Chapter 3 (Determinants): Properties of determinants

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

Row Operations

• The secret of determinants lies in how they change when row operations are performed.

Study Tip: In general, computation of a $3\times 3$ determinant by row reduction takes 10 multiplications (and divisions), but cofactor expansion only takes 9 multiplications. At $n=4$, the advantage switches to row reduction, which requires 23 multiplications, cofactor expansion 40 (9 for four $3\times 3$ determinants, plus 4 multiplications of $a_{ij}$ times $det{A}_{ij}$). Often, the best strategy is to combine the two techniques

• 推论：$|kA|=k^n|A|$

利用行变换计算行列式

• Suppose a square matrix $A$ has been reduced to an echelon form $U$ by row replacements and row interchanges.
  
• If there are $r$ interchanges, then
  It is interesting to note that although the echelon form $U$ described above is not unique, and the pivots are not unique, the $product$ of the pivots is unique, except for a possible minus sign.

---

EXAMPLE 1

Compute $detA$, where $A=\left[\begin{array}{ccc}1& -4& 2\\ -2& 8& -9\\ -1& 7& 0\end{array}\right]$.

SOLUTION

• The strategy is to reduce $A$ to echelon form and then to use the fact that the determinant of a triangular matrix is the product of the diagonal entries.
  

---

• A common use of Theorem 3(c）in hand calculations is to $factor$ $out$ $a$ $common$ $multiple$ $of$ $one$ $row$ of a matrix. For instance,

EXAMPLE 2

Compute $detA$, where $A=\left[\begin{array}{cccc}2& -8& 6& 8\\ 3& -9& 5& 10\\ -3& 0& 1& -2\\ 1& -4& 0& 6\end{array}\right]$.

SOLUTION

• To simplify the arithmetic, we want a 1 in the upper-left corner. We could interchange rows 1 and 4. Instead, we factor out 2 from the top row, and then proceed with row replacements in the first column:
  
• Next, we could factor out another 2 from row 3 or use the 3 in the second column as a pivot. We choose the latter operation, adding 4 times row 2 to row 3:

推论: 方阵 $A$ 可逆 $\iff$ $\text{det}A\ne 0$

• Formula (1) not only gives a concrete interpretation of $detA$ but also proves the main theorem of this section:
  Theorem 4 adds the statement “$detA\ne 0$” to the Invertible Matrix Theorem.
• A useful corollary is that $detA=0$ when the columns of $A$ are linearly dependent. Also, $detA=0$ when the rows of $A$ are linearly dependent.
  • Rows of $A$ are columns of $A^T$ , and linearly dependent columns of $A^T$ make $A^T$ singular. When $A^T$ is singular, so is $A$.
  • In practice, linear dependence is obvious when two columns or two rows are the same or a column or a row is zero.

可以用求行列式的方法判断矩阵是否可逆

推论: 分块矩阵的行列式计算

EXAMPLE

Let $A$, $B$, $C$, $D$, and $I$ be $n\times n$ matrices. Use the definition or properties of a determinant to justify the following formulas.

SOLUTION

• $a$, $b$ 都很简单，下面就只证明一下 $c$ 的前一个等式 (后一个类似)；下面分两种情况讨论:
  • 若 $A$ 不可逆，则 $detA=0$. 同时对 $A$ 作行等价变换一定可以形成 0 行，这样沿 0 行展开，$det\left[\begin{array}{cc}A& 0\\ C& D\end{array}\right]=0$, 因此左右两式相等
  • 若 $A$ 可逆，$A$ 的各行一定线性无关，因此 $A$ 的各行生成 ${\mathbb{R}}^n$，可以使用行倍加将 $\left[\begin{array}{cc}A& 0\\ C& D\end{array}\right]$ 化简为 $\left[\begin{array}{cc}A& 0\\ 0& D\end{array}\right]$。$A$ 和 $D$ 又可分别化为上三角矩阵，这样对角线元素之积$=detA\cdot detD=det\left[\begin{array}{cc}A& 0\\ C& D\end{array}\right]$

上面的三个结论可以直接使用

Proof of Thoerem 3

• It is convenient to prove Theorem 3 when it is stated in terms of the elementary matrices.
  • We call an elementary matrix $E$ a $row$ $replacement$ ($matrix$) if $E$ is obtained from the identity $I$ by adding a multiple of one row to another row;
  • $E$ is an $interchange$ if $E$ is obtained by interchanging two rows of $I$ ;
  • $E$ is a $scale$ $by$ $r$ if $E$ is obtained by multiplying a row of $I$ by a nonzero scalar $r$.

