betaFeedback

WorksMathematics

Almost Positive Sectional Curvature with Bounded Geometry on S2×S2S^2\times S^2

Abstract

Let M=S2×S2M=S^2\times S^2. We prove that there is a fixed smooth open ball U⊂MU\subset M such that, for every ε>0\varepsilon>0, one can find a smooth metric gεg_\varepsilon and a closed set SεS_\varepsilon with the following properties: sec⁡gε>0\sec_{g_\varepsilon}>0 on M∖SεM\setminus S_\varepsilon; the relative volume of SεS_\varepsilon is less than ε\varepsilon; every sectional curvature on UU is greater than 11; −ε<sec⁡gε≤Λ-\varepsilon<\sec_{g_\varepsilon}\leqΛ for one constant ΛΛ independent of ε\varepsilon; the volume and diameter have uniform positive lower and upper bounds; and the injectivity radius has a uniform positive lower bound. In fact, U∩Sε=∅U\cap S_\varepsilon=\varnothing. The construction starts from the product of the unit round metrics. A small conformal factor, obtained by smoothing the absolute value of the height function on each sphere, has negative definite Hessian away from two thin equatorial bands. It makes every formerly flat mixed plane strictly positive outside a set of arbitrarily small relative volume, while the negative curvature created in the bands is only O(ε)O(\varepsilon). A second conformal deformation supported in a very small region produces a core on which all sectional curvatures are greater than 11; its transition has Hessian bounded above by O(ε)O(\varepsilon). Finally, a diffeomorphism compresses the fixed topological ball UU into that core. Pullback preserves all intrinsic bounded-geometry estimates. The main content of this paper is generated by ChatGPT 5.6 and verified by the author.

Introduction and Statement

The main purpose of this paper is to try to produce a metric on S2×S2S^2 \times S^2 with quasi-positive sectional curvature, which is not successful. We note that recently Brendle and Hung constructed metrics with positive sectional curvature on S2×S2S^2 \times S^2, resolving the Hopf conjecture [1]. However, since the metrics constructed in this paper is relatively simple and explicit, we post it as a note of independent interest. Main techniques used in this paper can be found in standard textbooks like [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.

Let hh be the unit round metric on S2S^2, and put

g0=h⊕honM=S2×S2.g_0 = h \oplus h \quad\text{on} \quad M = S^2 \times S^2.

An open ball below means a subset diffeomorphic to the standard open 4-ball; it is fixed as a subset of the smooth manifold and is not required to have a fixed intrinsic radius for the varying metrics.

Theorem 1.1. There exist a smooth open ball U⊂MU \subset M and constants

0<v−≤v+<∞,0<d−≤d+<∞,i0>0,Λ<∞,0 < v_- \leq v_+ < \infty,\qquad0 < d_- \leq d_+ < \infty,\qquad i_0 > 0,\qquad\Lambda< \infty,

such that, for every ε>0\varepsilon> 0, there are a smooth Riemannian metric gεg_{\varepsilon} on MM and a closed subset Sε⊂MS_{\varepsilon} \subset M satisfying

  1. sec⁡gε>0\operatorname{sec}_{g_{\varepsilon}} > 0 on M∖SεM \setminus S_{\varepsilon};

2.

Vol⁡(Sε,gε)Vol⁡(M,gε)<ε;\frac{\operatorname{Vol}(S_{\varepsilon},g_{\varepsilon})}{\operatorname{Vol}(M,g_{\varepsilon})} < \varepsilon;
  1. U∩Sε=∅U \cap S_{\varepsilon} = \varnothing, and every two-plane based at a point of UU has sectional curvature greater than 11;

4.

−ε<sec⁡gε≤Λon M;-\varepsilon< \operatorname{sec}_{g_{\varepsilon}} \leq\Lambda\quad\text{on } M;

5.

v−≤Vol⁡(M,gε)≤v+,d−≤diam⁡(M,gε)≤d+;v_{-} \leq\operatorname{Vol}(M,g_{\varepsilon}) \leq v_{+}, \qquad d_{-} \leq\operatorname{diam}(M,g_{\varepsilon}) \leq d_{+};
  1. inj⁡(M,gε)≥i0\operatorname{inj}(M,g_{\varepsilon}) \geq i_{0}.

