张量基础起步

规定记号，上下标同时出现表示求和，即
$$\sum_{i} a_ib^i = a_ib^i.$$
$$\sum_{i} \frac{\partial x^j}{\partial y^i}b^i = \frac{\partial x^j}{\partial y^i}b^i.$$
规定，偏导数记号的分子为上标，分母为下标.

$n$维线性空间$V$，任意两组基$\{e_i\}_{i=1}^{n}$和$\{\tilde{e}_i\}_{i=1}^{n}$，规定记号
$$\tilde{e}_j = \sum_{i=1}^{n} \frac{\partial x^i}{\partial \tilde{x}^j} e_i = \frac{\partial x^i}{\partial \tilde{x}^j} e_i.$$
即
$$[\tilde{e}_1, \ldots, \tilde{e}_n] = [e_1, \ldots, e_n] \begin{bmatrix}
\frac{\partial x^1}{\partial \tilde{x}^1} & \ldots & \frac{\partial x^1}{\partial \tilde{x}^n} \\
\vdots & & \vdots \\
\frac{\partial x^n}{\partial \tilde{x}^1} & \ldots & \frac{\partial x^n}{\partial \tilde{x}^n}
\end{bmatrix}.$$
$$[\tilde{e}_i] = [e_i] \left[ \frac{\partial x^i}{\partial \tilde{x}^j} \right].$$
显然，任意向量$v$，坐标$[x^j]$和$[\tilde{x}^j]$有
$$v = \tilde{x}^j \tilde{e}_j = \tilde{x}^j \frac{\partial x^i}{\partial \tilde{x}^j} e_i.$$
所以
$$x^i = \tilde{x}^j \frac{\partial x^i}{\partial \tilde{x}^j}.$$
注意，偏导数只是记号，用以表示过渡矩阵.

显然
$$[e_i] = [\tilde{e}_i] \left[ \frac{\partial \tilde{x}^i}{\partial x^j} \right] = [\bar{e}_i] \left[ \frac{\partial \bar{x}^i}{\partial \tilde{x}^j} \right] \left[ \frac{\partial \tilde{x}^i}{\partial x^j} \right] = [\bar{e}_i] \left[ \frac{\partial \bar{x}^i}{\partial x^j} \right].$$
所以
$$\left[ \frac{\partial \bar{x}^i}{\partial \tilde{x}^j} \right] \left[ \frac{\partial \tilde{x}^i}{\partial x^j} \right] = \left[ \frac{\partial \bar{x}^i}{\partial x^j} \right].$$
$$\frac{\partial \bar{x}^i}{\partial \tilde{x}^k} \frac{\partial \tilde{x}^k}{\partial x^j} = \frac{\partial \bar{x}^i}{\partial x^j}.$$

显然
$$[e_i] = [e_i] \left[ \frac{\partial x^i}{\partial x^j} \right].$$
所以
$$\left[ \frac{\partial x^i}{\partial x^j} \right] = I_n.$$
$$\frac{\partial x^i}{\partial x^j} = \delta_j^i.$$
所以
$$\left[ \frac{\partial x^i}{\partial \tilde{x}^j} \right] \left[ \frac{\partial \tilde{x}^i}{\partial x^j} \right] = I_n.$$
$$\left[ \frac{\partial x^i}{\partial \tilde{x}^j} \right]^{-1} = \left[ \frac{\partial \tilde{x}^i}{\partial x^j} \right].$$
$$\frac{\partial x^i}{\partial \tilde{x}^k} \frac{\partial \tilde{x}^k}{\partial x^j} = \delta_j^i.$$

