(r).r^{-(2d-2)}\Gamma(r)=\left(4d+2(p-1)+\frac{2(p-1)}{d}\right)\mathrm{B}+8\mathrm{Z}+y_d+O(r).

Using (53), we obtain

lim inf⁡r↓0r−(2d−2)Γ(r)≥(4d+2(p−1)+2(p−1)d)B+8Z−4(d+p−1)dN.(56)\liminf_{r\downarrow0}r^{-(2d-2)}\Gamma(r)\geq \left(4d+2(p-1)+\frac{2(p-1)}{d}\right)\mathrm{B} +8\mathrm{Z}-\frac{4(d+p-1)}{d}\mathbb{N}. \tag*{(56)}

We now derive the opposite estimate. Pull the horizontal metric back to TxM−T_xM_- by horizontal lifting and denote it by qrq_r. The tangential part of the soliton equation and f=r2/4f=r^2/4 give

r∂rqr=−4EH♯.(57)r\partial_r q_r=-4E_H^{\sharp}. \tag*{(57)}

Indeed, choose horizontal lifts of fixed vector fields on M−M_-. Their brackets with ν\nu are vertical. The resulting correction terms in the radial derivative of their metric pairing vanish by orthogonality. Thus

r∂rqr(X,Y)=2g(X,Y)−4Ric⁡(X,Y)=−4g(EHX,Y).r\partial_rq_r(X,Y)=2g(X,Y)-4\operatorname{Ric}(X,Y)=-4g(E_HX,Y).

The cancellation follows directly from AH=12Id⁡H+EHA_H=\frac{1}{2}\operatorname{Id}_H+E_H. Integrating (57) and using (6.1)–(6.2) gives

qr=gM−+rdQd+O(rd+1),Qd=−4dHd♭.(58)q_r=g_{M_-}+r^dQ_d+O(r^{d+1}),\qquad Q_d=-\frac{4}{d}H_d^{\flat}. \tag*{(58)}

To estimate Γ(r)\Gamma(r), we need an integrated formula in which the horizontal mean-curvature term cancels. We give the pointwise computation and the weighted integration together, so that all sign and density conventions are explicit.

Lemma 8.2 (Weighted Walczak identity). For fixed r>0r>0, let TΣr=H⊕VT\Sigma_r=\mathcal{H}\oplus\mathcal{V} be the orthogonal splitting induced by the normal projection. Let BHB_H, THT_H, HHH_H be the second fundamental form, integrability tensor and mean curvature of H\mathcal{H}, and let BVB_V, HVH_V be the corresponding objects for the integrable vertical distribution V\mathcal{V}. Then

HH=∇Vlog⁡ρ,(59)H_H=\nabla_V\log\rho, \tag*{(59)}

and

Γ(r)=∫ρ(∣BH∣2−∣TH∣2+⟨∇Hlog⁡ρ,HV⟩) dμ0.(60)\Gamma(r)=\int\rho\left(\lvert B_H\rvert^2-\lvert T_H\rvert^2+\langle\nabla_H\log\rho,H_V\rangle\right)\,d\mu_0. \tag*{(60)}

Proof. In this proof, ∇\nabla and div⁡\operatorname{div} are the connection and divergence of the induced metric on Σr\Sigma_r. For X,Y∈HX,Y\in\mathcal{H}, we use the conventions

BH(X,Y)=12((∇XY)V+(∇YX)V),TH(X,Y)=12[X,Y]V,B_H(X,Y)=\frac{1}{2}\left((\nabla_XY)^V+(\nabla_YX)^V\right),\qquad T_H(X,Y)=\frac{1}{2}[X,Y]^V,

and HH=tr⁡HBHH_H=\operatorname{tr}_{\mathcal{H}}B_H. The analogous tensors for V\mathcal{V} are defined by interchanging H\mathcal{H} and V\mathcal{V}. Since V\mathcal{V} is tangent to the actual fibers Sx,rS_{x,r}, it is integrable and hence TV=0T_V=0.

