Seoul National University Instructor: Junho Song Dept. of Civil and Environmental Engineering junhosong@snu.ac.kr
1
457.643 Structural Random Vibrations In-Class Material: Class 02
I. Basic Elements (Contd.)
Joint characteristic function: alternative complete description to ______ PDF 𝑀𝑿(𝜽) ≡ E𝑿{exp[𝑖𝜽T𝑿]}
= E𝑿{exp[𝑖(𝜃1𝑋1+ 𝜃2𝑋2+ ⋯ + 𝜃𝑛𝑋𝑛)]}
= ∫ ⋯ ∫ exp[𝑖(𝜃1𝑥1+ 𝜃2𝑥2+ ⋯ + 𝜃𝑛𝑥𝑛)] 𝑓𝑿(𝒙)𝑑𝒙
m____variate F______ transform of _____ PDF
Therefore,
𝑓𝑿(𝒙) =( )1 𝑛∫ ⋯ ∫ exp[−𝑖(𝜃1𝑥1+ 𝜃2𝑥2+ ⋯ + 𝜃𝑛𝑥𝑛)] 𝑑𝜽
Can show __________-generating property for the joint characteristic function, i.e.
1 𝑖𝑚1+⋯+𝑚𝑛
𝜕𝑚1+⋯+𝑚𝑛𝑀𝑿(𝜽)
𝜕𝜃1𝑚1⋯ 𝜃𝑛𝑚𝑛 |
𝜽=𝟎
= E[𝑋1𝑚1𝑋2𝑚2⋯ 𝑋𝑛𝑚𝑛]
Some observations:
1) Consistency rule: 𝑀𝑿(𝜃1, ⋯ , 𝜃𝑘, 0, ⋯ ,0) =
2) For statistically independent random variables, 𝑀𝑿(𝜽) =
Joint log characteristic function
Remember 𝑀𝑋(𝜃) = E𝑋[exp(𝑖𝜃𝑋)] = ∫ exp(𝑖𝜃𝑋) 𝑓−∞∞ 𝑋(𝑥)𝑑𝑥 Joint log characteristic function 𝐿𝑿(𝜽) =
Joint cumulant function 𝜅(𝑿) = 1
𝑖𝑛
𝜕𝑛𝐿𝑿(𝜽)
𝜕𝜃1⋯ 𝜃𝑛|
𝜽=𝟎
𝜅(𝑋𝑖) =
κ(𝑋𝑖, 𝑋𝑗) =
κ(𝑋𝑖, 𝑋𝑗, 𝑋𝑘) =
Seoul National University Instructor: Junho Song Dept. of Civil and Environmental Engineering junhosong@snu.ac.kr
2
Multivariate normal (Gaussian) distribution Joint PDF:
𝑓𝑿(𝒙) = 1
(2𝜋)𝑛/2√|det 𝚺|exp [−1
2(𝒙 − 𝑴)T𝚺−1(𝒙 − 𝑴)]
completely determined by _____ and ______ order moments
denoted by 𝑿~𝑁(𝑴, 𝜮)
can show 𝒀 = 𝑨𝑿 + 𝒃 ~ 𝑁(𝑨𝑴 + 𝒃, 𝑨𝜮𝑨T)
e.g. 𝑛 = 1, 𝑋~𝑁(𝜇, 𝜎2) 𝑓𝑋(𝑥) = 1
√2𝜋𝜎exp [−1 2(𝑥 − 𝜇
𝜎 )2] Can show
M𝐗(𝜽) = exp (𝑖𝑴T𝜽 −1
2𝜽T𝚺𝜽) LX(𝜽) =
_________ function of 𝜽
Higher order (𝑛 ≥ ) cumulants are zero
Example: κ(𝑋𝑖, 𝑋𝑗) for bivariate normal random variables
Example: Characteristic function of 𝑌 = 𝑋1+ 𝑋2+ ⋯ + 𝑋𝑛 (and PDF?)
Seoul National University Instructor: Junho Song Dept. of Civil and Environmental Engineering junhosong@snu.ac.kr
3
Proof of Central Limit Theorem using characteristic functions
Consider 𝑍 = 𝑋1+ 𝑋2+ ⋯ + 𝑋𝑛 where 𝑋𝑖, 𝑖 = 1, … , 𝑛 are statistically independent, identically distributed (SIID) random variables. Try
𝑍′= 1
√𝑛∑ (𝑋𝑖− 𝜇 𝜎 )
𝑛
𝑖=1
where 𝜇 and 𝜎 respectively denote the common mean and standard deviation of 𝑋𝑖’s.