$n$维线性空间$V$，量$T$满足：
1. $V$的任意一组基$\{e_i\}_{i=1}^{n}$下，存在唯一数组$(t_{j_1,\ldots,j_s}^{i_1,\ldots,i_r})$与$T$对应，其中，上下标每个都取遍$1,\ldots,n$.
2. $V$的任意两组基$\{e_i\}_{i=1}^{n}$和$\{\tilde{e}_i\}_{i=1}^{n}$下，$T$对应的数组满足
$$\tilde{t}_{j_1,\ldots,j_s}^{i_1,\ldots,i_r} = \frac{\partial \tilde{x}^{i_1}}{\partial x^{k_1}} \cdots \frac{\partial \tilde{x}^{i_r}}{\partial x^{k_r}} t_{l_1,\ldots,l_s}^{k_1,\ldots,k_r} \frac{\partial x^{l_1}}{\partial \tilde{x}^{j_1}} \cdots \frac{\partial x^{l_s}}{\partial \tilde{x}^{j_s}}.$$
简记
$$\frac{\partial \tilde{x}^i}{\partial x^k} = \frac{\partial \tilde{x}^{(i_1, \ldots, i_r)}}{\partial x^{(k_1, \ldots, k_r)}} = \frac{\partial \tilde{x}^{i_1}}{\partial x^{k_1}} \cdots \frac{\partial \tilde{x}^{i_r}}{\partial x^{k_r}}.$$
$$\tilde{t}_j^i = \frac{\partial \tilde{x}^i}{\partial x^k} t_l^k \frac{\partial x^l}{\partial \tilde{x}^j}.$$
称$T$为$(r,s)$型张量.$r$称为逆变阶数，$s$称为协变阶数.显然，$T$在一组基下的分量个数为$n^{r+s}$.

显然，只需要给出一组基下的分量，以及任意一组基下的分量和该分量之间的关系满足上式，那么就确定了一个张量.

假设基$\{e_i\}$给定分量$(t_j^i)$，那么任意两组基$\{\tilde{e}_i\}$和$\{\bar{e}_i\}$下，有
$$\tilde{t}_j^i = \frac{\partial \tilde{x}^i}{\partial x^k} t_l^k \frac{\partial x^l}{\partial \tilde{x}^j}.$$
所以
$$\frac{\partial x^a}{\partial \tilde{x}^i} \tilde{t}_j^i \frac{\partial \tilde{x}^j}{\partial x^b} = \frac{\partial x^a}{\partial \tilde{x}^i} \frac{\partial \tilde{x}^i}{\partial x^k} t_l^k \frac{\partial x^l}{\partial \tilde{x}^j} \frac{\partial \tilde{x}^j}{\partial x^b} = \delta_k^a t_l^k \delta_b^l.$$
令$a=k,b=l$则
$$t_l^k = \frac{\partial x^k}{\partial \tilde{x}^i} \tilde{t}_j^i \frac{\partial \tilde{x}^j}{\partial x^l}.$$
所以
$$\bar{t}_q^p = \frac{\partial \bar{x}^p}{\partial x^k} t_l^k \frac{\partial x^l}{\partial \bar{x}^q} = \frac{\partial \bar{x}^p}{\partial x^k} \frac{\partial x^k}{\partial \tilde{x}^i} \tilde{t}_j^i \frac{\partial \tilde{x}^j}{\partial x^l} \frac{\partial x^l}{\partial \bar{x}^q} = \frac{\partial \bar{x}^p}{\partial \tilde{x}^i} \tilde{t}_j^i \frac{\partial \tilde{x}^j}{\partial \bar{x}^q}.$$
得证.

线性空间$V$的线性函数全体构成对偶空间$V^*$，$V$的基为$\{e_i\}$，取$V^*$的基$\{e^{*i}\}$满足
$$e^{*i}(e_j) = \delta_j^i.$$
那么，任意$V$的基与$V^*$的基构成一对一映射.假设
$$\tilde{e}^{*i} = y_j e^{*j}.$$
则
$$\delta_j^i = \tilde{e}^{*i}(\tilde{e}_j) = y_k e^{*k} \left( \frac{\partial x^l}{\partial \tilde{x}^j} e_l \right) = y_k \frac{\partial x^l}{\partial \tilde{x}^j} \delta_l^k = y_k \frac{\partial x^k}{\partial \tilde{x}^j}.$$
$$\delta_j^i \frac{\partial \tilde{x}^j}{\partial x^w} = y_k \frac{\partial x^k}{\partial \tilde{x}^j} \frac{\partial \tilde{x}^j}{\partial x^w} = y_k \delta_w^k.$$
所以
$$y_k = \frac{\partial \tilde{x}^i}{\partial x^k}.$$
$$\tilde{e}^{*i} = \frac{\partial \tilde{x}^i}{\partial x^j} e^{*j}.$$
所以
$$\tilde{e}_i = \frac{\partial x^j}{\partial \tilde{x}^i} e_j \quad \Leftrightarrow \quad \tilde{e}^{*i} = \frac{\partial \tilde{x}^i}{\partial x^j} e^{*j}.$$

显然，函数和元素之间没有区别，即$V^*$的对偶空间就是$V$，即
$$e_{j}(e^{*i}) = \delta_j^i.$$
此时，$e_j$是作用子，代表$V^*$上的线性函数.