The pointwise Walczak identity in these conventions is

div⁡Σr(HH+HV)=M+∣BH∣2+∣BV∣2−∣HH∣2−∣HV∣2−∣TH∣2.(61)\operatorname{div}_{\Sigma_r}(H_H+H_V)=\mathcal{M}+\lvert B_H\rvert^2+\lvert B_V\rvert^2-\lvert H_H\rvert^2-\lvert H_V\rvert^2-\lvert T_H\rvert^2. \tag*{(61)}

Equivalently, using G=∣HV∣2−∣BV∣2−MG=\lvert H_V\rvert^2-\lvert B_V\rvert^2-\mathcal{M},

G=∣BH∣2−∣TH∣2−∣HH∣2−div⁡Σr(HH+HV).(62)G=\lvert B_H\rvert^2-\lvert T_H\rvert^2-\lvert H_H\rvert^2-\operatorname{div}_{\Sigma_r}(H_H+H_V). \tag*{(62)}

Here is a component verification of (61). Use orthonormal frames eα∈He_\alpha\in\mathcal{H} and ea∈Ve_a\in\mathcal{V} whose projected connection forms vanish at the point under consideration. For this computation only, write

Cαβa=⟨∇eαeβ,ea⟩,Dabα=⟨∇eaeb,eα⟩.C_{\alpha\beta a}=\langle\nabla_{e_\alpha}e_\beta,e_a\rangle,\qquad D_{ab\alpha}=\langle\nabla_{e_a}e_b,e_\alpha\rangle.

The symmetric and skew parts of C\mathcal{C} are BHB_H and THT_H. Similarly, D\mathcal{D} represents BVB_V and is symmetric because V\mathcal{V} is integrable. Metric compatibility gives

[ea,eα]=−∑bDabαeb+∑βCαβaeβ[e_a,e_\alpha] = -\sum_b D_{ab\alpha}e_b+\sum_\beta C_{\alpha\beta a}e_\beta

at that point. With the convention Rm⁡(X,Y)Z=∇X∇YZ−∇Y∇XZ−∇[X,Y]Z\operatorname{Rm}(X,Y)Z=\nabla_X\nabla_Y Z-\nabla_Y\nabla_X Z-\nabla_{[X,Y]}Z, expansion of the mixed curvature yields

M=∑a,α(eaCααa+eαDaaα)−∑a,α,βCαβaCβαa−∑a,b,αDabαDbaα.\mathcal{M}=\sum_{a,\alpha}\left(e_aC_{\alpha\alpha a}+e_\alpha D_{aa\alpha}\right)-\sum_{a,\alpha,\beta}C_{\alpha\beta a}C_{\beta\alpha a}-\sum_{a,b,\alpha}D_{ab\alpha}D_{ba\alpha}.

The derivative sum equals

div⁡(HH+HV)+∣HH∣2+∣HV∣2.\operatorname{div}(H_H+H_V)+\lvert H_H\rvert^2+\lvert H_V\rvert^2.

For example, the horizontal and vertical contributions to the divergence of HHH_H are, respectively,

∑α⟨∇eαHH,eα⟩=−∣HH∣2,∑a,αeaCααa.\sum_\alpha\langle\nabla_{e_\alpha}H_H,e_\alpha\rangle=-\lvert H_H\rvert^2,\qquad\sum_{a,\alpha}e_aC_{\alpha\alpha a}.

Finally,

∑a,α,βCαβaCβαa=∣BH∣2−∣TH∣2,∑a,b,αDabαDbaα=∣BV∣2.\sum_{a,\alpha,\beta}C_{\alpha\beta a}C_{\beta\alpha a}=\lvert B_H\rvert^2-\lvert T_H\rvert^2,\qquad\sum_{a,b,\alpha}D_{ab\alpha}D_{ba\alpha}=\lvert B_V\rvert^2.

