第 21 章
特征值与特征向量:矩阵作用下方向不变的向量
从投影、置换、旋转与三角矩阵四类例子出发,用 det(A−λI)=0 求特征值,并说明实特征值、复特征值与特征向量短缺三种可能。
Eigenvalues and Eigenvectors: Directions the Matrix Leaves Alone
Four examples — projection, permutation, rotation, triangular — build intuition; then det(A−λI)=0 finds eigenvalues, and three failure modes appear: complex values, repeats, and too few eigenvectors.
篇内目录
- What an eigenvector asks
- What a zero eigenvalue means
- Timeline
- Four examples for intuition
- Projection matrix
- Permutation matrix
- Rotation matrix
- Triangular matrix
- The method: find first, then
- The key rearrangement
- The convenient form
- A full example
- Three structural facts
- Adding shifts eigenvalues
- But does not work that way
- The trace and determinant identities always hold
- Three "bad cases"
- Core review
- Common misconceptions
- Check your understanding
- How this connects
特征向量在问什么
矩阵是作用在向量上的函数:输入 ,输出 。大多数向量的方向会被改变。少数特殊向量的方向不变,只是长度被缩放——这些就是特征向量:
称为特征值。它可以是负数或零,甚至可以是复数;「方向不变」允许反向。
注意这个方程和 完全不同: 与 都是未知量,而且乘在一起。消元法不能直接用,需要换一个思路。
零特征值的含义
若 ,则 。所以零特征值存在,等价于 奇异,对应的特征向量就是零空间中的向量。这样一来,「行列式为零」与「有零特征值」是同一件事。
时间轴
| 时间 | 内容 |
|---|---|
| 0:07 | 特征值与特征向量的总体介绍 |
| 3:10 | 与 同向意味着什么 |
| 6:43 | 判断一个向量是否为特征向量 |
| 9:43 | 投影矩阵:零空间与列空间方向 |
| 12:50 | 置换与反射矩阵的例子 |
| 15:57 | 的方向 |
| 19:15 | 求法:解 |
| 22:25 | 特征向量构成整条直线 |
| 25:32 | 2×2 的特征多项式 |
| 31:34 | 平移 的特征值加 3 |
| 34:40 | 迹与行列式给出特征值的和与积 |
| 37:49 | 旋转 的复特征值 |
| 41:02 | 复特征值成共轭对 |
| 47:05 | 重根与退化情形 |
| 50:11 | 退化矩阵的例子 |
四个例子,先建立直觉
投影矩阵
设 把向量投影到一个平面。
- 平面内的向量投影后不变:,所以 ,整个平面都是特征向量。
- 垂直于平面的向量投影为零:,所以 ,对应的特征向量是平面的法向量。
投影矩阵的特征值只有 和 ,与「投影两次等于投影一次」完全吻合: 迫使 。
置换矩阵
它交换两个分量。
- :交换后不变,。
- :交换后变成 ,。
两个特征向量垂直。这个矩阵是对称的,后面会看到这不是巧合。
旋转矩阵
逆时针旋转 :
从几何上看,任何非零向量旋转 后都不可能保持方向,所以应当「没有实特征向量」。代数上:
解得 。实矩阵可以有复特征值,而且成共轭对出现。,是反对称矩阵,这类矩阵的特征值纯虚。
三角矩阵
特征值可以直接从对角线读出:。但求特征向量时:
零空间只有一维,基是 。只有一个线性无关的特征向量,无法张成整个平面。这是重复特征值带来的「特征向量短缺」。
求法:先求 λ,再求 x
关键变形
把 移项:
要存在非零的 ,矩阵 必须奇异,于是
这称为特征方程。它是关于 的 次多项式,有 个根(计重数)。
二阶的便捷形式
对 矩阵,
其中迹是主对角线之和。因此:
这两个关系对任意 也成立:所有特征值之和等于迹,之积等于行列式。
完整例子
先算特征多项式:
因式分解:
所以 ,。验证:和 等于迹,积 等于行列式。
再求特征向量。
:
零空间由 张成。
:
零空间由 张成(也可取 )。
两个特征向量垂直。矩阵是对称的,特征值都是实数。
三个结构性事实
加 只平移特征值
若 ,则
所以特征向量完全不变,特征值整体加 。课堂上正是这样从置换矩阵 得到 :特征值从 变成 ,特征向量仍是 与 。
但 没有这种性质
一个常见错误是:若 、,就断言 的特征值是 。这不成立,因为 一般不是 的特征向量。 只有在 是 的倍数时才碰巧成立。
同理, 的特征值一般也不是特征值的乘积。
迹与行列式总是对的
无论矩阵是否对称、是否重复,下面的恒等式都成立:
这也是判断计算是否出错的第一道检查。
三种「坏情况」
| 情况 | 例子 | 后果 |
|---|---|---|
| 复特征值 | 旋转 | 没有实特征方向;特征值成共轭对 |
| 重复特征值 | 可能只有一维特征空间 | |
| 特征向量不足 | 同上 | 无法对角化,下一讲处理 |
对称矩阵是「好情况」的代表:特征值全为实,特征向量两两正交。反对称矩阵是另一个极端:特征值纯虚。一般矩阵介于两者之间。
核心回顾
- :矩阵作用后方向不变的向量。
- 零特征值 矩阵奇异。
- 求法:解 得 ,再解 得 。
- 二阶:。
- 特征值之和等于迹,之积等于行列式。
- 加 平移特征值、保持特征向量; 一般没有这种关系。
- 实矩阵可能有复特征值;重复特征值可能导致特征向量不足。
常见误区
- 把 当成普通方程组用消元解。两个未知量相乘,必须先消去 。
- 认为特征向量唯一。特征向量张成一个空间,任何非零倍数都有效。
- 忘记零特征值的几何含义。它就是零空间。
- 对 直接相加特征值。只有 时才成立。
- 以为实矩阵的特征值一定是实数。旋转矩阵给出反例。
理解检验
1. 投影矩阵的特征值为什么只能是 0 和 1?
