Files
2026-10-04 11:44:56 +08:00

195 lines
6.0 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# 题型五:协方差与相关系数
## 识别特征
- 题干涉及 $\text{Cov}(X,Y)$ 或 $\rho_{XY}$
- 要求**计算**协方差或相关系数
- 利用性质:$\text{Cov}(aX+b, cY+d) = ac \cdot \text{Cov}(X,Y)$
## 解题流程
```mermaid
flowchart TD
A[识别:Cov 或 ρ] --> B{已知联合密度?}
B -->|是| C["Cov X,Y = ∬ x-E X y-E Y f x,y dxdy"]
B -->|否| D["Cov X,Y = E XY - E X E Y"]
C --> E[计算 E XY 和 E X E Y]
D --> E
E --> F{求相关系数?}
F -->|是| G["ρ = Cov / σ_X σ_Y"]
F -->|否| H[得 Cov X,Y]
```
## 通法步骤
### 一、协方差的定义与等价公式
**定义**:
$$
\text{Cov}(X,Y) = E[(X - E(X))(Y - E(Y))]
$$
**等价公式**(**计算首选**):
$$
\boxed{\text{Cov}(X,Y) = E(XY) - E(X) E(Y)}
$$
> *极简例子:* $X, Y$ 独立时 $E(XY) = E(X)E(Y)$,故 $\text{Cov}=0$
### 二、相关系数 $\rho_{XY}$
**定义**(标准化协方差):
$$
\rho_{XY} = \frac{\text{Cov}(X,Y)}{\sqrt{D(X)} \sqrt{D(Y)}} = \frac{\text{Cov}(X,Y)}{\sigma_X \sigma_Y}
$$
**性质**:
- $-1 \le \rho_{XY} \le 1$(Cauchy-Schwarz 不等式)
- $\rho_{XY} = 0$ $\iff$ $X, Y$ **不相关**
- $|\rho_{XY}| = 1$ $\iff$ $\exists a \ne 0, b$ 使得 $P(Y = aX+b) = 1$
- $|\rho_{XY}|$ 越大,$X, Y$ **线性关系**越强
### 三、协方差的性质 ⭐
| 性质 | 公式 |
|------|------|
| **自身** | $\text{Cov}(X,X) = D(X)$ |
| **对称** | $\text{Cov}(X,Y) = \text{Cov}(Y,X)$ |
| **伸缩** | $\text{Cov}(aX, bY) = ab \cdot \text{Cov}(X,Y)$ |
| **平移** | $\text{Cov}(aX+b, cY+d) = ac \cdot \text{Cov}(X,Y)$(常数消失) |
| **分配律** | $\text{Cov}(X_1+X_2, Y) = \text{Cov}(X_1,Y) + \text{Cov}(X_2,Y)$ |
| **与方差** | $D(X \pm Y) = D(X) + D(Y) \pm 2\text{Cov}(X,Y)$ |
### 四、计算路径
**路径 A**:已知联合密度 $f(x,y)$
$$
\text{Cov}(X,Y) = \iint (x-\mu_X)(y-\mu_Y) f(x,y) dxdy
$$
**路径 B**(**更常用**):先求 $E(X), E(Y), E(XY)$
$$
\text{Cov}(X,Y) = E(XY) - E(X) E(Y)
$$
**路径 C**:利用性质分解
$$
\text{Cov}(aX+b, cY+d) = ac \cdot \text{Cov}(X,Y)
$$
### 五、相关系数的计算
求 $\rho_{XY}$ 需先算:
1. $\text{Cov}(X,Y)$(用上述路径)
2. $D(X) = E(X^2) - E^2(X)$
3. $D(Y) = E(Y^2) - E^2(Y)$
4. $\rho_{XY} = \text{Cov}(X,Y) / \sqrt{D(X) D(Y)}$
## 常见陷阱
- **$\rho_{XY}=0$ 不等价独立**:除二维正态外,独立是不相关的**充分非必要**条件
- **正态分布的 $\rho$**:$(X,Y) \sim N(\mu_1, \mu_2, \sigma_1^2, \sigma_2^2, \rho)$ 的 $\rho$ 就是 $\rho_{XY}$
- **$\rho$ 只测线性**:$Y = X^2$($X \sim U(-1,1)$)的 $\rho_{XY}=0$,但有强非线性关系
- **$\text{Cov}$ 的伸缩**:$\text{Cov}(aX+b, cY+d) = ac \cdot \text{Cov}(X,Y)$,**不要忘了系数相乘**
- **$D(X \pm Y)$ 中的协方差符号**:减法对应 $-2\text{Cov}(X,Y)$
## 经典母题
> **题目 1**:$(X,Y)$ 联合密度 $f(x,y) = \dfrac{1}{8}(x+y), 0 \le x, y \le 2$。求 $\text{Cov}(X,Y)$ 和 $\rho_{XY}$。
**解析**:
由对称性 $E(X) = E(Y)$。
**求 $E(X)$**:边缘密度 $f_X(x) = \int_0^2 \frac{1}{8}(x+y) dy = \frac{x+1}{4}$
$$