Substitution proves (61), with the same conventions as the integrated formula of Walczak [19].

We now identify the weight. By the exact level-set volume law (29), after pulling back by Fr(x,θ)=exp⁡x(rθ)F_r(x,\theta)=\exp_x(r\theta),

(8.12)

Fr∗dVΣr=ωprp dVM− dω.F_r^*dV_{\Sigma_r}=\omega_p r^p\,dV_{M_-}\,d\omega.

On the other hand, the actual vertical metric is r2hrr^2h_r, so (27) gives

(8.13)

dVV=ωprpρ dω.dV_{\mathcal{V}}=\omega_p r^p\rho\,d\omega.

Let qrq_r be the horizontal metric pulled back to TxM−T_xM_-, and write

dVqr=JH dVM−.dV_{q_r}=J_H\,dV_{M_-}.

Orthogonality of H\mathcal{H} and V\mathcal{V} implies that the total volume density is the product of the horizontal and vertical densities. Comparing (8.12) and (8.13) gives the exact identity

(8.14)

JHρ=1.J_H\rho=1.

Let U∈VU\in\mathcal{V}. Choose local vector fields E‾1,…,E‾k\overline{E}_1,\ldots,\overline{E}_k on M−M_- and let EαE_\alpha be their projectable horizontal lifts, orthonormal at the point under consideration. Differentiating the horizontal volume form gives

U(log⁡JH)=∑α⟨∇UEα,Eα⟩.U(\log J_H)=\sum_\alpha\langle\nabla_U E_\alpha,E_\alpha\rangle.

Because EαE_\alpha is projectable and UU is vertical, [U,Eα][U,E_\alpha] is vertical. Torsion-freeness and orthogonality therefore give

⟨∇UEα,Eα⟩=⟨∇EαU,Eα⟩=−⟨U,∇EαEα⟩.\langle\nabla_U E_\alpha,E_\alpha\rangle=\langle\nabla_{E_\alpha}U,E_\alpha\rangle=-\langle U,\nabla_{E_\alpha}E_\alpha\rangle.

Summing yields

U(log⁡JH)=−⟨HH,U⟩.(63)U(\log J_{\mathcal H})=-\langle H_{\mathcal H},U\rangle. \tag*{(63)}

Using log⁡JH=−log⁡ρ\log J_{\mathcal H}=-\log\rho from (8.14), we obtain

⟨HH,U⟩=U(log⁡ρ)\langle H_{\mathcal H},U\rangle=U(\log\rho)

for every vertical UU, which is (59).

Finally, at fixed rr, (8.12) shows that dVΣrdV_{\Sigma_r} is a constant multiple of dμ0d\mu_0. For every tangent vector field XX on the compact boundaryless level set Σr\Sigma_r, the divergence theorem gives

0=∫div⁡(ρX) dμ0=∫(ρdiv⁡X+⟨∇ρ,X⟩) dμ0.0=\int\operatorname{div}(\rho X)\,d\mu_0=\int\left(\rho\operatorname{div}X+\langle\nabla\rho,X\rangle\right)\,d\mu_0.

Using ∇ρ=ρ∇log⁡ρ\nabla\rho=\rho\nabla\log\rho, we obtain

∫ρdiv⁡X dμ0=−∫ρ⟨∇log⁡ρ,X⟩ dμ0.(64)\int\rho\operatorname{div}X\,d\mu_0=-\int\rho\langle\nabla\log\rho,X\rangle\,d\mu_0. \tag*{(64)}

Multiply (62) by ρ\rho and integrate. Taking X=HH+HVX=H_{\mathcal H}+H_{\mathcal V} in (64), and using HH∈VH_{\mathcal H}\in\mathcal V, HV∈HH_{\mathcal V}\in\mathcal H, together with (59), gives

⟨∇log⁡ρ,HH⟩=∣HH∣2,⟨∇log⁡ρ,HV⟩=⟨∇Hlog⁡ρ,HV⟩.\langle\nabla\log\rho,H_{\mathcal H}\rangle=|H_{\mathcal H}|^2,\qquad \langle\nabla\log\rho,H_{\mathcal V}\rangle=\langle\nabla_{\mathcal H}\log\rho,H_{\mathcal V}\rangle.

