一般的积分是指黎曼积分, 其计算是把区域无限细分求和并取极限, 有另外一种积分是把区域无限细分求积并取极限, 这个在生存模型中有很多应用.
生存模型
设生存的时间为随机变量
T
T
T, 则生存函数定义为
S
(
t
)
:
=
P
r
(
T
≥
t
)
,
t
>
0
,
S(t):= \mathrm{Pr} (T \ge t), \: t>0,
S(t):=Pr(T≥t),t>0,
显然
S
(
0
)
=
0
S(0)=0
S(0)=0. 生存函数表示, 一个个体生存时间超过
t
t
t的概率.
连续情形
设随机变量
T
T
T所对应的密度函数为
f
(
t
)
f(t)
f(t), 并定义hazard rate为
α
(
t
)
:
=
lim
h
→
0
P
r
(
t
≤
T
≤
t
+
h
∣
T
≥
t
)
h
,
\alpha (t) := \mathop{\lim} \limits_{h \rightarrow 0} \frac{\mathrm{Pr}(t \le T \le t+h|T \ge t)}{h},
α(t):=h→0limhPr(t≤T≤t+h∣T≥t),
注意到
P
r
(
t
≤
T
≤
t
+
h
∣
T
≥
t
)
h
=
P
r
(
t
≤
T
≤
t
+
h
)
h
⋅
P
r
(
T
≥
t
)
,
\frac{\mathrm{Pr}(t \le T \le t+h|T \ge t)}{h}= \frac{\mathrm{Pr}(t\le T \le t+h)}{h \cdot \mathrm{Pr}(T\ge t)},
hPr(t≤T≤t+h∣T≥t)=h⋅Pr(T≥t)Pr(t≤T≤t+h),
故
α
(
t
)
=
f
(
t
)
/
S
(
t
)
.
\alpha(t)=f(t)/S(t).
α(t)=f(t)/S(t).
又
f
(
t
)
=
d
F
(
t
)
d
t
=
d
(
1
−
S
(
t
)
)
d
t
=
−
d
d
t
S
(
t
)
=
:
S
′
(
t
)
.
f(t) =\frac{\mathrm{d}F(t)}{\mathrm{d}t} = \frac{\mathrm{d}(1-S(t))}{\mathrm{d}t}=-\frac{d}{dt}S(t)=:S'(t).
f(t)=dtdF(t)=dtd(1−S(t))=−dtdS(t)=:S′(t).
所以
α
(
t
)
=
−
S
′
(
t
)
S
(
t
)
=
−
d
d
t
log
S
(
t
)
,
\alpha(t)=-\frac{S'(t)}{S(t)}=-\frac{\mathrm{d}}{\mathrm{d}t} \log S(t),
α(t)=−S(t)S′(t)=−dtdlogS(t),
故
S
(
t
)
=
exp
{
−
∫
0
t
α
(
s
)
d
s
}
,
t
>
0.
S(t)=\exp \{ -\int_{0}^t \alpha(s) \mathrm{d}s\}, \: t>0.
S(t)=exp{−∫0tα(s)ds},t>0.
离散情形
此时假设
f
(
t
)
=
P
r
(
T
=
t
)
f(t)=\mathrm{Pr}(T=t)
f(t)=Pr(T=t),
α
(
t
)
:
=
P
r
(
T
=
t
∣
T
≥
t
)
=
f
(
t
)
/
S
(
t
)
,
\alpha(t):=\mathrm{Pr}(T=t|T\ge t)=f(t)/S(t),
α(t):=Pr(T=t∣T≥t)=f(t)/S(t),
可以证明
S
(
t
)
=
∏
0
t
(
1
−
α
(
s
)
)
,
S(t)= \prod_0^t (1-\alpha(s)),
S(t)=0∏t(1−α(s)),
注意, 这里的
∏
\prod
∏个人感觉都没法用极限去理解, 只能用无限(即便是不可数)个1相乘仍为1理解.
不妨设
f
(
t
)
f(t)
f(t)仅在
0
<
t
1
<
t
2
<
⋯
0<t_1 < t_2 < \cdots
0<t1<t2<⋯处非零, 则
S
(
t
)
=
1
,
t
≤
t
1
,
S
(
t
)
=
1
−
f
(
t
1
)
=
1
−
α
(
t
1
)
,
t
1
<
t
≤
t
2
,
S(t)=1, \: t\le t_1, \\ S(t)=1-f(t_1)=1-\alpha(t_1), \: t_1 < t \le t_2, \\
S(t)=1,t≤t1,S(t)=1−f(t1)=1−α(t1),t1<t≤t2,
S
(
t
)
=
1
−
f
(
t
1
)
−
f
(
t
2
)
=
1
−
α
(
t
1
)
−
α
(
t
2
)
S
(
t
2
)
=
(
1
−
α
(
t
1
)
(
1
−
α
(
t
2
)
)
,
t
2
<
t
≤
t
3
⋯
S(t)=1-f(t_1)-f(t_2)=1-\alpha(t_1)- \alpha(t_2)S(t_2)=(1-\alpha(t_1)(1-\alpha(t_2)), \: t_2 < t \le t_3 \\ \cdots
S(t)=1−f(t1)−f(t2)=1−α(t1)−α(t2)S(t2)=(1−α(t1)(1−α(t2)),t2<t≤t3⋯
统一
记连续情况下
A
(
t
)
=
∫
0
t
α
(
s
)
d
s
A(t) = \int_0^t \alpha(s) \mathrm{d}s
A(t)=∫0tα(s)ds
离散情况下
A
(
t
)
=
∑
0
t
α
(
s
)
,
A(t) =\sum_0^t \alpha(s),
A(t)=0∑tα(s),
这里的
∑
\sum
∑请用勒贝格积分理解, 二者在实变函数下统一为
A
(
t
)
=
∫
0
t
1
S
(
s
)
d
S
(
s
)
.
A(t) = \int_0^t \frac{1}{S(s)} \mathrm{d}S(s).
A(t)=∫0tS(s)1dS(s).
A
(
t
+
h
)
−
A
(
t
)
A(t+h)-A(t)
A(t+h)−A(t)可以理解为个体在
[
t
,
t
+
h
]
[t,t+h]
[t,t+h]内死亡的概率, 则
S
(
t
)
=
lim
max
∣
t
i
−
t
i
−
1
∣
→
0
∏
0
t
(
1
−
(
A
(
t
i
)
−
A
(
t
i
−
1
)
)
=
:
∏
0
t
(
1
−
d
A
(
s
)
)
S(t)= \lim_{\max |t_i - t_{i-1}| \rightarrow 0} \prod_0^t (1-(A(t_i)-A(t_{i-1}))=:\prod_0^t (1-dA(s))
S(t)=max∣ti−ti−1∣→0lim0∏t(1−(A(ti)−A(ti−1))=:0∏t(1−dA(s))
意思就是, 个体想活过
t
t
t, 必须前面的每一个阶段都是活着的(严格的推导, 以及极限存在等等不知).
还有在矩阵和马尔可夫上的推广, 一知半解, 就不记录了.