Let 𝑌𝑖 =𝑋𝑖−𝜇
𝜎 Then, 𝑍′= 1
√𝑛∑𝑛𝑖=1𝑌𝑖
The characteristic function of 𝑍′ is then derived as 𝑀𝑍′(𝜃) = E[exp(i𝜃𝑍′)] = E {exp [i𝜃
√𝑛∑ 𝑌𝑗
𝑛 𝑗=1
]} = E [∏ exp (i𝜃𝑌𝑗
√𝑛)
𝑛 𝑗=1
]
= ∏ E [exp (i𝜃𝑌𝑗
√𝑛)]
𝑛 𝑗=1
= ∏ 𝑀𝑌𝑗(𝜃
√𝑛)
𝑛 𝑗=1
= [𝑀𝑌(𝜃
√𝑛)]
𝑛
Let us consider the characteristic function of 𝑌, 𝑀𝑌(𝜃). Note that the mean of 𝑌 is zero and its standard deviation is one. From the moment generating property of the characteristic function, we confirm
𝑑𝑀𝑌(𝜃) 𝑑𝜃 |
𝜃=0
= iE[𝑌] = 𝑖 ⋅ 𝜇𝑌= 0
𝑑2𝑀𝑌(𝜃) 𝑑𝜃2 |
𝜃=0
= i2E[𝑌2] = −(𝜎𝑌2+ 𝜇𝑌2) = −1
Therefore, the characteristic function 𝑀𝑌(𝜃) can be constructed by a Taylor series:
𝑀𝑌(𝜃) = 1 −𝜃2
2 + 𝑜(𝜃2) 𝑀𝑌(𝜃
√𝑛) = 1 −𝜃2
2𝑛+ 𝑜 (𝜃2 𝑛) 𝑀𝑍′(𝜃) = [1 −𝜃2
2𝑛+ 𝑜 (𝜃2 𝑛)]
𝑛
Since lim
n→∞(1 +𝑥
𝑛)𝑛= 𝑒𝑥,
𝑛→∞lim 𝑀𝑍′(𝜃) = exp (−𝜃2 2)
The end result is the characteristic function of the standard ______ distribution. Thus, we hereby proved that 𝑍′ asymptotically follows the standard _______ distribution as 𝑛 → ∞.
Since 𝑍 is a linear function of 𝑍′, 𝑍 also asymptotically follows a _______ distribution.
statistically independent identically distributed
Seoul National University Instructor: Junho Song Dept. of Civil and Environmental Engineering junhosong@snu.ac.kr
4 II. Introduction to Random Process
II-1. Random Process
Definitions
Random (stochastic) process: {𝑋(𝑡)} or 𝑋(𝑡) (cf. 𝑥(𝑡))
e.g. earthquake ground motion
Definition 1: Random process is an
“e_______” (collection) of possible t____ h______ {𝑥(1)(𝑡), 𝑥(2)(𝑡), … }
Definition 2: “Continuously indexed” r_______ v________, or a family of random variables {𝑋(0), … 𝑋(𝑡𝑘), … 𝑋(𝑡𝑚), … }
Note: the concept of random process can be generalized
1) Random field 𝑋(𝑡, 𝑢, 𝑣), e.g. wind pressure at location (𝑢, 𝑣) of the roof at time 𝑡 2) Vector random process:
𝐗(𝑡) = { 𝑋
1(𝑡) 𝑋
2(𝑡)
⋮ 𝑋
𝑛(𝑡)
} e.g. 𝐗
g(𝑡) = { 𝑥
𝑔(𝑡) 𝑥̇
𝑔(𝑡) 𝑥̈
𝑔(𝑡) }
3)
Vector random field:
𝐗(𝑡, 𝑢, 𝑣) = {
𝑋
1(𝑡, 𝑢, 𝑣) 𝑋
2(𝑡, 𝑢, 𝑣)
⋮ 𝑋
𝑛(𝑡, 𝑢, 𝑣)
}
Figure credit: http://www.wind.arch.t-
kougei.ac.jp/info_center/windpressure/lowrise/Intro ductionofthedatabase.pdf