The first term cancels the −∣HH∣2-|H_{\mathcal H}|^2 in (62), leaving precisely (60). □

Proof of Proposition 8.1. We next expand the terms in (60). In a projectable horizontal frame X,YX,Y and for U∈VU\in\mathcal V, Koszul’s formula gives

2⟨BH(X,Y),U⟩=−U(qr(X,Y)).2\langle B_{\mathcal H}(X,Y),U\rangle=-U\bigl(q_r(X,Y)\bigr).

Indeed the bracket term ⟨[X,Y],U⟩\langle[X,Y],U\rangle is skew in X,YX,Y and therefore disappears from the symmetric part, while [U,X][U,X] and [U,Y][U,Y] are vertical. Let eae_a be a round unit angular vector and choose a gM−g_{M_-}-orthonormal basis XαX_\alpha at the base point. The corresponding unit vertical vector is r−1(ea+O(rd))r^{-1}(e_a+O(r^d)), because hr=g0+O(rd)h_r=g_0+O(r^d). Since gM−g_{M_-} is independent of θ\theta, (58) gives

⟨BH(Xα,Xβ),r−1(ea+O(rd))⟩=−12rd−1∇ea(Qd)αβ+O(rd).\left\langle B_{\mathcal H}(X_\alpha,X_\beta),r^{-1}(e_a+O(r^d))\right\rangle=-\frac{1}{2}r^{d-1}\nabla_{e_a}(Q_d)_{\alpha\beta}+O(r^d).

The horizontal frame differs from a qrq_r-orthonormal frame by O(rd)O(r^d). Squaring and summing over a,α,βa,\alpha,\beta therefore yields

∣BH∣2=14r2d−2∣∇SpQd∣2+O(r2d−1).(65)|B_{\mathcal H}|^2=\frac{1}{4}r^{2d-2}|\nabla_{S^{p}}Q_d|^2+O(r^{2d-1}). \tag*{(65)}

We also need to estimate the remaining weighted term. In smooth normal bundle coordinates, a horizontal lift has bounded coefficients in the base and angular variables. Differentiating log⁡ρ=O(r2d)\log\rho=O(r^{2d}) horizontally therefore gives ∇Hlog⁡ρ=O(r2d)\nabla_{\mathcal H}\log\rho=O(r^{2d}).

To estimate HVH_{\mathcal V}, fix x∈M−x\in M_- and consider the normal slice exp⁡x(νxM−)\exp_x(\nu_xM_-). Every radial line in this slice is an ambient geodesic. At its center, the second fundamental form of the slice consequently vanishes on every pair (ξ,ξ)(\xi,\xi); polarization shows that it is zero. Smoothness then gives an O(r)O(r) bound near the center. The radial component of the second fundamental form of the small sphere is tangent to the slice, so its projection to H\mathcal{H} is zero. Thus the horizontal mean curvature of the sphere satisfies HV=O(r)H_V=O(r), and

⟨∇Hlog⁡ρ,HV⟩=O(r2d+1).\langle\nabla_{\mathcal{H}}\log\rho,H_V\rangle=O(r^{2d+1}).

