Chapter 2 (Matrix Algebra): The inverse of a matrix (矩阵的逆)

本文为《Linear algebra and its applications》的读书笔记

The inverse of a matrix

• Recall that the multiplicative inverse (乘法逆) of a number such as 5 is 1/5 or $5^{-1}$. This inverse satisfies the equations
  $$5^{-1}\cdot 5=1,5\cdot 5^{-1}=1$$
• The matrix generalization requires $both$ equations and avoids the slanted-line (斜线) notation (for division) because matrix multiplication is not commutative (可交换的).
• Furthermore, a full generalization is possible only if the matrices involved are square. (Check Example 1~3)

---

EXAMPLE 1

Suppose $CA=I_n$ (the $n\times n$ identity matrix). Show that $A$ cannot have more columns than rows.

SOLUTION

• If $\mathbf{x}$ satisfies $A\mathbf{x}=0$, then $CA\mathbf{x}=C0=0$ and so $I_n\mathbf{x}=0$ and $\mathbf{x}=0$. This implies that $A\mathbf{x}=0$ only has the trivial solution and the echelon form of $A$ must have no free variables. Thus $A$ cannot have more columns than rows.

---

EXAMPLE 2

Suppose $AD=I_m$. Show that $A$ cannot have more rows than columns.

SOLUTION

• For any $\mathbf{b}$ in ${\mathbb{R}}^m$, the equation $Ax=\mathbf{b}$ has a solution. ($AD\mathbf{b}=\mathbf{b}$) Thus $A$ cannot have more rows than columns.
• We can also apply the transpose operation to obtain $D^T{A}^T=I_m$. Then the result from example 1 can lead us to the conclusion.

---

EXAMPLE 3

Suppose $A$ is an $m\times n$ matrix and there exist $n\times m$ matrices $C$ and $D$ such that $CA=I_n$ and $AD=I_m$: Prove that $m=n$ and $C=D$.

SOLUTION

• By EXAMPLE 1, the equation $CA=I_n$ implies that $m\ge n$. By EXAMPLE 2, the equation $AD=I_m$ implies that $m\le n$. Thus $m=n$.
• To prove the second statement, $C=C{I}_n=C(AD)=(CA)D=I_{n}D=D$

代数结构中，若运算满足结合律且左右逆元均存在，则左右逆元相同 ($C=D$)。同时，逆元唯一，即 $A$ 的逆唯一

---

• An $n\times n$ matrix $A$ is said to be invertible (可逆的) if there is an $n\times n$ matrix $C$ such that
  $$CA=I_n,AC=I_n$$In this case, $C$ is an inverse (逆) of $A$. $C$ is uniquely determined by $A$. This unique inverse is denoted by $A^{-1}$, so that
  $$A^{-1}A=I_n,A{A}^{-1}=I_n$$
• A matrix that is not invertible is sometimes called a singular matrix (奇异矩阵), and an invertible matrix is called a nonsingular matrix (非奇异矩阵).

已知 $A$ 为 $n\times n$ 矩阵，且 $CA=I$，则 $A\mathbf{x}=0$ 只有平凡解，$A$ 有 $n$ 个主元列，$A\sim I$，因此 $A$ 可逆，$C=A^{-1}$
同理已知 $A$ 为 $n\times n$ 矩阵，且 $AD=I$，则 $A$ 可逆，$D=A^{-1}$

---

• Here is a simple formula for the inverse of a $2\times 2$ matrix, along with a test to tell if the inverse exists.

• The quantity $ad-bc$ is called the determinant (行列式) of $A$, and we write
  $$detA=ad-bc$$
• Theorem 4 says that a $2\times 2$ matrix $A$ is invertible if and only if $detA\ne 0$.

---

• Numerical Notes:
  • The formula in Theorem 5 is seldom used to solve an equation $A\mathbf{x}=\mathbf{b}$ numerically because row reduction of $[A\mathbf{b}]$ is nearly always faster. (Row reduction is usually more accurate, too, when computations involve rounding off numbers.)
  • One possible exception is the $2\times 2$ case. In this case, mental computations to solve $A\mathbf{x}=\mathbf{b}$ are sometimes easier using the formula for $A^{-1}$, as in the next example.

---

EXAMPLE 4

Use the inverse of the matrix to solve the system

SOLUTION

---

• The following generalization of Theorem 6(b) is needed later.
  

Elementary Matrices

初等矩阵

• There is an important connection between invertible matrices and row operations that leads to a method for computing inverses. As we shall see, an invertible matrix $A$ is row equivalent to an identity matrix, and we can find $A^{-1}$ by watching the row reduction of $A$ to $I$.

---

• An elementary matrix is one that is obtained by performing a single elementary row operation on an identity matrix.
• The next example illustrates the three kinds of elementary matrices.

EXAMPLE 5

Let

Compute $E_{1}A$, $E_{2}A$, and $E_{3}A$, and describe how these products can be obtained by elementary row operations on $A$.

SOLUTION

• Addition of $-4$ times row 1 of $A$ to row 3 produces $E_{1}A$. (This is a row replacement operation.) An interchange of rows 1 and 2 of $A$ produces $E_{2}A$, and multiplication of row 3 of $A$ by 5 produces $E_{3}A$.
  
• Left-multiplication by $E_1$ in Example 5 has the same effect on any $3\times n$ matrix. It adds $-4$ times row 1 to row 3. In particular, since $E_1\cdot I=E_1$, we see that $E_1$ itself is produced by this same row operation on the identity.

---

• Example 5 illustrates the following general fact about elementary matrices.

