ScalingStacks

3.1. Normal coordinates for an embedded submanifold [04ZD]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

3.1. Normal coordinates for an embedded submanifold

We start our discussion by introducing the basic notions of the normal exponential map and normal coordinates with respect to an embedded submanifold. This part seems to be standard in Riemannian geometry. For the consistence of the notations and the completeness of the paper, here we include detailed discussions and proofs.

Let (Q,g)(Q,g) be an oriented Riemannian manifold of dimension mm and let P⊂QP\subset Q be a closed embedded oriented submanifold of codimension k0k_{0} in QQ. In our later applications, we only need the case k0=3k_{0}=3. Denote by N{N} the normal bundle of PP in QQ, equipped with the induced fiberwise Riemannian inner products. For any p∈Pp\in P, N⁡(p)N(p) denotes the fiber of pp in NN. The normal exponential map of PP in QQ is defined as

(3.1) ExpP:N→Q,(p,v)↦Expp⁡(v),v∈N⁡(p),\Exp_{P}:{N}\rightarrow Q,\ (p,v)\mapsto\Exp_{p}(v),\ v\in N(p),

where Expp:Tp​Q→Q\Exp_{p}:T_{p}Q\to Q is the standard exponential map at p∈Qp\in Q. By standard implicit function theorem, it is straightforward that ExpP:N→Q\Exp_{P}:{N}\to Q is a diffeomorphism from some neighborhood of the zero section of N{N} to some tubular neighborhood of PP in QQ.

Now we define the normal coordinates. Fix p∈Pp\in P. First we choose local coordinates {x1′,…,xm−k0′}\{x_{1}^{\prime},\ldots,x_{m-k_{0}}^{\prime}\} in a small neighborhood U⊂PU\subset P of pp such that

(3.2) ⟨∂xi′,∂xj′⟩=δi​j, 1≤i,j≤m−k0,\langle\partial_{x_{i}^{\prime}},\partial_{x_{j}^{\prime}}\rangle=\delta_{ij},\ 1\leq i,j\leq m-k_{0},

at pp. We also assume that ∂x1′∧⋯∧∂xm−k0′\partial_{x_{1}^{\prime}}\wedge\cdots\wedge\partial_{x_{m-k_{0}}^{\prime}} is compatible with the orientation on UU. Next, we pick local orthonormal sections {e1,…,ek0}\{e_{1},\ldots,e_{k_{0}}\} of the normal bundle NN such that

(3.3) ⟨eα,eβ⟩=δα​β, 1≤α,β≤k0,\langle e_{\alpha},e_{\beta}\rangle=\delta_{\alpha\beta},\ 1\leq\alpha,\beta\leq k_{0},

on UU. Again we assume that e1∧⋯∧ek0e_{1}\wedge\cdots\wedge e_{k_{0}} is compatible with the orientation on NN, i.e. e1∧⋯∧ek0∧∂x1′∧⋯∂xk−m0′e_{1}\wedge\cdots\wedge e_{k_{0}}\wedge\partial_{x_{1}^{\prime}}\wedge\cdots\partial_{x_{k-m_{0}}^{\prime}} is compatible with the orientation on QQ. Then we can find ϵ>0\epsilon>0 such that Expp\Exp_{p} restricts to a diffeomorphism from

(3.4) 𝔖0={(q,v)∈N|q∈U,|v|<ϵ}\mathfrak{S}_{0}=\{(q,v)\in N|q\in U,\ |v|<\epsilon\}

to a neighborhood 𝒰\mathcal{U} of pp. In particular, U=𝒰∩PU=\mathcal{U}\cap P.

Definition 3.1 (Normal coordinates).