After division by r2d−2r^{2d-2}, this term is O(r3)O(r^3) and tends to zero. Discarding the nonpositive term −∣TH∣2-\lvert T_{\mathcal{H}}\rvert^2 in (8.9) and using (8.17) gives

lim sup⁡r↓0r−(2d−2)Γ(r)≤14∫∣∇SpQd∣2 dμ0.(66)\limsup_{r\downarrow0}r^{-(2d-2)}\Gamma(r)\leq\frac{1}{4}\int\lvert\nabla_{S^p}Q_d\rvert^2\,d\mu_0. \tag*{(66)}

Since the components of QdQ_d in a fixed basis of TxM−T_xM_- are spherical polynomials of degree at most dd, the scalar harmonic estimate gives

∫∣∇SpQd∣2 dμ0≤d(d+p−1)∫∣Qd∣2 dμ0.\int\lvert\nabla_{S^p}Q_d\rvert^2\,d\mu_0\leq d(d+p-1)\int\lvert Q_d\rvert^2\,d\mu_0.

Using Qd=−4Hd♭/dQ_d=-4H_d^\flat/d gives

lim sup⁡r↓0r−(2d−2)Γ(r)≤4(d+p−1)dH.(67)\limsup_{r\downarrow0}r^{-(2d-2)}\Gamma(r)\leq\frac{4(d+p-1)}{d}H. \tag*{(67)}

Combining (8.5) and (8.19), and using B=H+N+2ZB=H+N+2Z, gives

cd,p(H+N)+[8d+8+4(p−1)(1+1d)]Z≤0,(68)c_{d,p}(H+N)+\left[8d+8+4(p-1)\left(1+\frac{1}{d}\right)\right]Z\leq0, \tag*{(68)}

where

cd,p=4(d−1)+2(p−1)(1−1d).(69)c_{d,p}=4(d-1)+2(p-1)\left(1-\frac{1}{d}\right). \tag*{(69)}

Indeed, the coefficient of H+NH+N is

4d+2(p−1)+2(p−1)d−4(d+p−1)d=cd,p,4d+2(p-1)+\frac{2(p-1)}{d}-\frac{4(d+p-1)}{d}=c_{d,p},

and the coefficient of ZZ is 8d+8+4(p−1)(1+1/d)8d+8+4(p-1)(1+1/d). Both are strictly positive for d≥2d\geq2 and p≥1p\geq1. Since the three energies are nonnegative, (8.20) forces H=N=Z=0H=N=Z=0, contrary to (6.4). □

Remark 8.1 (The critical first mode). The coefficient (8.21) satisfies c1,p=0c_{1,p}=0 for every pp. Thus the argument is exactly critical at degree one. By Lemma 5.1, this degree-one mode is precisely the second fundamental form of M−M_-. Total geodesy removes the only mode not controlled by the spherical quadratic gap.

Analytic continuation and global splitting

The preceding argument gives vanishing to every normal order. To turn this local information into global rigidity, we apply real analyticity to a polynomial in the Ricci endomorphism. This avoids requiring the auxiliary horizontal distribution to be analytic.

Proof of Theorem 1.2. By Corollary 5.1 and Proposition 8.1, for every integer j≥1j \ge1,

E=O(rj)(70)E = O(r^j) \tag*{(70)}

uniformly on the compact unit normal sphere bundle. All normal Taylor coefficients of EE vanish as smooth tensor fields along M−M_-.

We do not need to establish analyticity of H\mathcal{H}. Instead, we pass to the intrinsic polynomial in the Ricci endomorphism

F=A2−12A.(71)F = A^2 - \frac{1}{2}A. \tag*{(71)}

Using A=12ΠH+EA = \frac{1}{2}\Pi_{\mathcal{H}} + E and ΠH2=ΠH\Pi_{\mathcal{H}}^2 = \Pi_{\mathcal{H}},

F=12(ΠHE+EΠH−E)+E2.(72)F = \frac{1}{2}(\Pi_{\mathcal{H}}E + E\Pi_{\mathcal{H}} - E) + E^2. \tag*{(72)}

Hence every normal Taylor coefficient of FF vanishes along M−M_-.