The freedom removed from the earlier one-region formulation is crucial. The exceptional set constructed here is distributed along two thin equatorial slabs, together with one small transition annulus. Its volume tends to zero, but it is not forced to have small diameter.

The product metric and conformal curvature

For a two-plane P⊂T(x,y)MP \subset T_{(x,y)}M, choose a g0g_{0}-orthonormal basis X=(X1,X2)X=(X_{1},X_{2}), Y=(Y1,Y2)Y=(Y_{1},Y_{2}). The product curvature tensor gives

sec⁡g0(P)=∣X1∧Y1∣h2+∣X2∧Y2∣h2.(1)\operatorname{sec}_{g_{0}}(P)=\lvert X_{1}\wedge Y_{1}\rvert_{h}^{2}+\lvert X_{2}\wedge Y_{2}\rvert_{h}^{2}. \tag*{(1)}

Hence

0≤sec⁡g0≤1.(2)0 \leq\operatorname{sec}_{g_{0}} \leq1. \tag*{(2)}

The zero-curvature planes are precisely the mixed planes.

We shall repeatedly use the conformal curvature formula. If

g~=e2fg\widetilde{g}=e^{2f}g

and e1,e2e_{1},e_{2} are gg-orthonormal vectors spanning PP, then

e2fsec⁡g~(P)=sec⁡g(P)−Hess⁡gf(e1,e1)−Hess⁡gf(e2,e2)+df(e1)2+df(e2)2−∣∇gf∣2.(3)e^{2f}\operatorname{sec}_{\widetilde{g}}(P)=\operatorname{sec}_{g}(P)-\operatorname{Hess}_{g}f(e_{1},e_{1})-\operatorname{Hess}_{g}f(e_{2},e_{2})+df(e_{1})^{2}+df(e_{2})^{2}-\lvert\nabla_{g}f\rvert^{2}. \tag*{(3)}

In particular,

df(e1)2+df(e2)2−∣∇f∣2≥−∣∇f∣2.(4)df(e_{1})^{2}+df(e_{2})^{2}-\lvert\nabla f\rvert^{2}\geq-\lvert\nabla f\rvert^{2}. \tag*{(4)}

An almost concave function on the sphere

Let z:S2→[−1,1]z:S^{2}\to[-1,1] be the height function. For the unit round metric,

∣∇z∣2=1−z2,Hess⁡hz=−zh.(5)\lvert\nabla z\rvert^{2}=1-z^{2}, \qquad\operatorname{Hess}_{h}z=-zh. \tag*{(5)}

Choose once and for all an even nonnegative function β∈Cc∞((−1,1))\beta\in C_{c}^{\infty}((-1,1)) with ∫Rβ=1\int_{\mathbb{R}}\beta=1, and put

βδ(t)=δ−1β(t/δ),ψδ=∣⋅∣∗βδ,0<δ<14.\beta_{\delta}(t)=\delta^{-1}\beta(t/\delta), \qquad\psi_{\delta}=\lvert\cdot\rvert*\beta_{\delta}, \qquad0<\delta<\frac{1}{4}.

Let

Cβ=2∥β∥∞.C_{\beta}=2\lVert\beta\rVert_{\infty}.

Lemma 3.1. The function ψδ\psi_{\delta} is smooth, even and convex, and it satisfies

(i) ψδ(t)=∣t∣\psi_{\delta}(t)=\lvert t\rvert whenever ∣t∣≥δ\lvert t\rvert\geq\delta;

(ii) ∣ψδ′∣≤1|\psi_\delta'| \le1, tψδ′(t)≥0t\psi_\delta'(t) \ge0, and

0≤ψδ′′(t)≤Cβδ;0 \le\psi_\delta''(t) \le\frac{C_\beta}{\delta};

(iii) 0<ψδ(t)≤1+δ0 < \psi_\delta(t) \le1+\delta for ∣t∣≤1|t| \le1.

If ϕδ=ψδ∘z\phi_\delta=\psi_\delta\circ z, then

Hess⁡hϕδ=ψδ′′(z) dz⊗dz−zψδ′(z)h.(6)\operatorname{Hess}_h\phi_\delta=\psi_\delta''(z)\,dz\otimes dz-z\psi_\delta'(z)h. \tag*{(6)}

Consequently, as quadratic forms,

−h≤Hess⁡hϕδ≤Cβδh,∣∇ϕδ∣≤1,(7)-h\le\operatorname{Hess}_h\phi_\delta\le\frac{C_\beta}{\delta}h,\qquad|\nabla\phi_\delta|\le1, \tag*{(7)}

and on {∣z∣>δ}\{|z|>\delta\},

Hess⁡hϕδ=−∣z∣h≤−δh.(8)\operatorname{Hess}_h\phi_\delta=-|z|h\le-\delta h. \tag*{(8)}

Proof. Convolution preserves smoothness, evenness and convexity. If t≥δt\ge\delta, then t−s≥0t-s\ge0 for every ss in the support of βδ\beta_\delta, and the evenness of βδ\beta_\delta gives

ψδ(t)=∫(t−s)βδ(s) ds=t.\psi_\delta(t)=\int(t-s)\beta_\delta(s)\,ds=t.

The case t≤−δt\le-\delta is identical. Since the distributional second derivative of ∣t∣|t| is twice the Dirac mass at the origin,

ψδ′′=2βδ.\psi_\delta''=2\beta_\delta.

The remaining one-dimensional assertions follow immediately. Formula (6) follows from the chain rule and (5). Because zψδ′(z)≥0z\psi_\delta'(z)\ge0, the second term in (6) is nonpositive. The upper bound follows from dz⊗dz≤hdz\otimes dz\le h, and the lower bound follows from ∣zψδ′(z)∣≤1|z\psi_\delta'(z)|\le1. On ∣z∣>δ|z|>\delta one has ψδ(z)=∣z∣\psi_\delta(z)=|z|, which gives (8). □

For Eδ={∣z∣≤δ}⊂S2E_\delta=\{|z|\le\delta\}\subset S^2, spherical coordinates give

Area⁡(Eδ,h)Area⁡(S2,h)=δ.(9)\frac{\operatorname{Area}(E_\delta,h)}{\operatorname{Area}(S^2,h)}=\delta. \tag*{(9)}

Define the two equatorial slabs

Σδ=(Eδ×S2)∪(S2×Eδ).(10)\Sigma_\delta=(E_\delta\times S^2)\cup(S^2\times E_\delta). \tag*{(10)}

Then

Vol⁡(Σδ,g0)Vol⁡(M,g0)=2δ−δ2<2δ.(11)\frac{\operatorname{Vol}(\Sigma_\delta,g_0)}{\operatorname{Vol}(M,g_0)}=2\delta-\delta^2<2\delta. \tag*{(11)}

For a>0a>0, set

ua,δ(x,y)=a(ϕδ(x)+ϕδ(y)).(12)u_{a,\delta}(x,y)=a\bigl(\phi_\delta(x)+\phi_\delta(y)\bigr). \tag*{(12)}

Lemma 3.1 gives

−ag0≤Hess⁡g0ua,δ≤aCβδg0,∣∇ua,δ∣2≤2a2,(13)-ag_0\le\operatorname{Hess}_{g_0}u_{a,\delta}\le\frac{aC_\beta}{\delta}g_0,\qquad|\nabla u_{a,\delta}|^2\le2a^2, \tag*{(13)}

and, on M∖ΣδM\setminus\Sigma_\delta,

Hess⁡g0ua,δ≤−aδg0.(14)\operatorname{Hess}_{g_0}u_{a,\delta}\le-a\delta g_0. \tag*{(14)}

Proposition 3.2. If 0<a<δ0<a<\delta, then the metric

gˉa,δ=e2ua,δg0\bar g_{a,\delta}=e^{2u_{a,\delta}}g_0

has strictly positive sectional curvature on M∖ΣδM\setminus\Sigma_\delta. Globally,

sec⁡gˉa,δ≥−2aCβδ−2a2,(15)\operatorname{sec}_{\bar g_{a,\delta}}\ge-\frac{2aC_\beta}{\delta}-2a^2, \tag*{(15)}

and

sec⁡gˉa,δ≤1+2a+2a2.(16)\operatorname{sec}_{\bar g_{a,\delta}} \le1 + 2a + 2a^2. \tag*{(16)}

Proof. Apply (3) with g=g0g=g_0 and f=ua,δf=u_{a,\delta}. On M∖ΣδM\setminus\Sigma_\delta, (14), (4), and (2) give

e2ua,δsec⁡gˉa,δ(P)≥2aδ−2a2>0.e^{2u_{a,\delta}}\operatorname{sec}_{\bar g_{a,\delta}}(P) \ge2a\delta-2a^2>0.

The global lower bound follows from the upper Hessian estimate in (13). For the upper bound, use

Hess⁡ua,δ≥−ag0\operatorname{Hess}u_{a,\delta}\ge-ag_0

and

du(e1)2+du(e2)2−∣∇u∣2≤∣∇u∣2.du(e_1)^2+du(e_2)^2-|\nabla u|^2\le|\nabla u|^2.

Finally ua,δ≥0u_{a,\delta}\ge0, so the factor e−2ua,δe^{-2u_{a,\delta}} never enlarges a positive upper bound or a negative lower bound in the estimates above. □

A bounded-curvature positive needle

We next construct a conformal factor whose Hessian is strongly negative on a very small core, while its positive Hessian is arbitrarily small everywhere.

Lemma 4.1 (local needle). Let (N,g)(N,g) be a smooth closed Riemannian manifold and let q∈Nq\in N. There are constants

R∗>0,c∗>0,C∗>1,R_* > 0,\qquad c_* > 0,\qquad C_* > 1,

depending only on (N,g,q)(N,g,q), with the following property. For every 0<σ≤10<\sigma\le1 and every 0<R≤R∗0<R\le R_* there is a smooth function w=wσ,R:N→[0,∞)w=w_{\sigma,R}:N\to[0,\infty) such that

  1. w=0w=0 outside Bg(q,R)B_g(q,R) and in a neighborhood of its boundary;

−C∗g≤Hess⁡gw≤σg;-C_*g\le\operatorname{Hess}_g w\le\sigma g;
  1. for some ρ≥c∗σR\rho\ge c_*\sigma R,

Hess⁡gw≤−3gon Bg(q,ρ);\operatorname{Hess}_g w\le-3g\quad\text{on }B_g(q,\rho);

4.

∣dw∣g≤C∗σR,0≤w≤C∗σR2.|dw|_g\le C_*\sigma R,\qquad0\le w\le C_*\sigma R^2.

Proof. Let r=dist⁡g(q,⋅)r=\operatorname{dist}_g(q,\mathord\cdot). After decreasing R∗R_*, geodesic polar coordinates are smooth on 0<r<R∗0<r<R_* and

Hess⁡gr=1r(g−dr2)+Er,∣Er∣g≤Cr.(17)\operatorname{Hess}_g r=\frac{1}{r}(g-dr^2)+E_r,\qquad|E_r|_g\le Cr. \tag*{(17)}

for a fixed constant CC. Equivalently, Hess⁡g(r2/2)=g+O(r2)\operatorname{Hess}_g(r^2/2)=g+O(r^2).

We construct a one-variable function FF. Put A=4A=4. Choose smooth functions α:[0,∞)→[0,1]\alpha:[0,\infty)\to[0,1] and γ∈Cc∞((0,1))\gamma\in C^\infty_c((0,1)) such that

α=1 on [0,1],α=0 on [2,∞),γ≥0,∫01γ=1.\alpha=1\text{ on }[0,1],\qquad\alpha=0\text{ on }[2,\infty),\qquad\gamma\ge0,\qquad\int_0^1\gamma=1.

Choose a sufficiently small fixed constant c>0c>0 and put ℓ=cσR\ell=c\sigma R. Define

n(r)=−Aα(r/ℓ),D=−∫0Rn(s) ds,n(r)=-A\alpha(r/\ell),\qquad D=-\int_0^R n(s)\,ds,

and put a positive bump of total mass DD in the interval (3ℓ,R−ℓ)(3\ell,R-\ell):

p(r)=DR−4ℓγ(r−3ℓR−4ℓ)p(r)=\frac{D}{R-4\ell}\gamma\left(\frac{r-3\ell}{R-4\ell}\right)

there, and p=0p=0 elsewhere. Taking cc small, independently of σ\sigma and RR, ensures

0≤p≤σ4,4ℓ<R.0\le p\le\frac{\sigma}{4},\qquad4\ell<R.

Set

q1=n+p,Q(r)=∫0rq1(s) ds.q_1=n+p,\qquad Q(r)=\int_0^r q_1(s)\,ds.

The negative bump occurs before the positive one, the two total masses cancel, and therefore

Q≤0,Q=0 near R,∣Q∣≤C1σR.(18)Q\leq0,\qquad Q=0\text{ near }R,\qquad|Q|\leq C_1\sigma R. \tag*{(18)}

Moreover,

−A≤q1≤σ4,Q(r)r≥−C2A.(19)-A\leq q_1\leq\frac{\sigma}{4},\qquad\frac{Q(r)}{r}\geq-C_2A. \tag*{(19)}

Define

F(r)=−∫rRQ(s) ds.F(r)=-\int_r^R Q(s)\,ds.

Then F≥0F\geq0, F′=QF'=Q, F′′=q1F''=q_1, and F=0F=0 near RR. On 0≤r≤ℓ0\leq r\leq\ell one has

F(r)=F(0)−A2r2.F(r)=F(0)-\frac{A}{2}r^2.

The radial function w=F(r)w=F(r), extended by zero, is smooth at both qq and ∂Bg(q,R)\partial B_g(q,R). For r>0r>0,

Hess⁡gw=q1(r) dr2+Q(r)Hess⁡gr.(20)\operatorname{Hess}_g w=q_1(r)\,dr^2+Q(r)\operatorname{Hess}_g r. \tag*{(20)}

The main tangential term Q(r)r−1(g−dr2)Q(r)r^{-1}(g-dr^2) is nonpositive. The error in (17) has norm at most C∣Q∣r≤CC1σR2C|Q|r\leq CC_1\sigma R^2. Decrease R∗R_* so that this is at most 3σ/43\sigma/4. Equations (19) and (20) then give Hess⁡w≤σg\operatorname{Hess}w\leq\sigma g, while their lower bounds give Hess⁡w≥−C∗g\operatorname{Hess}w\geq-C_*g for a fixed C∗C_*. On r≤ℓr\leq\ell,

Hess⁡gw=−AHess⁡g(r2/2).\operatorname{Hess}_g w=-A\operatorname{Hess}_g(r^2/2).

After one more decrease of R∗R_*, this is at most −3g-3g. Thus one may take ρ=ℓ=cσR\rho=\ell=c\sigma R. Finally, (18) gives

∣dw∣=∣Q∣≤C∗σR,0≤w≤Rmax⁡∣Q∣≤C∗σR2.|dw|=|Q|\leq C_*\sigma R,\qquad0\leq w\leq R\max|Q|\leq C_*\sigma R^2.

Construction before the final pullback

Let N∈S2N\in S^2 be the north pole and set

q=(N,N)∈M.q=(N,N)\in M.

Apply Lemma 4.1 to (M,g0,q)(M,g_0,q), and retain its constants R∗,c∗,C∗R_*,c_*,C_*. Decrease R∗R_* so that Bg0(q,R∗)B_{g_0}(q,R_*) is disjoint from Σδ\Sigma_\delta for every 0<δ≤1/1000<\delta\leq1/100.

Choose a small constant τ0∈(0,1)\tau_0\in(0,1), to be fixed below. Given ε>0\varepsilon>0, put

τ=min⁡{ε,τ0},δ=τ100,a=τδ100(Cβ+1),σ=τ100(C∗+1).(21)\tau=\min\{\varepsilon,\tau_0\},\qquad\delta=\frac{\tau}{100},\qquad a=\frac{\tau\delta}{100(C_\beta+1)},\qquad\sigma=\frac{\tau}{100(C_*+1)}. \tag*{(21)}

Choose R=Rτ≤min⁡{R∗,1}R=R_\tau\leq\min\{R_*,1\} so small that

Vol⁡(Bg0(q,R),g0)Vol⁡(M,g0)<τ100.(22)\frac{\operatorname{Vol}(B_{g_0}(q,R),g_0)}{\operatorname{Vol}(M,g_0)}<\frac{\tau}{100}. \tag*{(22)}

Let w=wσ,Rw=w_{\sigma,R} be given by Lemma 4.1, and let ρ≥c∗σR\rho\geq c_*\sigma R be the radius in part (iii) of that lemma.

Define

f=ua,δ+w,g^τ=e2fg0.(23)f=u_{a,\delta}+w,\qquad\widehat g_\tau=e^{2f}g_0. \tag*{(23)}

Both summands are nonnegative. By decreasing τ0\tau_0 if necessary, the preceding estimates give, uniformly in τ\tau,

0≤f≤110,∣df∣g0≤τ20.(24)0\leq f\leq\frac{1}{10},\qquad|df|_{g_0}\leq\frac{\tau}{20}. \tag*{(24)}

Furthermore, (13), Lemma 4.1, and (21) imply

Hess⁡g0f≤τ50g0(25)\operatorname{Hess}_{g_0} f \le\frac{\tau}{50}g_0 \tag*{(25)}

and

Hess⁡g0f≥−(C∗+1)g0.(26)\operatorname{Hess}_{g_0} f \ge-(C_*+1)g_0. \tag*{(26)}

Proposition 5.1. After choosing τ0\tau_0 sufficiently small, the metrics g^τ\widehat{g}_\tau satisfy the following estimates.

  1. On all of MM,

    −τ<sec⁡g^τ≤Λ,Λ:=2C∗+4.-\tau< \sec_{\widehat{g}_\tau} \le\Lambda,\qquad\Lambda:= 2C_*+4.
  1. On Bg0(q,ρ)B_{g_0}(q,\rho), every sectional curvature is greater than 1.

  1. On M∖(Σδ∪Bg0(q,R))M \setminus\left(\Sigma_\delta\cup B_{g_0}(q,R)\right), every sectional curvature is strictly positive.

Proof. For a g0g_0-orthonormal basis e1,e2e_1,e_2 of a plane, equations (3), (2), (24), and (25) give

e2fsec⁡g^τ(P)≥−2τ50−τ2400>−τ.e^{2f}\sec_{\widehat{g}_\tau}(P)\ge-\frac{2\tau}{50}-\frac{\tau^2}{400}>-\tau.

Since f≥0f\ge0, multiplication by e−2fe^{-2f} cannot make this lower bound more negative. The lower Hessian estimate gives

e2fsec⁡g^τ(P)≤1+2(C∗+1)+τ2400≤2C∗+4,e^{2f}\sec_{\widehat{g}_\tau}(P)\le1+2(C_*+1)+\frac{\tau^2}{400}\le2C_*+4,

after decreasing τ0\tau_0.

On Bg0(q,ρ)B_{g_0}(q,\rho), Lemma 4.1 and (13) give

Hess⁡g0f≤(−3+τ100)g0.\operatorname{Hess}_{g_0} f\le\left(-3+\frac{\tau}{100}\right)g_0.

Thus

e2fsec⁡g^τ(P)≥2(3−τ100)−τ2400.e^{2f}\sec_{\widehat{g}_\tau}(P)\ge2\left(3-\frac{\tau}{100}\right)-\frac{\tau^2}{400}.

Together with f≤1/10f\le1/10, this is greater than 1 for small τ0\tau_0.

Finally, outside Σδ∪Bg0(q,R)\Sigma_\delta\cup B_{g_0}(q,R) one has w=0w=0. Proposition 3.2 applies because

aδ=τ100(Cβ+1)<1,\frac{a}{\delta}=\frac{\tau}{100(C_\beta+1)}<1,

and gives strict positivity there. □

Define the closed set

S^τ=Σδ∪(Bg0(q,R)‾∖Bg0(q,ρ)).(27)\widehat{S}_\tau=\Sigma_\delta\cup\left(\overline{B_{g_0}(q,R)}\setminus B_{g_0}(q,\rho)\right). \tag*{(27)}

Proposition 5.1 shows that

sec⁡g^τ>0on M∖S^τ,(28)\sec_{\widehat{g}_\tau}>0\quad\text{on }M\setminus\widehat{S}_\tau, \tag*{(28)}

and all sectional curvatures on the core Bg0(q,ρ)B_{g_0}(q,\rho) are greater than 1.

By (11), (22), and 0≤f≤1/100\le f\le1/10,

Vol⁡(S^τ,g^τ)Vol⁡(M,g^τ)≤e4/10Vol⁡(S^τ,g0)Vol⁡(M,g0)<e4/10(τ50+τ100)<τ.(29)\begin{aligned} \frac{\operatorname{Vol}(\widehat{S}_\tau,\widehat{g}_\tau)} {\operatorname{Vol}(M,\widehat{g}_\tau)} \le e^{4/10}\frac{\operatorname{Vol}(\widehat{S}_\tau,g_0)} {\operatorname{Vol}(M,g_0)} \\ &<e^{4/10}\left(\frac{\tau}{50}+\frac{\tau}{100}\right)<\tau. \tag*{(29)} \end{aligned}

Uniform geometry and the fixed open ball

The inequalities 0≤f≤1/100 \le f \le1/10 imply

g0≤g^τ≤e1/5g0.(30)g_0 \le\widehat{g}_\tau\le e^{1/5}g_0. \tag*{(30)}

Thus the volume and diameter of g^τ\widehat{g}_\tau have uniform positive lower and upper bounds. We also need a uniform injectivity-radius bound.

Lemma 6.1 (injectivity from first-order control). Let NN be a closed manifold and gjg_j a family of smooth metrics. Assume that

(i) the gjg_j are uniformly bilipschitz to one fixed smooth metric;

(ii) in one fixed finite coordinate atlas, the Christoffel symbols of the gjg_j are uniformly bounded;

(iii) sec⁡gj≤Λ\operatorname{sec}_{g_j} \le\Lambda for one constant Λ\Lambda.

Then inj⁡(N,gj)\operatorname{inj}(N,g_j) has a uniform positive lower bound.

Proof. By the Rauch comparison theorem, the conjugate radius is bounded below by a positive constant depending only on Λ\Lambda. If the injectivity radii tended to zero, the standard injectivity-radius lemma would therefore produce geodesic loops γj:[0,Lj]→N\gamma_j : [0,L_j] \to N with Lj→0L_j \to0.

Uniform bilipschitz control implies that each sufficiently short loop lies in one chart of the fixed finite atlas. Parametrize by gjg_j-arclength and write its coordinates as xj(t)x_j(t). Uniform ellipticity gives constants c,C>0c,C>0 such that

c≤∣x˙j(t)∣≤C.c \le\lvert\dot{x}_j(t)\rvert\le C.

The geodesic equation and the uniform Christoffel bound give ∣x¨j(t)∣≤A\lvert\ddot{x}_j(t)\rvert\le A for one constant AA. Since xj(Lj)=xj(0)x_j(L_j)=x_j(0),

0=∫0Ljx˙j(t) dt=Ljx˙j(0)+∫0Lj(Lj−t)x¨j(t) dt.0=\int_0^{L_j}\dot{x}_j(t)\,\mathrm{d}t =L_j\dot{x}_j(0)+\int_0^{L_j}(L_j-t)\ddot{x}_j(t)\,\mathrm{d}t.

Hence

∣x˙j(0)∣≤A2Lj,\lvert\dot{x}_j(0)\rvert\le\frac{A}{2}L_j,

contradicting the lower bound cc. □

For a conformal metric e2fg0e^{2f}g_0, the Levi-Civita connections are related by

∇~XY=∇X0Y+df(X)Y+df(Y)X−g0(X,Y)∇0f.(31)\widetilde{\nabla}_X Y =\nabla^0_XY+df(X)Y+df(Y)X-g_0(X,Y)\nabla^0f. \tag*{(31)}

Thus (24) gives a uniform Christoffel bound in a fixed atlas. Lemma 6.1, Proposition 5.1, and (30) therefore give

inj⁡(M,g^τ)≥i0>0(32)\operatorname{inj}(M,\widehat{g}_\tau) \ge i_0>0 \tag*{(32)}

for one constant i0i_0 independent of τ\tau.