线性函数
$$T(x^1,\ldots,x^r,x_1, \ldots, x_s) = t_{j_1, \ldots, j_s}^{i_1, \ldots, i_r} e_{i_1}(x^1) \cdots e_{i_r}(x^r) e^{*j_1}(x_1) \cdots e^{*j_s}(x_s).$$
其中
$$e_{i_1}, \ldots, e_{i_r} \in \{e_i\}_{i=1}^{n}, \quad e^{*j_1}, \ldots, e^{*j_s} \in \{e^{*i}\}_{i=1}^{n}.$$
$$x^1, \ldots, x^r \in V^*, \quad x_1, \ldots, x_s \in V.$$
简记
$$T = t_{j_1, \ldots, j_s}^{i_1, \ldots, i_r} e_{i_1} \otimes \cdots \otimes e_{i_r} \otimes e^{*j_1} \otimes \cdots \otimes e^{*j_s}.$$
换基有
$$T = t_{j_1, \ldots, j_s}^{i_1, \ldots, i_r} \frac{\partial \tilde{x}^{k_1}}{\partial x^{i_1}} \cdots \frac{\partial \tilde{x}^{k_r}}{\partial x^{i_r}} \frac{\partial x^{j_1}}{\partial \tilde{x}^{l_1}} \cdots \frac{\partial x^{j_s}}{\partial \tilde{x}^{l_s}} \tilde{e}_{k_1} \otimes \cdots \otimes \tilde{e}_{k_r} \otimes \tilde{e}^{*l_1} \otimes \cdots \otimes \tilde{e}^{*l_s}$$
所以
$$\tilde{t}_{l_1, \ldots, l_s}^{k_1, \ldots, k_r} = t_{j_1, \ldots, j_s}^{i_1, \ldots, i_r} \frac{\partial \tilde{x}^{k_1}}{\partial x^{i_1}} \cdots \frac{\partial \tilde{x}^{k_r}}{\partial x^{i_r}} \frac{\partial x^{j_1}}{\partial \tilde{x}^{l_1}} \cdots \frac{\partial x^{j_s}}{\partial \tilde{x}^{l_s}}.$$
所以，张量与线性函数是一对一映射.$(r,s)$型张量表示，$r$个$V^*$上的变量和$s$个$V$上的变量的线性函数.所以，$r$为逆变阶数，$s$为协变阶数.

在相同的基下，$(r,s)$型张量按分量的加法和数乘，构成$n^{r+s}$维线性空间$T^{r,s}(V)$，也表示张量对应的线性函数构成的线性空间.

$A \in T^{r, s}(V), B \in T^{p, q}(V)$，有
$$A = (a_{j_1, \ldots, j_{s}}^{i_1, \ldots, i_r}), \quad B = (b_{j_1, \ldots, j_{q}}^{i_1, \ldots, i_p}).$$
令
$$A \otimes B = (a_{j_1, \ldots, j_{s}}^{i_1, \ldots, i_r} \cdot b_{j_{s+1}, \ldots, j_{s+q}}^{i_{r+1}, \ldots, i_{r+p}}).$$
称为张量积.显然，$A \otimes B \in T^{r+p, s+q}(V)$.使用线性函数表示
$$A = a_{j_1, \ldots, j_{s}}^{i_1, \ldots, i_r} e_{i_1} \otimes \cdots \otimes e_{i_r} \otimes e^{*j_1} \otimes \cdots \otimes e^{*j_s}, \quad B = b_{l_1, \ldots, l_{q}}^{k_1, \ldots, k_p} e_{k_1} \otimes \cdots \otimes e_{k_r} \otimes e^{*l_1} \otimes \cdots \otimes e^{*l_s}.$$
$$AB = A \otimes B = a_{j_1, \ldots, j_{s}}^{i_1, \ldots, i_r} b_{l_1, \ldots, l_{q}}^{k_1, \ldots, k_p} e_{i_1} \otimes \cdots \otimes e_{i_r} \otimes e^{*j_1} \otimes \cdots \otimes e^{*j_s} \otimes e_{k_1} \otimes \cdots \otimes e_{k_r} \otimes e^{*l_1} \otimes \cdots \otimes e^{*l_s}.$$
显然，线性运算和张量积，与基的选择无关，即不同基下的表示对应同一个张量.