For any (q,v)∈𝔖0(q,v)\in\mathfrak{S}_{0} with v=∑α=1k0vα​eαv=\sum\limits_{\alpha=1}^{k_{0}}v_{\alpha}e_{\alpha}, the local normal coordinates are defined as follows

(3.5) {xj​(ExpP⁡(q,v))≡xj′​(q),1≤j≤m−k0,yβ​(ExpP⁡(q,v))≡vβ,1≤β≤k0.\displaystyle\begin{cases}x_{j}(\Exp_{P}(q,v))\equiv x_{j}^{\prime}(q),&1\leq j\leq m-k_{0},\\ y_{\beta}(\Exp_{P}(q,v))\equiv v_{\beta},&1\leq\beta\leq k_{0}.\end{cases}

By definition for each fixed point (y1,…,yk0,x1,…,xm−k0)(y_{1},\ldots,y_{k_{0}},x_{1},\ldots,x_{m-k_{0}}) in 𝒰\mathcal{U}, the curve

(3.6) ϑ⁡(t)≡(t​y1,…,t​yk0,x1,…,xm−k0),\vartheta(t)\equiv(ty_{1},\ldots,ty_{k_{0}},x_{1},\ldots,x_{m-k_{0}}),

is a normal geodesic which is orthogonal to U⊂PU\subset P. Let

(3.7) ∂y1,…,∂yk0,∂x1,…,∂xm−k0\partial_{y_{1}},\ldots,\partial_{y_{k_{0}}},\partial_{x_{1}},\ldots,\partial_{x_{m-k_{0}}}

be the induced coordinate vector fields. Then ∂yα|y=0=eα\partial_{y_{\alpha}}|_{y=0}=e_{\alpha} when both are viewed as sections of NN over U⊂PU\subset P. For 1≤i,j≤m−k01\leq i,j\leq m-k_{0} and 1≤α≤k01\leq\alpha\leq k_{0}, we denote

(3.8) gi​j≡⟨∂xi,∂xj⟩,gi​α≡⟨∂xi,∂yα⟩,gα​β≡⟨∂yα,∂yβ⟩.g_{ij}\equiv\langle\partial_{x_{i}},\partial_{x_{j}}\rangle,\ g_{i\alpha}\equiv\langle\partial_{x_{i}},\partial_{y_{\alpha}}\rangle,\ g_{\alpha\beta}\equiv\langle\partial_{y_{\alpha}},\partial_{y_{\beta}}\rangle.

By definition, we have

(3.9) gi​j​(0,0)=δi​j,gα​β​(x,0)=δα​β,gi​α​(x,0)=0,g_{ij}(0,0)=\delta_{ij},\ g_{\alpha\beta}(x,0)=\delta_{\alpha\beta},\ g_{i\alpha}(x,0)=0,

for all x∈U⊂Px\in U\subset P. The second fundamental form of the embedding U↪𝒰U\hookrightarrow\mathcal{U} can be written as II=IIi​jα∂yα⊗(dxi⊗dxj)\IIs=\IIs_{ij}^{\alpha}\partial_{y_{\alpha}}\otimes(dx_{i}\otimes dx_{j}), where

(3.10) IIi​jα≡⟨∇∂xieα,∂xj⟩\IIs^{\alpha}_{ij}\equiv\langle\nabla_{\partial_{x_{i}}}e_{\alpha},\partial_{x_{j}}\rangle

Denote by H→=Hα​eα\overrightarrow{H}=H^{\alpha}e_{\alpha} the mean curvature vector, then

(3.11) Hα≡gi​j​IIi​jα.H^{\alpha}\equiv g^{ij}\IIs^{\alpha}_{ij}.

In the above coordinates, we define the normal distance function

(3.12) r≡|y|=(∑α=1k0yα2)12.r\equiv|y|=\Big(\sum\limits_{\alpha=1}^{k_{0}}y_{\alpha}^{2}\Big)^{\frac{1}{2}}.

A straightforward extension of the usual Gauss Lemma gives the following and we omit the proof.

Lemma 3.2 (Generalized Gauss Lemma).

For any p∈Pp\in P, there is some sufficiently small neighborhood U⊂PU\subset P of pp and a tubular neighborhood 𝒰⊂Q\mathcal{U}\subset Q with U=𝒰∩PU=\mathcal{U}\cap P such that the function rr defined by (3.12) satisfies the following properties:

  1. (1)

    ∇r=∂r\nabla r=\partial_{r} holds in 𝒰∖U\mathcal{U}\setminus U. In particular, rr is the normal distance function in 𝒰\mathcal{U}, i.e., r⁡(q)=d⁡(q,P)r(q)=d(q,P) for all q∈𝒰q\in\mathcal{U}.

  2. (2)

    ∂r\partial_{r} is orthogonal to ∂xi\partial_{x_{i}}’s in 𝒰⊂Q\mathcal{U}\subset Q, and hence

    (3.13) ⟨∂xi,r∂r⟩=⟨∂xi,∑α=1k0yα∂yα⟩=∑α=1k0yαgi​α=0, 1≤i≤m−k0.\langle\partial_{x_{i}},r\partial_{r}\rangle=\langle\partial_{x_{i}},\sum_{\alpha=1}^{k_{0}}y_{\alpha}\partial_{y_{\alpha}}\rangle=\sum_{\alpha=1}^{k_{0}}y_{\alpha}g_{i\alpha}=0,\ 1\leq i\leq m-k_{0}.

For the convenience of later discussion, we introduce several notations concerning the normal regularity order near the submanifold PP. It will be frequently used throughout the paper.

Definition 3.3 (Normal regularity order).

Let T⁡(x,y)T(x,y) be a tensor locally defined in 𝒰\mathcal{U} which is C∞C^{\infty} on 𝒰∖U\mathcal{U}\setminus U, then for a non-negative integer kk we say as r→0r\rightarrow 0

  1. (1)

    T⁡(x,y)=O′​(rk)T(x,y)=O^{\prime}(r^{k}) if for each ϵ>0\epsilon>0

    (3.14) |∂xI∂yJT⁡(x,y)|={O⁡(r−ϵ),|J|≤k,O⁡(rk−|J|−ϵ),|J|>k,\displaystyle\Big|\partial_{x}^{I}\partial_{y}^{J}T(x,y)\Big|=\begin{cases}O(r^{-\epsilon}),&|J|\leq k,\\ O(r^{k-|J|-\epsilon}),&|J|>k,\end{cases}

    for all multi-indices II and JJ. In particular, if T∈C∞​(𝒰)T\in C^{\infty}(\mathcal{U}), then T=O′​(rk)T=O^{\prime}(r^{k}) for all k∈ℤk\in\mathbb{Z}.

  2. (2)

    T⁡(x,y)=rk​O′​(1)T(x,y)=r^{k}O^{\prime}(1) if r−k​T​(x,y)=O′​(1)r^{-k}T(x,y)=O^{\prime}(1).

  3. (3)

    T​(x,y)=O~​(rk)T(x,y)=\widetilde{O}(r^{k}) if T⁡(x,y)∈C∞​(𝒰)T(x,y)\in C^{\infty}(\mathcal{U}) and

    (3.15) T⁡(x,y)=rk​O′​(1).T(x,y)=r^{k}O^{\prime}(1).

    In other words, T⁡(x,y)T(x,y) is smooth in 𝒰\mathcal{U} and has vanishing normal derivatives along UU up to order k−1k-1.

Notice that the defining condition does not depend on the choice of the local coordinates, since if we have another coordinate system {y~1,⋯,y~k0,x~1,⋯,x~m−k0}\{\tilde{y}_{1},\cdots,\tilde{y}_{k_{0}},\tilde{x}_{1},\cdots,\tilde{x}_{m-k_{0}}\}, then we have

(3.16) ∂y~α=∂yβ∂y~α⋅∂yβ+∂xl∂y~α⋅∂xl,\displaystyle\partial_{\tilde{y}_{\alpha}}=\frac{\partial y_{\beta}}{\partial\tilde{y}_{\alpha}}\cdot\partial_{y_{\beta}}+\frac{\partial x_{l}}{\partial\tilde{y}_{\alpha}}\cdot\partial_{x_{l}},
(3.17) ∂x~j=∂xl∂x~j⋅∂xl+∂yα∂x~j⋅∂yα,\displaystyle\partial_{\tilde{x}_{j}}=\frac{\partial x_{l}}{\partial\tilde{x}_{j}}\cdot\partial_{x_{l}}+\frac{\partial y_{\alpha}}{\partial\tilde{x}_{j}}\cdot\partial_{y_{\alpha}},

and ∂yα∂x~j∈r​O′​(1)\frac{\partial y_{\alpha}}{\partial\tilde{x}_{j}}\in rO^{\prime}(1). Similarly, we can also use any local coordinate system {yα,xi}\{y_{\alpha},x_{i}\} such that yα=0y_{\alpha}=0 along PP for 1≤α≤k01\leq\alpha\leq k_{0}.

Lemma 3.4.

In the above normal coordinates, we have the following expansions of the metric tensor gg of QQ along the normal directions,

(3.18) gα​β|(x,y)\displaystyle g_{\alpha\beta}|_{(x,y)} =δα​β−13​Rmα​γ​ξ​β|(x,0)​yγ​yξ+O~​(r3),\displaystyle=\delta_{\alpha\beta}-\frac{1}{3}\Rm_{\alpha\gamma\xi\beta}\Big|_{(x,0)}y_{\gamma}y_{\xi}+\widetilde{O}(r^{3}),
(3.19) gi​j|(x,y)\displaystyle g_{ij}|_{(x,y)} =gi​jP(x)+2IIi​jα|(x,0)yα−(Rmi​γ​ξ​j+⟨∇∂xi∂yγ,∇∂xj∂yξ⟩)|(x,0)yγyξ+O~(r3),\displaystyle=g^{P}_{ij}(x)+2\IIs^{\alpha}_{ij}\Big|_{(x,0)}y_{\alpha}-(\Rm_{i\gamma\xi j}+\langle\nabla_{\partial_{x_{i}}}\partial_{y_{\gamma}},\nabla_{\partial_{x_{j}}}\partial_{y_{\xi}}\rangle)\Big|_{(x,0)}y_{\gamma}y_{\xi}+\widetilde{O}(r^{3}),
(3.20) gi​α|(x,y)\displaystyle g_{i\alpha}|_{(x,y)} =⟨∇∂xi∂yγ,∂yα⟩|(x,0)yγ−23Rmi​γ​ξ​α|(x,0)yγyξ+O~(r3),\displaystyle=\langle\nabla_{\partial_{x_{i}}}\partial_{y_{\gamma}},\partial_{y_{\alpha}}\rangle\Big|_{(x,0)}y_{\gamma}-\frac{2}{3}\Rm_{i\gamma\xi\alpha}\Big|_{(x,0)}y_{\gamma}y_{\xi}+\widetilde{O}(r^{3}),

where gP=(gi​jP)g^{P}=(g^{P}_{ij}) denotes the restriction of the metric gg to U⊂PU\subset P, Rm\Rm denotes the Riemann curvature tensor of gg.

Proof.

The above expansions can be proved using the Jacobi fields. Fix a point q=(0k0,x1,…,xm−k0)∈U⊂Pq=(0^{k_{0}},x_{1},\ldots,x_{m-k_{0}})\in U\subset P, we we choose a unit vector v=∑α=1m−k0vα∂yα∈N(q)≅ℝk0v=\sum\limits_{\alpha=1}^{m-k_{0}}v_{\alpha}\partial_{y_{\alpha}}\in N(q)\cong\mathbb{R}^{k_{0}} with |v|=1|v|=1. Let ϑ\vartheta be the following radial geodesic in 𝒰\mathcal{U},

(3.21) ϑ⁡(t)=ExpP⁡(q,t​v)≡Expq⁡(t​v)\vartheta(t)=\Exp_{P}(q,tv)\equiv\Exp_{q}(tv)

such that ϑ′​(0)=v\vartheta^{\prime}(0)=v. In the normal coordinates, the geodesic ϑ\vartheta can represented as ϑ⁡(t)=(t​v1,…,t​vk0,x1,…,xm−k0)\vartheta(t)=(tv_{1},\ldots,tv_{k_{0}},x_{1},\ldots,x_{m-k_{0}}).

For each 1≤α≤k01\leq\alpha\leq k_{0} and 1≤i≤m−k01\leq i\leq m-k_{0}, we define the geodesic variations

(3.22) σα​(t,s)\displaystyle\sigma_{\alpha}(t,s) ≡((t​v1,…,t⁡(vα+s),…,t​vk0,x1,…,xm−k0)CLOSE,\displaystyle\equiv((tv_{1},\ldots,t(v_{\alpha}+s),\ldots,tv_{k_{0}},x_{1},\ldots,x_{m-k_{0}}),
(3.23) σi​(t,s)\displaystyle\sigma_{i}(t,s) ≡(t​v1,…,t​vk0,x1,…,xi+s,…​xm−k0).\displaystyle\equiv(tv_{1},\ldots,tv_{k_{0}},x_{1},\ldots,x_{i}+s,\ldots x_{m-k_{0}}).

Then variation fields of σα​(t,s)\sigma_{\alpha}(t,s) and σi​(t,s)\sigma_{i}(t,s) give the following Jacobi fields along the radial geodesic ϑ⁡(t)\vartheta(t) respectively:

(3.24) {Jα(t)=t⋅∂yα,1≤α≤k0Ji(t)=∂xi,1≤i≤m−k0.\displaystyle\begin{cases}J_{\alpha}(t)=t\cdot\partial_{y_{\alpha}},&1\leq\alpha\leq k_{0}\\ J_{i}(t)=\partial_{x_{i}},&1\leq i\leq m-k_{0}.\end{cases}

By definition,

(3.25) Jα(0)=0,Ji(0)=∂xi.J_{\alpha}(0)=0,\ J_{i}(0)=\partial_{x_{i}}.

Taking first derivatives at t=0t=0,

(3.26) Jα′(0)=∂yα,Ji′(0)=vα∇∂xi∂yα.J_{\alpha}^{\prime}(0)=\partial_{y_{\alpha}},\ J_{i}^{\prime}(0)=v_{\alpha}\nabla_{\partial_{x_{i}}}\partial_{y_{\alpha}}.

Then applying the Jacobi equation along the geodesic ϑ\vartheta,

(3.27) {Jα′′+Rm⁡(Jα,ϑ′)​ϑ′=0,Ji′′+Rm⁡(Ji,ϑ′)​ϑ′=0,\displaystyle\begin{cases}J_{\alpha}^{\prime\prime}+\Rm(J_{\alpha},\vartheta^{\prime})\vartheta^{\prime}=0,\\ J_{i}^{\prime\prime}+\Rm(J_{i},\vartheta^{\prime})\vartheta^{\prime}=0,\\ \end{cases}

where Rm⁡(X,Y)​Z≡∇X∇Y​Z−∇Y∇X​Z−∇[X,Y]Z\Rm(X,Y)Z\equiv\nabla_{X}\nabla_{Y}Z-\nabla_{Y}\nabla_{X}Z-\nabla_{[X,Y]}Z denotes the Riemann curvature tensor of gg, so it follows that

(3.28) Jα′′(0)=0,Jα′′′(0)=−vγvξRm(∂yα,∂yγ)∂yξ,\displaystyle J^{\prime\prime}_{\alpha}(0)=0,\ J^{\prime\prime\prime}_{\alpha}(0)=-v_{\gamma}v_{\xi}\Rm(\partial_{y_{\alpha}},\partial_{y_{\gamma}})\partial_{y_{\xi}},
(3.29) Ji′′(0)=−vγvξRm(∂xi,∂yγ)∂yξ.\displaystyle J^{\prime\prime}_{i}(0)=-v_{\gamma}v_{\xi}\Rm(\partial_{x_{i}},\partial_{y_{\gamma}})\partial_{y_{\xi}}.

Therefore,

(3.30) gα​β\displaystyle g_{\alpha\beta} =t−2​g​(Jα,Jβ)=δα​β−13​Rmα​γ​ξ​β​vγ​vξ​t2+O~​(t3),\displaystyle=t^{-2}g(J_{\alpha},J_{\beta})=\delta_{\alpha\beta}-\frac{1}{3}\Rm_{\alpha\gamma\xi\beta}v_{\gamma}v_{\xi}t^{2}+\widetilde{O}(t^{3}),
(3.31) gi​j\displaystyle g_{ij} =g(Ji,Jj)=gi​jP+2vαIIi​jαt−(Rmi​γ​ξ​j+⟨∇∂xi∂yγ,∇∂xj∂yξ⟩)vγvξt2+O~(t3),\displaystyle=g(J_{i},J_{j})=g^{P}_{ij}+2v_{\alpha}\IIs_{ij}^{\alpha}t-\Big(\Rm_{i\gamma\xi j}+\langle\nabla_{\partial_{x_{i}}}\partial_{y_{\gamma}},\nabla_{\partial_{x_{j}}}\partial_{y_{\xi}}\rangle\Big)v_{\gamma}v_{\xi}t^{2}+\widetilde{O}(t^{3}),
(3.32) gi​α\displaystyle g_{i\alpha} =t−1g(Ji,Jα)=⟨∇∂xi∂yγ,∂yα⟩vγt−23Rmi​γ​ξ​αvγvξt2+O~(t3).\displaystyle=t^{-1}g(J_{i},J_{\alpha})=\langle\nabla_{\partial_{x_{i}}}\partial_{y_{\gamma}},\partial_{y_{\alpha}}\rangle v_{\gamma}t-\frac{2}{3}\Rm_{i\gamma\xi\alpha}v_{\gamma}v_{\xi}t^{2}+\widetilde{O}(t^{3}).

Let yα=t​vαy_{\alpha}=tv_{\alpha}, then we obtain the desired expansions. ∎

As a digression we briefly discuss the intrinsic meaning of the above expansion. The point is that locally the Riemannian metric gg is approximated by a Riemannian metric gNg_{N} on the normal bundle NN up to the first order. Notice we have the natural projection π:N→P\pi:N\rightarrow P and NN is a Riemannian vector bundle together with an induced “normal” connection, given by the normal component of the Levi-Civita connection of gg. The latter hence gives rise to a distribution of horizontal subspaces at each point of NN, which in our coordinates is spanned by ∂xi−Ai​α​βyβ∂yα\partial_{x_{i}}-A_{i\alpha\beta}y_{\beta}\partial_{y_{\alpha}}, where

(3.33) Ai​α​β(x)≡⟨∇∂xi∂yβ,∂yα⟩|y=0A_{i\alpha\beta}(x)\equiv\langle\nabla_{\partial_{x_{i}}}\partial_{y_{\beta}},\partial_{y_{\alpha}}\rangle|_{y=0}

is a smooth function on UU. We define gNg_{N} so that at each point of NN, the vertical and horizontal subspaces are orthogonal and on the vertical part is given by the bundle metric on NN, and on the horizontal part is given by the perturbation of the base metric gPg_{P} using the second fundamental form. In this way we get a coordinate free description of the above expansion up to the first order.

We also define

(3.34) Ai​j​α​β≡12​(∂xiAj​α​β−∂xjAi​α​β).A_{ij\alpha\beta}\equiv\frac{1}{2}(\partial_{x_{i}}A_{j\alpha\beta}-\partial_{x_{j}}A_{i\alpha\beta}).

Then the curvature of the normal connection is given by

(3.35) Ωi​j​α​β≡Ai​j​α​β+12​(Ai​α​γ​Aj​γ​β−Ai​β​γ​Aj​γ​α).\Omega_{ij\alpha\beta}\equiv A_{ij\alpha\beta}+\frac{1}{2}(A_{i\alpha\gamma}A_{j\gamma\beta}-A_{i\beta\gamma}A_{j\gamma\alpha}).

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.