证明施瓦茨不等式(协方差和方差的关系)
定理:
对于任意二维随机变量(X,Y),若X与Y的方差都存在,且记
σ
x
2
=
V
a
r
(
X
)
\sigma^2_x=Var(X)
σx2=Var(X),
σ
Y
2
=
V
a
r
(
Y
)
\sigma^2_Y=Var(Y)
σY2=Var(Y),则有
[
C
o
v
(
X
,
Y
)
]
2
≤
σ
x
2
σ
Y
2
[Cov(X,Y)]^2 \leq \sigma^2_x \sigma^2_Y
[Cov(X,Y)]2≤σx2σY2
证明:
∀
t
∈
R
\forall t \in R
∀t∈R,考虑二次函数:
g
(
t
)
=
E
[
t
(
X
−
E
X
)
+
(
Y
−
E
Y
)
]
2
=
E
[
t
2
(
X
−
E
X
)
2
+
2
t
(
X
−
E
X
)
(
Y
−
E
Y
)
+
(
Y
−
E
Y
)
2
]
=
t
2
E
(
X
−
E
X
)
2
+
2
t
E
[
(
X
−
E
X
)
(
Y
−
E
Y
)
]
+
E
(
Y
−
E
Y
)
2
=
t
2
σ
x
2
+
2
t
C
o
v
(
X
,
Y
)
+
σ
Y
2
\begin{aligned} g(t)& = E[t(X-EX) + (Y-EY)]^2 \\ & = E[t^2(X-EX)^2 + 2t(X-EX)(Y-EY) + (Y-EY)^2] \\ & = t^2E(X-EX)^2 + 2tE[(X-EX)(Y-EY)] + E(Y-EY)^2 \\ & = t^2\sigma^2_x + 2tCov(X,Y) + \sigma^2_Y \end{aligned}
g(t)=E[t(X−EX)+(Y−EY)]2=E[t2(X−EX)2+2t(X−EX)(Y−EY)+(Y−EY)2]=t2E(X−EX)2+2tE[(X−EX)(Y−EY)]+E(Y−EY)2=t2σx2+2tCov(X,Y)+σY2
根据期望的定义以及
[
t
(
X
−
E
X
)
+
(
Y
−
E
Y
)
]
2
≥
0
[t(X-EX) + (Y-EY)]^2 \ge 0
[t(X−EX)+(Y−EY)]2≥0,可以得知
g
(
t
)
≥
0
g(t)\ge 0
g(t)≥0,所以根据二次函数的判别式就有
▽
=
[
2
C
o
v
(
X
,
Y
)
]
2
−
4
σ
X
2
σ
Y
2
≤
0
\large \begin{aligned} \bigtriangledown = [2Cov(X,Y)]^2 - 4\sigma^2_X\sigma^2_Y \leq 0 \end{aligned}
▽=[2Cov(X,Y)]2−4σX2σY2≤0
上面的不等式移项后即为结论中的施瓦茨不等式。