显然，张量积满足数乘，结合律，分配律.如果不考虑基顺序，即只考虑线性函数的情况，则
$$AB = BA.$$
如果考虑基顺序，即只考虑张量的情况，则
$$A \otimes B \neq B \otimes A.$$

$T\in T^{r, s}(V)$，有
$$T = (t_{j_1,\ldots,j_s}^{i_1,\ldots,i_r}).$$
令
$$c_{p,q}(T) = (t_{j_1,\ldots,j_{q-1},m,j_{q+1},\ldots,j_s}^{i_1,\ldots,i_{p-1},m,i_{p+1},\ldots,i_r}).$$
称为缩并.
$$\tilde{t}_{m,j_1, \ldots, j_s}^{m,i_1, \ldots, i_r} = \frac{\partial \tilde{x}^{m}}{\partial x^{k_p}} \frac{\partial \tilde{x}^{i_1}}{\partial x^{k_1}} \cdots \frac{\partial \tilde{x}^{i_r}}{\partial x^{k_r}} t_{l_q,l_1, \ldots, l_s}^{k_p,k_1, \ldots, k_r} \frac{\partial x^{l_q}}{\partial \tilde{x}^m} \frac{\partial x^{l_1}}{\partial \tilde{x}^{j_1}} \cdots \frac{\partial x^{l_s}}{\partial \tilde{x}^{j_s}}.$$
所以
$$\tilde{t}_{m,j_1, \ldots, j_s}^{m,i_1, \ldots, i_r} = \frac{\partial \tilde{x}^{i_1}}{\partial x^{k_1}} \cdots \frac{\partial \tilde{x}^{i_r}}{\partial x^{k_r}} t_{n,l_1, \ldots, l_s}^{n,k_1, \ldots, k_r} \frac{\partial x^{l_1}}{\partial \tilde{x}^{j_1}} \cdots \frac{\partial x^{l_s}}{\partial \tilde{x}^{j_s}}.$$
所以，$c_{p,q}(T) \in T^{r-1, s-1}(V)$，并且与基的选择无关.

矩阵$G \in T^{0,2}(V)$对称可逆，则
$$\tilde{g}_{i,j} = g_{k,l} \frac{\partial x^k}{\partial \tilde{x}^i} \frac{\partial x^l}{\partial \tilde{x}^j}.$$
$$[\tilde{g}_{i,j}] = \left[ \frac{\partial x^k}{\partial \tilde{x}^i} \right]^T [g_{k,l}] \left[ \frac{\partial x^l}{\partial \tilde{x}^j} \right].$$
所以
$$[\tilde{g}^{i,j}] = [\tilde{g}_{i,j}]^{-1} = \left[ \frac{\partial \tilde{x}^j}{\partial x^l} \right] [g^{k,l}] \left[ \frac{\partial \tilde{x}^i}{\partial x^k} \right]^T = \left[ \frac{\partial \tilde{x}^i}{\partial x^k} \right] [g^{k,l}] \left[ \frac{\partial \tilde{x}^j}{\partial x^l} \right]^T.$$
$$\tilde{g}^{i,j} = g^{k,l} \frac{\partial \tilde{x}^i}{\partial x^k} \frac{\partial \tilde{x}^j}{\partial x^l}.$$
所以，逆矩阵$G^{-1} \in T^{2,0}(V)$.

$T\in T^{r, s}(V)$，有
$$T = (t_{j_1,\ldots,j_s}^{i_1,\ldots,i_r}).$$
令
$$c_{p,q}(G \otimes T).$$
称为指标下降.
$$c_{p,q}(G^{-1} \otimes T).$$
称为指标上升.