PROOF (下面的证明以 $3\times 3$ 矩阵为例)
$$\begin{array}{rl}\because A& =I\cdot A\\ \therefore ro{w}_i(A)& =ro{w}_i(I)\cdot A\end{array}$$a. Show that if rows 1 and 2 of $A$ are interchanged, then the result may be written as $EA$, where $E$ is an elementary matrix formed by interchanging rows 1 and 2 of $I$.

b. Show that if row 3 of $A$ is multiplied by 5, then the result may be written as $EA$, where $E$ is formed by multiplying row 3 of $I$ by 5.

---

• Since row operations are reversible, elementary matrices are invertible. Hence there is an elementary matrix $F$ such that $FE=I$. Since $E$ and $F$ correspond to reverse operations, $EF=I$, too.

---

EXAMPLE 6
Find the inverse of $E_1=\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ -4& 0& 1\end{array}\right]$.
SOLUTION
$$E_1^{-1}=\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 4& 0& 1\end{array}\right]$$

---

• The following theorem provides the best way to “visualize” an invertible matrix, and the theorem leads immediately to a method for finding the inverse of a matrix.

PROOF

• Suppose that $A$ is invertible. Then, since the equation $A\mathbf{x}=\mathbf{b}$ has a solution for each $\mathbf{b}$, $A$ has a pivot position in every row. Because $A$ is square, the $n$ pivot positions must be on the diagonal, which implies that the reduced echelon form of $A$ is $I_n$. That is, $A\sim I_n$.
• Now suppose, conversely, that $A\sim I_n$. Then, since each step of the row reduction of $A$ corresponds to left-multiplication by an elementary matrix, there exist elementary matrices $E_1,...,E_p$ such that
  $$A\sim E_{1}A\sim E_2(E_{1}A)\sim ...\sim E_p(E_{p-1}...E_{1}A)=I_n$$That is,
  $$E_p...E_{1}A=I_n(1)$$ Since the product $E_p...E_1$ of invertible matrices is invertible, $(1)$ leads to
  $$\begin{array}{rl}(E_p...E_1{)}^{-1}(E_p...E_1)A& =(E_p...E_1{)}^{-1}{I}_n\\ A& =(E_p...E_1{)}^{-1}\end{array}$$Thus $A$ is invertible, as it is the inverse of an invertible matrix (Theorem 6). Also,
  $$A^{-1}=[(E_p...E_1{)}^{-1}{]}^{-1}=E_p...E_1$$Then $A^{-1}=E_p...E_1\cdot I_n$, which says that $A^{-1}$ results from applying $E_1,...,E_p$ successively to $I_n$. This is the same sequence in (1) that reduced $A$ to $I_n$.

An Algorithm for Finding $A^{–1}$

EXAMPLE 7
Find the inverse of the matrix $A=\left[\begin{array}{ccc}0& 1& 2\\ 1& 0& 3\\ 4& -3& 8\end{array}\right]$, if it exists.
SOLUTION

---

举一反三：

• 如果想计算 $A^{-1}B$ 的话，也可以由行化简求得：若 $[AB]\sim ...\sim [IX]$，则 $X=A^{-1}B$
  • 如果 $A$ 是大于 $2\times 2$ 的矩阵，则用行化简的方法求要比直接求快得多
• We can even accelerate this process by using the LU factorization of $A$, which is adapted by MATLAB.

Another View of Matrix Inversion

• Denote the columns of $I_n$ by ${\mathbf{e}}_1,...,{\mathbf{e}}_n$. Then row reduction of $[AI]$ to $[I{A}^{-1}]$ can be viewed as the simultaneous solution of the $n$ systems
  $$A{\mathbf{x}}_1={\mathbf{e}}_1,A{\mathbf{x}}_2={\mathbf{e}}_2,...,A{\mathbf{x}}_n={\mathbf{e}}_n(2)$$where the “augmented columns” of these systems have all been placed next to $A$ to form $[A{\mathbf{e}}_1{\mathbf{e}}_2...{\mathbf{e}}_n]=[AI]$. The equation $A{A}^{-1}=I$ and the definition of matrix multiplication show that the columns of $A^{-1}$ are precisely the solutions of the systems in (2).
• This observation is useful because some applied problems may require finding only one or two columns of $A^{-1}$. In this case, only the corresponding systems in (2) need be solved.

---

EXAMPLE 8

Let $A=\left[\begin{array}{cccccc}1& 0& ...& ...& ...& 0\\ 1& 1& 0& ...& ...& 0\\ 1& 1& 1& 0& ...& 0\\ ...& ...& ...& ...& ...& ...\\ 1& 1& 1& 1& 1& 0\\ 1& 1& 1& 1& 1& 1\end{array}\right]$. $A$ is an $n\times n$ matrix. Let $B$ be its inverse. Guess the form of $B$, and then prove that $AB=I$ and $BA=I$.

SOLUTION

• Let ${\mathbf{a}}_j$, ${\mathbf{b}}_j$, and ${\mathbf{e}}_j$ denote the $j$th columns of $A$, $B$, and $I$, respectively. Note that for $j=1,\dots ,n–1$, ${\mathbf{a}}_j–{\mathbf{a}}_{j+1}={\mathbf{e}}_j$.
• Suppose $B=A^{-1}$, then $A{\mathbf{b}}_j={\mathbf{e}}_j={\mathbf{a}}_j-{\mathbf{a}}_{j+1}=A{\mathbf{e}}_j-A{\mathbf{e}}_{j+1}=A({\mathbf{e}}_j-{\mathbf{e}}_{j+1})$, $A{\mathbf{b}}_n={\mathbf{e}}_n={\mathbf{a}}_n=A{\mathbf{e}}_n$. Thus ${\mathbf{b}}_j={\mathbf{e}}_j–{\mathbf{e}}_{j+1}$ and ${\mathbf{b}}_n={\mathbf{e}}_n$.

也可以从初等行变换的角度来考虑