To make the passage to the full Taylor jet explicit, choose Fermi coordinates (x,y)(x,y) near a point of M−M_-, where xx are coordinates on M−M_- and yy are normal coordinates. For every multi-index β\beta in the normal variables,

∂yβF(x,0)=0\partial_y^\beta F(x,0) = 0

as a smooth tensor-valued function of xx. Differentiating this identity in the tangential variables shows

∂xα∂yβF(x,0)=0\partial_x^\alpha\partial_y^\beta F(x,0) = 0

for all multi-indices α,β\alpha,\beta. Thus the complete coordinate Taylor jet of FF vanishes at every point of M−M_-. Vanishing of the complete jet is invariant under smooth coordinate changes: every derivative in a new chart is a finite sum involving components of FF and their derivatives in the old chart. It therefore also holds in the analytic coordinates used below.

By Lemma 2.3, gg is locally real analytic in suitable coordinates. Its Ricci endomorphism and the polynomial FF are therefore real analytic. A real-analytic tensor with vanishing complete Taylor jet at a point vanishes in a neighborhood of that point. By connectedness and analytic continuation,

A2=12Aon M.(73)A^2 = \frac{1}{2}A \qquad\text{on } M. \tag*{(73)}

Since tr⁡A=k/2\operatorname{tr} A = k/2, the spectrum of AA consists everywhere of exactly kk copies of 1/21/2 and mm copies of 00. Thus rank⁡Ric⁡≡k\operatorname{rank}\operatorname{Ric} \equiv k. The fixed-rank rigidity theorem of Fernández-López and García-Río [8] gives

(M~,g~)≅(Nk,gN)×Rm,Ric⁡gN=12gN.(\widetilde{M},\widetilde{g}) \cong(N^k,g_N) \times\mathbb{R}^m,\qquad\operatorname{Ric}_{g_N} = \frac{1}{2}g_N.

The factor NN is complete with positive Ricci curvature, hence compact by Bonnet–Myers. Since the universal cover and Rm\mathbb{R}^m are simply connected, so is NN. □

Proof of Corollary 1.1. In dimension one the metric is flat and the soliton equation gives the Gaussian shrinker on the universal cover, so the conclusion holds. Assume n≥2n \ge2. By Lemma 2.4, we can write R=k/2R = k/2 for an integer $0 \le k \le n$. If k=0k = 0, k=n−1k = n - 1, or k=nk = n, rigidity is already known. The case k=1k = 1 does not occur. Every remaining value satisfies 2≤k≤n−22 \le k \le n - 2. Setting $m = n - k gives k,m≥2k,m \ge2, so Theorem 1.2 applies and proves rigidity. These cases exhaust all possible constant scalar curvatures. □

Remark 9.1. The proof separates the global geometry of the minimum set from the local vanishing argument. The global normal coordinates follow from the soliton identities and completeness. Total geodesy is then used to remove the first normal derivative; all subsequent finite degrees are excluded by the strict inequality (68). Thus proving total geodesy of M−M_{-} under the constant-scalar-curvature assumption alone would, together with Corollary 1.1, establish Cao’s rigidity conjecture in full.

Declaration of AI use. The authors used ChatGPT (OpenAI) during the preparation of the manuscript for language editing and exploratory calculations. All mathematical arguments and computations were independently checked by the authors.

(Fengjiang Li) Mathematical Science Research Center, Chongqing University of Technology, Chongqing 400054, China

Email address: [email protected]

School of Mathematical Sciences, Key Laboratory of MEA (Ministry of Education) and Shanghai Key Laboratory of PMMP, East China Normal University, Shanghai 200241, China

Email address: [email protected]

(Jianyu Ou) Department of Mathematics, Xiamen University, Xiamen 361005, China

Email address: [email protected]

(Yuanyuan Qu) School of Mathematical Sciences, Shanghai Key Laboratory of PMMP, East China Normal University, Shanghai 200241, China

Email address: [email protected]

School of Science, Zhejiang Sci-Tech University, Hangzhou 310018, China

Email address: [email protected]