E(X) = \int_0^2 x \cdot \frac{x+1}{4} dx = \frac{1}{4} \left[ \frac{x^3}{3} + \frac{x^2}{2} \right]_0^2 = \frac{1}{4} \left( \frac{8}{3} + 2 \right) = \frac{7}{6}
$$
**求 $E(XY)$**:
$$
E(XY) = \int_0^2 \int_0^2 xy \cdot \frac{1}{8}(x+y) dy dx
$$
先算内层:$\int_0^2 y(x+y) dy = x \cdot 2 + 8/3 = 2x + 8/3$(再乘 $\frac{1}{8}$,得 $\frac{x}{4} + \frac{1}{3}$)
外层:$\int_0^2 xy \cdot \frac{1}{8}(x+y) dx = \int_0^2 \left( \frac{x^2}{4} + \frac{x}{3} \right) x dx = \int_0^2 \left( \frac{x^3}{4} + \frac{x^2}{3} \right) dx$
$$
= \frac{1}{4} \cdot 4 + \frac{1}{3} \cdot \frac{8}{3} = 1 + \frac{8}{9} = \frac{17}{9}
$$
Wait, redo carefully:
$$
E(XY) = \int_0^2 \int_0^2 xy \cdot \frac{1}{8}(x+y) dy dx = \frac{1}{8} \int_0^2 \int_0^2 (x^2 y + xy^2) dy dx
$$
外层对 $x$:$\int_0^2 (x^2 y + xy^2) dx = \frac{8y}{3} + 2y^2$
再对 $y$:$\int_0^2 \left(\frac{8y}{3} + 2y^2\right) dy = \frac{16}{3} + \frac{16}{3} = \frac{32}{3}$
故 $E(XY) = \frac{1}{8} \cdot \frac{32}{3} = \frac{4}{3}$
**协方差**:
$$
\text{Cov}(X,Y) = E(XY) - E(X)E(Y) = \frac{4}{3} - \frac{7}{6} \cdot \frac{7}{6} = \frac{4}{3} - \frac{49}{36} = \frac{48 - 49}{36} = -\frac{1}{36}
$$
**复盘**:$\text{Cov}<0$ 说明 $X, Y$ 负相关。
---
> **题目 2**:设 $X \sim N(0,1), Y = X^2$。求 $\rho_{XY}$。
**解析**:
- $E(X) = 0, D(X) = 1$
- $E(Y) = E(X^2) = 1, D(Y) = E(X^4) - 1 = 3 - 1 = 2$
- $E(XY) = E(X^3) = 0$(正态的奇数阶原点矩为 0)
- $\text{Cov}(X,Y) = 0 - 0 \cdot 1 = 0$
- $\rho_{XY} = 0$
**复盘**:$\rho_{XY} = 0$ 但 $Y = X^2$($Y$ 完全由 $X$ 决定)——**不相关但不独立**的经典反例!
---
> **题目 3**:$D(X) = 4, D(Y) = 9, \rho_{XY} = 0.5$。求 $\text{Cov}(X,Y)$ 和 $D(X+Y)$。
**解析**:
$$
\text{Cov}(X,Y) = \rho_{XY} \sigma_X \sigma_Y = 0.5 \times 2 \times 3 = 3
$$
$$
D(X+Y) = D(X) + D(Y) + 2\text{Cov}(X,Y) = 4 + 9 + 2 \times 3 = 19
$$
**复盘**:$\rho$ 是"标准化协方差",已知 $\rho$ 和 $\sigma$ 可还原 $\text{Cov}$。
---
> **题目 4**:设 $(X,Y) \sim N(0, 0, 1, 1, \rho)$。$Z = X + Y$,$W = X - Y$。求 $\rho_{ZW}$。
**解析**:
- $\text{Cov}(Z,W) = \text{Cov}(X+Y, X-Y) = \text{Cov}(X,X) - \text{Cov}(X,Y) + \text{Cov}(Y,X) - \text{Cov}(Y,Y)$
- $= D(X) - D(Y) = 1 - 1 = 0$
**复盘**:$\text{Cov}(Z,W) = 0 \Rightarrow Z, W$ **不相关**;二维正态下进一步 ⇒ $Z, W$ **独立**。
---
> **变式 1**:$X, Y$ 独立,$\text{Cov}(2X+3, 4Y-5) = ?$
>
> **解**:$\text{Cov}(2X+3, 4Y-5) = 2 \times 4 \cdot \text{Cov}(X,Y) = 8 \times 0 = 0$。
> **变式 2**:$D(X) = D(Y) = 4, \rho_{XY} = -0.8$,求 $D(2X+3Y)$。
>
> **解**:$D(2X+3Y) = 4 D(X) + 9 D(Y) + 12 \text{Cov}(X,Y) = 16 + 36 + 12 \times (-0.8) \times 2 \times 2 = 52 - 38.4 = 13.6$。
---
> **口诀速记**:
> - "**$\text{Cov}(X,Y) = E(XY) - E(X)E(Y)$**"
> - "**$\rho = \text{Cov} / (\sigma_X \sigma_Y)$**"
> - "**$|\rho| \le 1$**,$|\rho|=1$ ⇔ 线性"
> - "**$\rho=0$ 不相关,独立 ⇐ 不相关(一般),独立 ⇔ 不相关(二维正态)**"