---

• With this terminology, Theorem 3 can be reformulated as follows:
  $$detEA=(detE)(detA)$$$where$
  
• The proof is by induction on the size of $A$.
  • The theorem is obvious for the case of a $2\times 2$ matrix.
  • Suppose the theorem has been verified for determinants of $k\times k$ matrices with $k\ge 2$, let $n=k+1$, and let $A$ be $n\times n$. The action of $E$ on $A$ involves either two rows or only one row. So we can expand $detEA$ across a row that is unchanged by the action of $E$, say, row $i$ . Let $A_{ij}$ (respectively, $B_{ij}$) be the matrix obtained by deleting row $i$ and column $j$ from $A$ (respectively, $EA$). Then the rows of $B_{ij}$ are obtained from the rows of $A_{ij}$ by the same type of elementary row operation that $E$ performs on $A$. Since these submatrices are only $k\times k$, the induction assumption implies that
    $$det{B}_{ij}=\alpha \cdot det{A}_{ij}$$where $\alpha =1,-1$, or $r$, depending on the nature of $E$. The cofactor expansion across row $i$ is
    In particular, taking $A=I_n$, we see that $detE=1,-1$, or $r$, depending on the nature of $E$.

Column Operations

• Because of Theorem 5, each statement in Theorem 3 is true when the word row is replaced everywhere by column. (i.e. column operations have the same effects on determinants as row operations)

PROOF

• Using the Principle of Mathematical Induction
  • The theorem is obvious for $n=1$.
  • Suppose the theorem is true for $k\times k$ determinants and let $n=k+1$. Then the cofactor of $a_{1j}$ in $A$ equals the cofactor of $a_{j1}$ in $A^T$. Hence the cofactor expansion of $detA$ along the first row equals the cofactor expansion of $det{A}^T$ down the first column. That is, $A$ and $A^T$ have equal determinants.
  • By the Principle of Mathematical Induction, the theorem is true for all $n\ge 1$.

$\text{det}AB=(\text{det}A)(\text{det}B)$

• $\mathbf{W}\mathbf{a}\mathbf{r}\mathbf{n}\mathbf{i}\mathbf{n}\mathbf{g}$: A common misconception is that Theorem 6 has an analogue for sums of matrices. However, $det(A+B)$ $is$ $not$ equal to $detA+detB$, in general.

---

推论

• 若 $A$ 可逆，则 $det{A}^{-1}=\frac{1}{detA}$
• 如果 $A$ 为方阵, 则 $|\text{det}A|$ 为 $A$ 的奇异值的积
  • Proof: $|detA|=|det(U\Sigma V^T)|=|detU\cdot det\Sigma \cdot det{V}^T|=|\pm 1\cdot det\Sigma \cdot \pm 1|=det\Sigma$

Check The Singular Value Decomposition

---

Proof of Theorem 6

• If $A$ is not invertible, then neither is $AB$. In this case, $detAB=(detA)(detB)=0$.
• If $A$ is invertible, then $A$ and the identity matrix $I_n$ are row equivalent. So there exist elementary matrices $E_1,...,E_p$ such that
  $$A=E_p,E_{p-1}...E_1\cdot I_n=E_p,E_{p-1}...E_1$$Then repeated application of Theorem 3 shows that
  

A Linearity Property of the Determinant Function

行列式函数的一个线性性质

• For an $n\times n$ matrix $A$, we can consider $detA$ as a function of the $n$ column vectors in $A$. We will show that if all columns except one are held fixed, then $detA$ is a linear function of that one (vector) variable.
• Suppose that the $j$ th column of $A$ is allowed to vary, and write
  Define a transformation $T$ from ${\mathbb{R}}^n$ to $\mathbb{R}$ by
  Then,
  

This (multi-) linearity property of the determinant turns out to have many useful consequences that are studied in more advanced courses.

---

EXAMPLE

Compute $detA$, where

SOLUTION 1

• a. Subtract row 2 from row 1, row 3 from row 2, and so on.
• b. With the resulting matrix from part (a), add column 1 to column 2, then add this new column 2 to column 3, and so on.
• c. Find the determinant of the resulting matrix from (b).
  $$detA=(a-b{)}^{n-1}[a+(n-1)b]$$

SOLUTION 2

• 观察到各行(列)的元素之和都相等，因此可以先将第 2 至 $n$ 列都加到第 1 列，这样得到的新矩阵中第一列中每个元素都相等，再用每行都减第一行即可得到一个对角矩阵

SOLUTION 3

• Notice that $A$, $B$, and $C$ are nearly the same except that the first column of $A$ equals the sum of the first columns of $B$ and $C$. A linearity property of the determinant function, says that $detA=detB+detC$. We can use this fact to prove the formula above by induction on the size of matrix $A$.