Instituto de Matemática e Estatística, Universidade Federal Fluminense, São Domingos, Niterói, RJ 24210-201, Brazil

Email address: [email protected]

References

  1. [1]A. L. Besse, Einstein Manifolds, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 10, Springer-Verlag, Berlin, 1987.DOI
  2. [2]H.-D. Cao, Recent progress on Ricci solitons, in Recent Advances in Geometric Analysis, Adv. Lect. Math. 11 (2010), 1–38.arxiv.org/abs/0908.2006
  3. [3]H.-D. Cao and Q. Chen, On Bach-flat gradient shrinking Ricci solitons, Duke Math. J. 162 (2013), 1149–1169.DOI
  4. [4]H.-D. Cao and D. Zhou, On complete gradient shrinking Ricci solitons, J. Differential Geom. 85 (2010), 175–186.arxiv.org/abs/0903.3932
  5. [5]G. Catino, P. Mastrolia and D. D. Monticelli, Gradient Ricci solitons with vanishing conditions on Weyl, J. Math. Pures Appl. 108 (2017), 1–13.arxiv.org/abs/1602.00534
  6. [6]X. Cheng and D. Zhou, Rigidity of four-dimensional gradient shrinking Ricci solitons, J. Reine Angew. Math. 802 (2023), 255–274.arxiv.org/abs/2105.10744
  7. [7]M. Eminenti, G. La Nave and C. Mantegazza, Ricci solitons: the equation point of view, Manuscripta Math. 127 (2008), 345–367.arxiv.org/abs/math/0607546
  8. [8]M. Fernández-López and E. García-Río, On gradient Ricci solitons with constant scalar curvature, Proc. Amer. Math. Soc. 144 (2016), 369–378.
  9. [9]M. Fernández-López and E. García-Río, Rigidity of shrinking Ricci solitons, Math. Z. 269 (2011), 461–466.
  10. [10]R. S. Hamilton, Three-manifolds with positive Ricci curvature, J. Differential Geom. 17 (1982), 255–306.DOI
  11. [11]B. Kotschwar, A local version of Bando’s theorem on the real-analyticity of solutions to the Ricci flow, Bull. Lond. Math. Soc. 45 (2013), 153–158.
  12. [12]F. Li, J. Ou, Y. Qu and G. Wu, Rigidity of five-dimensional shrinking gradient Ricci solitons with constant scalar curvature, arXiv:2411.10712, 2024, revised 2025.arxiv.org/abs/2411.10712
  13. [13]F. Li, Y. Qu and G. Wu, Rigidity of shrinking gradient Ricci soliton with constant scalar curvature, arXiv:2608.20040, 2026.arxiv.org/abs/2608.20040
  14. [14]O. Munteanu and N. Šešum, On gradient Ricci solitons, J. Geom. Anal. 23 (2013), 539–561.
  15. [15]L. Ni and N. Wallach, On a classification of gradient shrinking solitons, Math. Res. Lett. 15 (2008), 941–955.DOI
  16. [16]J. Ou, Y. Qu and G. Wu, Some rigidity results on shrinking gradient Ricci soliton, J. Geom. Anal. 35 (2025), Paper No. 311.arxiv.org/abs/2411.06395
  17. [17]G. Perelman, The entropy formula for the Ricci flow and its geometric applications, arXiv:math/0211159, 2002.
  18. [18]P. Petersen and W. Wylie, Rigidity of gradient Ricci solitons, Pacific J. Math. 241 (2009), 329–345.arxiv.org/abs/0710.3174
  19. [19]P. G. Walczak, An integral formula for a Riemannian manifold with two orthogonal complementary distributions, Colloq. Math. 58 (1990), 243–252.DOI
  20. [20]Q.-M. Wang, Isoparametric functions on Riemannian manifolds. I, Math. Ann. 277 (1987), 639–646.DOI

Paper details

vlnc:W_hfdg2jq9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents