betaFeedback

WorksMathematics

A Smooth Nonnegatively Curved Sphere Whose Zero-Curvature Point Locus Has Hausdorff Dimension 1/2

Abstract

For every integer m≥3m\geq 3, we construct a smooth Riemannian metric gg on a manifold diffeomorphic to SmS^m such that secg≥0\text{sec}_g\geq 0, every sectional curvature is strictly positive away from a closed nowhere-dense set, and the point locus on which some sectional curvature vanishes has Hausdorff dimension exactly 12\frac{1}{2}. The metric is induced on the boundary of a smooth convex body in Rm+1\mathbb{R}^{m+1}. The essential local model is a convex graph whose Hessian has a one-dimensional kernel precisely on a Cantor subset of one coordinate axis; a regularized maximum then inserts this graph into a round sphere without introducing additional degenerate points. The main content of this paper is generated by ChatGPT 5.6 and verified by the author.

Statement and convention

Part of the motivation of this construction comes from recent advances in the study of positively curved manifolds, e.g. [1, 2, 3].

Acknowledgement: the author thanks Xi’an Jiaotong-Liverpool University for providing financial and academic support. He also thanks Zipei Nie for providing access to ChatGPT and insights about AI usage. His gratitude also goes to his daughter and son who were born last year, and all of his family members who have always been supportive during his academic career.

For a Riemannian manifold (M,g)(M,g) with nonnegative sectional curvature, define the zero-curvature point locus

Z(g):={p∈M:there is a two-plane Π⊂TpM with sec⁡g(Π)=0}.Z(g) := \{p \in M : \text{there is a two-plane } \Pi\subset T_pM \text{ with } \operatorname{sec}_g(\Pi) = 0\}.

This is a subset of the manifold, rather than a subset of the Grassmann bundle of two-planes.

The main conclusion of this paper is the following theorem:

Theorem 1.1. For every integer m≥3m \ge3, there is a smooth closed hypersurface Σm⊂Rm+1\Sigma^m \subset\mathbb{R}^{m+1}, diffeomorphic to SmS^m, such that its induced metric gg satisfies

  1. sec⁡g≥0\operatorname{sec}_g \ge0 everywhere;

  2. every two-plane has strictly positive sectional curvature at every point of the open dense set Σ∖Z(g)\Sigma\setminus Z(g);

  3. dim⁡HZ(g)=12\dim_{\mathrm{H}} Z(g) = \frac{1}{2}.

In fact, Z(g)Z(g) is bi-Lipschitz equivalent to a self-similar Cantor set of Hausdorff dimension 1/21/2.

We note that in the construction below, degeneracy is confined simultaneously in the transverse variables, so the exceptional point set itself is only a Cantor set.

2010 Mathematics Subject Classification. 53C21.

Key words and phrases. Sectional Curvature, Hausdorff Dimension.

Three elementary tools

A Cantor set of dimension 1/2

Let C⊂[0,1]C \subset[0,1] be the unique nonempty compact set satisfying

C=14C∪(34+14C).(1)C=\frac{1}{4}C\cup\left(\frac{3}{4}+\frac{1}{4}C\right). \tag*{(1)}

Equivalently, at each stage one retains the left and right subintervals of relative length 1/41/4. Put

E:=18(C−12)⊂[−116,116].(2)E:=\frac{1}{8}\left(C-\frac{1}{2}\right)\subset\left[-\frac{1}{16},\frac{1}{16}\right]. \tag*{(2)}

Then EE is compact, nowhere dense, and symmetric under t↦−tt\mapsto-t.

Lemma 2.1. The Hausdorff dimension of EE is 1/21/2.

Proof. Translation and nonzero scaling do not change Hausdorff dimension, so it is enough to treat CC. At level kk, the set CC is covered by 2k2^k intervals of length 4−k4^{-k}. Hence, for every s>1/2s>1/2,

2k(4−k)s=(2 4−s)k⟶0,2^k(4^{-k})^s=(2\,4^{-s})^k\longrightarrow0,

which gives dim⁡HC≤1/2\dim_{\mathrm H}C\le1/2.

For the reverse inequality, let μ\mu be the natural probability measure assigning mass 2−k2^{-k} to every level-kk interval. Distinct level-kk intervals have length 4−k4^{-k} and are separated by gaps of length at least 2⋅4−k2\cdot4^{-k}. Therefore an interval II of length at most 4−k4^{-k} meets at most two level-kk intervals. If

4−(k+1)<∣I∣≤4−k,4^{-(k+1)}<|I|\le4^{-k},

then

μ(I)≤2⋅2−k=2(4−k)1/2<4∣I∣1/2.\mu(I)\le2\cdot2^{-k}=2(4^{-k})^{1/2}<4|I|^{1/2}.

Thus, for every countable interval cover C⊂⋃jIjC\subset\bigcup_j I_j by sufficiently short intervals,

1=μ(C)≤∑jμ(Ij)≤4∑j∣Ij∣1/2.1=\mu(C)\le\sum_j\mu(I_j)\le4\sum_j|I_j|^{1/2}.

It follows that the 1/21/2-dimensional Hausdorff measure of CC is positive, and hence dim⁡HC≥1/2\dim_{\mathrm H}C\ge1/2. □

A smooth function with a prescribed zero set

Lemma 2.2. If A⊂RA\subset\mathbb{R} is closed, then there is a function q∈C∞(R)q\in C^\infty(\mathbb{R}) such that q≥0q\ge0 and q−1(0)=Aq^{-1}(0)=A. If A=−AA=-A, the function may be chosen even and bounded by 11.

Proof. Let O=R∖AO=\mathbb{R}\setminus A. Choose nonnegative functions ϕj∈Cc∞(O)\phi_j\in C^\infty_c(O) such that the open sets {ϕj>0}\{\phi_j>0\} cover OO. Choose aj>0a_j>0 so rapidly decreasing that

aj(1+max⁡0≤r≤j∥ϕj(r)∥∞)≤2−j.a_j\left(1+\max_{0\le r\le j}\left\|\phi_j^{(r)}\right\|_\infty\right)\le2^{-j}.

For every fixed derivative order, the corresponding differentiated series converges uniformly. Hence

q0:=∑j=1∞ajϕjq_0:=\sum_{j=1}^{\infty}a_j\phi_j

is smooth, nonnegative, and has zero set exactly AA. If A=−AA=-A, replace q0q_0 by q0(t)+q0(−t)q_0(t)+q_0(-t) and then by

q(t):=q0(t)+q0(−t)1+q0(t)+q0(−t).q(t):=\frac{q_0(t)+q_0(-t)}{1+q_0(t)+q_0(-t)}.

The resulting function is smooth, even, takes values in [0,1)[0,1), and has zero set AA. □

Apply Lemma 2.2 to the symmetric set EE in (2). Fix from now on an even function

q∈C∞(R),0≤q≤1,q−1(0)=E.(3)q \in C^\infty(\mathbb{R}), \qquad0 \le q \le1, \qquad q^{-1}(0) = E. \tag*{(3)}

A regularized maximum

Lemma 2.3. For every δ>0\delta> 0, there is a smooth function Mδ:R2→RM_\delta: \mathbb{R}^2 \to\mathbb{R} with the following properties:

(a) max⁡{a,b}≤Mδ(a,b)≤max⁡{a,b}+δ/2\max\{a,b\} \le M_\delta(a,b) \le\max\{a,b\} + \delta/2;

(b) Mδ(a,b)=max⁡{a,b}M_\delta(a,b) = \max\{a,b\} whenever ∣a−b∣≥δ|a-b| \ge\delta;

(c) if AA and BB are smooth convex functions on a convex domain, then Mδ(A,B)M_\delta(A,B) is convex;

(d) at a point where ∣A−B∣<δ|A-B| < \delta, if either D2AD^2A or D2BD^2B is positive definite, then D2Mδ(A,B)D^2M_\delta(A,B) is positive definite.

Proof. Choose an even nonnegative mollifier ρ∈C∞(R)\rho\in C^\infty(\mathbb{R}) supported in [−δ,δ][-\delta,\delta], positive for ∣r∣<δ|r| < \delta, and satisfying ∫Rρ=1\int_{\mathbb{R}} \rho= 1. Define

ϑδ(s):=(∣⋅∣∗ρ)(s)=∫R∣s−r∣ρ(r) dr.\vartheta_\delta(s) := (|\cdot| * \rho)(s) = \int_{\mathbb{R}} |s-r|\rho(r)\,dr.