参考答案:投影满足 。设 ,。两边再作用一次 :
但 ,所以 。由于 ,得 ,即 ,故 或 。几何上,投影到平面内的向量不动(),法向向量被压成零()。
2. 用迹与行列式验证 的特征值。
参考答案:迹 ,行列式 。特征多项式为
所以 。检验:和 等于迹,积 等于行列式,与正文一致。
3. 旋转 的矩阵为什么没有实特征向量?
参考答案:几何上, 与 垂直且长度不变,一个非零向量不可能等于自己的实倍数,所以不存在实特征向量。代数上,
方程 在实数中没有解,只有 。 是反对称矩阵(),其特征值必为纯虚数或零。
4. 一个 矩阵有特征值 ,它的迹与行列式各是多少?
参考答案:迹等于特征值之和:。行列式等于特征值之积(计重数):。这两个恒等式对任意矩阵都成立,是检查计算的第一道关口。
5. 为什么 只有一维特征空间?
参考答案:特征值是 (二重)。求特征向量解 :
方程只有 , 任意,所以零空间是 ,维数为 1。特征值的代数重数是 2,但几何重数只有 1,缺少一个线性无关的特征向量,因此这个矩阵不能对角化。
与前后课程的关系
这一讲定义特征值与特征向量,并给出求法。下一讲讨论对角化:当矩阵有足够多的线性无关特征向量时,可以写成 ,从而快速计算幂与理解长期行为。重复特征值导致的短缺,将由 Jordan 形式处理。
What an eigenvector asks
A matrix is a function on vectors: input , output . Most vectors change direction. A few keep their direction and only change length — those are the eigenvectors:
The scalar is the eigenvalue. It may be negative or zero, or even complex; "same direction" allows reversal.
This equation is not : and are both unknown, and they multiply. Elimination does not apply directly, so a different idea is needed.
What a zero eigenvalue means
If , then . A zero eigenvalue therefore exists exactly when is singular, and its eigenvectors are the null space. "Determinant zero" and "a zero eigenvalue" are the same statement.
Timeline
| Time | Topic |
|---|---|
| 0:07 | The big picture: eigenvalues and eigenvectors |
| 3:10 | What it means for to be parallel to |
| 6:43 | Testing whether a vector is an eigenvector |
| 9:43 | Projection: nullspace and column-space directions |
| 12:50 | Permutation and reflection examples |
| 15:57 | A direction with |
| 19:15 | The method: solve |
| 22:25 | Eigenvectors form whole lines |
| 25:32 | The 2×2 characteristic polynomial |
| 31:34 | Shifting by adds 3 to the eigenvalues |
| 34:40 | Trace and determinant give the sum and product |
| 37:49 | A rotation has complex eigenvalues |
| 41:02 | Complex eigenvalues come in conjugate pairs |
| 47:05 | Repeated eigenvalues and degenerate cases |
| 50:11 | An example of a degenerate matrix |
Four examples for intuition
Projection matrix
Let project onto a plane.
- A vector already in the plane is unchanged: , so , and the whole plane consists of eigenvectors.
- A vector perpendicular to the plane projects to zero: , so , and the normal direction supplies eigenvectors.
The eigenvalues are only and , consistent with , which forces .
Permutation matrix
It swaps the two components.
- : unchanged, so .
- : becomes , so .
The two eigenvectors are perpendicular. The matrix is symmetric, and that is not a coincidence.
Rotation matrix
Rotate counterclockwise by :
Geometrically, no nonzero vector can keep its direction after a quarter turn, so there should be no real eigenvector. Algebraically,
so . A real matrix can have complex eigenvalues, and they occur in conjugate pairs. Here : the matrix is antisymmetric, and such matrices have purely imaginary eigenvalues.
Triangular matrix
The eigenvalues can be read directly off the diagonal: . But for eigenvectors,
whose null space is only one-dimensional, spanned by . There is only one independent eigenvector, not enough to span the plane. A repeated eigenvalue can create a shortage.
The method: find first, then
The key rearrangement
Move everything to one side:
For a nonzero to exist, must be singular, so
This is the characteristic equation, a polynomial of degree in with roots counted with multiplicity.
The convenient form
For a matrix,
Hence
Both identities hold for every : the eigenvalues sum to the trace and multiply to the determinant.
A full example
The characteristic polynomial:
Factor:
So and . Check: the sum is the trace, and the product is the determinant.
Now the eigenvectors.
For :
Its null space is spanned by .
For :
Its null space is spanned by (or ).
The two eigenvectors are perpendicular. The matrix is symmetric, and both eigenvalues are real.
Three structural facts
Adding shifts eigenvalues
If , then
The eigenvectors are unchanged and every eigenvalue increases by . In the lecture, adding to the permutation matrix produces : the eigenvalues move from to , while the eigenvectors stay and .
But does not work that way
A common mistake: given and , conclude that has eigenvalue . This is false, because is generally not an eigenvector of . It works only when is a multiple of .
Likewise, the eigenvalues of are generally not the products of the eigenvalues.
The trace and determinant identities always hold
Whatever the matrix, symmetric or not, repeated eigenvalues or not:
These are the first check on any eigenvalue computation.
Three "bad cases"
| Case | Example | Consequence |
|---|---|---|
| Complex eigenvalues | rotation | No real eigendirection; values come in conjugate pairs |
| Repeated eigenvalues | The eigenspace may be one-dimensional | |
| Too few eigenvectors | Same example | Cannot be diagonalized; the next lecture addresses it |
Symmetric matrices are the good case: real eigenvalues and orthogonal eigenvectors. Antisymmetric matrices are the opposite extreme: purely imaginary eigenvalues. General matrices sit in between.
Core review
- : vectors whose direction survives the matrix.
- A zero eigenvalue is equivalent to singularity.
- Method: solve for , then for .
- For : .
- Eigenvalues sum to the trace and multiply to the determinant.
- Adding shifts eigenvalues and preserves eigenvectors; generally does not.
- Real matrices can have complex eigenvalues; repeated eigenvalues can leave too few eigenvectors.
Common misconceptions
- Treating as an ordinary system to eliminate. Two unknowns multiply; eliminate first.
- Expecting a unique eigenvector. Eigenvectors fill a space; any nonzero multiple works.
- Missing the geometry of a zero eigenvalue. It is the null space.
- Adding eigenvalues of and . Only valid when one matrix is a multiple of .
- Assuming a real matrix has real eigenvalues. The rotation matrix is the counterexample.
Check your understanding
1. Why can a projection matrix only have eigenvalues 0 and 1?
Reference answer: A projection satisfies . Suppose with . Apply again:
But , so . Since , , i.e. , hence or . Geometrically, vectors in the target plane stay put () and the normal direction collapses to zero ().
2. Use the trace and determinant to verify the eigenvalues of .
Reference answer: Trace and determinant . The characteristic polynomial is
so . Check: the sum equals the trace and the product equals the determinant, matching the text.
3. Why does a rotation have no real eigenvector?
Reference answer: Geometrically, is perpendicular to and of the same length, so a nonzero vector can never be a real multiple of its image. Algebraically,
and has no real solution, only . Since is antisymmetric (), its eigenvalues must be purely imaginary or zero.
4. A matrix has eigenvalues . What are its trace and determinant?
Reference answer: Trace equals the sum of the eigenvalues: . Determinant equals their product counted with multiplicity: . Both identities hold for every matrix and are the first check when computing.
5. Why does have only a one-dimensional eigenspace?
Reference answer: The only eigenvalue is (double). Solve :
whose only equation is with free, so the nullspace is , of dimension 1. The algebraic multiplicity is 2 but the geometric multiplicity is 1; one independent eigenvector is missing, so the matrix is not diagonalizable.
How this connects
This lecture defines eigenvalues and eigenvectors and shows how to compute them. The next discusses diagonalization: when there are enough independent eigenvectors, makes powers and long-term behavior easy to read. The shortage caused by repeated eigenvalues leads to Jordan form.