Bochner公式给出了黎曼流形上函数的Laplace与曲率之间的关系。
$$
\Delta v=\mathrm{Ric}(\nabla u,\nabla u)+\langle\nabla u,\nabla\Delta u\rangle+|\mathrm{Hess}_u|^2.
$$
\begin{align*}
\mathrm{Hess}_u(X,Y)&=(D^2u)(X,Y)=(D_X(\nabla u))(Y)\\
&=D_XD_Yu-\nabla_{\nabla_XY}u\\
&=D_X\langle\nabla u,Y\rangle-\langle D_XY,\nabla u\rangle\\
&=\langle D_X\nabla u, Y \rangle.
\end{align*}
下面, 我们说明$\mathrm{Hess}u(X,Y)$关于$X,Y$是对称的。
\begin{align*}
D^2u(X,Y)-D^2u(Y,X)&=\langle D_X\nabla u,Y \rangle\langle D_Y\nabla u,X \rangle\\
&=D_X\langle \nabla u,Y \rangle-D_Y\langle\nabla u,X\rangle-\langle \nabla u,D_XY-D_YX \rangle\\
&=X(Yu)-Y(Xu)-[X,Y]u\\
&=0,
\end{align*}
这里, 我们用到了黎曼联络的度量相容性与无挠性。
令$v=\frac{1}{2}|\nabla u|^2$, 则(假设$\left\{ e_i \right\}_{i=1}^{\dim M}$是一个幺正标架)
\begin{align*}
\langle \nabla v,e_j \rangle&\nabla_{e_j}v=\langle \nabla_{e_j}\nabla u,\nabla u \rangle\\
&=\mathrm{Hess}_u(e_j,\nabla u)\\
&=\mathrm{Hess}_u(\nabla u,e_j)\\
&=\langle \nabla_{\nabla u}\nabla u,e_j \rangle,
\end{align*}
可见
$$
\nabla v=\nabla_{\nabla u}\nabla u.
$$
我们按照Petersen书上(参考[1,p.33,38])的曲率符号约定, 计算得到
\begin{align*}
\Delta v&=\mathrm{Hess}_v(e_i,e_i)=\langle \nabla_{e_i}\nabla v,e_i \rangle\\
&=\langle \nabla_{e_i}\nabla_{\nabla u} \nabla u,e_i\rangle\\
&=\langle R(e_i,\nabla u)\nabla u,e_i\rangle+\langle \nabla_{\nabla u}\nabla_{e_i} \nabla u,e_i\rangle+\langle \nabla_{[e_i,\nabla u]}\nabla u,e_i\rangle\\
&=\mathrm{Ric}(\nabla u,\nabla u)+\nabla u\langle \nabla_{e_i}\nabla u,e_i \rangle -\langle \nabla_{e_i} \nabla u, \nabla_{\nabla u}e_i \rangle+\langle \nabla_{[e_i,\nabla u]}\nabla u,e_i\rangle.
\end{align*}
现在, 注意到
\begin{align*}
\nabla u\langle \nabla_{e_i}\nabla u,e_i \rangle&=\langle\nabla u,\nabla\Delta u\rangle,\\
\langle \nabla_{[e_i,\nabla u]}\nabla u,e_i\rangle&=\mathrm{Hess}_u([e_i,\nabla u],e_i)=\langle \nabla_{e_i}\nabla u,[e_i,\nabla u] \rangle\\
&=\langle\nabla_{e_i}\nabla u,\nabla_{e_i}\nabla u-\nabla_{\nabla u}e_i\rangle\\
&=\left\langle\langle\nabla_{e_i}\nabla u,e_j\rangle e_j,\langle\nabla_{e_i}\nabla u,e_k\rangle e_k\right\rangle-\langle\nabla_{e_i}\nabla u,\nabla_{\nabla u}e_i\rangle\\
&=|\mathrm{Hess}_u|^2-\langle\nabla_{e_i}\nabla u,\nabla_{\nabla u}e_i\rangle.
\end{align*}
因此
\begin{equation}\label{eq:Bochner-temp}
\Delta v=\mathrm{Ric}(\nabla u,\nabla u)+\langle\nabla u,\nabla\Delta u\rangle+|\mathrm{Hess}_u|^2-2\langle\nabla_{e_i}\nabla u,\nabla_{\nabla u}e_i\rangle.