Then ϑδ\vartheta_\delta is smooth, even, and convex. Since ρ\rho is even, it has mean zero, and therefore ϑδ(s)=∣s∣\vartheta_\delta(s) = |s| for ∣s∣≥δ|s| \ge\delta. Jensen’s inequality and the triangle inequality give

∣s∣≤ϑδ(s)≤∣s∣+δ.|s| \le\vartheta_\delta(s) \le|s| + \delta.

Moreover, ∣ϑδ′(s)∣<1|\vartheta'_\delta(s)| < 1 for ∣s∣<δ|s| < \delta, and ϑδ′′=2ρ≥0\vartheta''_\delta= 2\rho\ge0.

Define

Mδ(a,b):=a+b+ϑδ(a−b)2.M_\delta(a,b) := \frac{a+b+\vartheta_\delta(a-b)}{2}.

The first two assertions follow immediately. For F=Mδ(A,B)F = M_\delta(A,B), put d=A−Bd = A-B and

λ:=1+ϑδ′(d)2∈[0,1].\lambda:= \frac{1+\vartheta'_\delta(d)}{2} \in[0,1].

A direct differentiation gives

D2F=λD2A+(1−λ)D2B+12ϑδ′′(d) d(A−B)⊗d(A−B).(4)D^2F = \lambda D^2A + (1-\lambda)D^2B + \frac{1}{2}\vartheta''_\delta(d)\,d(A-B)\otimes d(A-B). \tag*{(4)}

All three terms are positive semidefinite when AA and BB are convex. If ∣d∣<δ|d| < \delta, then 0<λ<10 < \lambda< 1, so a positive-definite Hessian among D2A,D2BD^2A,D^2B makes the right-hand side positive definite. □

A convex local model with Cantor degeneracy

Write a point of Rm\mathbb{R}^m as

x=(t,y),t∈R,y∈Rm−1.x = (t,y), \qquad t \in\mathbb{R}, \quad y \in\mathbb{R}^{m-1}.

Let Bm={x∈Rm:∣x∣<1}B^m = \{x \in\mathbb{R}^m : |x| < 1\} and fix

ε:=1100.\varepsilon:= \frac{1}{100}.

Define

u(t):=ε∫0t(t−s)q(s) ds.(5)u(t) := \varepsilon\int_0^t (t-s)q(s)\,ds. \tag*{(5)}

Since qq is even, the function uu is even. Moreover,

u′′(t)=εq(t),u(0)=u′(0)=0,0≤u(t)≤ε2t2(∣t∣≤1).(6)u''(t) = \varepsilon q(t), \qquad u(0) = u'(0) = 0, \qquad0 \le u(t) \le\frac{\varepsilon}{2}t^2 \quad(|t| \le1). \tag*{(6)}

Set

U0(t,y):=u(t)+ε2(4+t2)∣y∣2,U(t,y):=U0(t,y)−12.(7)U_0(t,y) := u(t) + \frac{\varepsilon}{2}(4+t^2)|y|^2, \qquad U(t,y) := U_0(t,y) - \frac{1}{2}. \tag*{(7)}

Let

K:=E×{0}⊂Bm.(8)K := E \times\{0\} \subset B^m. \tag*{(8)}

Proposition 3.1. The function UU is convex on BmB^m. Its Hessian is positive definite on Bm∖KB^m \setminus K, and at every point of KK its Hessian has rank exactly m−1m-1. Moreover,

0≤U0(x)≤3ε=3100(x∈Bm).(9)0 \le U_0(x) \le3\varepsilon= \frac{3}{100} \qquad(x \in B^m). \tag*{(9)}

Proof. Since the constant −1/2-1/2 does not affect the Hessian, it is enough to consider U0U_0. In the splitting R⊕Rm−1\mathbb{R} \oplus\mathbb{R}^{m-1},

D2U0(t,y)=ε(q(t)+∣y∣22tyT2ty(4+t2)Im−1).(10)D^2U_0(t,y) = \varepsilon \begin{pmatrix} q(t)+|y|^2 & 2ty^T \\ 2ty & (4+t^2)I_{m-1} \end{pmatrix}. \tag*{(10)}

The lower-right block is positive definite. Its Schur complement is

ε(q(t)+∣y∣2−4t24+t2∣y∣2)=ε(q(t)+4−3t24+t2∣y∣2).(11)\varepsilon\left(q(t)+|y|^2-\frac{4t^2}{4+t^2}|y|^2\right) = \varepsilon\left(q(t)+\frac{4-3t^2}{4+t^2}|y|^2\right). \tag*{(11)}

Because ∣t∣<1|t|<1, the coefficient (4−3t2)/(4+t2)(4-3t^2)/(4+t^2) is positive. Thus the Schur complement is nonnegative, and it vanishes exactly when q(t)=0q(t)=0 and y=0y=0, namely at the points of KK. This proves convexity and positive definiteness off KK. At a point (t,0)∈K(t,0)\in K, the matrix in (10) is

ε(000(4+t2)Im−1),\varepsilon \begin{pmatrix} 0 & 0 \\ 0 & (4+t^2)I_{m-1} \end{pmatrix},

which has rank m−1m-1.

Finally, (6) and ∣t∣,∣y∣<1|t|,|y|<1 give

0≤U0(t,y)≤ε2+5ε2=3ε.0 \le U_0(t,y) \le\frac{\varepsilon}{2}+\frac{5\varepsilon}{2}=3\varepsilon.

The mixed term in (7) is the localization mechanism. The tt-direction Hessian may vanish when t∈Et\in E, but for y≠0y\ne0 the positive term ε∣y∣2\varepsilon|y|^2 in UttU_{tt} restores strict convexity. Hence the degeneracy does not extend to an (m−1)(m-1)-dimensional fiber.

Insertion into a round sphere

On BmB^m, write the lower and upper unit hemispheres as graphs

f−(x):=−1−∣x∣2,f+(x):=1−∣x∣2.(12)f_-(x):=-\sqrt{1-|x|^2},\qquad f_+(x):=\sqrt{1-|x|^2}. \tag*{(12)}

The function f−f_- is strictly convex, since

D2f−(x)=11−∣x∣2Im+x⊗x(1−∣x∣2)3/2>0.(13)D^2f_-(x)=\frac{1}{\sqrt{1-|x|^2}}I_m+\frac{x\otimes x}{(1-|x|^2)^{3/2}}>0. \tag*{(13)}

Fix

δ:=110.\delta:=\frac{1}{10}.

and define

F(x):=Mδ(U(x),f−(x)).(14)F(x):=M_\delta(U(x),f_-(x)). \tag*{(14)}

Proposition 4.1. The function FF has the following properties.

  1. FF is smooth and convex on BmB^m.

  2. D2FD^2F is singular exactly on KK; there it has rank m−1m-1, and it is positive definite on Bm∖KB^m\setminus K.

  3. F=f−F=f_- whenever ∣x∣≥24/25|x|\ge24/25.

  4. F(x)<f+(x)F(x)<f_+(x) for every x∈Bmx\in B^m.

Proof. Convexity and smoothness follow from Lemma 2.3, Proposition 3.1, and (13). Put

d(x):=U(x)−f−(x)=U0(x)−12+1−∣x∣2.d(x) := U(x) - f_-(x) = U_0(x) - \frac{1}{2} + \sqrt{1-|x|^2}.

If x=(t,0)∈Kx=(t,0)\in K, then ∣t∣≤1/16|t|\le1/16 and

d(x)≥−12+1−1256>49100>δ.(15)d(x) \ge-\frac{1}{2}+\sqrt{1-\frac{1}{256}}>\frac{49}{100}>\delta. \tag*{(15)}

Thus F=UF=U on a neighborhood of KK, so D2FD^2F has rank m−1m-1 there.

If d≥δd\ge\delta away from KK, then F=UF=U and D2F>0D^2F>0. If d≤−δd\le-\delta, then F=f−F=f_- and D2F>0D^2F>0. In the transition region ∣d∣<δ|d|<\delta, formula (4) contains a positive multiple of the positive-definite matrix D2f−D^2f_-, so D2F>0D^2F>0. At the boundary value d=δd=\delta, the equality F=UF=U holds; such a point is not in KK by (15), and hence D2U>0D^2U>0. The case d=−δd=-\delta reduces to f−f_-. This proves (2).

If ∣x∣≥24/25|x|\ge24/25, then

1−∣x∣2≤1−576625=725.\sqrt{1-|x|^2}\le\sqrt{1-\frac{576}{625}}=\frac{7}{25}.

Using (9),

d(x)≤3100−12+725=−19100<−δ.d(x)\le\frac{3}{100}-\frac{1}{2}+\frac{7}{25}=-\frac{19}{100}<-\delta.

Therefore F=f−F=f_- on this outer collar, proving (3).

It remains to compare FF with f+f_+. If ∣x∣≤24/25|x|\le24/25, then

f+(x)≥725.f_+(x)\ge\frac{7}{25}.

Both U(x)≤−47/100U(x)\le-47/100 and f−(x)≤0f_-(x)\le0, so Lemma 2.3 gives

F(x)≤max⁡{U(x),f−(x)}+δ2≤120<725≤f+(x).F(x)\le\max\{U(x),f_-(x)\}+\frac{\delta}{2}\le\frac{1}{20}<\frac{7}{25}\le f_+(x).

If ∣x∣≥24/25|x|\ge24/25, then F=f−<f+F=f_-<f_+. This proves (4).

Define a closed hypersurface in Rm×R\mathbb{R}^m\times\mathbb{R} by

Σ:={(x,F(x)):x∈Bm}∪{(x,f+(x)):x∈Bm}∪{(x,0):∣x∣=1}.(16)\Sigma:=\{(x,F(x)):x\in B^m\}\cup\{(x,f_+(x)):x\in B^m\}\cup\{(x,0):|x|=1\}. \tag*{(16)}

Because F=f−F=f_- on an outer collar, the first graph agrees there with the lower unit hemisphere. Hence the union in (16) is a smooth embedded hypersurface and is diffeomorphic to Sm\mathbb{S}^m. Proposition 4.1(4) shows that the two graph pieces are disjoint away from the equator.

The hypersurface is also the boundary of the convex body

Ω:={(x,z):x∈Bm‾, F(x)≤z≤f+(x)}.(17)\Omega:=\{(x,z):x\in\overline{B^m},\ F(x)\le z\le f_+(x)\}. \tag*{(17)}

Indeed, FF is convex and f+f_+ is concave, so the set in (17) is convex. This global convexity is consistent with the local curvature calculation below, but the calculation itself will give the precise zero-curvature locus.

Sectional curvature of the hypersurface

Let gg be the metric induced on Σ\Sigma from the Euclidean metric. We first record the graph calculation. For a smooth function GG on a domain in Rm\mathbb{R}^m, parameterize its graph by

X(x)=(x,G(x)).X(x)=(x,G(x)).

For tangent vectors v,w∈Rmv,w\in\mathbb{R}^m, the first and second fundamental forms, using the upward unit normal, are

I(v,w)=⟨v,w⟩+dG(v)dG(w),II(v,w)=D2G(v,w)1+∣∇G∣2.(18)I(v,w)=\langle v,w\rangle+dG(v)dG(w),\qquad II(v,w)=\frac{D^2G(v,w)}{\sqrt{1+|\nabla G|^2}}. \tag*{(18)}

The Euclidean Gauss equation therefore gives

sec⁡g(dX(span⁡{v,w}))=D2G(v,v)D2G(w,w)−D2G(v,w)2(1+∣∇G∣2)(I(v,v)I(w,w)−I(v,w)2)(19)\operatorname{sec}_g\bigl(dX(\operatorname{span}\{v,w\})\bigr)=\frac{D^2G(v,v)D^2G(w,w)-D^2G(v,w)^2}{(1+|\nabla G|^2)\bigl(I(v,v)I(w,w)-I(v,w)^2\bigr)} \tag*{(19)}

for linearly independent v,wv,w.

If D2GD^2G is positive semidefinite, the numerator in (19) is nonnegative by the Cauchy–Schwarz inequality for a positive-semidefinite bilinear form. If D2GD^2G is positive definite, its restriction to every two-dimensional subspace is positive definite, so the numerator is strictly positive.

On the upper hemisphere, the induced metric is the standard round metric and has sectional curvature 11. On the lower graph, take G=FG=F. Proposition 4.1 shows that D2F≥0D^2F\geq0, and hence every sectional curvature is nonnegative. At every point outside the graph of KK, the Hessian is positive definite, so every sectional curvature is strictly positive.

At a point (t,0)∈K(t,0)\in K, the function FF agrees with UU, and (10) shows that

D2F(∂t,∂t)=0,D2F(∂t,v)=0for every v∈Rm.D^2F(\partial_t,\partial_t)=0,\qquad D^2F(\partial_t,v)=0\quad\text{for every }v\in\mathbb{R}^m.

Taking v=∂tv=\partial_t and any nonzero ww in a yy-direction in (19) gives a zero-curvature two-plane. Consequently,

Z(g)={((t,0),F(t,0)):t∈E}.(20)\mathcal{Z}(g)=\{((t,0),F(t,0)):t\in E\}. \tag*{(20)}

This equality also proves that no additional zero-curvature points are introduced by the global patching.

Hausdorff dimension and density

Let

Φ:E⟶Z(g),Φ(t)=((t,0),F(t,0)).\Phi:E\longrightarrow\mathcal{Z}(g),\qquad\Phi(t)=((t,0),F(t,0)).

For every ∣t∣≤1/16|t|\leq1/16 and y=0y=0, the estimate used in (15) gives U−f−>δU-f_->\delta, not merely at points of EE. Hence

F(t,0)=U(t,0)=u(t)−12(∣t∣≤116).F(t,0)=U(t,0)=u(t)-\frac{1}{2}\qquad\left(|t|\leq\frac{1}{16}\right).

The Euclidean distance between Φ(s)\Phi(s) and Φ(t)\Phi(t) is at least ∣s−t∣|s-t|. Conversely, the curve

r⟼((r,0),u(r)−12),r∈[s,t],r\longmapsto\left((r,0),u(r)-\frac{1}{2}\right),\qquad r\in[s,t],

lies in Σ\Sigma and has length at most

(1+sup⁡∣r∣≤1/16∣u′(r)∣2)1/2∣s−t∣.\left(1+\sup_{|r|\leq1/16}|u'(r)|^2\right)^{1/2}|s-t|.

Since intrinsic distance is at least Euclidean distance and at most the length of this curve, Φ\Phi is bi-Lipschitz from EE onto Z(g)\mathcal{Z}(g). Lemma 2.1 now yields

dim⁡HZ(g)=dim⁡HE=12.\dim_{\mathrm{H}}\mathcal{Z}(g)=\dim_{\mathrm{H}}E=\frac{1}{2}.

The set Z(g)\mathcal{Z}(g) is compact and has Hausdorff dimension smaller than mm, so it has empty interior. Its complement is therefore open and dense. Every sectional curvature is strictly positive at every point of this complement, as proved in the preceding section. The induced metric is C∞C^\infty, and hence in particular is of class C2C^2. This completes the proof of Theorem 1.1.

Remark 6.1. The same construction gives any prescribed dimension α∈(0,1)\alpha\in(0,1). Replace the ratio 1/41/4 in (1) by λ=2−1/α∈(0,1/2)\lambda=2^{-1/\alpha}\in(0,1/2). The resulting two-branch self-similar Cantor set has Hausdorff dimension α\alpha, while all convexity and curvature arguments are unchanged.

References

References

  1. [1]L. Kennard. Torus actions and positive curvature. Virtual seminar on geometry with symmetries, May 6, 2020.
  2. [2]A. Krishnan and M. Wiemeler. 10-dimensional positively curved manifolds with T^3-symmetry. Commun. Contemp. Math., Vol. 27, No. 10, 2550018 (2025).
  3. [3]Y. Liu. On Closed Six-Manifolds Admitting Riemannian Metrics with Positive Sectional Curvature and Non-Abelian Symmetry. Acta Mathematica Sinica-English Series, 40(12), 2024, 3003–3026.

Paper details

vlnc:W_md8qa2hk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

ChatGPT 5.6

View statement
Looking for related papers…
Contents