It remains to place the uniformly positively curved core over one fixed open ball. Choose once and for all a small g0g_0-geodesic ball

U=Bg0(q,rU)U=B_{g_0}(q,r_U)

whose closure is contained in a normal coordinate neighborhood of qq. For every τ\tau, there is a diffeomorphism

Fτ:M⟶MF_\tau:M\longrightarrow M

such that

Fτ(U)⊂Bg0(q,ρ/2).(33)F_\tau(U)\subset B_{g_0}(q,\rho/2). \tag*{(33)}

Indeed, in polar coordinates about qq, use a smooth increasing radial map that sends [0,rU][0,r_U] into [0,ρ/2][0,\rho/2] and agrees with the identity outside a slightly larger fixed coordinate ball.

For the original parameter ε\varepsilon, with τ\tau as in (21), set

gε=Fτ∗g^τ,Sε=Fτ−1(S^τ).(34)g_\varepsilon=F_\tau^*\widehat{g}_\tau,\qquad S_\varepsilon=F_\tau^{-1}(\widehat{S}_\tau). \tag*{(34)}

The map

Fτ:(M,gε)⟶(M,g^τ)F_{\tau} : (M, g_{\varepsilon}) \longrightarrow(M, \widehat{g}_{\tau})

is an isometry. Therefore all sectional-curvature bounds, relative volumes, total volumes, diameters, and injectivity radii are unchanged. Equations (33) and (27) also give U∩Sε=∅U \cap S_{\varepsilon} = \varnothing, while Proposition 5.1 gives sec⁡gε>1\operatorname{sec}_{g_{\varepsilon}} > 1 on UU.

Taking, for example,

v−=Vol⁡(M,g0),v+=e2/5Vol⁡(M,g0),v_{-} = \operatorname{Vol}(M,g_{0}), \qquad v_{+} = e^{2/5}\operatorname{Vol}(M,g_{0}),
d−=diam⁡(M,g0),d+=e1/10diam⁡(M,g0),d_{-} = \operatorname{diam}(M,g_{0}), \qquad d_{+} = e^{1/10}\operatorname{diam}(M,g_{0}),

and using (29), (32), and Proposition 5.1, proves Theorem 1.1. Notice that when ε≥τ0\varepsilon\geq\tau_{0}, one has τ=τ0\tau= \tau_{0}; then −τ0>−ε-\tau_{0} > -\varepsilon and the relative-volume estimate <τ0≤ε< \tau_{0} \leq\varepsilon still gives the required assertions.

Why this does not produce a quasi-positively curved limit

The construction exploits two freedoms that are absent from a genuine compactness argument. First, the set Σδ\Sigma_{\delta} is thin in volume but spread across the manifold. It intersects the closed directions that obstruct the existence of a globally strictly concave function, while its relative volume tends to zero. The positive curvature created outside it is very small, of order aδa\delta.

Second, the region on which all sectional curvatures exceed 1 has intrinsic size tending to zero before the final pullback. The diffeomorphism FτF_{\tau} places the fixed topological ball UU over that shrinking core without changing any intrinsic global bound. Consequently the final metrics need not be precompact in C2C^{2} in the fixed differentiable gauge. The family therefore does not yield a quasi-positively curved limit metric on S2×S2S^{2} \times S^{2}.

References

References

  1. [1]S. Brendle, P-K. Hung, A Metric on S^2 × S^2 with Positive Sectional Curvature, arXiv:2608.19068v1, 2026.arxiv.org/abs/2608.19068
  2. [2]W. Klingenberg, Riemannian Geometry, second edition, de Gruyter, 1995.
  3. [3]P. Petersen, Riemannian Geometry, third edition, Graduate Texts in Mathematics 171, Springer, 2016.

Paper details

vlnc:W_799knrc3
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