\end{equation}
下面, 我们将证明
$$
\langle\nabla_{e_i}\nabla u,\nabla_{\nabla u}e_i\rangle=0.
$$
一种简单的方式是, 注意到, 它不依赖于坐标的选取(由\eqref{eq:Bochner-temp})。取法坐标$\{e_i\}$, 使得$\nabla_{e_i}e_j=0$ 在给某给定点$p$处成立。若设$\nabla u=\langle\nabla u,e_i\rangle e_i=(e_iu)e_i:=u_ie_i$, 则在$p$点处
$$
\nabla_{\nabla u}e_i=\nabla_{u_je_j}e_i=u_j\nabla_{e_j}e_i=0.
$$
另一种, 不利用法坐标的办法是:假设
$$
\nabla_{e_i}\nabla u=a_{ij}e_j,\quad
\nabla_{\nabla u}e_i=b_{ij}e_j.
$$
则
$$
\langle\nabla_{e_i}\nabla u,\nabla_{\nabla u}e_i\rangle=a_{ij}b_{ij}.
$$
注意到,
\begin{align*}
a_{ij}&=\langle\nabla_{e_i}\nabla u,e_j\rangle=\mathrm{Hess}_u(e_i,e_j)=a_{ji}\\
b_{ij}&=\langle\nabla_{\nabla u}e_i,e_j\rangle=\nabla u\langle e_i,e_j\rangle-\langle e_i,\nabla_{\nabla u}e_j\rangle\\
&=-\langle e_i, b_{jk}e_k\rangle=-b_{ji}.
\end{align*}
于是
$$
\sum_{i,j}a_{ij}b_{ij}=\sum_{j,i}a_{ji}b_{ji}=-\sum_{j,i}a_{ij}b_{ij}.
$$
由此即得结论。
最后, 作为注记, 我们利用局部分量来计算, 和前面一样假设$\set{e_i}$是$p$的一个邻域上的法坐标系。定义
\begin{align*}
\nabla u&=\inner{\nabla u,e_i}e_i:=u_ie_i,\\
D^2u(e_i,e_j)&=\inner{\nabla_{e_i}\nabla u,e_j}=\inner{\nabla_{e_i}(u_ke_k),e_j}\\
&=e_iu_k\delta_{kj}=e_iu_j:=u_{ji}=u_{ij},\\
\Delta u&=u_{ii},\quad\nabla\Delta u=\nabla u_{ii}=u_{iik}e_k,\quad\Delta u_k=u_{kii},\\
\nabla_{e_i}\nabla u&=\nabla_{e_i}(u_ke_k)=u_{ki}e_k,\\
\nabla_{\nabla u}e_i&=\nabla_{u_ke_k}e_i=0,\\
[e_i,\nabla u]&=\nabla_{e_i}\nabla u-\nabla_{\nabla u}e_i=u_{ki}e_k,\\
\nabla_{[e_i,\nabla u]}\nabla u&=\nabla_{u_{ki}e_k}(u_le_l)=u_{ki}u_{lk}e_l,\\
\nabla_{\nabla u}\nabla u&=\nabla_{u_ke_k}(u_le_l)=u_ku_{lk}e_l,\\
\nabla_{\nabla u}(u_{ki}e_k)&=\nabla_{\nabla u}u_{ki}e_k=u_lu_{kil}e_k,\\
\langle R(e_i,\nabla u)\nabla u,e_j\rangle&=\langle ([\nabla_{e_i},\nabla_{\nabla u}]-\nabla_{[e_i,\nabla u]})\nabla u,e_j\rangle\\
&=\langle \nabla_{e_i}(u_ku_{lk}e_l)-\nabla_{\nabla u}(u_{ki}e_k)-u_{ki}u_{lk}e_l,e_j\rangle\\
&=\langle (u_{ki}u_{lk}+u_ku_{lki})e_l-u_lu_{kil}e_k-u_{ki}u_{lk}e_l,e_j\rangle\\
&=u_ku_{lki}\delta_{lj}-u_lu_{kil}\delta_{kj}\\
&=u_ku_{jki}-u_lu_{jil}\\
&=u_k(u_{jki}-u_{jik})\\
\mathrm{Ric}(\nabla u,\nabla u)&=\mathrm{Ric}(u_ie_i,u_je_j)=u_iu_j\mathrm{Ric}(e_i,e_j):=R_{ij}u_iu_j\\
\end{align*}
可见
$$
\langle R(e_i,\nabla u)\nabla u,e_i\rangle=\mathrm{Ric}(\nabla u,\nabla u)=
R_{ij}u_iu_j=u_j(u_{iji}-u_{iij})=u_j(u_{jii}-u_{iij})=u_ju_{jii}-\inner{\nabla u,\nabla\Delta u}
$$
则
\begin{align*}
v&=\frac{1}{2}|\nabla u|^2=\frac{1}{2}u_k^2\\
\Delta v&=D^2v(e_i,e_i)=v_{ii}=u_{ki}u_{ki}+u_ku_{kii}\\
&=|\mathrm{Hess u}|^2+\mathrm{Ric}(\nabla u,\nabla u)+\inner{\nabla u,\nabla\Delta u}.
\end{align*}
$$
u_{iki}-u_{iik}=u_{kii}-u_{iik}=R_{ik}u_i.
$$
- Riemannian geometry, Second, Graduate Texts in Mathematics, Springer, New York, vol. 171, 2006. xvi+401. MR2243772 ,